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 | 41 | #include <net/net_namespace.h> |
8f408ab6 DJ |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
324e227e | 44 | #include <linux/hashtable.h> |
b2cbae2c | 45 | #include <rdma/rdma_netlink.h> |
03db3a2d MB |
46 | #include <rdma/ib_addr.h> |
47 | #include <rdma/ib_cache.h> | |
413d3347 | 48 | #include <rdma/rdma_counter.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 | ||
9cd58817 | 96 | static u32 highest_client_id; |
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 | 100 | |
621e55ff JG |
101 | static void ib_client_put(struct ib_client *client) |
102 | { | |
103 | if (refcount_dec_and_test(&client->uses)) | |
104 | complete(&client->uses_zero); | |
105 | } | |
106 | ||
1da177e4 | 107 | /* |
0df91bb6 JG |
108 | * If client_data is registered then the corresponding client must also still |
109 | * be registered. | |
110 | */ | |
111 | #define CLIENT_DATA_REGISTERED XA_MARK_1 | |
4e0f7b90 | 112 | |
1d2fedd8 | 113 | unsigned int rdma_dev_net_id; |
4e0f7b90 PP |
114 | |
115 | /* | |
116 | * A list of net namespaces is maintained in an xarray. This is necessary | |
117 | * because we can't get the locking right using the existing net ns list. We | |
118 | * would require a init_net callback after the list is updated. | |
119 | */ | |
120 | static DEFINE_XARRAY_FLAGS(rdma_nets, XA_FLAGS_ALLOC); | |
121 | /* | |
122 | * rwsem to protect accessing the rdma_nets xarray entries. | |
123 | */ | |
124 | static DECLARE_RWSEM(rdma_nets_rwsem); | |
125 | ||
cb7e0e13 | 126 | bool ib_devices_shared_netns = true; |
a56bc45b PP |
127 | module_param_named(netns_mode, ib_devices_shared_netns, bool, 0444); |
128 | MODULE_PARM_DESC(netns_mode, | |
129 | "Share device among net namespaces; default=1 (shared)"); | |
41c61401 | 130 | /** |
d6537c1a | 131 | * rdma_dev_access_netns() - Return whether an rdma device can be accessed |
41c61401 | 132 | * from a specified net namespace or not. |
d6537c1a | 133 | * @dev: Pointer to rdma device which needs to be checked |
41c61401 PP |
134 | * @net: Pointer to net namesapce for which access to be checked |
135 | * | |
d6537c1a | 136 | * When the rdma device is in shared mode, it ignores the net namespace. |
137 | * When the rdma device is exclusive to a net namespace, rdma device net | |
138 | * namespace is checked against the specified one. | |
41c61401 PP |
139 | */ |
140 | bool rdma_dev_access_netns(const struct ib_device *dev, const struct net *net) | |
141 | { | |
142 | return (ib_devices_shared_netns || | |
143 | net_eq(read_pnet(&dev->coredev.rdma_net), net)); | |
144 | } | |
145 | EXPORT_SYMBOL(rdma_dev_access_netns); | |
146 | ||
0df91bb6 JG |
147 | /* |
148 | * xarray has this behavior where it won't iterate over NULL values stored in | |
149 | * allocated arrays. So we need our own iterator to see all values stored in | |
150 | * the array. This does the same thing as xa_for_each except that it also | |
151 | * returns NULL valued entries if the array is allocating. Simplified to only | |
152 | * work on simple xarrays. | |
153 | */ | |
154 | static void *xan_find_marked(struct xarray *xa, unsigned long *indexp, | |
155 | xa_mark_t filter) | |
156 | { | |
157 | XA_STATE(xas, xa, *indexp); | |
158 | void *entry; | |
159 | ||
160 | rcu_read_lock(); | |
161 | do { | |
162 | entry = xas_find_marked(&xas, ULONG_MAX, filter); | |
163 | if (xa_is_zero(entry)) | |
164 | break; | |
165 | } while (xas_retry(&xas, entry)); | |
166 | rcu_read_unlock(); | |
167 | ||
168 | if (entry) { | |
169 | *indexp = xas.xa_index; | |
170 | if (xa_is_zero(entry)) | |
171 | return NULL; | |
172 | return entry; | |
173 | } | |
174 | return XA_ERROR(-ENOENT); | |
175 | } | |
176 | #define xan_for_each_marked(xa, index, entry, filter) \ | |
177 | for (index = 0, entry = xan_find_marked(xa, &(index), filter); \ | |
178 | !xa_is_err(entry); \ | |
179 | (index)++, entry = xan_find_marked(xa, &(index), filter)) | |
180 | ||
324e227e JG |
181 | /* RCU hash table mapping netdevice pointers to struct ib_port_data */ |
182 | static DEFINE_SPINLOCK(ndev_hash_lock); | |
183 | static DECLARE_HASHTABLE(ndev_hash, 5); | |
184 | ||
c2261dd7 | 185 | static void free_netdevs(struct ib_device *ib_dev); |
d0899892 JG |
186 | static void ib_unregister_work(struct work_struct *work); |
187 | static void __ib_unregister_device(struct ib_device *device); | |
8f408ab6 DJ |
188 | static int ib_security_change(struct notifier_block *nb, unsigned long event, |
189 | void *lsm_data); | |
190 | static void ib_policy_change_task(struct work_struct *work); | |
191 | static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task); | |
192 | ||
923abb9d GP |
193 | static void __ibdev_printk(const char *level, const struct ib_device *ibdev, |
194 | struct va_format *vaf) | |
195 | { | |
196 | if (ibdev && ibdev->dev.parent) | |
197 | dev_printk_emit(level[1] - '0', | |
198 | ibdev->dev.parent, | |
199 | "%s %s %s: %pV", | |
200 | dev_driver_string(ibdev->dev.parent), | |
201 | dev_name(ibdev->dev.parent), | |
202 | dev_name(&ibdev->dev), | |
203 | vaf); | |
204 | else if (ibdev) | |
205 | printk("%s%s: %pV", | |
206 | level, dev_name(&ibdev->dev), vaf); | |
207 | else | |
208 | printk("%s(NULL ib_device): %pV", level, vaf); | |
209 | } | |
210 | ||
211 | void ibdev_printk(const char *level, const struct ib_device *ibdev, | |
212 | const char *format, ...) | |
213 | { | |
214 | struct va_format vaf; | |
215 | va_list args; | |
216 | ||
217 | va_start(args, format); | |
218 | ||
219 | vaf.fmt = format; | |
220 | vaf.va = &args; | |
221 | ||
222 | __ibdev_printk(level, ibdev, &vaf); | |
223 | ||
224 | va_end(args); | |
225 | } | |
226 | EXPORT_SYMBOL(ibdev_printk); | |
227 | ||
228 | #define define_ibdev_printk_level(func, level) \ | |
229 | void func(const struct ib_device *ibdev, const char *fmt, ...) \ | |
230 | { \ | |
231 | struct va_format vaf; \ | |
232 | va_list args; \ | |
233 | \ | |
234 | va_start(args, fmt); \ | |
235 | \ | |
236 | vaf.fmt = fmt; \ | |
237 | vaf.va = &args; \ | |
238 | \ | |
239 | __ibdev_printk(level, ibdev, &vaf); \ | |
240 | \ | |
241 | va_end(args); \ | |
242 | } \ | |
243 | EXPORT_SYMBOL(func); | |
244 | ||
245 | define_ibdev_printk_level(ibdev_emerg, KERN_EMERG); | |
246 | define_ibdev_printk_level(ibdev_alert, KERN_ALERT); | |
247 | define_ibdev_printk_level(ibdev_crit, KERN_CRIT); | |
248 | define_ibdev_printk_level(ibdev_err, KERN_ERR); | |
249 | define_ibdev_printk_level(ibdev_warn, KERN_WARNING); | |
250 | define_ibdev_printk_level(ibdev_notice, KERN_NOTICE); | |
251 | define_ibdev_printk_level(ibdev_info, KERN_INFO); | |
252 | ||
8f408ab6 DJ |
253 | static struct notifier_block ibdev_lsm_nb = { |
254 | .notifier_call = ib_security_change, | |
255 | }; | |
1da177e4 | 256 | |
decbc7a6 PP |
257 | static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, |
258 | struct net *net); | |
259 | ||
324e227e JG |
260 | /* Pointer to the RCU head at the start of the ib_port_data array */ |
261 | struct ib_port_data_rcu { | |
262 | struct rcu_head rcu_head; | |
263 | struct ib_port_data pdata[]; | |
264 | }; | |
265 | ||
deee3c7e | 266 | static void ib_device_check_mandatory(struct ib_device *device) |
1da177e4 | 267 | { |
3023a1e9 | 268 | #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device_ops, x), #x } |
1da177e4 LT |
269 | static const struct { |
270 | size_t offset; | |
271 | char *name; | |
272 | } mandatory_table[] = { | |
273 | IB_MANDATORY_FUNC(query_device), | |
274 | IB_MANDATORY_FUNC(query_port), | |
1da177e4 LT |
275 | IB_MANDATORY_FUNC(alloc_pd), |
276 | IB_MANDATORY_FUNC(dealloc_pd), | |
1da177e4 LT |
277 | IB_MANDATORY_FUNC(create_qp), |
278 | IB_MANDATORY_FUNC(modify_qp), | |
279 | IB_MANDATORY_FUNC(destroy_qp), | |
280 | IB_MANDATORY_FUNC(post_send), | |
281 | IB_MANDATORY_FUNC(post_recv), | |
282 | IB_MANDATORY_FUNC(create_cq), | |
283 | IB_MANDATORY_FUNC(destroy_cq), | |
284 | IB_MANDATORY_FUNC(poll_cq), | |
285 | IB_MANDATORY_FUNC(req_notify_cq), | |
286 | IB_MANDATORY_FUNC(get_dma_mr), | |
44ce37bc | 287 | IB_MANDATORY_FUNC(reg_user_mr), |
7738613e IW |
288 | IB_MANDATORY_FUNC(dereg_mr), |
289 | IB_MANDATORY_FUNC(get_port_immutable) | |
1da177e4 LT |
290 | }; |
291 | int i; | |
292 | ||
6780c4fa | 293 | device->kverbs_provider = true; |
9a6b090c | 294 | for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { |
3023a1e9 KH |
295 | if (!*(void **) ((void *) &device->ops + |
296 | mandatory_table[i].offset)) { | |
6780c4fa GP |
297 | device->kverbs_provider = false; |
298 | break; | |
1da177e4 LT |
299 | } |
300 | } | |
1da177e4 LT |
301 | } |
302 | ||
f8978bd9 | 303 | /* |
01b67117 PP |
304 | * Caller must perform ib_device_put() to return the device reference count |
305 | * when ib_device_get_by_index() returns valid device pointer. | |
f8978bd9 | 306 | */ |
37eeab55 | 307 | struct ib_device *ib_device_get_by_index(const struct net *net, u32 index) |
f8978bd9 LR |
308 | { |
309 | struct ib_device *device; | |
310 | ||
921eab11 | 311 | down_read(&devices_rwsem); |
0df91bb6 | 312 | device = xa_load(&devices, index); |
01b67117 | 313 | if (device) { |
37eeab55 PP |
314 | if (!rdma_dev_access_netns(device, net)) { |
315 | device = NULL; | |
316 | goto out; | |
317 | } | |
318 | ||
d79af724 | 319 | if (!ib_device_try_get(device)) |
01b67117 PP |
320 | device = NULL; |
321 | } | |
37eeab55 | 322 | out: |
921eab11 | 323 | up_read(&devices_rwsem); |
f8978bd9 LR |
324 | return device; |
325 | } | |
326 | ||
d79af724 JG |
327 | /** |
328 | * ib_device_put - Release IB device reference | |
329 | * @device: device whose reference to be released | |
330 | * | |
331 | * ib_device_put() releases reference to the IB device to allow it to be | |
332 | * unregistered and eventually free. | |
333 | */ | |
01b67117 PP |
334 | void ib_device_put(struct ib_device *device) |
335 | { | |
336 | if (refcount_dec_and_test(&device->refcount)) | |
337 | complete(&device->unreg_completion); | |
338 | } | |
d79af724 | 339 | EXPORT_SYMBOL(ib_device_put); |
01b67117 | 340 | |
1da177e4 LT |
341 | static struct ib_device *__ib_device_get_by_name(const char *name) |
342 | { | |
343 | struct ib_device *device; | |
0df91bb6 | 344 | unsigned long index; |
1da177e4 | 345 | |
0df91bb6 | 346 | xa_for_each (&devices, index, device) |
896de009 | 347 | if (!strcmp(name, dev_name(&device->dev))) |
1da177e4 LT |
348 | return device; |
349 | ||
350 | return NULL; | |
351 | } | |
352 | ||
6cc2c8e5 JG |
353 | /** |
354 | * ib_device_get_by_name - Find an IB device by name | |
355 | * @name: The name to look for | |
356 | * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) | |
357 | * | |
358 | * Find and hold an ib_device by its name. The caller must call | |
359 | * ib_device_put() on the returned pointer. | |
360 | */ | |
361 | struct ib_device *ib_device_get_by_name(const char *name, | |
362 | enum rdma_driver_id driver_id) | |
363 | { | |
364 | struct ib_device *device; | |
365 | ||
366 | down_read(&devices_rwsem); | |
367 | device = __ib_device_get_by_name(name); | |
368 | if (device && driver_id != RDMA_DRIVER_UNKNOWN && | |
b9560a41 | 369 | device->ops.driver_id != driver_id) |
6cc2c8e5 JG |
370 | device = NULL; |
371 | ||
372 | if (device) { | |
373 | if (!ib_device_try_get(device)) | |
374 | device = NULL; | |
375 | } | |
376 | up_read(&devices_rwsem); | |
377 | return device; | |
378 | } | |
379 | EXPORT_SYMBOL(ib_device_get_by_name); | |
380 | ||
4e0f7b90 PP |
381 | static int rename_compat_devs(struct ib_device *device) |
382 | { | |
383 | struct ib_core_device *cdev; | |
384 | unsigned long index; | |
385 | int ret = 0; | |
386 | ||
387 | mutex_lock(&device->compat_devs_mutex); | |
388 | xa_for_each (&device->compat_devs, index, cdev) { | |
389 | ret = device_rename(&cdev->dev, dev_name(&device->dev)); | |
390 | if (ret) { | |
391 | dev_warn(&cdev->dev, | |
392 | "Fail to rename compatdev to new name %s\n", | |
393 | dev_name(&device->dev)); | |
394 | break; | |
395 | } | |
396 | } | |
397 | mutex_unlock(&device->compat_devs_mutex); | |
398 | return ret; | |
399 | } | |
400 | ||
d21943dd LR |
401 | int ib_device_rename(struct ib_device *ibdev, const char *name) |
402 | { | |
dc1435c0 LR |
403 | unsigned long index; |
404 | void *client_data; | |
e3593b56 | 405 | int ret; |
d21943dd | 406 | |
921eab11 | 407 | down_write(&devices_rwsem); |
e3593b56 | 408 | if (!strcmp(name, dev_name(&ibdev->dev))) { |
dc1435c0 LR |
409 | up_write(&devices_rwsem); |
410 | return 0; | |
e3593b56 JG |
411 | } |
412 | ||
344684e6 | 413 | if (__ib_device_get_by_name(name)) { |
dc1435c0 LR |
414 | up_write(&devices_rwsem); |
415 | return -EEXIST; | |
d21943dd LR |
416 | } |
417 | ||
418 | ret = device_rename(&ibdev->dev, name); | |
dc1435c0 LR |
419 | if (ret) { |
420 | up_write(&devices_rwsem); | |
421 | return ret; | |
422 | } | |
423 | ||
d21943dd | 424 | strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); |
4e0f7b90 | 425 | ret = rename_compat_devs(ibdev); |
dc1435c0 LR |
426 | |
427 | downgrade_write(&devices_rwsem); | |
428 | down_read(&ibdev->client_data_rwsem); | |
429 | xan_for_each_marked(&ibdev->client_data, index, client_data, | |
430 | CLIENT_DATA_REGISTERED) { | |
431 | struct ib_client *client = xa_load(&clients, index); | |
432 | ||
433 | if (!client || !client->rename) | |
434 | continue; | |
435 | ||
