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); |
53d0bd1e | 75 | DEFINE_IDR(ib_uverbs_xrcd_idr); |
bc38a6ab | 76 | |
6276e08a | 77 | static DEFINE_SPINLOCK(map_lock); |
bc38a6ab RD |
78 | static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES); |
79 | ||
80 | static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file, | |
81 | const char __user *buf, int in_len, | |
82 | int out_len) = { | |
9afed76d AC |
83 | [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context, |
84 | [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device, | |
85 | [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port, | |
86 | [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd, | |
87 | [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd, | |
88 | [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr, | |
89 | [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr, | |
6b73597e | 90 | [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel, |
9afed76d AC |
91 | [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq, |
92 | [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq, | |
93 | [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq, | |
94 | [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq, | |
95 | [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq, | |
96 | [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp, | |
97 | [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp, | |
98 | [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp, | |
99 | [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp, | |
100 | [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send, | |
101 | [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv, | |
102 | [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv, | |
103 | [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah, | |
104 | [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah, | |
105 | [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast, | |
106 | [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast, | |
107 | [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq, | |
108 | [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq, | |
109 | [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq, | |
110 | [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq, | |
53d0bd1e SH |
111 | [IB_USER_VERBS_CMD_OPEN_XRCD] = ib_uverbs_open_xrcd, |
112 | [IB_USER_VERBS_CMD_CLOSE_XRCD] = ib_uverbs_close_xrcd, | |
42849b26 SH |
113 | [IB_USER_VERBS_CMD_CREATE_XSRQ] = ib_uverbs_create_xsrq, |
114 | [IB_USER_VERBS_CMD_OPEN_QP] = ib_uverbs_open_qp | |
bc38a6ab RD |
115 | }; |
116 | ||
bc38a6ab RD |
117 | static void ib_uverbs_add_one(struct ib_device *device); |
118 | static void ib_uverbs_remove_one(struct ib_device *device); | |
119 | ||
70a30e16 RD |
120 | static void ib_uverbs_release_dev(struct kref *ref) |
121 | { | |
122 | struct ib_uverbs_device *dev = | |
123 | container_of(ref, struct ib_uverbs_device, ref); | |
124 | ||
fd60ae40 | 125 | complete(&dev->comp); |
70a30e16 RD |
126 | } |
127 | ||
04d29b0e RD |
128 | static void ib_uverbs_release_event_file(struct kref *ref) |
129 | { | |
130 | struct ib_uverbs_event_file *file = | |
131 | container_of(ref, struct ib_uverbs_event_file, ref); | |
132 | ||
133 | kfree(file); | |
134 | } | |
135 | ||
70a30e16 RD |
136 | void ib_uverbs_release_ucq(struct ib_uverbs_file *file, |
137 | struct ib_uverbs_event_file *ev_file, | |
138 | struct ib_ucq_object *uobj) | |
139 | { | |
140 | struct ib_uverbs_event *evt, *tmp; | |
141 | ||
142 | if (ev_file) { | |
143 | spin_lock_irq(&ev_file->lock); | |
144 | list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) { | |
145 | list_del(&evt->list); | |
146 | kfree(evt); | |
147 | } | |
148 | spin_unlock_irq(&ev_file->lock); | |
149 | ||
150 | kref_put(&ev_file->ref, ib_uverbs_release_event_file); | |
151 | } | |
152 | ||
153 | spin_lock_irq(&file->async_file->lock); | |
154 | list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) { | |
