]>
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> |
166c9775 | 26 | #include <linux/kvm_irqfd.h> |
721eecbf GH |
27 | #include <linux/workqueue.h> |
28 | #include <linux/syscalls.h> | |
29 | #include <linux/wait.h> | |
30 | #include <linux/poll.h> | |
31 | #include <linux/file.h> | |
32 | #include <linux/list.h> | |
33 | #include <linux/eventfd.h> | |
d34e6b17 | 34 | #include <linux/kernel.h> |
719d93cd | 35 | #include <linux/srcu.h> |
5a0e3ad6 | 36 | #include <linux/slab.h> |
56f89f36 | 37 | #include <linux/seqlock.h> |
9016cfb5 | 38 | #include <linux/irqbypass.h> |
e4d57e1e | 39 | #include <trace/events/kvm.h> |
d34e6b17 | 40 | |
af669ac6 | 41 | #include <kvm/iodev.h> |
721eecbf | 42 | |
297e2105 | 43 | #ifdef CONFIG_HAVE_KVM_IRQFD |
721eecbf GH |
44 | |
45 | static struct workqueue_struct *irqfd_cleanup_wq; | |
46 | ||
47 | static void | |
48 | irqfd_inject(struct work_struct *work) | |
49 | { | |
166c9775 EA |
50 | struct kvm_kernel_irqfd *irqfd = |
51 | container_of(work, struct kvm_kernel_irqfd, inject); | |
721eecbf GH |
52 | struct kvm *kvm = irqfd->kvm; |
53 | ||
7a84428a | 54 | if (!irqfd->resampler) { |
aa2fbe6d YZ |
55 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1, |
56 | false); | |
57 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0, | |
58 | false); | |
7a84428a AW |
59 | } else |
60 | kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
aa2fbe6d | 61 | irqfd->gsi, 1, false); |
7a84428a AW |
62 | } |
63 | ||
64 | /* | |
65 | * Since resampler irqfds share an IRQ source ID, we de-assert once | |
66 | * then notify all of the resampler irqfds using this GSI. We can't | |
67 | * do multiple de-asserts or we risk racing with incoming re-asserts. | |
68 | */ | |
69 | static void | |
70 | irqfd_resampler_ack(struct kvm_irq_ack_notifier *kian) | |
71 | { | |
166c9775 | 72 | struct kvm_kernel_irqfd_resampler *resampler; |
719d93cd | 73 | struct kvm *kvm; |
166c9775 | 74 | struct kvm_kernel_irqfd *irqfd; |
719d93cd | 75 | int idx; |
7a84428a | 76 | |
166c9775 EA |
77 | resampler = container_of(kian, |
78 | struct kvm_kernel_irqfd_resampler, notifier); | |
719d93cd | 79 | kvm = resampler->kvm; |
7a84428a | 80 | |
719d93cd | 81 | kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, |
aa2fbe6d | 82 | resampler->notifier.gsi, 0, false); |
7a84428a | 83 | |
719d93cd | 84 | idx = srcu_read_lock(&kvm->irq_srcu); |
7a84428a AW |
85 | |
86 | list_for_each_entry_rcu(irqfd, &resampler->list, resampler_link) | |
87 | eventfd_signal(irqfd->resamplefd, 1); | |
88 | ||
719d93cd | 89 | srcu_read_unlock(&kvm->irq_srcu, idx); |
7a84428a AW |
90 | } |
91 | ||
92 | static void | |
166c9775 | 93 | irqfd_resampler_shutdown(struct kvm_kernel_irqfd *irqfd) |
7a84428a | 94 | { |
166c9775 | 95 | struct kvm_kernel_irqfd_resampler *resampler = irqfd->resampler; |
7a84428a AW |
96 | struct kvm *kvm = resampler->kvm; |
97 | ||
98 | mutex_lock(&kvm->irqfds.resampler_lock); | |
99 | ||
100 | list_del_rcu(&irqfd->resampler_link); | |
719d93cd | 101 | synchronize_srcu(&kvm->irq_srcu); |
7a84428a AW |
102 | |
103 | if (list_empty(&resampler->list)) { | |
104 | list_del(&resampler->link); | |
105 | kvm_unregister_irq_ack_notifier(kvm, &resampler->notifier); | |
106 | kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
aa2fbe6d | 107 | resampler->notifier.gsi, 0, false); |
7a84428a AW |
108 | kfree(resampler); |
109 | } | |
110 | ||
111 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
721eecbf GH |
112 | } |
113 | ||
114 | /* | |
115 | * Race-free decouple logic (ordering is critical) | |
116 | */ | |
117 | static void | |
118 | irqfd_shutdown(struct work_struct *work) | |
119 | { | |
166c9775 EA |
120 | struct kvm_kernel_irqfd *irqfd = |
121 | container_of(work, struct kvm_kernel_irqfd, shutdown); | |
b6a114d2 | 122 | u64 cnt; |
721eecbf GH |
123 | |
124 | /* | |
125 | * Synchronize with the wait-queue and unhook ourselves to prevent | |
