blob: 148b2392c762ba763a6ad09b314699c451b463d0 [file] [log] [blame]
Gregory Haskins721eecbf2009-05-20 10:30:49 -04001/*
2 * kvm eventfd support - use eventfd objects to signal various KVM events
3 *
4 * Copyright 2009 Novell. All Rights Reserved.
Avi Kivity221d0592010-05-23 18:37:00 +03005 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Gregory Haskins721eecbf2009-05-20 10:30:49 -04006 *
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>
Gregory Haskinsd34e6b12009-07-07 17:08:49 -040025#include <linux/kvm.h>
Gregory Haskins721eecbf2009-05-20 10:30:49 -040026#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>
Gregory Haskinsd34e6b12009-07-07 17:08:49 -040033#include <linux/kernel.h>
Christian Borntraeger719d93c2014-01-16 13:44:20 +010034#include <linux/srcu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090035#include <linux/slab.h>
Paul Mackerras56f89f32014-06-30 20:51:09 +100036#include <linux/seqlock.h>
Paul Mackerrase4d57e12014-06-30 20:51:12 +100037#include <trace/events/kvm.h>
Gregory Haskinsd34e6b12009-07-07 17:08:49 -040038
39#include "iodev.h"
Gregory Haskins721eecbf2009-05-20 10:30:49 -040040
Paul Mackerras297e2102014-06-30 20:51:13 +100041#ifdef CONFIG_HAVE_KVM_IRQFD
Gregory Haskins721eecbf2009-05-20 10:30:49 -040042/*
43 * --------------------------------------------------------------------
44 * irqfd: Allows an fd to be used to inject an interrupt to the guest
45 *
46 * Credit goes to Avi Kivity for the original idea.
47 * --------------------------------------------------------------------
48 */
49
Alex Williamson7a844282012-09-21 11:58:03 -060050/*
51 * Resampling irqfds are a special variety of irqfds used to emulate
52 * level triggered interrupts. The interrupt is asserted on eventfd
53 * trigger. On acknowledgement through the irq ack notifier, the
54 * interrupt is de-asserted and userspace is notified through the
55 * resamplefd. All resamplers on the same gsi are de-asserted
56 * together, so we don't need to track the state of each individual
57 * user. We can also therefore share the same irq source ID.
58 */
59struct _irqfd_resampler {
60 struct kvm *kvm;
61 /*
62 * List of resampling struct _irqfd objects sharing this gsi.
63 * RCU list modified under kvm->irqfds.resampler_lock
64 */
65 struct list_head list;
66 struct kvm_irq_ack_notifier notifier;
67 /*
68 * Entry in list of kvm->irqfd.resampler_list. Use for sharing
69 * resamplers among irqfds on the same gsi.
70 * Accessed and modified under kvm->irqfds.resampler_lock
71 */
72 struct list_head link;
73};
74
Gregory Haskins721eecbf2009-05-20 10:30:49 -040075struct _irqfd {
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +020076 /* Used for MSI fast-path */
77 struct kvm *kvm;
78 wait_queue_t wait;
79 /* Update side is protected by irqfds.lock */
Paul Mackerras56f89f32014-06-30 20:51:09 +100080 struct kvm_kernel_irq_routing_entry irq_entry;
81 seqcount_t irq_entry_sc;
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +020082 /* Used for level IRQ fast-path */
83 int gsi;
84 struct work_struct inject;
Alex Williamson7a844282012-09-21 11:58:03 -060085 /* The resampler used by this irqfd (resampler-only) */
86 struct _irqfd_resampler *resampler;
87 /* Eventfd notified on resample (resampler-only) */
88 struct eventfd_ctx *resamplefd;
89 /* Entry in list of irqfds for a resampler (resampler-only) */
90 struct list_head resampler_link;
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +020091 /* Used for setup/shutdown */
92 struct eventfd_ctx *eventfd;
93 struct list_head list;
94 poll_table pt;
95 struct work_struct shutdown;
Gregory Haskins721eecbf2009-05-20 10:30:49 -040096};
97
98static struct workqueue_struct *irqfd_cleanup_wq;
99
100static void
101irqfd_inject(struct work_struct *work)
102{
103 struct _irqfd *irqfd = container_of(work, struct _irqfd, inject);
104 struct kvm *kvm = irqfd->kvm;
105
Alex Williamson7a844282012-09-21 11:58:03 -0600106 if (!irqfd->resampler) {
Yang Zhangaa2fbe62013-04-11 19:21:40 +0800107 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1,
108 false);
109 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0,
110 false);
Alex Williamson7a844282012-09-21 11:58:03 -0600111 } else
112 kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
Yang Zhangaa2fbe62013-04-11 19:21:40 +0800113 irqfd->gsi, 1, false);
Alex Williamson7a844282012-09-21 11:58:03 -0600114}
115
116/*
117 * Since resampler irqfds share an IRQ source ID, we de-assert once
118 * then notify all of the resampler irqfds using this GSI. We can't
119 * do multiple de-asserts or we risk racing with incoming re-asserts.
