2 * kvm eventfd support - use eventfd objects to signal various KVM events
4 * Copyright 2009 Novell. All Rights Reserved.
5 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
8 * Gregory Haskins <ghaskins@novell.com>
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.
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.
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.
24 #include <linux/kvm_host.h>
25 #include <linux/kvm.h>
26 #include <linux/kvm_irqfd.h>
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>
34 #include <linux/kernel.h>
35 #include <linux/srcu.h>
36 #include <linux/slab.h>
37 #include <linux/seqlock.h>
38 #include <linux/irqbypass.h>
39 #include <trace/events/kvm.h>
41 #include <kvm/iodev.h>
43 #ifdef CONFIG_HAVE_KVM_IRQFD
45 static struct workqueue_struct
*irqfd_cleanup_wq
;
48 irqfd_inject(struct work_struct
*work
)
50 struct kvm_kernel_irqfd
*irqfd
=
51 container_of(work
, struct kvm_kernel_irqfd
, inject
);
52 struct kvm
*kvm
= irqfd
->kvm
;
54 if (!irqfd
->resampler
) {
55 kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
, irqfd
->gsi
, 1,
57 kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
, irqfd
->gsi
, 0,
60 kvm_set_irq(kvm
, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
61 irqfd
->gsi
, 1, false);
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.
70 irqfd_resampler_ack(struct kvm_irq_ack_notifier
*kian
)
72 struct kvm_kernel_irqfd_resampler
*resampler
;
74 struct kvm_kernel_irqfd
*irqfd
;
77 resampler
= container_of(kian
,
78 struct kvm_kernel_irqfd_resampler
, notifier
);
81 kvm_set_irq(kvm
, KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
82 resampler
->notifier
.gsi
, 0, false);
84 idx
= srcu_read_lock(&kvm
->irq_srcu
);
86 list_for_each_entry_rcu(irqfd
, &resampler
->list
, resampler_link
)
87 eventfd_signal(irqfd
->resamplefd
, 1);
89 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
93 irqfd_resampler_shutdown(struct kvm_kernel_irqfd
*irqfd
)
95 struct kvm_kernel_irqfd_resampler
*resampler
= irqfd
->resampler
;
96 struct kvm
*kvm
= resampler
->kvm
;
98 mutex_lock(&kvm
->irqfds
.resampler_lock
);
100 list_del_rcu(&irqfd
->resampler_link
);
101 synchronize_srcu(&kvm
->irq_srcu
);
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
,
107 resampler
->notifier
.gsi
, 0, false);
111 mutex_unlock(&kvm
->irqfds
.resampler_lock
);
115 * Race-free decouple logic (ordering is critical)
118 irqfd_shutdown(struct work_struct
*work
)
120 struct kvm_kernel_irqfd
*irqfd
=
121 container_of(work
, struct kvm_kernel_irqfd
, shutdown
);
125 * Synchronize with the wait-queue and unhook ourselves to prevent
128 eventfd_ctx_remove_wait_queue(irqfd
->eventfd
, &irqfd
->wait
, &cnt
);
131 * We know no new events will be scheduled at this point, so block
132 * until all previously outstanding events have completed
134 flush_work(&irqfd
->inject
);
136 if (irqfd
->resampler
) {
137 irqfd_resampler_shutdown(irqfd
);
138 eventfd_ctx_put(irqfd
->resamplefd
);
142 * It is now safe to release the object's resources