126 | * further events. | |
127 | */ | |
b6a114d2 | 128 | eventfd_ctx_remove_wait_queue(irqfd->eventfd, &irqfd->wait, &cnt); |
721eecbf GH |
129 | |
130 | /* | |
131 | * We know no new events will be scheduled at this point, so block | |
132 | * until all previously outstanding events have completed | |
133 | */ | |
43829731 | 134 | flush_work(&irqfd->inject); |
721eecbf | 135 | |
7a84428a AW |
136 | if (irqfd->resampler) { |
137 | irqfd_resampler_shutdown(irqfd); | |
138 | eventfd_ctx_put(irqfd->resamplefd); | |
139 | } | |
140 | ||
721eecbf GH |
141 | /* |
142 | * It is now safe to release the object's resources | |
143 | */ | |
9016cfb5 EA |
144 | #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS |
145 | irq_bypass_unregister_consumer(&irqfd->consumer); | |
146 | #endif | |
721eecbf GH |
147 | eventfd_ctx_put(irqfd->eventfd); |
148 | kfree(irqfd); | |
149 | } | |
150 | ||
151 | ||
152 | /* assumes kvm->irqfds.lock is held */ | |
153 | static bool | |
166c9775 | 154 | irqfd_is_active(struct kvm_kernel_irqfd *irqfd) |
721eecbf GH |
155 | { |
156 | return list_empty(&irqfd->list) ? false : true; | |
157 | } | |
158 | ||
159 | /* | |
160 | * Mark the irqfd as inactive and schedule it for removal | |
161 | * | |
162 | * assumes kvm->irqfds.lock is held | |
163 | */ | |
164 | static void | |
166c9775 | 165 | irqfd_deactivate(struct kvm_kernel_irqfd *irqfd) |
721eecbf GH |
166 | { |
167 | BUG_ON(!irqfd_is_active(irqfd)); | |
168 | ||
169 | list_del_init(&irqfd->list); | |
170 | ||
171 | queue_work(irqfd_cleanup_wq, &irqfd->shutdown); | |
172 | } | |
173 | ||
174 | /* | |
175 | * Called with wqh->lock held and interrupts disabled | |
176 | */ | |
177 | static int | |
178 | irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
179 | { | |
166c9775 EA |
180 | struct kvm_kernel_irqfd *irqfd = |
181 | container_of(wait, struct kvm_kernel_irqfd, wait); | |
721eecbf | 182 | unsigned long flags = (unsigned long)key; |
56f89f36 | 183 | struct kvm_kernel_irq_routing_entry irq; |
bd2b53b2 | 184 | struct kvm *kvm = irqfd->kvm; |
56f89f36 | 185 | unsigned seq; |
719d93cd | 186 | int idx; |
721eecbf | 187 | |
bd2b53b2 | 188 | if (flags & POLLIN) { |
719d93cd | 189 | idx = srcu_read_lock(&kvm->irq_srcu); |
56f89f36 PM |
190 | do { |
191 | seq = read_seqcount_begin(&irqfd->irq_entry_sc); | |
192 | irq = irqfd->irq_entry; | |
193 | } while (read_seqcount_retry(&irqfd->irq_entry_sc, seq)); | |
721eecbf | 194 | /* An event has been signaled, inject an interrupt */ |
56f89f36 PM |
195 | if (irq.type == KVM_IRQ_ROUTING_MSI) |
196 | kvm_set_msi(&irq, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1, | |
aa2fbe6d | 197 | false); |
bd2b53b2 MT |
198 | else |
199 | schedule_work(&irqfd->inject); | |
719d93cd | 200 | srcu_read_unlock(&kvm->irq_srcu, idx); |
bd2b53b2 | 201 | } |
721eecbf GH |
202 | |
203 | if (flags & POLLHUP) { | |
204 | /* The eventfd is closing, detach from KVM */ | |
721eecbf GH |
205 | unsigned long flags; |
206 | ||
207 | spin_lock_irqsave(&kvm->irqfds.lock, flags); | |
208 | ||
209 | /* | |
210 | * We must check if someone deactivated the irqfd before | |
211 | * we could acquire the irqfds.lock since the item is | |
212 | * deactivated from the KVM side before it is unhooked from | |
213 | * the wait-queue. If it is already deactivated, we can | |
214 | * simply return knowing the other side will cleanup for us. | |
215 | * We cannot race against the irqfd going away since the | |
216 | * other side is required to acquire wqh->lock, which we hold | |
217 | */ | |
218 | if (irqfd_is_active(irqfd)) | |
219 | irqfd_deactivate(irqfd); | |
220 | ||
221 | spin_unlock_irqrestore(&kvm->irqfds.lock, flags); | |
222 | } | |
223 | ||
224 | return 0; | |
225 | } | |
226 | ||
227 | static void | |
228 | irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh, | |
229 | poll_table *pt) | |
230 | { | |
166c9775 EA |
231 | struct kvm_kernel_irqfd *irqfd = |