436 | client->rename(ibdev, client_data); | |
437 | } | |
438 | up_read(&ibdev->client_data_rwsem); | |
439 | up_read(&devices_rwsem); | |
440 | return 0; | |
d21943dd LR |
441 | } |
442 | ||
f8fc8cd9 YF |
443 | int ib_device_set_dim(struct ib_device *ibdev, u8 use_dim) |
444 | { | |
445 | if (use_dim > 1) | |
446 | return -EINVAL; | |
447 | ibdev->use_cq_dim = use_dim; | |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
e349f858 | 452 | static int alloc_name(struct ib_device *ibdev, const char *name) |
1da177e4 | 453 | { |
1da177e4 | 454 | struct ib_device *device; |
0df91bb6 | 455 | unsigned long index; |
3b88afd3 JG |
456 | struct ida inuse; |
457 | int rc; | |
1da177e4 LT |
458 | int i; |
459 | ||
9ffbe8ac | 460 | lockdep_assert_held_write(&devices_rwsem); |
3b88afd3 | 461 | ida_init(&inuse); |
0df91bb6 | 462 | xa_for_each (&devices, index, device) { |
e349f858 JG |
463 | char buf[IB_DEVICE_NAME_MAX]; |
464 | ||
896de009 | 465 | if (sscanf(dev_name(&device->dev), name, &i) != 1) |
1da177e4 | 466 | continue; |
3b88afd3 | 467 | if (i < 0 || i >= INT_MAX) |
1da177e4 LT |
468 | continue; |
469 | snprintf(buf, sizeof buf, name, i); | |
3b88afd3 JG |
470 | if (strcmp(buf, dev_name(&device->dev)) != 0) |
471 | continue; | |
472 | ||
473 | rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); | |
474 | if (rc < 0) | |
475 | goto out; | |
1da177e4 LT |
476 | } |
477 | ||
3b88afd3 JG |
478 | rc = ida_alloc(&inuse, GFP_KERNEL); |
479 | if (rc < 0) | |
480 | goto out; | |
1da177e4 | 481 | |
3b88afd3 JG |
482 | rc = dev_set_name(&ibdev->dev, name, rc); |
483 | out: | |
484 | ida_destroy(&inuse); | |
485 | return rc; | |
1da177e4 LT |
486 | } |
487 | ||
55aeed06 JG |
488 | static void ib_device_release(struct device *device) |
489 | { | |
490 | struct ib_device *dev = container_of(device, struct ib_device, dev); | |
491 | ||
c2261dd7 | 492 | free_netdevs(dev); |
652432f3 | 493 | WARN_ON(refcount_read(&dev->refcount)); |
46bdf370 KH |
494 | if (dev->port_data) { |
495 | ib_cache_release_one(dev); | |
496 | ib_security_release_port_pkey_list(dev); | |
413d3347 | 497 | rdma_counter_release(dev); |
324e227e JG |
498 | kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, |
499 | pdata[0]), | |
500 | rcu_head); | |
46bdf370 | 501 | } |
413d3347 | 502 | |
56594ae1 PP |
503 | mutex_destroy(&dev->unregistration_lock); |
504 | mutex_destroy(&dev->compat_devs_mutex); | |
505 | ||
46bdf370 KH |
506 | xa_destroy(&dev->compat_devs); |
507 | xa_destroy(&dev->client_data); | |
324e227e | 508 | kfree_rcu(dev, rcu_head); |
55aeed06 JG |
509 | } |
510 | ||
511 | static int ib_device_uevent(struct device *device, | |
512 | struct kobj_uevent_env *env) | |
513 | { | |
896de009 | 514 | if (add_uevent_var(env, "NAME=%s", dev_name(device))) |
55aeed06 JG |
515 | return -ENOMEM; |
516 | ||
517 | /* | |
518 | * It would be nice to pass the node GUID with the event... | |
519 | */ | |
520 | ||
521 | return 0; | |
522 | } | |
523 | ||
62dfa795 PP |
524 | static const void *net_namespace(struct device *d) |
525 | { | |
4e0f7b90 PP |
526 | struct ib_core_device *coredev = |
527 | container_of(d, struct ib_core_device, dev); | |
528 | ||
529 | return read_pnet(&coredev->rdma_net); | |
62dfa795 PP |
530 | } |
531 | ||
55aeed06 JG |
532 | static struct class ib_class = { |
533 | .name = "infiniband", | |
534 | .dev_release = ib_device_release, | |
535 | .dev_uevent = ib_device_uevent, | |
62dfa795 PP |
536 | .ns_type = &net_ns_type_operations, |
537 | .namespace = net_namespace, | |
55aeed06 JG |
538 | }; |
539 | ||
cebe556b | 540 | static void rdma_init_coredev(struct ib_core_device *coredev, |
4e0f7b90 | 541 | struct ib_device *dev, struct net *net) |
cebe556b PP |
542 | { |
543 | /* This BUILD_BUG_ON is intended to catch layout change | |
544 | * of union of ib_core_device and device. | |
545 | * dev must be the first element as ib_core and providers | |
546 | * driver uses it. Adding anything in ib_core_device before | |
547 | * device will break this assumption. | |
548 | */ | |
549 | BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != | |
550 | offsetof(struct ib_device, dev)); | |
551 | ||
552 | coredev->dev.class = &ib_class; | |
553 | coredev->dev.groups = dev->groups; | |
554 | device_initialize(&coredev->dev); | |
555 | coredev->owner = dev; | |
556 | INIT_LIST_HEAD(&coredev->port_list); | |
4e0f7b90 | 557 | write_pnet(&coredev->rdma_net, net); |
cebe556b PP |
558 | } |
559 | ||
1da177e4 | 560 | /** |
459cc69f | 561 | * _ib_alloc_device - allocate an IB device struct |
1da177e4 LT |
562 | * @size:size of structure to allocate |
563 | * | |
564 | * Low-level drivers should use ib_alloc_device() to allocate &struct | |
565 | * ib_device. @size is the size of the structure to be allocated, | |
566 | * including any private data used by the low-level driver. | |
567 | * ib_dealloc_device() must be used to free structures allocated with | |
568 | * ib_alloc_device(). | |
569 | */ | |
459cc69f | 570 | struct ib_device *_ib_alloc_device(size_t size) |
1da177e4 | 571 | { |
55aeed06 JG |
572 | struct ib_device *device; |
573 | ||
574 | if (WARN_ON(size < sizeof(struct ib_device))) | |
575 | return NULL; | |
576 | ||
577 | device = kzalloc(size, GFP_KERNEL); | |
578 | if (!device) | |
579 | return NULL; | |
580 | ||
41eda65c LR |
581 | if (rdma_restrack_init(device)) { |
582 | kfree(device); | |
583 | return NULL; | |
584 | } | |
02d8883f | 585 | |
5f8f5499 | 586 | device->groups[0] = &ib_dev_attr_group; |
4e0f7b90 | 587 | rdma_init_coredev(&device->coredev, device, &init_net); |
55aeed06 | 588 | |
55aeed06 | 589 | INIT_LIST_HEAD(&device->event_handler_list); |
40adf686 | 590 | spin_lock_init(&device->qp_open_list_lock); |
6b57cea9 | 591 | init_rwsem(&device->event_handler_rwsem); |
d0899892 | 592 | mutex_init(&device->unregistration_lock); |
0df91bb6 JG |
593 | /* |
594 | * client_data needs to be alloc because we don't want our mark to be | |
595 | * destroyed if the user stores NULL in the client data. | |
596 | */ | |
597 | xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); | |
921eab11 | 598 | init_rwsem(&device->client_data_rwsem); |
4e0f7b90 PP |
599 | xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); |
600 | mutex_init(&device->compat_devs_mutex); | |
01b67117 | 601 | init_completion(&device->unreg_completion); |
d0899892 | 602 | INIT_WORK(&device->unregistration_work, ib_unregister_work); |
1da177e4 | 603 | |
c074bb1e | 604 | device->uverbs_cmd_mask = |
44ce37bc | 605 | BIT_ULL(IB_USER_VERBS_CMD_ALLOC_MW) | |
c074bb1e | 606 | BIT_ULL(IB_USER_VERBS_CMD_ALLOC_PD) | |
44ce37bc JG |
607 | BIT_ULL(IB_USER_VERBS_CMD_ATTACH_MCAST) | |
608 | BIT_ULL(IB_USER_VERBS_CMD_CLOSE_XRCD) | | |
676a80ad | 609 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_AH) | |
c074bb1e JG |
610 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | |
611 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_CQ) | | |
612 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_QP) | | |
44ce37bc | 613 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_SRQ) | |
652caba5 | 614 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_XSRQ) | |
44ce37bc | 615 | BIT_ULL(IB_USER_VERBS_CMD_DEALLOC_MW) | |
c074bb1e JG |
616 | BIT_ULL(IB_USER_VERBS_CMD_DEALLOC_PD) | |
617 | BIT_ULL(IB_USER_VERBS_CMD_DEREG_MR) | | |
676a80ad | 618 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_AH) | |
c074bb1e JG |
619 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_CQ) | |
620 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_QP) | | |
44ce37bc JG |
621 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_SRQ) | |
622 | BIT_ULL(IB_USER_VERBS_CMD_DETACH_MCAST) | | |
c074bb1e JG |
623 | BIT_ULL(IB_USER_VERBS_CMD_GET_CONTEXT) | |
624 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_QP) | | |
44ce37bc JG |
625 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_SRQ) | |
626 | BIT_ULL(IB_USER_VERBS_CMD_OPEN_QP) | | |
627 | BIT_ULL(IB_USER_VERBS_CMD_OPEN_XRCD) | | |
c074bb1e JG |
628 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_DEVICE) | |
629 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_PORT) | | |
630 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_QP) | | |
44ce37bc JG |
631 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_SRQ) | |
632 | BIT_ULL(IB_USER_VERBS_CMD_REG_MR) | | |
633 | BIT_ULL(IB_USER_VERBS_CMD_REREG_MR) | | |
634 | BIT_ULL(IB_USER_VERBS_CMD_RESIZE_CQ); | |
55aeed06 | 635 | return device; |
1da177e4 | 636 | } |
459cc69f | 637 | EXPORT_SYMBOL(_ib_alloc_device); |
1da177e4 LT |
638 | |
639 | /** | |
640 | * ib_dealloc_device - free an IB device struct | |
641 | * @device:structure to free | |
642 | * | |
643 | * Free a structure allocated with ib_alloc_device(). | |
644 | */ | |
645 | void ib_dealloc_device(struct ib_device *device) | |
646 | { | |
d0899892 JG |
647 | if (device->ops.dealloc_driver) |
648 | device->ops.dealloc_driver(device); | |
649 | ||
650 | /* | |
651 | * ib_unregister_driver() requires all devices to remain in the xarray | |
652 | * while their ops are callable. The last op we call is dealloc_driver | |
653 | * above. This is needed to create a fence on op callbacks prior to | |
654 | * allowing the driver module to unload. | |
655 | */ | |
656 | down_write(&devices_rwsem); | |
657 | if (xa_load(&devices, device->index) == device) | |
658 | xa_erase(&devices, device->index); | |
659 | up_write(&devices_rwsem); | |
660 | ||
c2261dd7 JG |
661 | /* Expedite releasing netdev references */ |
662 | free_netdevs(device); | |
663 | ||
4e0f7b90 | 664 | WARN_ON(!xa_empty(&device->compat_devs)); |
0df91bb6 | 665 | WARN_ON(!xa_empty(&device->client_data)); |
652432f3 | 666 | WARN_ON(refcount_read(&device->refcount)); |
0ad699c0 | 667 | rdma_restrack_clean(device); |
e155755e | 668 | /* Balances with device_initialize */ |
924b8900 | 669 | put_device(&device->dev); |
1da177e4 LT |
670 | } |
671 | EXPORT_SYMBOL(ib_dealloc_device); | |
672 | ||
921eab11 JG |
673 | /* |
674 | * add_client_context() and remove_client_context() must be safe against | |
675 | * parallel calls on the same device - registration/unregistration of both the | |
676 | * device and client can be occurring in parallel. | |
677 | * | |
678 | * The routines need to be a fence, any caller must not return until the add | |
679 | * or remove is fully completed. | |
680 | */ | |
681 | static int add_client_context(struct ib_device *device, | |
682 | struct ib_client *client) | |
1da177e4 | 683 | { |
921eab11 | 684 | int ret = 0; |
1da177e4 | 685 | |
6780c4fa | 686 | if (!device->kverbs_provider && !client->no_kverbs_req) |
921eab11 JG |
687 | return 0; |
688 | ||
689 | down_write(&device->client_data_rwsem); | |
621e55ff JG |
690 | /* |
691 | * So long as the client is registered hold both the client and device | |
692 | * unregistration locks. | |
693 | */ | |
694 | if (!refcount_inc_not_zero(&client->uses)) | |
695 | goto out_unlock; | |
696 | refcount_inc(&device->refcount); | |
697 | ||
921eab11 JG |
698 | /* |
699 | * Another caller to add_client_context got here first and has already | |
700 | * completely initialized context. | |
701 | */ | |
702 | if (xa_get_mark(&device->client_data, client->client_id, | |
703 | CLIENT_DATA_REGISTERED)) | |
704 | goto out; | |
705 | ||
706 | ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, | |
707 | GFP_KERNEL)); | |
708 | if (ret) | |
709 | goto out; | |
710 | downgrade_write(&device->client_data_rwsem); | |
11a0ae4c JG |
711 | if (client->add) { |
712 | if (client->add(device)) { | |
713 | /* | |
714 | * If a client fails to add then the error code is | |
715 | * ignored, but we won't call any more ops on this | |
716 | * client. | |
717 | */ | |
718 | xa_erase(&device->client_data, client->client_id); | |
719 | up_read(&device->client_data_rwsem); | |
720 | ib_device_put(device); | |
721 | ib_client_put(client); | |
722 | return 0; | |
723 | } | |
724 | } | |
921eab11 JG |
725 | |
726 | /* Readers shall not see a client until add has been completed */ | |
727 | xa_set_mark(&device->client_data, client->client_id, | |
728 | CLIENT_DATA_REGISTERED); | |
729 | up_read(&device->client_data_rwsem); | |
730 | return 0; | |
731 | ||
732 | out: | |
621e55ff JG |
733 | ib_device_put(device); |
734 | ib_client_put(client); | |
735 | out_unlock: | |
921eab11 JG |
736 | up_write(&device->client_data_rwsem); |
737 | return ret; | |
738 | } | |
739 | ||
740 | static void remove_client_context(struct ib_device *device, | |
741 | unsigned int client_id) | |
742 | { | |
743 | struct ib_client *client; | |
744 | void *client_data; | |
6780c4fa | 745 | |
921eab11 JG |
746 | down_write(&device->client_data_rwsem); |
747 | if (!xa_get_mark(&device->client_data, client_id, | |
748 | CLIENT_DATA_REGISTERED)) { | |
749 | up_write(&device->client_data_rwsem); | |
750 | return; | |
751 | } | |
752 | client_data = xa_load(&device->client_data, client_id); | |
753 | xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); | |
754 | client = xa_load(&clients, client_id); | |
621e55ff | 755 | up_write(&device->client_data_rwsem); |
1da177e4 | 756 | |
921eab11 JG |
757 | /* |
758 | * Notice we cannot be holding any exclusive locks when calling the | |
759 | * remove callback as the remove callback can recurse back into any | |
760 | * public functions in this module and thus try for any locks those | |
