]>
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 | } | |
f70c20aa FW |
269 | |
270 | int __attribute__((weak)) kvm_arch_update_irqfd_routing( | |
271 | struct kvm *kvm, unsigned int host_irq, | |
272 | uint32_t guest_irq, bool set) | |
273 | { | |
274 | return 0; | |
275 | } | |
1a02b270 EA |
276 | #endif |
277 | ||
721eecbf | 278 | static int |
d4db2935 | 279 | kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 280 | { |
166c9775 | 281 | struct kvm_kernel_irqfd *irqfd, *tmp; |
cffe78d9 | 282 | struct fd f; |
7a84428a | 283 | struct eventfd_ctx *eventfd = NULL, *resamplefd = NULL; |
721eecbf GH |
284 | int ret; |
285 | unsigned int events; | |
9957c86d | 286 | int idx; |
721eecbf | 287 | |
01c94e64 EA |
288 | if (!kvm_arch_intc_initialized(kvm)) |
289 | return -EAGAIN; | |
290 | ||
721eecbf GH |
291 | irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL); |
292 | if (!irqfd) | |
293 | return -ENOMEM; | |
294 | ||
295 | irqfd->kvm = kvm; | |
d4db2935 | 296 | irqfd->gsi = args->gsi; |
721eecbf GH |
297 | INIT_LIST_HEAD(&irqfd->list); |
298 | INIT_WORK(&irqfd->inject, irqfd_inject); | |
299 | INIT_WORK(&irqfd->shutdown, irqfd_shutdown); | |
56f89f36 | 300 | seqcount_init(&irqfd->irq_entry_sc); |
721eecbf | 301 | |
cffe78d9 AV |
302 | f = fdget(args->fd); |
303 | if (!f.file) { | |
304 | ret = -EBADF; | |
305 | goto out; | |
721eecbf GH |
306 | } |
307 | ||
cffe78d9 | 308 | eventfd = eventfd_ctx_fileget(f.file); |
721eecbf GH |
309 | if (IS_ERR(eventfd)) { |
310 | ret = PTR_ERR(eventfd); | |
311 | goto fail; | |
312 | } | |
313 | ||
314 | irqfd->eventfd = eventfd; | |
315 | ||
7a84428a | 316 | if (args->flags & KVM_IRQFD_FLAG_RESAMPLE) { |
166c9775 | 317 | struct kvm_kernel_irqfd_resampler *resampler; |
7a84428a AW |
318 | |
319 | resamplefd = eventfd_ctx_fdget(args->resamplefd); | |
320 | if (IS_ERR(resamplefd)) { | |
321 | ret = PTR_ERR(resamplefd); | |
322 | goto fail; | |
323 | } | |
324 | ||
325 | irqfd->resamplefd = resamplefd; | |
326 | INIT_LIST_HEAD(&irqfd->resampler_link); | |
327 | ||
328 | mutex_lock(&kvm->irqfds.resampler_lock); | |
329 | ||
330 | list_for_each_entry(resampler, | |
49f8a1a5 | 331 | &kvm->irqfds.resampler_list, link) { |
7a84428a AW |
332 | if (resampler->notifier.gsi == irqfd->gsi) { |
333 | irqfd->resampler = resampler; | |
334 | break; | |
335 | } | |
336 | } | |
337 | ||
338 | if (!irqfd->resampler) { | |
339 | resampler = kzalloc(sizeof(*resampler), GFP_KERNEL); | |
340 | if (!resampler) { | |
341 | ret = -ENOMEM; | |
342 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
343 | goto fail; | |
344 | } | |
345 | ||
346 | resampler->kvm = kvm; | |
347 | INIT_LIST_HEAD(&resampler->list); | |
348 | resampler->notifier.gsi = irqfd->gsi; | |
349 | resampler->notifier.irq_acked = irqfd_resampler_ack; | |
350 | INIT_LIST_HEAD(&resampler->link); | |
351 | ||
352 | list_add(&resampler->link, &kvm->irqfds.resampler_list); | |
353 | kvm_register_irq_ack_notifier(kvm, | |
354 | &resampler->notifier); | |
355 | irqfd->resampler = resampler; | |
356 | } | |
357 | ||
358 | list_add_rcu(&irqfd->resampler_link, &irqfd->resampler->list); | |
719d93cd | 359 | synchronize_srcu(&kvm->irq_srcu); |
7a84428a AW |
360 | |
361 | mutex_unlock(&kvm->irqfds.resampler_lock); | |
362 | } | |
363 | ||
721eecbf GH |
364 | /* |
