blob: b7c78a403dc209dc12b72c2b769352cd4400187b [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
Hollis Blancharde2174022007-12-03 15:30:24 -060018#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080019
Avi Kivityedf88412007-12-16 11:02:48 +020020#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080021#include <linux/kvm.h>
22#include <linux/module.h>
23#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080024#include <linux/percpu.h>
25#include <linux/gfp.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Avi Kivity59ae6c62007-02-12 00:54:48 -080033#include <linux/sysdev.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080046
Avi Kivitye4956062007-06-28 14:15:57 -040047#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040048#include <asm/io.h>
49#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020050#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080051
Laurent Vivier5f94c172008-05-30 16:05:54 +020052#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53#include "coalesced_mmio.h"
54#endif
55
Xiantao Zhang8a98f662008-10-06 13:47:38 +080056#ifdef KVM_CAP_DEVICE_ASSIGNMENT
57#include <linux/pci.h>
58#include <linux/interrupt.h>
59#include "irq.h"
60#endif
61
Marcelo Tosatti229456f2009-06-17 09:22:14 -030062#define CREATE_TRACE_POINTS
63#include <trace/events/kvm.h>
64
Avi Kivity6aa8b732006-12-10 02:21:36 -080065MODULE_AUTHOR("Qumranet");
66MODULE_LICENSE("GPL");
67
Marcelo Tosattifa40a822009-06-04 15:08:24 -030068/*
69 * Ordering of locks:
70 *
Michael S. Tsirkin22fc0292009-06-29 22:24:45 +030071 * kvm->slots_lock --> kvm->lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030072 */
73
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080074DEFINE_SPINLOCK(kvm_lock);
75LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080076
Rusty Russell7f59f492008-12-07 21:25:45 +103077static cpumask_var_t cpus_hardware_enabled;
Avi Kivity1b6c0162007-05-24 13:03:52 +030078
Rusty Russellc16f8622007-07-30 21:12:19 +100079struct kmem_cache *kvm_vcpu_cache;
80EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030081
Avi Kivity15ad7142007-07-11 18:17:21 +030082static __read_mostly struct preempt_ops kvm_preempt_ops;
83
Hollis Blanchard76f7c872008-04-15 16:05:42 -050084struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080085
Avi Kivitybccf2152007-02-21 18:04:26 +020086static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 unsigned long arg);
88
Hannes Edere8ba5d32008-11-28 17:02:42 +010089static bool kvm_rebooting;
Avi Kivity4ecac3f2008-05-13 13:23:38 +030090
Marcelo Tosatti54dee992009-06-11 12:07:44 -030091static bool largepages_enabled = true;
92
Xiantao Zhang8a98f662008-10-06 13:47:38 +080093#ifdef KVM_CAP_DEVICE_ASSIGNMENT
94static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
95 int assigned_dev_id)
96{
97 struct list_head *ptr;
98 struct kvm_assigned_dev_kernel *match;
99
100 list_for_each(ptr, head) {
101 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
102 if (match->assigned_dev_id == assigned_dev_id)
103 return match;
104 }
105 return NULL;
106}
107
Sheng Yang2350bd12009-02-25 17:22:27 +0800108static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
109 *assigned_dev, int irq)
110{
111 int i, index;
112 struct msix_entry *host_msix_entries;
113
114 host_msix_entries = assigned_dev->host_msix_entries;
115
116 index = -1;
117 for (i = 0; i < assigned_dev->entries_nr; i++)
118 if (irq == host_msix_entries[i].vector) {
119 index = i;
120 break;
121 }
122 if (index < 0) {
123 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
124 return 0;
125 }
126
127 return index;
128}
129
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800130static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
131{
132 struct kvm_assigned_dev_kernel *assigned_dev;
Sheng Yang2350bd12009-02-25 17:22:27 +0800133 struct kvm *kvm;
Sheng Yang968a6342009-04-30 10:58:42 +0800134 int i;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800135
136 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
137 interrupt_work);
Sheng Yang2350bd12009-02-25 17:22:27 +0800138 kvm = assigned_dev->kvm;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800139
Marcelo Tosattifa40a822009-06-04 15:08:24 -0300140 mutex_lock(&kvm->irq_lock);
Marcelo Tosatti547de292009-05-07 17:55:13 -0300141 spin_lock_irq(&assigned_dev->assigned_dev_lock);
Sheng Yange56d5322009-03-12 21:45:39 +0800142 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
Sheng Yang2350bd12009-02-25 17:22:27 +0800143 struct kvm_guest_msix_entry *guest_entries =
144 assigned_dev->guest_msix_entries;
145 for (i = 0; i < assigned_dev->entries_nr; i++) {
146 if (!(guest_entries[i].flags &
147 KVM_ASSIGNED_MSIX_PENDING))
148 continue;
149 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
150 kvm_set_irq(assigned_dev->kvm,
151 assigned_dev->irq_source_id,
152 guest_entries[i].vector, 1);
Sheng Yang2350bd12009-02-25 17:22:27 +0800153 }
Sheng Yang968a6342009-04-30 10:58:42 +0800154 } else
Sheng Yang2350bd12009-02-25 17:22:27 +0800155 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
156 assigned_dev->guest_irq, 1);
Sheng Yang2350bd12009-02-25 17:22:27 +0800157
Marcelo Tosatti547de292009-05-07 17:55:13 -0300158 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
Marcelo Tosattifa40a822009-06-04 15:08:24 -0300159 mutex_unlock(&assigned_dev->kvm->irq_lock);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800160}
161
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800162static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
163{
Marcelo Tosatti547de292009-05-07 17:55:13 -0300164 unsigned long flags;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800165 struct kvm_assigned_dev_kernel *assigned_dev =
166 (struct kvm_assigned_dev_kernel *) dev_id;
167
Marcelo Tosatti547de292009-05-07 17:55:13 -0300168 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
Sheng Yange56d5322009-03-12 21:45:39 +0800169 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
Sheng Yang2350bd12009-02-25 17:22:27 +0800170 int index = find_index_from_host_irq(assigned_dev, irq);
171 if (index < 0)
Marcelo Tosatti547de292009-05-07 17:55:13 -0300172 goto out;
Sheng Yang2350bd12009-02-25 17:22:27 +0800173 assigned_dev->guest_msix_entries[index].flags |=
174 KVM_ASSIGNED_MSIX_PENDING;
175 }
176
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800177 schedule_work(&assigned_dev->interrupt_work);
Mark McLoughlindefaf152008-12-02 12:16:33 +0000178
Sheng Yang968a6342009-04-30 10:58:42 +0800179 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
180 disable_irq_nosync(irq);
181 assigned_dev->host_irq_disabled = true;
182 }
Mark McLoughlindefaf152008-12-02 12:16:33 +0000183
Marcelo Tosatti547de292009-05-07 17:55:13 -0300184out:
185 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800186 return IRQ_HANDLED;
187}
188
189/* Ack the irq line for an assigned device */
190static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
191{
192 struct kvm_assigned_dev_kernel *dev;
Marcelo Tosatti547de292009-05-07 17:55:13 -0300193 unsigned long flags;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800194
195 if (kian->gsi == -1)
196 return;
197
198 dev = container_of(kian, struct kvm_assigned_dev_kernel,
199 ack_notifier);
Mark McLoughlindefaf152008-12-02 12:16:33 +0000200
Sheng Yang5550af42008-10-15 20:15:06 +0800201 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
Mark McLoughlindefaf152008-12-02 12:16:33 +0000202
203 /* The guest irq may be shared so this ack may be
204 * from another device.
205 */
Marcelo Tosatti547de292009-05-07 17:55:13 -0300206 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
Mark McLoughlindefaf152008-12-02 12:16:33 +0000207 if (dev->host_irq_disabled) {
208 enable_irq(dev->host_irq);
209 dev->host_irq_disabled = false;
210 }
Marcelo Tosatti547de292009-05-07 17:55:13 -0300211 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800212}
213
Sheng Yange56d5322009-03-12 21:45:39 +0800214static void deassign_guest_irq(struct kvm *kvm,
215 struct kvm_assigned_dev_kernel *assigned_dev)
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800216{
Marcelo Tosattifa40a822009-06-04 15:08:24 -0300217 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
Sheng Yange56d5322009-03-12 21:45:39 +0800218 assigned_dev->ack_notifier.gsi = -1;
Mark McLoughlinf29b2672008-12-01 13:57:47 +0000219
220 if (assigned_dev->irq_source_id != -1)
221 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
222 assigned_dev->irq_source_id = -1;
Sheng Yange56d5322009-03-12 21:45:39 +0800223 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
224}
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800225
Sheng Yange56d5322009-03-12 21:45:39 +0800226/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
227static void deassign_host_irq(struct kvm *kvm,
228 struct kvm_assigned_dev_kernel *assigned_dev)
229{
Sheng Yangba4cef32009-01-06 10:03:03 +0800230 /*
231 * In kvm_free_device_irq, cancel_work_sync return true if:
232 * 1. work is scheduled, and then cancelled.
233 * 2. work callback is executed.
234 *
235 * The first one ensured that the irq is disabled and no more events
236 * would happen. But for the second one, the irq may be enabled (e.g.
237 * for MSI). So we disable irq here to prevent further events.
238 *
239 * Notice this maybe result in nested disable if the interrupt type is
240 * INTx, but it's OK for we are going to free it.
241 *
242 * If this function is a part of VM destroy, please ensure that till
243 * now, the kvm state is still legal for probably we also have to wait
244 * interrupt_work done.
245 */
Sheng Yange56d5322009-03-12 21:45:39 +0800246 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
Sheng Yangd510d6c2009-02-25 17:22:28 +0800247 int i;
248 for (i = 0; i < assigned_dev->entries_nr; i++)
249 disable_irq_nosync(assigned_dev->
250 host_msix_entries[i].vector);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800251
Sheng Yangd510d6c2009-02-25 17:22:28 +0800252 cancel_work_sync(&assigned_dev->interrupt_work);
Mark McLoughlin4a643be2008-12-01 13:57:49 +0000253
Sheng Yangd510d6c2009-02-25 17:22:28 +0800254 for (i = 0; i < assigned_dev->entries_nr; i++)
255 free_irq(assigned_dev->host_msix_entries[i].vector,
256 (void *)assigned_dev);
257
258 assigned_dev->entries_nr = 0;
259 kfree(assigned_dev->host_msix_entries);
260 kfree(assigned_dev->guest_msix_entries);
261 pci_disable_msix(assigned_dev->dev);
262 } else {
263 /* Deal with MSI and INTx */
264 disable_irq_nosync(assigned_dev->host_irq);
265 cancel_work_sync(&assigned_dev->interrupt_work);
266
267 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
268
Sheng Yange56d5322009-03-12 21:45:39 +0800269 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
Sheng Yangd510d6c2009-02-25 17:22:28 +0800270 pci_disable_msi(assigned_dev->dev);
271 }
Mark McLoughlin4a643be2008-12-01 13:57:49 +0000272
Sheng Yange56d5322009-03-12 21:45:39 +0800273 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
Mark McLoughlin4a643be2008-12-01 13:57:49 +0000274}
275
Sheng Yange56d5322009-03-12 21:45:39 +0800276static int kvm_deassign_irq(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *assigned_dev,
278 unsigned long irq_requested_type)
279{
280 unsigned long guest_irq_type, host_irq_type;
281
282 if (!irqchip_in_kernel(kvm))
283 return -EINVAL;
284 /* no irq assignment to deassign */
285 if (!assigned_dev->irq_requested_type)
286 return -ENXIO;
287
288 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
289 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
290
291 if (host_irq_type)
292 deassign_host_irq(kvm, assigned_dev);
293 if (guest_irq_type)
294 deassign_guest_irq(kvm, assigned_dev);
295
296 return 0;
297}
298
299static void kvm_free_assigned_irq(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *assigned_dev)
301{
302 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
303}
Mark McLoughlin4a643be2008-12-01 13:57:49 +0000304
305static void kvm_free_assigned_device(struct kvm *kvm,
306 struct kvm_assigned_dev_kernel
307 *assigned_dev)
308{
309 kvm_free_assigned_irq(kvm, assigned_dev);
310
Sheng Yang6eb55812008-10-31 12:37:41 +0800311 pci_reset_function(assigned_dev->dev);
312
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800313 pci_release_regions(assigned_dev->dev);
314 pci_disable_device(assigned_dev->dev);
315 pci_dev_put(assigned_dev->dev);
316
317 list_del(&assigned_dev->list);
318 kfree(assigned_dev);
319}
320
321void kvm_free_all_assigned_devices(struct kvm *kvm)
322{
323 struct list_head *ptr, *ptr2;
324 struct kvm_assigned_dev_kernel *assigned_dev;
325
326 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
327 assigned_dev = list_entry(ptr,
328 struct kvm_assigned_dev_kernel,
329 list);
330
331 kvm_free_assigned_device(kvm, assigned_dev);
332 }
333}
334
Sheng Yange56d5322009-03-12 21:45:39 +0800335static int assigned_device_enable_host_intx(struct kvm *kvm,
336 struct kvm_assigned_dev_kernel *dev)
Sheng Yang00e3ed32008-11-24 14:32:50 +0800337{
Sheng Yange56d5322009-03-12 21:45:39 +0800338 dev->host_irq = dev->dev->irq;
339 /* Even though this is PCI, we don't want to use shared
340 * interrupts. Sharing host devices with guest-assigned devices
341 * on the same interrupt line is not a happy situation: there
342 * are going to be long delays in accepting, acking, etc.