232 | container_of(pt, struct kvm_kernel_irqfd, pt); | |
721eecbf GH |
233 | add_wait_queue(wqh, &irqfd->wait); |
234 | } | |
235 | ||
bd2b53b2 | 236 | /* Must be called under irqfds.lock */ |
166c9775 | 237 | static void irqfd_update(struct kvm *kvm, struct kvm_kernel_irqfd *irqfd) |
bd2b53b2 MT |
238 | { |
239 | struct kvm_kernel_irq_routing_entry *e; | |
8ba918d4 PM |
240 | struct kvm_kernel_irq_routing_entry entries[KVM_NR_IRQCHIPS]; |
241 | int i, n_entries; | |
242 | ||
9957c86d | 243 | n_entries = kvm_irq_map_gsi(kvm, entries, irqfd->gsi); |
bd2b53b2 | 244 | |
56f89f36 PM |
245 | write_seqcount_begin(&irqfd->irq_entry_sc); |
246 | ||
247 | irqfd->irq_entry.type = 0; | |
bd2b53b2 | 248 | |
8ba918d4 PM |
249 | e = entries; |
250 | for (i = 0; i < n_entries; ++i, ++e) { | |
bd2b53b2 MT |
251 | /* Only fast-path MSI. */ |
252 | if (e->type == KVM_IRQ_ROUTING_MSI) | |
56f89f36 | 253 | irqfd->irq_entry = *e; |
bd2b53b2 | 254 | } |
56f89f36 | 255 | |
56f89f36 | 256 | write_seqcount_end(&irqfd->irq_entry_sc); |
bd2b53b2 MT |
257 | } |
258 | ||
1a02b270 EA |
259 | #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS |
260 | void __attribute__((weak)) kvm_arch_irq_bypass_stop( | |
261 | struct irq_bypass_consumer *cons) | |
262 | { | |
263 | } | |
264 | ||
265 | void __attribute__((weak)) kvm_arch_irq_bypass_start( | |
266 | struct irq_bypass_consumer *cons) | |
267 | { | |
268 | } | |
269 | #endif | |
270 | ||
721eecbf | 271 | static int |
d4db2935 | 272 | kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 273 | { |
166c9775 | 274 | struct kvm_kernel_irqfd *irqfd, *tmp; |
cffe78d9 | 275 | struct fd f; |
7a84428a | 276 | struct eventfd_ctx *eventfd = NULL, *resamplefd = NULL; |
721eecbf GH |
277 | int ret; |
278 | unsigned int events; | |
9957c86d | 279 | int idx; |
721eecbf | 280 | |
01c94e64 EA |
281 | if (!kvm_arch_intc_initialized(kvm)) |
282 | return -EAGAIN; | |
283 | ||
721eecbf GH |
284 | irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL); |
285 | if (!irqfd) | |
286 | return -ENOMEM; | |
287 | ||
288 | irqfd->kvm = kvm; | |
d4db2935 | 289 | irqfd->gsi = args->gsi; |
721eecbf GH |
290 | INIT_LIST_HEAD(&irqfd->list); |
291 | INIT_WORK(&irqfd->inject, irqfd_inject); | |
292 | INIT_WORK(&irqfd->shutdown, irqfd_shutdown); | |
56f89f36 | 293 | seqcount_init(&irqfd->irq_entry_sc); |
721eecbf | 294 | |
cffe78d9 AV |
295 | f = fdget(args->fd); |
296 | if (!f.file) { | |
297 | ret = -EBADF; | |
298 | goto out; | |
721eecbf GH |
299 | } |
300 | ||
cffe78d9 | 301 | eventfd = eventfd_ctx_fileget(f.file); |
721eecbf GH |
302 | if (IS_ERR(eventfd)) { |
303 | ret = PTR_ERR(eventfd); | |
304 | goto fail; | |
305 | } | |
306 | ||
307 | irqfd->eventfd = eventfd; | |
308 | ||
7a84428a | 309 | if (args->flags & KVM_IRQFD_FLAG_RESAMPLE) { |
166c9775 | 310 | struct kvm_kernel_irqfd_resampler *resampler; |
7a84428a AW |
311 | |
312 | resamplefd = eventfd_ctx_fdget(args->resamplefd); | |
313 | if (IS_ERR(resamplefd)) { | |
314 | ret = PTR_ERR(resamplefd); | |
315 | goto fail; | |
316 | } | |
317 | ||
318 | irqfd->resamplefd = resamplefd; | |
319 | INIT_LIST_HEAD(&irqfd->resampler_link); | |
320 | ||
321 | mutex_lock(&kvm->irqfds.resampler_lock); | |
322 | ||
323 | list_for_each_entry(resampler, | |
49f8a1a5 | 324 | &kvm->irqfds.resampler_list, link) { |
7a84428a AW |
325 | if (resampler->notifier.gsi == irqfd->gsi) { |
326 | irqfd->resampler = resampler; | |
327 | break; | |
328 | } | |
329 | } | |
330 | ||
331 | if (!irqfd->resampler) { | |
332 | resampler = kzalloc(sizeof(*resampler), GFP_KERNEL); | |
333 | if (!resampler) { | |
334 | ret = -ENOMEM; | |
335 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
336 | goto fail; | |
337 | } | |
338 | ||
339 | resampler->kvm = kvm; | |
340 | INIT_LIST_HEAD(&resampler->list); | |
341 | resampler->notifier.gsi = irqfd->gsi; | |
342 | resampler->notifier.irq_acked = irqfd_resampler_ack; | |
343 | INIT_LIST_HEAD(&resampler->link); | |