120 */
121static void
122irqfd_resampler_ack(struct kvm_irq_ack_notifier *kian)
123{
124 struct _irqfd_resampler *resampler;
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100125 struct kvm *kvm;
Alex Williamson7a844282012-09-21 11:58:03 -0600126 struct _irqfd *irqfd;
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100127 int idx;
Alex Williamson7a844282012-09-21 11:58:03 -0600128
129 resampler = container_of(kian, struct _irqfd_resampler, notifier);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100130 kvm = resampler->kvm;
Alex Williamson7a844282012-09-21 11:58:03 -0600131
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100132 kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
Yang Zhangaa2fbe62013-04-11 19:21:40 +0800133 resampler->notifier.gsi, 0, false);
Alex Williamson7a844282012-09-21 11:58:03 -0600134
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100135 idx = srcu_read_lock(&kvm->irq_srcu);
Alex Williamson7a844282012-09-21 11:58:03 -0600136
137 list_for_each_entry_rcu(irqfd, &resampler->list, resampler_link)
138 eventfd_signal(irqfd->resamplefd, 1);
139
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100140 srcu_read_unlock(&kvm->irq_srcu, idx);
Alex Williamson7a844282012-09-21 11:58:03 -0600141}
142
143static void
144irqfd_resampler_shutdown(struct _irqfd *irqfd)
145{
146 struct _irqfd_resampler *resampler = irqfd->resampler;
147 struct kvm *kvm = resampler->kvm;
148
149 mutex_lock(&kvm->irqfds.resampler_lock);
150
151 list_del_rcu(&irqfd->resampler_link);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100152 synchronize_srcu(&kvm->irq_srcu);
Alex Williamson7a844282012-09-21 11:58:03 -0600153
154 if (list_empty(&resampler->list)) {
155 list_del(&resampler->link);
156 kvm_unregister_irq_ack_notifier(kvm, &resampler->notifier);
157 kvm_set_irq(kvm, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
Yang Zhangaa2fbe62013-04-11 19:21:40 +0800158 resampler->notifier.gsi, 0, false);
Alex Williamson7a844282012-09-21 11:58:03 -0600159 kfree(resampler);
160 }
161
162 mutex_unlock(&kvm->irqfds.resampler_lock);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400163}
164
165/*
166 * Race-free decouple logic (ordering is critical)
167 */
168static void
169irqfd_shutdown(struct work_struct *work)
170{
171 struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown);
Michael S. Tsirkinb6a114d2010-01-13 19:12:30 +0200172 u64 cnt;
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400173
174 /*
175 * Synchronize with the wait-queue and unhook ourselves to prevent
176 * further events.
177 */
Michael S. Tsirkinb6a114d2010-01-13 19:12:30 +0200178 eventfd_ctx_remove_wait_queue(irqfd->eventfd, &irqfd->wait, &cnt);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400179
180 /*
181 * We know no new events will be scheduled at this point, so block
182 * until all previously outstanding events have completed
183 */
Tejun Heo43829732012-08-20 14:51:24 -0700184 flush_work(&irqfd->inject);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400185
Alex Williamson7a844282012-09-21 11:58:03 -0600186 if (irqfd->resampler) {
187 irqfd_resampler_shutdown(irqfd);
188 eventfd_ctx_put(irqfd->resamplefd);
189 }
190
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400191 /*
192 * It is now safe to release the object's resources
193 */
194 eventfd_ctx_put(irqfd->eventfd);
195 kfree(irqfd);
196}
197
198
199/* assumes kvm->irqfds.lock is held */
200static bool
201irqfd_is_active(struct _irqfd *irqfd)
202{
203 return list_empty(&irqfd->list) ? false : true;
204}
205
206/*
207 * Mark the irqfd as inactive and schedule it for removal
208 *
209 * assumes kvm->irqfds.lock is held
210 */
211static void
212irqfd_deactivate(struct _irqfd *irqfd)
213{
214 BUG_ON(!irqfd_is_active(irqfd));
215
216 list_del_init(&irqfd->list);
217
218 queue_work(irqfd_cleanup_wq, &irqfd->shutdown);
219}
220
221/*
222 * Called with wqh->lock held and interrupts disabled
223 */
224static int
225irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key)
226{
227 struct _irqfd *irqfd = container_of(wait, struct _irqfd, wait);
228 unsigned long flags = (unsigned long)key;
