]>
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. | |
5 | * | |
6 | * Author: | |
7 | * Gregory Haskins <ghaskins@novell.com> | |
8 | * | |
9 | * This file is free software; you can redistribute it and/or modify | |
10 | * it under the terms of version 2 of the GNU General Public License | |
11 | * as published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software Foundation, | |
20 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. | |
21 | */ | |
22 | ||
23 | #include <linux/kvm_host.h> | |
24 | #include <linux/workqueue.h> | |
25 | #include <linux/syscalls.h> | |
26 | #include <linux/wait.h> | |
27 | #include <linux/poll.h> | |
28 | #include <linux/file.h> | |
29 | #include <linux/list.h> | |
30 | #include <linux/eventfd.h> | |
31 | ||
32 | /* | |
33 | * -------------------------------------------------------------------- | |
34 | * irqfd: Allows an fd to be used to inject an interrupt to the guest | |
35 | * | |
36 | * Credit goes to Avi Kivity for the original idea. | |
37 | * -------------------------------------------------------------------- | |
38 | */ | |
39 | ||
40 | struct _irqfd { | |
41 | struct kvm *kvm; | |
42 | struct eventfd_ctx *eventfd; | |
43 | int gsi; | |
44 | struct list_head list; | |
45 | poll_table pt; | |
46 | wait_queue_head_t *wqh; | |
47 | wait_queue_t wait; | |
48 | struct work_struct inject; | |
49 | struct work_struct shutdown; | |
50 | }; | |
51 | ||
52 | static struct workqueue_struct *irqfd_cleanup_wq; | |
53 | ||
54 | static void | |
55 | irqfd_inject(struct work_struct *work) | |
56 | { | |
57 | struct _irqfd *irqfd = container_of(work, struct _irqfd, inject); | |
58 | struct kvm *kvm = irqfd->kvm; | |
59 | ||
fa40a821 | 60 | mutex_lock(&kvm->irq_lock); |
721eecbf GH |
61 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1); |
62 | kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0); | |
fa40a821 | 63 | mutex_unlock(&kvm->irq_lock); |
721eecbf GH |
64 | } |
65 | ||
66 | /* | |
67 | * Race-free decouple logic (ordering is critical) | |
68 | */ | |
69 | static void | |
70 | irqfd_shutdown(struct work_struct *work) | |
71 | { | |
72 | struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown); | |
73 | ||
74 | /* | |
75 | * Synchronize with the wait-queue and unhook ourselves to prevent | |
76 | * further events. | |
77 | */ | |
78 | remove_wait_queue(irqfd->wqh, &irqfd->wait); | |
79 | ||
80 | /* | |
81 | * We know no new events will be scheduled at this point, so block | |
82 | * until all previously outstanding events have completed | |
83 | */ | |
84 | flush_work(&irqfd->inject); | |
85 | ||
86 | /* | |
87 | * It is now safe to release the object's resources | |
88 | */ | |
89 | eventfd_ctx_put(irqfd->eventfd); | |
90 | kfree(irqfd); | |
91 | } | |
92 | ||
93 | ||
94 | /* assumes kvm->irqfds.lock is held */ | |
95 | static bool | |
96 | irqfd_is_active(struct _irqfd *irqfd) | |
97 | { | |
98 | return list_empty(&irqfd->list) ? false : true; | |
99 | } | |
100 | ||
101 | /* | |
102 | * Mark the irqfd as inactive and schedule it for removal | |
103 | * | |
104 | * assumes kvm->irqfds.lock is held | |
105 | */ | |
106 | static void | |
107 | irqfd_deactivate(struct _irqfd *irqfd) | |
108 | { | |
109 | BUG_ON(!irqfd_is_active(irqfd)); | |
110 | ||
111 | list_del_init(&irqfd->list); | |
112 | ||
113 | queue_work(irqfd_cleanup_wq, &irqfd->shutdown); | |
114 | } | |
115 | ||
116 | /* | |