155 | list_del(&evt->list); | |
156 | kfree(evt); | |
157 | } | |
158 | spin_unlock_irq(&file->async_file->lock); | |
159 | } | |
160 | ||
161 | void ib_uverbs_release_uevent(struct ib_uverbs_file *file, | |
162 | struct ib_uevent_object *uobj) | |
163 | { | |
164 | struct ib_uverbs_event *evt, *tmp; | |
165 | ||
166 | spin_lock_irq(&file->async_file->lock); | |
167 | list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) { | |
168 | list_del(&evt->list); | |
169 | kfree(evt); | |
170 | } | |
171 | spin_unlock_irq(&file->async_file->lock); | |
172 | } | |
173 | ||
f4e40156 JM |
174 | static void ib_uverbs_detach_umcast(struct ib_qp *qp, |
175 | struct ib_uqp_object *uobj) | |
176 | { | |
177 | struct ib_uverbs_mcast_entry *mcast, *tmp; | |
178 | ||
179 | list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) { | |
180 | ib_detach_mcast(qp, &mcast->gid, mcast->lid); | |
181 | list_del(&mcast->list); | |
182 | kfree(mcast); | |
183 | } | |
184 | } | |
185 | ||
70a30e16 RD |
186 | static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file, |
187 | struct ib_ucontext *context) | |
bc38a6ab RD |
188 | { |
189 | struct ib_uobject *uobj, *tmp; | |
190 | ||
191 | if (!context) | |
192 | return 0; | |
193 | ||
f7c6a7b5 RD |
194 | context->closing = 1; |
195 | ||
67cdb40c | 196 | list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) { |
9ead190b RD |
197 | struct ib_ah *ah = uobj->object; |
198 | ||
199 | idr_remove_uobj(&ib_uverbs_ah_idr, uobj); | |
67cdb40c | 200 | ib_destroy_ah(ah); |
67cdb40c RD |
201 | kfree(uobj); |
202 | } | |
bc38a6ab RD |
203 | |
204 | list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) { | |
9ead190b | 205 | struct ib_qp *qp = uobj->object; |
f4e40156 JM |
206 | struct ib_uqp_object *uqp = |
207 | container_of(uobj, struct ib_uqp_object, uevent.uobject); | |
9ead190b RD |
208 | |
209 | idr_remove_uobj(&ib_uverbs_qp_idr, uobj); | |
0e0ec7e0 SH |
210 | if (qp != qp->real_qp) { |
211 | ib_close_qp(qp); | |
b93f3c18 SH |
212 | } else { |
213 | ib_uverbs_detach_umcast(qp, uqp); | |
214 | ib_destroy_qp(qp); | |
215 | } | |
f4e40156 JM |
216 | ib_uverbs_release_uevent(file, &uqp->uevent); |
217 | kfree(uqp); | |
bc38a6ab RD |
218 | } |
219 | ||
220 | list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) { | |
9ead190b | 221 | struct ib_cq *cq = uobj->object; |
70a30e16 RD |
222 | struct ib_uverbs_event_file *ev_file = cq->cq_context; |
223 | struct ib_ucq_object *ucq = | |
224 | container_of(uobj, struct ib_ucq_object, uobject); | |
9ead190b RD |
225 | |
226 | idr_remove_uobj(&ib_uverbs_cq_idr, uobj); | |
bc38a6ab | 227 | ib_destroy_cq(cq); |
70a30e16 RD |
228 | ib_uverbs_release_ucq(file, ev_file, ucq); |
229 | kfree(ucq); | |
bc38a6ab RD |
230 | } |
231 | ||
f520ba5a | 232 | list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) { |
9ead190b | 233 | struct ib_srq *srq = uobj->object; |
70a30e16 RD |
234 | struct ib_uevent_object *uevent = |
235 | container_of(uobj, struct ib_uevent_object, uobject); | |
9ead190b RD |
236 | |
237 | idr_remove_uobj(&ib_uverbs_srq_idr, uobj); | |
f520ba5a | 238 | ib_destroy_srq(srq); |
70a30e16 RD |
239 | ib_uverbs_release_uevent(file, uevent); |
240 | kfree(uevent); | |
f520ba5a RD |
241 | } |
242 | ||
bc38a6ab RD |
243 | /* XXX Free MWs */ |
244 | ||
245 | list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) { | |
9ead190b | 246 | struct ib_mr *mr = uobj->object; |
bc38a6ab | 247 | |
9ead190b | 248 | idr_remove_uobj(&ib_uverbs_mr_idr, uobj); |
bc38a6ab | 249 | ib_dereg_mr(mr); |
f7c6a7b5 | 250 | kfree(uobj); |
bc38a6ab RD |
251 | } |
252 | ||
53d0bd1e SH |
253 | mutex_lock(&file->device->xrcd_tree_mutex); |
254 | list_for_each_entry_safe(uobj, tmp, &context->xrcd_list, list) { | |
255 | struct ib_xrcd *xrcd = uobj->object; | |
256 | struct ib_uxrcd_object *uxrcd = | |
257 | container_of(uobj, struct ib_uxrcd_object, uobject); | |
258 | ||
259 | idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj); | |
