]>
Commit | Line | Data |
---|---|---|
721eecbf GH |
1 | /* |
2 | * kvm eventfd support - use eventfd objects to signal various KVM events | |
3 | * | |
4 | * Copyright 2009 Novell. All Rights Reserved. | |
221d059d | 5 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
721eecbf GH |
6 | * |
7 | * Author: | |
8 | * Gregory Haskins <ghaskins@novell.com> | |
9 | * | |
10 | * This file is free software; you can redistribute it and/or modify | |
11 | * it under the terms of version 2 of the GNU General Public License | |
12 | * as published by the Free Software Foundation. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software Foundation, | |
21 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. | |
22 | */ | |
23 | ||
24 | #include <linux/kvm_host.h> | |
d34e6b17 | 25 | #include <linux/kvm.h> |
721eecbf GH |
26 | #include <linux/workqueue.h> |
27 | #include <linux/syscalls.h> | |
28 | #include <linux/wait.h> | |
29 | #include <linux/poll.h> | |
30 | #include <linux/file.h> | |
31 | #include <linux/list.h> | |
32 | #include <linux/eventfd.h> | |
d34e6b17 | 33 | #include <linux/kernel.h> |
5a0e3ad6 | 34 | #include <linux/slab.h> |
d34e6b17 GH |
35 | |
36 | #include "iodev.h" | |
721eecbf GH |
37 | |
38 | /* | |
39 | * -------------------------------------------------------------------- | |
40 | * irqfd: Allows an fd to be used to inject an interrupt to the guest | |
41 | * | |
42 | * Credit goes to Avi Kivity for the original idea. | |
43 | * -------------------------------------------------------------------- | |
44 | */ | |
45 | ||
7a84428a AW |
46 | /* |
47 | * Resampling irqfds are a special variety of irqfds used to emulate | |
48 | * level triggered interrupts. The interrupt is asserted on eventfd | |
49 | * trigger. On acknowledgement through the irq ack notifier, the | |
50 | * interrupt is de-asserted and userspace is notified through the | |
51 | * resamplefd. All resamplers on the same gsi are de-asserted | |
52 | * together, so we don't need to track the state of each individual | |
53 | * user. We can also therefore share the same irq source ID. | |
54 | */ | |
55 | struct _irqfd_resampler { | |
56 | struct kvm *kvm; | |
57 | /* | |
58 | * List of resampling struct _irqfd objects sharing this gsi. | |
59 | * RCU list modified under kvm->irqfds.resampler_lock | |
60 | */ | |
61 | struct list_head list; | |
62 | struct kvm_irq_ack_notifier notifier; | |
63 | /* | |
64 | * Entry in list of kvm->irqfd.resampler_list. Use for sharing | |
65 | * resamplers among irqfds on the same gsi. | |
66 | * Accessed and modified under kvm->irqfds.resampler_lock | |
67 | */ | |
68 | struct list_head link; | |
69 | }; | |
70 | ||
721eecbf | 71 | struct _irqfd { |
bd2b53b2 MT |
72 | /* Used for MSI fast-path */ |
73 | struct kvm *kvm; | |
74 | wait_queue_t wait; | |
75 | /* Update side is protected by irqfds.lock */ | |
76 | struct kvm_kernel_irq_routing_entry __rcu *irq_entry; | |
77 | /* Used for level IRQ fast-path */ | |
78 | int gsi; | |
79 | struct work_struct inject; | |
7a84428a AW |
80 | /* The resampler used by this irqfd (resampler-only) */ |
81 | struct _irqfd_resampler *resampler; | |
82 | /* Eventfd notified on resample (resampler-only) */ | |
83 | struct eventfd_ctx *resamplefd; | |
84 | /* Entry in list of irqfds for a resampler (resampler-only) */ | |
85 | struct list_head resampler_link; | |
bd2b53b2 MT |
