const: constify remaining file_operations
[linux-2.6.git] / virt / kvm / kvm_main.c
1 /*
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
18 #include "iodev.h"
19
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46
47 #include <asm/processor.h>
48 #include <asm/io.h>
49 #include <asm/uaccess.h>
50 #include <asm/pgtable.h>
51
52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53 #include "coalesced_mmio.h"
54 #endif
55
56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
57 #include <linux/pci.h>
58 #include <linux/interrupt.h>
59 #include "irq.h"
60 #endif
61
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/kvm.h>
64
65 MODULE_AUTHOR("Qumranet");
66 MODULE_LICENSE("GPL");
67
68 /*
69  * Ordering of locks:
70  *
71  *              kvm->slots_lock --> kvm->lock --> kvm->irq_lock
72  */
73
74 DEFINE_SPINLOCK(kvm_lock);
75 LIST_HEAD(vm_list);
76
77 static cpumask_var_t cpus_hardware_enabled;
78
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
83
84 struct dentry *kvm_debugfs_dir;
85
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87                            unsigned long arg);
88
89 static bool kvm_rebooting;
90
91 static bool largepages_enabled = true;
92
93 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
94 static 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
108 static 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
130 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
131 {
132         struct kvm_assigned_dev_kernel *assigned_dev;
133         struct kvm *kvm;
134         int i;
135
136         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
137                                     interrupt_work);
138         kvm = assigned_dev->kvm;
139
140         mutex_lock(&kvm->irq_lock);
141         spin_lock_irq(&assigned_dev->assigned_dev_lock);
142         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
143                 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);
153                 }
154         } else
155                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
156                             assigned_dev->guest_irq, 1);
157
158         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
159         mutex_unlock(&assigned_dev->kvm->irq_lock);
160 }
161
162 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
163 {
164         unsigned long flags;
165         struct kvm_assigned_dev_kernel *assigned_dev =
166                 (struct kvm_assigned_dev_kernel *) dev_id;
167
168         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
169         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
170                 int index = find_index_from_host_irq(assigned_dev, irq);
171                 if (index < 0)
172                         goto out;
173                 assigned_dev->guest_msix_entries[index].flags |=
174                         KVM_ASSIGNED_MSIX_PENDING;
175         }
176
177         schedule_work(&assigned_dev->interrupt_work);
178
179         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
180                 disable_irq_nosync(irq);
181                 assigned_dev->host_irq_disabled = true;
182         }
183
184 out:
185         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
186         return IRQ_HANDLED;
187 }
188
189 /* Ack the irq line for an assigned device */
190 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
191 {
192         struct kvm_assigned_dev_kernel *dev;
193         unsigned long flags;
194
195         if (kian->gsi == -1)
196                 return;
197
198         dev = container_of(kian, struct kvm_assigned_dev_kernel,
199                            ack_notifier);
200
201         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
202
203         /* The guest irq may be shared so this ack may be
204          * from another device.
