]>
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 | ||
46 | struct _irqfd { | |
47 | struct kvm *kvm; | |
48 | struct eventfd_ctx *eventfd; | |
49 | int gsi; | |
50 | struct list_head list; | |
51 | poll_table pt; | |
721eecbf GH |
52 | wait_queue_t wait; |
53 | struct work_struct inject; | |
54 | struct work_struct shutdown; | |
55 | }; | |
56 | ||
57 | static struct workqueue_struct *irqfd_cleanup_wq; | |
58 | ||
59 | static void | |
60 | irqfd_inject(struct work_struct *work) | |
61 | { | |
62 | struct _irqfd *irqfd = container_of(work, struct _irqfd, inject); | |
63 | struct kvm *kvm = irqfd->kvm; | |
64 | ||
721eecbf GH |
65 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1); |
66 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0); | |
721eecbf GH |
67 | } |
68 | ||
69 | /* | |
70 | * Race-free decouple logic (ordering is critical) | |
71 | */ | |
72 | static void | |
73 | irqfd_shutdown(struct work_struct *work) | |
74 | { | |
75 | struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown); | |
b6a114d2 | 76 | u64 cnt; |
721eecbf GH |
77 | |
78 | /* | |
79 | * Synchronize with the wait-queue and unhook ourselves to prevent | |
80 | * further events. | |
81 | */ | |
b6a114d2 | 82 | eventfd_ctx_remove_wait_queue(irqfd->eventfd, &irqfd->wait, &cnt); |
721eecbf GH |
83 | |
84 | /* | |
85 | * We know no new events will be scheduled at this point, so block | |
86 | * until all previously outstanding events have completed | |
87 | */ | |
88 | flush_work(&irqfd->inject); | |
89 | ||
90 | /* | |
91 | * It is now safe to release the object's resources | |
92 | */ | |
93 | eventfd_ctx_put(irqfd->eventfd); | |
94 | kfree(irqfd); | |
95 | } | |
96 | ||
97 | ||
98 | /* assumes kvm->irqfds.lock is held */ | |
99 | static bool | |
100 | irqfd_is_active(struct _irqfd *irqfd) | |
101 | { | |
102 | return list_empty(&irqfd->list) ? false : true; | |
103 | } | |
104 | ||
105 | /* | |
106 | * Mark the irqfd as inactive and schedule it for removal | |
107 | * | |
108 | * assumes kvm->irqfds.lock is held | |
109 | */ | |
110 | static void | |
111 | irqfd_deactivate(struct _irqfd *irqfd) | |
112 | { | |
113 | BUG_ON(!irqfd_is_active(irqfd)); | |
114 | ||
115 | list_del_init(&irqfd->list); | |
116 | ||
117 | queue_work(irqfd_cleanup_wq, &irqfd->shutdown); | |
118 | } | |
119 | ||
120 | /* | |
121 | * Called with wqh->lock held and interrupts disabled | |
122 | */ | |
123 | static int | |
124 | irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
125 | { | |
126 | struct _irqfd *irqfd = container_of(wait, struct _irqfd, wait); | |
127 | unsigned long flags = (unsigned long)key; | |
128 | ||
129 | if (flags & POLLIN) | |
130 | /* An event has been signaled, inject an interrupt */ | |
131 | schedule_work(&irqfd->inject); | |
132 | ||
133 | if (flags & POLLHUP) { | |
134 | /* The eventfd is closing, detach from KVM */ | |
135 | struct kvm *kvm = irqfd->kvm; | |
136 | unsigned long flags; | |
137 | ||
138 | spin_lock_irqsave(&kvm->irqfds.lock, flags); | |
139 | ||
140 | /* | |
141 | * We must check if someone deactivated the irqfd before | |
142 | * we could acquire the irqfds.lock since the item is | |
143 | * deactivated from the KVM side before it is unhooked from | |
144 | * the wait-queue. If it is already deactivated, we can | |
145 | * simply return knowing the other side will cleanup for us. | |
146 | * We cannot race against the irqfd going away since the | |
