Commit | Line | Data |
---|---|---|
bc38a6ab RD |
1 | /* |
2 | * Copyright (c) 2005 Topspin Communications. All rights reserved. | |
33b9b3ee | 3 | * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. |
2a1d9b7f RD |
4 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. |
5 | * Copyright (c) 2005 Voltaire, Inc. All rights reserved. | |
67cdb40c | 6 | * Copyright (c) 2005 PathScale, Inc. All rights reserved. |
bc38a6ab RD |
7 | * |
8 | * This software is available to you under a choice of one of two | |
9 | * licenses. You may choose to be licensed under the terms of the GNU | |
10 | * General Public License (GPL) Version 2, available from the file | |
11 | * COPYING in the main directory of this source tree, or the | |
12 | * OpenIB.org BSD license below: | |
13 | * | |
14 | * Redistribution and use in source and binary forms, with or | |
15 | * without modification, are permitted provided that the following | |
16 | * conditions are met: | |
17 | * | |
18 | * - Redistributions of source code must retain the above | |
19 | * copyright notice, this list of conditions and the following | |
20 | * disclaimer. | |
21 | * | |
22 | * - Redistributions in binary form must reproduce the above | |
23 | * copyright notice, this list of conditions and the following | |
24 | * disclaimer in the documentation and/or other materials | |
25 | * provided with the distribution. | |
26 | * | |
27 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
28 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
29 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
30 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
31 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
32 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
33 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
34 | * SOFTWARE. | |
bc38a6ab RD |
35 | */ |
36 | ||
37 | #include <linux/module.h> | |
38 | #include <linux/init.h> | |
39 | #include <linux/device.h> | |
40 | #include <linux/err.h> | |
41 | #include <linux/fs.h> | |
42 | #include <linux/poll.h> | |
a99bbaf5 | 43 | #include <linux/sched.h> |
bc38a6ab | 44 | #include <linux/file.h> |
70a30e16 | 45 | #include <linux/cdev.h> |
a265e558 | 46 | #include <linux/anon_inodes.h> |
5a0e3ad6 | 47 | #include <linux/slab.h> |
bc38a6ab RD |
48 | |
49 | #include <asm/uaccess.h> | |
50 | ||
51 | #include "uverbs.h" | |
52 | ||
53 | MODULE_AUTHOR("Roland Dreier"); | |
54 | MODULE_DESCRIPTION("InfiniBand userspace verbs access"); | |
55 | MODULE_LICENSE("Dual BSD/GPL"); | |
56 | ||
bc38a6ab RD |
57 | enum { |
58 | IB_UVERBS_MAJOR = 231, | |
59 | IB_UVERBS_BASE_MINOR = 192, | |
60 | IB_UVERBS_MAX_DEVICES = 32 | |
61 | }; | |
62 | ||
63 | #define IB_UVERBS_BASE_DEV MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR) | |
64 | ||
70a30e16 RD |
65 | static struct class *uverbs_class; |
66 | ||
9ead190b | 67 | DEFINE_SPINLOCK(ib_uverbs_idr_lock); |
bc38a6ab RD |
68 | DEFINE_IDR(ib_uverbs_pd_idr); |
69 | DEFINE_IDR(ib_uverbs_mr_idr); | |
70 | DEFINE_IDR(ib_uverbs_mw_idr); | |
71 | DEFINE_IDR(ib_uverbs_ah_idr); | |
72 | DEFINE_IDR(ib_uverbs_cq_idr); | |
73 | DEFINE_IDR(ib_uverbs_qp_idr); | |
f520ba5a | 74 | DEFINE_IDR(ib_uverbs_srq_idr); |
bc38a6ab | 75 | |
6276e08a | 76 | static DEFINE_SPINLOCK(map_lock); |
bc38a6ab RD |
77 | static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES); |
78 | ||
79 | static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file, | |
80 | const char __user *buf, int in_len, | |
81 | int out_len) = { | |
9afed76d AC |
82 | [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context, |
83 | [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device, | |
84 | [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port, | |
85 | [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd, | |
86 | [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd, | |
87 | [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr, | |
88 | [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr, | |
6b73597e | 89 | [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel, |
9afed76d AC |
90 | [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq, |
91 | [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq, | |
92 | [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq, | |