365 | * Install our own custom wake-up handling so we are notified via | |
366 | * a callback whenever someone signals the underlying eventfd | |
367 | */ | |
368 | init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); | |
369 | init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); | |
370 | ||
f1d1c309 MT |
371 | spin_lock_irq(&kvm->irqfds.lock); |
372 | ||
373 | ret = 0; | |
374 | list_for_each_entry(tmp, &kvm->irqfds.items, list) { | |
375 | if (irqfd->eventfd != tmp->eventfd) | |
376 | continue; | |
377 | /* This fd is used for another irq already. */ | |
378 | ret = -EBUSY; | |
379 | spin_unlock_irq(&kvm->irqfds.lock); | |
380 | goto fail; | |
381 | } | |
382 | ||
9957c86d PM |
383 | idx = srcu_read_lock(&kvm->irq_srcu); |
384 | irqfd_update(kvm, irqfd); | |
385 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
bd2b53b2 | 386 | |
721eecbf | 387 | list_add_tail(&irqfd->list, &kvm->irqfds.items); |
721eecbf | 388 | |
684a0b71 CH |
389 | spin_unlock_irq(&kvm->irqfds.lock); |
390 | ||
721eecbf GH |
391 | /* |
392 | * Check if there was an event already pending on the eventfd | |
393 | * before we registered, and trigger it as if we didn't miss it. | |
394 | */ | |
684a0b71 CH |
395 | events = f.file->f_op->poll(f.file, &irqfd->pt); |
396 | ||
721eecbf GH |
397 | if (events & POLLIN) |
398 | schedule_work(&irqfd->inject); | |
399 | ||
400 | /* | |
401 | * do not drop the file until the irqfd is fully initialized, otherwise | |
402 | * we might race against the POLLHUP | |
403 | */ | |
cffe78d9 | 404 | fdput(f); |
9016cfb5 EA |
405 | #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS |
406 | irqfd->consumer.token = (void *)irqfd->eventfd; | |
407 | irqfd->consumer.add_producer = kvm_arch_irq_bypass_add_producer; | |
408 | irqfd->consumer.del_producer = kvm_arch_irq_bypass_del_producer; | |
409 | irqfd->consumer.stop = kvm_arch_irq_bypass_stop; | |
410 | irqfd->consumer.start = kvm_arch_irq_bypass_start; | |
411 | ret = irq_bypass_register_consumer(&irqfd->consumer); | |
412 | if (ret) | |
413 | pr_info("irq bypass consumer (token %p) registration fails: %d\n", | |
414 | irqfd->consumer.token, ret); | |
415 | #endif | |
721eecbf GH |
416 | |
417 | return 0; | |
418 | ||
419 | fail: | |
7a84428a AW |
420 | if (irqfd->resampler) |
421 | irqfd_resampler_shutdown(irqfd); | |
422 | ||
423 | if (resamplefd && !IS_ERR(resamplefd)) | |
424 | eventfd_ctx_put(resamplefd); | |
425 | ||
721eecbf GH |
426 | if (eventfd && !IS_ERR(eventfd)) |
427 | eventfd_ctx_put(eventfd); | |
428 | ||
cffe78d9 | 429 | fdput(f); |
721eecbf | 430 | |
cffe78d9 | 431 | out: |
721eecbf GH |
432 | kfree(irqfd); |
433 | return ret; | |
434 | } | |
c77dcacb PB |
435 | |
436 | bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin) | |
437 | { | |
438 | struct kvm_irq_ack_notifier *kian; | |
439 | int gsi, idx; | |
440 | ||
441 | idx = srcu_read_lock(&kvm->irq_srcu); | |
442 | gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin); | |
443 | if (gsi != -1) | |
444 | hlist_for_each_entry_rcu(kian, &kvm->irq_ack_notifier_list, | |
445 | link) | |
446 | if (kian->gsi == gsi) { | |
447 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
448 | return true; | |
449 | } | |
450 | ||
451 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
452 | ||