Paul Mackerras56f89f32014-06-30 20:51:09 +1000229 struct kvm_kernel_irq_routing_entry irq;
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200230 struct kvm *kvm = irqfd->kvm;
Paul Mackerras56f89f32014-06-30 20:51:09 +1000231 unsigned seq;
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100232 int idx;
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400233
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200234 if (flags & POLLIN) {
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100235 idx = srcu_read_lock(&kvm->irq_srcu);
Paul Mackerras56f89f32014-06-30 20:51:09 +1000236 do {
237 seq = read_seqcount_begin(&irqfd->irq_entry_sc);
238 irq = irqfd->irq_entry;
239 } while (read_seqcount_retry(&irqfd->irq_entry_sc, seq));
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400240 /* An event has been signaled, inject an interrupt */
Paul Mackerras56f89f32014-06-30 20:51:09 +1000241 if (irq.type == KVM_IRQ_ROUTING_MSI)
242 kvm_set_msi(&irq, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1,
Yang Zhangaa2fbe62013-04-11 19:21:40 +0800243 false);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200244 else
245 schedule_work(&irqfd->inject);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100246 srcu_read_unlock(&kvm->irq_srcu, idx);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200247 }
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400248
249 if (flags & POLLHUP) {
250 /* The eventfd is closing, detach from KVM */
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400251 unsigned long flags;
252
253 spin_lock_irqsave(&kvm->irqfds.lock, flags);
254
255 /*
256 * We must check if someone deactivated the irqfd before
257 * we could acquire the irqfds.lock since the item is
258 * deactivated from the KVM side before it is unhooked from
259 * the wait-queue. If it is already deactivated, we can
260 * simply return knowing the other side will cleanup for us.
261 * We cannot race against the irqfd going away since the
262 * other side is required to acquire wqh->lock, which we hold
263 */
264 if (irqfd_is_active(irqfd))
265 irqfd_deactivate(irqfd);
266
267 spin_unlock_irqrestore(&kvm->irqfds.lock, flags);
268 }
269
270 return 0;
271}
272
273static void
274irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh,
275 poll_table *pt)
276{
277 struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400278 add_wait_queue(wqh, &irqfd->wait);
279}
280
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200281/* Must be called under irqfds.lock */
Paul Mackerras9957c862014-06-30 20:51:11 +1000282static void irqfd_update(struct kvm *kvm, struct _irqfd *irqfd)
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200283{
284 struct kvm_kernel_irq_routing_entry *e;
Paul Mackerras8ba918d2014-06-30 20:51:10 +1000285 struct kvm_kernel_irq_routing_entry entries[KVM_NR_IRQCHIPS];
286 int i, n_entries;
287
Paul Mackerras9957c862014-06-30 20:51:11 +1000288 n_entries = kvm_irq_map_gsi(kvm, entries, irqfd->gsi);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200289
Paul Mackerras56f89f32014-06-30 20:51:09 +1000290 write_seqcount_begin(&irqfd->irq_entry_sc);
291
292 irqfd->irq_entry.type = 0;
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200293
Paul Mackerras8ba918d2014-06-30 20:51:10 +1000294 e = entries;
295 for (i = 0; i < n_entries; ++i, ++e) {
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200296 /* Only fast-path MSI. */
297 if (e->type == KVM_IRQ_ROUTING_MSI)
Paul Mackerras56f89f32014-06-30 20:51:09 +1000298 irqfd->irq_entry = *e;
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200299 }
Paul Mackerras56f89f32014-06-30 20:51:09 +1000300
Paul Mackerras56f89f32014-06-30 20:51:09 +1000301 write_seqcount_end(&irqfd->irq_entry_sc);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200302}
303
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400304static int
Alex Williamsond4db2932012-06-29 09:56:08 -0600305kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args)
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400306{
Michael S. Tsirkinf1d1c302010-01-13 18:58:09 +0200307 struct _irqfd *irqfd, *tmp;
Al Virocffe78d2013-08-30 15:47:17 -0400308 struct fd f;