260 | ib_uverbs_dealloc_xrcd(file->device, xrcd); | |
261 | kfree(uxrcd); | |
262 | } | |
263 | mutex_unlock(&file->device->xrcd_tree_mutex); | |
264 | ||
bc38a6ab | 265 | list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) { |
9ead190b RD |
266 | struct ib_pd *pd = uobj->object; |
267 | ||
268 | idr_remove_uobj(&ib_uverbs_pd_idr, uobj); | |
bc38a6ab | 269 | ib_dealloc_pd(pd); |
bc38a6ab RD |
270 | kfree(uobj); |
271 | } | |
272 | ||
bc38a6ab RD |
273 | return context->device->dealloc_ucontext(context); |
274 | } | |
275 | ||
276 | static void ib_uverbs_release_file(struct kref *ref) | |
277 | { | |
278 | struct ib_uverbs_file *file = | |
279 | container_of(ref, struct ib_uverbs_file, ref); | |
280 | ||
281 | module_put(file->device->ib_dev->owner); | |
70a30e16 RD |
282 | kref_put(&file->device->ref, ib_uverbs_release_dev); |
283 | ||
bc38a6ab RD |
284 | kfree(file); |
285 | } | |
286 | ||
287 | static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf, | |
288 | size_t count, loff_t *pos) | |
289 | { | |
290 | struct ib_uverbs_event_file *file = filp->private_data; | |
63aaf647 | 291 | struct ib_uverbs_event *event; |
bc38a6ab RD |
292 | int eventsz; |
293 | int ret = 0; | |
294 | ||
295 | spin_lock_irq(&file->lock); | |
296 | ||
6b73597e | 297 | while (list_empty(&file->event_list)) { |
bc38a6ab RD |
298 | spin_unlock_irq(&file->lock); |
299 | ||
300 | if (filp->f_flags & O_NONBLOCK) | |
301 | return -EAGAIN; | |
302 | ||
303 | if (wait_event_interruptible(file->poll_wait, | |
6b73597e | 304 | !list_empty(&file->event_list))) |
bc38a6ab RD |
305 | return -ERESTARTSYS; |
306 | ||
307 | spin_lock_irq(&file->lock); | |
308 | } | |
309 | ||
63aaf647 RD |
310 | event = list_entry(file->event_list.next, struct ib_uverbs_event, list); |
311 | ||
312 | if (file->is_async) | |
bc38a6ab | 313 | eventsz = sizeof (struct ib_uverbs_async_event_desc); |
63aaf647 | 314 | else |
bc38a6ab | 315 | eventsz = sizeof (struct ib_uverbs_comp_event_desc); |
bc38a6ab RD |
316 | |
317 | if (eventsz > count) { | |
318 | ret = -EINVAL; | |
319 | event = NULL; | |
63aaf647 | 320 | } else { |
bc38a6ab | 321 | list_del(file->event_list.next); |
63aaf647 RD |
322 | if (event->counter) { |
323 | ++(*event->counter); | |
324 | list_del(&event->obj_list); | |
325 | } | |
326 | } | |
bc38a6ab RD |
327 | |
328 | spin_unlock_irq(&file->lock); | |
329 | ||
330 | if (event) { | |
331 | if (copy_to_user(buf, event, eventsz)) | |
332 | ret = -EFAULT; | |
333 | else | |
334 | ret = eventsz; | |
335 | } | |
336 | ||
337 | kfree(event); | |
338 | ||
339 | return ret; | |
340 | } | |
341 | ||
342 | static unsigned int ib_uverbs_event_poll(struct file *filp, | |
343 | struct poll_table_struct *wait) | |
344 | { | |
345 | unsigned int pollflags = 0; | |
346 | struct ib_uverbs_event_file *file = filp->private_data; | |
347 | ||
348 | poll_wait(filp, &file->poll_wait, wait); | |
349 | ||
350 | spin_lock_irq(&file->lock); | |
6b73597e | 351 | if (!list_empty(&file->event_list)) |
bc38a6ab RD |
352 | pollflags = POLLIN | POLLRDNORM; |
353 | spin_unlock_irq(&file->lock); | |
354 | ||
355 | return pollflags; | |
356 | } | |
357 | ||
abdf119b GN |
358 | static int ib_uverbs_event_fasync(int fd, struct file *filp, int on) |
359 | { | |
360 | struct ib_uverbs_event_file *file = filp->private_data; | |
361 | ||
362 | return fasync_helper(fd, filp, on, &file->async_queue); | |
363 | } | |
364 | ||
bc38a6ab RD |
365 | static int ib_uverbs_event_close(struct inode *inode, struct file *filp) |
366 | { | |
367 | struct ib_uverbs_event_file *file = filp->private_data; | |
6b73597e RD |
368 | struct ib_uverbs_event *entry, *tmp; |
369 | ||
370 | spin_lock_irq(&file->lock); | |
1ae5c187 | 371 | file->is_closed = 1; |
6b73597e RD |
372 | list_for_each_entry_safe(entry, tmp, &file->event_list, list) { |
373 | if (entry->counter) | |
374 | list_del(&entry->obj_list); | |
375 | kfree(entry); | |
376 | } | |
377 | spin_unlock_irq(&file->lock); | |
bc38a6ab | 378 | |
6b73597e RD |