453 | return false; | |
454 | } | |
455 | EXPORT_SYMBOL_GPL(kvm_irq_has_notifier); | |
456 | ||
457 | void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin) | |
458 | { | |
459 | struct kvm_irq_ack_notifier *kian; | |
460 | int gsi, idx; | |
461 | ||
462 | trace_kvm_ack_irq(irqchip, pin); | |
463 | ||
464 | idx = srcu_read_lock(&kvm->irq_srcu); | |
465 | gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin); | |
466 | if (gsi != -1) | |
467 | hlist_for_each_entry_rcu(kian, &kvm->irq_ack_notifier_list, | |
468 | link) | |
469 | if (kian->gsi == gsi) | |
470 | kian->irq_acked(kian); | |
471 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
472 | } | |
473 | ||
474 | void kvm_register_irq_ack_notifier(struct kvm *kvm, | |
475 | struct kvm_irq_ack_notifier *kian) | |
476 | { | |
477 | mutex_lock(&kvm->irq_lock); | |
478 | hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list); | |
479 | mutex_unlock(&kvm->irq_lock); | |
c77dcacb | 480 | kvm_vcpu_request_scan_ioapic(kvm); |
c77dcacb PB |
481 | } |
482 | ||
483 | void kvm_unregister_irq_ack_notifier(struct kvm *kvm, | |
484 | struct kvm_irq_ack_notifier *kian) | |
485 | { | |
486 | mutex_lock(&kvm->irq_lock); | |
487 | hlist_del_init_rcu(&kian->link); | |
488 | mutex_unlock(&kvm->irq_lock); | |
489 | synchronize_srcu(&kvm->irq_srcu); | |
c77dcacb | 490 | kvm_vcpu_request_scan_ioapic(kvm); |
c77dcacb | 491 | } |
914daba8 | 492 | #endif |
721eecbf GH |
493 | |
494 | void | |
d34e6b17 | 495 | kvm_eventfd_init(struct kvm *kvm) |
721eecbf | 496 | { |
297e2105 | 497 | #ifdef CONFIG_HAVE_KVM_IRQFD |
721eecbf GH |
498 | spin_lock_init(&kvm->irqfds.lock); |
499 | INIT_LIST_HEAD(&kvm->irqfds.items); | |
7a84428a AW |
500 | INIT_LIST_HEAD(&kvm->irqfds.resampler_list); |
501 | mutex_init(&kvm->irqfds.resampler_lock); | |
914daba8 | 502 | #endif |
d34e6b17 | 503 | INIT_LIST_HEAD(&kvm->ioeventfds); |
721eecbf GH |
504 | } |
505 | ||
297e2105 | 506 | #ifdef CONFIG_HAVE_KVM_IRQFD |
721eecbf GH |
507 | /* |
508 | * shutdown any irqfd's that match fd+gsi | |
509 | */ | |
510 | static int | |
d4db2935 | 511 | kvm_irqfd_deassign(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 512 | { |
166c9775 | 513 | struct kvm_kernel_irqfd *irqfd, *tmp; |
721eecbf GH |
514 | struct eventfd_ctx *eventfd; |
515 | ||
d4db2935 | 516 | eventfd = eventfd_ctx_fdget(args->fd); |
721eecbf GH |
517 | if (IS_ERR(eventfd)) |
518 | return PTR_ERR(eventfd); | |
519 | ||
520 | spin_lock_irq(&kvm->irqfds.lock); | |
521 | ||
522 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) { | |
d4db2935 | 523 | if (irqfd->eventfd == eventfd && irqfd->gsi == args->gsi) { |
bd2b53b2 | 524 | /* |
56f89f36 | 525 | * This clearing of irq_entry.type is needed for when |
c8ce057e MT |
526 | * another thread calls kvm_irq_routing_update before |
527 | * we flush workqueue below (we synchronize with | |
528 | * kvm_irq_routing_update using irqfds.lock). | |
bd2b53b2 | 529 | */ |
56f89f36 PM |
530 | write_seqcount_begin(&irqfd->irq_entry_sc); |
531 | irqfd->irq_entry.type = 0; | |
532 | write_seqcount_end(&irqfd->irq_entry_sc); | |
721eecbf | 533 | irqfd_deactivate(irqfd); |
bd2b53b2 | 534 | } |
721eecbf GH |
535 | } |
536 | ||
537 | spin_unlock_irq(&kvm->irqfds.lock); | |
538 | eventfd_ctx_put(eventfd); | |
539 | ||
540 | /* | |