86 | /* Used for setup/shutdown */ |
87 | struct eventfd_ctx *eventfd; | |
88 | struct list_head list; | |
89 | poll_table pt; | |
90 | struct work_struct shutdown; | |
721eecbf GH |
91 | }; |
92 | ||
93 | static struct workqueue_struct *irqfd_cleanup_wq; | |
94 | ||
95 | static void | |
96 | irqfd_inject(struct work_struct *work) | |
97 | { | |
98 | struct _irqfd *irqfd = container_of(work, struct _irqfd, inject); | |
99 | struct kvm *kvm = irqfd->kvm; | |
100 | ||
7a84428a AW |
101 | if (!irqfd->resampler) { |
102 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1); | |
103 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0); | |
104 | } else | |
105 | kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
106 | irqfd->gsi, 1); | |
107 | } | |
108 | ||
109 | /* | |
110 | * Since resampler irqfds share an IRQ source ID, we de-assert once | |
111 | * then notify all of the resampler irqfds using this GSI. We can't | |
112 | * do multiple de-asserts or we risk racing with incoming re-asserts. | |
113 | */ | |
114 | static void | |
115 | irqfd_resampler_ack(struct kvm_irq_ack_notifier *kian) | |
116 | { | |
117 | struct _irqfd_resampler *resampler; | |
118 | struct _irqfd *irqfd; | |
119 | ||
120 | resampler = container_of(kian, struct _irqfd_resampler, notifier); | |
121 | ||
122 | kvm_set_irq(resampler->kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
123 | resampler->notifier.gsi, 0); | |
124 | ||
125 | rcu_read_lock(); | |
126 | ||
127 | list_for_each_entry_rcu(irqfd, &resampler->list, resampler_link) | |
128 | eventfd_signal(irqfd->resamplefd, 1); | |
129 | ||
130 | rcu_read_unlock(); | |
131 | } | |
132 | ||
133 | static void | |
134 | irqfd_resampler_shutdown(struct _irqfd *irqfd) | |
135 | { | |
136 | struct _irqfd_resampler *resampler = irqfd->resampler; | |
137 | struct kvm *kvm = resampler->kvm; | |
138 | ||
139 | mutex_lock(&kvm->irqfds.resampler_lock); | |
140 | ||
141 | list_del_rcu(&irqfd->resampler_link); | |
142 | synchronize_rcu(); | |
143 | ||
144 | if (list_empty(&resampler->list)) { | |
145 | list_del(&resampler->link); | |
146 | kvm_unregister_irq_ack_notifier(kvm, &resampler->notifier); | |
147 | kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
148 | resampler->notifier.gsi, 0); | |
149 | kfree(resampler); | |
150 | } | |
151 | ||
152 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
721eecbf GH |
153 | } |
154 | ||
155 | /* | |
156 | * Race-free decouple logic (ordering is critical) | |
157 | */ | |
158 | static void | |
159 | irqfd_shutdown(struct work_struct *work) | |
160 | { | |
161 | struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown); | |
b6a114d2 | 162 | u64 cnt; |
721eecbf GH |
163 | |
164 | /* | |
165 | * Synchronize with the wait-queue and unhook ourselves to prevent | |
166 | * further events. | |
167 | */ | |
b6a114d2 | 168 | eventfd_ctx_remove_wait_queue(irqfd->eventfd, &irqfd->wait, &cnt); |
721eecbf GH |
169 | |
170 | /* | |
171 | * We know no new events will be scheduled at this point, so block | |
172 | * until all previously outstanding events have completed | |
173 | */ | |
43829731 | 174 | flush_work(&irqfd->inject); |
721eecbf | 175 | |
7a84428a AW |
176 | if (irqfd->resampler) { |
177 | irqfd_resampler_shutdown(irqfd); | |
178 | eventfd_ctx_put(irqfd->resamplefd); | |
179 | } | |
180 | ||
721eecbf GH |
181 | /* |
182 | * It is now safe to release the object's resources | |