344 | ||
345 | list_add(&resampler->link, &kvm->irqfds.resampler_list); | |
346 | kvm_register_irq_ack_notifier(kvm, | |
347 | &resampler->notifier); | |
348 | irqfd->resampler = resampler; | |
349 | } | |
350 | ||
351 | list_add_rcu(&irqfd->resampler_link, &irqfd->resampler->list); | |
719d93cd | 352 | synchronize_srcu(&kvm->irq_srcu); |
7a84428a AW |
353 | |
354 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
355 | } | |
356 | ||
721eecbf GH |
357 | /* |
358 | * Install our own custom wake-up handling so we are notified via | |
359 | * a callback whenever someone signals the underlying eventfd | |
360 | */ | |
361 | init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); | |
362 | init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); | |
363 | ||
f1d1c309 MT |
364 | spin_lock_irq(&kvm->irqfds.lock); |
365 | ||
366 | ret = 0; | |
367 | list_for_each_entry(tmp, &kvm->irqfds.items, list) { | |
368 | if (irqfd->eventfd != tmp->eventfd) | |
369 | continue; | |
370 | /* This fd is used for another irq already. */ | |
371 | ret = -EBUSY; | |
372 | spin_unlock_irq(&kvm->irqfds.lock); | |
373 | goto fail; | |
374 | } | |
375 | ||
9957c86d PM |
376 | idx = srcu_read_lock(&kvm->irq_srcu); |
377 | irqfd_update(kvm, irqfd); | |
378 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
bd2b53b2 | 379 | |
721eecbf | 380 | list_add_tail(&irqfd->list, &kvm->irqfds.items); |
721eecbf | 381 | |
684a0b71 CH |
382 | spin_unlock_irq(&kvm->irqfds.lock); |
383 | ||
721eecbf GH |
384 | /* |
385 | * Check if there was an event already pending on the eventfd | |
386 | * before we registered, and trigger it as if we didn't miss it. | |
387 | */ | |
684a0b71 CH |
388 | events = f.file->f_op->poll(f.file, &irqfd->pt); |
389 | ||
721eecbf GH |
390 | if (events & POLLIN) |
391 | schedule_work(&irqfd->inject); | |
392 | ||
393 | /* | |
394 | * do not drop the file until the irqfd is fully initialized, otherwise | |
395 | * we might race against the POLLHUP | |
396 | */ | |
cffe78d9 | 397 | fdput(f); |
9016cfb5 EA |
398 | #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS |
399 | irqfd->consumer.token = (void *)irqfd->eventfd; | |
400 | irqfd->consumer.add_producer = kvm_arch_irq_bypass_add_producer; | |
401 | irqfd->consumer.del_producer = kvm_arch_irq_bypass_del_producer; | |
402 | irqfd->consumer.stop = kvm_arch_irq_bypass_stop; | |
403 | irqfd->consumer.start = kvm_arch_irq_bypass_start; | |
404 | ret = irq_bypass_register_consumer(&irqfd->consumer); | |
405 | if (ret) | |
406 | pr_info("irq bypass consumer (token %p) registration fails: %d\n", | |
407 | irqfd->consumer.token, ret); | |
408 | #endif | |
721eecbf GH |
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 | ||
cffe78d9 | 422 | fdput(f); |
721eecbf | 423 | |
cffe78d9 | 424 | out: |
721eecbf GH |
425 | kfree(irqfd); |
426 | return ret; | |
427 | } | |
c77dcacb PB |
428 | |
429 | bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin) | |
430 | { | |
431 | struct kvm_irq_ack_notifier *kian; | |
432 | int gsi, idx; | |
433 | ||
434 | idx = srcu_read_lock(&kvm->irq_srcu); | |
435 | gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin); | |
436 | if (gsi != -1) | |
437 | hlist_for_each_entry_rcu(kian, &kvm->irq_ack_notifier_list, | |
438 | link) | |
439 | if (kian->gsi == gsi) { | |
440 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
441 | return true; | |
442 | } | |
443 | ||
444 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
445 | ||
446 | return false; | |
447 | } | |
448 | EXPORT_SYMBOL_GPL(kvm_irq_has_notifier); | |
449 | ||
450 | void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin) | |
451 | { | |
452 | struct kvm_irq_ack_notifier *kian; | |
453 | int gsi, idx; | |
454 | ||
455 | trace_kvm_ack_irq(irqchip, pin); | |
456 | ||
457 | idx = srcu_read_lock(&kvm->irq_srcu); | |
458 | gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin); | |
459 | if (gsi != -1) | |