93 | [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq, | |
94 | [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq, | |
95 | [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp, | |
96 | [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp, | |
97 | [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp, | |
98 | [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp, | |
99 | [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send, | |
100 | [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv, | |
101 | [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv, | |
102 | [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah, | |
103 | [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah, | |
104 | [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast, | |
105 | [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast, | |
106 | [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq, | |
107 | [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq, | |
108 | [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq, | |
109 | [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq, | |
bc38a6ab RD |
110 | }; |
111 | ||
bc38a6ab RD |
112 | static void ib_uverbs_add_one(struct ib_device *device); |
113 | static void ib_uverbs_remove_one(struct ib_device *device); | |
114 | ||
70a30e16 RD |
115 | static void ib_uverbs_release_dev(struct kref *ref) |
116 | { | |
117 | struct ib_uverbs_device *dev = | |
118 | container_of(ref, struct ib_uverbs_device, ref); | |
119 | ||
fd60ae40 | 120 | complete(&dev->comp); |
70a30e16 RD |
121 | } |
122 | ||
04d29b0e RD |
123 | static void ib_uverbs_release_event_file(struct kref *ref) |
124 | { | |
125 | struct ib_uverbs_event_file *file = | |
126 | container_of(ref, struct ib_uverbs_event_file, ref); | |
127 | ||
128 | kfree(file); | |
129 | } | |
130 | ||
70a30e16 RD |
131 | void ib_uverbs_release_ucq(struct ib_uverbs_file *file, |
132 | struct ib_uverbs_event_file *ev_file, | |
133 | struct ib_ucq_object *uobj) | |
134 | { | |
135 | struct ib_uverbs_event *evt, *tmp; | |
136 | ||
137 | if (ev_file) { | |
138 | spin_lock_irq(&ev_file->lock); | |
139 | list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) { | |
140 | list_del(&evt->list); | |
141 | kfree(evt); | |
142 | } | |
143 | spin_unlock_irq(&ev_file->lock); | |
144 | ||
145 | kref_put(&ev_file->ref, ib_uverbs_release_event_file); | |
146 | } | |
147 | ||
148 | spin_lock_irq(&file->async_file->lock); | |
149 | list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) { | |
150 | list_del(&evt->list); | |
151 | kfree(evt); | |
152 | } | |
153 | spin_unlock_irq(&file->async_file->lock); | |
154 | } | |
155 | ||
156 | void ib_uverbs_release_uevent(struct ib_uverbs_file *file, | |
157 | struct ib_uevent_object *uobj) | |
158 | { | |
159 | struct ib_uverbs_event *evt, *tmp; | |
160 | ||
161 | spin_lock_irq(&file->async_file->lock); | |
162 | list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) { | |
163 | list_del(&evt->list); | |
164 | kfree(evt); | |
165 | } | |
166 | spin_unlock_irq(&file->async_file->lock); | |
167 | } | |
168 | ||
f4e40156 JM |
169 | static void ib_uverbs_detach_umcast(struct ib_qp *qp, |
170 | struct ib_uqp_object *uobj) | |
171 | { | |
172 | struct ib_uverbs_mcast_entry *mcast, *tmp; | |
173 | ||
174 | list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) { | |
175 | ib_detach_mcast(qp, &mcast->gid, mcast->lid); | |
176 | list_del(&mcast->list); | |
177 | kfree(mcast); | |
178 | } | |
179 | } | |
180 | ||
70a30e16 RD |
181 | static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file, |
182 | struct ib_ucontext *context) | |
bc38a6ab RD |
183 | { |
184 | struct ib_uobject *uobj, *tmp; | |
185 | ||
186 | if (!context) | |
187 | return 0; | |
188 | ||
f7c6a7b5 RD |
189 | context->closing = 1; |
190 | ||
67cdb40c | 191 | list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) { |
9ead190b RD |
192 | struct ib_ah *ah = uobj->object; |
193 | ||
194 | idr_remove_uobj(&ib_uverbs_ah_idr, uobj); | |
67cdb40c | 195 | ib_destroy_ah(ah); |
67cdb40c RD |
196 | kfree(uobj); |
197 | } | |
bc38a6ab RD |
198 | |
199 | list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) { | |
9ead190b | 200 | struct ib_qp *qp = uobj->object; |