541 | * Block until we know all outstanding shutdown jobs have completed | |
542 | * so that we guarantee there will not be any more interrupts on this | |
543 | * gsi once this deassign function returns. | |
544 | */ | |
545 | flush_workqueue(irqfd_cleanup_wq); | |
546 | ||
547 | return 0; | |
548 | } | |
549 | ||
550 | int | |
d4db2935 | 551 | kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args) |
721eecbf | 552 | { |
7a84428a | 553 | if (args->flags & ~(KVM_IRQFD_FLAG_DEASSIGN | KVM_IRQFD_FLAG_RESAMPLE)) |
326cf033 AW |
554 | return -EINVAL; |
555 | ||
d4db2935 AW |
556 | if (args->flags & KVM_IRQFD_FLAG_DEASSIGN) |
557 | return kvm_irqfd_deassign(kvm, args); | |
721eecbf | 558 | |
d4db2935 | 559 | return kvm_irqfd_assign(kvm, args); |
721eecbf GH |
560 | } |
561 | ||
562 | /* | |
563 | * This function is called as the kvm VM fd is being released. Shutdown all | |
564 | * irqfds that still remain open | |
565 | */ | |
566 | void | |
567 | kvm_irqfd_release(struct kvm *kvm) | |
568 | { | |
166c9775 | 569 | struct kvm_kernel_irqfd *irqfd, *tmp; |
721eecbf GH |
570 | |
571 | spin_lock_irq(&kvm->irqfds.lock); | |
572 | ||
573 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) | |
574 | irqfd_deactivate(irqfd); | |
575 | ||
576 | spin_unlock_irq(&kvm->irqfds.lock); | |
577 | ||
578 | /* | |
579 | * Block until we know all outstanding shutdown jobs have completed | |
580 | * since we do not take a kvm* reference. | |
581 | */ | |
582 | flush_workqueue(irqfd_cleanup_wq); | |
583 | ||
584 | } | |
585 | ||
bd2b53b2 | 586 | /* |
9957c86d | 587 | * Take note of a change in irq routing. |
719d93cd | 588 | * Caller must invoke synchronize_srcu(&kvm->irq_srcu) afterwards. |
bd2b53b2 | 589 | */ |
9957c86d | 590 | void kvm_irq_routing_update(struct kvm *kvm) |
bd2b53b2 | 591 | { |
166c9775 | 592 | struct kvm_kernel_irqfd *irqfd; |
bd2b53b2 MT |
593 | |
594 | spin_lock_irq(&kvm->irqfds.lock); | |
595 | ||
f70c20aa | 596 | list_for_each_entry(irqfd, &kvm->irqfds.items, list) { |
9957c86d | 597 | irqfd_update(kvm, irqfd); |
bd2b53b2 | 598 | |
f70c20aa FW |
599 | #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS |
600 | if (irqfd->producer) { | |
601 | int ret = kvm_arch_update_irqfd_routing( | |
602 | irqfd->kvm, irqfd->producer->irq, | |
603 | irqfd->gsi, 1); | |
604 | WARN_ON(ret); | |
605 | } | |
606 | #endif | |
607 | } | |
608 | ||
bd2b53b2 MT |
609 | spin_unlock_irq(&kvm->irqfds.lock); |
610 | } | |
611 | ||
721eecbf GH |
612 | /* |
613 | * create a host-wide workqueue for issuing deferred shutdown requests | |
614 | * aggregated from all vm* instances. We need our own isolated single-thread | |
615 | * queue to prevent deadlock against flushing the normal work-queue. | |
616 | */ | |
a0f155e9 | 617 | int kvm_irqfd_init(void) |
721eecbf GH |
618 | { |
619 | irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup"); | |
620 | if (!irqfd_cleanup_wq) | |
621 | return -ENOMEM; | |
622 | ||
623 | return 0; | |
624 | } | |
625 | ||
a0f155e9 | 626 | void kvm_irqfd_exit(void) |
721eecbf GH |
627 | { |
628 | destroy_workqueue(irqfd_cleanup_wq); | |
629 | } | |
914daba8 | 630 | #endif |
d34e6b17 GH |
631 | |
632 | /* | |
633 | * -------------------------------------------------------------------- | |
634 | * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal. | |