183 | */ | |
184 | eventfd_ctx_put(irqfd->eventfd); | |
185 | kfree(irqfd); | |
186 | } | |
187 | ||
188 | ||
189 | /* assumes kvm->irqfds.lock is held */ | |
190 | static bool | |
191 | irqfd_is_active(struct _irqfd *irqfd) | |
192 | { | |
193 | return list_empty(&irqfd->list) ? false : true; | |
194 | } | |
195 | ||
196 | /* | |
197 | * Mark the irqfd as inactive and schedule it for removal | |
198 | * | |
199 | * assumes kvm->irqfds.lock is held | |
200 | */ | |
201 | static void | |
202 | irqfd_deactivate(struct _irqfd *irqfd) | |
203 | { | |
204 | BUG_ON(!irqfd_is_active(irqfd)); | |
205 | ||
206 | list_del_init(&irqfd->list); | |
207 | ||
208 | queue_work(irqfd_cleanup_wq, &irqfd->shutdown); | |
209 | } | |
210 | ||
211 | /* | |
212 | * Called with wqh->lock held and interrupts disabled | |
213 | */ | |
214 | static int | |
215 | irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
216 | { | |
217 | struct _irqfd *irqfd = container_of(wait, struct _irqfd, wait); | |
218 | unsigned long flags = (unsigned long)key; | |
bd2b53b2 MT |
219 | struct kvm_kernel_irq_routing_entry *irq; |
220 | struct kvm *kvm = irqfd->kvm; | |
721eecbf | 221 | |
bd2b53b2 MT |
222 | if (flags & POLLIN) { |
223 | rcu_read_lock(); | |
224 | irq = rcu_dereference(irqfd->irq_entry); | |
721eecbf | 225 | /* An event has been signaled, inject an interrupt */ |
bd2b53b2 MT |
226 | if (irq) |
227 | kvm_set_msi(irq, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1); | |
228 | else | |
229 | schedule_work(&irqfd->inject); | |
230 | rcu_read_unlock(); | |
231 | } | |
721eecbf GH |
232 | |
233 | if (flags & POLLHUP) { | |
234 | /* The eventfd is closing, detach from KVM */ | |
721eecbf GH |
235 | unsigned long flags; |
236 | ||
237 | spin_lock_irqsave(&kvm->irqfds.lock, flags); | |
238 | ||
239 | /* | |
240 | * We must check if someone deactivated the irqfd before | |
241 | * we could acquire the irqfds.lock since the item is | |
242 | * deactivated from the KVM side before it is unhooked from | |
243 | * the wait-queue. If it is already deactivated, we can | |
244 | * simply return knowing the other side will cleanup for us. | |
245 | * We cannot race against the irqfd going away since the | |
246 | * other side is required to acquire wqh->lock, which we hold | |
247 | */ | |
248 | if (irqfd_is_active(irqfd)) | |
249 | irqfd_deactivate(irqfd); | |
250 | ||
251 | spin_unlock_irqrestore(&kvm->irqfds.lock, flags); | |
252 | } | |
253 | ||
254 | return 0; | |
255 | } | |
256 | ||
257 | static void | |
258 | irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh, | |
259 | poll_table *pt) | |
260 | { | |
261 | struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt); | |
721eecbf GH |
262 | add_wait_queue(wqh, &irqfd->wait); |
263 | } | |
264 | ||
bd2b53b2 MT |
265 | /* Must be called under irqfds.lock */ |
266 | static void irqfd_update(struct kvm *kvm, struct _irqfd *irqfd, | |
267 | struct kvm_irq_routing_table *irq_rt) | |
268 | { | |
269 | struct kvm_kernel_irq_routing_entry *e; | |
270 | struct hlist_node *n; | |
271 | ||
272 | if (irqfd->gsi >= irq_rt->nr_rt_entries) { | |
273 | rcu_assign_pointer(irqfd->irq_entry, NULL); | |
274 | return; | |
275 | } | |
276 | ||
277 | hlist_for_each_entry(e, n, &irq_rt->map[irqfd->gsi], link) { | |
278 | /* Only fast-path MSI. */ | |
279 | if (e->type == KVM_IRQ_ROUTING_MSI) | |