f4e40156 JM |
201 | struct ib_uqp_object *uqp = |
202 | container_of(uobj, struct ib_uqp_object, uevent.uobject); | |
9ead190b RD |
203 | |
204 | idr_remove_uobj(&ib_uverbs_qp_idr, uobj); | |
f4e40156 | 205 | ib_uverbs_detach_umcast(qp, uqp); |
bc38a6ab | 206 | ib_destroy_qp(qp); |
f4e40156 JM |
207 | ib_uverbs_release_uevent(file, &uqp->uevent); |
208 | kfree(uqp); | |
bc38a6ab RD |
209 | } |
210 | ||
211 | list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) { | |
9ead190b | 212 | struct ib_cq *cq = uobj->object; |
70a30e16 RD |
213 | struct ib_uverbs_event_file *ev_file = cq->cq_context; |
214 | struct ib_ucq_object *ucq = | |
215 | container_of(uobj, struct ib_ucq_object, uobject); | |
9ead190b RD |
216 | |
217 | idr_remove_uobj(&ib_uverbs_cq_idr, uobj); | |
bc38a6ab | 218 | ib_destroy_cq(cq); |
70a30e16 RD |
219 | ib_uverbs_release_ucq(file, ev_file, ucq); |
220 | kfree(ucq); | |
bc38a6ab RD |
221 | } |
222 | ||
f520ba5a | 223 | list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) { |
9ead190b | 224 | struct ib_srq *srq = uobj->object; |
70a30e16 RD |
225 | struct ib_uevent_object *uevent = |
226 | container_of(uobj, struct ib_uevent_object, uobject); | |
9ead190b RD |
227 | |
228 | idr_remove_uobj(&ib_uverbs_srq_idr, uobj); | |
f520ba5a | 229 | ib_destroy_srq(srq); |
70a30e16 RD |
230 | ib_uverbs_release_uevent(file, uevent); |
231 | kfree(uevent); | |
f520ba5a RD |
232 | } |
233 | ||
bc38a6ab RD |
234 | /* XXX Free MWs */ |
235 | ||
236 | list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) { | |
9ead190b | 237 | struct ib_mr *mr = uobj->object; |
bc38a6ab | 238 | |
9ead190b | 239 | idr_remove_uobj(&ib_uverbs_mr_idr, uobj); |
bc38a6ab | 240 | ib_dereg_mr(mr); |
f7c6a7b5 | 241 | kfree(uobj); |
bc38a6ab RD |
242 | } |
243 | ||
244 | list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) { | |
9ead190b RD |
245 | struct ib_pd *pd = uobj->object; |
246 | ||
247 | idr_remove_uobj(&ib_uverbs_pd_idr, uobj); | |
bc38a6ab | 248 | ib_dealloc_pd(pd); |
bc38a6ab RD |
249 | kfree(uobj); |
250 | } | |
251 | ||
bc38a6ab RD |
252 | return context->device->dealloc_ucontext(context); |
253 | } | |
254 | ||
255 | static void ib_uverbs_release_file(struct kref *ref) | |
256 | { | |
257 | struct ib_uverbs_file *file = | |
258 | container_of(ref, struct ib_uverbs_file, ref); | |
259 | ||
260 | module_put(file->device->ib_dev->owner); | |
70a30e16 RD |
261 | kref_put(&file->device->ref, ib_uverbs_release_dev); |
262 | ||
bc38a6ab RD |
263 | kfree(file); |
264 | } | |
265 | ||
266 | static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf, | |
267 | size_t count, loff_t *pos) | |
268 | { | |
269 | struct ib_uverbs_event_file *file = filp->private_data; | |
63aaf647 | 270 | struct ib_uverbs_event *event; |
bc38a6ab RD |
271 | int eventsz; |
272 | int ret = 0; | |
273 | ||
274 | spin_lock_irq(&file->lock); | |
275 | ||
6b73597e | 276 | while (list_empty(&file->event_list)) { |
bc38a6ab RD |
277 | spin_unlock_irq(&file->lock); |
278 | ||
279 | if (filp->f_flags & O_NONBLOCK) | |
280 | return -EAGAIN; | |
281 | ||
282 | if (wait_event_interruptible(file->poll_wait, | |
6b73597e | 283 | !list_empty(&file->event_list))) |
bc38a6ab RD |
284 | return -ERESTARTSYS; |
285 | ||
286 | spin_lock_irq(&file->lock); | |
287 | } | |
288 | ||
63aaf647 RD |
289 | event = list_entry(file->event_list.next, struct ib_uverbs_event, list); |
290 | ||
291 | if (file->is_async) | |
bc38a6ab | 292 | eventsz = sizeof (struct ib_uverbs_async_event_desc); |
63aaf647 | 293 | else |
bc38a6ab | 294 | eventsz = sizeof (struct ib_uverbs_comp_event_desc); |
bc38a6ab RD |
295 | |
296 | if (eventsz > count) { | |
297 | ret = -EINVAL; | |
298 | event = NULL; | |
63aaf647 | 299 | } else { |
bc38a6ab | 300 | list_del(file->event_list.next); |
63aaf647 RD |
301 | if (event->counter) { |
302 | ++(*event->counter); | |
303 | list_del(&event->obj_list); | |
304 | } | |
305 | } | |
bc38a6ab RD |
306 | |
307 | spin_unlock_irq(&file->lock); | |
308 | ||
309 | if (event) { | |
310 | if (copy_to_user(buf, event, eventsz)) | |
311 | ret = -EFAULT; | |
312 | else | |
313 | ret = eventsz; | |
314 | } | |
315 | ||
316 | kfree(event); | |
317 | ||
318 | return ret; | |
319 | } | |
320 | ||
321 | static unsigned int ib_uverbs_event_poll(struct file *filp, | |