Alex Williamson7a844282012-09-21 11:58:03 -0600309 struct eventfd_ctx *eventfd = NULL, *resamplefd = NULL;
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400310 int ret;
311 unsigned int events;
Paul Mackerras9957c862014-06-30 20:51:11 +1000312 int idx;
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400313
314 irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL);
315 if (!irqfd)
316 return -ENOMEM;
317
318 irqfd->kvm = kvm;
Alex Williamsond4db2932012-06-29 09:56:08 -0600319 irqfd->gsi = args->gsi;
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400320 INIT_LIST_HEAD(&irqfd->list);
321 INIT_WORK(&irqfd->inject, irqfd_inject);
322 INIT_WORK(&irqfd->shutdown, irqfd_shutdown);
Paul Mackerras56f89f32014-06-30 20:51:09 +1000323 seqcount_init(&irqfd->irq_entry_sc);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400324
Al Virocffe78d2013-08-30 15:47:17 -0400325 f = fdget(args->fd);
326 if (!f.file) {
327 ret = -EBADF;
328 goto out;
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400329 }
330
Al Virocffe78d2013-08-30 15:47:17 -0400331 eventfd = eventfd_ctx_fileget(f.file);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400332 if (IS_ERR(eventfd)) {
333 ret = PTR_ERR(eventfd);
334 goto fail;
335 }
336
337 irqfd->eventfd = eventfd;
338
Alex Williamson7a844282012-09-21 11:58:03 -0600339 if (args->flags & KVM_IRQFD_FLAG_RESAMPLE) {
340 struct _irqfd_resampler *resampler;
341
342 resamplefd = eventfd_ctx_fdget(args->resamplefd);
343 if (IS_ERR(resamplefd)) {
344 ret = PTR_ERR(resamplefd);
345 goto fail;
346 }
347
348 irqfd->resamplefd = resamplefd;
349 INIT_LIST_HEAD(&irqfd->resampler_link);
350
351 mutex_lock(&kvm->irqfds.resampler_lock);
352
353 list_for_each_entry(resampler,
Alex Williamson49f8a1a2012-12-06 14:44:59 -0700354 &kvm->irqfds.resampler_list, link) {
Alex Williamson7a844282012-09-21 11:58:03 -0600355 if (resampler->notifier.gsi == irqfd->gsi) {
356 irqfd->resampler = resampler;
357 break;
358 }
359 }
360
361 if (!irqfd->resampler) {
362 resampler = kzalloc(sizeof(*resampler), GFP_KERNEL);
363 if (!resampler) {
364 ret = -ENOMEM;
365 mutex_unlock(&kvm->irqfds.resampler_lock);
366 goto fail;
367 }
368
369 resampler->kvm = kvm;
370 INIT_LIST_HEAD(&resampler->list);
371 resampler->notifier.gsi = irqfd->gsi;
372 resampler->notifier.irq_acked = irqfd_resampler_ack;
373 INIT_LIST_HEAD(&resampler->link);
374
375 list_add(&resampler->link, &kvm->irqfds.resampler_list);
376 kvm_register_irq_ack_notifier(kvm,
377 &resampler->notifier);
378 irqfd->resampler = resampler;
379 }
380
381 list_add_rcu(&irqfd->resampler_link, &irqfd->resampler->list);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100382 synchronize_srcu(&kvm->irq_srcu);
Alex Williamson7a844282012-09-21 11:58:03 -0600383
384 mutex_unlock(&kvm->irqfds.resampler_lock);
385 }
386
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400387 /*
388 * Install our own custom wake-up handling so we are notified via
389 * a callback whenever someone signals the underlying eventfd
390 */
391 init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup);
392 init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc);
393
Michael S. Tsirkinf1d1c302010-01-13 18:58:09 +0200394 spin_lock_irq(&kvm->irqfds.lock);
395
396 ret = 0;
397 list_for_each_entry(tmp, &kvm->irqfds.items, list) {
398 if (irqfd->eventfd != tmp->eventfd)
399 continue;
400 /* This fd is used for another irq already. */
401 ret = -EBUSY;
402 spin_unlock_irq(&kvm->irqfds.lock);
403 goto fail;
404 }
405
Paul Mackerras9957c862014-06-30 20:51:11 +1000406 idx = srcu_read_lock(&kvm->irq_srcu);
407 irqfd_update(kvm, irqfd);
408 srcu_read_unlock(&kvm->irq_srcu, idx);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200409
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400410 list_add_tail(&irqfd->list, &kvm->irqfds.items);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400411
Cornelia Huck684a0b72014-03-17 19:11:35 +0100412 spin_unlock_irq(&kvm->irqfds.lock);
413
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400414 /*
415 * Check if there was an event already pending on the eventfd
416 * before we registered, and trigger it as if we didn't miss it.