280 | rcu_assign_pointer(irqfd->irq_entry, e); | |
281 | else | |
282 | rcu_assign_pointer(irqfd->irq_entry, NULL); | |
283 | } | |
284 | } | |
285 | ||
721eecbf | 286 | static int |
d4db2935 | 287 | kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 288 | { |
bd2b53b2 | 289 | struct kvm_irq_routing_table *irq_rt; |
f1d1c309 | 290 | struct _irqfd *irqfd, *tmp; |
721eecbf | 291 | struct file *file = NULL; |
7a84428a | 292 | struct eventfd_ctx *eventfd = NULL, *resamplefd = NULL; |
721eecbf GH |
293 | int ret; |
294 | unsigned int events; | |
295 | ||
296 | irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL); | |
297 | if (!irqfd) | |
298 | return -ENOMEM; | |
299 | ||
300 | irqfd->kvm = kvm; | |
d4db2935 | 301 | irqfd->gsi = args->gsi; |
721eecbf GH |
302 | INIT_LIST_HEAD(&irqfd->list); |
303 | INIT_WORK(&irqfd->inject, irqfd_inject); | |
304 | INIT_WORK(&irqfd->shutdown, irqfd_shutdown); | |
305 | ||
d4db2935 | 306 | file = eventfd_fget(args->fd); |
721eecbf GH |
307 | if (IS_ERR(file)) { |
308 | ret = PTR_ERR(file); | |
309 | goto fail; | |
310 | } | |
311 | ||
312 | eventfd = eventfd_ctx_fileget(file); | |
313 | if (IS_ERR(eventfd)) { | |
314 | ret = PTR_ERR(eventfd); | |
315 | goto fail; | |
316 | } | |
317 | ||
318 | irqfd->eventfd = eventfd; | |
319 | ||
7a84428a AW |
320 | if (args->flags & KVM_IRQFD_FLAG_RESAMPLE) { |
321 | struct _irqfd_resampler *resampler; | |
322 | ||
323 | resamplefd = eventfd_ctx_fdget(args->resamplefd); | |
324 | if (IS_ERR(resamplefd)) { | |
325 | ret = PTR_ERR(resamplefd); | |
326 | goto fail; | |
327 | } | |
328 | ||
329 | irqfd->resamplefd = resamplefd; | |
330 | INIT_LIST_HEAD(&irqfd->resampler_link); | |
331 | ||
332 | mutex_lock(&kvm->irqfds.resampler_lock); | |
333 | ||
334 | list_for_each_entry(resampler, | |
335 | &kvm->irqfds.resampler_list, list) { | |
336 | if (resampler->notifier.gsi == irqfd->gsi) { | |
337 | irqfd->resampler = resampler; | |
338 | break; | |
339 | } | |
340 | } | |
341 | ||
342 | if (!irqfd->resampler) { | |
343 | resampler = kzalloc(sizeof(*resampler), GFP_KERNEL); | |
344 | if (!resampler) { | |
345 | ret = -ENOMEM; | |
346 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
347 | goto fail; | |
348 | } | |
349 | ||
350 | resampler->kvm = kvm; | |
351 | INIT_LIST_HEAD(&resampler->list); | |
352 | resampler->notifier.gsi = irqfd->gsi; | |
353 | resampler->notifier.irq_acked = irqfd_resampler_ack; | |
354 | INIT_LIST_HEAD(&resampler->link); | |
355 | ||
356 | list_add(&resampler->link, &kvm->irqfds.resampler_list); | |
357 | kvm_register_irq_ack_notifier(kvm, | |
358 | &resampler->notifier); | |
359 | irqfd->resampler = resampler; | |
360 | } | |
361 | ||
362 | list_add_rcu(&irqfd->resampler_link, &irqfd->resampler->list); | |
363 | synchronize_rcu(); | |
364 | ||
365 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
366 | } | |
367 | ||
721eecbf GH |
368 | /* |
369 | * Install our own custom wake-up handling so we are notified via | |
370 | * a callback whenever someone signals the underlying eventfd | |
371 | */ | |
372 | init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); | |
373 | init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); | |
374 | ||
f1d1c309 MT |
375 | spin_lock_irq(&kvm->irqfds.lock); |
376 | ||
377 | ret = 0; | |
378 | list_for_each_entry(tmp, &kvm->irqfds.items, list) { | |
379 | if (irqfd->eventfd != tmp->eventfd) | |