144 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
145 irq_bypass_unregister_consumer(&irqfd
->consumer
);
147 eventfd_ctx_put(irqfd
->eventfd
);
152 /* assumes kvm->irqfds.lock is held */
154 irqfd_is_active(struct kvm_kernel_irqfd
*irqfd
)
156 return list_empty(&irqfd
->list
) ? false : true;
160 * Mark the irqfd as inactive and schedule it for removal
162 * assumes kvm->irqfds.lock is held
165 irqfd_deactivate(struct kvm_kernel_irqfd
*irqfd
)
167 BUG_ON(!irqfd_is_active(irqfd
));
169 list_del_init(&irqfd
->list
);
171 queue_work(irqfd_cleanup_wq
, &irqfd
->shutdown
);
174 int __attribute__((weak
)) kvm_arch_set_irq_inatomic(
175 struct kvm_kernel_irq_routing_entry
*irq
,
176 struct kvm
*kvm
, int irq_source_id
,
184 * Called with wqh->lock held and interrupts disabled
187 irqfd_wakeup(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
189 struct kvm_kernel_irqfd
*irqfd
=
190 container_of(wait
, struct kvm_kernel_irqfd
, wait
);
191 unsigned long flags
= (unsigned long)key
;
192 struct kvm_kernel_irq_routing_entry irq
;
193 struct kvm
*kvm
= irqfd
->kvm
;
197 if (flags
& POLLIN
) {
198 idx
= srcu_read_lock(&kvm
->irq_srcu
);
200 seq
= read_seqcount_begin(&irqfd
->irq_entry_sc
);
201 irq
= irqfd
->irq_entry
;
202 } while (read_seqcount_retry(&irqfd
->irq_entry_sc
, seq
));
203 /* An event has been signaled, inject an interrupt */
204 if (kvm_arch_set_irq_inatomic(&irq
, kvm
,
205 KVM_USERSPACE_IRQ_SOURCE_ID
, 1,
206 false) == -EWOULDBLOCK
)
207 schedule_work(&irqfd
->inject
);
208 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
211 if (flags
& POLLHUP
) {
212 /* The eventfd is closing, detach from KVM */
215 spin_lock_irqsave(&kvm
->irqfds
.lock
, flags
);
218 * We must check if someone deactivated the irqfd before
219 * we could acquire the irqfds.lock since the item is
220 * deactivated from the KVM side before it is unhooked from
221 * the wait-queue. If it is already deactivated, we can
222 * simply return knowing the other side will cleanup for us.
223 * We cannot race against the irqfd going away since the
224 * other side is required to acquire wqh->lock, which we hold
226 if (irqfd_is_active(irqfd
))
227 irqfd_deactivate(irqfd
);
229 spin_unlock_irqrestore(&kvm
->irqfds
.lock
, flags
);
236 irqfd_ptable_queue_proc(struct file
*file
, wait_queue_head_t
*wqh
,
239 struct kvm_kernel_irqfd
*irqfd
=
240 container_of(pt
, struct kvm_kernel_irqfd
, pt
);
241 add_wait_queue(wqh
, &irqfd
->wait
);
244 /* Must be called under irqfds.lock */
245 static void irqfd_update(struct kvm
*kvm
, struct kvm_kernel_irqfd
*irqfd
)
247 struct kvm_kernel_irq_routing_entry
*e
;
248 struct kvm_kernel_irq_routing_entry entries
[KVM_NR_IRQCHIPS
];
251 n_entries
= kvm_irq_map_gsi(kvm
, entries
, irqfd
->gsi
);
253 write_seqcount_begin(&irqfd
->irq_entry_sc
);
257 irqfd
->irq_entry
= *e
;
259 irqfd
->irq_entry
.type
= 0;
261 write_seqcount_end(&irqfd
->irq_entry_sc
);
264 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
265 void __attribute__((weak
