]>
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. | |
5 | * | |
6 | * Author: | |
7 | * Gregory Haskins <ghaskins@novell.com> | |
8 | * | |
9 | * This file is free software; you can redistribute it and/or modify | |
10 | * it under the terms of version 2 of the GNU General Public License | |
11 | * as published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software Foundation, | |
20 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. | |
21 | */ | |
22 | ||
23 | #include <linux/kvm_host.h> | |
d34e6b17 | 24 | #include <linux/kvm.h> |
721eecbf GH |
25 | #include <linux/workqueue.h> |
26 | #include <linux/syscalls.h> | |
27 | #include <linux/wait.h> | |
28 | #include <linux/poll.h> | |
29 | #include <linux/file.h> | |
30 | #include <linux/list.h> | |
31 | #include <linux/eventfd.h> | |
d34e6b17 GH |
32 | #include <linux/kernel.h> |
33 | ||
34 | #include "iodev.h" | |
721eecbf GH |
35 | |
36 | /* | |
37 | * -------------------------------------------------------------------- | |
38 | * irqfd: Allows an fd to be used to inject an interrupt to the guest | |
39 | * | |
40 | * Credit goes to Avi Kivity for the original idea. | |
41 | * -------------------------------------------------------------------- | |
42 | */ | |
43 | ||
44 | struct _irqfd { | |
45 | struct kvm *kvm; | |
46 | struct eventfd_ctx *eventfd; | |
47 | int gsi; | |
48 | struct list_head list; | |
49 | poll_table pt; | |
721eecbf GH |
50 | wait_queue_t wait; |
51 | struct work_struct inject; | |
52 | struct work_struct shutdown; | |
53 | }; | |
54 | ||
55 | static struct workqueue_struct *irqfd_cleanup_wq; | |
56 | ||
57 | static void | |
58 | irqfd_inject(struct work_struct *work) | |
59 | { | |
60 | struct _irqfd *irqfd = container_of(work, struct _irqfd, inject); | |
61 | struct kvm *kvm = irqfd->kvm; | |
62 | ||
721eecbf GH |
63 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1); |
64 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0); | |
721eecbf GH |
65 | } |
66 | ||
67 | /* | |
68 | * Race-free decouple logic (ordering is critical) | |
69 | */ | |
70 | static void | |
71 | irqfd_shutdown(struct work_struct *work) | |
72 | { | |
73 | struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown); | |
b6a114d2 | 74 | u64 cnt; |
721eecbf GH |
75 | |
76 | /* | |
77 | * Synchronize with the wait-queue and unhook ourselves to prevent | |
78 | * further events. | |
79 | */ | |
b6a114d2 | 80 | eventfd_ctx_remove_wait_queue(irqfd->eventfd, &irqfd->wait, &cnt); |
721eecbf GH |
81 | |
82 | /* | |
83 | * We know no new events will be scheduled at this point, so block | |
84 | * until all previously outstanding events have completed | |
85 | */ | |
86 | flush_work(&irqfd->inject); | |
87 | ||
88 | /* | |
89 | * It is now safe to release the object's resources | |
90 | */ | |
91 | eventfd_ctx_put(irqfd->eventfd); | |
92 | kfree(irqfd); | |
93 | } | |
94 | ||
95 | ||
96 | /* assumes kvm->irqfds.lock is held */ | |
97 | static bool | |
98 | irqfd_is_active(struct _irqfd *irqfd) | |
99 | { | |
100 | return list_empty(&irqfd->list) ? false : true; | |
101 | } | |
102 | ||
103 | /* | |
104 | * Mark the irqfd as inactive and schedule it for removal | |
105 | * | |
106 | * assumes kvm->irqfds.lock is held | |
107 | */ | |
108 | static void | |
109 | irqfd_deactivate(struct _irqfd *irqfd) | |
110 | { | |
111 | BUG_ON(!irqfd_is_active(irqfd)); | |
112 | ||
113 | list_del_init(&irqfd->list); | |
114 | ||
115 | queue_work(irqfd_cleanup_wq, &irqfd->shutdown); | |
116 | } | |
117 | ||
118 | /* | |
119 | * Called with wqh->lock held and interrupts disabled | |
120 | */ | |
121 | static int | |
122 | irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