380 | continue; | |
381 | /* This fd is used for another irq already. */ | |
382 | ret = -EBUSY; | |
383 | spin_unlock_irq(&kvm->irqfds.lock); | |
384 | goto fail; | |
385 | } | |
386 | ||
bd2b53b2 MT |
387 | irq_rt = rcu_dereference_protected(kvm->irq_routing, |
388 | lockdep_is_held(&kvm->irqfds.lock)); | |
389 | irqfd_update(kvm, irqfd, irq_rt); | |
390 | ||
721eecbf GH |
391 | events = file->f_op->poll(file, &irqfd->pt); |
392 | ||
721eecbf | 393 | list_add_tail(&irqfd->list, &kvm->irqfds.items); |
721eecbf GH |
394 | |
395 | /* | |
396 | * Check if there was an event already pending on the eventfd | |
397 | * before we registered, and trigger it as if we didn't miss it. | |
398 | */ | |
399 | if (events & POLLIN) | |
400 | schedule_work(&irqfd->inject); | |
401 | ||
6bbfb265 MT |
402 | spin_unlock_irq(&kvm->irqfds.lock); |
403 | ||
721eecbf GH |
404 | /* |
405 | * do not drop the file until the irqfd is fully initialized, otherwise | |
406 | * we might race against the POLLHUP | |
407 | */ | |
408 | fput(file); | |
409 | ||
410 | return 0; | |
411 | ||
412 | fail: | |
7a84428a AW |
413 | if (irqfd->resampler) |
414 | irqfd_resampler_shutdown(irqfd); | |
415 | ||
416 | if (resamplefd && !IS_ERR(resamplefd)) | |
417 | eventfd_ctx_put(resamplefd); | |
418 | ||
721eecbf GH |
419 | if (eventfd && !IS_ERR(eventfd)) |
420 | eventfd_ctx_put(eventfd); | |
421 | ||
6223011f | 422 | if (!IS_ERR(file)) |
721eecbf GH |
423 | fput(file); |
424 | ||
425 | kfree(irqfd); | |
426 | return ret; | |
427 | } | |
428 | ||
429 | void | |
d34e6b17 | 430 | kvm_eventfd_init(struct kvm *kvm) |
721eecbf GH |
431 | { |
432 | spin_lock_init(&kvm->irqfds.lock); | |
433 | INIT_LIST_HEAD(&kvm->irqfds.items); | |
7a84428a AW |
434 | INIT_LIST_HEAD(&kvm->irqfds.resampler_list); |
435 | mutex_init(&kvm->irqfds.resampler_lock); | |
d34e6b17 | 436 | INIT_LIST_HEAD(&kvm->ioeventfds); |
721eecbf GH |
437 | } |
438 | ||
439 | /* | |
440 | * shutdown any irqfd's that match fd+gsi | |
441 | */ | |
442 | static int | |
d4db2935 | 443 | kvm_irqfd_deassign(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf GH |
444 | { |
445 | struct _irqfd *irqfd, *tmp; | |
446 | struct eventfd_ctx *eventfd; | |
447 | ||
d4db2935 | 448 | eventfd = eventfd_ctx_fdget(args->fd); |
721eecbf GH |
449 | if (IS_ERR(eventfd)) |
450 | return PTR_ERR(eventfd); | |
451 | ||
452 | spin_lock_irq(&kvm->irqfds.lock); | |
453 | ||
454 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) { | |
d4db2935 | 455 | if (irqfd->eventfd == eventfd && irqfd->gsi == args->gsi) { |
bd2b53b2 MT |
456 | /* |
457 | * This rcu_assign_pointer is needed for when | |
c8ce057e MT |
458 | * another thread calls kvm_irq_routing_update before |
459 | * we flush workqueue below (we synchronize with | |
460 | * kvm_irq_routing_update using irqfds.lock). | |
bd2b53b2 MT |
461 | * It is paired with synchronize_rcu done by caller |
462 | * of that function. | |
463 | */ | |
464 | rcu_assign_pointer(irqfd->irq_entry, NULL); | |
721eecbf | 465 | irqfd_deactivate(irqfd); |
bd2b53b2 | 466 | } |
721eecbf GH |
467 | } |
468 | ||
469 | spin_unlock_irq(&kvm->irqfds.lock); | |
470 | eventfd_ctx_put(eventfd); | |
471 | ||
472 | /* | |
473 | * Block until we know all outstanding shutdown jobs have completed | |
474 | * so that we guarantee there will not be any more interrupts on this | |
475 | * gsi once this deassign function returns. | |