322 | struct poll_table_struct *wait) | |
323 | { | |
324 | unsigned int pollflags = 0; | |
325 | struct ib_uverbs_event_file *file = filp->private_data; | |
326 | ||
327 | poll_wait(filp, &file->poll_wait, wait); | |
328 | ||
329 | spin_lock_irq(&file->lock); | |
6b73597e | 330 | if (!list_empty(&file->event_list)) |
bc38a6ab RD |
331 | pollflags = POLLIN | POLLRDNORM; |
332 | spin_unlock_irq(&file->lock); | |
333 | ||
334 | return pollflags; | |
335 | } | |
336 | ||
abdf119b GN |
337 | static int ib_uverbs_event_fasync(int fd, struct file *filp, int on) |
338 | { | |
339 | struct ib_uverbs_event_file *file = filp->private_data; | |
340 | ||
341 | return fasync_helper(fd, filp, on, &file->async_queue); | |
342 | } | |
343 | ||
bc38a6ab RD |
344 | static int ib_uverbs_event_close(struct inode *inode, struct file *filp) |
345 | { | |
346 | struct ib_uverbs_event_file *file = filp->private_data; | |
6b73597e RD |
347 | struct ib_uverbs_event *entry, *tmp; |
348 | ||
349 | spin_lock_irq(&file->lock); | |
1ae5c187 | 350 | file->is_closed = 1; |
6b73597e RD |
351 | list_for_each_entry_safe(entry, tmp, &file->event_list, list) { |
352 | if (entry->counter) | |
353 | list_del(&entry->obj_list); | |
354 | kfree(entry); | |
355 | } | |
356 | spin_unlock_irq(&file->lock); | |
bc38a6ab | 357 | |
6b73597e RD |
358 | if (file->is_async) { |
359 | ib_unregister_event_handler(&file->uverbs_file->event_handler); | |
360 | kref_put(&file->uverbs_file->ref, ib_uverbs_release_file); | |
361 | } | |
362 | kref_put(&file->ref, ib_uverbs_release_event_file); | |
bc38a6ab RD |
363 | |
364 | return 0; | |
365 | } | |
366 | ||
2b8693c0 | 367 | static const struct file_operations uverbs_event_fops = { |
6b73597e | 368 | .owner = THIS_MODULE, |
9afed76d | 369 | .read = ib_uverbs_event_read, |
bc38a6ab | 370 | .poll = ib_uverbs_event_poll, |
abdf119b | 371 | .release = ib_uverbs_event_close, |
bc1db9af RD |
372 | .fasync = ib_uverbs_event_fasync, |
373 | .llseek = no_llseek, | |
bc38a6ab RD |
374 | }; |
375 | ||
376 | void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context) | |
377 | { | |
6b73597e RD |
378 | struct ib_uverbs_event_file *file = cq_context; |
379 | struct ib_ucq_object *uobj; | |
380 | struct ib_uverbs_event *entry; | |
381 | unsigned long flags; | |
382 | ||
383 | if (!file) | |
384 | return; | |
385 | ||
386 | spin_lock_irqsave(&file->lock, flags); | |
1ae5c187 | 387 | if (file->is_closed) { |
6b73597e RD |
388 | spin_unlock_irqrestore(&file->lock, flags); |
389 | return; | |
390 | } | |
bc38a6ab RD |
391 | |
392 | entry = kmalloc(sizeof *entry, GFP_ATOMIC); | |
305a7e87 RD |
393 | if (!entry) { |
394 | spin_unlock_irqrestore(&file->lock, flags); | |
bc38a6ab | 395 | return; |
305a7e87 | 396 | } |
bc38a6ab | 397 | |
63aaf647 RD |
398 | uobj = container_of(cq->uobject, struct ib_ucq_object, uobject); |
399 | ||
400 | entry->desc.comp.cq_handle = cq->uobject->user_handle; | |
401 | entry->counter = &uobj->comp_events_reported; | |
bc38a6ab | 402 | |
6b73597e | 403 | list_add_tail(&entry->list, &file->event_list); |
63aaf647 | 404 | list_add_tail(&entry->obj_list, &uobj->comp_list); |
6b73597e | 405 | spin_unlock_irqrestore(&file->lock, flags); |
bc38a6ab | 406 | |
6b73597e RD |
407 | wake_up_interruptible(&file->poll_wait); |
408 | kill_fasync(&file->async_queue, SIGIO, POLL_IN); | |
bc38a6ab RD |
409 | } |
410 | ||
411 | static void ib_uverbs_async_handler(struct ib_uverbs_file *file, | |
63aaf647 RD |
412 | __u64 element, __u64 event, |
413 | struct list_head *obj_list, | |
414 | u32 *counter) | |
bc38a6ab | 415 | { |
63aaf647 | 416 | struct ib_uverbs_event *entry; |
bc38a6ab RD |
417 | unsigned long flags; |
418 | ||
6b73597e | 419 | spin_lock_irqsave(&file->async_file->lock, flags); |
fb77bcef | 420 | if (file->async_file->is_closed) { |
6b73597e RD |
421 | spin_unlock_irqrestore(&file->async_file->lock, flags); |
422 | return; | |
423 | } | |
424 | ||
bc38a6ab | 425 | entry = kmalloc(sizeof *entry, GFP_ATOMIC); |
305a7e87 RD |
426 | if (!entry) { |
427 | spin_unlock_irqrestore(&file->async_file->lock, flags); | |
bc38a6ab | 428 | return; |
305a7e87 | 429 | } |
bc38a6ab | 430 | |
63aaf647 RD |
431 | entry->desc.async.element = element; |
432 | entry->desc.async.event_type = event; | |