147 | * other side is required to acquire wqh->lock, which we hold | |
148 | */ | |
149 | if (irqfd_is_active(irqfd)) | |
150 | irqfd_deactivate(irqfd); | |
151 | ||
152 | spin_unlock_irqrestore(&kvm->irqfds.lock, flags); | |
153 | } | |
154 | ||
155 | return 0; | |
156 | } | |
157 | ||
158 | static void | |
159 | irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh, | |
160 | poll_table *pt) | |
161 | { | |
162 | struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt); | |
721eecbf GH |
163 | add_wait_queue(wqh, &irqfd->wait); |
164 | } | |
165 | ||
166 | static int | |
167 | kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi) | |
168 | { | |
f1d1c309 | 169 | struct _irqfd *irqfd, *tmp; |
721eecbf GH |
170 | struct file *file = NULL; |
171 | struct eventfd_ctx *eventfd = NULL; | |
172 | int ret; | |
173 | unsigned int events; | |
174 | ||
175 | irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL); | |
176 | if (!irqfd) | |
177 | return -ENOMEM; | |
178 | ||
179 | irqfd->kvm = kvm; | |
180 | irqfd->gsi = gsi; | |
181 | INIT_LIST_HEAD(&irqfd->list); | |
182 | INIT_WORK(&irqfd->inject, irqfd_inject); | |
183 | INIT_WORK(&irqfd->shutdown, irqfd_shutdown); | |
184 | ||
185 | file = eventfd_fget(fd); | |
186 | if (IS_ERR(file)) { | |
187 | ret = PTR_ERR(file); | |
188 | goto fail; | |
189 | } | |
190 | ||
191 | eventfd = eventfd_ctx_fileget(file); | |
192 | if (IS_ERR(eventfd)) { | |
193 | ret = PTR_ERR(eventfd); | |
194 | goto fail; | |
195 | } | |
196 | ||
197 | irqfd->eventfd = eventfd; | |
198 | ||
199 | /* | |
200 | * Install our own custom wake-up handling so we are notified via | |
201 | * a callback whenever someone signals the underlying eventfd | |
202 | */ | |
203 | init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); | |
204 | init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); | |
205 | ||
f1d1c309 MT |
206 | spin_lock_irq(&kvm->irqfds.lock); |
207 | ||
208 | ret = 0; | |
209 | list_for_each_entry(tmp, &kvm->irqfds.items, list) { | |
210 | if (irqfd->eventfd != tmp->eventfd) | |
211 | continue; | |
212 | /* This fd is used for another irq already. */ | |
213 | ret = -EBUSY; | |
214 | spin_unlock_irq(&kvm->irqfds.lock); | |
215 | goto fail; | |
216 | } | |
217 | ||
721eecbf GH |
218 | events = file->f_op->poll(file, &irqfd->pt); |
219 | ||
721eecbf GH |
220 | list_add_tail(&irqfd->list, &kvm->irqfds.items); |
221 | spin_unlock_irq(&kvm->irqfds.lock); | |
222 | ||
223 | /* | |
224 | * Check if there was an event already pending on the eventfd | |
225 | * before we registered, and trigger it as if we didn't miss it. | |
226 | */ | |
227 | if (events & POLLIN) | |
228 | schedule_work(&irqfd->inject); | |
229 | ||
230 | /* | |
231 | * do not drop the file until the irqfd is fully initialized, otherwise | |
232 | * we might race against the POLLHUP | |
233 | */ | |
234 | fput(file); | |
235 | ||
236 | return 0; | |
237 | ||
238 | fail: | |
239 | if (eventfd && !IS_ERR(eventfd)) | |
240 | eventfd_ctx_put(eventfd); | |
241 | ||
6223011f | 242 | if (!IS_ERR(file)) |
721eecbf GH |
243 | fput(file); |
244 | ||
245 | kfree(irqfd); | |
246 | return ret; | |
247 | } | |
248 | ||
249 | void | |
d34e6b17 | 250 | kvm_eventfd_init(struct kvm *kvm) |
721eecbf GH |
251 | { |
252 | spin_lock_init(&kvm->irqfds.lock); | |
253 | INIT_LIST_HEAD(&kvm->irqfds.items); | |
d34e6b17 | 254 | INIT_LIST_HEAD(&kvm->ioeventfds); |
721eecbf GH |