123 | { | |
124 | struct _irqfd *irqfd = container_of(wait, struct _irqfd, wait); | |
125 | unsigned long flags = (unsigned long)key; | |
126 | ||
127 | if (flags & POLLIN) | |
128 | /* An event has been signaled, inject an interrupt */ | |
129 | schedule_work(&irqfd->inject); | |
130 | ||
131 | if (flags & POLLHUP) { | |
132 | /* The eventfd is closing, detach from KVM */ | |
133 | struct kvm *kvm = irqfd->kvm; | |
134 | unsigned long flags; | |
135 | ||
136 | spin_lock_irqsave(&kvm->irqfds.lock, flags); | |
137 | ||
138 | /* | |
139 | * We must check if someone deactivated the irqfd before | |
140 | * we could acquire the irqfds.lock since the item is | |
141 | * deactivated from the KVM side before it is unhooked from | |
142 | * the wait-queue. If it is already deactivated, we can | |
143 | * simply return knowing the other side will cleanup for us. | |
144 | * We cannot race against the irqfd going away since the | |
145 | * other side is required to acquire wqh->lock, which we hold | |
146 | */ | |
147 | if (irqfd_is_active(irqfd)) | |
148 | irqfd_deactivate(irqfd); | |
149 | ||
150 | spin_unlock_irqrestore(&kvm->irqfds.lock, flags); | |
151 | } | |
152 | ||
153 | return 0; | |
154 | } | |
155 | ||
156 | static void | |
157 | irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh, | |
158 | poll_table *pt) | |
159 | { | |
160 | struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt); | |
721eecbf GH |
161 | add_wait_queue(wqh, &irqfd->wait); |
162 | } | |
163 | ||
164 | static int | |
165 | kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi) | |
166 | { | |
f1d1c309 | 167 | struct _irqfd *irqfd, *tmp; |
721eecbf GH |
168 | struct file *file = NULL; |
169 | struct eventfd_ctx *eventfd = NULL; | |
170 | int ret; | |
171 | unsigned int events; | |
172 | ||
173 | irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL); | |
174 | if (!irqfd) | |
175 | return -ENOMEM; | |
176 | ||
177 | irqfd->kvm = kvm; | |
178 | irqfd->gsi = gsi; | |
179 | INIT_LIST_HEAD(&irqfd->list); | |
180 | INIT_WORK(&irqfd->inject, irqfd_inject); | |
181 | INIT_WORK(&irqfd->shutdown, irqfd_shutdown); | |
182 | ||
183 | file = eventfd_fget(fd); | |
184 | if (IS_ERR(file)) { | |
185 | ret = PTR_ERR(file); | |
186 | goto fail; | |
187 | } | |
188 | ||
189 | eventfd = eventfd_ctx_fileget(file); | |
190 | if (IS_ERR(eventfd)) { | |
191 | ret = PTR_ERR(eventfd); | |
192 | goto fail; | |
193 | } | |
194 | ||
195 | irqfd->eventfd = eventfd; | |
196 | ||
197 | /* | |
198 | * Install our own custom wake-up handling so we are notified via | |
199 | * a callback whenever someone signals the underlying eventfd | |
200 | */ | |
201 | init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); | |
202 | init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); | |
203 | ||
f1d1c309 MT |
204 | spin_lock_irq(&kvm->irqfds.lock); |
205 | ||
206 | ret = 0; | |
207 | list_for_each_entry(tmp, &kvm->irqfds.items, list) { | |
208 | if (irqfd->eventfd != tmp->eventfd) | |
209 | continue; | |
210 | /* This fd is used for another irq already. */ | |
211 | ret = -EBUSY; | |
212 | spin_unlock_irq(&kvm->irqfds.lock); | |
213 | goto fail; | |
214 | } | |
215 | ||
721eecbf GH |
216 | events = file->f_op->poll(file, &irqfd->pt); |
217 | ||
721eecbf GH |
218 | list_add_tail(&irqfd->list, &kvm->irqfds.items); |
219 | spin_unlock_irq(&kvm->irqfds.lock); | |
220 | ||
221 | /* | |
222 | * Check if there was an event already pending on the eventfd | |
223 | * before we registered, and trigger it as if we didn't miss it. | |
224 | */ | |
225 | if (events & POLLIN) | |
226 | schedule_work(&irqfd->inject); | |
227 | ||