433 | entry->counter = counter; | |
bc38a6ab | 434 | |
6b73597e | 435 | list_add_tail(&entry->list, &file->async_file->event_list); |
63aaf647 RD |
436 | if (obj_list) |
437 | list_add_tail(&entry->obj_list, obj_list); | |
6b73597e | 438 | spin_unlock_irqrestore(&file->async_file->lock, flags); |
bc38a6ab | 439 | |
6b73597e RD |
440 | wake_up_interruptible(&file->async_file->poll_wait); |
441 | kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN); | |
bc38a6ab RD |
442 | } |
443 | ||
444 | void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr) | |
445 | { | |
7162a3e0 RD |
446 | struct ib_ucq_object *uobj = container_of(event->element.cq->uobject, |
447 | struct ib_ucq_object, uobject); | |
63aaf647 | 448 | |
7162a3e0 | 449 | ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle, |
63aaf647 RD |
450 | event->event, &uobj->async_list, |
451 | &uobj->async_events_reported); | |
bc38a6ab RD |
452 | } |
453 | ||
454 | void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr) | |
455 | { | |
63aaf647 RD |
456 | struct ib_uevent_object *uobj; |
457 | ||
458 | uobj = container_of(event->element.qp->uobject, | |
459 | struct ib_uevent_object, uobject); | |
460 | ||
461 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
462 | event->event, &uobj->event_list, | |
463 | &uobj->events_reported); | |
bc38a6ab RD |
464 | } |
465 | ||
f520ba5a RD |
466 | void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr) |
467 | { | |
63aaf647 RD |
468 | struct ib_uevent_object *uobj; |
469 | ||
470 | uobj = container_of(event->element.srq->uobject, | |
471 | struct ib_uevent_object, uobject); | |
472 | ||
473 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
474 | event->event, &uobj->event_list, | |
475 | &uobj->events_reported); | |
f520ba5a RD |
476 | } |
477 | ||
6b73597e RD |
478 | void ib_uverbs_event_handler(struct ib_event_handler *handler, |
479 | struct ib_event *event) | |
bc38a6ab RD |
480 | { |
481 | struct ib_uverbs_file *file = | |
482 | container_of(handler, struct ib_uverbs_file, event_handler); | |
483 | ||
63aaf647 RD |
484 | ib_uverbs_async_handler(file, event->element.port_num, event->event, |
485 | NULL, NULL); | |
bc38a6ab RD |
486 | } |
487 | ||
6b73597e | 488 | struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file, |
b1e4594b | 489 | int is_async) |
bc38a6ab | 490 | { |
6b73597e | 491 | struct ib_uverbs_event_file *ev_file; |
bc38a6ab RD |
492 | struct file *filp; |
493 | ||
6b73597e RD |
494 | ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL); |
495 | if (!ev_file) | |
496 | return ERR_PTR(-ENOMEM); | |
497 | ||
498 | kref_init(&ev_file->ref); | |
499 | spin_lock_init(&ev_file->lock); | |
500 | INIT_LIST_HEAD(&ev_file->event_list); | |
501 | init_waitqueue_head(&ev_file->poll_wait); | |
502 | ev_file->uverbs_file = uverbs_file; | |
503 | ev_file->async_queue = NULL; | |
504 | ev_file->is_async = is_async; | |
1ae5c187 | 505 | ev_file->is_closed = 0; |
6b73597e | 506 | |
b1e4594b | 507 | filp = anon_inode_getfile("[infinibandevent]", &uverbs_event_fops, |
a265e558 | 508 | ev_file, O_RDONLY); |
b1e4594b AV |
509 | if (IS_ERR(filp)) |
510 | kfree(ev_file); | |
a7dab9e8 | 511 | |
6b73597e | 512 | return filp; |
6b73597e RD |
513 | } |
514 | ||
515 | /* | |
516 | * Look up a completion event file by FD. If lookup is successful, | |
517 | * takes a ref to the event file struct that it returns; if | |
518 | * unsuccessful, returns NULL. | |
519 | */ | |
520 | struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd) | |
521 | { | |
522 | struct ib_uverbs_event_file *ev_file = NULL; | |
523 | struct file *filp; | |
524 | ||
525 | filp = fget(fd); | |
526 | if (!filp) | |
527 | return NULL; | |
528 | ||
529 | if (filp->f_op != &uverbs_event_fops) | |
530 | goto out; | |
531 | ||
532 | ev_file = filp->private_data; | |
533 | if (ev_file->is_async) { | |
534 | ev_file = NULL; | |
535 | goto out; | |
536 | } | |
537 | ||
538 | kref_get(&ev_file->ref); | |
539 | ||
540 | out: | |
541 | fput(filp); | |
542 | return ev_file; | |
bc38a6ab RD |
543 | } |
544 | ||
545 | static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf, | |
546 | size_t count, loff_t *pos) | |
547 | { | |
548 | struct ib_uverbs_file *file = filp->private_data; | |