)) kvm_arch_irq_bypass_stop(
266 struct irq_bypass_consumer
*cons
)
270 void __attribute__((weak
)) kvm_arch_irq_bypass_start(
271 struct irq_bypass_consumer
*cons
)
275 int __attribute__((weak
)) kvm_arch_update_irqfd_routing(
276 struct kvm
*kvm
, unsigned int host_irq
,
277 uint32_t guest_irq
, bool set
)
284 kvm_irqfd_assign(struct kvm
*kvm
, struct kvm_irqfd
*args
)
286 struct kvm_kernel_irqfd
*irqfd
, *tmp
;
288 struct eventfd_ctx
*eventfd
= NULL
, *resamplefd
= NULL
;
293 if (!kvm_arch_intc_initialized(kvm
))
296 irqfd
= kzalloc(sizeof(*irqfd
), GFP_KERNEL
);
301 irqfd
->gsi
= args
->gsi
;
302 INIT_LIST_HEAD(&irqfd
->list
);
303 INIT_WORK(&irqfd
->inject
, irqfd_inject
);
304 INIT_WORK(&irqfd
->shutdown
, irqfd_shutdown
);
305 seqcount_init(&irqfd
->irq_entry_sc
);
313 eventfd
= eventfd_ctx_fileget(f
.file
);
314 if (IS_ERR(eventfd
)) {
315 ret
= PTR_ERR(eventfd
);
319 irqfd
->eventfd
= eventfd
;
321 if (args
->flags
& KVM_IRQFD_FLAG_RESAMPLE
) {
322 struct kvm_kernel_irqfd_resampler
*resampler
;
324 resamplefd
= eventfd_ctx_fdget(args
->resamplefd
);
325 if (IS_ERR(resamplefd
)) {
326 ret
= PTR_ERR(resamplefd
);
330 irqfd
->resamplefd
= resamplefd
;
331 INIT_LIST_HEAD(&irqfd
->resampler_link
);
333 mutex_lock(&kvm
->irqfds
.resampler_lock
);
335 list_for_each_entry(resampler
,
336 &kvm
->irqfds
.resampler_list
, link
) {
337 if (resampler
->notifier
.gsi
== irqfd
->gsi
) {
338 irqfd
->resampler
= resampler
;
343 if (!irqfd
->resampler
) {
344 resampler
= kzalloc(sizeof(*resampler
), GFP_KERNEL
);
347 mutex_unlock(&kvm
->irqfds
.resampler_lock
);
351 resampler
->kvm
= kvm
;
352 INIT_LIST_HEAD(&resampler
->list
);
353 resampler
->notifier
.gsi
= irqfd
->gsi
;
354 resampler
->notifier
.irq_acked
= irqfd_resampler_ack
;
355 INIT_LIST_HEAD(&resampler
->link
);
357 list_add(&resampler
->link
, &kvm
->irqfds
.resampler_list
);
358 kvm_register_irq_ack_notifier(kvm
,
359 &resampler
->notifier
);
360 irqfd
->resampler
= resampler
;
363 list_add_rcu(&irqfd
->resampler_link
, &irqfd
->resampler
->list
);
364 synchronize_srcu(&kvm
->irq_srcu
);
366 mutex_unlock(&kvm
->irqfds
.resampler_lock
);
370 * Install our own custom wake-up handling so we are notified via
371 * a callback whenever someone signals the underlying eventfd
373 init_waitqueue_func_entry(&irqfd
->wait
, irqfd_wakeup
);
374 init_poll_funcptr(&irqfd
->pt
, irqfd_ptable_queue_proc
);
376 spin_lock_irq(&kvm
->irqfds
.lock
);
379 list_for_each_entry(tmp
, &kvm
->irqfds
.items
, list
) {
380 if (irqfd
->eventfd
!= tmp
->eventfd
)
382 /* This fd is used for another irq already. */
384 spin_unlock_irq(&kvm
->irqfds
.lock
);
388 idx
= srcu_read_lock(&kvm
->irq_srcu
);
389 irqfd_update(kvm
, irqfd
);
390 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
392 list_add_tail(&irqfd
->list
, &kvm
->irqfds
.items
);
394 spin_unlock_irq(&kvm
->irqfds
.lock
);
397 * Check if there was an event already pending on the eventfd
398 * before we registered, and trigger it as if we didn't miss it.