635 | * | |
636 | * userspace can register a PIO/MMIO address with an eventfd for receiving | |
637 | * notification when the memory has been touched. | |
638 | * -------------------------------------------------------------------- | |
639 | */ | |
640 | ||
641 | struct _ioeventfd { | |
642 | struct list_head list; | |
643 | u64 addr; | |
644 | int length; | |
645 | struct eventfd_ctx *eventfd; | |
646 | u64 datamatch; | |
647 | struct kvm_io_device dev; | |
05e07f9b | 648 | u8 bus_idx; |
d34e6b17 GH |
649 | bool wildcard; |
650 | }; | |
651 | ||
652 | static inline struct _ioeventfd * | |
653 | to_ioeventfd(struct kvm_io_device *dev) | |
654 | { | |
655 | return container_of(dev, struct _ioeventfd, dev); | |
656 | } | |
657 | ||
658 | static void | |
659 | ioeventfd_release(struct _ioeventfd *p) | |
660 | { | |
661 | eventfd_ctx_put(p->eventfd); | |
662 | list_del(&p->list); | |
663 | kfree(p); | |
664 | } | |
665 | ||
666 | static bool | |
667 | ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val) | |
668 | { | |
669 | u64 _val; | |
670 | ||
f848a5a8 MT |
671 | if (addr != p->addr) |
672 | /* address must be precise for a hit */ | |
673 | return false; | |
674 | ||
675 | if (!p->length) | |
676 | /* length = 0 means only look at the address, so always a hit */ | |
677 | return true; | |
678 | ||
679 | if (len != p->length) | |
d34e6b17 GH |
680 | /* address-range must be precise for a hit */ |
681 | return false; | |
682 | ||
683 | if (p->wildcard) | |
684 | /* all else equal, wildcard is always a hit */ | |
685 | return true; | |
686 | ||
687 | /* otherwise, we have to actually compare the data */ | |
688 | ||
689 | BUG_ON(!IS_ALIGNED((unsigned long)val, len)); | |
690 | ||
691 | switch (len) { | |
692 | case 1: | |
693 | _val = *(u8 *)val; | |
694 | break; | |
695 | case 2: | |
696 | _val = *(u16 *)val; | |
697 | break; | |
698 | case 4: | |
699 | _val = *(u32 *)val; | |
700 | break; | |
701 | case 8: | |
702 | _val = *(u64 *)val; | |
703 | break; | |
704 | default: | |
705 | return false; | |
706 | } | |
707 | ||
708 | return _val == p->datamatch ? true : false; | |
709 | } | |
710 | ||
711 | /* MMIO/PIO writes trigger an event if the addr/val match */ | |
712 | static int | |
e32edf4f NN |
713 | ioeventfd_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this, gpa_t addr, |
714 | int len, const void *val) | |
d34e6b17 GH |
715 | { |
716 | struct _ioeventfd *p = to_ioeventfd(this); | |
717 | ||
718 | if (!ioeventfd_in_range(p, addr, len, val)) | |
719 | return -EOPNOTSUPP; | |
720 | ||
721 | eventfd_signal(p->eventfd, 1); | |
722 | return 0; | |
723 | } | |
724 | ||
725 | /* | |
726 | * This function is called as KVM is completely shutting down. We do not | |
727 | * need to worry about locking just nuke anything we have as quickly as possible | |
728 | */ | |
729 | static void | |
730 | ioeventfd_destructor(struct kvm_io_device *this) | |
731 | { | |
732 | struct _ioeventfd *p = to_ioeventfd(this); | |
733 | ||
734 | ioeventfd_release(p); | |
735 | } | |
736 | ||
737 | static const struct kvm_io_device_ops ioeventfd_ops = { | |
738 | .write = ioeventfd_write, | |
739 | .destructor = ioeventfd_destructor, | |
740 | }; | |
741 | ||
742 | /* assumes kvm->slots_lock held */ | |
743 | static bool | |