343 */
344 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
345 0, "kvm_assigned_intx_device", (void *)dev))
346 return -EIO;
Sheng Yang00e3ed32008-11-24 14:32:50 +0800347 return 0;
348}
349
Sheng Yange56d5322009-03-12 21:45:39 +0800350#ifdef __KVM_HAVE_MSI
351static int assigned_device_enable_host_msi(struct kvm *kvm,
352 struct kvm_assigned_dev_kernel *dev)
Sheng Yang6b9cc7f2008-11-24 14:32:56 +0800353{
354 int r;
355
Sheng Yange56d5322009-03-12 21:45:39 +0800356 if (!dev->dev->msi_enabled) {
357 r = pci_enable_msi(dev->dev);
358 if (r)
359 return r;
Sheng Yang5319c662008-11-24 14:32:57 +0800360 }
Sheng Yang6b9cc7f2008-11-24 14:32:56 +0800361
Sheng Yange56d5322009-03-12 21:45:39 +0800362 dev->host_irq = dev->dev->irq;
363 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
364 "kvm_assigned_msi_device", (void *)dev)) {
365 pci_disable_msi(dev->dev);
366 return -EIO;
Sheng Yang6b9cc7f2008-11-24 14:32:56 +0800367 }
368
Sheng Yang6b9cc7f2008-11-24 14:32:56 +0800369 return 0;
370}
371#endif
372
Sheng Yangd510d6c2009-02-25 17:22:28 +0800373#ifdef __KVM_HAVE_MSIX
Sheng Yange56d5322009-03-12 21:45:39 +0800374static int assigned_device_enable_host_msix(struct kvm *kvm,
375 struct kvm_assigned_dev_kernel *dev)
Sheng Yangd510d6c2009-02-25 17:22:28 +0800376{
Sheng Yange56d5322009-03-12 21:45:39 +0800377 int i, r = -EINVAL;
Sheng Yangd510d6c2009-02-25 17:22:28 +0800378
Sheng Yange56d5322009-03-12 21:45:39 +0800379 /* host_msix_entries and guest_msix_entries should have been
380 * initialized */
381 if (dev->entries_nr == 0)
382 return r;
Sheng Yangd510d6c2009-02-25 17:22:28 +0800383
Sheng Yange56d5322009-03-12 21:45:39 +0800384 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
385 if (r)
386 return r;
Sheng Yangd510d6c2009-02-25 17:22:28 +0800387
Sheng Yange56d5322009-03-12 21:45:39 +0800388 for (i = 0; i < dev->entries_nr; i++) {
389 r = request_irq(dev->host_msix_entries[i].vector,
390 kvm_assigned_dev_intr, 0,
391 "kvm_assigned_msix_device",
392 (void *)dev);
393 /* FIXME: free requested_irq's on failure */
Sheng Yangd510d6c2009-02-25 17:22:28 +0800394 if (r)
395 return r;
Sheng Yangd510d6c2009-02-25 17:22:28 +0800396 }
397
Sheng Yange56d5322009-03-12 21:45:39 +0800398 return 0;
399}
Sheng Yangd510d6c2009-02-25 17:22:28 +0800400
Sheng Yange56d5322009-03-12 21:45:39 +0800401#endif
402
403static int assigned_device_enable_guest_intx(struct kvm *kvm,
404 struct kvm_assigned_dev_kernel *dev,
405 struct kvm_assigned_irq *irq)
406{
407 dev->guest_irq = irq->guest_irq;
408 dev->ack_notifier.gsi = irq->guest_irq;
409 return 0;
410}
411
412#ifdef __KVM_HAVE_MSI
413static int assigned_device_enable_guest_msi(struct kvm *kvm,
414 struct kvm_assigned_dev_kernel *dev,
415 struct kvm_assigned_irq *irq)
416{
417 dev->guest_irq = irq->guest_irq;
418 dev->ack_notifier.gsi = -1;
Sheng Yang968a6342009-04-30 10:58:42 +0800419 dev->host_irq_disabled = false;
Sheng Yange56d5322009-03-12 21:45:39 +0800420 return 0;
421}
422#endif
423#ifdef __KVM_HAVE_MSIX
424static int assigned_device_enable_guest_msix(struct kvm *kvm,
425 struct kvm_assigned_dev_kernel *dev,
426 struct kvm_assigned_irq *irq)
427{
428 dev->guest_irq = irq->guest_irq;
429 dev->ack_notifier.gsi = -1;
Sheng Yang968a6342009-04-30 10:58:42 +0800430 dev->host_irq_disabled = false;
Sheng Yangd510d6c2009-02-25 17:22:28 +0800431 return 0;
432}
433#endif
434
Sheng Yange56d5322009-03-12 21:45:39 +0800435static int assign_host_irq(struct kvm *kvm,
436 struct kvm_assigned_dev_kernel *dev,
437 __u32 host_irq_type)
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800438{
Sheng Yange56d5322009-03-12 21:45:39 +0800439 int r = -EEXIST;
440
441 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
442 return r;
443
444 switch (host_irq_type) {
445 case KVM_DEV_IRQ_HOST_INTX:
446 r = assigned_device_enable_host_intx(kvm, dev);
447 break;
448#ifdef __KVM_HAVE_MSI
449 case KVM_DEV_IRQ_HOST_MSI:
450 r = assigned_device_enable_host_msi(kvm, dev);
451 break;
452#endif
453#ifdef __KVM_HAVE_MSIX
454 case KVM_DEV_IRQ_HOST_MSIX:
455 r = assigned_device_enable_host_msix(kvm, dev);
456 break;
457#endif
458 default:
459 r = -EINVAL;
460 }
461
462 if (!r)
463 dev->irq_requested_type |= host_irq_type;
464
465 return r;
466}
467
468static int assign_guest_irq(struct kvm *kvm,
469 struct kvm_assigned_dev_kernel *dev,
470 struct kvm_assigned_irq *irq,
471 unsigned long guest_irq_type)
472{
473 int id;
474 int r = -EEXIST;
475
476 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
477 return r;
478
479 id = kvm_request_irq_source_id(kvm);
480 if (id < 0)
481 return id;
482
483 dev->irq_source_id = id;
484
485 switch (guest_irq_type) {
486 case KVM_DEV_IRQ_GUEST_INTX:
487 r = assigned_device_enable_guest_intx(kvm, dev, irq);
488 break;
489#ifdef __KVM_HAVE_MSI
490 case KVM_DEV_IRQ_GUEST_MSI:
491 r = assigned_device_enable_guest_msi(kvm, dev, irq);
492 break;
493#endif
494#ifdef __KVM_HAVE_MSIX
495 case KVM_DEV_IRQ_GUEST_MSIX:
496 r = assigned_device_enable_guest_msix(kvm, dev, irq);
497 break;
498#endif
499 default:
500 r = -EINVAL;
501 }
502
503 if (!r) {
504 dev->irq_requested_type |= guest_irq_type;
505 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
506 } else
507 kvm_free_irq_source_id(kvm, dev->irq_source_id);
508
509 return r;
510}
511
512/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
513static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
514 struct kvm_assigned_irq *assigned_irq)
515{
516 int r = -EINVAL;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800517 struct kvm_assigned_dev_kernel *match;
Sheng Yange56d5322009-03-12 21:45:39 +0800518 unsigned long host_irq_type, guest_irq_type;
519
520 if (!capable(CAP_SYS_RAWIO))
521 return -EPERM;
522
523 if (!irqchip_in_kernel(kvm))
524 return r;
525
526 mutex_lock(&kvm->lock);
527 r = -ENODEV;
528 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
529 assigned_irq->assigned_dev_id);
530 if (!match)
531 goto out;
532
533 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
534 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
535
536 r = -EINVAL;
537 /* can only assign one type at a time */
538 if (hweight_long(host_irq_type) > 1)
539 goto out;
540 if (hweight_long(guest_irq_type) > 1)
541 goto out;
542 if (host_irq_type == 0 && guest_irq_type == 0)
543 goto out;
544
545 r = 0;
546 if (host_irq_type)
547 r = assign_host_irq(kvm, match, host_irq_type);
548 if (r)
549 goto out;
550
551 if (guest_irq_type)
552 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
553out:
554 mutex_unlock(&kvm->lock);
555 return r;
556}
557
558static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
559 struct kvm_assigned_irq
560 *assigned_irq)
561{
562 int r = -ENODEV;
563 struct kvm_assigned_dev_kernel *match;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800564
565 mutex_lock(&kvm->lock);
566
567 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
568 assigned_irq->assigned_dev_id);
Sheng Yange56d5322009-03-12 21:45:39 +0800569 if (!match)
570 goto out;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800571
Sheng Yange56d5322009-03-12 21:45:39 +0800572 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
573out:
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800574 mutex_unlock(&kvm->lock);
575 return r;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800576}
577
578static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
579 struct kvm_assigned_pci_dev *assigned_dev)
580{
581 int r = 0;
582 struct kvm_assigned_dev_kernel *match;
583 struct pci_dev *dev;
584
Mark McLoughlin682edb42009-02-05 18:23:46 +0000585 down_read(&kvm->slots_lock);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800586 mutex_lock(&kvm->lock);
587
588 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
589 assigned_dev->assigned_dev_id);
590 if (match) {
591 /* device already assigned */
Sheng Yange56d5322009-03-12 21:45:39 +0800592 r = -EEXIST;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800593 goto out;
594 }
595
596 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
597 if (match == NULL) {
598 printk(KERN_INFO "%s: Couldn't allocate memory\n",
599 __func__);
600 r = -ENOMEM;
601 goto out;
602 }
603 dev = pci_get_bus_and_slot(assigned_dev->busnr,
604 assigned_dev->devfn);
605 if (!dev) {
606 printk(KERN_INFO "%s: host device not found\n", __func__);
607 r = -EINVAL;
608 goto out_free;
609 }
610 if (pci_enable_device(dev)) {
611 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
612 r = -EBUSY;
613 goto out_put;
614 }
615 r = pci_request_regions(dev, "kvm_assigned_device");
616 if (r) {
617 printk(KERN_INFO "%s: Could not get access to device regions\n",
618 __func__);
619 goto out_disable;
620 }
Sheng Yang6eb55812008-10-31 12:37:41 +0800621
622 pci_reset_function(dev);
623
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800624 match->assigned_dev_id = assigned_dev->assigned_dev_id;
625 match->host_busnr = assigned_dev->busnr;
626 match->host_devfn = assigned_dev->devfn;
Weidong Hanb6535742008-12-08 23:29:53 +0800627 match->flags = assigned_dev->flags;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800628 match->dev = dev;
Marcelo Tosatti547de292009-05-07 17:55:13 -0300629 spin_lock_init(&match->assigned_dev_lock);
Mark McLoughlinf29b2672008-12-01 13:57:47 +0000630 match->irq_source_id = -1;
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800631 match->kvm = kvm;
Sheng Yange56d5322009-03-12 21:45:39 +0800632 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
633 INIT_WORK(&match->interrupt_work,
634 kvm_assigned_dev_interrupt_work_handler);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800635
636 list_add(&match->list, &kvm->arch.assigned_dev_head);
637
638 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
Joerg Roedel19de40a2008-12-03 14:43:34 +0100639 if (!kvm->arch.iommu_domain) {
Weidong Han260782b2008-12-02 21:03:39 +0800640 r = kvm_iommu_map_guest(kvm);
641 if (r)
642 goto out_list_del;
643 }
644 r = kvm_assign_device(kvm, match);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800645 if (r)
646 goto out_list_del;
647 }
648
649out:
650 mutex_unlock(&kvm->lock);
Mark McLoughlin682edb42009-02-05 18:23:46 +0000651 up_read(&kvm->slots_lock);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800652 return r;
653out_list_del:
654 list_del(&match->list);
655 pci_release_regions(dev);
656out_disable:
657 pci_disable_device(dev);
658out_put:
659 pci_dev_put(dev);
660out_free:
661 kfree(match);
662 mutex_unlock(&kvm->lock);
Mark McLoughlin682edb42009-02-05 18:23:46 +0000663 up_read(&kvm->slots_lock);
Xiantao Zhang8a98f662008-10-06 13:47:38 +0800664 return r;
665}
666#endif
667
Weidong Han0a920352008-12-02 21:24:23 +0800668#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
669static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
670 struct kvm_assigned_pci_dev *assigned_dev)
671{
672 int r = 0;
673 struct kvm_assigned_dev_kernel *match;
674
675 mutex_lock(&kvm->lock);
676
677 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
678 assigned_dev->assigned_dev_id);
679 if (!match) {
680 printk(KERN_INFO "%s: device hasn't been assigned before, "
681 "so cannot be deassigned\n", __func__);
682 r = -EINVAL;
683 goto out;
684 }
685
Weidong Han4a906e42009-02-13 17:27:51 +0800686 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
Weidong Han0a920352008-12-02 21:24:23 +0800687 kvm_deassign_device(kvm, match);
688
689 kvm_free_assigned_device(kvm, match);
690
691out:
692 mutex_unlock(&kvm->lock);
693 return r;
694}
695#endif
696
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +0800697inline int kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300698{
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100699 if (pfn_valid(pfn)) {
700 struct page *page = compound_head(pfn_to_page(pfn));
701 return PageReserved(page);
702 }
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300703
704 return true;
705}
706
Avi Kivity6aa8b732006-12-10 02:21:36 -0800707/*
708 * Switches to specified vcpu, until a matching vcpu_put()
709 */