761 | * functions take. | |
762 | * | |
763 | * For this reason clients and drivers should not call the | |
764 | * unregistration functions will holdling any locks. | |
921eab11 JG |
765 | */ |
766 | if (client->remove) | |
767 | client->remove(device, client_data); | |
768 | ||
769 | xa_erase(&device->client_data, client_id); | |
621e55ff JG |
770 | ib_device_put(device); |
771 | ib_client_put(client); | |
1da177e4 LT |
772 | } |
773 | ||
c2261dd7 | 774 | static int alloc_port_data(struct ib_device *device) |
5eb620c8 | 775 | { |
324e227e | 776 | struct ib_port_data_rcu *pdata_rcu; |
ea1075ed | 777 | unsigned int port; |
c2261dd7 JG |
778 | |
779 | if (device->port_data) | |
780 | return 0; | |
781 | ||
782 | /* This can only be called once the physical port range is defined */ | |
783 | if (WARN_ON(!device->phys_port_cnt)) | |
784 | return -EINVAL; | |
7738613e | 785 | |
8ceb1357 JG |
786 | /* |
787 | * device->port_data is indexed directly by the port number to make | |
7738613e IW |
788 | * access to this data as efficient as possible. |
789 | * | |
8ceb1357 JG |
790 | * Therefore port_data is declared as a 1 based array with potential |
791 | * empty slots at the beginning. | |
7738613e | 792 | */ |
324e227e JG |
793 | pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, |
794 | rdma_end_port(device) + 1), | |
795 | GFP_KERNEL); | |
796 | if (!pdata_rcu) | |
55aeed06 | 797 | return -ENOMEM; |
324e227e JG |
798 | /* |
799 | * The rcu_head is put in front of the port data array and the stored | |
800 | * pointer is adjusted since we never need to see that member until | |
801 | * kfree_rcu. | |
802 | */ | |
803 | device->port_data = pdata_rcu->pdata; | |
5eb620c8 | 804 | |
ea1075ed | 805 | rdma_for_each_port (device, port) { |
8ceb1357 JG |
806 | struct ib_port_data *pdata = &device->port_data[port]; |
807 | ||
324e227e | 808 | pdata->ib_dev = device; |
8ceb1357 JG |
809 | spin_lock_init(&pdata->pkey_list_lock); |
810 | INIT_LIST_HEAD(&pdata->pkey_list); | |
c2261dd7 | 811 | spin_lock_init(&pdata->netdev_lock); |
324e227e | 812 | INIT_HLIST_NODE(&pdata->ndev_hash_link); |
c2261dd7 JG |
813 | } |
814 | return 0; | |
815 | } | |
816 | ||
817 | static int verify_immutable(const struct ib_device *dev, u8 port) | |
818 | { | |
819 | return WARN_ON(!rdma_cap_ib_mad(dev, port) && | |
820 | rdma_max_mad_size(dev, port) != 0); | |
821 | } | |
822 | ||
823 | static int setup_port_data(struct ib_device *device) | |
824 | { | |
825 | unsigned int port; | |
826 | int ret; | |
827 | ||
828 | ret = alloc_port_data(device); | |
829 | if (ret) | |
830 | return ret; | |
831 | ||
832 | rdma_for_each_port (device, port) { | |
833 | struct ib_port_data *pdata = &device->port_data[port]; | |
8ceb1357 JG |
834 | |
835 | ret = device->ops.get_port_immutable(device, port, | |
836 | &pdata->immutable); | |
5eb620c8 | 837 | if (ret) |
55aeed06 | 838 | return ret; |
337877a4 | 839 | |
55aeed06 JG |
840 | if (verify_immutable(device, port)) |
841 | return -EINVAL; | |
5eb620c8 | 842 | } |
55aeed06 | 843 | return 0; |
5eb620c8 YE |
844 | } |
845 | ||
9abb0d1b | 846 | void ib_get_device_fw_str(struct ib_device *dev, char *str) |
5fa76c20 | 847 | { |
3023a1e9 KH |
848 | if (dev->ops.get_dev_fw_str) |
849 | dev->ops.get_dev_fw_str(dev, str); | |
5fa76c20 IW |
850 | else |
851 | str[0] = '\0'; | |
852 | } | |
853 | EXPORT_SYMBOL(ib_get_device_fw_str); | |
854 | ||
8f408ab6 DJ |
855 | static void ib_policy_change_task(struct work_struct *work) |
856 | { | |
857 | struct ib_device *dev; | |
0df91bb6 | 858 | unsigned long index; |
8f408ab6 | 859 | |
921eab11 | 860 | down_read(&devices_rwsem); |
0df91bb6 | 861 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { |
ea1075ed | 862 | unsigned int i; |
8f408ab6 | 863 | |
ea1075ed | 864 | rdma_for_each_port (dev, i) { |
8f408ab6 DJ |
865 | u64 sp; |
866 | int ret = ib_get_cached_subnet_prefix(dev, | |
867 | i, | |
868 | &sp); | |
869 | ||
870 | WARN_ONCE(ret, | |
871 | "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", | |
872 | ret); | |
a750cfde DJ |
873 | if (!ret) |
874 | ib_security_cache_change(dev, i, sp); | |
8f408ab6 DJ |
875 | } |
876 | } | |
921eab11 | 877 | up_read(&devices_rwsem); |
8f408ab6 DJ |
878 | } |
879 | ||
880 | static int ib_security_change(struct notifier_block *nb, unsigned long event, | |
881 | void *lsm_data) | |
882 | { | |
883 | if (event != LSM_POLICY_CHANGE) | |
884 | return NOTIFY_DONE; | |
885 | ||
886 | schedule_work(&ib_policy_change_work); | |
c66f6741 | 887 | ib_mad_agent_security_change(); |
8f408ab6 DJ |
888 | |
889 | return NOTIFY_OK; | |
890 | } | |
891 | ||
4e0f7b90 PP |
892 | static void compatdev_release(struct device *dev) |
893 | { | |
894 | struct ib_core_device *cdev = | |
895 | container_of(dev, struct ib_core_device, dev); | |
896 | ||
897 | kfree(cdev); | |
898 | } | |
899 | ||
900 | static int add_one_compat_dev(struct ib_device *device, | |
901 | struct rdma_dev_net *rnet) | |
902 | { | |
903 | struct ib_core_device *cdev; | |
904 | int ret; | |
905 | ||
2b34c558 | 906 | lockdep_assert_held(&rdma_nets_rwsem); |
a56bc45b PP |
907 | if (!ib_devices_shared_netns) |
908 | return 0; | |
909 | ||
4e0f7b90 PP |
910 | /* |
911 | * Create and add compat device in all namespaces other than where it | |
912 | * is currently bound to. | |
913 | */ | |
914 | if (net_eq(read_pnet(&rnet->net), | |
915 | read_pnet(&device->coredev.rdma_net))) | |
916 | return 0; | |
917 | ||
918 | /* | |
919 | * The first of init_net() or ib_register_device() to take the | |
920 | * compat_devs_mutex wins and gets to add the device. Others will wait | |
921 | * for completion here. | |
922 | */ | |
923 | mutex_lock(&device->compat_devs_mutex); | |
924 | cdev = xa_load(&device->compat_devs, rnet->id); | |
925 | if (cdev) { | |
926 | ret = 0; | |
927 | goto done; | |
928 | } | |
929 | ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); | |
930 | if (ret) | |
931 | goto done; | |
932 | ||
933 | cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); | |
934 | if (!cdev) { | |
935 | ret = -ENOMEM; | |
936 | goto cdev_err; | |
937 | } | |
938 | ||
939 | cdev->dev.parent = device->dev.parent; | |
940 | rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); | |
941 | cdev->dev.release = compatdev_release; | |
f2f2b3bb JG |
942 | ret = dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); |
943 | if (ret) | |
944 | goto add_err; | |
4e0f7b90 PP |
945 | |
946 | ret = device_add(&cdev->dev); | |
947 | if (ret) | |
948 | goto add_err; | |
eb15c78b | 949 | ret = ib_setup_port_attrs(cdev); |
5417783e PP |
950 | if (ret) |
951 | goto port_err; | |
4e0f7b90 PP |
952 | |
953 | ret = xa_err(xa_store(&device->compat_devs, rnet->id, | |
954 | cdev, GFP_KERNEL)); | |
955 | if (ret) | |
956 | goto insert_err; | |
957 | ||
958 | mutex_unlock(&device->compat_devs_mutex); | |
959 | return 0; | |
960 | ||
961 | insert_err: | |
5417783e PP |
962 | ib_free_port_attrs(cdev); |
963 | port_err: | |
4e0f7b90 PP |
964 | device_del(&cdev->dev); |
965 | add_err: | |
966 | put_device(&cdev->dev); | |
967 | cdev_err: | |
968 | xa_release(&device->compat_devs, rnet->id); | |
969 | done: | |
970 | mutex_unlock(&device->compat_devs_mutex); | |
971 | return ret; | |
972 | } | |
973 | ||
974 | static void remove_one_compat_dev(struct ib_device *device, u32 id) | |
975 | { | |
976 | struct ib_core_device *cdev; | |
977 | ||
978 | mutex_lock(&device->compat_devs_mutex); | |
979 | cdev = xa_erase(&device->compat_devs, id); | |
980 | mutex_unlock(&device->compat_devs_mutex); | |
981 | if (cdev) { | |
5417783e | 982 | ib_free_port_attrs(cdev); |
4e0f7b90 PP |
983 | device_del(&cdev->dev); |
984 | put_device(&cdev->dev); | |
985 | } | |
986 | } | |
987 | ||
988 | static void remove_compat_devs(struct ib_device *device) | |
989 | { | |
990 | struct ib_core_device *cdev; | |
991 | unsigned long index; | |
992 | ||
993 | xa_for_each (&device->compat_devs, index, cdev) | |
994 | remove_one_compat_dev(device, index); | |
995 | } | |
996 | ||
997 | static int add_compat_devs(struct ib_device *device) | |
998 | { | |
999 | struct rdma_dev_net *rnet; | |
1000 | unsigned long index; | |
1001 | int ret = 0; | |
1002 | ||
decbc7a6 PP |
1003 | lockdep_assert_held(&devices_rwsem); |
1004 | ||
4e0f7b90 PP |
1005 | down_read(&rdma_nets_rwsem); |
1006 | xa_for_each (&rdma_nets, index, rnet) { | |
1007 | ret = add_one_compat_dev(device, rnet); | |
1008 | if (ret) | |
1009 | break; | |
1010 | } | |
1011 | up_read(&rdma_nets_rwsem); | |
1012 | return ret; | |
1013 | } | |
1014 | ||
2b34c558 PP |
1015 | static void remove_all_compat_devs(void) |
1016 | { | |
1017 | struct ib_compat_device *cdev; | |
1018 | struct ib_device *dev; | |
1019 | unsigned long index; | |
1020 | ||
1021 | down_read(&devices_rwsem); | |
1022 | xa_for_each (&devices, index, dev) { | |
1023 | unsigned long c_index = 0; | |
1024 | ||
1025 | /* Hold nets_rwsem so that any other thread modifying this | |
1026 | * system param can sync with this thread. | |
1027 | */ | |
1028 | down_read(&rdma_nets_rwsem); | |
1029 | xa_for_each (&dev->compat_devs, c_index, cdev) | |
1030 | remove_one_compat_dev(dev, c_index); | |
1031 | up_read(&rdma_nets_rwsem); | |
1032 | } | |
1033 | up_read(&devices_rwsem); | |
1034 | } | |
1035 | ||
1036 | static int add_all_compat_devs(void) | |
1037 | { | |
1038 | struct rdma_dev_net *rnet; | |
1039 | struct ib_device *dev; | |
1040 | unsigned long index; | |
1041 | int ret = 0; | |
1042 | ||
1043 | down_read(&devices_rwsem); | |
1044 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { | |
1045 | unsigned long net_index = 0; | |
1046 | ||
1047 | /* Hold nets_rwsem so that any other thread modifying this | |
1048 | * system param can sync with this thread. | |
1049 | */ | |
1050 | down_read(&rdma_nets_rwsem); | |
1051 | xa_for_each (&rdma_nets, net_index, rnet) { | |
1052 | ret = add_one_compat_dev(dev, rnet); | |
1053 | if (ret) | |
1054 | break; | |
1055 | } | |
1056 | up_read(&rdma_nets_rwsem); | |
1057 | } | |
1058 | up_read(&devices_rwsem); | |
1059 | if (ret) | |
1060 | remove_all_compat_devs(); | |
1061 | return ret; | |
1062 | } | |
1063 | ||
1064 | int rdma_compatdev_set(u8 enable) | |
1065 | { | |
1066 | struct rdma_dev_net *rnet; | |
1067 | unsigned long index; | |
1068 | int ret = 0; | |
1069 | ||
1070 | down_write(&rdma_nets_rwsem); | |
1071 | if (ib_devices_shared_netns == enable) { | |
1072 | up_write(&rdma_nets_rwsem); | |
1073 | return 0; | |
1074 | } | |
1075 | ||
1076 | /* enable/disable of compat devices is not supported | |
1077 | * when more than default init_net exists. | |
1078 | */ | |
1079 | xa_for_each (&rdma_nets, index, rnet) { | |
1080 | ret++; | |
1081 | break; | |
1082 | } | |
1083 | if (!ret) | |
1084 | ib_devices_shared_netns = enable; | |
1085 | up_write(&rdma_nets_rwsem); | |
1086 | if (ret) | |
1087 | return -EBUSY; | |
1088 | ||
1089 | if (enable) | |
1090 | ret = add_all_compat_devs(); | |
1091 | else | |
1092 | remove_all_compat_devs(); | |
1093 | return ret; | |
1094 | } | |
1095 | ||
4e0f7b90 PP |
1096 | static void rdma_dev_exit_net(struct net *net) |
1097 | { | |
1d2fedd8 | 1098 | struct rdma_dev_net *rnet = rdma_net_to_dev_net(net); |
4e0f7b90 PP |
1099 | struct ib_device *dev; |
1100 | unsigned long index; | |
1101 | int ret; | |
1102 | ||
1103 | down_write(&rdma_nets_rwsem); | |
1104 | /* | |
1105 | * Prevent the ID from being re-used and hide the id from xa_for_each. | |
1106 | */ | |
1107 | ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); | |
1108 | WARN_ON(ret); | |
1109 | up_write(&rdma_nets_rwsem); | |
1110 | ||
1111 | down_read(&devices_rwsem); | |
1112 | xa_for_each (&devices, index, dev) { | |
1113 | get_device(&dev->dev); | |
1114 | /* | |
1115 | * Release the devices_rwsem so that pontentially blocking | |
1116 | * device_del, doesn't hold the devices_rwsem for too long. | |
1117 | */ | |
1118 | up_read(&devices_rwsem); | |
1119 | ||
1120 | remove_one_compat_dev(dev, rnet->id); | |
1121 | ||
decbc7a6 PP |
1122 | /* |
1123 | * If the real device is in the NS then move it back to init. | |
1124 | */ | |
1125 | rdma_dev_change_netns(dev, net, &init_net); | |
1126 | ||
4e0f7b90 PP |
1127 | put_device(&dev->dev); |
1128 | down_read(&devices_rwsem); | |
1129 | } | |
1130 | up_read(&devices_rwsem); | |
1131 | ||
1d2fedd8 | 1132 | rdma_nl_net_exit(rnet); |
4e0f7b90 PP |
1133 | xa_erase(&rdma_nets, rnet->id); |
1134 | } | |
1135 | ||
1136 | static __net_init int rdma_dev_init_net(struct net *net) | |
1137 | { | |
1d2fedd8 | 1138 | struct rdma_dev_net *rnet = rdma_net_to_dev_net(net); |
4e0f7b90 PP |
1139 | unsigned long index; |
1140 | struct ib_device *dev; | |
1141 | int ret; | |
1142 | ||
1d2fedd8 PP |
1143 | write_pnet(&rnet->net, net); |
1144 | ||
1145 | ret = rdma_nl_net_init(rnet); | |
1146 | if (ret) | |
1147 | return ret; | |
1148 | ||
4e0f7b90 PP |
1149 | /* No need to create any compat devices in default init_net. */ |
1150 | if (net_eq(net, &init_net)) | |
1151 | return 0; | |
1152 | ||
4e0f7b90 | 1153 | ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); |
1d2fedd8 PP |
1154 | if (ret) { |
1155 | rdma_nl_net_exit(rnet); | |
4e0f7b90 | 1156 | return ret; |
1d2fedd8 | 1157 | } |
4e0f7b90 PP |
1158 | |
1159 | down_read(&devices_rwsem); | |
1160 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { | |
2b34c558 PP |
1161 | /* Hold nets_rwsem so that netlink command cannot change |
1162 | * system configuration for device sharing mode. | |
1163 | */ | |
1164 | down_read(&rdma_nets_rwsem); | |
4e0f7b90 | 1165 | ret = add_one_compat_dev(dev, rnet); |
2b34c558 | 1166 | up_read(&rdma_nets_rwsem); |
4e0f7b90 PP |
1167 | if (ret) |
1168 | break; | |
1169 | } | |
1170 | up_read(&devices_rwsem); | |
1171 | ||
1172 | if (ret) | |
1173 | rdma_dev_exit_net(net); | |
1174 | ||
1175 | return ret; | |
1176 | } | |
1177 | ||
0df91bb6 | 1178 | /* |
d0899892 JG |
1179 | * Assign the unique string device name and the unique device index. This is |
1180 | * undone by ib_dealloc_device. | |
ecc82c53 | 1181 | */ |
0df91bb6 | 1182 | static int assign_name(struct ib_device *device, const char *name) |
ecc82c53 | 1183 | { |
0df91bb6 JG |
1184 | static u32 last_id; |
1185 | int ret; | |
ecc82c53 | 1186 | |
921eab11 | 1187 | down_write(&devices_rwsem); |
0df91bb6 JG |
1188 | /* Assign a unique name to the device */ |
1189 | if (strchr(name, '%')) | |
1190 | ret = alloc_name(device, name); | |
1191 | else | |
1192 | ret = dev_set_name(&device->dev, name); | |
1193 | if (ret) | |
1194 | goto out; | |
1195 | ||
1196 | if (__ib_device_get_by_name(dev_name(&device->dev))) { | |
1197 | ret = -ENFILE; | |
1198 | goto out; | |
1199 | } | |
1200 | strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); | |
ecc82c53 | 1201 | |
ea295481 LT |
1202 | ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, |
1203 | &last_id, GFP_KERNEL); | |
1204 | if (ret > 0) | |
1205 | ret = 0; | |
921eab11 | 1206 | |
0df91bb6 | 1207 | out: |
921eab11 | 1208 | up_write(&devices_rwsem); |
0df91bb6 JG |
1209 | return ret; |
1210 | } | |
1211 | ||
e0477b34 JG |
1212 | static void setup_dma_device(struct ib_device *device, |
1213 | struct device *dma_device) | |
1da177e4 | 1214 | { |
e0477b34 JG |
1215 | /* |
1216 | * If the caller does not provide a DMA capable device then the IB | |
1217 | * device will be used. In this case the caller should fully setup the | |
1218 | * ibdev for DMA. This usually means using dma_virt_ops. | |
1219 | */ | |
1220 | #ifdef CONFIG_DMA_VIRT_OPS | |
1221 | if (!dma_device) { | |
1222 | device->dev.dma_ops = &dma_virt_ops; | |
1223 | dma_device = &device->dev; | |
c9121262 | 1224 | } |
e0477b34 JG |
1225 | #endif |
1226 | WARN_ON(!dma_device); | |
1227 | device->dma_device = dma_device; | |
1228 | WARN_ON(!device->dma_device->dma_parms); | |
548cb4fb | 1229 | } |
1da177e4 | 1230 | |
921eab11 JG |
1231 | /* |
1232 | * setup_device() allocates memory and sets up data that requires calling the | |
1233 | * device ops, this is the only reason these actions are not done during | |
1234 | * ib_alloc_device. It is undone by ib_dealloc_device(). | |
1235 | */ | |
548cb4fb PP |
1236 | static int setup_device(struct ib_device *device) |
1237 | { | |
1238 | struct ib_udata uhw = {.outlen = 0, .inlen = 0}; | |
1239 | int ret; | |
1da177e4 | 1240 | |
deee3c7e | 1241 | ib_device_check_mandatory(device); |
1da177e4 | 1242 | |
8ceb1357 | 1243 | ret = setup_port_data(device); |
5eb620c8 | 1244 | if (ret) { |
8ceb1357 | 1245 | dev_warn(&device->dev, "Couldn't create per-port data\n"); |
548cb4fb PP |
1246 | return ret; |
1247 | } | |
1248 | ||
1249 | memset(&device->attrs, 0, sizeof(device->attrs)); | |
3023a1e9 | 1250 | ret = device->ops.query_device(device, &device->attrs, &uhw); |
548cb4fb PP |
1251 | if (ret) { |
1252 | dev_warn(&device->dev, | |
1253 | "Couldn't query the device attributes\n"); | |
d45f89d5 | 1254 | return ret; |
5eb620c8 YE |
1255 | } |
1256 | ||
d45f89d5 | 1257 | return 0; |
548cb4fb PP |
1258 | } |
1259 | ||
921eab11 JG |
1260 | static void disable_device(struct ib_device *device) |
1261 | { | |
9cd58817 | 1262 | u32 cid; |
921eab11 JG |
1263 | |
1264 | WARN_ON(!refcount_read(&device->refcount)); | |
1265 | ||
1266 | down_write(&devices_rwsem); | |
1267 | xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); | |
1268 | up_write(&devices_rwsem); | |
1269 | ||
9cd58817 JG |
1270 | /* |
1271 | * Remove clients in LIFO order, see assign_client_id. This could be | |
1272 | * more efficient if xarray learns to reverse iterate. Since no new | |
1273 | * clients can be added to this ib_device past this point we only need | |
1274 | * the maximum possible client_id value here. | |
1275 | */ | |
921eab11 | 1276 | down_read(&clients_rwsem); |
9cd58817 | 1277 | cid = highest_client_id; |
921eab11 | 1278 | up_read(&clients_rwsem); |
9cd58817 JG |
1279 | while (cid) { |
1280 | cid--; | |
1281 | remove_client_context(device, cid); | |
1282 | } | |
921eab11 | 1283 | |
4aa16152 JG |
1284 | ib_cq_pool_destroy(device); |
1285 | ||
921eab11 JG |
1286 | /* Pairs with refcount_set in enable_device */ |
1287 | ib_device_put(device); | |
1288 | wait_for_completion(&device->unreg_completion); | |
c2261dd7 | 1289 | |
4e0f7b90 PP |
1290 | /* |
1291 | * compat devices must be removed after device refcount drops to zero. | |
1292 | * Otherwise init_net() may add more compatdevs after removing compat | |
1293 | * devices and before device is disabled. | |
1294 | */ | |
1295 | remove_compat_devs(device); | |
921eab11 JG |
1296 | } |
1297 | ||
1298 | /* | |
1299 | * An enabled device is visible to all clients and to all the public facing | |
d0899892 JG |
1300 | * APIs that return a device pointer. This always returns with a new get, even |
1301 | * if it fails. | |
921eab11 | 1302 | */ |
d0899892 | 1303 | static int enable_device_and_get(struct ib_device *device) |
921eab11 JG |
1304 | { |
1305 | struct ib_client *client; | |
1306 | unsigned long index; | |
d0899892 | 1307 | int ret = 0; |
921eab11 | 1308 | |
d0899892 JG |
1309 | /* |
1310 | * One ref belongs to the xa and the other belongs to this | |
1311 | * thread. This is needed to guard against parallel unregistration. | |
1312 | */ | |
1313 | refcount_set(&device->refcount, 2); | |
921eab11 JG |
1314 | down_write(&devices_rwsem); |
1315 | xa_set_mark(&devices, device->index, DEVICE_REGISTERED); | |
d0899892 JG |
1316 | |
1317 | /* | |
1318 | * By using downgrade_write() we ensure that no other thread can clear | |
1319 | * DEVICE_REGISTERED while we are completing the client setup. | |
1320 | */ | |
1321 | downgrade_write(&devices_rwsem); | |
921eab11 | 1322 | |
ca22354b JG |
1323 | if (device->ops.enable_driver) { |
1324 | ret = device->ops.enable_driver(device); | |
1325 | if (ret) | |
1326 | goto out; | |
1327 | } | |
1328 | ||
4aa16152 JG |
1329 | ib_cq_pool_init(device); |
1330 | ||
921eab11 JG |
1331 | down_read(&clients_rwsem); |
1332 | xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { | |
1333 | ret = add_client_context(device, client); | |
d0899892 JG |
1334 | if (ret) |
1335 | break; | |
921eab11 JG |
1336 | } |
1337 | up_read(&clients_rwsem); | |
4e0f7b90 PP |
1338 | if (!ret) |
1339 | ret = add_compat_devs(device); | |
ca22354b | 1340 | out: |
d0899892 JG |
1341 | up_read(&devices_rwsem); |
1342 | return ret; | |
921eab11 JG |
1343 | } |
1344 | ||
0cb42c02 JG |
1345 | static void prevent_dealloc_device(struct ib_device *ib_dev) |
1346 | { | |
1347 | } | |
1348 | ||
548cb4fb PP |
1349 | /** |
1350 | * ib_register_device - Register an IB device with IB core | |
d6537c1a | 1351 | * @device: Device to register |
1352 | * @name: unique string device name. This may include a '%' which will | |
e0477b34 JG |
1353 | * cause a unique index to be added to the passed device name. |
1354 | * @dma_device: pointer to a DMA-capable device. If %NULL, then the IB | |
1355 | * device will be used. In this case the caller should fully | |
1356 | * setup the ibdev for DMA. This usually means using dma_virt_ops. | |
548cb4fb PP |
1357 | * |
1358 | * Low-level drivers use ib_register_device() to register their | |
1359 | * devices with the IB core. All registered clients will receive a | |
1360 | * callback for each device that is added. @device must be allocated | |
1361 | * with ib_alloc_device(). | |
d0899892 JG |
1362 | * |
1363 | * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() | |
1364 | * asynchronously then the device pointer may become freed as soon as this | |
1365 | * function returns. | |
548cb4fb | 1366 | */ |
e0477b34 JG |
1367 | int ib_register_device(struct ib_device *device, const char *name, |
1368 | struct device *dma_device) | |
548cb4fb PP |
1369 | { |
1370 | int ret; | |
548cb4fb | 1371 | |
0df91bb6 JG |
1372 | ret = assign_name(device, name); |
1373 | if (ret) | |
921eab11 | 1374 | return ret; |
548cb4fb | 1375 | |
e0477b34 | 1376 | setup_dma_device(device, dma_device); |
548cb4fb PP |
1377 | ret = setup_device(device); |
1378 | if (ret) | |
d0899892 | 1379 | return ret; |
03db3a2d | 1380 | |
d45f89d5 JG |
1381 | ret = ib_cache_setup_one(device); |
1382 | if (ret) { | |
1383 | dev_warn(&device->dev, | |
1384 | "Couldn't set up InfiniBand P_Key/GID cache\n"); | |
d0899892 | 1385 | return ret; |
d45f89d5 JG |
1386 | } |
1387 | ||
7527a7b1 | 1388 | ib_device_register_rdmacg(device); |
3e153a93 | 1389 | |
413d3347 MZ |
1390 | rdma_counter_init(device); |
1391 | ||
e7a5b4aa LR |
1392 | /* |
1393 | * Ensure that ADD uevent is not fired because it | |
1394 | * is too early amd device is not initialized yet. | |
1395 | */ | |
1396 | dev_set_uevent_suppress(&device->dev, true); | |
5f8f5499 PP |
1397 | ret = device_add(&device->dev); |
1398 | if (ret) | |
1399 | goto cg_cleanup; | |
1400 | ||
ea4baf7f | 1401 | ret = ib_device_register_sysfs(device); |
1da177e4 | 1402 | if (ret) { |
43c7c851 JG |
1403 | dev_warn(&device->dev, |
1404 | "Couldn't register device with driver model\n"); | |
5f8f5499 | 1405 | goto dev_cleanup; |
1da177e4 LT |
1406 | } |
1407 | ||
d0899892 | 1408 | ret = enable_device_and_get(device); |
e7a5b4aa LR |
1409 | dev_set_uevent_suppress(&device->dev, false); |
1410 | /* Mark for userspace that device is ready */ | |
1411 | kobject_uevent(&device->dev.kobj, KOBJ_ADD); | |
d0899892 JG |
1412 | if (ret) { |
1413 | void (*dealloc_fn)(struct ib_device *); | |
1414 | ||
1415 | /* | |
1416 | * If we hit this error flow then we don't want to | |
1417 | * automatically dealloc the device since the caller is | |
1418 | * expected to call ib_dealloc_device() after | |
1419 | * ib_register_device() fails. This is tricky due to the | |
1420 | * possibility for a parallel unregistration along with this | |
1421 | * error flow. Since we have a refcount here we know any | |
1422 | * parallel flow is stopped in disable_device and will see the | |
0cb42c02 | 1423 | * special dealloc_driver pointer, causing the responsibility to |
d0899892 JG |
1424 | * ib_dealloc_device() to revert back to this thread. |
1425 | */ | |
1426 | dealloc_fn = device->ops.dealloc_driver; | |
0cb42c02 | 1427 | device->ops.dealloc_driver = prevent_dealloc_device; |
d0899892 JG |
1428 | ib_device_put(device); |
1429 | __ib_unregister_device(device); | |
1430 | device->ops.dealloc_driver = dealloc_fn; | |
1431 | return ret; | |
1432 | } | |
1433 | ib_device_put(device); | |
1da177e4 | 1434 | |
4be3a4fa PP |
1435 | return 0; |
1436 | ||
5f8f5499 PP |
1437 | dev_cleanup: |
1438 | device_del(&device->dev); | |
2fb4f4ea | 1439 | cg_cleanup: |
e7a5b4aa | 1440 | dev_set_uevent_suppress(&device->dev, false); |
2fb4f4ea | 1441 | ib_device_unregister_rdmacg(device); |
d45f89d5 | 1442 | ib_cache_cleanup_one(device); |
1da177e4 LT |
1443 | return ret; |
1444 | } | |
1445 | EXPORT_SYMBOL(ib_register_device); | |
1446 | ||
d0899892 JG |
1447 | /* Callers must hold a get on the device. */ |
1448 | static void __ib_unregister_device(struct ib_device *ib_dev) | |
1449 | { | |
1450 | /* | |
1451 | * We have a registration lock so that all the calls to unregister are | |
1452 | * fully fenced, once any unregister returns the device is truely | |
1453 | * unregistered even if multiple callers are unregistering it at the | |
1454 | * same time. This also interacts with the registration flow and | |
1455 | * provides sane semantics if register and unregister are racing. | |
1456 | */ | |
1457 | mutex_lock(&ib_dev->unregistration_lock); | |
1458 | if (!refcount_read(&ib_dev->refcount)) | |
1459 | goto out; | |
1460 | ||
1461 | disable_device(ib_dev); | |
3042492b PP |
1462 | |
1463 | /* Expedite removing unregistered pointers from the hash table */ | |
1464 | free_netdevs(ib_dev); | |
1465 | ||
d0899892 JG |
1466 | ib_device_unregister_sysfs(ib_dev); |
1467 | device_del(&ib_dev->dev); | |
1468 | ib_device_unregister_rdmacg(ib_dev); | |
1469 | ib_cache_cleanup_one(ib_dev); | |
1470 | ||
1471 | /* | |
1472 | * Drivers using the new flow may not call ib_dealloc_device except | |
1473 | * in error unwind prior to registration success. | |
1474 | */ | |
0cb42c02 JG |
1475 | if (ib_dev->ops.dealloc_driver && |
1476 | ib_dev->ops.dealloc_driver != prevent_dealloc_device) { | |
d0899892 JG |
1477 | WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); |
1478 | ib_dealloc_device(ib_dev); | |
1479 | } | |
1480 | out: | |
1481 | mutex_unlock(&ib_dev->unregistration_lock); | |
1482 | } | |
1483 | ||
1da177e4 LT |
1484 | /** |
1485 | * ib_unregister_device - Unregister an IB device | |
d6537c1a | 1486 | * @ib_dev: The device to unregister |
1da177e4 LT |
1487 | * |
1488 | * Unregister an IB device. All clients will receive a remove callback. | |
d0899892 JG |
1489 | * |