228 | /* | |
229 | * do not drop the file until the irqfd is fully initialized, otherwise | |
230 | * we might race against the POLLHUP | |
231 | */ | |
232 | fput(file); | |
233 | ||
234 | return 0; | |
235 | ||
236 | fail: | |
237 | if (eventfd && !IS_ERR(eventfd)) | |
238 | eventfd_ctx_put(eventfd); | |
239 | ||
6223011f | 240 | if (!IS_ERR(file)) |
721eecbf GH |
241 | fput(file); |
242 | ||
243 | kfree(irqfd); | |
244 | return ret; | |
245 | } | |
246 | ||
247 | void | |
d34e6b17 | 248 | kvm_eventfd_init(struct kvm *kvm) |
721eecbf GH |
249 | { |
250 | spin_lock_init(&kvm->irqfds.lock); | |
251 | INIT_LIST_HEAD(&kvm->irqfds.items); | |
d34e6b17 | 252 | INIT_LIST_HEAD(&kvm->ioeventfds); |
721eecbf GH |
253 | } |
254 | ||
255 | /* | |
256 | * shutdown any irqfd's that match fd+gsi | |
257 | */ | |
258 | static int | |
259 | kvm_irqfd_deassign(struct kvm *kvm, int fd, int gsi) | |
260 | { | |
261 | struct _irqfd *irqfd, *tmp; | |
262 | struct eventfd_ctx *eventfd; | |
263 | ||
264 | eventfd = eventfd_ctx_fdget(fd); | |
265 | if (IS_ERR(eventfd)) | |
266 | return PTR_ERR(eventfd); | |
267 | ||
268 | spin_lock_irq(&kvm->irqfds.lock); | |
269 | ||
270 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) { | |
271 | if (irqfd->eventfd == eventfd && irqfd->gsi == gsi) | |
272 | irqfd_deactivate(irqfd); | |
273 | } | |
274 | ||
275 | spin_unlock_irq(&kvm->irqfds.lock); | |
276 | eventfd_ctx_put(eventfd); | |
277 | ||
278 | /* | |
279 | * Block until we know all outstanding shutdown jobs have completed | |
280 | * so that we guarantee there will not be any more interrupts on this | |
281 | * gsi once this deassign function returns. | |
282 | */ | |
283 | flush_workqueue(irqfd_cleanup_wq); | |
284 | ||
285 | return 0; | |
286 | } | |
287 | ||
288 | int | |
289 | kvm_irqfd(struct kvm *kvm, int fd, int gsi, int flags) | |
290 | { | |
291 | if (flags & KVM_IRQFD_FLAG_DEASSIGN) | |
292 | return kvm_irqfd_deassign(kvm, fd, gsi); | |
293 | ||
294 | return kvm_irqfd_assign(kvm, fd, gsi); | |
295 | } | |
296 | ||
297 | /* | |
298 | * This function is called as the kvm VM fd is being released. Shutdown all | |
299 | * irqfds that still remain open | |
300 | */ | |
301 | void | |
302 | kvm_irqfd_release(struct kvm *kvm) | |
303 | { | |
304 | struct _irqfd *irqfd, *tmp; | |
305 | ||
306 | spin_lock_irq(&kvm->irqfds.lock); | |
307 | ||
308 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) | |
309 | irqfd_deactivate(irqfd); | |
310 | ||
311 | spin_unlock_irq(&kvm->irqfds.lock); | |
312 | ||
313 | /* | |
314 | * Block until we know all outstanding shutdown jobs have completed | |
315 | * since we do not take a kvm* reference. | |
316 | */ | |
317 | flush_workqueue(irqfd_cleanup_wq); | |
318 | ||
319 | } | |
320 | ||
321 | /* | |
322 | * create a host-wide workqueue for issuing deferred shutdown requests | |
323 | * aggregated from all vm* instances. We need our own isolated single-thread | |
324 | * queue to prevent deadlock against flushing the normal work-queue. | |
325 | */ | |
326 | static int __init irqfd_module_init(void) | |
327 | { | |
328 | irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup"); | |
329 | if (!irqfd_cleanup_wq) | |
330 | return -ENOMEM; | |
331 | ||
332 | return 0; | |
333 | } | |
334 | ||
335 | static void __exit irqfd_module_exit(void) | |
336 | { | |
337 | destroy_workqueue(irqfd_cleanup_wq); | |
338 | } | |
339 | ||
340 | module_init(irqfd_module_init); | |
341 | module_exit(irqfd_module_exit); | |
d34e6b17 GH |
342 | |
343 | /* | |
344 | * -------------------------------------------------------------------- | |
345 | * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal. | |