417 */
Cornelia Huck684a0b72014-03-17 19:11:35 +0100418 events = f.file->f_op->poll(f.file, &irqfd->pt);
419
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400420 if (events & POLLIN)
421 schedule_work(&irqfd->inject);
422
423 /*
424 * do not drop the file until the irqfd is fully initialized, otherwise
425 * we might race against the POLLHUP
426 */
Al Virocffe78d2013-08-30 15:47:17 -0400427 fdput(f);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400428
429 return 0;
430
431fail:
Alex Williamson7a844282012-09-21 11:58:03 -0600432 if (irqfd->resampler)
433 irqfd_resampler_shutdown(irqfd);
434
435 if (resamplefd && !IS_ERR(resamplefd))
436 eventfd_ctx_put(resamplefd);
437
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400438 if (eventfd && !IS_ERR(eventfd))
439 eventfd_ctx_put(eventfd);
440
Al Virocffe78d2013-08-30 15:47:17 -0400441 fdput(f);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400442
Al Virocffe78d2013-08-30 15:47:17 -0400443out:
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400444 kfree(irqfd);
445 return ret;
446}
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200447
448bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin)
449{
450 struct kvm_irq_ack_notifier *kian;
451 int gsi, idx;
452
453 idx = srcu_read_lock(&kvm->irq_srcu);
454 gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin);
455 if (gsi != -1)
456 hlist_for_each_entry_rcu(kian, &kvm->irq_ack_notifier_list,
457 link)
458 if (kian->gsi == gsi) {
459 srcu_read_unlock(&kvm->irq_srcu, idx);
460 return true;
461 }
462
463 srcu_read_unlock(&kvm->irq_srcu, idx);
464
465 return false;
466}
467EXPORT_SYMBOL_GPL(kvm_irq_has_notifier);
468
469void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin)
470{
471 struct kvm_irq_ack_notifier *kian;
472 int gsi, idx;
473
474 trace_kvm_ack_irq(irqchip, pin);
475
476 idx = srcu_read_lock(&kvm->irq_srcu);
477 gsi = kvm_irq_map_chip_pin(kvm, irqchip, pin);
478 if (gsi != -1)
479 hlist_for_each_entry_rcu(kian, &kvm->irq_ack_notifier_list,
480 link)
481 if (kian->gsi == gsi)
482 kian->irq_acked(kian);
483 srcu_read_unlock(&kvm->irq_srcu, idx);
484}
485
486void kvm_register_irq_ack_notifier(struct kvm *kvm,
487 struct kvm_irq_ack_notifier *kian)
488{
489 mutex_lock(&kvm->irq_lock);
490 hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list);
491 mutex_unlock(&kvm->irq_lock);
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200492 kvm_vcpu_request_scan_ioapic(kvm);
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200493}
494
495void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
496 struct kvm_irq_ack_notifier *kian)
497{
498 mutex_lock(&kvm->irq_lock);
499 hlist_del_init_rcu(&kian->link);
500 mutex_unlock(&kvm->irq_lock);
501 synchronize_srcu(&kvm->irq_srcu);
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200502 kvm_vcpu_request_scan_ioapic(kvm);
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200503}
Alexander Graf914daba2012-10-09 00:22:59 +0200504#endif
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400505
506void
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400507kvm_eventfd_init(struct kvm *kvm)
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400508{
Paul Mackerras297e2102014-06-30 20:51:13 +1000509#ifdef CONFIG_HAVE_KVM_IRQFD
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400510 spin_lock_init(&kvm->irqfds.lock);
511 INIT_LIST_HEAD(&kvm->irqfds.items);
Alex Williamson7a844282012-09-21 11:58:03 -0600512 INIT_LIST_HEAD(&kvm->irqfds.resampler_list);
513 mutex_init(&kvm->irqfds.resampler_lock);
Alexander Graf914daba2012-10-09 00:22:59 +0200514#endif
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400515 INIT_LIST_HEAD(&kvm->ioeventfds);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400516}
517
Paul Mackerras297e2102014-06-30 20:51:13 +1000518#ifdef CONFIG_HAVE_KVM_IRQFD
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400519/*
520 * shutdown any irqfd's that match fd+gsi
521 */
522static int
Alex Williamsond4db2932012-06-29 09:56:08 -0600523kvm_irqfd_deassign(struct kvm *kvm, struct kvm_irqfd *args)
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400524{
525 struct _irqfd *irqfd, *tmp;
526 struct eventfd_ctx *eventfd;
527
Alex Williamsond4db2932012-06-29 09:56:08 -0600528 eventfd = eventfd_ctx_fdget(args->fd);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400529 if (IS_ERR(eventfd))
530 return PTR_ERR(eventfd);
531
532 spin_lock_irq(&kvm->irqfds.lock);
533
534 list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) {
Alex Williamsond4db2932012-06-29 09:56:08 -0600535 if (irqfd->eventfd == eventfd && irqfd->gsi == args->gsi) {
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200536 /*
Paul Mackerras56f89f32014-06-30 20:51:09 +1000537 * This clearing of irq_entry.type is needed for when
Michael S. Tsirkinc8ce0572011-03-06 13:03:26 +0200538 * another thread calls kvm_irq_routing_update before
539 * we flush workqueue below (we synchronize with
540 * kvm_irq_routing_update using irqfds.lock).