549 | struct ib_uverbs_cmd_hdr hdr; | |
550 | ||
551 | if (count < sizeof hdr) | |
552 | return -EINVAL; | |
553 | ||
554 | if (copy_from_user(&hdr, buf, sizeof hdr)) | |
555 | return -EFAULT; | |
556 | ||
557 | if (hdr.in_words * 4 != count) | |
558 | return -EINVAL; | |
559 | ||
63c47c28 RD |
560 | if (hdr.command < 0 || |
561 | hdr.command >= ARRAY_SIZE(uverbs_cmd_table) || | |
b1b8afb8 | 562 | !uverbs_cmd_table[hdr.command]) |
bc38a6ab RD |
563 | return -EINVAL; |
564 | ||
6b73597e | 565 | if (!file->ucontext && |
bc38a6ab RD |
566 | hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT) |
567 | return -EINVAL; | |
568 | ||
b1b8afb8 JM |
569 | if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command))) |
570 | return -ENOSYS; | |
571 | ||
bc38a6ab RD |
572 | return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr, |
573 | hdr.in_words * 4, hdr.out_words * 4); | |
574 | } | |
575 | ||
576 | static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma) | |
577 | { | |
578 | struct ib_uverbs_file *file = filp->private_data; | |
579 | ||
580 | if (!file->ucontext) | |
581 | return -ENODEV; | |
582 | else | |
583 | return file->device->ib_dev->mmap(file->ucontext, vma); | |
584 | } | |
585 | ||
5b2d281a RD |
586 | /* |
587 | * ib_uverbs_open() does not need the BKL: | |
588 | * | |
2a72f212 | 589 | * - the ib_uverbs_device structures are properly reference counted and |
5b2d281a RD |
590 | * everything else is purely local to the file being created, so |
591 | * races against other open calls are not a problem; | |
592 | * - there is no ioctl method to race against; | |
2a72f212 AC |
593 | * - the open method will either immediately run -ENXIO, or all |
594 | * required initialization will be done. | |
5b2d281a | 595 | */ |
bc38a6ab RD |
596 | static int ib_uverbs_open(struct inode *inode, struct file *filp) |
597 | { | |
70a30e16 | 598 | struct ib_uverbs_device *dev; |
bc38a6ab | 599 | struct ib_uverbs_file *file; |
70a30e16 | 600 | int ret; |
bc38a6ab | 601 | |
2a72f212 | 602 | dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev); |
70a30e16 RD |
603 | if (dev) |
604 | kref_get(&dev->ref); | |
2a72f212 | 605 | else |
70a30e16 RD |
606 | return -ENXIO; |
607 | ||
608 | if (!try_module_get(dev->ib_dev->owner)) { | |
609 | ret = -ENODEV; | |
610 | goto err; | |
611 | } | |
bc38a6ab | 612 | |
6b73597e | 613 | file = kmalloc(sizeof *file, GFP_KERNEL); |
63c47c28 | 614 | if (!file) { |
70a30e16 RD |
615 | ret = -ENOMEM; |
616 | goto err_module; | |
63c47c28 | 617 | } |
bc38a6ab | 618 | |
70a30e16 RD |
619 | file->device = dev; |
620 | file->ucontext = NULL; | |
621 | file->async_file = NULL; | |
bc38a6ab | 622 | kref_init(&file->ref); |
95ed644f | 623 | mutex_init(&file->mutex); |
bc38a6ab | 624 | |
bc38a6ab RD |
625 | filp->private_data = file; |
626 | ||
bc1db9af | 627 | return nonseekable_open(inode, filp); |
70a30e16 RD |
628 | |
629 | err_module: | |
630 | module_put(dev->ib_dev->owner); | |
631 | ||
632 | err: | |
633 | kref_put(&dev->ref, ib_uverbs_release_dev); | |
70a30e16 | 634 | return ret; |
bc38a6ab RD |
635 | } |
636 | ||
637 | static int ib_uverbs_close(struct inode *inode, struct file *filp) | |
638 | { | |
639 | struct ib_uverbs_file *file = filp->private_data; | |
bc38a6ab | 640 | |
70a30e16 RD |
641 | ib_uverbs_cleanup_ucontext(file, file->ucontext); |
642 | ||
643 | if (file->async_file) | |
644 | kref_put(&file->async_file->ref, ib_uverbs_release_event_file); | |
bc38a6ab | 645 | |
bc38a6ab RD |
646 | kref_put(&file->ref, ib_uverbs_release_file); |
647 | ||
648 | return 0; | |
649 | } | |
650 | ||
2b8693c0 | 651 | static const struct file_operations uverbs_fops = { |
9afed76d AC |
652 | .owner = THIS_MODULE, |
653 | .write = ib_uverbs_write, | |
654 | .open = ib_uverbs_open, | |
bc1db9af RD |
655 | .release = ib_uverbs_close, |
656 | .llseek = no_llseek, | |
bc38a6ab RD |
657 | }; |
658 | ||
2b8693c0 | 659 | static const struct file_operations uverbs_mmap_fops = { |
9afed76d AC |
660 | .owner = THIS_MODULE, |
661 | .write = ib_uverbs_write, | |
bc38a6ab | 662 | .mmap = ib_uverbs_mmap, |
9afed76d | 663 | .open = ib_uverbs_open, |
bc1db9af RD |
664 | .release = ib_uverbs_close, |
665 | .llseek = no_llseek, | |
bc38a6ab RD |
666 | }; |
667 | ||
668 | static struct ib_client uverbs_client = { | |