346 | * | |
347 | * userspace can register a PIO/MMIO address with an eventfd for receiving | |
348 | * notification when the memory has been touched. | |
349 | * -------------------------------------------------------------------- | |
350 | */ | |
351 | ||
352 | struct _ioeventfd { | |
353 | struct list_head list; | |
354 | u64 addr; | |
355 | int length; | |
356 | struct eventfd_ctx *eventfd; | |
357 | u64 datamatch; | |
358 | struct kvm_io_device dev; | |
359 | bool wildcard; | |
360 | }; | |
361 | ||
362 | static inline struct _ioeventfd * | |
363 | to_ioeventfd(struct kvm_io_device *dev) | |
364 | { | |
365 | return container_of(dev, struct _ioeventfd, dev); | |
366 | } | |
367 | ||
368 | static void | |
369 | ioeventfd_release(struct _ioeventfd *p) | |
370 | { | |
371 | eventfd_ctx_put(p->eventfd); | |
372 | list_del(&p->list); | |
373 | kfree(p); | |
374 | } | |
375 | ||
376 | static bool | |
377 | ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val) | |
378 | { | |
379 | u64 _val; | |
380 | ||
381 | if (!(addr == p->addr && len == p->length)) | |
382 | /* address-range must be precise for a hit */ | |
383 | return false; | |
384 | ||
385 | if (p->wildcard) | |
386 | /* all else equal, wildcard is always a hit */ | |
387 | return true; | |
388 | ||
389 | /* otherwise, we have to actually compare the data */ | |
390 | ||
391 | BUG_ON(!IS_ALIGNED((unsigned long)val, len)); | |
392 | ||
393 | switch (len) { | |
394 | case 1: | |
395 | _val = *(u8 *)val; | |
396 | break; | |
397 | case 2: | |
398 | _val = *(u16 *)val; | |
399 | break; | |
400 | case 4: | |
401 | _val = *(u32 *)val; | |
402 | break; | |
403 | case 8: | |
404 | _val = *(u64 *)val; | |
405 | break; | |
406 | default: | |
407 | return false; | |
408 | } | |
409 | ||
410 | return _val == p->datamatch ? true : false; | |
411 | } | |
412 | ||
413 | /* MMIO/PIO writes trigger an event if the addr/val match */ | |
414 | static int | |
415 | ioeventfd_write(struct kvm_io_device *this, gpa_t addr, int len, | |
416 | const void *val) | |
417 | { | |
418 | struct _ioeventfd *p = to_ioeventfd(this); | |
419 | ||
420 | if (!ioeventfd_in_range(p, addr, len, val)) | |
421 | return -EOPNOTSUPP; | |
422 | ||
423 | eventfd_signal(p->eventfd, 1); | |
424 | return 0; | |
425 | } | |
426 | ||
427 | /* | |
428 | * This function is called as KVM is completely shutting down. We do not | |
429 | * need to worry about locking just nuke anything we have as quickly as possible | |
430 | */ | |
431 | static void | |
432 | ioeventfd_destructor(struct kvm_io_device *this) | |
433 | { | |
434 | struct _ioeventfd *p = to_ioeventfd(this); | |
435 | ||
436 | ioeventfd_release(p); | |
437 | } | |
438 | ||
439 | static const struct kvm_io_device_ops ioeventfd_ops = { | |
440 | .write = ioeventfd_write, | |
441 | .destructor = ioeventfd_destructor, | |
442 | }; | |
443 | ||
444 | /* assumes kvm->slots_lock held */ | |
445 | static bool | |
446 | ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p) | |
447 | { | |
448 | struct _ioeventfd *_p; | |
449 | ||
450 | list_for_each_entry(_p, &kvm->ioeventfds, list) | |
451 | if (_p->addr == p->addr && _p->length == p->length && | |
452 | (_p->wildcard || p->wildcard || | |
453 | _p->datamatch == p->datamatch)) | |
454 | return true; | |
455 | ||
456 | return false; | |
457 | } | |
458 | ||
459 | static int | |
460 | kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
461 | { | |
462 | int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; | |
e93f8a0f | 463 | enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS; |
d34e6b17 GH |
464 | struct _ioeventfd *p; |
465 | struct eventfd_ctx *eventfd; | |