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200541 */
Paul Mackerras56f89f32014-06-30 20:51:09 +1000542 write_seqcount_begin(&irqfd->irq_entry_sc);
543 irqfd->irq_entry.type = 0;
544 write_seqcount_end(&irqfd->irq_entry_sc);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400545 irqfd_deactivate(irqfd);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200546 }
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400547 }
548
549 spin_unlock_irq(&kvm->irqfds.lock);
550 eventfd_ctx_put(eventfd);
551
552 /*
553 * Block until we know all outstanding shutdown jobs have completed
554 * so that we guarantee there will not be any more interrupts on this
555 * gsi once this deassign function returns.
556 */
557 flush_workqueue(irqfd_cleanup_wq);
558
559 return 0;
560}
561
562int
Alex Williamsond4db2932012-06-29 09:56:08 -0600563kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400564{
Alex Williamson7a844282012-09-21 11:58:03 -0600565 if (args->flags & ~(KVM_IRQFD_FLAG_DEASSIGN | KVM_IRQFD_FLAG_RESAMPLE))
Alex Williamson326cf032012-06-29 09:56:24 -0600566 return -EINVAL;
567
Alex Williamsond4db2932012-06-29 09:56:08 -0600568 if (args->flags & KVM_IRQFD_FLAG_DEASSIGN)
569 return kvm_irqfd_deassign(kvm, args);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400570
Alex Williamsond4db2932012-06-29 09:56:08 -0600571 return kvm_irqfd_assign(kvm, args);
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400572}
573
574/*
575 * This function is called as the kvm VM fd is being released. Shutdown all
576 * irqfds that still remain open
577 */
578void
579kvm_irqfd_release(struct kvm *kvm)
580{
581 struct _irqfd *irqfd, *tmp;
582
583 spin_lock_irq(&kvm->irqfds.lock);
584
585 list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list)
586 irqfd_deactivate(irqfd);
587
588 spin_unlock_irq(&kvm->irqfds.lock);
589
590 /*
591 * Block until we know all outstanding shutdown jobs have completed
592 * since we do not take a kvm* reference.
593 */
594 flush_workqueue(irqfd_cleanup_wq);
595
596}
597
598/*
Paul Mackerras9957c862014-06-30 20:51:11 +1000599 * Take note of a change in irq routing.
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100600 * Caller must invoke synchronize_srcu(&kvm->irq_srcu) afterwards.
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200601 */
Paul Mackerras9957c862014-06-30 20:51:11 +1000602void kvm_irq_routing_update(struct kvm *kvm)
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200603{
604 struct _irqfd *irqfd;
605
606 spin_lock_irq(&kvm->irqfds.lock);
607
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200608 list_for_each_entry(irqfd, &kvm->irqfds.items, list)
Paul Mackerras9957c862014-06-30 20:51:11 +1000609 irqfd_update(kvm, irqfd);
Michael S. Tsirkinbd2b53b2010-11-18 19:09:08 +0200610
611 spin_unlock_irq(&kvm->irqfds.lock);
612}
613
614/*
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400615 * create a host-wide workqueue for issuing deferred shutdown requests
616 * aggregated from all vm* instances. We need our own isolated single-thread
617 * queue to prevent deadlock against flushing the normal work-queue.
618 */
Cornelia Hucka0f155e2013-02-28 12:33:18 +0100619int kvm_irqfd_init(void)
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400620{
621 irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup");
622 if (!irqfd_cleanup_wq)
623 return -ENOMEM;
624
625 return 0;
626}
627
Cornelia Hucka0f155e2013-02-28 12:33:18 +0100628void kvm_irqfd_exit(void)
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400629{
630 destroy_workqueue(irqfd_cleanup_wq);
631}
Alexander Graf914daba2012-10-09 00:22:59 +0200632#endif
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400633
634/*
635 * --------------------------------------------------------------------
636 * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal.
637 *
638 * userspace can register a PIO/MMIO address with an eventfd for receiving
639 * notification when the memory has been touched.