669 | .name = "uverbs", | |
670 | .add = ib_uverbs_add_one, | |
671 | .remove = ib_uverbs_remove_one | |
672 | }; | |
673 | ||
f4e91eb4 TJ |
674 | static ssize_t show_ibdev(struct device *device, struct device_attribute *attr, |
675 | char *buf) | |
bc38a6ab | 676 | { |
f4e91eb4 | 677 | struct ib_uverbs_device *dev = dev_get_drvdata(device); |
70a30e16 RD |
678 | |
679 | if (!dev) | |
680 | return -ENODEV; | |
bc38a6ab RD |
681 | |
682 | return sprintf(buf, "%s\n", dev->ib_dev->name); | |
683 | } | |
f4e91eb4 | 684 | static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); |
bc38a6ab | 685 | |
f4e91eb4 TJ |
686 | static ssize_t show_dev_abi_version(struct device *device, |
687 | struct device_attribute *attr, char *buf) | |
274c0891 | 688 | { |
f4e91eb4 | 689 | struct ib_uverbs_device *dev = dev_get_drvdata(device); |
70a30e16 RD |
690 | |
691 | if (!dev) | |
692 | return -ENODEV; | |
274c0891 RD |
693 | |
694 | return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver); | |
695 | } | |
f4e91eb4 | 696 | static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL); |
274c0891 | 697 | |
0933e2d9 AK |
698 | static CLASS_ATTR_STRING(abi_version, S_IRUGO, |
699 | __stringify(IB_USER_VERBS_ABI_VERSION)); | |
bc38a6ab | 700 | |
6d6a0e71 AC |
701 | static dev_t overflow_maj; |
702 | static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES); | |
703 | ||
704 | /* | |
705 | * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by | |
706 | * requesting a new major number and doubling the number of max devices we | |
707 | * support. It's stupid, but simple. | |
708 | */ | |
709 | static int find_overflow_devnum(void) | |
710 | { | |
711 | int ret; | |
712 | ||
713 | if (!overflow_maj) { | |
714 | ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES, | |
715 | "infiniband_verbs"); | |
716 | if (ret) { | |
717 | printk(KERN_ERR "user_verbs: couldn't register dynamic device number\n"); | |
718 | return ret; | |
719 | } | |
720 | } | |
721 | ||
722 | ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES); | |
723 | if (ret >= IB_UVERBS_MAX_DEVICES) | |
724 | return -1; | |
725 | ||
726 | return ret; | |
727 | } | |
728 | ||
bc38a6ab RD |
729 | static void ib_uverbs_add_one(struct ib_device *device) |
730 | { | |
38707980 | 731 | int devnum; |
ddbd6883 | 732 | dev_t base; |
bc38a6ab RD |
733 | struct ib_uverbs_device *uverbs_dev; |
734 | ||
735 | if (!device->alloc_ucontext) | |
736 | return; | |
737 | ||
de6eb66b | 738 | uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL); |
bc38a6ab RD |
739 | if (!uverbs_dev) |
740 | return; | |
741 | ||
70a30e16 | 742 | kref_init(&uverbs_dev->ref); |
fd60ae40 | 743 | init_completion(&uverbs_dev->comp); |
70a30e16 | 744 | |
bc38a6ab | 745 | spin_lock(&map_lock); |
38707980 AC |
746 | devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES); |
747 | if (devnum >= IB_UVERBS_MAX_DEVICES) { | |
bc38a6ab | 748 | spin_unlock(&map_lock); |
6d6a0e71 AC |
749 | devnum = find_overflow_devnum(); |
750 | if (devnum < 0) | |
751 | goto err; | |
752 | ||
753 | spin_lock(&map_lock); | |
754 | uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES; | |
755 | base = devnum + overflow_maj; | |
756 | set_bit(devnum, overflow_map); | |
757 | } else { | |
758 | uverbs_dev->devnum = devnum; | |
759 | base = devnum + IB_UVERBS_BASE_DEV; | |
760 | set_bit(devnum, dev_map); | |
bc38a6ab | 761 | } |
bc38a6ab RD |
762 | spin_unlock(&map_lock); |
763 | ||
6b73597e | 764 | uverbs_dev->ib_dev = device; |
f4fd0b22 | 765 | uverbs_dev->num_comp_vectors = device->num_comp_vectors; |
bc38a6ab | 766 | |
055422dd AC |
767 | cdev_init(&uverbs_dev->cdev, NULL); |
768 | uverbs_dev->cdev.owner = THIS_MODULE; | |
769 | uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops; | |
770 | kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum); | |
ddbd6883 | 771 | if (cdev_add(&uverbs_dev->cdev, base, 1)) |
70a30e16 | 772 | goto err_cdev; |
bc38a6ab | 773 | |
91bd418f | 774 | uverbs_dev->dev = device_create(uverbs_class, device->dma_device, |
055422dd | 775 | uverbs_dev->cdev.dev, uverbs_dev, |
91bd418f | 776 | "uverbs%d", uverbs_dev->devnum); |
f4e91eb4 | 777 | if (IS_ERR(uverbs_dev->dev)) |
bc38a6ab RD |
778 | goto err_cdev; |
779 | ||