Carsten Otte313a3dc2007-10-11 19:16:52 +0200710void vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800711{
Avi Kivity15ad7142007-07-11 18:17:21 +0300712 int cpu;
713
Avi Kivitybccf2152007-02-21 18:04:26 +0200714 mutex_lock(&vcpu->mutex);
Avi Kivity15ad7142007-07-11 18:17:21 +0300715 cpu = get_cpu();
716 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200717 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300718 put_cpu();
Avi Kivitybccf2152007-02-21 18:04:26 +0200719}
720
Carsten Otte313a3dc2007-10-11 19:16:52 +0200721void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800722{
Avi Kivity15ad7142007-07-11 18:17:21 +0300723 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200724 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300725 preempt_notifier_unregister(&vcpu->preempt_notifier);
726 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800727 mutex_unlock(&vcpu->mutex);
728}
729
Avi Kivityd9e368d2007-06-07 19:18:30 +0300730static void ack_flush(void *_completed)
731{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300732}
733
Rusty Russell49846892008-12-08 20:26:24 +1030734static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300735{
Avi Kivity597a5f52008-07-20 14:24:22 +0300736 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030737 cpumask_var_t cpus;
738 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300739 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300740
Li Zefan79f55992009-06-15 14:58:26 +0800741 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030742
Marcelo Tosatti84261922009-06-17 10:53:47 -0300743 spin_lock(&kvm->requests_lock);
Jan Kiszkae601e3b2009-07-20 11:30:12 +0200744 me = smp_processor_id();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300745 kvm_for_each_vcpu(i, vcpu, kvm) {
Rusty Russell49846892008-12-08 20:26:24 +1030746 if (test_and_set_bit(req, &vcpu->requests))
Avi Kivityd9e368d2007-06-07 19:18:30 +0300747 continue;
748 cpu = vcpu->cpu;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030749 if (cpus != NULL && cpu != -1 && cpu != me)
750 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300751 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030752 if (unlikely(cpus == NULL))
753 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
754 else if (!cpumask_empty(cpus))
755 smp_call_function_many(cpus, ack_flush, NULL, 1);
756 else
757 called = false;
Marcelo Tosatti84261922009-06-17 10:53:47 -0300758 spin_unlock(&kvm->requests_lock);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030759 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030760 return called;
761}
762
763void kvm_flush_remote_tlbs(struct kvm *kvm)
764{
765 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
766 ++kvm->stat.remote_tlb_flush;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300767}
768
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500769void kvm_reload_remote_mmus(struct kvm *kvm)
770{
Rusty Russell49846892008-12-08 20:26:24 +1030771 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500772}
773
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000774int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
775{
776 struct page *page;
777 int r;
778
779 mutex_init(&vcpu->mutex);
780 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000781 vcpu->kvm = kvm;
782 vcpu->vcpu_id = id;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300783 init_waitqueue_head(&vcpu->wq);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000784
785 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
786 if (!page) {
787 r = -ENOMEM;
788 goto fail;
789 }
790 vcpu->run = page_address(page);
791
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800792 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000793 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800794 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000795 return 0;
796
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000797fail_free_run:
798 free_page((unsigned long)vcpu->run);
799fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000800 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000801}
802EXPORT_SYMBOL_GPL(kvm_vcpu_init);
803
804void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
805{
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800806 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000807 free_page((unsigned long)vcpu->run);
808}
809EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
810
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200811#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
812static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
813{
814 return container_of(mn, struct kvm, mmu_notifier);
815}
816
817static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
818 struct mm_struct *mm,
819 unsigned long address)
820{
821 struct kvm *kvm = mmu_notifier_to_kvm(mn);
822 int need_tlb_flush;
823
824 /*
825 * When ->invalidate_page runs, the linux pte has been zapped
826 * already but the page is still allocated until
827 * ->invalidate_page returns. So if we increase the sequence
828 * here the kvm page fault will notice if the spte can't be
829 * established because the page is going to be freed. If
830 * instead the kvm page fault establishes the spte before
831 * ->invalidate_page runs, kvm_unmap_hva will release it
832 * before returning.
833 *
834 * The sequence increase only need to be seen at spin_unlock
835 * time, and not at spin_lock time.
836 *
837 * Increasing the sequence after the spin_unlock would be
838 * unsafe because the kvm page fault could then establish the
839 * pte after kvm_unmap_hva returned, without noticing the page
840 * is going to be freed.
841 */
842 spin_lock(&kvm->mmu_lock);
843 kvm->mmu_notifier_seq++;
844 need_tlb_flush = kvm_unmap_hva(kvm, address);
845 spin_unlock(&kvm->mmu_lock);
846
847 /* we've to flush the tlb before the pages can be freed */
848 if (need_tlb_flush)
849 kvm_flush_remote_tlbs(kvm);
850
851}
852
Izik Eidus3da0dd42009-09-23 21:47:18 +0300853static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
854 struct mm_struct *mm,
855 unsigned long address,
856 pte_t pte)
857{
858 struct kvm *kvm = mmu_notifier_to_kvm(mn);
859
860 spin_lock(&kvm->mmu_lock);
861 kvm->mmu_notifier_seq++;
862 kvm_set_spte_hva(kvm, address, pte);
863 spin_unlock(&kvm->mmu_lock);
864}
865
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200866static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
867 struct mm_struct *mm,
868 unsigned long start,
869 unsigned long end)
870{
871 struct kvm *kvm = mmu_notifier_to_kvm(mn);
872 int need_tlb_flush = 0;
873
874 spin_lock(&kvm->mmu_lock);
875 /*
876 * The count increase must become visible at unlock time as no
877 * spte can be established without taking the mmu_lock and
878 * count is also read inside the mmu_lock critical section.
879 */
880 kvm->mmu_notifier_count++;
881 for (; start < end; start += PAGE_SIZE)
882 need_tlb_flush |= kvm_unmap_hva(kvm, start);
883 spin_unlock(&kvm->mmu_lock);
884
885 /* we've to flush the tlb before the pages can be freed */
886 if (need_tlb_flush)
887 kvm_flush_remote_tlbs(kvm);
888}
889
890static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
891 struct mm_struct *mm,
892 unsigned long start,
893 unsigned long end)
894{
895 struct kvm *kvm = mmu_notifier_to_kvm(mn);
896
897 spin_lock(&kvm->mmu_lock);
898 /*
899 * This sequence increase will notify the kvm page fault that
900 * the page that is going to be mapped in the spte could have
901 * been freed.
902 */
903 kvm->mmu_notifier_seq++;
904 /*
905 * The above sequence increase must be visible before the
906 * below count decrease but both values are read by the kvm
907 * page fault under mmu_lock spinlock so we don't need to add
908 * a smb_wmb() here in between the two.
909 */
910 kvm->mmu_notifier_count--;
911 spin_unlock(&kvm->mmu_lock);
912
913 BUG_ON(kvm->mmu_notifier_count < 0);
914}
915
916static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
917 struct mm_struct *mm,
918 unsigned long address)
919{
920 struct kvm *kvm = mmu_notifier_to_kvm(mn);
921 int young;
922
923 spin_lock(&kvm->mmu_lock);
924 young = kvm_age_hva(kvm, address);
925 spin_unlock(&kvm->mmu_lock);
926
927 if (young)
928 kvm_flush_remote_tlbs(kvm);
929
930 return young;
931}
932
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100933static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
934 struct mm_struct *mm)
935{
936 struct kvm *kvm = mmu_notifier_to_kvm(mn);
937 kvm_arch_flush_shadow(kvm);
938}
939
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200940static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
941 .invalidate_page = kvm_mmu_notifier_invalidate_page,
942 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
943 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
944 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300945 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100946 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200947};
948#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
949
Avi Kivityf17abe92007-02-21 19:28:04 +0200950static struct kvm *kvm_create_vm(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800951{
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800952 struct kvm *kvm = kvm_arch_create_vm();
Laurent Vivier5f94c172008-05-30 16:05:54 +0200953#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
954 struct page *page;
955#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800956
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800957 if (IS_ERR(kvm))
958 goto out;
Avi Kivity75858a82009-01-04 17:10:50 +0200959#ifdef CONFIG_HAVE_KVM_IRQCHIP
Avi Kivity399ec802008-11-19 13:58:46 +0200960 INIT_LIST_HEAD(&kvm->irq_routing);
Avi Kivity75858a82009-01-04 17:10:50 +0200961 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
962#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800963
Laurent Vivier5f94c172008-05-30 16:05:54 +0200964#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
965 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
966 if (!page) {
967 kfree(kvm);
968 return ERR_PTR(-ENOMEM);
969 }
970 kvm->coalesced_mmio_ring =
971 (struct kvm_coalesced_mmio_ring *)page_address(page);
972#endif
973
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200974#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
975 {
976 int err;
977 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
978 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
979 if (err) {
980#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
981 put_page(page);
982#endif
983 kfree(kvm);
984 return ERR_PTR(err);
985 }
986 }
987#endif
988
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200989 kvm->mm = current->mm;
990 atomic_inc(&kvm->mm->mm_count);
Marcelo Tosattiaaee2c92007-12-20 19:18:26 -0500991 spin_lock_init(&kvm->mmu_lock);
Marcelo Tosatti84261922009-06-17 10:53:47 -0300992 spin_lock_init(&kvm->requests_lock);
Eddie Dong74906342007-06-19 18:05:03 +0300993 kvm_io_bus_init(&kvm->pio_bus);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400994 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800995 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300996 mutex_init(&kvm->irq_lock);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -0400997 kvm_io_bus_init(&kvm->mmio_bus);
Izik Eidus72dc67a2008-02-10 18:04:15 +0200998 init_rwsem(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300999 atomic_set(&kvm->users_count, 1);
Rusty Russell5e58cfe2007-07-23 17:08:21 +10001000 spin_lock(&kvm_lock);
1001 list_add(&kvm->vm_list, &vm_list);
1002 spin_unlock(&kvm_lock);
Laurent Vivier5f94c172008-05-30 16:05:54 +02001003#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1004 kvm_coalesced_mmio_init(kvm);
1005#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08001006out:
Avi Kivityf17abe92007-02-21 19:28:04 +02001007 return kvm;
1008}
1009
Avi Kivity6aa8b732006-12-10 02:21:36 -08001010/*
1011 * Free any memory in @free but not in @dont.