400 events
= f
.file
->f_op
->poll(f
.file
, &irqfd
->pt
);
403 schedule_work(&irqfd
->inject
);
406 * do not drop the file until the irqfd is fully initialized, otherwise
407 * we might race against the POLLHUP
410 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
411 if (kvm_arch_has_irq_bypass()) {
412 irqfd
->consumer
.token
= (void *)irqfd
->eventfd
;
413 irqfd
->consumer
.add_producer
= kvm_arch_irq_bypass_add_producer
;
414 irqfd
->consumer
.del_producer
= kvm_arch_irq_bypass_del_producer
;
415 irqfd
->consumer
.stop
= kvm_arch_irq_bypass_stop
;
416 irqfd
->consumer
.start
= kvm_arch_irq_bypass_start
;
417 ret
= irq_bypass_register_consumer(&irqfd
->consumer
);
419 pr_info("irq bypass consumer (token %p) registration fails: %d\n",
420 irqfd
->consumer
.token
, ret
);
427 if (irqfd
->resampler
)
428 irqfd_resampler_shutdown(irqfd
);
430 if (resamplefd
&& !IS_ERR(resamplefd
))
431 eventfd_ctx_put(resamplefd
);
433 if (eventfd
&& !IS_ERR(eventfd
))
434 eventfd_ctx_put(eventfd
);
443 bool kvm_irq_has_notifier(struct kvm
*kvm
, unsigned irqchip
, unsigned pin
)
445 struct kvm_irq_ack_notifier
*kian
;
448 idx
= srcu_read_lock(&kvm
->irq_srcu
);
449 gsi
= kvm_irq_map_chip_pin(kvm
, irqchip
, pin
);
451 hlist_for_each_entry_rcu(kian
, &kvm
->irq_ack_notifier_list
,
453 if (kian
->gsi
== gsi
) {
454 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
458 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
462 EXPORT_SYMBOL_GPL(kvm_irq_has_notifier
);
464 void kvm_notify_acked_gsi(struct kvm
*kvm
, int gsi
)
466 struct kvm_irq_ack_notifier
*kian
;
468 hlist_for_each_entry_rcu(kian
, &kvm
->irq_ack_notifier_list
,
470 if (kian
->gsi
== gsi
)
471 kian
->irq_acked(kian
);
474 void kvm_notify_acked_irq(struct kvm
*kvm
, unsigned irqchip
, unsigned pin
)
478 trace_kvm_ack_irq(irqchip
, pin
);
480 idx
= srcu_read_lock(&kvm
->irq_srcu
);
481 gsi
= kvm_irq_map_chip_pin(kvm
, irqchip
, pin
);
483 kvm_notify_acked_gsi(kvm
, gsi
);
484 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
487 void kvm_register_irq_ack_notifier(struct kvm
*kvm
,
488 struct kvm_irq_ack_notifier
*kian
)
490 mutex_lock(&kvm
->irq_lock
);
491 hlist_add_head_rcu(&kian
->link
, &kvm
->irq_ack_notifier_list
);
492 mutex_unlock(&kvm
->irq_lock
);
493 kvm_arch_post_irq_ack_notifier_list_update(kvm
);
496 void kvm_unregister_irq_ack_notifier(struct kvm
*kvm
,
497 struct kvm_irq_ack_notifier
*kian
)
499 mutex_lock(&kvm
->irq_lock
);
500 hlist_del_init_rcu(&kian
->link
);
501 mutex_unlock(&kvm
->irq_lock
);
502 synchronize_srcu(&kvm
->irq_srcu
);
503 kvm_arch_post_irq_ack_notifier_list_update(kvm
);
508 kvm_eventfd_init(struct kvm
*kvm
)
510 #ifdef CONFIG_HAVE_KVM_IRQFD
511 spin_lock_init(&kvm
->irqfds
.lock
);
512 INIT_LIST_HEAD(&kvm
->irqfds
.items
);
513 INIT_LIST_HEAD(&kvm
->irqfds
.resampler_list
);
514 mutex_init(&kvm
->irqfds
.resampler_lock
);
516 INIT_LIST_HEAD(&kvm
->ioeventfds
);
519 #ifdef CONFIG_HAVE_KVM_IRQFD
521 * shutdown any irqfd's that match fd+gsi
524 kvm_irqfd_deassign(struct kvm
*kvm
, struct kvm_irqfd
*args
)
526 struct kvm_kernel_irqfd
*irqfd
, *tmp
;
527 struct eventfd_ctx
*eventfd
;
529 eventfd
= eventfd_ctx_fdget(args
->fd
);
531 return PTR_ERR(eventfd
);
533 spin_lock_irq(&kvm
->irqfds
.lock
);
535 list_for_each_entry_safe(irqfd
, tmp
, &kvm
->irqfds
.items
, list
) {
536 if (irqfd
->eventfd
== eventfd
&& irqfd
->gsi
== args
->gsi
) {
538 * This clearing of irq_entry.type is needed for when
539 * another thread calls kvm_irq_routing_update before
540 * we flush workqueue below (we synchronize with
541 * kvm_irq_routing_update using irqfds.lock).