1490 | * Callers should call this routine only once, and protect against races with | |
1491 | * registration. Typically it should only be called as part of a remove | |
1492 | * callback in an implementation of driver core's struct device_driver and | |
1493 | * related. | |
1494 | * | |
1495 | * If ops.dealloc_driver is used then ib_dev will be freed upon return from | |
1496 | * this function. | |
1da177e4 | 1497 | */ |
d0899892 | 1498 | void ib_unregister_device(struct ib_device *ib_dev) |
1da177e4 | 1499 | { |
d0899892 JG |
1500 | get_device(&ib_dev->dev); |
1501 | __ib_unregister_device(ib_dev); | |
1502 | put_device(&ib_dev->dev); | |
1da177e4 LT |
1503 | } |
1504 | EXPORT_SYMBOL(ib_unregister_device); | |
1505 | ||
d0899892 JG |
1506 | /** |
1507 | * ib_unregister_device_and_put - Unregister a device while holding a 'get' | |
d6537c1a | 1508 | * @ib_dev: The device to unregister |
d0899892 JG |
1509 | * |
1510 | * This is the same as ib_unregister_device(), except it includes an internal | |
1511 | * ib_device_put() that should match a 'get' obtained by the caller. | |
1512 | * | |
1513 | * It is safe to call this routine concurrently from multiple threads while | |
1514 | * holding the 'get'. When the function returns the device is fully | |
1515 | * unregistered. | |
1516 | * | |
1517 | * Drivers using this flow MUST use the driver_unregister callback to clean up | |
1518 | * their resources associated with the device and dealloc it. | |
1519 | */ | |
1520 | void ib_unregister_device_and_put(struct ib_device *ib_dev) | |
1521 | { | |
1522 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1523 | get_device(&ib_dev->dev); | |
1524 | ib_device_put(ib_dev); | |
1525 | __ib_unregister_device(ib_dev); | |
1526 | put_device(&ib_dev->dev); | |
1527 | } | |
1528 | EXPORT_SYMBOL(ib_unregister_device_and_put); | |
1529 | ||
1530 | /** | |
1531 | * ib_unregister_driver - Unregister all IB devices for a driver | |
1532 | * @driver_id: The driver to unregister | |
1533 | * | |
1534 | * This implements a fence for device unregistration. It only returns once all | |
1535 | * devices associated with the driver_id have fully completed their | |
1536 | * unregistration and returned from ib_unregister_device*(). | |
1537 | * | |
1538 | * If device's are not yet unregistered it goes ahead and starts unregistering | |
1539 | * them. | |
1540 | * | |
1541 | * This does not block creation of new devices with the given driver_id, that | |
1542 | * is the responsibility of the caller. | |
1543 | */ | |
1544 | void ib_unregister_driver(enum rdma_driver_id driver_id) | |
1545 | { | |
1546 | struct ib_device *ib_dev; | |
1547 | unsigned long index; | |
1548 | ||
1549 | down_read(&devices_rwsem); | |
1550 | xa_for_each (&devices, index, ib_dev) { | |
b9560a41 | 1551 | if (ib_dev->ops.driver_id != driver_id) |
d0899892 JG |
1552 | continue; |
1553 | ||
1554 | get_device(&ib_dev->dev); | |
1555 | up_read(&devices_rwsem); | |
1556 | ||
1557 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1558 | __ib_unregister_device(ib_dev); | |
1559 | ||
1560 | put_device(&ib_dev->dev); | |
1561 | down_read(&devices_rwsem); | |
1562 | } | |
1563 | up_read(&devices_rwsem); | |
1564 | } | |
1565 | EXPORT_SYMBOL(ib_unregister_driver); | |
1566 | ||
1567 | static void ib_unregister_work(struct work_struct *work) | |
1568 | { | |
1569 | struct ib_device *ib_dev = | |
1570 | container_of(work, struct ib_device, unregistration_work); | |
1571 | ||
1572 | __ib_unregister_device(ib_dev); | |
1573 | put_device(&ib_dev->dev); | |
1574 | } | |
1575 | ||
1576 | /** | |
1577 | * ib_unregister_device_queued - Unregister a device using a work queue | |
d6537c1a | 1578 | * @ib_dev: The device to unregister |
d0899892 JG |
1579 | * |
1580 | * This schedules an asynchronous unregistration using a WQ for the device. A | |
1581 | * driver should use this to avoid holding locks while doing unregistration, | |
1582 | * such as holding the RTNL lock. | |
1583 | * | |
1584 | * Drivers using this API must use ib_unregister_driver before module unload | |
1585 | * to ensure that all scheduled unregistrations have completed. | |
1586 | */ | |
1587 | void ib_unregister_device_queued(struct ib_device *ib_dev) | |
1588 | { | |
1589 | WARN_ON(!refcount_read(&ib_dev->refcount)); | |
1590 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1591 | get_device(&ib_dev->dev); | |
1592 | if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) | |
1593 | put_device(&ib_dev->dev); | |
1594 | } | |
1595 | EXPORT_SYMBOL(ib_unregister_device_queued); | |
1596 | ||
decbc7a6 PP |
1597 | /* |
1598 | * The caller must pass in a device that has the kref held and the refcount | |
1599 | * released. If the device is in cur_net and still registered then it is moved | |
1600 | * into net. | |
1601 | */ | |
1602 | static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, | |
1603 | struct net *net) | |
1604 | { | |
1605 | int ret2 = -EINVAL; | |
1606 | int ret; | |
1607 | ||
1608 | mutex_lock(&device->unregistration_lock); | |
1609 | ||
1610 | /* | |
2e5b8a01 PP |
1611 | * If a device not under ib_device_get() or if the unregistration_lock |
1612 | * is not held, the namespace can be changed, or it can be unregistered. | |
1613 | * Check again under the lock. | |
decbc7a6 PP |
1614 | */ |
1615 | if (refcount_read(&device->refcount) == 0 || | |
1616 | !net_eq(cur_net, read_pnet(&device->coredev.rdma_net))) { | |
1617 | ret = -ENODEV; | |
1618 | goto out; | |
1619 | } | |
1620 | ||
1621 | kobject_uevent(&device->dev.kobj, KOBJ_REMOVE); | |
1622 | disable_device(device); | |
1623 | ||
1624 | /* | |
1625 | * At this point no one can be using the device, so it is safe to | |
1626 | * change the namespace. | |
1627 | */ | |
1628 | write_pnet(&device->coredev.rdma_net, net); | |
1629 | ||
2e5b8a01 | 1630 | down_read(&devices_rwsem); |
decbc7a6 PP |
1631 | /* |
1632 | * Currently rdma devices are system wide unique. So the device name | |
1633 | * is guaranteed free in the new namespace. Publish the new namespace | |
1634 | * at the sysfs level. | |
1635 | */ | |
decbc7a6 PP |
1636 | ret = device_rename(&device->dev, dev_name(&device->dev)); |
1637 | up_read(&devices_rwsem); | |
1638 | if (ret) { | |
1639 | dev_warn(&device->dev, | |
1640 | "%s: Couldn't rename device after namespace change\n", | |
1641 | __func__); | |
1642 | /* Try and put things back and re-enable the device */ | |
1643 | write_pnet(&device->coredev.rdma_net, cur_net); | |
1644 | } | |
1645 | ||
1646 | ret2 = enable_device_and_get(device); | |
2e5b8a01 | 1647 | if (ret2) { |
decbc7a6 PP |
1648 | /* |
1649 | * This shouldn't really happen, but if it does, let the user | |
1650 | * retry at later point. So don't disable the device. | |
1651 | */ | |
1652 | dev_warn(&device->dev, | |
1653 | "%s: Couldn't re-enable device after namespace change\n", | |
1654 | __func__); | |
2e5b8a01 | 1655 | } |
decbc7a6 | 1656 | kobject_uevent(&device->dev.kobj, KOBJ_ADD); |
2e5b8a01 | 1657 | |
decbc7a6 PP |
1658 | ib_device_put(device); |
1659 | out: | |
1660 | mutex_unlock(&device->unregistration_lock); | |
1661 | if (ret) | |
1662 | return ret; | |
1663 | return ret2; | |
1664 | } | |
1665 | ||
2e5b8a01 PP |
1666 | int ib_device_set_netns_put(struct sk_buff *skb, |
1667 | struct ib_device *dev, u32 ns_fd) | |
1668 | { | |
1669 | struct net *net; | |
1670 | int ret; | |
1671 | ||
1672 | net = get_net_ns_by_fd(ns_fd); | |
1673 | if (IS_ERR(net)) { | |
1674 | ret = PTR_ERR(net); | |
1675 | goto net_err; | |
1676 | } | |
1677 | ||
1678 | if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) { | |
1679 | ret = -EPERM; | |
1680 | goto ns_err; | |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * Currently supported only for those providers which support | |
1685 | * disassociation and don't do port specific sysfs init. Once a | |
1686 | * port_cleanup infrastructure is implemented, this limitation will be | |
1687 | * removed. | |
1688 | */ | |
1689 | if (!dev->ops.disassociate_ucontext || dev->ops.init_port || | |
1690 | ib_devices_shared_netns) { | |
1691 | ret = -EOPNOTSUPP; | |
1692 | goto ns_err; | |
1693 | } | |
1694 | ||
1695 | get_device(&dev->dev); | |
1696 | ib_device_put(dev); | |
1697 | ret = rdma_dev_change_netns(dev, current->nsproxy->net_ns, net); | |
1698 | put_device(&dev->dev); | |
1699 | ||
1700 | put_net(net); | |
1701 | return ret; | |
1702 | ||
1703 | ns_err: | |
1704 | put_net(net); | |
1705 | net_err: | |
1706 | ib_device_put(dev); | |
1707 | return ret; | |
1708 | } | |
1709 | ||
4e0f7b90 PP |
1710 | static struct pernet_operations rdma_dev_net_ops = { |
1711 | .init = rdma_dev_init_net, | |
1712 | .exit = rdma_dev_exit_net, | |
1713 | .id = &rdma_dev_net_id, | |
1714 | .size = sizeof(struct rdma_dev_net), | |
1715 | }; | |
1716 | ||
e59178d8 JG |
1717 | static int assign_client_id(struct ib_client *client) |
1718 | { | |
1719 | int ret; | |
1720 | ||
921eab11 | 1721 | down_write(&clients_rwsem); |
e59178d8 JG |
1722 | /* |
1723 | * The add/remove callbacks must be called in FIFO/LIFO order. To | |
1724 | * achieve this we assign client_ids so they are sorted in | |
9cd58817 | 1725 | * registration order. |
e59178d8 | 1726 | */ |
9cd58817 | 1727 | client->client_id = highest_client_id; |
ea295481 | 1728 | ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); |
e59178d8 JG |
1729 | if (ret) |
1730 | goto out; | |
1731 | ||
9cd58817 | 1732 | highest_client_id++; |
921eab11 | 1733 | xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); |
921eab11 | 1734 | |
e59178d8 | 1735 | out: |
921eab11 | 1736 | up_write(&clients_rwsem); |
e59178d8 JG |
1737 | return ret; |
1738 | } | |
1739 | ||
9cd58817 JG |
1740 | static void remove_client_id(struct ib_client *client) |
1741 | { | |
1742 | down_write(&clients_rwsem); | |
1743 | xa_erase(&clients, client->client_id); | |
1744 | for (; highest_client_id; highest_client_id--) | |
1745 | if (xa_load(&clients, highest_client_id - 1)) | |
1746 | break; | |
1747 | up_write(&clients_rwsem); | |
1748 | } | |
1749 | ||
1da177e4 LT |
1750 | /** |
1751 | * ib_register_client - Register an IB client | |
1752 | * @client:Client to register | |
1753 | * | |
1754 | * Upper level users of the IB drivers can use ib_register_client() to | |
1755 | * register callbacks for IB device addition and removal. When an IB | |
1756 | * device is added, each registered client's add method will be called | |
1757 | * (in the order the clients were registered), and when a device is | |
1758 | * removed, each client's remove method will be called (in the reverse | |
1759 | * order that clients were registered). In addition, when | |
1760 | * ib_register_client() is called, the client will receive an add | |
1761 | * callback for all devices already registered. | |
1762 | */ | |
1763 | int ib_register_client(struct ib_client *client) | |
1764 | { | |
1765 | struct ib_device *device; | |
0df91bb6 | 1766 | unsigned long index; |
e59178d8 | 1767 | int ret; |
1da177e4 | 1768 | |
621e55ff JG |
1769 | refcount_set(&client->uses, 1); |
1770 | init_completion(&client->uses_zero); | |
e59178d8 | 1771 | ret = assign_client_id(client); |
921eab11 | 1772 | if (ret) |
e59178d8 | 1773 | return ret; |
1da177e4 | 1774 | |
921eab11 JG |
1775 | down_read(&devices_rwsem); |
1776 | xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { | |
1777 | ret = add_client_context(device, client); | |
1778 | if (ret) { | |
1779 | up_read(&devices_rwsem); | |
1780 | ib_unregister_client(client); | |
1781 | return ret; | |
1782 | } | |
1783 | } | |
1784 | up_read(&devices_rwsem); | |
1da177e4 LT |
1785 | return 0; |
1786 | } | |
1787 | EXPORT_SYMBOL(ib_register_client); | |
1788 | ||
1789 | /** | |
1790 | * ib_unregister_client - Unregister an IB client | |
1791 | * @client:Client to unregister | |
1792 | * | |
1793 | * Upper level users use ib_unregister_client() to remove their client | |
1794 | * registration. When ib_unregister_client() is called, the client | |
1795 | * will receive a remove callback for each IB device still registered. | |
921eab11 JG |
1796 | * |
1797 | * This is a full fence, once it returns no client callbacks will be called, | |
1798 | * or are running in another thread. | |
1da177e4 LT |
1799 | */ |
1800 | void ib_unregister_client(struct ib_client *client) | |
1801 | { | |
1da177e4 | 1802 | struct ib_device *device; |
0df91bb6 | 1803 | unsigned long index; |
1da177e4 | 1804 | |
921eab11 | 1805 | down_write(&clients_rwsem); |
621e55ff | 1806 | ib_client_put(client); |
e59178d8 | 1807 | xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); |
921eab11 | 1808 | up_write(&clients_rwsem); |
621e55ff JG |
1809 | |
1810 | /* We do not want to have locks while calling client->remove() */ | |
1811 | rcu_read_lock(); | |
1812 | xa_for_each (&devices, index, device) { | |
1813 | if (!ib_device_try_get(device)) | |
1814 | continue; | |
1815 | rcu_read_unlock(); | |
1816 | ||
921eab11 | 1817 | remove_client_context(device, client->client_id); |
1da177e4 | 1818 | |
621e55ff JG |
1819 | ib_device_put(device); |
1820 | rcu_read_lock(); | |
1821 | } | |
1822 | rcu_read_unlock(); | |
1823 | ||
921eab11 | 1824 | /* |
621e55ff JG |
1825 | * remove_client_context() is not a fence, it can return even though a |
1826 | * removal is ongoing. Wait until all removals are completed. | |
921eab11 | 1827 | */ |