460 | hlist_for_each_entry_rcu(kian, &kvm->irq_ack_notifier_list, | |
461 | link) | |
462 | if (kian->gsi == gsi) | |
463 | kian->irq_acked(kian); | |
464 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
465 | } | |
466 | ||
467 | void kvm_register_irq_ack_notifier(struct kvm *kvm, | |
468 | struct kvm_irq_ack_notifier *kian) | |
469 | { | |
470 | mutex_lock(&kvm->irq_lock); | |
471 | hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list); | |
472 | mutex_unlock(&kvm->irq_lock); | |
c77dcacb | 473 | kvm_vcpu_request_scan_ioapic(kvm); |
c77dcacb PB |
474 | } |
475 | ||
476 | void kvm_unregister_irq_ack_notifier(struct kvm *kvm, | |
477 | struct kvm_irq_ack_notifier *kian) | |
478 | { | |
479 | mutex_lock(&kvm->irq_lock); | |
480 | hlist_del_init_rcu(&kian->link); | |
481 | mutex_unlock(&kvm->irq_lock); | |
482 | synchronize_srcu(&kvm->irq_srcu); | |
c77dcacb | 483 | kvm_vcpu_request_scan_ioapic(kvm); |
c77dcacb | 484 | } |
914daba8 | 485 | #endif |
721eecbf GH |
486 | |
487 | void | |
d34e6b17 | 488 | kvm_eventfd_init(struct kvm *kvm) |
721eecbf | 489 | { |
297e2105 | 490 | #ifdef CONFIG_HAVE_KVM_IRQFD |
721eecbf GH |
491 | spin_lock_init(&kvm->irqfds.lock); |
492 | INIT_LIST_HEAD(&kvm->irqfds.items); | |
7a84428a AW |
493 | INIT_LIST_HEAD(&kvm->irqfds.resampler_list); |
494 | mutex_init(&kvm->irqfds.resampler_lock); | |
914daba8 | 495 | #endif |
d34e6b17 | 496 | INIT_LIST_HEAD(&kvm->ioeventfds); |
721eecbf GH |
497 | } |
498 | ||
297e2105 | 499 | #ifdef CONFIG_HAVE_KVM_IRQFD |
721eecbf GH |
500 | /* |
501 | * shutdown any irqfd's that match fd+gsi | |
502 | */ | |
503 | static int | |
d4db2935 | 504 | kvm_irqfd_deassign(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 505 | { |
166c9775 | 506 | struct kvm_kernel_irqfd *irqfd, *tmp; |
721eecbf GH |
507 | struct eventfd_ctx *eventfd; |
508 | ||
d4db2935 | 509 | eventfd = eventfd_ctx_fdget(args->fd); |
721eecbf GH |
510 | if (IS_ERR(eventfd)) |
511 | return PTR_ERR(eventfd); | |
512 | ||
513 | spin_lock_irq(&kvm->irqfds.lock); | |
514 | ||
515 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) { | |
d4db2935 | 516 | if (irqfd->eventfd == eventfd && irqfd->gsi == args->gsi) { |
bd2b53b2 | 517 | /* |
56f89f36 | 518 | * This clearing of irq_entry.type is needed for when |
c8ce057e MT |
519 | * another thread calls kvm_irq_routing_update before |
520 | * we flush workqueue below (we synchronize with | |
521 | * kvm_irq_routing_update using irqfds.lock). | |
bd2b53b2 | 522 | */ |
56f89f36 PM |
523 | write_seqcount_begin(&irqfd->irq_entry_sc); |
524 | irqfd->irq_entry.type = 0; | |
525 | write_seqcount_end(&irqfd->irq_entry_sc); | |
721eecbf | 526 | irqfd_deactivate(irqfd); |
bd2b53b2 | 527 | } |
721eecbf GH |
528 | } |
529 | ||
530 | spin_unlock_irq(&kvm->irqfds.lock); | |
531 | eventfd_ctx_put(eventfd); | |
532 | ||
533 | /* | |
534 | * Block until we know all outstanding shutdown jobs have completed | |
535 | * so that we guarantee there will not be any more interrupts on this | |
536 | * gsi once this deassign function returns. | |
537 | */ | |
538 | flush_workqueue(irqfd_cleanup_wq); | |
539 | ||
540 | return 0; | |
541 | } | |
542 | ||
543 | int | |
d4db2935 | 544 | kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 545 | { |
7a84428a | 546 | if (args->flags & ~(KVM_IRQFD_FLAG_DEASSIGN | KVM_IRQFD_FLAG_RESAMPLE)) |
326cf033 AW |
547 | return -EINVAL; |
548 | ||
d4db2935 AW |
549 | if (args->flags & KVM_IRQFD_FLAG_DEASSIGN) |
550 | return kvm_irqfd_deassign(kvm, args); | |
721eecbf | 551 | |
d4db2935 | 552 | return kvm_irqfd_assign(kvm, args); |
721eecbf GH |
553 | } |
554 | ||
555 | /* | |
556 | * This function is called as the kvm VM fd is being released. Shutdown all | |
557 | * irqfds that still remain open | |
558 | */ | |
559 | void | |
560 | kvm_irqfd_release(struct kvm *kvm) | |
561 | { | |
166c9775 | 562 | struct kvm_kernel_irqfd *irqfd, *tmp; |
721eecbf GH |
563 | |
564 | spin_lock_irq(&kvm->irqfds.lock); | |