744 | ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p) | |
745 | { | |
746 | struct _ioeventfd *_p; | |
747 | ||
748 | list_for_each_entry(_p, &kvm->ioeventfds, list) | |
05e07f9b | 749 | if (_p->bus_idx == p->bus_idx && |
f848a5a8 MT |
750 | _p->addr == p->addr && |
751 | (!_p->length || !p->length || | |
752 | (_p->length == p->length && | |
753 | (_p->wildcard || p->wildcard || | |
754 | _p->datamatch == p->datamatch)))) | |
d34e6b17 GH |
755 | return true; |
756 | ||
757 | return false; | |
758 | } | |
759 | ||
2b83451b CH |
760 | static enum kvm_bus ioeventfd_bus_from_flags(__u32 flags) |
761 | { | |
762 | if (flags & KVM_IOEVENTFD_FLAG_PIO) | |
763 | return KVM_PIO_BUS; | |
764 | if (flags & KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY) | |
765 | return KVM_VIRTIO_CCW_NOTIFY_BUS; | |
766 | return KVM_MMIO_BUS; | |
767 | } | |
768 | ||
85da11ca JW |
769 | static int kvm_assign_ioeventfd_idx(struct kvm *kvm, |
770 | enum kvm_bus bus_idx, | |
771 | struct kvm_ioeventfd *args) | |
d34e6b17 | 772 | { |
d34e6b17 | 773 | |
85da11ca JW |
774 | struct eventfd_ctx *eventfd; |
775 | struct _ioeventfd *p; | |
776 | int ret; | |
f848a5a8 | 777 | |
d34e6b17 GH |
778 | eventfd = eventfd_ctx_fdget(args->fd); |
779 | if (IS_ERR(eventfd)) | |
780 | return PTR_ERR(eventfd); | |
781 | ||
782 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
783 | if (!p) { | |
784 | ret = -ENOMEM; | |
785 | goto fail; | |
786 | } | |
787 | ||
788 | INIT_LIST_HEAD(&p->list); | |
789 | p->addr = args->addr; | |
05e07f9b | 790 | p->bus_idx = bus_idx; |
d34e6b17 GH |
791 | p->length = args->len; |
792 | p->eventfd = eventfd; | |
793 | ||
794 | /* The datamatch feature is optional, otherwise this is a wildcard */ | |
795 | if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH) | |
796 | p->datamatch = args->datamatch; | |
797 | else | |
798 | p->wildcard = true; | |
799 | ||
79fac95e | 800 | mutex_lock(&kvm->slots_lock); |
d34e6b17 | 801 | |
25985edc | 802 | /* Verify that there isn't a match already */ |
d34e6b17 GH |
803 | if (ioeventfd_check_collision(kvm, p)) { |
804 | ret = -EEXIST; | |
805 | goto unlock_fail; | |
806 | } | |
807 | ||
808 | kvm_iodevice_init(&p->dev, &ioeventfd_ops); | |
809 | ||
743eeb0b SL |
810 | ret = kvm_io_bus_register_dev(kvm, bus_idx, p->addr, p->length, |
811 | &p->dev); | |
d34e6b17 GH |
812 | if (ret < 0) |
813 | goto unlock_fail; | |
814 | ||
6ea34c9b | 815 | kvm->buses[bus_idx]->ioeventfd_count++; |
d34e6b17 GH |
816 | list_add_tail(&p->list, &kvm->ioeventfds); |
817 | ||
79fac95e | 818 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
819 | |
820 | return 0; | |
821 | ||
822 | unlock_fail: | |
79fac95e | 823 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
824 | |
825 | fail: | |
826 | kfree(p); | |
827 | eventfd_ctx_put(eventfd); | |
828 | ||
829 | return ret; | |
830 | } | |
831 | ||
832 | static int | |
85da11ca JW |
833 | kvm_deassign_ioeventfd_idx(struct kvm *kvm, enum kvm_bus bus_idx, |
834 | struct kvm_ioeventfd *args) | |
d34e6b17 | 835 | { |
d34e6b17 GH |
836 | struct _ioeventfd *p, *tmp; |
837 | struct eventfd_ctx *eventfd; | |
838 | int ret = -ENOENT; | |
839 | ||
840 | eventfd = eventfd_ctx_fdget(args->fd); | |
841 | if (IS_ERR(eventfd)) | |
842 | return PTR_ERR(eventfd); | |
843 | ||
79fac95e | 844 | mutex_lock(&kvm->slots_lock); |