379 | if (file->is_async) { |
380 | ib_unregister_event_handler(&file->uverbs_file->event_handler); | |
381 | kref_put(&file->uverbs_file->ref, ib_uverbs_release_file); | |
382 | } | |
383 | kref_put(&file->ref, ib_uverbs_release_event_file); | |
bc38a6ab RD |
384 | |
385 | return 0; | |
386 | } | |
387 | ||
2b8693c0 | 388 | static const struct file_operations uverbs_event_fops = { |
6b73597e | 389 | .owner = THIS_MODULE, |
9afed76d | 390 | .read = ib_uverbs_event_read, |
bc38a6ab | 391 | .poll = ib_uverbs_event_poll, |
abdf119b | 392 | .release = ib_uverbs_event_close, |
bc1db9af RD |
393 | .fasync = ib_uverbs_event_fasync, |
394 | .llseek = no_llseek, | |
bc38a6ab RD |
395 | }; |
396 | ||
397 | void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context) | |
398 | { | |
6b73597e RD |
399 | struct ib_uverbs_event_file *file = cq_context; |
400 | struct ib_ucq_object *uobj; | |
401 | struct ib_uverbs_event *entry; | |
402 | unsigned long flags; | |
403 | ||
404 | if (!file) | |
405 | return; | |
406 | ||
407 | spin_lock_irqsave(&file->lock, flags); | |
1ae5c187 | 408 | if (file->is_closed) { |
6b73597e RD |
409 | spin_unlock_irqrestore(&file->lock, flags); |
410 | return; | |
411 | } | |
bc38a6ab RD |
412 | |
413 | entry = kmalloc(sizeof *entry, GFP_ATOMIC); | |
305a7e87 RD |
414 | if (!entry) { |
415 | spin_unlock_irqrestore(&file->lock, flags); | |
bc38a6ab | 416 | return; |
305a7e87 | 417 | } |
bc38a6ab | 418 | |
63aaf647 RD |
419 | uobj = container_of(cq->uobject, struct ib_ucq_object, uobject); |
420 | ||
421 | entry->desc.comp.cq_handle = cq->uobject->user_handle; | |
422 | entry->counter = &uobj->comp_events_reported; | |
bc38a6ab | 423 | |
6b73597e | 424 | list_add_tail(&entry->list, &file->event_list); |
63aaf647 | 425 | list_add_tail(&entry->obj_list, &uobj->comp_list); |
6b73597e | 426 | spin_unlock_irqrestore(&file->lock, flags); |
bc38a6ab | 427 | |
6b73597e RD |
428 | wake_up_interruptible(&file->poll_wait); |
429 | kill_fasync(&file->async_queue, SIGIO, POLL_IN); | |
bc38a6ab RD |
430 | } |
431 | ||
432 | static void ib_uverbs_async_handler(struct ib_uverbs_file *file, | |
63aaf647 RD |
433 | __u64 element, __u64 event, |
434 | struct list_head *obj_list, | |
435 | u32 *counter) | |
bc38a6ab | 436 | { |
63aaf647 | 437 | struct ib_uverbs_event *entry; |
bc38a6ab RD |
438 | unsigned long flags; |
439 | ||
6b73597e | 440 | spin_lock_irqsave(&file->async_file->lock, flags); |
fb77bcef | 441 | if (file->async_file->is_closed) { |
6b73597e RD |
442 | spin_unlock_irqrestore(&file->async_file->lock, flags); |
443 | return; | |
444 | } | |
445 | ||
bc38a6ab | 446 | entry = kmalloc(sizeof *entry, GFP_ATOMIC); |
305a7e87 RD |
447 | if (!entry) { |
448 | spin_unlock_irqrestore(&file->async_file->lock, flags); | |
bc38a6ab | 449 | return; |
305a7e87 | 450 | } |
bc38a6ab | 451 | |
63aaf647 RD |
452 | entry->desc.async.element = element; |
453 | entry->desc.async.event_type = event; | |
454 | entry->counter = counter; | |
bc38a6ab | 455 | |
6b73597e | 456 | list_add_tail(&entry->list, &file->async_file->event_list); |
63aaf647 RD |
457 | if (obj_list) |
458 | list_add_tail(&entry->obj_list, obj_list); | |
6b73597e | 459 | spin_unlock_irqrestore(&file->async_file->lock, flags); |
bc38a6ab | 460 | |
6b73597e RD |
461 | wake_up_interruptible(&file->async_file->poll_wait); |
462 | kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN); | |
bc38a6ab RD |
463 | } |
464 | ||
465 | void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr) | |
466 | { | |
7162a3e0 RD |
467 | struct ib_ucq_object *uobj = container_of(event->element.cq->uobject, |
468 | struct ib_ucq_object, uobject); | |
63aaf647 | 469 | |
7162a3e0 | 470 | ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle, |
63aaf647 RD |
471 | event->event, &uobj->async_list, |
472 | &uobj->async_events_reported); | |
bc38a6ab RD |
473 | } |
474 | ||
475 | void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr) | |
476 | { | |
63aaf647 RD |
477 | struct ib_uevent_object *uobj; |