1012 */
1013static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1014 struct kvm_memory_slot *dont)
1015{
Joerg Roedelec04b262009-06-19 15:16:23 +02001016 int i;
1017
Izik Eidus290fc382007-09-27 14:11:22 +02001018 if (!dont || free->rmap != dont->rmap)
1019 vfree(free->rmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001020
1021 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1022 vfree(free->dirty_bitmap);
1023
Joerg Roedelec04b262009-06-19 15:16:23 +02001024
1025 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1026 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1027 vfree(free->lpage_info[i]);
1028 free->lpage_info[i] = NULL;
1029 }
1030 }
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001031
Avi Kivity6aa8b732006-12-10 02:21:36 -08001032 free->npages = 0;
Al Viro8b6d44c2007-02-09 16:38:40 +00001033 free->dirty_bitmap = NULL;
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001034 free->rmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001035}
1036
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08001037void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001038{
1039 int i;
1040
1041 for (i = 0; i < kvm->nmemslots; ++i)
Al Viro8b6d44c2007-02-09 16:38:40 +00001042 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001043}
1044
Avi Kivityf17abe92007-02-21 19:28:04 +02001045static void kvm_destroy_vm(struct kvm *kvm)
1046{
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001047 struct mm_struct *mm = kvm->mm;
1048
Sheng Yangad8ba2c2009-01-06 10:03:02 +08001049 kvm_arch_sync_events(kvm);
Avi Kivity133de902007-02-12 00:54:44 -08001050 spin_lock(&kvm_lock);
1051 list_del(&kvm->vm_list);
1052 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +02001053 kvm_free_irq_routing(kvm);
Eddie Dong74906342007-06-19 18:05:03 +03001054 kvm_io_bus_destroy(&kvm->pio_bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04001055 kvm_io_bus_destroy(&kvm->mmio_bus);
Laurent Vivier5f94c172008-05-30 16:05:54 +02001056#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1057 if (kvm->coalesced_mmio_ring != NULL)
1058 free_page((unsigned long)kvm->coalesced_mmio_ring);
1059#endif
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001060#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1061 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +02001062#else
1063 kvm_arch_flush_shadow(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +02001064#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +08001065 kvm_arch_destroy_vm(kvm);
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001066 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +02001067}
1068
Izik Eidusd39f13b2008-03-30 16:01:25 +03001069void kvm_get_kvm(struct kvm *kvm)
1070{
1071 atomic_inc(&kvm->users_count);
1072}
1073EXPORT_SYMBOL_GPL(kvm_get_kvm);
1074
1075void kvm_put_kvm(struct kvm *kvm)
1076{
1077 if (atomic_dec_and_test(&kvm->users_count))
1078 kvm_destroy_vm(kvm);
1079}
1080EXPORT_SYMBOL_GPL(kvm_put_kvm);
1081
1082
Avi Kivityf17abe92007-02-21 19:28:04 +02001083static int kvm_vm_release(struct inode *inode, struct file *filp)
1084{
1085 struct kvm *kvm = filp->private_data;
1086
Gregory Haskins721eecb2009-05-20 10:30:49 -04001087 kvm_irqfd_release(kvm);
1088
Izik Eidusd39f13b2008-03-30 16:01:25 +03001089 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001090 return 0;
1091}
1092
Avi Kivity6aa8b732006-12-10 02:21:36 -08001093/*
Avi Kivity6aa8b732006-12-10 02:21:36 -08001094 * Allocate some memory and give it an address in the guest physical address
1095 * space.
1096 *
1097 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +08001098 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -05001099 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -08001100 */
Sheng Yangf78e0e22007-10-29 09:40:42 +08001101int __kvm_set_memory_region(struct kvm *kvm,
1102 struct kvm_userspace_memory_region *mem,
1103 int user_alloc)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001104{
1105 int r;
1106 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +02001107 unsigned long npages;
1108 unsigned long i;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001109 struct kvm_memory_slot *memslot;
1110 struct kvm_memory_slot old, new;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001111
1112 r = -EINVAL;
1113 /* General sanity checks */
1114 if (mem->memory_size & (PAGE_SIZE - 1))
1115 goto out;
1116 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1117 goto out;
Sheng Yange7cacd42008-11-11 15:30:40 +08001118 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
Hollis Blanchard78749802008-11-07 13:32:12 -06001119 goto out;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001120 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001121 goto out;
1122 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1123 goto out;
1124
1125 memslot = &kvm->memslots[mem->slot];
1126 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1127 npages = mem->memory_size >> PAGE_SHIFT;
1128
1129 if (!npages)
1130 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1131
Avi Kivity6aa8b732006-12-10 02:21:36 -08001132 new = old = *memslot;
1133
1134 new.base_gfn = base_gfn;
1135 new.npages = npages;
1136 new.flags = mem->flags;
1137
1138 /* Disallow changing a memory slot's size. */
1139 r = -EINVAL;
1140 if (npages && old.npages && npages != old.npages)
Sheng Yangf78e0e22007-10-29 09:40:42 +08001141 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001142
1143 /* Check for overlaps */
1144 r = -EEXIST;
1145 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1146 struct kvm_memory_slot *s = &kvm->memslots[i];
1147
Jan Kiszka4cd481f2009-04-13 11:59:32 +02001148 if (s == memslot || !s->npages)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001149 continue;
1150 if (!((base_gfn + npages <= s->base_gfn) ||
1151 (base_gfn >= s->base_gfn + s->npages)))
Sheng Yangf78e0e22007-10-29 09:40:42 +08001152 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001153 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001154
Avi Kivity6aa8b732006-12-10 02:21:36 -08001155 /* Free page dirty bitmap if unneeded */
1156 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +00001157 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001158
1159 r = -ENOMEM;
1160
1161 /* Allocate if a slot is being created */
Carsten Otteeff01142008-06-27 15:05:31 +02001162#ifndef CONFIG_S390
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001163 if (npages && !new.rmap) {
Mike Dayd77c26f2007-10-08 09:02:08 -04001164 new.rmap = vmalloc(npages * sizeof(struct page *));
Izik Eidus290fc382007-09-27 14:11:22 +02001165
1166 if (!new.rmap)
Sheng Yangf78e0e22007-10-29 09:40:42 +08001167 goto out_free;
Izik Eidus290fc382007-09-27 14:11:22 +02001168
Izik Eidus290fc382007-09-27 14:11:22 +02001169 memset(new.rmap, 0, npages * sizeof(*new.rmap));
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001170
Izik Eidus80b14b52007-10-25 11:54:04 +02001171 new.user_alloc = user_alloc;
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02001172 /*
1173 * hva_to_rmmap() serialzies with the mmu_lock and to be
1174 * safe it has to ignore memslots with !user_alloc &&
1175 * !userspace_addr.
1176 */
1177 if (user_alloc)
1178 new.userspace_addr = mem->userspace_addr;
1179 else
1180 new.userspace_addr = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001181 }
Joerg Roedelec04b262009-06-19 15:16:23 +02001182 if (!npages)
1183 goto skip_lpage;
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001184
Joerg Roedelec04b262009-06-19 15:16:23 +02001185 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
Heiko Carstens28bcb112009-09-03 17:35:35 +02001186 unsigned long ugfn;
1187 unsigned long j;
1188 int lpages;
Joerg Roedelec04b262009-06-19 15:16:23 +02001189 int level = i + 2;
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001190
Joerg Roedelec04b262009-06-19 15:16:23 +02001191 /* Avoid unused variable warning if no large pages */
1192 (void)level;
1193
1194 if (new.lpage_info[i])
1195 continue;
1196
1197 lpages = 1 + (base_gfn + npages - 1) /
1198 KVM_PAGES_PER_HPAGE(level);
1199 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1200
1201 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1202
1203 if (!new.lpage_info[i])
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001204 goto out_free;
1205
Joerg Roedelec04b262009-06-19 15:16:23 +02001206 memset(new.lpage_info[i], 0,
1207 lpages * sizeof(*new.lpage_info[i]));
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001208
Joerg Roedelec04b262009-06-19 15:16:23 +02001209 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1210 new.lpage_info[i][0].write_count = 1;
1211 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1212 new.lpage_info[i][lpages - 1].write_count = 1;
Avi Kivityac045272009-06-08 15:52:39 +03001213 ugfn = new.userspace_addr >> PAGE_SHIFT;
1214 /*
1215 * If the gfn and userspace address are not aligned wrt each
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001216 * other, or if explicitly asked to, disable large page
1217 * support for this slot
Avi Kivityac045272009-06-08 15:52:39 +03001218 */
Joerg Roedelec04b262009-06-19 15:16:23 +02001219 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001220 !largepages_enabled)
Joerg Roedelec04b262009-06-19 15:16:23 +02001221 for (j = 0; j < lpages; ++j)
1222 new.lpage_info[i][j].write_count = 1;
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001223 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001224
Joerg Roedelec04b262009-06-19 15:16:23 +02001225skip_lpage:
1226
Avi Kivity6aa8b732006-12-10 02:21:36 -08001227 /* Allocate page dirty bitmap if needed */
1228 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1229 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1230
1231 new.dirty_bitmap = vmalloc(dirty_bytes);
1232 if (!new.dirty_bitmap)
Sheng Yangf78e0e22007-10-29 09:40:42 +08001233 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001234 memset(new.dirty_bitmap, 0, dirty_bytes);
Izik Eiduse2445842009-06-10 19:23:24 +03001235 if (old.npages)
1236 kvm_arch_flush_shadow(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001237 }
Christian Borntraeger3eea8432009-06-23 17:24:06 +02001238#else /* not defined CONFIG_S390 */
1239 new.user_alloc = user_alloc;
1240 if (user_alloc)
1241 new.userspace_addr = mem->userspace_addr;
Carsten Otteeff01142008-06-27 15:05:31 +02001242#endif /* not defined CONFIG_S390 */
Avi Kivity6aa8b732006-12-10 02:21:36 -08001243
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03001244 if (!npages)
1245 kvm_arch_flush_shadow(kvm);
1246
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02001247 spin_lock(&kvm->mmu_lock);
1248 if (mem->slot >= kvm->nmemslots)
1249 kvm->nmemslots = mem->slot + 1;
1250
Avi Kivity6aa8b732006-12-10 02:21:36 -08001251 *memslot = new;
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02001252 spin_unlock(&kvm->mmu_lock);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001253
Zhang Xiantao0de10342007-11-20 16:25:04 +08001254 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1255 if (r) {
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02001256 spin_lock(&kvm->mmu_lock);
Zhang Xiantao0de10342007-11-20 16:25:04 +08001257 *memslot = old;
Andrea Arcangeli604b38a2008-07-25 16:32:03 +02001258 spin_unlock(&kvm->mmu_lock);
Zhang Xiantao0de10342007-11-20 16:25:04 +08001259 goto out_free;
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08001260 }
1261
Glauber Costa6f897242008-12-03 13:40:51 -02001262 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1263 /* Slot deletion case: we have to update the current slot */
Marcelo Tosattib43b1902009-05-12 18:55:44 -03001264 spin_lock(&kvm->mmu_lock);
Glauber Costa6f897242008-12-03 13:40:51 -02001265 if (!npages)
1266 *memslot = old;
Marcelo Tosattib43b1902009-05-12 18:55:44 -03001267 spin_unlock(&kvm->mmu_lock);
Xiantao Zhang8a98f662008-10-06 13:47:38 +08001268#ifdef CONFIG_DMAR
Ben-Ami Yassour62c476c2008-09-14 03:48:28 +03001269 /* map the pages in iommu page table */
1270 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1271 if (r)
1272 goto out;
Xiantao Zhang8a98f662008-10-06 13:47:38 +08001273#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -08001274 return 0;
1275
Sheng Yangf78e0e22007-10-29 09:40:42 +08001276out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -08001277 kvm_free_physmem_slot(&new, &old);
1278out:
1279 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +02001280
1281}
Sheng Yangf78e0e22007-10-29 09:40:42 +08001282EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1283
1284int kvm_set_memory_region(struct kvm *kvm,
1285 struct kvm_userspace_memory_region *mem,
1286 int user_alloc)
1287{
1288 int r;
1289
Izik Eidus72dc67a2008-02-10 18:04:15 +02001290 down_write(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +08001291 r = __kvm_set_memory_region(kvm, mem, user_alloc);
Izik Eidus72dc67a2008-02-10 18:04:15 +02001292 up_write(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +08001293 return r;
1294}
Izik Eidus210c7c42007-10-24 23:52:57 +02001295EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1296
Carsten Otte1fe779f2007-10-29 16:08:35 +01001297int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1298 struct
1299 kvm_userspace_memory_region *mem,
1300 int user_alloc)
Izik Eidus210c7c42007-10-24 23:52:57 +02001301{
Izik Eiduse0d62c72007-10-24 23:57:46 +02001302 if (mem->slot >= KVM_MEMORY_SLOTS)
1303 return -EINVAL;
Izik Eidus210c7c42007-10-24 23:52:57 +02001304 return kvm_set_memory_region(kvm, mem, user_alloc);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001305}
1306
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08001307int kvm_get_dirty_log(struct kvm *kvm,
1308 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001309{
1310 struct kvm_memory_slot *memslot;
1311 int r, i;
1312 int n;
1313 unsigned long any = 0;
1314
Avi Kivity6aa8b732006-12-10 02:21:36 -08001315 r = -EINVAL;
1316 if (log->slot >= KVM_MEMORY_SLOTS)
1317 goto out;
1318
1319 memslot = &kvm->memslots[log->slot];
1320 r = -ENOENT;
1321 if (!memslot->dirty_bitmap)
1322 goto out;
1323
Uri Lublincd1a4a92007-02-22 16:43:09 +02001324 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001325
Uri Lublincd1a4a92007-02-22 16:43:09 +02001326 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001327 any = memslot->dirty_bitmap[i];
1328
1329 r = -EFAULT;
1330 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1331 goto out;