117 | * Called with wqh->lock held and interrupts disabled | |
118 | */ | |
119 | static int | |
120 | irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
121 | { | |
122 | struct _irqfd *irqfd = container_of(wait, struct _irqfd, wait); | |
123 | unsigned long flags = (unsigned long)key; | |
124 | ||
125 | if (flags & POLLIN) | |
126 | /* An event has been signaled, inject an interrupt */ | |
127 | schedule_work(&irqfd->inject); | |
128 | ||
129 | if (flags & POLLHUP) { | |
130 | /* The eventfd is closing, detach from KVM */ | |
131 | struct kvm *kvm = irqfd->kvm; | |
132 | unsigned long flags; | |
133 | ||
134 | spin_lock_irqsave(&kvm->irqfds.lock, flags); | |
135 | ||
136 | /* | |
137 | * We must check if someone deactivated the irqfd before | |
138 | * we could acquire the irqfds.lock since the item is | |
139 | * deactivated from the KVM side before it is unhooked from | |
140 | * the wait-queue. If it is already deactivated, we can | |
141 | * simply return knowing the other side will cleanup for us. | |
142 | * We cannot race against the irqfd going away since the | |
143 | * other side is required to acquire wqh->lock, which we hold | |
144 | */ | |
145 | if (irqfd_is_active(irqfd)) | |
146 | irqfd_deactivate(irqfd); | |
147 | ||
148 | spin_unlock_irqrestore(&kvm->irqfds.lock, flags); | |
149 | } | |
150 | ||
151 | return 0; | |
152 | } | |
153 | ||
154 | static void | |
155 | irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh, | |
156 | poll_table *pt) | |
157 | { | |
158 | struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt); | |
159 | ||
160 | irqfd->wqh = wqh; | |
161 | add_wait_queue(wqh, &irqfd->wait); | |
162 | } | |
163 | ||
164 | static int | |
165 | kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi) | |
166 | { | |
167 | struct _irqfd *irqfd; | |
168 | struct file *file = NULL; | |
169 | struct eventfd_ctx *eventfd = NULL; | |
170 | int ret; | |
171 | unsigned int events; | |
172 | ||
173 | irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL); | |
174 | if (!irqfd) | |
175 | return -ENOMEM; | |
176 | ||
177 | irqfd->kvm = kvm; | |
178 | irqfd->gsi = gsi; | |
179 | INIT_LIST_HEAD(&irqfd->list); | |
180 | INIT_WORK(&irqfd->inject, irqfd_inject); | |
181 | INIT_WORK(&irqfd->shutdown, irqfd_shutdown); | |
182 | ||
183 | file = eventfd_fget(fd); | |
184 | if (IS_ERR(file)) { | |
185 | ret = PTR_ERR(file); | |
186 | goto fail; | |
187 | } | |
188 | ||
189 | eventfd = eventfd_ctx_fileget(file); | |
190 | if (IS_ERR(eventfd)) { | |
191 | ret = PTR_ERR(eventfd); | |
192 | goto fail; | |
193 | } | |
194 | ||
195 | irqfd->eventfd = eventfd; | |
196 | ||
197 | /* | |
198 | * Install our own custom wake-up handling so we are notified via | |
199 | * a callback whenever someone signals the underlying eventfd | |
200 | */ | |
201 | init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup); | |
202 | init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc); | |
203 | ||
204 | events = file->f_op->poll(file, &irqfd->pt); | |
205 | ||
206 | spin_lock_irq(&kvm->irqfds.lock); | |
207 | list_add_tail(&irqfd->list, &kvm->irqfds.items); | |
208 | spin_unlock_irq(&kvm->irqfds.lock); | |
209 | ||
210 | /* | |
211 | * Check if there was an event already pending on the eventfd | |
212 | * before we registered, and trigger it as if we didn't miss it. | |
213 | */ | |
214 | if (events & POLLIN) | |
215 | schedule_work(&irqfd->inject); | |