543 write_seqcount_begin(&irqfd
->irq_entry_sc
);
544 irqfd
->irq_entry
.type
= 0;
545 write_seqcount_end(&irqfd
->irq_entry_sc
);
546 irqfd_deactivate(irqfd
);
550 spin_unlock_irq(&kvm
->irqfds
.lock
);
551 eventfd_ctx_put(eventfd
);
554 * Block until we know all outstanding shutdown jobs have completed
555 * so that we guarantee there will not be any more interrupts on this
556 * gsi once this deassign function returns.
558 flush_workqueue(irqfd_cleanup_wq
);
564 kvm_irqfd(struct kvm
*kvm
, struct kvm_irqfd
*args
)
566 if (args
->flags
& ~(KVM_IRQFD_FLAG_DEASSIGN
| KVM_IRQFD_FLAG_RESAMPLE
))
568 if (args
->gsi
>= KVM_MAX_IRQ_ROUTES
)
571 if (args
->flags
& KVM_IRQFD_FLAG_DEASSIGN
)
572 return kvm_irqfd_deassign(kvm
, args
);
574 return kvm_irqfd_assign(kvm
, args
);
578 * This function is called as the kvm VM fd is being released. Shutdown all
579 * irqfds that still remain open
582 kvm_irqfd_release(struct kvm
*kvm
)
584 struct kvm_kernel_irqfd
*irqfd
, *tmp
;
586 spin_lock_irq(&kvm
->irqfds
.lock
);
588 list_for_each_entry_safe(irqfd
, tmp
, &kvm
->irqfds
.items
, list
)
589 irqfd_deactivate(irqfd
);
591 spin_unlock_irq(&kvm
->irqfds
.lock
);
594 * Block until we know all outstanding shutdown jobs have completed
595 * since we do not take a kvm* reference.
597 flush_workqueue(irqfd_cleanup_wq
);
602 * Take note of a change in irq routing.
603 * Caller must invoke synchronize_srcu(&kvm->irq_srcu) afterwards.
605 void kvm_irq_routing_update(struct kvm
*kvm
)
607 struct kvm_kernel_irqfd
*irqfd
;
609 spin_lock_irq(&kvm
->irqfds
.lock
);
611 list_for_each_entry(irqfd
, &kvm
->irqfds
.items
, list
) {
612 irqfd_update(kvm
, irqfd
);
614 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
615 if (irqfd
->producer
) {
616 int ret
= kvm_arch_update_irqfd_routing(
617 irqfd
->kvm
, irqfd
->producer
->irq
,
624 spin_unlock_irq(&kvm
->irqfds
.lock
);
628 * create a host-wide workqueue for issuing deferred shutdown requests
629 * aggregated from all vm* instances. We need our own isolated
630 * queue to ease flushing work items when a VM exits.
632 int kvm_irqfd_init(void)
634 irqfd_cleanup_wq
= alloc_workqueue("kvm-irqfd-cleanup", 0, 0);
635 if (!irqfd_cleanup_wq
)
641 void kvm_irqfd_exit(void)
643 destroy_workqueue(irqfd_cleanup_wq
);
648 * --------------------------------------------------------------------
649 * ioeventfd: translate a PIO/MMIO memory write to an eventfd signal.