1332
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08001333 if (any)
1334 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001335
1336 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001337out:
Avi Kivity6aa8b732006-12-10 02:21:36 -08001338 return r;
1339}
1340
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001341void kvm_disable_largepages(void)
1342{
1343 largepages_enabled = false;
1344}
1345EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1346
Izik Eiduscea7bb22007-10-17 19:17:48 +02001347int is_error_page(struct page *page)
1348{
1349 return page == bad_page;
1350}
1351EXPORT_SYMBOL_GPL(is_error_page);
1352
Anthony Liguori35149e22008-04-02 14:46:56 -05001353int is_error_pfn(pfn_t pfn)
1354{
1355 return pfn == bad_pfn;
1356}
1357EXPORT_SYMBOL_GPL(is_error_pfn);
1358
Izik Eidusf9d46eb2007-11-11 22:02:22 +02001359static inline unsigned long bad_hva(void)
1360{
1361 return PAGE_OFFSET;
1362}
1363
1364int kvm_is_error_hva(unsigned long addr)
1365{
1366 return addr == bad_hva();
1367}
1368EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1369
Izik Eidus28430992008-10-03 17:40:32 +03001370struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001371{
1372 int i;
1373
1374 for (i = 0; i < kvm->nmemslots; ++i) {
1375 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1376
1377 if (gfn >= memslot->base_gfn
1378 && gfn < memslot->base_gfn + memslot->npages)
1379 return memslot;
1380 }
Al Viro8b6d44c2007-02-09 16:38:40 +00001381 return NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001382}
Izik Eidus28430992008-10-03 17:40:32 +03001383EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
Avi Kivitye8207542007-03-30 16:54:30 +03001384
1385struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1386{
1387 gfn = unalias_gfn(kvm, gfn);
Izik Eidus28430992008-10-03 17:40:32 +03001388 return gfn_to_memslot_unaliased(kvm, gfn);
Avi Kivitye8207542007-03-30 16:54:30 +03001389}
Avi Kivity6aa8b732006-12-10 02:21:36 -08001390
Izik Eiduse0d62c72007-10-24 23:57:46 +02001391int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1392{
1393 int i;
1394
1395 gfn = unalias_gfn(kvm, gfn);
1396 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1397 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1398
1399 if (gfn >= memslot->base_gfn
1400 && gfn < memslot->base_gfn + memslot->npages)
1401 return 1;
1402 }
1403 return 0;
1404}
1405EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1406
Marcelo Tosatti05da4552008-02-23 11:44:30 -03001407unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
Izik Eidus539cb662007-11-11 22:05:04 +02001408{
1409 struct kvm_memory_slot *slot;
1410
1411 gfn = unalias_gfn(kvm, gfn);
Izik Eidus28430992008-10-03 17:40:32 +03001412 slot = gfn_to_memslot_unaliased(kvm, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001413 if (!slot)
1414 return bad_hva();
1415 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1416}
Sheng Yang0d150292008-04-25 21:44:50 +08001417EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001418
Anthony Liguori35149e22008-04-02 14:46:56 -05001419pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
Avi Kivity954bbbc22007-03-30 14:02:32 +03001420{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001421 struct page *page[1];
Izik Eidus539cb662007-11-11 22:05:04 +02001422 unsigned long addr;
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001423 int npages;
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001424 pfn_t pfn;
Avi Kivity954bbbc22007-03-30 14:02:32 +03001425
Avi Kivity60395222007-10-21 11:03:36 +02001426 might_sleep();
1427
Izik Eidus539cb662007-11-11 22:05:04 +02001428 addr = gfn_to_hva(kvm, gfn);
1429 if (kvm_is_error_hva(addr)) {
Izik Eidus8a7ae052007-10-18 11:09:33 +02001430 get_page(bad_page);
Anthony Liguori35149e22008-04-02 14:46:56 -05001431 return page_to_pfn(bad_page);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001432 }
Izik Eidus8a7ae052007-10-18 11:09:33 +02001433
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03001434 npages = get_user_pages_fast(addr, 1, 1, page);
Izik Eidus539cb662007-11-11 22:05:04 +02001435
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001436 if (unlikely(npages != 1)) {
1437 struct vm_area_struct *vma;
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001438
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03001439 down_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001440 vma = find_vma(current->mm, addr);
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03001441
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001442 if (vma == NULL || addr < vma->vm_start ||
1443 !(vma->vm_flags & VM_PFNMAP)) {
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03001444 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001445 get_page(bad_page);
1446 return page_to_pfn(bad_page);
1447 }
1448
1449 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Marcelo Tosatti4c2155c2008-09-16 20:54:47 -03001450 up_read(&current->mm->mmap_sem);
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001451 BUG_ON(!kvm_is_mmio_pfn(pfn));
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001452 } else
1453 pfn = page_to_pfn(page[0]);
1454
1455 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001456}
1457
1458EXPORT_SYMBOL_GPL(gfn_to_pfn);
1459
1460struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1461{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001462 pfn_t pfn;
1463
1464 pfn = gfn_to_pfn(kvm, gfn);
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001465 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001466 return pfn_to_page(pfn);
1467
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001468 WARN_ON(kvm_is_mmio_pfn(pfn));
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001469
1470 get_page(bad_page);
1471 return bad_page;
Avi Kivity954bbbc22007-03-30 14:02:32 +03001472}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001473
Avi Kivity954bbbc22007-03-30 14:02:32 +03001474EXPORT_SYMBOL_GPL(gfn_to_page);
1475
Izik Eidusb4231d62007-11-20 11:49:33 +02001476void kvm_release_page_clean(struct page *page)
1477{
Anthony Liguori35149e22008-04-02 14:46:56 -05001478 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001479}
1480EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1481
Anthony Liguori35149e22008-04-02 14:46:56 -05001482void kvm_release_pfn_clean(pfn_t pfn)
1483{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001484 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001485 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001486}
1487EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1488
Izik Eidusb4231d62007-11-20 11:49:33 +02001489void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001490{
Anthony Liguori35149e22008-04-02 14:46:56 -05001491 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001492}
Izik Eidusb4231d62007-11-20 11:49:33 +02001493EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001494
Anthony Liguori35149e22008-04-02 14:46:56 -05001495void kvm_release_pfn_dirty(pfn_t pfn)
1496{
1497 kvm_set_pfn_dirty(pfn);
1498 kvm_release_pfn_clean(pfn);
1499}
1500EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1501
1502void kvm_set_page_dirty(struct page *page)
1503{
1504 kvm_set_pfn_dirty(page_to_pfn(page));
1505}
1506EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1507
1508void kvm_set_pfn_dirty(pfn_t pfn)
1509{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001510 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001511 struct page *page = pfn_to_page(pfn);
1512 if (!PageReserved(page))
1513 SetPageDirty(page);
1514 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001515}
1516EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1517
1518void kvm_set_pfn_accessed(pfn_t pfn)
1519{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001520 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001521 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001522}
1523EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1524
1525void kvm_get_pfn(pfn_t pfn)
1526{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001527 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001528 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001529}
1530EXPORT_SYMBOL_GPL(kvm_get_pfn);
1531
Izik Eidus195aefd2007-10-01 22:14:18 +02001532static int next_segment(unsigned long len, int offset)
1533{
1534 if (len > PAGE_SIZE - offset)
1535 return PAGE_SIZE - offset;
1536 else
1537 return len;
1538}
1539
1540int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1541 int len)
1542{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001543 int r;
1544 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001545
Izik Eiduse0506bc2007-11-11 22:10:22 +02001546 addr = gfn_to_hva(kvm, gfn);
1547 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001548 return -EFAULT;
Izik Eiduse0506bc2007-11-11 22:10:22 +02001549 r = copy_from_user(data, (void __user *)addr + offset, len);
1550 if (r)
1551 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001552 return 0;
1553}
1554EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1555
1556int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1557{
1558 gfn_t gfn = gpa >> PAGE_SHIFT;
1559 int seg;
1560 int offset = offset_in_page(gpa);
1561 int ret;
1562
1563 while ((seg = next_segment(len, offset)) != 0) {
1564 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1565 if (ret < 0)
1566 return ret;
1567 offset = 0;
1568 len -= seg;
1569 data += seg;
1570 ++gfn;
1571 }
1572 return 0;
1573}
1574EXPORT_SYMBOL_GPL(kvm_read_guest);
1575
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001576int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1577 unsigned long len)
1578{
1579 int r;
1580 unsigned long addr;
1581 gfn_t gfn = gpa >> PAGE_SHIFT;
1582 int offset = offset_in_page(gpa);
1583
1584 addr = gfn_to_hva(kvm, gfn);
1585 if (kvm_is_error_hva(addr))
1586 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001587 pagefault_disable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001588 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001589 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001590 if (r)
1591 return -EFAULT;
1592 return 0;
1593}
1594EXPORT_SYMBOL(kvm_read_guest_atomic);
1595
Izik Eidus195aefd2007-10-01 22:14:18 +02001596int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1597 int offset, int len)
1598{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001599 int r;
1600 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001601
Izik Eiduse0506bc2007-11-11 22:10:22 +02001602 addr = gfn_to_hva(kvm, gfn);
1603 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001604 return -EFAULT;
Izik Eiduse0506bc2007-11-11 22:10:22 +02001605 r = copy_to_user((void __user *)addr + offset, data, len);
1606 if (r)
1607 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001608 mark_page_dirty(kvm, gfn);
1609 return 0;
1610}
1611EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1612
1613int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1614 unsigned long len)
1615{
1616 gfn_t gfn = gpa >> PAGE_SHIFT;
1617 int seg;
1618 int offset = offset_in_page(gpa);
1619 int ret;
1620
1621 while ((seg = next_segment(len, offset)) != 0) {
1622 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1623 if (ret < 0)
1624 return ret;
1625 offset = 0;
1626 len -= seg;
1627 data += seg;
1628 ++gfn;
1629 }
1630 return 0;
1631}
1632
1633int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1634{
Izik Eidus3e021bf2007-11-19 11:16:57 +02001635 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001636}
1637EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1638
1639int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1640{
1641 gfn_t gfn = gpa >> PAGE_SHIFT;
1642 int seg;
1643 int offset = offset_in_page(gpa);
1644 int ret;
1645
1646 while ((seg = next_segment(len, offset)) != 0) {
1647 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1648 if (ret < 0)
1649 return ret;
1650 offset = 0;
1651 len -= seg;
1652 ++gfn;
1653 }
1654 return 0;
1655}
1656EXPORT_SYMBOL_GPL(kvm_clear_guest);
1657
Avi Kivity6aa8b732006-12-10 02:21:36 -08001658void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1659{
Nguyen Anh Quynh31389942007-06-05 10:35:19 +03001660 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001661
Uri Lublin3b6fff12007-10-30 10:42:09 +02001662 gfn = unalias_gfn(kvm, gfn);
Izik Eidus28430992008-10-03 17:40:32 +03001663 memslot = gfn_to_memslot_unaliased(kvm, gfn);
Rusty Russell7e9d6192007-07-31 20:41:14 +10001664 if (memslot && memslot->dirty_bitmap) {
1665 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001666
Rusty Russell7e9d6192007-07-31 20:41:14 +10001667 /* avoid RMW */
1668 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1669 set_bit(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001670 }
1671}
1672
Eddie Dongb6958ce2007-07-18 12:15:21 +03001673/*
1674 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1675 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001676void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001677{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001678 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001679
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001680 for (;;) {
1681 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001682
Gleb Natapova1b37102009-07-09 15:33:52 +03001683 if (kvm_arch_vcpu_runnable(vcpu)) {
Marcelo Tosattid7690172008-09-08 15:23:48 -03001684 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001685 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001686 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001687 if (kvm_cpu_has_pending_timer(vcpu))
1688 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001689 if (signal_pending(current))
1690 break;
1691
Eddie Dongb6958ce2007-07-18 12:15:21 +03001692 vcpu_put(vcpu);
1693 schedule();
1694 vcpu_load(vcpu);
1695 }
1696
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001697 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001698}
1699
Avi Kivity6aa8b732006-12-10 02:21:36 -08001700void kvm_resched(struct kvm_vcpu *vcpu)
1701{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001702 if (!need_resched())
1703 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001704 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001705}
1706EXPORT_SYMBOL_GPL(kvm_resched);