f4e91eb4 | 780 | if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev)) |
bc38a6ab | 781 | goto err_class; |
f4e91eb4 | 782 | if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version)) |
274c0891 | 783 | goto err_class; |
bc38a6ab RD |
784 | |
785 | ib_set_client_data(device, &uverbs_client, uverbs_dev); | |
786 | ||
787 | return; | |
788 | ||
789 | err_class: | |
055422dd | 790 | device_destroy(uverbs_class, uverbs_dev->cdev.dev); |
bc38a6ab RD |
791 | |
792 | err_cdev: | |
055422dd | 793 | cdev_del(&uverbs_dev->cdev); |
6d6a0e71 AC |
794 | if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES) |
795 | clear_bit(devnum, dev_map); | |
796 | else | |
797 | clear_bit(devnum, overflow_map); | |
bc38a6ab RD |
798 | |
799 | err: | |
70a30e16 | 800 | kref_put(&uverbs_dev->ref, ib_uverbs_release_dev); |
fd60ae40 JM |
801 | wait_for_completion(&uverbs_dev->comp); |
802 | kfree(uverbs_dev); | |
bc38a6ab RD |
803 | return; |
804 | } | |
805 | ||
806 | static void ib_uverbs_remove_one(struct ib_device *device) | |
807 | { | |
808 | struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client); | |
809 | ||
810 | if (!uverbs_dev) | |
811 | return; | |
812 | ||
f4e91eb4 | 813 | dev_set_drvdata(uverbs_dev->dev, NULL); |
055422dd AC |
814 | device_destroy(uverbs_class, uverbs_dev->cdev.dev); |
815 | cdev_del(&uverbs_dev->cdev); | |
70a30e16 | 816 | |
6d6a0e71 AC |
817 | if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES) |
818 | clear_bit(uverbs_dev->devnum, dev_map); | |
819 | else | |
820 | clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map); | |
fd60ae40 | 821 | |
70a30e16 | 822 | kref_put(&uverbs_dev->ref, ib_uverbs_release_dev); |
fd60ae40 JM |
823 | wait_for_completion(&uverbs_dev->comp); |
824 | kfree(uverbs_dev); | |
bc38a6ab RD |
825 | } |
826 | ||
71c29bd5 RD |
827 | static char *uverbs_devnode(struct device *dev, mode_t *mode) |
828 | { | |
b2bc4782 GR |
829 | if (mode) |
830 | *mode = 0666; | |
71c29bd5 RD |
831 | return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev)); |
832 | } | |
833 | ||
bc38a6ab RD |
834 | static int __init ib_uverbs_init(void) |
835 | { | |
836 | int ret; | |
837 | ||
bc38a6ab RD |
838 | ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES, |
839 | "infiniband_verbs"); | |
840 | if (ret) { | |
841 | printk(KERN_ERR "user_verbs: couldn't register device number\n"); | |
842 | goto out; | |
843 | } | |
844 | ||
70a30e16 RD |
845 | uverbs_class = class_create(THIS_MODULE, "infiniband_verbs"); |
846 | if (IS_ERR(uverbs_class)) { | |
847 | ret = PTR_ERR(uverbs_class); | |
bc38a6ab RD |
848 | printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n"); |
849 | goto out_chrdev; | |
850 | } | |
851 | ||
71c29bd5 RD |
852 | uverbs_class->devnode = uverbs_devnode; |
853 | ||
0933e2d9 | 854 | ret = class_create_file(uverbs_class, &class_attr_abi_version.attr); |
bc38a6ab RD |
855 | if (ret) { |
856 | printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n"); | |
857 | goto out_class; | |
858 | } | |
859 | ||
bc38a6ab RD |
860 | ret = ib_register_client(&uverbs_client); |
861 | if (ret) { | |
862 | printk(KERN_ERR "user_verbs: couldn't register client\n"); | |
a265e558 | 863 | goto out_class; |
bc38a6ab RD |
864 | } |
865 | ||
866 | return 0; | |
867 | ||
bc38a6ab | 868 | out_class: |
70a30e16 | 869 | class_destroy(uverbs_class); |
bc38a6ab RD |
870 | |
871 | out_chrdev: | |
872 | unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES); | |
873 | ||
874 | out: | |
875 | return ret; | |
876 | } | |
877 | ||
878 | static void __exit ib_uverbs_cleanup(void) | |
879 | { | |
880 | ib_unregister_client(&uverbs_client); | |
70a30e16 | 881 | class_destroy(uverbs_class); |
bc38a6ab | 882 | unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES); |
6d6a0e71 AC |
883 | if (overflow_maj) |
884 | unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES); | |
5d7edb3c RD |
885 | idr_destroy(&ib_uverbs_pd_idr); |
886 | idr_destroy(&ib_uverbs_mr_idr); | |
887 | idr_destroy(&ib_uverbs_mw_idr); | |
888 | idr_destroy(&ib_uverbs_ah_idr); | |
889 | idr_destroy(&ib_uverbs_cq_idr); | |
890 | idr_destroy(&ib_uverbs_qp_idr); | |
891 | idr_destroy(&ib_uverbs_srq_idr); | |
bc38a6ab RD |
892 | } |
893 | ||
894 | module_init(ib_uverbs_init); | |
895 | module_exit(ib_uverbs_cleanup); |