621e55ff | 1828 | wait_for_completion(&client->uses_zero); |
9cd58817 | 1829 | remove_client_id(client); |
1da177e4 LT |
1830 | } |
1831 | EXPORT_SYMBOL(ib_unregister_client); | |
1832 | ||
0e2d00eb JG |
1833 | static int __ib_get_global_client_nl_info(const char *client_name, |
1834 | struct ib_client_nl_info *res) | |
1835 | { | |
1836 | struct ib_client *client; | |
1837 | unsigned long index; | |
1838 | int ret = -ENOENT; | |
1839 | ||
1840 | down_read(&clients_rwsem); | |
1841 | xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { | |
1842 | if (strcmp(client->name, client_name) != 0) | |
1843 | continue; | |
1844 | if (!client->get_global_nl_info) { | |
1845 | ret = -EOPNOTSUPP; | |
1846 | break; | |
1847 | } | |
1848 | ret = client->get_global_nl_info(res); | |
1849 | if (WARN_ON(ret == -ENOENT)) | |
1850 | ret = -EINVAL; | |
1851 | if (!ret && res->cdev) | |
1852 | get_device(res->cdev); | |
1853 | break; | |
1854 | } | |
1855 | up_read(&clients_rwsem); | |
1856 | return ret; | |
1857 | } | |
1858 | ||
1859 | static int __ib_get_client_nl_info(struct ib_device *ibdev, | |
1860 | const char *client_name, | |
1861 | struct ib_client_nl_info *res) | |
1862 | { | |
1863 | unsigned long index; | |
1864 | void *client_data; | |
1865 | int ret = -ENOENT; | |
1866 | ||
1867 | down_read(&ibdev->client_data_rwsem); | |
1868 | xan_for_each_marked (&ibdev->client_data, index, client_data, | |
1869 | CLIENT_DATA_REGISTERED) { | |
1870 | struct ib_client *client = xa_load(&clients, index); | |
1871 | ||
1872 | if (!client || strcmp(client->name, client_name) != 0) | |
1873 | continue; | |
1874 | if (!client->get_nl_info) { | |
1875 | ret = -EOPNOTSUPP; | |
1876 | break; | |
1877 | } | |
1878 | ret = client->get_nl_info(ibdev, client_data, res); | |
1879 | if (WARN_ON(ret == -ENOENT)) | |
1880 | ret = -EINVAL; | |
1881 | ||
1882 | /* | |
1883 | * The cdev is guaranteed valid as long as we are inside the | |
1884 | * client_data_rwsem as remove_one can't be called. Keep it | |
1885 | * valid for the caller. | |
1886 | */ | |
1887 | if (!ret && res->cdev) | |
1888 | get_device(res->cdev); | |
1889 | break; | |
1890 | } | |
1891 | up_read(&ibdev->client_data_rwsem); | |
1892 | ||
1893 | return ret; | |
1894 | } | |
1895 | ||
1896 | /** | |
1897 | * ib_get_client_nl_info - Fetch the nl_info from a client | |
1898 | * @device - IB device | |
1899 | * @client_name - Name of the client | |
1900 | * @res - Result of the query | |
1901 | */ | |
1902 | int ib_get_client_nl_info(struct ib_device *ibdev, const char *client_name, | |
1903 | struct ib_client_nl_info *res) | |
1904 | { | |
1905 | int ret; | |
1906 | ||
1907 | if (ibdev) | |
1908 | ret = __ib_get_client_nl_info(ibdev, client_name, res); | |
1909 | else | |
1910 | ret = __ib_get_global_client_nl_info(client_name, res); | |
1911 | #ifdef CONFIG_MODULES | |
1912 | if (ret == -ENOENT) { | |
1913 | request_module("rdma-client-%s", client_name); | |
1914 | if (ibdev) | |
1915 | ret = __ib_get_client_nl_info(ibdev, client_name, res); | |
1916 | else | |
1917 | ret = __ib_get_global_client_nl_info(client_name, res); | |
1918 | } | |
1919 | #endif | |
1920 | if (ret) { | |
1921 | if (ret == -ENOENT) | |
1922 | return -EOPNOTSUPP; | |
1923 | return ret; | |
1924 | } | |
1925 | ||
1926 | if (WARN_ON(!res->cdev)) | |
1927 | return -EINVAL; | |
1928 | return 0; | |
1929 | } | |
1930 | ||
1da177e4 | 1931 | /** |
9cd330d3 | 1932 | * ib_set_client_data - Set IB client context |
1da177e4 LT |
1933 | * @device:Device to set context for |
1934 | * @client:Client to set context for | |
1935 | * @data:Context to set | |
1936 | * | |
0df91bb6 JG |
1937 | * ib_set_client_data() sets client context data that can be retrieved with |
1938 | * ib_get_client_data(). This can only be called while the client is | |
1939 | * registered to the device, once the ib_client remove() callback returns this | |
1940 | * cannot be called. | |
1da177e4 LT |
1941 | */ |
1942 | void ib_set_client_data(struct ib_device *device, struct ib_client *client, | |
1943 | void *data) | |
1944 | { | |
0df91bb6 | 1945 | void *rc; |
1da177e4 | 1946 | |
0df91bb6 JG |
1947 | if (WARN_ON(IS_ERR(data))) |
1948 | data = NULL; | |
1da177e4 | 1949 | |
0df91bb6 JG |
1950 | rc = xa_store(&device->client_data, client->client_id, data, |
1951 | GFP_KERNEL); | |
1952 | WARN_ON(xa_is_err(rc)); | |
1da177e4 LT |
1953 | } |
1954 | EXPORT_SYMBOL(ib_set_client_data); | |
1955 | ||
1956 | /** | |
1957 | * ib_register_event_handler - Register an IB event handler | |
1958 | * @event_handler:Handler to register | |
1959 | * | |
1960 | * ib_register_event_handler() registers an event handler that will be | |
1961 | * called back when asynchronous IB events occur (as defined in | |
6b57cea9 PP |
1962 | * chapter 11 of the InfiniBand Architecture Specification). This |
1963 | * callback occurs in workqueue context. | |
1da177e4 | 1964 | */ |
dcc9881e | 1965 | void ib_register_event_handler(struct ib_event_handler *event_handler) |
1da177e4 | 1966 | { |
6b57cea9 | 1967 | down_write(&event_handler->device->event_handler_rwsem); |
1da177e4 LT |
1968 | list_add_tail(&event_handler->list, |
1969 | &event_handler->device->event_handler_list); | |
6b57cea9 | 1970 | up_write(&event_handler->device->event_handler_rwsem); |
1da177e4 LT |
1971 | } |
1972 | EXPORT_SYMBOL(ib_register_event_handler); | |
1973 | ||
1974 | /** | |
1975 | * ib_unregister_event_handler - Unregister an event handler | |
1976 | * @event_handler:Handler to unregister | |
1977 | * | |
1978 | * Unregister an event handler registered with | |
1979 | * ib_register_event_handler(). | |
1980 | */ | |
dcc9881e | 1981 | void ib_unregister_event_handler(struct ib_event_handler *event_handler) |
1da177e4 | 1982 | { |
6b57cea9 | 1983 | down_write(&event_handler->device->event_handler_rwsem); |
1da177e4 | 1984 | list_del(&event_handler->list); |
6b57cea9 | 1985 | up_write(&event_handler->device->event_handler_rwsem); |
1da177e4 LT |
1986 | } |
1987 | EXPORT_SYMBOL(ib_unregister_event_handler); | |
1988 | ||
6b57cea9 | 1989 | void ib_dispatch_event_clients(struct ib_event *event) |
1da177e4 | 1990 | { |
1da177e4 LT |
1991 | struct ib_event_handler *handler; |
1992 | ||
6b57cea9 | 1993 | down_read(&event->device->event_handler_rwsem); |
1da177e4 LT |
1994 | |
1995 | list_for_each_entry(handler, &event->device->event_handler_list, list) | |
1996 | handler->handler(handler, event); | |
1997 | ||
6b57cea9 | 1998 | up_read(&event->device->event_handler_rwsem); |
1da177e4 | 1999 | } |
1da177e4 | 2000 | |
4929116b KH |
2001 | static int iw_query_port(struct ib_device *device, |
2002 | u8 port_num, | |
2003 | struct ib_port_attr *port_attr) | |
1da177e4 | 2004 | { |
4929116b KH |
2005 | struct in_device *inetdev; |
2006 | struct net_device *netdev; | |
fad61ad4 | 2007 | |
4929116b KH |
2008 | memset(port_attr, 0, sizeof(*port_attr)); |
2009 | ||
2010 | netdev = ib_device_get_netdev(device, port_num); | |
2011 | if (!netdev) | |
2012 | return -ENODEV; | |
2013 | ||
4929116b KH |
2014 | port_attr->max_mtu = IB_MTU_4096; |
2015 | port_attr->active_mtu = ib_mtu_int_to_enum(netdev->mtu); | |
2016 | ||
2017 | if (!netif_carrier_ok(netdev)) { | |
2018 | port_attr->state = IB_PORT_DOWN; | |
2019 | port_attr->phys_state = IB_PORT_PHYS_STATE_DISABLED; | |
2020 | } else { | |
390d3fdc MK |
2021 | rcu_read_lock(); |
2022 | inetdev = __in_dev_get_rcu(netdev); | |
4929116b KH |
2023 | |
2024 | if (inetdev && inetdev->ifa_list) { | |
2025 | port_attr->state = IB_PORT_ACTIVE; | |
2026 | port_attr->phys_state = IB_PORT_PHYS_STATE_LINK_UP; | |
4929116b KH |
2027 | } else { |
2028 | port_attr->state = IB_PORT_INIT; | |
2029 | port_attr->phys_state = | |
2030 | IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING; | |
2031 | } | |
390d3fdc MK |
2032 | |
2033 | rcu_read_unlock(); | |
4929116b KH |
2034 | } |
2035 | ||
390d3fdc | 2036 | dev_put(netdev); |
1e123d96 | 2037 | return device->ops.query_port(device, port_num, port_attr); |
4929116b KH |
2038 | } |
2039 | ||
2040 | static int __ib_query_port(struct ib_device *device, | |
2041 | u8 port_num, | |
2042 | struct ib_port_attr *port_attr) | |
2043 | { | |
2044 | union ib_gid gid = {}; | |
2045 | int err; | |
116c0074 | 2046 | |
fad61ad4 | 2047 | memset(port_attr, 0, sizeof(*port_attr)); |
4929116b | 2048 | |
3023a1e9 | 2049 | err = device->ops.query_port(device, port_num, port_attr); |
fad61ad4 EC |
2050 | if (err || port_attr->subnet_prefix) |
2051 | return err; | |
2052 | ||
4929116b KH |
2053 | if (rdma_port_get_link_layer(device, port_num) != |
2054 | IB_LINK_LAYER_INFINIBAND) | |
d7012467 EC |
2055 | return 0; |
2056 | ||
3023a1e9 | 2057 | err = device->ops.query_gid(device, port_num, 0, &gid); |
fad61ad4 EC |
2058 | if (err) |
2059 | return err; | |
2060 | ||
2061 | port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); | |
2062 | return 0; | |
1da177e4 | 2063 | } |
4929116b KH |
2064 | |
2065 | /** | |
2066 | * ib_query_port - Query IB port attributes | |
2067 | * @device:Device to query | |
2068 | * @port_num:Port number to query | |
2069 | * @port_attr:Port attributes | |
2070 | * | |
2071 | * ib_query_port() returns the attributes of a port through the | |
2072 | * @port_attr pointer. | |
2073 | */ | |
2074 | int ib_query_port(struct ib_device *device, | |
2075 | u8 port_num, | |
2076 | struct ib_port_attr *port_attr) | |
2077 | { | |
2078 | if (!rdma_is_port_valid(device, port_num)) | |
2079 | return -EINVAL; | |
2080 | ||
2081 | if (rdma_protocol_iwarp(device, port_num)) | |
2082 | return iw_query_port(device, port_num, port_attr); | |
2083 | else | |
2084 | return __ib_query_port(device, port_num, port_attr); | |
2085 | } | |
1da177e4 LT |
2086 | EXPORT_SYMBOL(ib_query_port); |
2087 | ||
324e227e JG |
2088 | static void add_ndev_hash(struct ib_port_data *pdata) |
2089 | { | |
2090 | unsigned long flags; | |
2091 | ||
2092 | might_sleep(); | |
2093 | ||
2094 | spin_lock_irqsave(&ndev_hash_lock, flags); | |
2095 | if (hash_hashed(&pdata->ndev_hash_link)) { | |
2096 | hash_del_rcu(&pdata->ndev_hash_link); | |
2097 | spin_unlock_irqrestore(&ndev_hash_lock, flags); | |
2098 | /* | |
2099 | * We cannot do hash_add_rcu after a hash_del_rcu until the | |
2100 | * grace period | |
2101 | */ | |
2102 | synchronize_rcu(); | |
2103 | spin_lock_irqsave(&ndev_hash_lock, flags); | |
2104 | } | |
2105 | if (pdata->netdev) | |
2106 | hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, | |
2107 | (uintptr_t)pdata->netdev); | |
2108 | spin_unlock_irqrestore(&ndev_hash_lock, flags); | |
2109 | } | |
2110 | ||
c2261dd7 JG |
2111 | /** |
2112 | * ib_device_set_netdev - Associate the ib_dev with an underlying net_device | |
2113 | * @ib_dev: Device to modify | |
2114 | * @ndev: net_device to affiliate, may be NULL | |
2115 | * @port: IB port the net_device is connected to | |
2116 | * | |
2117 | * Drivers should use this to link the ib_device to a netdev so the netdev | |
2118 | * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be | |
2119 | * affiliated with any port. | |
2120 | * | |
2121 | * The caller must ensure that the given ndev is not unregistered or | |
2122 | * unregistering, and that either the ib_device is unregistered or | |
2123 | * ib_device_set_netdev() is called with NULL when the ndev sends a | |
2124 | * NETDEV_UNREGISTER event. | |
2125 | */ | |
2126 | int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, | |
2127 | unsigned int port) | |
2128 | { | |
2129 | struct net_device *old_ndev; | |
2130 | struct ib_port_data *pdata; | |
2131 | unsigned long flags; | |
2132 | int ret; | |
2133 | ||
2134 | /* | |
2135 | * Drivers wish to call this before ib_register_driver, so we have to | |
2136 | * setup the port data early. | |
2137 | */ | |
2138 | ret = alloc_port_data(ib_dev); | |
2139 | if (ret) | |
2140 | return ret; | |
2141 | ||
2142 | if (!rdma_is_port_valid(ib_dev, port)) | |
2143 | return -EINVAL; | |
2144 | ||
2145 | pdata = &ib_dev->port_data[port]; | |
2146 | spin_lock_irqsave(&pdata->netdev_lock, flags); | |
324e227e JG |
2147 | old_ndev = rcu_dereference_protected( |
2148 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
2149 | if (old_ndev == ndev) { | |
c2261dd7 JG |
2150 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); |
2151 | return 0; | |
2152 | } | |
c2261dd7 JG |
2153 | |
2154 | if (ndev) | |
2155 | dev_hold(ndev); | |
324e227e | 2156 | rcu_assign_pointer(pdata->netdev, ndev); |
c2261dd7 JG |
2157 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); |
2158 | ||
324e227e | 2159 | add_ndev_hash(pdata); |
c2261dd7 JG |
2160 | if (old_ndev) |
2161 | dev_put(old_ndev); | |
2162 | ||
2163 | return 0; | |
2164 | } | |
2165 | EXPORT_SYMBOL(ib_device_set_netdev); | |
2166 | ||
2167 | static void free_netdevs(struct ib_device *ib_dev) | |
2168 | { | |
2169 | unsigned long flags; | |
2170 | unsigned int port; | |
2171 | ||
46bdf370 KH |
2172 | if (!ib_dev->port_data) |
2173 | return; | |
2174 | ||
c2261dd7 JG |
2175 | rdma_for_each_port (ib_dev, port) { |
2176 | struct ib_port_data *pdata = &ib_dev->port_data[port]; | |
324e227e | 2177 | struct net_device *ndev; |
c2261dd7 JG |
2178 | |
2179 | spin_lock_irqsave(&pdata->netdev_lock, flags); | |
324e227e JG |
2180 | ndev = rcu_dereference_protected( |
2181 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
2182 | if (ndev) { | |