478 | ||
479 | uobj = container_of(event->element.qp->uobject, | |
480 | struct ib_uevent_object, uobject); | |
481 | ||
482 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
483 | event->event, &uobj->event_list, | |
484 | &uobj->events_reported); | |
bc38a6ab RD |
485 | } |
486 | ||
f520ba5a RD |
487 | void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr) |
488 | { | |
63aaf647 RD |
489 | struct ib_uevent_object *uobj; |
490 | ||
491 | uobj = container_of(event->element.srq->uobject, | |
492 | struct ib_uevent_object, uobject); | |
493 | ||
494 | ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle, | |
495 | event->event, &uobj->event_list, | |
496 | &uobj->events_reported); | |
f520ba5a RD |
497 | } |
498 | ||
6b73597e RD |
499 | void ib_uverbs_event_handler(struct ib_event_handler *handler, |
500 | struct ib_event *event) | |
bc38a6ab RD |
501 | { |
502 | struct ib_uverbs_file *file = | |
503 | container_of(handler, struct ib_uverbs_file, event_handler); | |
504 | ||
63aaf647 RD |
505 | ib_uverbs_async_handler(file, event->element.port_num, event->event, |
506 | NULL, NULL); | |
bc38a6ab RD |
507 | } |
508 | ||
6b73597e | 509 | struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file, |
b1e4594b | 510 | int is_async) |
bc38a6ab | 511 | { |
6b73597e | 512 | struct ib_uverbs_event_file *ev_file; |
bc38a6ab RD |
513 | struct file *filp; |
514 | ||
6b73597e RD |
515 | ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL); |
516 | if (!ev_file) | |
517 | return ERR_PTR(-ENOMEM); | |
518 | ||
519 | kref_init(&ev_file->ref); | |
520 | spin_lock_init(&ev_file->lock); | |
521 | INIT_LIST_HEAD(&ev_file->event_list); | |
522 | init_waitqueue_head(&ev_file->poll_wait); | |
523 | ev_file->uverbs_file = uverbs_file; | |
524 | ev_file->async_queue = NULL; | |
525 | ev_file->is_async = is_async; | |
1ae5c187 | 526 | ev_file->is_closed = 0; |
6b73597e | 527 | |
b1e4594b | 528 | filp = anon_inode_getfile("[infinibandevent]", &uverbs_event_fops, |
a265e558 | 529 | ev_file, O_RDONLY); |
b1e4594b AV |
530 | if (IS_ERR(filp)) |
531 | kfree(ev_file); | |
a7dab9e8 | 532 | |
6b73597e | 533 | return filp; |
6b73597e RD |
534 | } |
535 | ||
536 | /* | |
537 | * Look up a completion event file by FD. If lookup is successful, | |
538 | * takes a ref to the event file struct that it returns; if | |
539 | * unsuccessful, returns NULL. | |
540 | */ | |
541 | struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd) | |
542 | { | |
543 | struct ib_uverbs_event_file *ev_file = NULL; | |
544 | struct file *filp; | |
545 | ||
546 | filp = fget(fd); | |
547 | if (!filp) | |
548 | return NULL; | |
549 | ||
550 | if (filp->f_op != &uverbs_event_fops) | |
551 | goto out; | |
552 | ||
553 | ev_file = filp->private_data; | |
554 | if (ev_file->is_async) { | |
555 | ev_file = NULL; | |
556 | goto out; | |
557 | } | |
558 | ||
559 | kref_get(&ev_file->ref); | |
560 | ||
561 | out: | |
562 | fput(filp); | |
563 | return ev_file; | |
bc38a6ab RD |
564 | } |
565 | ||
566 | static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf, | |
567 | size_t count, loff_t *pos) | |
568 | { | |
569 | struct ib_uverbs_file *file = filp->private_data; | |
570 | struct ib_uverbs_cmd_hdr hdr; | |
571 | ||
572 | if (count < sizeof hdr) | |
573 | return -EINVAL; | |
574 | ||
575 | if (copy_from_user(&hdr, buf, sizeof hdr)) | |
576 | return -EFAULT; | |
577 | ||
578 | if (hdr.in_words * 4 != count) | |
579 | return -EINVAL; | |
580 | ||
caf6e3f2 | 581 | if (hdr.command >= ARRAY_SIZE(uverbs_cmd_table) || |
b1b8afb8 | 582 | !uverbs_cmd_table[hdr.command]) |
bc38a6ab RD |
583 | return -EINVAL; |
584 | ||
6b73597e | 585 | if (!file->ucontext && |
bc38a6ab RD |
586 | hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT) |
587 | return -EINVAL; | |
588 | ||
b1b8afb8 JM |
589 | if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command))) |
590 | return -ENOSYS; | |
591 | ||
bc38a6ab RD |
592 | return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr, |