1707
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001708static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001709{
1710 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001711 struct page *page;
1712
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001713 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001714 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001715#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001716 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001717 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001718#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001719#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1720 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1721 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1722#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001723 else
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001724 return VM_FAULT_SIGBUS;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001725 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001726 vmf->page = page;
1727 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001728}
1729
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001730static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001731 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001732};
1733
1734static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1735{
1736 vma->vm_ops = &kvm_vcpu_vm_ops;
1737 return 0;
1738}
1739
Avi Kivitybccf2152007-02-21 18:04:26 +02001740static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1741{
1742 struct kvm_vcpu *vcpu = filp->private_data;
1743
Al Viro66c0b392008-04-19 20:33:56 +01001744 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001745 return 0;
1746}
1747
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001748static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001749 .release = kvm_vcpu_release,
1750 .unlocked_ioctl = kvm_vcpu_ioctl,
1751 .compat_ioctl = kvm_vcpu_ioctl,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001752 .mmap = kvm_vcpu_mmap,
Avi Kivitybccf2152007-02-21 18:04:26 +02001753};
1754
1755/*
1756 * Allocates an inode for the vcpu.
1757 */
1758static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1759{
Gleb Natapov73880c82009-06-09 15:56:28 +03001760 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
Avi Kivitybccf2152007-02-21 18:04:26 +02001761}
1762
Avi Kivityc5ea7662007-02-20 18:41:05 +02001763/*
1764 * Creates some virtual cpus. Good luck creating more than one.
1765 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001766static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001767{
1768 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001769 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001770
Gleb Natapov73880c82009-06-09 15:56:28 +03001771 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001772 if (IS_ERR(vcpu))
1773 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001774
Avi Kivity15ad7142007-07-11 18:17:21 +03001775 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1776
Avi Kivity26e52152007-11-20 15:30:24 +02001777 r = kvm_arch_vcpu_setup(vcpu);
1778 if (r)
Glauber Costa7d8fece2008-09-17 23:16:59 -03001779 return r;
Avi Kivity26e52152007-11-20 15:30:24 +02001780
Shaohua Li11ec2802007-07-23 14:51:37 +08001781 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03001782 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1783 r = -EINVAL;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08001784 goto vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001785 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001786
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001787 kvm_for_each_vcpu(r, v, kvm)
1788 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001789 r = -EEXIST;
1790 goto vcpu_destroy;
1791 }
1792
1793 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001794
1795 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001796 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001797 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001798 if (r < 0) {
1799 kvm_put_kvm(kvm);
1800 goto vcpu_destroy;
1801 }
1802
1803 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1804 smp_wmb();
1805 atomic_inc(&kvm->online_vcpus);
1806
1807#ifdef CONFIG_KVM_APIC_ARCHITECTURE
1808 if (kvm->bsp_vcpu_id == id)
1809 kvm->bsp_vcpu = vcpu;
1810#endif
1811 mutex_unlock(&kvm->lock);
Avi Kivitybccf2152007-02-21 18:04:26 +02001812 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001813
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08001814vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001815 mutex_unlock(&kvm->lock);
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001816 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001817 return r;
1818}
1819
Avi Kivity1961d272007-03-05 19:46:05 +02001820static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1821{
1822 if (sigset) {
1823 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1824 vcpu->sigset_active = 1;
1825 vcpu->sigset = *sigset;
1826 } else
1827 vcpu->sigset_active = 0;
1828 return 0;
1829}
1830
Sheng Yangc1e01512009-02-25 17:22:26 +08001831#ifdef __KVM_HAVE_MSIX
1832static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1833 struct kvm_assigned_msix_nr *entry_nr)
1834{
1835 int r = 0;
1836 struct kvm_assigned_dev_kernel *adev;
1837
1838 mutex_lock(&kvm->lock);
1839
1840 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1841 entry_nr->assigned_dev_id);
1842 if (!adev) {
1843 r = -EINVAL;
1844 goto msix_nr_out;
1845 }
1846
1847 if (adev->entries_nr == 0) {
1848 adev->entries_nr = entry_nr->entry_nr;
1849 if (adev->entries_nr == 0 ||
1850 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1851 r = -EINVAL;
1852 goto msix_nr_out;
1853 }
1854
1855 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1856 entry_nr->entry_nr,
1857 GFP_KERNEL);
1858 if (!adev->host_msix_entries) {
1859 r = -ENOMEM;
1860 goto msix_nr_out;
1861 }
1862 adev->guest_msix_entries = kzalloc(
1863 sizeof(struct kvm_guest_msix_entry) *
1864 entry_nr->entry_nr, GFP_KERNEL);
1865 if (!adev->guest_msix_entries) {
1866 kfree(adev->host_msix_entries);
1867 r = -ENOMEM;
1868 goto msix_nr_out;
1869 }
1870 } else /* Not allowed set MSI-X number twice */
1871 r = -EINVAL;
1872msix_nr_out:
1873 mutex_unlock(&kvm->lock);
1874 return r;
1875}
1876
1877static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1878 struct kvm_assigned_msix_entry *entry)
1879{
1880 int r = 0, i;
1881 struct kvm_assigned_dev_kernel *adev;
1882
1883 mutex_lock(&kvm->lock);
1884
1885 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1886 entry->assigned_dev_id);
1887
1888 if (!adev) {
1889 r = -EINVAL;
1890 goto msix_entry_out;
1891 }
1892
1893 for (i = 0; i < adev->entries_nr; i++)
1894 if (adev->guest_msix_entries[i].vector == 0 ||
1895 adev->guest_msix_entries[i].entry == entry->entry) {
1896 adev->guest_msix_entries[i].entry = entry->entry;
1897 adev->guest_msix_entries[i].vector = entry->gsi;
1898 adev->host_msix_entries[i].entry = entry->entry;
1899 break;
1900 }
1901 if (i == adev->entries_nr) {
1902 r = -ENOSPC;
1903 goto msix_entry_out;
1904 }
1905
1906msix_entry_out:
1907 mutex_unlock(&kvm->lock);
1908
1909 return r;
1910}
1911#endif
1912
Avi Kivitybccf2152007-02-21 18:04:26 +02001913static long kvm_vcpu_ioctl(struct file *filp,
1914 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001915{
Avi Kivitybccf2152007-02-21 18:04:26 +02001916 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f366982007-02-09 16:38:35 +00001917 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001918 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001919 struct kvm_fpu *fpu = NULL;
1920 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001921
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001922 if (vcpu->kvm->mm != current->mm)
1923 return -EIO;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001924 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001925 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001926 r = -EINVAL;
1927 if (arg)
1928 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001929 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001930 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001931 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001932 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001933
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001934 r = -ENOMEM;
1935 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1936 if (!kvm_regs)
1937 goto out;
1938 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001939 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001940 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001941 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001942 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1943 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001944 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001945out_free1:
1946 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001947 break;
1948 }
1949 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001950 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001951
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001952 r = -ENOMEM;
1953 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1954 if (!kvm_regs)
1955 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001956 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001957 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1958 goto out_free2;
1959 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001960 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001961 goto out_free2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001962 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001963out_free2:
1964 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001965 break;
1966 }
1967 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07001968 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1969 r = -ENOMEM;
1970 if (!kvm_sregs)
1971 goto out;
1972 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001973 if (r)
1974 goto out;
1975 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001976 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001977 goto out;
1978 r = 0;
1979 break;
1980 }
1981 case KVM_SET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07001982 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1983 r = -ENOMEM;
1984 if (!kvm_sregs)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001985 goto out;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001986 r = -EFAULT;
1987 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1988 goto out;
1989 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001990 if (r)
1991 goto out;
1992 r = 0;
1993 break;
1994 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03001995 case KVM_GET_MP_STATE: {
1996 struct kvm_mp_state mp_state;
1997
1998 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1999 if (r)
2000 goto out;
2001 r = -EFAULT;
2002 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2003 goto out;
2004 r = 0;
2005 break;
2006 }
2007 case KVM_SET_MP_STATE: {
2008 struct kvm_mp_state mp_state;
2009
2010 r = -EFAULT;
2011 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2012 goto out;
2013 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2014 if (r)
2015 goto out;
2016 r = 0;
2017 break;
2018 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002019 case KVM_TRANSLATE: {
2020 struct kvm_translation tr;
2021
2022 r = -EFAULT;
Al Viro2f366982007-02-09 16:38:35 +00002023 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002024 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002025 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002026 if (r)
2027 goto out;
2028 r = -EFAULT;
Al Viro2f366982007-02-09 16:38:35 +00002029 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002030 goto out;
2031 r = 0;
2032 break;
2033 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002034 case KVM_SET_GUEST_DEBUG: {
2035 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002036
2037 r = -EFAULT;
Al Viro2f366982007-02-09 16:38:35 +00002038 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002039 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002040 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002041 if (r)
2042 goto out;
2043 r = 0;
2044 break;
2045 }
Avi Kivity1961d272007-03-05 19:46:05 +02002046 case KVM_SET_SIGNAL_MASK: {
2047 struct kvm_signal_mask __user *sigmask_arg = argp;
2048 struct kvm_signal_mask kvm_sigmask;
2049 sigset_t sigset, *p;
2050
2051 p = NULL;
2052 if (argp) {
2053 r = -EFAULT;
2054 if (copy_from_user(&kvm_sigmask, argp,
2055 sizeof kvm_sigmask))
2056 goto out;
2057 r = -EINVAL;
2058 if (kvm_sigmask.len != sizeof sigset)
2059 goto out;
2060 r = -EFAULT;
2061 if (copy_from_user(&sigset, sigmask_arg->sigset,
2062 sizeof sigset))
2063 goto out;
2064 p = &sigset;
2065 }
2066 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2067 break;
2068 }
Avi Kivityb8836732007-04-01 16:34:31 +03002069 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002070 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2071 r = -ENOMEM;
2072 if (!fpu)
2073 goto out;
2074 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002075 if (r)
2076 goto out;
2077 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002078 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002079 goto out;
2080 r = 0;
2081 break;
2082 }
2083 case KVM_SET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002084 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2085 r = -ENOMEM;
2086 if (!fpu)
Avi Kivityb8836732007-04-01 16:34:31 +03002087 goto out;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002088 r = -EFAULT;
2089 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2090 goto out;
2091 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002092 if (r)
2093 goto out;
2094 r = 0;
2095 break;
2096 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002097 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002098 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002099 }
2100out:
Dave Hansenfa3795a2008-08-11 10:01:46 -07002101 kfree(fpu);
2102 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002103 return r;
2104}
2105
2106static long kvm_vm_ioctl(struct file *filp,