476 | */ | |
477 | flush_workqueue(irqfd_cleanup_wq); | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | int | |
d4db2935 | 483 | kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 484 | { |
7a84428a | 485 | if (args->flags & ~(KVM_IRQFD_FLAG_DEASSIGN | KVM_IRQFD_FLAG_RESAMPLE)) |
326cf033 AW |
486 | return -EINVAL; |
487 | ||
d4db2935 AW |
488 | if (args->flags & KVM_IRQFD_FLAG_DEASSIGN) |
489 | return kvm_irqfd_deassign(kvm, args); | |
721eecbf | 490 | |
d4db2935 | 491 | return kvm_irqfd_assign(kvm, args); |
721eecbf GH |
492 | } |
493 | ||
494 | /* | |
495 | * This function is called as the kvm VM fd is being released. Shutdown all | |
496 | * irqfds that still remain open | |
497 | */ | |
498 | void | |
499 | kvm_irqfd_release(struct kvm *kvm) | |
500 | { | |
501 | struct _irqfd *irqfd, *tmp; | |
502 | ||
503 | spin_lock_irq(&kvm->irqfds.lock); | |
504 | ||
505 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) | |
506 | irqfd_deactivate(irqfd); | |
507 | ||
508 | spin_unlock_irq(&kvm->irqfds.lock); | |
509 | ||
510 | /* | |
511 | * Block until we know all outstanding shutdown jobs have completed | |
512 | * since we do not take a kvm* reference. | |
513 | */ | |
514 | flush_workqueue(irqfd_cleanup_wq); | |
515 | ||
516 | } | |
517 | ||
bd2b53b2 MT |
518 | /* |
519 | * Change irq_routing and irqfd. | |
520 | * Caller must invoke synchronize_rcu afterwards. | |
521 | */ | |
522 | void kvm_irq_routing_update(struct kvm *kvm, | |
523 | struct kvm_irq_routing_table *irq_rt) | |
524 | { | |
525 | struct _irqfd *irqfd; | |
526 | ||
527 | spin_lock_irq(&kvm->irqfds.lock); | |
528 | ||
529 | rcu_assign_pointer(kvm->irq_routing, irq_rt); | |
530 | ||
531 | list_for_each_entry(irqfd, &kvm->irqfds.items, list) | |
532 | irqfd_update(kvm, irqfd, irq_rt); | |
533 | ||
534 | spin_unlock_irq(&kvm->irqfds.lock); | |
535 | } | |
536 | ||
721eecbf GH |
537 | /* |
538 | * create a host-wide workqueue for issuing deferred shutdown requests | |
539 | * aggregated from all vm* instances. We need our own isolated single-thread | |
540 | * queue to prevent deadlock against flushing the normal work-queue. | |
541 | */ | |
542 | static int __init irqfd_module_init(void) | |
543 | { | |
544 | irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup"); | |
545 | if (!irqfd_cleanup_wq) | |
546 | return -ENOMEM; | |
547 | ||
548 | return 0; | |
549 | } | |
550 | ||
551 | static void __exit irqfd_module_exit(void) | |
552 | { | |
553 | destroy_workqueue(irqfd_cleanup_wq); | |
554 | } | |
555 | ||
556 | module_init(irqfd_module_init); | |
557 | module_exit(irqfd_module_exit); | |
d34e6b17 GH |
558 | |
559 | /* | |
560 | * -------------------------------------------------------------------- | |
561 | * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal. | |
562 | * | |
563 | * userspace can register a PIO/MMIO address with an eventfd for receiving | |
564 | * notification when the memory has been touched. | |
565 | * -------------------------------------------------------------------- | |
566 | */ | |
567 | ||
568 | struct _ioeventfd { | |
569 | struct list_head list; | |
570 | u64 addr; | |
571 | int length; | |
572 | struct eventfd_ctx *eventfd; | |
573 | u64 datamatch; | |
574 | struct kvm_io_device dev; | |
575 | bool wildcard; | |
576 | }; | |
577 | ||
578 | static inline struct _ioeventfd * | |
579 | to_ioeventfd(struct kvm_io_device *dev) | |