216 | ||
217 | /* | |
218 | * do not drop the file until the irqfd is fully initialized, otherwise | |
219 | * we might race against the POLLHUP | |
220 | */ | |
221 | fput(file); | |
222 | ||
223 | return 0; | |
224 | ||
225 | fail: | |
226 | if (eventfd && !IS_ERR(eventfd)) | |
227 | eventfd_ctx_put(eventfd); | |
228 | ||
229 | if (file && !IS_ERR(file)) | |
230 | fput(file); | |
231 | ||
232 | kfree(irqfd); | |
233 | return ret; | |
234 | } | |
235 | ||
236 | void | |
237 | kvm_irqfd_init(struct kvm *kvm) | |
238 | { | |
239 | spin_lock_init(&kvm->irqfds.lock); | |
240 | INIT_LIST_HEAD(&kvm->irqfds.items); | |
241 | } | |
242 | ||
243 | /* | |
244 | * shutdown any irqfd's that match fd+gsi | |
245 | */ | |
246 | static int | |
247 | kvm_irqfd_deassign(struct kvm *kvm, int fd, int gsi) | |
248 | { | |
249 | struct _irqfd *irqfd, *tmp; | |
250 | struct eventfd_ctx *eventfd; | |
251 | ||
252 | eventfd = eventfd_ctx_fdget(fd); | |
253 | if (IS_ERR(eventfd)) | |
254 | return PTR_ERR(eventfd); | |
255 | ||
256 | spin_lock_irq(&kvm->irqfds.lock); | |
257 | ||
258 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) { | |
259 | if (irqfd->eventfd == eventfd && irqfd->gsi == gsi) | |
260 | irqfd_deactivate(irqfd); | |
261 | } | |
262 | ||
263 | spin_unlock_irq(&kvm->irqfds.lock); | |
264 | eventfd_ctx_put(eventfd); | |
265 | ||
266 | /* | |
267 | * Block until we know all outstanding shutdown jobs have completed | |
268 | * so that we guarantee there will not be any more interrupts on this | |
269 | * gsi once this deassign function returns. | |
270 | */ | |
271 | flush_workqueue(irqfd_cleanup_wq); | |
272 | ||
273 | return 0; | |
274 | } | |
275 | ||
276 | int | |
277 | kvm_irqfd(struct kvm *kvm, int fd, int gsi, int flags) | |
278 | { | |
279 | if (flags & KVM_IRQFD_FLAG_DEASSIGN) | |
280 | return kvm_irqfd_deassign(kvm, fd, gsi); | |
281 | ||
282 | return kvm_irqfd_assign(kvm, fd, gsi); | |
283 | } | |
284 | ||
285 | /* | |
286 | * This function is called as the kvm VM fd is being released. Shutdown all | |
287 | * irqfds that still remain open | |
288 | */ | |
289 | void | |
290 | kvm_irqfd_release(struct kvm *kvm) | |
291 | { | |
292 | struct _irqfd *irqfd, *tmp; | |
293 | ||
294 | spin_lock_irq(&kvm->irqfds.lock); | |
295 | ||
296 | list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) | |
297 | irqfd_deactivate(irqfd); | |
298 | ||
299 | spin_unlock_irq(&kvm->irqfds.lock); | |
300 | ||
301 | /* | |
302 | * Block until we know all outstanding shutdown jobs have completed | |
303 | * since we do not take a kvm* reference. | |
304 | */ | |
305 | flush_workqueue(irqfd_cleanup_wq); | |
306 | ||
307 | } | |
308 | ||
309 | /* | |
310 | * create a host-wide workqueue for issuing deferred shutdown requests | |
311 | * aggregated from all vm* instances. We need our own isolated single-thread | |
312 | * queue to prevent deadlock against flushing the normal work-queue. | |
313 | */ | |
314 | static int __init irqfd_module_init(void) | |
315 | { | |
316 | irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup"); | |
317 | if (!irqfd_cleanup_wq) | |
318 | return -ENOMEM; | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
323 | static void __exit irqfd_module_exit(void) | |
324 | { | |
325 | destroy_workqueue(irqfd_cleanup_wq); | |
326 | } | |
327 | ||
328 | module_init(irqfd_module_init); | |
329 | module_exit(irqfd_module_exit); |