565 | ||
566 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) | |
567 | irqfd_deactivate(irqfd); | |
568 | ||
569 | spin_unlock_irq(&kvm->irqfds.lock); | |
570 | ||
571 | /* | |
572 | * Block until we know all outstanding shutdown jobs have completed | |
573 | * since we do not take a kvm* reference. | |
574 | */ | |
575 | flush_workqueue(irqfd_cleanup_wq); | |
576 | ||
577 | } | |
578 | ||
bd2b53b2 | 579 | /* |
9957c86d | 580 | * Take note of a change in irq routing. |
719d93cd | 581 | * Caller must invoke synchronize_srcu(&kvm->irq_srcu) afterwards. |
bd2b53b2 | 582 | */ |
9957c86d | 583 | void kvm_irq_routing_update(struct kvm *kvm) |
bd2b53b2 | 584 | { |
166c9775 | 585 | struct kvm_kernel_irqfd *irqfd; |
bd2b53b2 MT |
586 | |
587 | spin_lock_irq(&kvm->irqfds.lock); | |
588 | ||
bd2b53b2 | 589 | list_for_each_entry(irqfd, &kvm->irqfds.items, list) |
9957c86d | 590 | irqfd_update(kvm, irqfd); |
bd2b53b2 MT |
591 | |
592 | spin_unlock_irq(&kvm->irqfds.lock); | |
593 | } | |
594 | ||
721eecbf GH |
595 | /* |
596 | * create a host-wide workqueue for issuing deferred shutdown requests | |
597 | * aggregated from all vm* instances. We need our own isolated single-thread | |
598 | * queue to prevent deadlock against flushing the normal work-queue. | |
599 | */ | |
a0f155e9 | 600 | int kvm_irqfd_init(void) |
721eecbf GH |
601 | { |
602 | irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup"); | |
603 | if (!irqfd_cleanup_wq) | |
604 | return -ENOMEM; | |
605 | ||
606 | return 0; | |
607 | } | |
608 | ||
a0f155e9 | 609 | void kvm_irqfd_exit(void) |
721eecbf GH |
610 | { |
611 | destroy_workqueue(irqfd_cleanup_wq); | |
612 | } | |
914daba8 | 613 | #endif |
d34e6b17 GH |
614 | |
615 | /* | |
616 | * -------------------------------------------------------------------- | |
617 | * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal. | |
618 | * | |
619 | * userspace can register a PIO/MMIO address with an eventfd for receiving | |
620 | * notification when the memory has been touched. | |
621 | * -------------------------------------------------------------------- | |
622 | */ | |
623 | ||
624 | struct _ioeventfd { | |
625 | struct list_head list; | |
626 | u64 addr; | |
627 | int length; | |
628 | struct eventfd_ctx *eventfd; | |
629 | u64 datamatch; | |
630 | struct kvm_io_device dev; | |
05e07f9b | 631 | u8 bus_idx; |
d34e6b17 GH |
632 | bool wildcard; |
633 | }; | |
634 | ||
635 | static inline struct _ioeventfd * | |
636 | to_ioeventfd(struct kvm_io_device *dev) | |
637 | { | |
638 | return container_of(dev, struct _ioeventfd, dev); | |
639 | } | |
640 | ||
641 | static void | |
642 | ioeventfd_release(struct _ioeventfd *p) | |
643 | { | |
644 | eventfd_ctx_put(p->eventfd); | |
645 | list_del(&p->list); | |
646 | kfree(p); | |
647 | } | |
648 | ||
649 | static bool | |
650 | ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val) | |
651 | { | |
652 | u64 _val; | |
653 | ||
f848a5a8 MT |
654 | if (addr != p->addr) |
655 | /* address must be precise for a hit */ | |
656 | return false; | |
657 | ||
658 | if (!p->length) | |
659 | /* length = 0 means only look at the address, so always a hit */ | |
660 | return true; | |
661 | ||
662 | if (len != p->length) | |
d34e6b17 GH |
663 | /* address-range must be precise for a hit */ |
664 | return false; | |
665 | ||
666 | if (p->wildcard) | |
667 | /* all else equal, wildcard is always a hit */ | |
668 | return true; | |
669 | ||
670 | /* otherwise, we have to actually compare the data */ | |
671 | ||
672 | BUG_ON(!IS_ALIGNED((unsigned long)val, len)); | |
673 | ||
674 | switch (len) { | |
675 | case 1: | |
676 | _val = *(u8 *)val; | |
677 | break; | |
678 | case 2: | |
679 | _val = *(u16 *)val; | |
680 | break; | |
681 | case 4: | |
682 | _val = *(u32 *)val; | |
683 | break; | |
684 | case 8: | |
685 | _val = *(u64 *)val; | |
686 | break; | |
687 | default: | |
688 | return false; | |