2107 unsigned int ioctl, unsigned long arg)
2108{
2109 struct kvm *kvm = filp->private_data;
2110 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002111 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002112
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002113 if (kvm->mm != current->mm)
2114 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002115 switch (ioctl) {
2116 case KVM_CREATE_VCPU:
2117 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2118 if (r < 0)
2119 goto out;
2120 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002121 case KVM_SET_USER_MEMORY_REGION: {
2122 struct kvm_userspace_memory_region kvm_userspace_mem;
2123
2124 r = -EFAULT;
2125 if (copy_from_user(&kvm_userspace_mem, argp,
2126 sizeof kvm_userspace_mem))
2127 goto out;
2128
2129 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002130 if (r)
2131 goto out;
2132 break;
2133 }
2134 case KVM_GET_DIRTY_LOG: {
2135 struct kvm_dirty_log log;
2136
2137 r = -EFAULT;
Al Viro2f366982007-02-09 16:38:35 +00002138 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002139 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002140 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002141 if (r)
2142 goto out;
2143 break;
2144 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002145#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2146 case KVM_REGISTER_COALESCED_MMIO: {
2147 struct kvm_coalesced_mmio_zone zone;
2148 r = -EFAULT;
2149 if (copy_from_user(&zone, argp, sizeof zone))
2150 goto out;
2151 r = -ENXIO;
2152 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2153 if (r)
2154 goto out;
2155 r = 0;
2156 break;
2157 }
2158 case KVM_UNREGISTER_COALESCED_MMIO: {
2159 struct kvm_coalesced_mmio_zone zone;
2160 r = -EFAULT;
2161 if (copy_from_user(&zone, argp, sizeof zone))
2162 goto out;
2163 r = -ENXIO;
2164 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2165 if (r)
2166 goto out;
2167 r = 0;
2168 break;
2169 }
2170#endif
Xiantao Zhang8a98f662008-10-06 13:47:38 +08002171#ifdef KVM_CAP_DEVICE_ASSIGNMENT
2172 case KVM_ASSIGN_PCI_DEVICE: {
2173 struct kvm_assigned_pci_dev assigned_dev;
2174
2175 r = -EFAULT;
2176 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2177 goto out;
2178 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2179 if (r)
2180 goto out;
2181 break;
2182 }
2183 case KVM_ASSIGN_IRQ: {
Sheng Yange56d5322009-03-12 21:45:39 +08002184 r = -EOPNOTSUPP;
2185 break;
2186 }
2187#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2188 case KVM_ASSIGN_DEV_IRQ: {
Xiantao Zhang8a98f662008-10-06 13:47:38 +08002189 struct kvm_assigned_irq assigned_irq;
2190
2191 r = -EFAULT;
2192 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2193 goto out;
2194 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2195 if (r)
2196 goto out;
2197 break;
2198 }
Sheng Yange56d5322009-03-12 21:45:39 +08002199 case KVM_DEASSIGN_DEV_IRQ: {
2200 struct kvm_assigned_irq assigned_irq;
2201
2202 r = -EFAULT;
2203 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2204 goto out;
2205 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2206 if (r)
2207 goto out;
2208 break;
2209 }
2210#endif
Xiantao Zhang8a98f662008-10-06 13:47:38 +08002211#endif
Weidong Han0a920352008-12-02 21:24:23 +08002212#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2213 case KVM_DEASSIGN_PCI_DEVICE: {
2214 struct kvm_assigned_pci_dev assigned_dev;
2215
2216 r = -EFAULT;
2217 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2218 goto out;
2219 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2220 if (r)
2221 goto out;
2222 break;
2223 }
2224#endif
Avi Kivity399ec802008-11-19 13:58:46 +02002225#ifdef KVM_CAP_IRQ_ROUTING
2226 case KVM_SET_GSI_ROUTING: {
2227 struct kvm_irq_routing routing;
2228 struct kvm_irq_routing __user *urouting;
2229 struct kvm_irq_routing_entry *entries;
2230
2231 r = -EFAULT;
2232 if (copy_from_user(&routing, argp, sizeof(routing)))
2233 goto out;
2234 r = -EINVAL;
2235 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2236 goto out;
2237 if (routing.flags)
2238 goto out;
2239 r = -ENOMEM;
2240 entries = vmalloc(routing.nr * sizeof(*entries));
2241 if (!entries)
2242 goto out;
2243 r = -EFAULT;
2244 urouting = argp;
2245 if (copy_from_user(entries, urouting->entries,
2246 routing.nr * sizeof(*entries)))
2247 goto out_free_irq_routing;
2248 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2249 routing.flags);
2250 out_free_irq_routing:
2251 vfree(entries);
2252 break;
2253 }
Avi Kivity6621fbc2009-08-10 17:00:52 +03002254#endif /* KVM_CAP_IRQ_ROUTING */
Sheng Yangc1e01512009-02-25 17:22:26 +08002255#ifdef __KVM_HAVE_MSIX
2256 case KVM_ASSIGN_SET_MSIX_NR: {
2257 struct kvm_assigned_msix_nr entry_nr;
2258 r = -EFAULT;
2259 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2260 goto out;
2261 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2262 if (r)
2263 goto out;
2264 break;
2265 }
2266 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2267 struct kvm_assigned_msix_entry entry;
2268 r = -EFAULT;
2269 if (copy_from_user(&entry, argp, sizeof entry))
2270 goto out;
2271 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2272 if (r)
2273 goto out;
2274 break;
2275 }
Avi Kivity399ec802008-11-19 13:58:46 +02002276#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002277 case KVM_IRQFD: {
2278 struct kvm_irqfd data;
2279
2280 r = -EFAULT;
2281 if (copy_from_user(&data, argp, sizeof data))
2282 goto out;
2283 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2284 break;
2285 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002286 case KVM_IOEVENTFD: {
2287 struct kvm_ioeventfd data;
2288
2289 r = -EFAULT;
2290 if (copy_from_user(&data, argp, sizeof data))
2291 goto out;
2292 r = kvm_ioeventfd(kvm, &data);
2293 break;
2294 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002295#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2296 case KVM_SET_BOOT_CPU_ID:
2297 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002298 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002299 if (atomic_read(&kvm->online_vcpus) != 0)
2300 r = -EBUSY;
2301 else
2302 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002303 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002304 break;
2305#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002306 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002307 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002308 }
2309out:
2310 return r;
2311}
2312
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002313static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002314{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002315 struct page *page[1];
2316 unsigned long addr;
2317 int npages;
2318 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002319 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002320
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002321 addr = gfn_to_hva(kvm, gfn);
2322 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002323 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002324
2325 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2326 NULL);
2327 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002328 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002329
2330 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002331 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002332}
2333
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002334static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002335 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002336};
2337
2338static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2339{
2340 vma->vm_ops = &kvm_vm_vm_ops;
2341 return 0;
2342}
2343
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002344static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002345 .release = kvm_vm_release,
2346 .unlocked_ioctl = kvm_vm_ioctl,
2347 .compat_ioctl = kvm_vm_ioctl,
2348 .mmap = kvm_vm_mmap,
2349};
2350
2351static int kvm_dev_ioctl_create_vm(void)
2352{
Al Viro2030a422008-02-23 06:46:49 -05002353 int fd;
Avi Kivityf17abe92007-02-21 19:28:04 +02002354 struct kvm *kvm;
2355
Avi Kivityf17abe92007-02-21 19:28:04 +02002356 kvm = kvm_create_vm();
Avi Kivityd6d28162007-06-28 08:38:16 -04002357 if (IS_ERR(kvm))
2358 return PTR_ERR(kvm);
Ulrich Drepper7d9dbca2008-07-23 21:29:22 -07002359 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
Al Viro2030a422008-02-23 06:46:49 -05002360 if (fd < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002361 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002362
Avi Kivityf17abe92007-02-21 19:28:04 +02002363 return fd;
Avi Kivityf17abe92007-02-21 19:28:04 +02002364}
2365
Avi Kivity1a811b62008-12-08 18:25:27 +02002366static long kvm_dev_ioctl_check_extension_generic(long arg)
2367{
2368 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002369 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002370 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002371 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002372#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID:
2374#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002375 return 1;
Avi Kivity399ec802008-11-19 13:58:46 +02002376#ifdef CONFIG_HAVE_KVM_IRQCHIP
2377 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002378 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002379#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002380 default:
2381 break;
2382 }
2383 return kvm_dev_ioctl_check_extension(arg);
2384}
2385
Avi Kivityf17abe92007-02-21 19:28:04 +02002386static long kvm_dev_ioctl(struct file *filp,
2387 unsigned int ioctl, unsigned long arg)
2388{
Avi Kivity07c45a32007-03-07 13:05:38 +02002389 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002390
2391 switch (ioctl) {
2392 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002393 r = -EINVAL;
2394 if (arg)
2395 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002396 r = KVM_API_VERSION;
2397 break;
2398 case KVM_CREATE_VM:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002399 r = -EINVAL;
2400 if (arg)
2401 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002402 r = kvm_dev_ioctl_create_vm();
2403 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002404 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002405 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002406 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002407 case KVM_GET_VCPU_MMAP_SIZE:
2408 r = -EINVAL;
2409 if (arg)
2410 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002411 r = PAGE_SIZE; /* struct kvm_run */
2412#ifdef CONFIG_X86
2413 r += PAGE_SIZE; /* pio data page */
2414#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002415#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2416 r += PAGE_SIZE; /* coalesced mmio ring page */
2417#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002418 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002419 case KVM_TRACE_ENABLE:
2420 case KVM_TRACE_PAUSE:
2421 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002422 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002423 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002424 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002425 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002426 }
2427out:
2428 return r;
2429}
2430
Avi Kivity6aa8b732006-12-10 02:21:36 -08002431static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002432 .unlocked_ioctl = kvm_dev_ioctl,
2433 .compat_ioctl = kvm_dev_ioctl,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002434};
2435
2436static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002437 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002438 "kvm",
2439 &kvm_chardev_ops,
2440};
2441
Avi Kivity1b6c0162007-05-24 13:03:52 +03002442static void hardware_enable(void *junk)
2443{
2444 int cpu = raw_smp_processor_id();
2445
Rusty Russell7f59f492008-12-07 21:25:45 +10302446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002447 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302448 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002449 kvm_arch_hardware_enable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002450}
2451
2452static void hardware_disable(void *junk)
2453{
2454 int cpu = raw_smp_processor_id();
2455
Rusty Russell7f59f492008-12-07 21:25:45 +10302456 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002457 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302458 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002459 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002460}
2461
Avi Kivity774c47f2007-02-12 00:54:47 -08002462static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2463 void *v)
2464{
2465 int cpu = (long)v;
2466
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002467 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002468 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002469 case CPU_DYING:
Avi Kivity6ec8a852007-08-19 15:57:26 +03002470 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2471 cpu);
2472 hardware_disable(NULL);
2473 break;
Avi Kivity774c47f2007-02-12 00:54:47 -08002474 case CPU_UP_CANCELED:
Jeremy Katz43934a32007-02-19 14:37:46 +02002475 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2476 cpu);
Jens Axboe8691e5a2008-06-06 11:18:06 +02002477 smp_call_function_single(cpu, hardware_disable, NULL, 1);
Avi Kivity774c47f2007-02-12 00:54:47 -08002478 break;
Jeremy Katz43934a32007-02-19 14:37:46 +02002479 case CPU_ONLINE:
2480 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2481 cpu);
Jens Axboe8691e5a2008-06-06 11:18:06 +02002482 smp_call_function_single(cpu, hardware_enable, NULL, 1);
Avi Kivity774c47f2007-02-12 00:54:47 -08002483 break;
2484 }
2485 return NOTIFY_OK;
2486}
2487
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002488
2489asmlinkage void kvm_handle_fault_on_reboot(void)
2490{
2491 if (kvm_rebooting)
2492 /* spin while reset goes on */
2493 while (true)
2494 ;
2495 /* Fault while not rebooting. We want the trace. */
2496 BUG();
2497}
2498EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2499
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002500static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002501 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002502{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002503 /*
2504 * Some (well, at least mine) BIOSes hang on reboot if
2505 * in vmx root mode.