651 * userspace can register a PIO/MMIO address with an eventfd for receiving
652 * notification when the memory has been touched.
653 * --------------------------------------------------------------------
657 struct list_head list
;
660 struct eventfd_ctx
*eventfd
;
662 struct kvm_io_device dev
;
667 static inline struct _ioeventfd
*
668 to_ioeventfd(struct kvm_io_device
*dev
)
670 return container_of(dev
, struct _ioeventfd
, dev
);
674 ioeventfd_release(struct _ioeventfd
*p
)
676 eventfd_ctx_put(p
->eventfd
);
682 ioeventfd_in_range(struct _ioeventfd
*p
, gpa_t addr
, int len
, const void *val
)
687 /* address must be precise for a hit */
691 /* length = 0 means only look at the address, so always a hit */
694 if (len
!= p
->length
)
695 /* address-range must be precise for a hit */
699 /* all else equal, wildcard is always a hit */
702 /* otherwise, we have to actually compare the data */
704 BUG_ON(!IS_ALIGNED((unsigned long)val
, len
));
723 return _val
== p
->datamatch
? true : false;
726 /* MMIO/PIO writes trigger an event if the addr/val match */
728 ioeventfd_write(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this, gpa_t addr
,
729 int len
, const void *val
)
731 struct _ioeventfd
*p
= to_ioeventfd(this);
733 if (!ioeventfd_in_range(p
, addr
, len
, val
))
736 eventfd_signal(p
->eventfd
, 1);
741 * This function is called as KVM is completely shutting down. We do not
742 * need to worry about locking just nuke anything we have as quickly as possible
745 ioeventfd_destructor(struct kvm_io_device
*this)
747 struct _ioeventfd
*p
= to_ioeventfd(this);
749 ioeventfd_release(p
);
752 static const struct kvm_io_device_ops ioeventfd_ops
= {
753 .write
= ioeventfd_write
,
754 .destructor
= ioeventfd_destructor
,
757 /* assumes kvm->slots_lock held */
759 ioeventfd_check_collision(struct kvm
*kvm
, struct _ioeventfd
*p
)
761 struct _ioeventfd
*_p
;
763 list_for_each_entry(_p
, &kvm
->ioeventfds
, list
)
764 if (_p
->bus_idx
== p
->bus_idx
&&
765 _p
->addr
== p
->addr
&&
766 (!_p
->length
|| !p
->length
||
767 (_p
->length
== p
->length
&&
768 (_p
->wildcard
|| p
->wildcard
||
769 _p
->datamatch
== p
->datamatch
))))
775 static enum kvm_bus
ioeventfd_bus_from_flags(__u32 flags
)
777 if (flags
& KVM_IOEVENTFD_FLAG_PIO
)
779 if (flags
& KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY
)
780 return KVM_VIRTIO_CCW_NOTIFY_BUS
;
784 static int kvm_assign_ioeventfd_idx(struct kvm
*kvm
,
785 enum kvm_bus bus_idx
,
786 struct kvm_ioeventfd
*args
)
789 struct eventfd_ctx
*eventfd
;
790 struct _ioeventfd
*p
;
793 eventfd
= eventfd_ctx_fdget(args
->fd
);
795 return PTR_ERR(eventfd
);
797 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
803 INIT_LIST_HEAD(&p
->list
);
804 p
->addr
= args
->addr
;
805 p
->bus_idx
= bus_idx
;
806 p
->length
= args
->len
;
807 p
->eventfd
= eventfd
;
809 /* The datamatch feature is optional, otherwise this is a wildcard */