580 | { | |
581 | return container_of(dev, struct _ioeventfd, dev); | |
582 | } | |
583 | ||
584 | static void | |
585 | ioeventfd_release(struct _ioeventfd *p) | |
586 | { | |
587 | eventfd_ctx_put(p->eventfd); | |
588 | list_del(&p->list); | |
589 | kfree(p); | |
590 | } | |
591 | ||
592 | static bool | |
593 | ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val) | |
594 | { | |
595 | u64 _val; | |
596 | ||
597 | if (!(addr == p->addr && len == p->length)) | |
598 | /* address-range must be precise for a hit */ | |
599 | return false; | |
600 | ||
601 | if (p->wildcard) | |
602 | /* all else equal, wildcard is always a hit */ | |
603 | return true; | |
604 | ||
605 | /* otherwise, we have to actually compare the data */ | |
606 | ||
607 | BUG_ON(!IS_ALIGNED((unsigned long)val, len)); | |
608 | ||
609 | switch (len) { | |
610 | case 1: | |
611 | _val = *(u8 *)val; | |
612 | break; | |
613 | case 2: | |
614 | _val = *(u16 *)val; | |
615 | break; | |
616 | case 4: | |
617 | _val = *(u32 *)val; | |
618 | break; | |
619 | case 8: | |
620 | _val = *(u64 *)val; | |
621 | break; | |
622 | default: | |
623 | return false; | |
624 | } | |
625 | ||
626 | return _val == p->datamatch ? true : false; | |
627 | } | |
628 | ||
629 | /* MMIO/PIO writes trigger an event if the addr/val match */ | |
630 | static int | |
631 | ioeventfd_write(struct kvm_io_device *this, gpa_t addr, int len, | |
632 | const void *val) | |
633 | { | |
634 | struct _ioeventfd *p = to_ioeventfd(this); | |
635 | ||
636 | if (!ioeventfd_in_range(p, addr, len, val)) | |
637 | return -EOPNOTSUPP; | |
638 | ||
639 | eventfd_signal(p->eventfd, 1); | |
640 | return 0; | |
641 | } | |
642 | ||
643 | /* | |
644 | * This function is called as KVM is completely shutting down. We do not | |
645 | * need to worry about locking just nuke anything we have as quickly as possible | |
646 | */ | |
647 | static void | |
648 | ioeventfd_destructor(struct kvm_io_device *this) | |
649 | { | |
650 | struct _ioeventfd *p = to_ioeventfd(this); | |
651 | ||
652 | ioeventfd_release(p); | |
653 | } | |
654 | ||
655 | static const struct kvm_io_device_ops ioeventfd_ops = { | |
656 | .write = ioeventfd_write, | |
657 | .destructor = ioeventfd_destructor, | |
658 | }; | |
659 | ||
660 | /* assumes kvm->slots_lock held */ | |
661 | static bool | |
662 | ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p) | |
663 | { | |
664 | struct _ioeventfd *_p; | |
665 | ||
666 | list_for_each_entry(_p, &kvm->ioeventfds, list) | |
667 | if (_p->addr == p->addr && _p->length == p->length && | |
668 | (_p->wildcard || p->wildcard || | |
669 | _p->datamatch == p->datamatch)) | |
670 | return true; | |
671 | ||
672 | return false; | |
673 | } | |
674 | ||
675 | static int | |
676 | kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
677 | { | |
678 | int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; | |
e93f8a0f | 679 | enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS; |
d34e6b17 GH |
680 | struct _ioeventfd *p; |
681 | struct eventfd_ctx *eventfd; | |
682 | int ret; | |
683 | ||
684 | /* must be natural-word sized */ | |
685 | switch (args->len) { | |
686 | case 1: | |
687 | case 2: | |
688 | case 4: | |
689 | case 8: | |
690 | break; | |
691 | default: | |
692 | return -EINVAL; | |
693 | } | |
694 | ||
695 | /* check for range overflow */ | |