255 | } |
256 | ||
257 | /* | |
258 | * shutdown any irqfd's that match fd+gsi | |
259 | */ | |
260 | static int | |
261 | kvm_irqfd_deassign(struct kvm *kvm, int fd, int gsi) | |
262 | { | |
263 | struct _irqfd *irqfd, *tmp; | |
264 | struct eventfd_ctx *eventfd; | |
265 | ||
266 | eventfd = eventfd_ctx_fdget(fd); | |
267 | if (IS_ERR(eventfd)) | |
268 | return PTR_ERR(eventfd); | |
269 | ||
270 | spin_lock_irq(&kvm->irqfds.lock); | |
271 | ||
272 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) { | |
273 | if (irqfd->eventfd == eventfd && irqfd->gsi == gsi) | |
274 | irqfd_deactivate(irqfd); | |
275 | } | |
276 | ||
277 | spin_unlock_irq(&kvm->irqfds.lock); | |
278 | eventfd_ctx_put(eventfd); | |
279 | ||
280 | /* | |
281 | * Block until we know all outstanding shutdown jobs have completed | |
282 | * so that we guarantee there will not be any more interrupts on this | |
283 | * gsi once this deassign function returns. | |
284 | */ | |
285 | flush_workqueue(irqfd_cleanup_wq); | |
286 | ||
287 | return 0; | |
288 | } | |
289 | ||
290 | int | |
291 | kvm_irqfd(struct kvm *kvm, int fd, int gsi, int flags) | |
292 | { | |
293 | if (flags & KVM_IRQFD_FLAG_DEASSIGN) | |
294 | return kvm_irqfd_deassign(kvm, fd, gsi); | |
295 | ||
296 | return kvm_irqfd_assign(kvm, fd, gsi); | |
297 | } | |
298 | ||
299 | /* | |
300 | * This function is called as the kvm VM fd is being released. Shutdown all | |
301 | * irqfds that still remain open | |
302 | */ | |
303 | void | |
304 | kvm_irqfd_release(struct kvm *kvm) | |
305 | { | |
306 | struct _irqfd *irqfd, *tmp; | |
307 | ||
308 | spin_lock_irq(&kvm->irqfds.lock); | |
309 | ||
310 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) | |
311 | irqfd_deactivate(irqfd); | |
312 | ||
313 | spin_unlock_irq(&kvm->irqfds.lock); | |
314 | ||
315 | /* | |
316 | * Block until we know all outstanding shutdown jobs have completed | |
317 | * since we do not take a kvm* reference. | |
318 | */ | |
319 | flush_workqueue(irqfd_cleanup_wq); | |
320 | ||
321 | } | |
322 | ||
323 | /* | |
324 | * create a host-wide workqueue for issuing deferred shutdown requests | |
325 | * aggregated from all vm* instances. We need our own isolated single-thread | |
326 | * queue to prevent deadlock against flushing the normal work-queue. | |
327 | */ | |
328 | static int __init irqfd_module_init(void) | |
329 | { | |
330 | irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup"); | |
331 | if (!irqfd_cleanup_wq) | |
332 | return -ENOMEM; | |
333 | ||
334 | return 0; | |
335 | } | |
336 | ||
337 | static void __exit irqfd_module_exit(void) | |
338 | { | |
339 | destroy_workqueue(irqfd_cleanup_wq); | |
340 | } | |
341 | ||
342 | module_init(irqfd_module_init); | |
343 | module_exit(irqfd_module_exit); | |
d34e6b17 GH |
344 | |
345 | /* | |
346 | * -------------------------------------------------------------------- | |
347 | * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal. | |
348 | * | |
349 | * userspace can register a PIO/MMIO address with an eventfd for receiving | |
350 | * notification when the memory has been touched. | |
351 | * -------------------------------------------------------------------- | |
352 | */ | |
353 | ||
354 | struct _ioeventfd { | |
355 | struct list_head list; | |
356 | u64 addr; | |
357 | int length; | |
358 | struct eventfd_ctx *eventfd; | |
359 | u64 datamatch; | |
360 | struct kvm_io_device dev; | |