593 | hdr.in_words * 4, hdr.out_words * 4); | |
594 | } | |
595 | ||
596 | static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma) | |
597 | { | |
598 | struct ib_uverbs_file *file = filp->private_data; | |
599 | ||
600 | if (!file->ucontext) | |
601 | return -ENODEV; | |
602 | else | |
603 | return file->device->ib_dev->mmap(file->ucontext, vma); | |
604 | } | |
605 | ||
5b2d281a RD |
606 | /* |
607 | * ib_uverbs_open() does not need the BKL: | |
608 | * | |
2a72f212 | 609 | * - the ib_uverbs_device structures are properly reference counted and |
5b2d281a RD |
610 | * everything else is purely local to the file being created, so |
611 | * races against other open calls are not a problem; | |
612 | * - there is no ioctl method to race against; | |
2a72f212 AC |
613 | * - the open method will either immediately run -ENXIO, or all |
614 | * required initialization will be done. | |
5b2d281a | 615 | */ |
bc38a6ab RD |
616 | static int ib_uverbs_open(struct inode *inode, struct file *filp) |
617 | { | |
70a30e16 | 618 | struct ib_uverbs_device *dev; |
bc38a6ab | 619 | struct ib_uverbs_file *file; |
70a30e16 | 620 | int ret; |
bc38a6ab | 621 | |
2a72f212 | 622 | dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev); |
70a30e16 RD |
623 | if (dev) |
624 | kref_get(&dev->ref); | |
2a72f212 | 625 | else |
70a30e16 RD |
626 | return -ENXIO; |
627 | ||
628 | if (!try_module_get(dev->ib_dev->owner)) { | |
629 | ret = -ENODEV; | |
630 | goto err; | |
631 | } | |
bc38a6ab | 632 | |
6b73597e | 633 | file = kmalloc(sizeof *file, GFP_KERNEL); |
63c47c28 | 634 | if (!file) { |
70a30e16 RD |
635 | ret = -ENOMEM; |
636 | goto err_module; | |
63c47c28 | 637 | } |
bc38a6ab | 638 | |
70a30e16 RD |
639 | file->device = dev; |
640 | file->ucontext = NULL; | |
641 | file->async_file = NULL; | |
bc38a6ab | 642 | kref_init(&file->ref); |
95ed644f | 643 | mutex_init(&file->mutex); |
bc38a6ab | 644 | |
bc38a6ab RD |
645 | filp->private_data = file; |
646 | ||
bc1db9af | 647 | return nonseekable_open(inode, filp); |
70a30e16 RD |
648 | |
649 | err_module: | |
650 | module_put(dev->ib_dev->owner); | |
651 | ||
652 | err: | |
653 | kref_put(&dev->ref, ib_uverbs_release_dev); | |
70a30e16 | 654 | return ret; |
bc38a6ab RD |
655 | } |
656 | ||
657 | static int ib_uverbs_close(struct inode *inode, struct file *filp) | |
658 | { | |
659 | struct ib_uverbs_file *file = filp->private_data; | |
bc38a6ab | 660 | |
70a30e16 RD |
661 | ib_uverbs_cleanup_ucontext(file, file->ucontext); |
662 | ||
663 | if (file->async_file) | |
664 | kref_put(&file->async_file->ref, ib_uverbs_release_event_file); | |
bc38a6ab | 665 | |
bc38a6ab RD |
666 | kref_put(&file->ref, ib_uverbs_release_file); |
667 | ||
668 | return 0; | |
669 | } | |
670 | ||
2b8693c0 | 671 | static const struct file_operations uverbs_fops = { |
9afed76d AC |
672 | .owner = THIS_MODULE, |
673 | .write = ib_uverbs_write, | |
674 | .open = ib_uverbs_open, | |
bc1db9af RD |
675 | .release = ib_uverbs_close, |
676 | .llseek = no_llseek, | |
bc38a6ab RD |
677 | }; |
678 | ||
2b8693c0 | 679 | static const struct file_operations uverbs_mmap_fops = { |
9afed76d AC |
680 | .owner = THIS_MODULE, |
681 | .write = ib_uverbs_write, | |
bc38a6ab | 682 | .mmap = ib_uverbs_mmap, |
9afed76d | 683 | .open = ib_uverbs_open, |
bc1db9af RD |
684 | .release = ib_uverbs_close, |
685 | .llseek = no_llseek, | |
bc38a6ab RD |
686 | }; |
687 | ||
688 | static struct ib_client uverbs_client = { | |
689 | .name = "uverbs", | |
690 | .add = ib_uverbs_add_one, | |
691 | .remove = ib_uverbs_remove_one | |
692 | }; | |
693 | ||
f4e91eb4 TJ |
694 | static ssize_t show_ibdev(struct device *device, struct device_attribute *attr, |
695 | char *buf) | |
bc38a6ab | 696 | { |
f4e91eb4 | 697 | struct ib_uverbs_device *dev = dev_get_drvdata(device); |
70a30e16 RD |
698 | |
699 | if (!dev) | |
700 | return -ENODEV; | |
bc38a6ab RD |
701 | |
702 | return sprintf(buf, "%s\n", dev->ib_dev->name); | |
703 | } | |