640 * --------------------------------------------------------------------
641 */
642
643struct _ioeventfd {
644 struct list_head list;
645 u64 addr;
646 int length;
647 struct eventfd_ctx *eventfd;
648 u64 datamatch;
649 struct kvm_io_device dev;
Michael S. Tsirkin05e07f92013-04-04 13:27:21 +0300650 u8 bus_idx;
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400651 bool wildcard;
652};
653
654static inline struct _ioeventfd *
655to_ioeventfd(struct kvm_io_device *dev)
656{
657 return container_of(dev, struct _ioeventfd, dev);
658}
659
660static void
661ioeventfd_release(struct _ioeventfd *p)
662{
663 eventfd_ctx_put(p->eventfd);
664 list_del(&p->list);
665 kfree(p);
666}
667
668static bool
669ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val)
670{
671 u64 _val;
672
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +0300673 if (addr != p->addr)
674 /* address must be precise for a hit */
675 return false;
676
677 if (!p->length)
678 /* length = 0 means only look at the address, so always a hit */
679 return true;
680
681 if (len != p->length)
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400682 /* address-range must be precise for a hit */
683 return false;
684
685 if (p->wildcard)
686 /* all else equal, wildcard is always a hit */
687 return true;
688
689 /* otherwise, we have to actually compare the data */
690
691 BUG_ON(!IS_ALIGNED((unsigned long)val, len));
692
693 switch (len) {
694 case 1:
695 _val = *(u8 *)val;
696 break;
697 case 2:
698 _val = *(u16 *)val;
699 break;
700 case 4:
701 _val = *(u32 *)val;
702 break;
703 case 8:
704 _val = *(u64 *)val;
705 break;
706 default:
707 return false;
708 }
709
710 return _val == p->datamatch ? true : false;
711}
712
713/* MMIO/PIO writes trigger an event if the addr/val match */
714static int
715ioeventfd_write(struct kvm_io_device *this, gpa_t addr, int len,
716 const void *val)
717{
718 struct _ioeventfd *p = to_ioeventfd(this);
719
720 if (!ioeventfd_in_range(p, addr, len, val))
721 return -EOPNOTSUPP;
722
723 eventfd_signal(p->eventfd, 1);
724 return 0;
725}
726
727/*
728 * This function is called as KVM is completely shutting down. We do not
729 * need to worry about locking just nuke anything we have as quickly as possible
730 */
731static void
732ioeventfd_destructor(struct kvm_io_device *this)
733{
734 struct _ioeventfd *p = to_ioeventfd(this);
735
736 ioeventfd_release(p);
737}
738
739static const struct kvm_io_device_ops ioeventfd_ops = {
740 .write = ioeventfd_write,
741 .destructor = ioeventfd_destructor,
742};
743
744/* assumes kvm->slots_lock held */
745static bool
746ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p)
747{
748 struct _ioeventfd *_p;
749
750 list_for_each_entry(_p, &kvm->ioeventfds, list)
Michael S. Tsirkin05e07f92013-04-04 13:27:21 +0300751 if (_p->bus_idx == p->bus_idx &&
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +0300752 _p->addr == p->addr &&
753 (!_p->length || !p->length ||
754 (_p->length == p->length &&
755 (_p->wildcard || p->wildcard ||
756 _p->datamatch == p->datamatch))))
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400757 return true;
758
759 return false;
760}
761
Cornelia Huck2b834512013-02-28 12:33:20 +0100762static enum kvm_bus ioeventfd_bus_from_flags(__u32 flags)
763{
764 if (flags & KVM_IOEVENTFD_FLAG_PIO)
765 return KVM_PIO_BUS;
766 if (flags & KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY)
767 return KVM_VIRTIO_CCW_NOTIFY_BUS;
768 return KVM_MMIO_BUS;
769}
770
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400771static int
772kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
773{
Cornelia Huck2b834512013-02-28 12:33:20 +0100774 enum kvm_bus bus_idx;
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400775 struct _ioeventfd *p;
776 struct eventfd_ctx *eventfd;
777 int ret;
778
Cornelia Huck2b834512013-02-28 12:33:20 +0100779 bus_idx = ioeventfd_bus_from_flags(args->flags);
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +0300780 /* must be natural-word sized, or 0 to ignore length */
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400781 switch (args->len) {
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +0300782 case 0:
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400783 case 1:
784 case 2:
785 case 4:
786 case 8:
787 break;
788 default:
789 return -EINVAL;
790 }
791
792 /* check for range overflow */
793 if (args->addr + args->len < args->addr)
794 return -EINVAL;
795
796 /* check for extra flags that we don't understand */
797 if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK)
798 return -EINVAL;
799
Michael S. Tsirkinf848a5a2014-03-31 21:50:38 +0300800 /* ioeventfd with no length can't be combined with DATAMATCH */