2506 *
2507 * And Intel TXT required VMX off for all cpu when system shutdown.
2508 */
2509 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2510 kvm_rebooting = true;
2511 on_each_cpu(hardware_disable, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002512 return NOTIFY_OK;
2513}
2514
2515static struct notifier_block kvm_reboot_notifier = {
2516 .notifier_call = kvm_reboot,
2517 .priority = 0,
2518};
2519
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002520void kvm_io_bus_init(struct kvm_io_bus *bus)
2521{
2522 memset(bus, 0, sizeof(*bus));
2523}
2524
2525void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2526{
2527 int i;
2528
2529 for (i = 0; i < bus->dev_count; i++) {
2530 struct kvm_io_device *pos = bus->devs[i];
2531
2532 kvm_iodevice_destructor(pos);
2533 }
2534}
2535
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002536/* kvm_io_bus_write - called under kvm->slots_lock */
2537int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2538 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002539{
2540 int i;
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002541 for (i = 0; i < bus->dev_count; i++)
2542 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2543 return 0;
2544 return -EOPNOTSUPP;
2545}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002546
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002547/* kvm_io_bus_read - called under kvm->slots_lock */
2548int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2549{
2550 int i;
2551 for (i = 0; i < bus->dev_count; i++)
2552 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2553 return 0;
2554 return -EOPNOTSUPP;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002555}
2556
Gregory Haskins090b7af2009-07-07 17:08:44 -04002557int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002558 struct kvm_io_device *dev)
2559{
Gregory Haskins090b7af2009-07-07 17:08:44 -04002560 int ret;
2561
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002562 down_write(&kvm->slots_lock);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002563 ret = __kvm_io_bus_register_dev(bus, dev);
2564 up_write(&kvm->slots_lock);
2565
2566 return ret;
2567}
2568
2569/* An unlocked version. Caller must have write lock on slots_lock. */
2570int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2571 struct kvm_io_device *dev)
2572{
2573 if (bus->dev_count > NR_IOBUS_DEVS-1)
2574 return -ENOSPC;
2575
2576 bus->devs[bus->dev_count++] = dev;
2577
2578 return 0;
2579}
2580
2581void kvm_io_bus_unregister_dev(struct kvm *kvm,
2582 struct kvm_io_bus *bus,
2583 struct kvm_io_device *dev)
2584{
2585 down_write(&kvm->slots_lock);
2586 __kvm_io_bus_unregister_dev(bus, dev);
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002587 up_write(&kvm->slots_lock);
2588}
2589
2590/* An unlocked version. Caller must have write lock on slots_lock. */
Gregory Haskins090b7af2009-07-07 17:08:44 -04002591void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2592 struct kvm_io_device *dev)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002593{
Gregory Haskins090b7af2009-07-07 17:08:44 -04002594 int i;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002595
Gregory Haskins090b7af2009-07-07 17:08:44 -04002596 for (i = 0; i < bus->dev_count; i++)
2597 if (bus->devs[i] == dev) {
2598 bus->devs[i] = bus->devs[--bus->dev_count];
2599 break;
2600 }
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002601}
2602
Avi Kivity774c47f2007-02-12 00:54:47 -08002603static struct notifier_block kvm_cpu_notifier = {
2604 .notifier_call = kvm_cpu_hotplug,
2605 .priority = 20, /* must be > scheduler priority */
2606};
2607
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002608static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02002609{
2610 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02002611 struct kvm *kvm;
2612
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002613 *val = 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02002614 spin_lock(&kvm_lock);
2615 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002616 *val += *(u32 *)((void *)kvm + offset);
Avi Kivityba1389b2007-11-18 16:24:12 +02002617 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002618 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02002619}
2620
2621DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2622
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002623static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03002624{
2625 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002626 struct kvm *kvm;
2627 struct kvm_vcpu *vcpu;
2628 int i;
2629
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002630 *val = 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002631 spin_lock(&kvm_lock);
2632 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002633 kvm_for_each_vcpu(i, vcpu, kvm)
2634 *val += *(u32 *)((void *)vcpu + offset);
2635
Avi Kivity1165f5f2007-04-19 17:27:43 +03002636 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002637 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002638}
2639
Avi Kivityba1389b2007-11-18 16:24:12 +02002640DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2641
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002642static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02002643 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2644 [KVM_STAT_VM] = &vm_stat_fops,
2645};
Avi Kivity1165f5f2007-04-19 17:27:43 +03002646
Zhang Xiantaoa16b0432007-11-16 14:38:21 +08002647static void kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002648{
2649 struct kvm_stats_debugfs_item *p;
2650
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002651 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002652 for (p = debugfs_entries; p->name; ++p)
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002653 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03002654 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02002655 stat_fops[p->kind]);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002656}
2657
2658static void kvm_exit_debug(void)
2659{
2660 struct kvm_stats_debugfs_item *p;
2661
2662 for (p = debugfs_entries; p->name; ++p)
2663 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002664 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002665}
2666
Avi Kivity59ae6c62007-02-12 00:54:48 -08002667static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2668{
Avi Kivity4267c412007-05-24 13:09:41 +03002669 hardware_disable(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002670 return 0;
2671}
2672
2673static int kvm_resume(struct sys_device *dev)
2674{
Avi Kivity4267c412007-05-24 13:09:41 +03002675 hardware_enable(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002676 return 0;
2677}
2678
2679static struct sysdev_class kvm_sysdev_class = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +01002680 .name = "kvm",
Avi Kivity59ae6c62007-02-12 00:54:48 -08002681 .suspend = kvm_suspend,
2682 .resume = kvm_resume,
2683};
2684
2685static struct sys_device kvm_sysdev = {
2686 .id = 0,
2687 .cls = &kvm_sysdev_class,
2688};
2689
Izik Eiduscea7bb22007-10-17 19:17:48 +02002690struct page *bad_page;
Anthony Liguori35149e22008-04-02 14:46:56 -05002691pfn_t bad_pfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002692
Avi Kivity15ad7142007-07-11 18:17:21 +03002693static inline
2694struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2695{
2696 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2697}
2698
2699static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2700{
2701 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2702
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002703 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002704}
2705
2706static void kvm_sched_out(struct preempt_notifier *pn,
2707 struct task_struct *next)
2708{
2709 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2710
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002711 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002712}
2713
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002714int kvm_init(void *opaque, unsigned int vcpu_size,
Rusty Russellc16f8622007-07-30 21:12:19 +10002715 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002716{
2717 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002718 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002719
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002720 kvm_init_debug();
2721
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002722 r = kvm_arch_init(opaque);
2723 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002724 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002725
2726 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2727
2728 if (bad_page == NULL) {
2729 r = -ENOMEM;
2730 goto out;
2731 }
2732
Anthony Liguori35149e22008-04-02 14:46:56 -05002733 bad_pfn = page_to_pfn(bad_page);
2734
Avi Kivity8437a612009-06-06 14:52:35 -07002735 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10302736 r = -ENOMEM;
2737 goto out_free_0;
2738 }
2739
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002740 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002741 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10302742 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002743
Yang, Sheng002c7f72007-07-31 14:23:01 +03002744 for_each_online_cpu(cpu) {
2745 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002746 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02002747 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03002748 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002749 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002750 }
2751
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002752 on_each_cpu(hardware_enable, NULL, 1);
Avi Kivity774c47f2007-02-12 00:54:47 -08002753 r = register_cpu_notifier(&kvm_cpu_notifier);
2754 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002755 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002756 register_reboot_notifier(&kvm_reboot_notifier);
2757
Avi Kivity59ae6c62007-02-12 00:54:48 -08002758 r = sysdev_class_register(&kvm_sysdev_class);
2759 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002760 goto out_free_3;
Avi Kivity59ae6c62007-02-12 00:54:48 -08002761
2762 r = sysdev_register(&kvm_sysdev);
2763 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002764 goto out_free_4;
Avi Kivity59ae6c62007-02-12 00:54:48 -08002765
Rusty Russellc16f8622007-07-30 21:12:19 +10002766 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2767 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
Joe Perches56919c52007-11-12 20:06:51 -08002768 __alignof__(struct kvm_vcpu),
2769 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10002770 if (!kvm_vcpu_cache) {
2771 r = -ENOMEM;
Zhang Xiantaod23087842007-11-29 15:35:39 +08002772 goto out_free_5;
Rusty Russellc16f8622007-07-30 21:12:19 +10002773 }
2774
Avi Kivity6aa8b732006-12-10 02:21:36 -08002775 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002776 kvm_vm_fops.owner = module;
2777 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002778
2779 r = misc_register(&kvm_dev);
2780 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04002781 printk(KERN_ERR "kvm: misc device register failed\n");
Avi Kivity6aa8b732006-12-10 02:21:36 -08002782 goto out_free;
2783 }
2784
Avi Kivity15ad7142007-07-11 18:17:21 +03002785 kvm_preempt_ops.sched_in = kvm_sched_in;
2786 kvm_preempt_ops.sched_out = kvm_sched_out;
2787
Avi Kivityc7addb92007-09-16 18:58:32 +02002788 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002789
2790out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10002791 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002792out_free_5:
Avi Kivity59ae6c62007-02-12 00:54:48 -08002793 sysdev_unregister(&kvm_sysdev);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002794out_free_4:
Avi Kivity59ae6c62007-02-12 00:54:48 -08002795 sysdev_class_unregister(&kvm_sysdev_class);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002796out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002797 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08002798 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002799out_free_2:
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002800 on_each_cpu(hardware_disable, NULL, 1);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002801out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002802 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10302803out_free_0a:
2804 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002805out_free_0:
2806 __free_page(bad_page);
Avi Kivityca45aaa2007-03-01 19:21:03 +02002807out:
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002808 kvm_arch_exit();
Zhang Xiantaod23087842007-11-29 15:35:39 +08002809out_fail:
Xiao Guangrongaed665f2009-08-03 17:10:06 +08002810 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002811 return r;
2812}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002813EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002814
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002815void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002816{
Marcelo Tosatti229456f2009-06-17 09:22:14 -03002817 tracepoint_synchronize_unregister();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002818 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10002819 kmem_cache_destroy(kvm_vcpu_cache);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002820 sysdev_unregister(&kvm_sysdev);
2821 sysdev_class_unregister(&kvm_sysdev_class);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002822 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002823 unregister_cpu_notifier(&kvm_cpu_notifier);
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002824 on_each_cpu(hardware_disable, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002825 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002826 kvm_arch_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002827 kvm_exit_debug();
Rusty Russell7f59f492008-12-07 21:25:45 +10302828 free_cpumask_var(cpus_hardware_enabled);
Izik Eiduscea7bb22007-10-17 19:17:48 +02002829 __free_page(bad_page);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002830}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002831EXPORT_SYMBOL_GPL(kvm_exit);