d34e6b17 GH |
845 | |
846 | list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) { | |
847 | bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH); | |
848 | ||
05e07f9b MT |
849 | if (p->bus_idx != bus_idx || |
850 | p->eventfd != eventfd || | |
d34e6b17 GH |
851 | p->addr != args->addr || |
852 | p->length != args->len || | |
853 | p->wildcard != wildcard) | |
854 | continue; | |
855 | ||
856 | if (!p->wildcard && p->datamatch != args->datamatch) | |
857 | continue; | |
858 | ||
e93f8a0f | 859 | kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev); |
6ea34c9b | 860 | kvm->buses[bus_idx]->ioeventfd_count--; |
d34e6b17 GH |
861 | ioeventfd_release(p); |
862 | ret = 0; | |
863 | break; | |
864 | } | |
865 | ||
79fac95e | 866 | mutex_unlock(&kvm->slots_lock); |
d34e6b17 GH |
867 | |
868 | eventfd_ctx_put(eventfd); | |
869 | ||
870 | return ret; | |
871 | } | |
872 | ||
85da11ca JW |
873 | static int kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) |
874 | { | |
875 | enum kvm_bus bus_idx = ioeventfd_bus_from_flags(args->flags); | |
eefd6b06 JW |
876 | int ret = kvm_deassign_ioeventfd_idx(kvm, bus_idx, args); |
877 | ||
878 | if (!args->len && bus_idx == KVM_MMIO_BUS) | |
879 | kvm_deassign_ioeventfd_idx(kvm, KVM_FAST_MMIO_BUS, args); | |
85da11ca | 880 | |
eefd6b06 | 881 | return ret; |
85da11ca JW |
882 | } |
883 | ||
884 | static int | |
885 | kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
886 | { | |
887 | enum kvm_bus bus_idx; | |
eefd6b06 | 888 | int ret; |
85da11ca JW |
889 | |
890 | bus_idx = ioeventfd_bus_from_flags(args->flags); | |
891 | /* must be natural-word sized, or 0 to ignore length */ | |
892 | switch (args->len) { | |
893 | case 0: | |
894 | case 1: | |
895 | case 2: | |
896 | case 4: | |
897 | case 8: | |
898 | break; | |
899 | default: | |
900 | return -EINVAL; | |
901 | } | |
902 | ||
903 | /* check for range overflow */ | |
904 | if (args->addr + args->len < args->addr) | |
905 | return -EINVAL; | |
906 | ||
907 | /* check for extra flags that we don't understand */ | |
908 | if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK) | |
909 | return -EINVAL; | |
910 | ||
911 | /* ioeventfd with no length can't be combined with DATAMATCH */ | |
e9ea5069 | 912 | if (!args->len && (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH)) |
85da11ca JW |
913 | return -EINVAL; |
914 | ||
eefd6b06 JW |
915 | ret = kvm_assign_ioeventfd_idx(kvm, bus_idx, args); |
916 | if (ret) | |
917 | goto fail; | |
918 | ||
919 | /* When length is ignored, MMIO is also put on a separate bus, for | |
920 | * faster lookups. | |
921 | */ | |
922 | if (!args->len && bus_idx == KVM_MMIO_BUS) { | |
923 | ret = kvm_assign_ioeventfd_idx(kvm, KVM_FAST_MMIO_BUS, args); | |
924 | if (ret < 0) | |
925 | goto fast_fail; | |
926 | } | |
927 | ||
928 | return 0; | |
929 | ||
930 | fast_fail: | |
931 | kvm_deassign_ioeventfd_idx(kvm, bus_idx, args); | |
932 | fail: | |
933 | return ret; | |
85da11ca JW |
934 | } |
935 | ||
d34e6b17 GH |
936 | int |
937 | kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args) | |
938 | { | |
939 | if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN) | |
940 | return kvm_deassign_ioeventfd(kvm, args); | |
941 | ||
942 | return kvm_assign_ioeventfd(kvm, args); | |
943 | } |