696 | if (args->addr + args->len < args->addr) | |
697 | return -EINVAL; | |
698 | ||
699 | /* check for extra flags that we don't understand */ | |
700 | if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK) | |
701 | return -EINVAL; | |
702 | ||
703 | eventfd = eventfd_ctx_fdget(args->fd); | |
704 | if (IS_ERR(eventfd)) | |
705 | return PTR_ERR(eventfd); | |
706 | ||
707 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
708 | if (!p) { | |
709 | ret = -ENOMEM; | |
710 | goto fail; | |
711 | } | |
712 | ||
713 | INIT_LIST_HEAD(&p->list); | |
714 | p->addr = args->addr; | |
715 | p->length = args->len; | |
716 | p->eventfd = eventfd; | |
717 | ||
718 | /* The datamatch feature is optional, otherwise this is a wildcard */ | |
719 | if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH) | |
720 | p->datamatch = args->datamatch; | |
721 | else | |
722 | p->wildcard = true; | |
723 | ||
79fac95e | 724 | mutex_lock(&kvm->slots_lock); |
d34e6b17 | 725 | |
25985edc | 726 | /* Verify that there isn't a match already */ |
d34e6b17 GH |
727 | if (ioeventfd_check_collision(kvm, p)) { |
728 | ret = -EEXIST; | |
729 | goto unlock_fail; | |
730 | } | |
731 | ||
732 | kvm_iodevice_init(&p->dev, &ioeventfd_ops); | |
733 | ||
743eeb0b SL |
734 | ret = kvm_io_bus_register_dev(kvm, bus_idx, p->addr, p->length, |
735 | &p->dev); | |
d34e6b17 GH |
736 | if (ret < 0) |
737 | goto unlock_fail; | |
738 | ||
739 | list_add_tail(&p->list, &kvm->ioeventfds); | |
740 | ||
79fac95e | 741 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
742 | |
743 | return 0; | |
744 | ||
745 | unlock_fail: | |
79fac95e | 746 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
747 | |
748 | fail: | |
749 | kfree(p); | |
750 | eventfd_ctx_put(eventfd); | |
751 | ||
752 | return ret; | |
753 | } | |
754 | ||
755 | static int | |
756 | kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
757 | { | |
758 | int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; | |
e93f8a0f | 759 | enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS; |
d34e6b17 GH |
760 | struct _ioeventfd *p, *tmp; |
761 | struct eventfd_ctx *eventfd; | |
762 | int ret = -ENOENT; | |
763 | ||
764 | eventfd = eventfd_ctx_fdget(args->fd); | |
765 | if (IS_ERR(eventfd)) | |
766 | return PTR_ERR(eventfd); | |
767 | ||
79fac95e | 768 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
769 | |
770 | list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) { | |
771 | bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH); | |
772 | ||
773 | if (p->eventfd != eventfd || | |
774 | p->addr != args->addr || | |
775 | p->length != args->len || | |
776 | p->wildcard != wildcard) | |
777 | continue; | |
778 | ||
779 | if (!p->wildcard && p->datamatch != args->datamatch) | |
780 | continue; | |
781 | ||
e93f8a0f | 782 | kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev); |
d34e6b17 GH |
783 | ioeventfd_release(p); |
784 | ret = 0; | |
785 | break; | |
786 | } | |
787 | ||
79fac95e | 788 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
789 | |
790 | eventfd_ctx_put(eventfd); | |
791 | ||
792 | return ret; | |
793 | } | |
794 | ||
795 | int | |
796 | kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
797 | { | |
798 | if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN) | |
799 | return kvm_deassign_ioeventfd(kvm, args); | |
800 | ||
801 | return kvm_assign_ioeventfd(kvm, args); | |
802 | } |