801 if (!args->len &&
802 args->flags & (KVM_IOEVENTFD_FLAG_PIO |
803 KVM_IOEVENTFD_FLAG_DATAMATCH))
804 return -EINVAL;
805
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400806 eventfd = eventfd_ctx_fdget(args->fd);
807 if (IS_ERR(eventfd))
808 return PTR_ERR(eventfd);
809
810 p = kzalloc(sizeof(*p), GFP_KERNEL);
811 if (!p) {
812 ret = -ENOMEM;
813 goto fail;
814 }
815
816 INIT_LIST_HEAD(&p->list);
817 p->addr = args->addr;
Michael S. Tsirkin05e07f92013-04-04 13:27:21 +0300818 p->bus_idx = bus_idx;
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400819 p->length = args->len;
820 p->eventfd = eventfd;
821
822 /* The datamatch feature is optional, otherwise this is a wildcard */
823 if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH)
824 p->datamatch = args->datamatch;
825 else
826 p->wildcard = true;
827
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200828 mutex_lock(&kvm->slots_lock);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400829
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300830 /* Verify that there isn't a match already */
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400831 if (ioeventfd_check_collision(kvm, p)) {
832 ret = -EEXIST;
833 goto unlock_fail;
834 }
835
836 kvm_iodevice_init(&p->dev, &ioeventfd_ops);
837
Sasha Levin743eeb02011-07-27 16:00:48 +0300838 ret = kvm_io_bus_register_dev(kvm, bus_idx, p->addr, p->length,
839 &p->dev);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400840 if (ret < 0)
841 goto unlock_fail;
842
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +0300843 /* When length is ignored, MMIO is also put on a separate bus, for
844 * faster lookups.
845 */
846 if (!args->len && !(args->flags & KVM_IOEVENTFD_FLAG_PIO)) {
847 ret = kvm_io_bus_register_dev(kvm, KVM_FAST_MMIO_BUS,
848 p->addr, 0, &p->dev);
849 if (ret < 0)
850 goto register_fail;
851 }
852
Amos Kong6ea34c92013-05-25 06:44:15 +0800853 kvm->buses[bus_idx]->ioeventfd_count++;
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400854 list_add_tail(&p->list, &kvm->ioeventfds);
855
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200856 mutex_unlock(&kvm->slots_lock);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400857
858 return 0;
859
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +0300860register_fail:
861 kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400862unlock_fail:
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200863 mutex_unlock(&kvm->slots_lock);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400864
865fail:
866 kfree(p);
867 eventfd_ctx_put(eventfd);
868
869 return ret;
870}
871
872static int
873kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
874{
Cornelia Huck2b834512013-02-28 12:33:20 +0100875 enum kvm_bus bus_idx;
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400876 struct _ioeventfd *p, *tmp;
877 struct eventfd_ctx *eventfd;
878 int ret = -ENOENT;
879
Cornelia Huck2b834512013-02-28 12:33:20 +0100880 bus_idx = ioeventfd_bus_from_flags(args->flags);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400881 eventfd = eventfd_ctx_fdget(args->fd);
882 if (IS_ERR(eventfd))
883 return PTR_ERR(eventfd);
884
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200885 mutex_lock(&kvm->slots_lock);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400886
887 list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) {
888 bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH);
889
Michael S. Tsirkin05e07f92013-04-04 13:27:21 +0300890 if (p->bus_idx != bus_idx ||
891 p->eventfd != eventfd ||
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400892 p->addr != args->addr ||
893 p->length != args->len ||
894 p->wildcard != wildcard)
895 continue;
896
897 if (!p->wildcard && p->datamatch != args->datamatch)
898 continue;
899
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200900 kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev);
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +0300901 if (!p->length) {
902 kvm_io_bus_unregister_dev(kvm, KVM_FAST_MMIO_BUS,
903 &p->dev);
904 }
Amos Kong6ea34c92013-05-25 06:44:15 +0800905 kvm->buses[bus_idx]->ioeventfd_count--;
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400906 ioeventfd_release(p);
907 ret = 0;
908 break;
909 }
910
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200911 mutex_unlock(&kvm->slots_lock);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400912
913 eventfd_ctx_put(eventfd);
914
915 return ret;
916}
917
918int
919kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
920{
921 if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN)
922 return kvm_deassign_ioeventfd(kvm, args);
923
924 return kvm_assign_ioeventfd(kvm, args);
925}