2183 | spin_lock(&ndev_hash_lock); | |
2184 | hash_del_rcu(&pdata->ndev_hash_link); | |
2185 | spin_unlock(&ndev_hash_lock); | |
2186 | ||
2187 | /* | |
2188 | * If this is the last dev_put there is still a | |
2189 | * synchronize_rcu before the netdev is kfreed, so we | |
2190 | * can continue to rely on unlocked pointer | |
2191 | * comparisons after the put | |
2192 | */ | |
2193 | rcu_assign_pointer(pdata->netdev, NULL); | |
2194 | dev_put(ndev); | |
c2261dd7 JG |
2195 | } |
2196 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); | |
2197 | } | |
2198 | } | |
2199 | ||
2200 | struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, | |
2201 | unsigned int port) | |
2202 | { | |
2203 | struct ib_port_data *pdata; | |
2204 | struct net_device *res; | |
2205 | ||
2206 | if (!rdma_is_port_valid(ib_dev, port)) | |
2207 | return NULL; | |
2208 | ||
2209 | pdata = &ib_dev->port_data[port]; | |
2210 | ||
2211 | /* | |
2212 | * New drivers should use ib_device_set_netdev() not the legacy | |
2213 | * get_netdev(). | |
2214 | */ | |
2215 | if (ib_dev->ops.get_netdev) | |
2216 | res = ib_dev->ops.get_netdev(ib_dev, port); | |
2217 | else { | |
2218 | spin_lock(&pdata->netdev_lock); | |
324e227e JG |
2219 | res = rcu_dereference_protected( |
2220 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
c2261dd7 JG |
2221 | if (res) |
2222 | dev_hold(res); | |
2223 | spin_unlock(&pdata->netdev_lock); | |
2224 | } | |
2225 | ||
2226 | /* | |
2227 | * If we are starting to unregister expedite things by preventing | |
2228 | * propagation of an unregistering netdev. | |
2229 | */ | |
2230 | if (res && res->reg_state != NETREG_REGISTERED) { | |
2231 | dev_put(res); | |
2232 | return NULL; | |
2233 | } | |
2234 | ||
2235 | return res; | |
2236 | } | |
2237 | ||
324e227e JG |
2238 | /** |
2239 | * ib_device_get_by_netdev - Find an IB device associated with a netdev | |
2240 | * @ndev: netdev to locate | |
2241 | * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) | |
2242 | * | |
2243 | * Find and hold an ib_device that is associated with a netdev via | |
2244 | * ib_device_set_netdev(). The caller must call ib_device_put() on the | |
2245 | * returned pointer. | |
2246 | */ | |
2247 | struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, | |
2248 | enum rdma_driver_id driver_id) | |
2249 | { | |
2250 | struct ib_device *res = NULL; | |
2251 | struct ib_port_data *cur; | |
2252 | ||
2253 | rcu_read_lock(); | |
2254 | hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, | |
2255 | (uintptr_t)ndev) { | |
2256 | if (rcu_access_pointer(cur->netdev) == ndev && | |
2257 | (driver_id == RDMA_DRIVER_UNKNOWN || | |
b9560a41 | 2258 | cur->ib_dev->ops.driver_id == driver_id) && |
324e227e JG |
2259 | ib_device_try_get(cur->ib_dev)) { |
2260 | res = cur->ib_dev; | |
2261 | break; | |
2262 | } | |
2263 | } | |
2264 | rcu_read_unlock(); | |
2265 | ||
2266 | return res; | |
2267 | } | |
2268 | EXPORT_SYMBOL(ib_device_get_by_netdev); | |
2269 | ||
03db3a2d MB |
2270 | /** |
2271 | * ib_enum_roce_netdev - enumerate all RoCE ports | |
2272 | * @ib_dev : IB device we want to query | |
2273 | * @filter: Should we call the callback? | |
2274 | * @filter_cookie: Cookie passed to filter | |
2275 | * @cb: Callback to call for each found RoCE ports | |
2276 | * @cookie: Cookie passed back to the callback | |
2277 | * | |
2278 | * Enumerates all of the physical RoCE ports of ib_dev | |
2279 | * which are related to netdevice and calls callback() on each | |
2280 | * device for which filter() function returns non zero. | |
2281 | */ | |
2282 | void ib_enum_roce_netdev(struct ib_device *ib_dev, | |
2283 | roce_netdev_filter filter, | |
2284 | void *filter_cookie, | |
2285 | roce_netdev_callback cb, | |
2286 | void *cookie) | |
2287 | { | |
ea1075ed | 2288 | unsigned int port; |
03db3a2d | 2289 | |
ea1075ed | 2290 | rdma_for_each_port (ib_dev, port) |
03db3a2d | 2291 | if (rdma_protocol_roce(ib_dev, port)) { |
c2261dd7 JG |
2292 | struct net_device *idev = |
2293 | ib_device_get_netdev(ib_dev, port); | |
03db3a2d MB |
2294 | |
2295 | if (filter(ib_dev, port, idev, filter_cookie)) | |
2296 | cb(ib_dev, port, idev, cookie); | |
2297 | ||
2298 | if (idev) | |
2299 | dev_put(idev); | |
2300 | } | |
2301 | } | |
2302 | ||
2303 | /** | |
2304 | * ib_enum_all_roce_netdevs - enumerate all RoCE devices | |
2305 | * @filter: Should we call the callback? | |
2306 | * @filter_cookie: Cookie passed to filter | |
2307 | * @cb: Callback to call for each found RoCE ports | |
2308 | * @cookie: Cookie passed back to the callback | |
2309 | * | |
2310 | * Enumerates all RoCE devices' physical ports which are related | |
2311 | * to netdevices and calls callback() on each device for which | |
2312 | * filter() function returns non zero. | |
2313 | */ | |
2314 | void ib_enum_all_roce_netdevs(roce_netdev_filter filter, | |
2315 | void *filter_cookie, | |
2316 | roce_netdev_callback cb, | |
2317 | void *cookie) | |
2318 | { | |
2319 | struct ib_device *dev; | |
0df91bb6 | 2320 | unsigned long index; |
03db3a2d | 2321 | |
921eab11 | 2322 | down_read(&devices_rwsem); |
0df91bb6 | 2323 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) |
03db3a2d | 2324 | ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); |
921eab11 | 2325 | up_read(&devices_rwsem); |
8030c835 LR |
2326 | } |
2327 | ||
2328 | /** | |
2329 | * ib_enum_all_devs - enumerate all ib_devices | |
2330 | * @cb: Callback to call for each found ib_device | |
2331 | * | |
2332 | * Enumerates all ib_devices and calls callback() on each device. | |
2333 | */ | |
2334 | int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, | |
2335 | struct netlink_callback *cb) | |
2336 | { | |
0df91bb6 | 2337 | unsigned long index; |
8030c835 LR |
2338 | struct ib_device *dev; |
2339 | unsigned int idx = 0; | |
2340 | int ret = 0; | |
2341 | ||
921eab11 | 2342 | down_read(&devices_rwsem); |
0df91bb6 | 2343 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { |
37eeab55 PP |
2344 | if (!rdma_dev_access_netns(dev, sock_net(skb->sk))) |
2345 | continue; | |
2346 | ||
8030c835 LR |
2347 | ret = nldev_cb(dev, skb, cb, idx); |
2348 | if (ret) | |
2349 | break; | |
2350 | idx++; | |
2351 | } | |
921eab11 | 2352 | up_read(&devices_rwsem); |
8030c835 | 2353 | return ret; |
03db3a2d MB |
2354 | } |
2355 | ||
1da177e4 LT |
2356 | /** |
2357 | * ib_query_pkey - Get P_Key table entry | |
2358 | * @device:Device to query | |
2359 | * @port_num:Port number to query | |
2360 | * @index:P_Key table index to query | |
2361 | * @pkey:Returned P_Key | |
2362 | * | |
2363 | * ib_query_pkey() fetches the specified P_Key table entry. | |
2364 | */ | |
2365 | int ib_query_pkey(struct ib_device *device, | |
2366 | u8 port_num, u16 index, u16 *pkey) | |
2367 | { | |
9af3f5cf YS |
2368 | if (!rdma_is_port_valid(device, port_num)) |
2369 | return -EINVAL; | |
2370 | ||
ab75a6cb KH |
2371 | if (!device->ops.query_pkey) |
2372 | return -EOPNOTSUPP; | |
2373 | ||
3023a1e9 | 2374 | return device->ops.query_pkey(device, port_num, index, pkey); |
1da177e4 LT |
2375 | } |
2376 | EXPORT_SYMBOL(ib_query_pkey); | |
2377 | ||
2378 | /** | |
2379 | * ib_modify_device - Change IB device attributes | |
2380 | * @device:Device to modify | |
2381 | * @device_modify_mask:Mask of attributes to change | |
2382 | * @device_modify:New attribute values | |
2383 | * | |
2384 | * ib_modify_device() changes a device's attributes as specified by | |
2385 | * the @device_modify_mask and @device_modify structure. | |
2386 | */ | |
2387 | int ib_modify_device(struct ib_device *device, | |
2388 | int device_modify_mask, | |
2389 | struct ib_device_modify *device_modify) | |
2390 | { | |
3023a1e9 | 2391 | if (!device->ops.modify_device) |
d0f3ef36 | 2392 | return -EOPNOTSUPP; |
10e1b54b | 2393 | |
3023a1e9 KH |
2394 | return device->ops.modify_device(device, device_modify_mask, |
2395 | device_modify); | |
1da177e4 LT |
2396 | } |
2397 | EXPORT_SYMBOL(ib_modify_device); | |
2398 | ||
2399 | /** | |
2400 | * ib_modify_port - Modifies the attributes for the specified port. | |
2401 | * @device: The device to modify. | |
2402 | * @port_num: The number of the port to modify. | |
2403 | * @port_modify_mask: Mask used to specify which attributes of the port | |
2404 | * to change. | |
2405 | * @port_modify: New attribute values for the port. | |
2406 | * | |
2407 | * ib_modify_port() changes a port's attributes as specified by the | |
2408 | * @port_modify_mask and @port_modify structure. | |
2409 | */ | |
2410 | int ib_modify_port(struct ib_device *device, | |
2411 | u8 port_num, int port_modify_mask, | |
2412 | struct ib_port_modify *port_modify) | |
2413 | { | |
61e0962d | 2414 | int rc; |
10e1b54b | 2415 | |
24dc831b | 2416 | if (!rdma_is_port_valid(device, port_num)) |
116c0074 RD |
2417 | return -EINVAL; |
2418 | ||
3023a1e9 KH |
2419 | if (device->ops.modify_port) |
2420 | rc = device->ops.modify_port(device, port_num, | |
2421 | port_modify_mask, | |
2422 | port_modify); | |
55bfe905 KH |
2423 | else if (rdma_protocol_roce(device, port_num) && |
2424 | ((port_modify->set_port_cap_mask & ~IB_PORT_CM_SUP) == 0 || | |
2425 | (port_modify->clr_port_cap_mask & ~IB_PORT_CM_SUP) == 0)) | |
2426 | rc = 0; | |
61e0962d | 2427 | else |
55bfe905 | 2428 | rc = -EOPNOTSUPP; |
61e0962d | 2429 | return rc; |
1da177e4 LT |
2430 | } |
2431 | EXPORT_SYMBOL(ib_modify_port); | |
2432 | ||
5eb620c8 YE |
2433 | /** |
2434 | * ib_find_gid - Returns the port number and GID table index where | |
dbb12562 | 2435 | * a specified GID value occurs. Its searches only for IB link layer. |
5eb620c8 YE |
2436 | * @device: The device to query. |
2437 | * @gid: The GID value to search for. | |
2438 | * @port_num: The port number of the device where the GID value was found. | |
2439 | * @index: The index into the GID table where the GID was found. This | |
2440 | * parameter may be NULL. | |
2441 | */ | |
2442 | int ib_find_gid(struct ib_device *device, union ib_gid *gid, | |
b26c4a11 | 2443 | u8 *port_num, u16 *index) |
5eb620c8 YE |
2444 | { |
2445 | union ib_gid tmp_gid; | |
ea1075ed JG |
2446 | unsigned int port; |
2447 | int ret, i; | |
5eb620c8 | 2448 | |
ea1075ed | 2449 | rdma_for_each_port (device, port) { |
22d24f75 | 2450 | if (!rdma_protocol_ib(device, port)) |
b39ffa1d MB |
2451 | continue; |
2452 | ||
8ceb1357 JG |
2453 | for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; |
2454 | ++i) { | |
1dfce294 | 2455 | ret = rdma_query_gid(device, port, i, &tmp_gid); |
5eb620c8 YE |
2456 | if (ret) |
2457 | return ret; | |
2458 | if (!memcmp(&tmp_gid, gid, sizeof *gid)) { | |
2459 | *port_num = port; | |
2460 | if (index) | |
2461 | *index = i; | |
2462 | return 0; | |
2463 | } | |
2464 | } | |
2465 | } | |
2466 | ||
2467 | return -ENOENT; | |
2468 | } | |
2469 | EXPORT_SYMBOL(ib_find_gid); | |
2470 | ||
2471 | /** | |
2472 | * ib_find_pkey - Returns the PKey table index where a specified | |
2473 | * PKey value occurs. | |
2474 | * @device: The device to query. | |
2475 | * @port_num: The port number of the device to search for the PKey. | |
2476 | * @pkey: The PKey value to search for. | |
2477 | * @index: The index into the PKey table where the PKey was found. | |
2478 | */ | |
2479 | int ib_find_pkey(struct ib_device *device, | |
2480 | u8 port_num, u16 pkey, u16 *index) | |
2481 | { | |
2482 | int ret, i; | |
2483 | u16 tmp_pkey; | |
ff7166c4 | 2484 | int partial_ix = -1; |
5eb620c8 | 2485 | |
8ceb1357 JG |
2486 | for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; |
2487 | ++i) { | |
5eb620c8 YE |
2488 | ret = ib_query_pkey(device, port_num, i, &tmp_pkey); |
2489 | if (ret) | |
2490 | return ret; | |
36026ecc | 2491 | if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { |
ff7166c4 JM |
2492 | /* if there is full-member pkey take it.*/ |
2493 | if (tmp_pkey & 0x8000) { | |
2494 | *index = i; | |
2495 | return 0; | |
2496 | } | |
2497 | if (partial_ix < 0) | |
2498 | partial_ix = i; | |
5eb620c8 YE |
2499 | } |
2500 | } | |
2501 | ||
ff7166c4 JM |
2502 | /*no full-member, if exists take the limited*/ |
2503 | if (partial_ix >= 0) { | |
2504 | *index = partial_ix; | |
2505 | return 0; | |
2506 | } | |
5eb620c8 YE |
2507 | return -ENOENT; |
2508 | } | |
2509 | EXPORT_SYMBOL(ib_find_pkey); | |
2510 | ||
9268f72d YK |
2511 | /** |
2512 | * ib_get_net_dev_by_params() - Return the appropriate net_dev | |
2513 | * for a received CM request | |
2514 | * @dev: An RDMA device on which the request has been received. | |
2515 | * @port: Port number on the RDMA device. | |
2516 | * @pkey: The Pkey the request came on. | |
2517 | * @gid: A GID that the net_dev uses to communicate. | |
2518 | * @addr: Contains the IP address that the request specified as its | |
2519 | * destination. | |
921eab11 | 2520 | * |
9268f72d YK |
2521 | */ |
2522 | struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, | |
2523 | u8 port, | |
2524 | u16 pkey, | |
2525 | const union ib_gid *gid, | |
2526 | const struct sockaddr *addr) | |
2527 | { | |
2528 | struct net_device *net_dev = NULL; | |
0df91bb6 JG |
2529 | unsigned long index; |
2530 | void *client_data; | |
9268f72d YK |
2531 | |
2532 | if (!rdma_protocol_ib(dev, port)) | |
2533 | return NULL; | |
2534 | ||
921eab11 JG |
2535 | /* |
2536 | * Holding the read side guarantees that the client will not become | |