466 | int ret; | |
467 | ||
468 | /* must be natural-word sized */ | |
469 | switch (args->len) { | |
470 | case 1: | |
471 | case 2: | |
472 | case 4: | |
473 | case 8: | |
474 | break; | |
475 | default: | |
476 | return -EINVAL; | |
477 | } | |
478 | ||
479 | /* check for range overflow */ | |
480 | if (args->addr + args->len < args->addr) | |
481 | return -EINVAL; | |
482 | ||
483 | /* check for extra flags that we don't understand */ | |
484 | if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK) | |
485 | return -EINVAL; | |
486 | ||
487 | eventfd = eventfd_ctx_fdget(args->fd); | |
488 | if (IS_ERR(eventfd)) | |
489 | return PTR_ERR(eventfd); | |
490 | ||
491 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
492 | if (!p) { | |
493 | ret = -ENOMEM; | |
494 | goto fail; | |
495 | } | |
496 | ||
497 | INIT_LIST_HEAD(&p->list); | |
498 | p->addr = args->addr; | |
499 | p->length = args->len; | |
500 | p->eventfd = eventfd; | |
501 | ||
502 | /* The datamatch feature is optional, otherwise this is a wildcard */ | |
503 | if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH) | |
504 | p->datamatch = args->datamatch; | |
505 | else | |
506 | p->wildcard = true; | |
507 | ||
79fac95e | 508 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
509 | |
510 | /* Verify that there isnt a match already */ | |
511 | if (ioeventfd_check_collision(kvm, p)) { | |
512 | ret = -EEXIST; | |
513 | goto unlock_fail; | |
514 | } | |
515 | ||
516 | kvm_iodevice_init(&p->dev, &ioeventfd_ops); | |
517 | ||
e93f8a0f | 518 | ret = kvm_io_bus_register_dev(kvm, bus_idx, &p->dev); |
d34e6b17 GH |
519 | if (ret < 0) |
520 | goto unlock_fail; | |
521 | ||
522 | list_add_tail(&p->list, &kvm->ioeventfds); | |
523 | ||
79fac95e | 524 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
525 | |
526 | return 0; | |
527 | ||
528 | unlock_fail: | |
79fac95e | 529 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
530 | |
531 | fail: | |
532 | kfree(p); | |
533 | eventfd_ctx_put(eventfd); | |
534 | ||
535 | return ret; | |
536 | } | |
537 | ||
538 | static int | |
539 | kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
540 | { | |
541 | int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; | |
e93f8a0f | 542 | enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS; |
d34e6b17 GH |
543 | struct _ioeventfd *p, *tmp; |
544 | struct eventfd_ctx *eventfd; | |
545 | int ret = -ENOENT; | |
546 | ||
547 | eventfd = eventfd_ctx_fdget(args->fd); | |
548 | if (IS_ERR(eventfd)) | |
549 | return PTR_ERR(eventfd); | |
550 | ||
79fac95e | 551 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
552 | |
553 | list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) { | |
554 | bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH); | |
555 | ||
556 | if (p->eventfd != eventfd || | |
557 | p->addr != args->addr || | |
558 | p->length != args->len || | |
559 | p->wildcard != wildcard) | |
560 | continue; | |
561 | ||
562 | if (!p->wildcard && p->datamatch != args->datamatch) | |
563 | continue; | |
564 | ||
e93f8a0f | 565 | kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev); |
d34e6b17 GH |
566 | ioeventfd_release(p); |
567 | ret = 0; | |
568 | break; | |
569 | } | |
570 | ||
79fac95e | 571 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
572 | |
573 | eventfd_ctx_put(eventfd); | |
574 | ||
575 | return ret; | |
576 | } | |
577 | ||
578 | int | |
579 | kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
580 | { | |
581 | if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN) | |
582 | return kvm_deassign_ioeventfd(kvm, args); | |
583 | ||
584 | return kvm_assign_ioeventfd(kvm, args); | |
585 | } |