689 | } | |
690 | ||
691 | return _val == p->datamatch ? true : false; | |
692 | } | |
693 | ||
694 | /* MMIO/PIO writes trigger an event if the addr/val match */ | |
695 | static int | |
e32edf4f NN |
696 | ioeventfd_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this, gpa_t addr, |
697 | int len, const void *val) | |
d34e6b17 GH |
698 | { |
699 | struct _ioeventfd *p = to_ioeventfd(this); | |
700 | ||
701 | if (!ioeventfd_in_range(p, addr, len, val)) | |
702 | return -EOPNOTSUPP; | |
703 | ||
704 | eventfd_signal(p->eventfd, 1); | |
705 | return 0; | |
706 | } | |
707 | ||
708 | /* | |
709 | * This function is called as KVM is completely shutting down. We do not | |
710 | * need to worry about locking just nuke anything we have as quickly as possible | |
711 | */ | |
712 | static void | |
713 | ioeventfd_destructor(struct kvm_io_device *this) | |
714 | { | |
715 | struct _ioeventfd *p = to_ioeventfd(this); | |
716 | ||
717 | ioeventfd_release(p); | |
718 | } | |
719 | ||
720 | static const struct kvm_io_device_ops ioeventfd_ops = { | |
721 | .write = ioeventfd_write, | |
722 | .destructor = ioeventfd_destructor, | |
723 | }; | |
724 | ||
725 | /* assumes kvm->slots_lock held */ | |
726 | static bool | |
727 | ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p) | |
728 | { | |
729 | struct _ioeventfd *_p; | |
730 | ||
731 | list_for_each_entry(_p, &kvm->ioeventfds, list) | |
05e07f9b | 732 | if (_p->bus_idx == p->bus_idx && |
f848a5a8 MT |
733 | _p->addr == p->addr && |
734 | (!_p->length || !p->length || | |
735 | (_p->length == p->length && | |
736 | (_p->wildcard || p->wildcard || | |
737 | _p->datamatch == p->datamatch)))) | |
d34e6b17 GH |
738 | return true; |
739 | ||
740 | return false; | |
741 | } | |
742 | ||
2b83451b CH |
743 | static enum kvm_bus ioeventfd_bus_from_flags(__u32 flags) |
744 | { | |
745 | if (flags & KVM_IOEVENTFD_FLAG_PIO) | |
746 | return KVM_PIO_BUS; | |
747 | if (flags & KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY) | |
748 | return KVM_VIRTIO_CCW_NOTIFY_BUS; | |
749 | return KVM_MMIO_BUS; | |
750 | } | |
751 | ||
85da11ca JW |
752 | static int kvm_assign_ioeventfd_idx(struct kvm *kvm, |
753 | enum kvm_bus bus_idx, | |
754 | struct kvm_ioeventfd *args) | |
d34e6b17 | 755 | { |
d34e6b17 | 756 | |
85da11ca JW |
757 | struct eventfd_ctx *eventfd; |
758 | struct _ioeventfd *p; | |
759 | int ret; | |
f848a5a8 | 760 | |
d34e6b17 GH |
761 | eventfd = eventfd_ctx_fdget(args->fd); |
762 | if (IS_ERR(eventfd)) | |
763 | return PTR_ERR(eventfd); | |
764 | ||
765 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
766 | if (!p) { | |
767 | ret = -ENOMEM; | |
768 | goto fail; | |
769 | } | |
770 | ||
771 | INIT_LIST_HEAD(&p->list); | |
772 | p->addr = args->addr; | |
05e07f9b | 773 | p->bus_idx = bus_idx; |
d34e6b17 GH |
774 | p->length = args->len; |
775 | p->eventfd = eventfd; | |
776 | ||
777 | /* The datamatch feature is optional, otherwise this is a wildcard */ | |
778 | if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH) | |
779 | p->datamatch = args->datamatch; | |
780 | else | |
781 | p->wildcard = true; | |
782 | ||
79fac95e | 783 | mutex_lock(&kvm->slots_lock); |
d34e6b17 | 784 | |
25985edc | 785 | /* Verify that there isn't a match already */ |
d34e6b17 GH |
786 | if (ioeventfd_check_collision(kvm, p)) { |
787 | ret = -EEXIST; | |
788 | goto unlock_fail; | |
789 | } | |
790 | ||
791 | kvm_iodevice_init(&p->dev, &ioeventfd_ops); | |
792 | ||
743eeb0b SL |
793 | ret = kvm_io_bus_register_dev(kvm, bus_idx, p->addr, p->length, |
794 | &p->dev); | |
d34e6b17 GH |
795 | if (ret < 0) |
796 | goto unlock_fail; | |
797 | ||
6ea34c9b | 798 | kvm->buses[bus_idx]->ioeventfd_count++; |
d34e6b17 GH |
799 | list_add_tail(&p->list, &kvm->ioeventfds); |
800 | ||
79fac95e | 801 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
802 | |
803 | return 0; | |
804 | ||
805 | unlock_fail: | |