810 if (args
->flags
& KVM_IOEVENTFD_FLAG_DATAMATCH
)
811 p
->datamatch
= args
->datamatch
;
815 mutex_lock(&kvm
->slots_lock
);
817 /* Verify that there isn't a match already */
818 if (ioeventfd_check_collision(kvm
, p
)) {
823 kvm_iodevice_init(&p
->dev
, &ioeventfd_ops
);
825 ret
= kvm_io_bus_register_dev(kvm
, bus_idx
, p
->addr
, p
->length
,
830 kvm_get_bus(kvm
, bus_idx
)->ioeventfd_count
++;
831 list_add_tail(&p
->list
, &kvm
->ioeventfds
);
833 mutex_unlock(&kvm
->slots_lock
);
838 mutex_unlock(&kvm
->slots_lock
);
842 eventfd_ctx_put(eventfd
);
848 kvm_deassign_ioeventfd_idx(struct kvm
*kvm
, enum kvm_bus bus_idx
,
849 struct kvm_ioeventfd
*args
)
851 struct _ioeventfd
*p
, *tmp
;
852 struct eventfd_ctx
*eventfd
;
853 struct kvm_io_bus
*bus
;
856 eventfd
= eventfd_ctx_fdget(args
->fd
);
858 return PTR_ERR(eventfd
);
860 mutex_lock(&kvm
->slots_lock
);
862 list_for_each_entry_safe(p
, tmp
, &kvm
->ioeventfds
, list
) {
863 bool wildcard
= !(args
->flags
& KVM_IOEVENTFD_FLAG_DATAMATCH
);
865 if (p
->bus_idx
!= bus_idx
||
866 p
->eventfd
!= eventfd
||
867 p
->addr
!= args
->addr
||
868 p
->length
!= args
->len
||
869 p
->wildcard
!= wildcard
)
872 if (!p
->wildcard
&& p
->datamatch
!= args
->datamatch
)
875 kvm_io_bus_unregister_dev(kvm
, bus_idx
, &p
->dev
);
876 bus
= kvm_get_bus(kvm
, bus_idx
);
878 bus
->ioeventfd_count
--;
879 ioeventfd_release(p
);
884 mutex_unlock(&kvm
->slots_lock
);
886 eventfd_ctx_put(eventfd
);
891 static int kvm_deassign_ioeventfd(struct kvm
*kvm
, struct kvm_ioeventfd
*args
)
893 enum kvm_bus bus_idx
= ioeventfd_bus_from_flags(args
->flags
);
894 int ret
= kvm_deassign_ioeventfd_idx(kvm
, bus_idx
, args
);
896 if (!args
->len
&& bus_idx
== KVM_MMIO_BUS
)
897 kvm_deassign_ioeventfd_idx(kvm
, KVM_FAST_MMIO_BUS
, args
);
903 kvm_assign_ioeventfd(struct kvm
*kvm
, struct kvm_ioeventfd
*args
)
905 enum kvm_bus bus_idx
;
908 bus_idx
= ioeventfd_bus_from_flags(args
->flags
);
909 /* must be natural-word sized, or 0 to ignore length */
921 /* check for range overflow */
922 if (args
->addr
+ args
->len
< args
->addr
)
925 /* check for extra flags that we don't understand */
926 if (args
->flags
& ~KVM_IOEVENTFD_VALID_FLAG_MASK
)
929 /* ioeventfd with no length can't be combined with DATAMATCH */
930 if (!args
->len
&& (args
->flags
& KVM_IOEVENTFD_FLAG_DATAMATCH
))
933 ret
= kvm_assign_ioeventfd_idx(kvm
, bus_idx
, args
);
937 /* When length is ignored, MMIO is also put on a separate bus, for
940 if (!args
->len
&& bus_idx
== KVM_MMIO_BUS
) {
941 ret
= kvm_assign_ioeventfd_idx(kvm
, KVM_FAST_MMIO_BUS
, args
);
949 kvm_deassign_ioeventfd_idx(kvm
, bus_idx
, args
);
955 kvm_ioeventfd(struct kvm
*kvm
, struct kvm_ioeventfd
*args
)
957 if (args
->flags
& KVM_IOEVENTFD_FLAG_DEASSIGN
)
958 return kvm_deassign_ioeventfd(kvm
, args
);
960 return kvm_assign_ioeventfd(kvm
, args
);