2537 | * unregistered while we are calling get_net_dev_by_params() | |
2538 | */ | |
2539 | down_read(&dev->client_data_rwsem); | |
0df91bb6 JG |
2540 | xan_for_each_marked (&dev->client_data, index, client_data, |
2541 | CLIENT_DATA_REGISTERED) { | |
2542 | struct ib_client *client = xa_load(&clients, index); | |
9268f72d | 2543 | |
0df91bb6 | 2544 | if (!client || !client->get_net_dev_by_params) |
9268f72d YK |
2545 | continue; |
2546 | ||
0df91bb6 JG |
2547 | net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, |
2548 | addr, client_data); | |
2549 | if (net_dev) | |
2550 | break; | |
9268f72d | 2551 | } |
921eab11 | 2552 | up_read(&dev->client_data_rwsem); |
9268f72d YK |
2553 | |
2554 | return net_dev; | |
2555 | } | |
2556 | EXPORT_SYMBOL(ib_get_net_dev_by_params); | |
2557 | ||
521ed0d9 KH |
2558 | void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) |
2559 | { | |
3023a1e9 | 2560 | struct ib_device_ops *dev_ops = &dev->ops; |
521ed0d9 KH |
2561 | #define SET_DEVICE_OP(ptr, name) \ |
2562 | do { \ | |
2563 | if (ops->name) \ | |
2564 | if (!((ptr)->name)) \ | |
2565 | (ptr)->name = ops->name; \ | |
2566 | } while (0) | |
2567 | ||
30471d4b LR |
2568 | #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) |
2569 | ||
b9560a41 JG |
2570 | if (ops->driver_id != RDMA_DRIVER_UNKNOWN) { |
2571 | WARN_ON(dev_ops->driver_id != RDMA_DRIVER_UNKNOWN && | |
2572 | dev_ops->driver_id != ops->driver_id); | |
2573 | dev_ops->driver_id = ops->driver_id; | |
2574 | } | |
7a154142 JG |
2575 | if (ops->owner) { |
2576 | WARN_ON(dev_ops->owner && dev_ops->owner != ops->owner); | |
2577 | dev_ops->owner = ops->owner; | |
2578 | } | |
72c6ec18 JG |
2579 | if (ops->uverbs_abi_ver) |
2580 | dev_ops->uverbs_abi_ver = ops->uverbs_abi_ver; | |
b9560a41 | 2581 | |
8f71bb00 JG |
2582 | dev_ops->uverbs_no_driver_id_binding |= |
2583 | ops->uverbs_no_driver_id_binding; | |
2584 | ||
3023a1e9 | 2585 | SET_DEVICE_OP(dev_ops, add_gid); |
2f1927b0 | 2586 | SET_DEVICE_OP(dev_ops, advise_mr); |
3023a1e9 | 2587 | SET_DEVICE_OP(dev_ops, alloc_dm); |
3023a1e9 KH |
2588 | SET_DEVICE_OP(dev_ops, alloc_hw_stats); |
2589 | SET_DEVICE_OP(dev_ops, alloc_mr); | |
26bc7eae | 2590 | SET_DEVICE_OP(dev_ops, alloc_mr_integrity); |
3023a1e9 KH |
2591 | SET_DEVICE_OP(dev_ops, alloc_mw); |
2592 | SET_DEVICE_OP(dev_ops, alloc_pd); | |
2593 | SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); | |
2594 | SET_DEVICE_OP(dev_ops, alloc_ucontext); | |
2595 | SET_DEVICE_OP(dev_ops, alloc_xrcd); | |
2596 | SET_DEVICE_OP(dev_ops, attach_mcast); | |
2597 | SET_DEVICE_OP(dev_ops, check_mr_status); | |
c4ffee7c | 2598 | SET_DEVICE_OP(dev_ops, counter_alloc_stats); |
99fa331d MZ |
2599 | SET_DEVICE_OP(dev_ops, counter_bind_qp); |
2600 | SET_DEVICE_OP(dev_ops, counter_dealloc); | |
2601 | SET_DEVICE_OP(dev_ops, counter_unbind_qp); | |
c4ffee7c | 2602 | SET_DEVICE_OP(dev_ops, counter_update_stats); |
3023a1e9 KH |
2603 | SET_DEVICE_OP(dev_ops, create_ah); |
2604 | SET_DEVICE_OP(dev_ops, create_counters); | |
2605 | SET_DEVICE_OP(dev_ops, create_cq); | |
2606 | SET_DEVICE_OP(dev_ops, create_flow); | |
2607 | SET_DEVICE_OP(dev_ops, create_flow_action_esp); | |
2608 | SET_DEVICE_OP(dev_ops, create_qp); | |
2609 | SET_DEVICE_OP(dev_ops, create_rwq_ind_table); | |
2610 | SET_DEVICE_OP(dev_ops, create_srq); | |
676a80ad | 2611 | SET_DEVICE_OP(dev_ops, create_user_ah); |
3023a1e9 KH |
2612 | SET_DEVICE_OP(dev_ops, create_wq); |
2613 | SET_DEVICE_OP(dev_ops, dealloc_dm); | |
d0899892 | 2614 | SET_DEVICE_OP(dev_ops, dealloc_driver); |
3023a1e9 KH |
2615 | SET_DEVICE_OP(dev_ops, dealloc_mw); |
2616 | SET_DEVICE_OP(dev_ops, dealloc_pd); | |
2617 | SET_DEVICE_OP(dev_ops, dealloc_ucontext); | |
2618 | SET_DEVICE_OP(dev_ops, dealloc_xrcd); | |
2619 | SET_DEVICE_OP(dev_ops, del_gid); | |
2620 | SET_DEVICE_OP(dev_ops, dereg_mr); | |
2621 | SET_DEVICE_OP(dev_ops, destroy_ah); | |
2622 | SET_DEVICE_OP(dev_ops, destroy_counters); | |
2623 | SET_DEVICE_OP(dev_ops, destroy_cq); | |
2624 | SET_DEVICE_OP(dev_ops, destroy_flow); | |
2625 | SET_DEVICE_OP(dev_ops, destroy_flow_action); | |
2626 | SET_DEVICE_OP(dev_ops, destroy_qp); | |
2627 | SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); | |
2628 | SET_DEVICE_OP(dev_ops, destroy_srq); | |
2629 | SET_DEVICE_OP(dev_ops, destroy_wq); | |
2630 | SET_DEVICE_OP(dev_ops, detach_mcast); | |
2631 | SET_DEVICE_OP(dev_ops, disassociate_ucontext); | |
2632 | SET_DEVICE_OP(dev_ops, drain_rq); | |
2633 | SET_DEVICE_OP(dev_ops, drain_sq); | |
ca22354b | 2634 | SET_DEVICE_OP(dev_ops, enable_driver); |
211cd945 | 2635 | SET_DEVICE_OP(dev_ops, fill_res_cm_id_entry); |
9e2a187a | 2636 | SET_DEVICE_OP(dev_ops, fill_res_cq_entry); |
65959522 | 2637 | SET_DEVICE_OP(dev_ops, fill_res_cq_entry_raw); |
f4434529 | 2638 | SET_DEVICE_OP(dev_ops, fill_res_mr_entry); |
65959522 | 2639 | SET_DEVICE_OP(dev_ops, fill_res_mr_entry_raw); |
5cc34116 | 2640 | SET_DEVICE_OP(dev_ops, fill_res_qp_entry); |
65959522 | 2641 | SET_DEVICE_OP(dev_ops, fill_res_qp_entry_raw); |
f4434529 | 2642 | SET_DEVICE_OP(dev_ops, fill_stat_mr_entry); |
3023a1e9 KH |
2643 | SET_DEVICE_OP(dev_ops, get_dev_fw_str); |
2644 | SET_DEVICE_OP(dev_ops, get_dma_mr); | |
2645 | SET_DEVICE_OP(dev_ops, get_hw_stats); | |
2646 | SET_DEVICE_OP(dev_ops, get_link_layer); | |
2647 | SET_DEVICE_OP(dev_ops, get_netdev); | |
2648 | SET_DEVICE_OP(dev_ops, get_port_immutable); | |
2649 | SET_DEVICE_OP(dev_ops, get_vector_affinity); | |
2650 | SET_DEVICE_OP(dev_ops, get_vf_config); | |
bfcb3c5d | 2651 | SET_DEVICE_OP(dev_ops, get_vf_guid); |
3023a1e9 | 2652 | SET_DEVICE_OP(dev_ops, get_vf_stats); |
ea4baf7f | 2653 | SET_DEVICE_OP(dev_ops, init_port); |
dd05cb82 KH |
2654 | SET_DEVICE_OP(dev_ops, iw_accept); |
2655 | SET_DEVICE_OP(dev_ops, iw_add_ref); | |
2656 | SET_DEVICE_OP(dev_ops, iw_connect); | |
2657 | SET_DEVICE_OP(dev_ops, iw_create_listen); | |
2658 | SET_DEVICE_OP(dev_ops, iw_destroy_listen); | |
2659 | SET_DEVICE_OP(dev_ops, iw_get_qp); | |
2660 | SET_DEVICE_OP(dev_ops, iw_reject); | |
2661 | SET_DEVICE_OP(dev_ops, iw_rem_ref); | |
3023a1e9 | 2662 | SET_DEVICE_OP(dev_ops, map_mr_sg); |
2cdfcdd8 | 2663 | SET_DEVICE_OP(dev_ops, map_mr_sg_pi); |
3023a1e9 | 2664 | SET_DEVICE_OP(dev_ops, mmap); |
3411f9f0 | 2665 | SET_DEVICE_OP(dev_ops, mmap_free); |
3023a1e9 KH |
2666 | SET_DEVICE_OP(dev_ops, modify_ah); |
2667 | SET_DEVICE_OP(dev_ops, modify_cq); | |
2668 | SET_DEVICE_OP(dev_ops, modify_device); | |
2669 | SET_DEVICE_OP(dev_ops, modify_flow_action_esp); | |
2670 | SET_DEVICE_OP(dev_ops, modify_port); | |
2671 | SET_DEVICE_OP(dev_ops, modify_qp); | |
2672 | SET_DEVICE_OP(dev_ops, modify_srq); | |
2673 | SET_DEVICE_OP(dev_ops, modify_wq); | |
2674 | SET_DEVICE_OP(dev_ops, peek_cq); | |
2675 | SET_DEVICE_OP(dev_ops, poll_cq); | |
2676 | SET_DEVICE_OP(dev_ops, post_recv); | |
2677 | SET_DEVICE_OP(dev_ops, post_send); | |
2678 | SET_DEVICE_OP(dev_ops, post_srq_recv); | |
2679 | SET_DEVICE_OP(dev_ops, process_mad); | |
2680 | SET_DEVICE_OP(dev_ops, query_ah); | |
2681 | SET_DEVICE_OP(dev_ops, query_device); | |
2682 | SET_DEVICE_OP(dev_ops, query_gid); | |
2683 | SET_DEVICE_OP(dev_ops, query_pkey); | |
2684 | SET_DEVICE_OP(dev_ops, query_port); | |
2685 | SET_DEVICE_OP(dev_ops, query_qp); | |
2686 | SET_DEVICE_OP(dev_ops, query_srq); | |
1c8fb1ea | 2687 | SET_DEVICE_OP(dev_ops, query_ucontext); |
3023a1e9 KH |
2688 | SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); |
2689 | SET_DEVICE_OP(dev_ops, read_counters); | |
2690 | SET_DEVICE_OP(dev_ops, reg_dm_mr); | |
2691 | SET_DEVICE_OP(dev_ops, reg_user_mr); | |
2692 | SET_DEVICE_OP(dev_ops, req_ncomp_notif); | |
2693 | SET_DEVICE_OP(dev_ops, req_notify_cq); | |
2694 | SET_DEVICE_OP(dev_ops, rereg_user_mr); | |
2695 | SET_DEVICE_OP(dev_ops, resize_cq); | |
2696 | SET_DEVICE_OP(dev_ops, set_vf_guid); | |
2697 | SET_DEVICE_OP(dev_ops, set_vf_link_state); | |
21a428a0 | 2698 | |
d3456914 | 2699 | SET_OBJ_SIZE(dev_ops, ib_ah); |
3b023e1b | 2700 | SET_OBJ_SIZE(dev_ops, ib_counters); |
e39afe3d | 2701 | SET_OBJ_SIZE(dev_ops, ib_cq); |
d18bb3e1 | 2702 | SET_OBJ_SIZE(dev_ops, ib_mw); |
21a428a0 | 2703 | SET_OBJ_SIZE(dev_ops, ib_pd); |
c0a6b5ec | 2704 | SET_OBJ_SIZE(dev_ops, ib_rwq_ind_table); |
68e326de | 2705 | SET_OBJ_SIZE(dev_ops, ib_srq); |
a2a074ef | 2706 | SET_OBJ_SIZE(dev_ops, ib_ucontext); |
28ad5f65 | 2707 | SET_OBJ_SIZE(dev_ops, ib_xrcd); |
521ed0d9 KH |
2708 | } |
2709 | EXPORT_SYMBOL(ib_set_device_ops); | |
2710 | ||
d0e312fe | 2711 | static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { |
735c631a | 2712 | [RDMA_NL_LS_OP_RESOLVE] = { |
647c75ac | 2713 | .doit = ib_nl_handle_resolve_resp, |
e3a2b93d LR |
2714 | .flags = RDMA_NL_ADMIN_PERM, |
2715 | }, | |
735c631a | 2716 | [RDMA_NL_LS_OP_SET_TIMEOUT] = { |
647c75ac | 2717 | .doit = ib_nl_handle_set_timeout, |
e3a2b93d LR |
2718 | .flags = RDMA_NL_ADMIN_PERM, |
2719 | }, | |
ae43f828 | 2720 | [RDMA_NL_LS_OP_IP_RESOLVE] = { |
647c75ac | 2721 | .doit = ib_nl_handle_ip_res_resp, |
e3a2b93d LR |
2722 | .flags = RDMA_NL_ADMIN_PERM, |
2723 | }, | |
735c631a MB |
2724 | }; |
2725 | ||
1da177e4 LT |
2726 | static int __init ib_core_init(void) |
2727 | { | |
2728 | int ret; | |
2729 | ||
f0626710 TH |
2730 | ib_wq = alloc_workqueue("infiniband", 0, 0); |
2731 | if (!ib_wq) | |
2732 | return -ENOMEM; | |
2733 | ||
14d3a3b2 | 2734 | ib_comp_wq = alloc_workqueue("ib-comp-wq", |
b7363e67 | 2735 | WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); |
14d3a3b2 CH |
2736 | if (!ib_comp_wq) { |
2737 | ret = -ENOMEM; | |
2738 | goto err; | |
2739 | } | |
2740 | ||
f794809a JM |
2741 | ib_comp_unbound_wq = |
2742 | alloc_workqueue("ib-comp-unb-wq", | |
2743 | WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | | |
2744 | WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); | |
2745 | if (!ib_comp_unbound_wq) { | |
2746 | ret = -ENOMEM; | |
2747 | goto err_comp; | |
2748 | } | |
2749 | ||
55aeed06 | 2750 | ret = class_register(&ib_class); |
fd75c789 | 2751 | if (ret) { |
aba25a3e | 2752 | pr_warn("Couldn't create InfiniBand device class\n"); |
f794809a | 2753 | goto err_comp_unbound; |
fd75c789 | 2754 | } |
1da177e4 | 2755 | |
549af008 PP |
2756 | rdma_nl_init(); |
2757 | ||
e3f20f02 LR |
2758 | ret = addr_init(); |
2759 | if (ret) { | |
4469add9 | 2760 | pr_warn("Couldn't init IB address resolution\n"); |
e3f20f02 LR |
2761 | goto err_ibnl; |
2762 | } | |
2763 | ||
4c2cb422 MB |
2764 | ret = ib_mad_init(); |
2765 | if (ret) { | |
2766 | pr_warn("Couldn't init IB MAD\n"); | |
2767 | goto err_addr; | |
2768 | } | |
2769 | ||
c2e49c92 MB |
2770 | ret = ib_sa_init(); |
2771 | if (ret) { | |
2772 | pr_warn("Couldn't init SA\n"); | |
2773 | goto err_mad; | |
2774 | } | |
2775 | ||
42df744c | 2776 | ret = register_blocking_lsm_notifier(&ibdev_lsm_nb); |
8f408ab6 DJ |
2777 | if (ret) { |
2778 | pr_warn("Couldn't register LSM notifier. ret %d\n", ret); | |
c9901724 | 2779 | goto err_sa; |
8f408ab6 DJ |
2780 | } |
2781 | ||
4e0f7b90 PP |
2782 | ret = register_pernet_device(&rdma_dev_net_ops); |
2783 | if (ret) { | |
2784 | pr_warn("Couldn't init compat dev. ret %d\n", ret); | |
2785 | goto err_compat; | |
2786 | } | |
2787 | ||
6c80b41a | 2788 | nldev_init(); |
c9901724 | 2789 | rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); |
5ef8c0c1 | 2790 | roce_gid_mgmt_init(); |
1da177e4 | 2791 | |
fd75c789 NM |
2792 | return 0; |
2793 | ||
4e0f7b90 | 2794 | err_compat: |
42df744c | 2795 | unregister_blocking_lsm_notifier(&ibdev_lsm_nb); |
735c631a MB |
2796 | err_sa: |
2797 | ib_sa_cleanup(); | |
c2e49c92 MB |
2798 | err_mad: |
2799 | ib_mad_cleanup(); | |
4c2cb422 MB |
2800 | err_addr: |
2801 | addr_cleanup(); | |
e3f20f02 | 2802 | err_ibnl: |
55aeed06 | 2803 | class_unregister(&ib_class); |
f794809a JM |
2804 | err_comp_unbound: |
2805 | destroy_workqueue(ib_comp_unbound_wq); | |
14d3a3b2 CH |
2806 | err_comp: |
2807 | destroy_workqueue(ib_comp_wq); | |
fd75c789 NM |
2808 | err: |
2809 | destroy_workqueue(ib_wq); | |
1da177e4 LT |
2810 | return ret; |
2811 | } | |
2812 | ||
2813 | static void __exit ib_core_cleanup(void) | |
2814 | { | |
5ef8c0c1 | 2815 | roce_gid_mgmt_cleanup(); |
6c80b41a | 2816 | nldev_exit(); |
c9901724 | 2817 | rdma_nl_unregister(RDMA_NL_LS); |
4e0f7b90 | 2818 | unregister_pernet_device(&rdma_dev_net_ops); |
42df744c | 2819 | unregister_blocking_lsm_notifier(&ibdev_lsm_nb); |
c2e49c92 | 2820 | ib_sa_cleanup(); |
4c2cb422 | 2821 | ib_mad_cleanup(); |
e3f20f02 | 2822 | addr_cleanup(); |
c9901724 | 2823 | rdma_nl_exit(); |
55aeed06 | 2824 | class_unregister(&ib_class); |
f794809a | 2825 | destroy_workqueue(ib_comp_unbound_wq); |
14d3a3b2 | 2826 | destroy_workqueue(ib_comp_wq); |
f7c6a7b5 | 2827 | /* Make sure that any pending umem accounting work is done. */ |
f0626710 | 2828 | destroy_workqueue(ib_wq); |
d0899892 | 2829 | flush_workqueue(system_unbound_wq); |
e59178d8 | 2830 | WARN_ON(!xa_empty(&clients)); |
0df91bb6 | 2831 | WARN_ON(!xa_empty(&devices)); |
1da177e4 LT |
2832 | } |
2833 | ||
e3bf14bd JG |
2834 | MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); |
2835 | ||
62dfa795 PP |
2836 | /* ib core relies on netdev stack to first register net_ns_type_operations |
2837 | * ns kobject type before ib_core initialization. | |
2838 | */ | |
2839 | fs_initcall(ib_core_init); | |
1da177e4 | 2840 | module_exit(ib_core_cleanup); |