79fac95e | 806 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
807 | |
808 | fail: | |
809 | kfree(p); | |
810 | eventfd_ctx_put(eventfd); | |
811 | ||
812 | return ret; | |
813 | } | |
814 | ||
815 | static int | |
85da11ca JW |
816 | kvm_deassign_ioeventfd_idx(struct kvm *kvm, enum kvm_bus bus_idx, |
817 | struct kvm_ioeventfd *args) | |
d34e6b17 | 818 | { |
d34e6b17 GH |
819 | struct _ioeventfd *p, *tmp; |
820 | struct eventfd_ctx *eventfd; | |
821 | int ret = -ENOENT; | |
822 | ||
823 | eventfd = eventfd_ctx_fdget(args->fd); | |
824 | if (IS_ERR(eventfd)) | |
825 | return PTR_ERR(eventfd); | |
826 | ||
79fac95e | 827 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
828 | |
829 | list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) { | |
830 | bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH); | |
831 | ||
05e07f9b MT |
832 | if (p->bus_idx != bus_idx || |
833 | p->eventfd != eventfd || | |
d34e6b17 GH |
834 | p->addr != args->addr || |
835 | p->length != args->len || | |
836 | p->wildcard != wildcard) | |
837 | continue; | |
838 | ||
839 | if (!p->wildcard && p->datamatch != args->datamatch) | |
840 | continue; | |
841 | ||
e93f8a0f | 842 | kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev); |
6ea34c9b | 843 | kvm->buses[bus_idx]->ioeventfd_count--; |
d34e6b17 GH |
844 | ioeventfd_release(p); |
845 | ret = 0; | |
846 | break; | |
847 | } | |
848 | ||
79fac95e | 849 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
850 | |
851 | eventfd_ctx_put(eventfd); | |
852 | ||
853 | return ret; | |
854 | } | |
855 | ||
85da11ca JW |
856 | static int kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) |
857 | { | |
858 | enum kvm_bus bus_idx = ioeventfd_bus_from_flags(args->flags); | |
eefd6b06 JW |
859 | int ret = kvm_deassign_ioeventfd_idx(kvm, bus_idx, args); |
860 | ||
861 | if (!args->len && bus_idx == KVM_MMIO_BUS) | |
862 | kvm_deassign_ioeventfd_idx(kvm, KVM_FAST_MMIO_BUS, args); | |
85da11ca | 863 | |
eefd6b06 | 864 | return ret; |
85da11ca JW |
865 | } |
866 | ||
867 | static int | |
868 | kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
869 | { | |
870 | enum kvm_bus bus_idx; | |
eefd6b06 | 871 | int ret; |
85da11ca JW |
872 | |
873 | bus_idx = ioeventfd_bus_from_flags(args->flags); | |
874 | /* must be natural-word sized, or 0 to ignore length */ | |
875 | switch (args->len) { | |
876 | case 0: | |
877 | case 1: | |
878 | case 2: | |
879 | case 4: | |
880 | case 8: | |
881 | break; | |
882 | default: | |
883 | return -EINVAL; | |
884 | } | |
885 | ||
886 | /* check for range overflow */ | |
887 | if (args->addr + args->len < args->addr) | |
888 | return -EINVAL; | |
889 | ||
890 | /* check for extra flags that we don't understand */ | |
891 | if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK) | |
892 | return -EINVAL; | |
893 | ||
894 | /* ioeventfd with no length can't be combined with DATAMATCH */ | |
e9ea5069 | 895 | if (!args->len && (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH)) |
85da11ca JW |
896 | return -EINVAL; |
897 | ||
eefd6b06 JW |
898 | ret = kvm_assign_ioeventfd_idx(kvm, bus_idx, args); |
899 | if (ret) | |
900 | goto fail; | |
901 | ||
902 | /* When length is ignored, MMIO is also put on a separate bus, for | |
903 | * faster lookups. | |
904 | */ | |
905 | if (!args->len && bus_idx == KVM_MMIO_BUS) { | |
906 | ret = kvm_assign_ioeventfd_idx(kvm, KVM_FAST_MMIO_BUS, args); | |
907 | if (ret < 0) | |
908 | goto fast_fail; | |
909 | } | |
910 | ||
911 | return 0; | |
912 | ||
913 | fast_fail: | |
914 | kvm_deassign_ioeventfd_idx(kvm, bus_idx, args); | |
915 | fail: | |
916 | return ret; | |
85da11ca JW |
917 | } |
918 | ||
d34e6b17 GH |
919 | int |
920 | kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
921 | { | |
922 | if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN) | |
923 | return kvm_deassign_ioeventfd(kvm, args); | |
924 | ||
925 | return kvm_assign_ioeventfd(kvm, args); | |
926 | } |