f4e91eb4 | 704 | static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); |
bc38a6ab | 705 | |
f4e91eb4 TJ |
706 | static ssize_t show_dev_abi_version(struct device *device, |
707 | struct device_attribute *attr, char *buf) | |
274c0891 | 708 | { |
f4e91eb4 | 709 | struct ib_uverbs_device *dev = dev_get_drvdata(device); |
70a30e16 RD |
710 | |
711 | if (!dev) | |
712 | return -ENODEV; | |
274c0891 RD |
713 | |
714 | return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver); | |
715 | } | |
f4e91eb4 | 716 | static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL); |
274c0891 | 717 | |
0933e2d9 AK |
718 | static CLASS_ATTR_STRING(abi_version, S_IRUGO, |
719 | __stringify(IB_USER_VERBS_ABI_VERSION)); | |
bc38a6ab | 720 | |
6d6a0e71 AC |
721 | static dev_t overflow_maj; |
722 | static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES); | |
723 | ||
724 | /* | |
725 | * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by | |
726 | * requesting a new major number and doubling the number of max devices we | |
727 | * support. It's stupid, but simple. | |
728 | */ | |
729 | static int find_overflow_devnum(void) | |
730 | { | |
731 | int ret; | |
732 | ||
733 | if (!overflow_maj) { | |
734 | ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES, | |
735 | "infiniband_verbs"); | |
736 | if (ret) { | |
737 | printk(KERN_ERR "user_verbs: couldn't register dynamic device number\n"); | |
738 | return ret; | |
739 | } | |
740 | } | |
741 | ||
742 | ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES); | |
743 | if (ret >= IB_UVERBS_MAX_DEVICES) | |
744 | return -1; | |
745 | ||
746 | return ret; | |
747 | } | |
748 | ||
bc38a6ab RD |
749 | static void ib_uverbs_add_one(struct ib_device *device) |
750 | { | |
38707980 | 751 | int devnum; |
ddbd6883 | 752 | dev_t base; |
bc38a6ab RD |
753 | struct ib_uverbs_device *uverbs_dev; |
754 | ||
755 | if (!device->alloc_ucontext) | |
756 | return; | |
757 | ||
de6eb66b | 758 | uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL); |
bc38a6ab RD |
759 | if (!uverbs_dev) |
760 | return; | |
761 | ||
70a30e16 | 762 | kref_init(&uverbs_dev->ref); |
fd60ae40 | 763 | init_completion(&uverbs_dev->comp); |
53d0bd1e SH |
764 | uverbs_dev->xrcd_tree = RB_ROOT; |
765 | mutex_init(&uverbs_dev->xrcd_tree_mutex); | |
70a30e16 | 766 | |
bc38a6ab | 767 | spin_lock(&map_lock); |
38707980 AC |
768 | devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES); |
769 | if (devnum >= IB_UVERBS_MAX_DEVICES) { | |
bc38a6ab | 770 | spin_unlock(&map_lock); |
6d6a0e71 AC |
771 | devnum = find_overflow_devnum(); |
772 | if (devnum < 0) | |
773 | goto err; | |
774 | ||
775 | spin_lock(&map_lock); | |
776 | uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES; | |
777 | base = devnum + overflow_maj; | |
778 | set_bit(devnum, overflow_map); | |
779 | } else { | |
780 | uverbs_dev->devnum = devnum; | |
781 | base = devnum + IB_UVERBS_BASE_DEV; | |
782 | set_bit(devnum, dev_map); | |
bc38a6ab | 783 | } |
bc38a6ab RD |
784 | spin_unlock(&map_lock); |
785 | ||
6b73597e | 786 | uverbs_dev->ib_dev = device; |
f4fd0b22 | 787 | uverbs_dev->num_comp_vectors = device->num_comp_vectors; |
bc38a6ab | 788 | |
055422dd AC |
789 | cdev_init(&uverbs_dev->cdev, NULL); |
790 | uverbs_dev->cdev.owner = THIS_MODULE; | |
791 | uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops; | |
792 | kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum); | |
ddbd6883 | 793 | if (cdev_add(&uverbs_dev->cdev, base, 1)) |
70a30e16 | 794 | goto err_cdev; |
bc38a6ab | 795 | |
91bd418f | 796 | uverbs_dev->dev = device_create(uverbs_class, device->dma_device, |
055422dd | 797 | uverbs_dev->cdev.dev, uverbs_dev, |
91bd418f | 798 | "uverbs%d", uverbs_dev->devnum); |
f4e91eb4 | 799 | if (IS_ERR(uverbs_dev->dev)) |
bc38a6ab RD |
800 | goto err_cdev; |
801 | ||
f4e91eb4 | 802 | if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev)) |
bc38a6ab | 803 | goto err_class; |
f4e91eb4 | 804 | if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version)) |