361 | bool wildcard; | |
362 | }; | |
363 | ||
364 | static inline struct _ioeventfd * | |
365 | to_ioeventfd(struct kvm_io_device *dev) | |
366 | { | |
367 | return container_of(dev, struct _ioeventfd, dev); | |
368 | } | |
369 | ||
370 | static void | |
371 | ioeventfd_release(struct _ioeventfd *p) | |
372 | { | |
373 | eventfd_ctx_put(p->eventfd); | |
374 | list_del(&p->list); | |
375 | kfree(p); | |
376 | } | |
377 | ||
378 | static bool | |
379 | ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val) | |
380 | { | |
381 | u64 _val; | |
382 | ||
383 | if (!(addr == p->addr && len == p->length)) | |
384 | /* address-range must be precise for a hit */ | |
385 | return false; | |
386 | ||
387 | if (p->wildcard) | |
388 | /* all else equal, wildcard is always a hit */ | |
389 | return true; | |
390 | ||
391 | /* otherwise, we have to actually compare the data */ | |
392 | ||
393 | BUG_ON(!IS_ALIGNED((unsigned long)val, len)); | |
394 | ||
395 | switch (len) { | |
396 | case 1: | |
397 | _val = *(u8 *)val; | |
398 | break; | |
399 | case 2: | |
400 | _val = *(u16 *)val; | |
401 | break; | |
402 | case 4: | |
403 | _val = *(u32 *)val; | |
404 | break; | |
405 | case 8: | |
406 | _val = *(u64 *)val; | |
407 | break; | |
408 | default: | |
409 | return false; | |
410 | } | |
411 | ||
412 | return _val == p->datamatch ? true : false; | |
413 | } | |
414 | ||
415 | /* MMIO/PIO writes trigger an event if the addr/val match */ | |
416 | static int | |
417 | ioeventfd_write(struct kvm_io_device *this, gpa_t addr, int len, | |
418 | const void *val) | |
419 | { | |
420 | struct _ioeventfd *p = to_ioeventfd(this); | |
421 | ||
422 | if (!ioeventfd_in_range(p, addr, len, val)) | |
423 | return -EOPNOTSUPP; | |
424 | ||
425 | eventfd_signal(p->eventfd, 1); | |
426 | return 0; | |
427 | } | |
428 | ||
429 | /* | |
430 | * This function is called as KVM is completely shutting down. We do not | |
431 | * need to worry about locking just nuke anything we have as quickly as possible | |
432 | */ | |
433 | static void | |
434 | ioeventfd_destructor(struct kvm_io_device *this) | |
435 | { | |
436 | struct _ioeventfd *p = to_ioeventfd(this); | |
437 | ||
438 | ioeventfd_release(p); | |
439 | } | |
440 | ||
441 | static const struct kvm_io_device_ops ioeventfd_ops = { | |
442 | .write = ioeventfd_write, | |
443 | .destructor = ioeventfd_destructor, | |
444 | }; | |
445 | ||
446 | /* assumes kvm->slots_lock held */ | |
447 | static bool | |
448 | ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p) | |
449 | { | |
450 | struct _ioeventfd *_p; | |
451 | ||
452 | list_for_each_entry(_p, &kvm->ioeventfds, list) | |
453 | if (_p->addr == p->addr && _p->length == p->length && | |
454 | (_p->wildcard || p->wildcard || | |
455 | _p->datamatch == p->datamatch)) | |
456 | return true; | |
457 | ||
458 | return false; | |
459 | } | |
460 | ||
461 | static int | |
462 | kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
463 | { | |
464 | int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; | |
e93f8a0f | 465 | enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS; |
d34e6b17 GH |
466 | struct _ioeventfd *p; |
467 | struct eventfd_ctx *eventfd; | |
468 | int ret; | |
469 | ||
470 | /* must be natural-word sized */ | |
471 | switch (args->len) { | |
472 | case 1: | |
473 | case 2: | |
474 | case 4: | |
475 | case 8: | |
476 | break; | |
477 | default: | |
478 | return -EINVAL; | |