274c0891 | 805 | goto err_class; |
bc38a6ab RD |
806 | |
807 | ib_set_client_data(device, &uverbs_client, uverbs_dev); | |
808 | ||
809 | return; | |
810 | ||
811 | err_class: | |
055422dd | 812 | device_destroy(uverbs_class, uverbs_dev->cdev.dev); |
bc38a6ab RD |
813 | |
814 | err_cdev: | |
055422dd | 815 | cdev_del(&uverbs_dev->cdev); |
6d6a0e71 AC |
816 | if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES) |
817 | clear_bit(devnum, dev_map); | |
818 | else | |
819 | clear_bit(devnum, overflow_map); | |
bc38a6ab RD |
820 | |
821 | err: | |
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 | return; |
826 | } | |
827 | ||
828 | static void ib_uverbs_remove_one(struct ib_device *device) | |
829 | { | |
830 | struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client); | |
831 | ||
832 | if (!uverbs_dev) | |
833 | return; | |
834 | ||
f4e91eb4 | 835 | dev_set_drvdata(uverbs_dev->dev, NULL); |
055422dd AC |
836 | device_destroy(uverbs_class, uverbs_dev->cdev.dev); |
837 | cdev_del(&uverbs_dev->cdev); | |
70a30e16 | 838 | |
6d6a0e71 AC |
839 | if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES) |
840 | clear_bit(uverbs_dev->devnum, dev_map); | |
841 | else | |
842 | clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map); | |
fd60ae40 | 843 | |
70a30e16 | 844 | kref_put(&uverbs_dev->ref, ib_uverbs_release_dev); |
fd60ae40 JM |
845 | wait_for_completion(&uverbs_dev->comp); |
846 | kfree(uverbs_dev); | |
bc38a6ab RD |
847 | } |
848 | ||
2c9ede55 | 849 | static char *uverbs_devnode(struct device *dev, umode_t *mode) |
71c29bd5 | 850 | { |
b2bc4782 GR |
851 | if (mode) |
852 | *mode = 0666; | |
71c29bd5 RD |
853 | return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev)); |
854 | } | |
855 | ||
bc38a6ab RD |
856 | static int __init ib_uverbs_init(void) |
857 | { | |
858 | int ret; | |
859 | ||
bc38a6ab RD |
860 | ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES, |
861 | "infiniband_verbs"); | |
862 | if (ret) { | |
863 | printk(KERN_ERR "user_verbs: couldn't register device number\n"); | |
864 | goto out; | |
865 | } | |
866 | ||
70a30e16 RD |
867 | uverbs_class = class_create(THIS_MODULE, "infiniband_verbs"); |
868 | if (IS_ERR(uverbs_class)) { | |
869 | ret = PTR_ERR(uverbs_class); | |
bc38a6ab RD |
870 | printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n"); |
871 | goto out_chrdev; | |
872 | } | |
873 | ||
71c29bd5 RD |
874 | uverbs_class->devnode = uverbs_devnode; |
875 | ||
0933e2d9 | 876 | ret = class_create_file(uverbs_class, &class_attr_abi_version.attr); |
bc38a6ab RD |
877 | if (ret) { |
878 | printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n"); | |
879 | goto out_class; | |
880 | } | |
881 | ||
bc38a6ab RD |
882 | ret = ib_register_client(&uverbs_client); |
883 | if (ret) { | |
884 | printk(KERN_ERR "user_verbs: couldn't register client\n"); | |
a265e558 | 885 | goto out_class; |
bc38a6ab RD |
886 | } |
887 | ||
888 | return 0; | |
889 | ||
bc38a6ab | 890 | out_class: |
70a30e16 | 891 | class_destroy(uverbs_class); |
bc38a6ab RD |
892 | |
893 | out_chrdev: | |
894 | unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES); | |
895 | ||
896 | out: | |
897 | return ret; | |
898 | } | |
899 | ||
900 | static void __exit ib_uverbs_cleanup(void) | |
901 | { | |
902 | ib_unregister_client(&uverbs_client); | |
70a30e16 | 903 | class_destroy(uverbs_class); |
bc38a6ab | 904 | unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES); |
6d6a0e71 AC |
905 | if (overflow_maj) |
906 | unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES); | |
5d7edb3c RD |
907 | idr_destroy(&ib_uverbs_pd_idr); |
908 | idr_destroy(&ib_uverbs_mr_idr); | |
909 | idr_destroy(&ib_uverbs_mw_idr); | |
910 | idr_destroy(&ib_uverbs_ah_idr); | |
911 | idr_destroy(&ib_uverbs_cq_idr); | |
912 | idr_destroy(&ib_uverbs_qp_idr); | |
913 | idr_destroy(&ib_uverbs_srq_idr); | |
bc38a6ab RD |
914 | } |
915 | ||
916 | module_init(ib_uverbs_init); | |
917 | module_exit(ib_uverbs_cleanup); |