479 | } | |
480 | ||
481 | /* check for range overflow */ | |
482 | if (args->addr + args->len < args->addr) | |
483 | return -EINVAL; | |
484 | ||
485 | /* check for extra flags that we don't understand */ | |
486 | if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK) | |
487 | return -EINVAL; | |
488 | ||
489 | eventfd = eventfd_ctx_fdget(args->fd); | |
490 | if (IS_ERR(eventfd)) | |
491 | return PTR_ERR(eventfd); | |
492 | ||
493 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
494 | if (!p) { | |
495 | ret = -ENOMEM; | |
496 | goto fail; | |
497 | } | |
498 | ||
499 | INIT_LIST_HEAD(&p->list); | |
500 | p->addr = args->addr; | |
501 | p->length = args->len; | |
502 | p->eventfd = eventfd; | |
503 | ||
504 | /* The datamatch feature is optional, otherwise this is a wildcard */ | |
505 | if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH) | |
506 | p->datamatch = args->datamatch; | |
507 | else | |
508 | p->wildcard = true; | |
509 | ||
79fac95e | 510 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
511 | |
512 | /* Verify that there isnt a match already */ | |
513 | if (ioeventfd_check_collision(kvm, p)) { | |
514 | ret = -EEXIST; | |
515 | goto unlock_fail; | |
516 | } | |
517 | ||
518 | kvm_iodevice_init(&p->dev, &ioeventfd_ops); | |
519 | ||
e93f8a0f | 520 | ret = kvm_io_bus_register_dev(kvm, bus_idx, &p->dev); |
d34e6b17 GH |
521 | if (ret < 0) |
522 | goto unlock_fail; | |
523 | ||
524 | list_add_tail(&p->list, &kvm->ioeventfds); | |
525 | ||
79fac95e | 526 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
527 | |
528 | return 0; | |
529 | ||
530 | unlock_fail: | |
79fac95e | 531 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
532 | |
533 | fail: | |
534 | kfree(p); | |
535 | eventfd_ctx_put(eventfd); | |
536 | ||
537 | return ret; | |
538 | } | |
539 | ||
540 | static int | |
541 | kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
542 | { | |
543 | int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO; | |
e93f8a0f | 544 | enum kvm_bus bus_idx = pio ? KVM_PIO_BUS : KVM_MMIO_BUS; |
d34e6b17 GH |
545 | struct _ioeventfd *p, *tmp; |
546 | struct eventfd_ctx *eventfd; | |
547 | int ret = -ENOENT; | |
548 | ||
549 | eventfd = eventfd_ctx_fdget(args->fd); | |
550 | if (IS_ERR(eventfd)) | |
551 | return PTR_ERR(eventfd); | |
552 | ||
79fac95e | 553 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
554 | |
555 | list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) { | |
556 | bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH); | |
557 | ||
558 | if (p->eventfd != eventfd || | |
559 | p->addr != args->addr || | |
560 | p->length != args->len || | |
561 | p->wildcard != wildcard) | |
562 | continue; | |
563 | ||
564 | if (!p->wildcard && p->datamatch != args->datamatch) | |
565 | continue; | |
566 | ||
e93f8a0f | 567 | kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev); |
d34e6b17 GH |
568 | ioeventfd_release(p); |
569 | ret = 0; | |
570 | break; | |
571 | } | |
572 | ||
79fac95e | 573 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
574 | |
575 | eventfd_ctx_put(eventfd); | |
576 | ||
577 | return ret; | |
578 | } | |
579 | ||
580 | int | |
581 | kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
582 | { | |
583 | if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN) | |
584 | return kvm_deassign_ioeventfd(kvm, args); | |
585 | ||
586 | return kvm_assign_ioeventfd(kvm, args); | |
587 | } |