205          */
206         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
207         if (dev->host_irq_disabled) {
208                 enable_irq(dev->host_irq);
209                 dev->host_irq_disabled = false;
210         }
211         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
212 }
213
214 static void deassign_guest_irq(struct kvm *kvm,
215                                struct kvm_assigned_dev_kernel *assigned_dev)
216 {
217         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
218         assigned_dev->ack_notifier.gsi = -1;
219
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;
223         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
224 }
225
226 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
227 static void deassign_host_irq(struct kvm *kvm,
228                               struct kvm_assigned_dev_kernel *assigned_dev)
229 {
230         /*
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          */
246         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
247                 int i;
248                 for (i = 0; i < assigned_dev->entries_nr; i++)
249                         disable_irq_nosync(assigned_dev->
250                                            host_msix_entries[i].vector);
251
252                 cancel_work_sync(&assigned_dev->interrupt_work);
253
254                 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
269                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
270                         pci_disable_msi(assigned_dev->dev);
271         }
272
273         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
274 }
275
276 static 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
299 static 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 }
304
305 static 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
311         pci_reset_function(assigned_dev->dev);
312
313         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
321 void 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
335 static int assigned_device_enable_host_intx(struct kvm *kvm,
336                                             struct kvm_assigned_dev_kernel *dev)
337 {
338         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;
347         return 0;
348 }
349
350 #ifdef __KVM_HAVE_MSI
351 static int assigned_device_enable_host_msi(struct kvm *kvm,
352                                            struct kvm_assigned_dev_kernel *dev)
353 {
354         int r;
355
356         if (!dev->dev->msi_enabled) {
357                 r = pci_enable_msi(dev->dev);
358                 if (r)
359                         return r;
360         }
361
362         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;
367         }
368
369         return 0;
370 }
371 #endif
372
373 #ifdef __KVM_HAVE_MSIX
374 static int assigned_device_enable_host_msix(struct kvm *kvm,
375                                             struct kvm_assigned_dev_kernel *dev)
376 {
377         int i, r = -EINVAL;
378
379         /* host_msix_entries and guest_msix_entries should have been
380          * initialized */
381         if (dev->entries_nr == 0)
382                 return r;
383
384         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
385         if (r)
386                 return r;
387
388         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 */
394                 if (r)
395                         return r;
396         }
397
398         return 0;
399 }
400
401 #endif
402
403 static 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
413 static 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;
419         dev->host_irq_disabled = false;
420         return 0;
421 }
422 #endif
423 #ifdef __KVM_HAVE_MSIX
424 static 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;
430         dev->host_irq_disabled = false;
431         return 0;
432 }
433 #endif
434
435 static int assign_host_irq(struct kvm *kvm,
436                            struct kvm_assigned_dev_kernel *dev,
437                            __u32 host_irq_type)
438 {
439         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
468 static 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 */
513 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
514                                    struct kvm_assigned_irq *assigned_irq)
515 {
516         int r = -EINVAL;
517         struct kvm_assigned_dev_kernel *match;
518         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);
553 out:
554         mutex_unlock(&kvm->lock);
555         return r;
556 }
557
558 static 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;
564
565         mutex_lock(&kvm->lock);
566
567         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
568                                       assigned_irq->assigned_dev_id);
569         if (!match)
570                 goto out;
571
572         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
573 out:
574         mutex_unlock(&kvm->lock);
575         return r;
576 }
577
578 static 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
585         down_read(&kvm->slots_lock);
586         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 */
592                 r = -EEXIST;
593                 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         }
621
622         pci_reset_function(dev);
623
624         match->assigned_dev_id = assigned_dev->assigned_dev_id;
625         match->host_busnr = assigned_dev->busnr;
626         match->host_devfn = assigned_dev->devfn;
627         match->flags = assigned_dev->flags;
628         match->dev = dev;
629         spin_lock_init(&match->assigned_dev_lock);
630         match->irq_source_id = -1;
631         match->kvm = kvm;
632         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
633         INIT_WORK(&match->interrupt_work,
634                   kvm_assigned_dev_interrupt_work_handler);
635
636         list_add(&match->list, &kvm->arch.assigned_dev_head);
637
638         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
639                 if (!kvm->arch.iommu_domain) {
640                         r = kvm_iommu_map_guest(kvm);
641                         if (r)
642                                 goto out_list_del;
643                 }
644                 r = kvm_assign_device(kvm, match);
645                 if (r)
646                         goto out_list_del;
647         }
648
649 out:
650         mutex_unlock(&kvm->lock);
651         up_read(&kvm->slots_lock);
652         return r;
653 out_list_del:
654         list_del(&match->list);
655         pci_release_regions(dev);
656 out_disable:
657         pci_disable_device(dev);
658 out_put:
659         pci_dev_put(dev);
660 out_free:
661         kfree(match);
662         mutex_unlock(&kvm->lock);
663         up_read(&kvm->slots_lock);
664         return r;
665 }
666 #endif
667
668 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
669 static 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
686         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
687                 kvm_deassign_device(kvm, match);
688
689         kvm_free_assigned_device(kvm, match);
690
691 out:
692         mutex_unlock(&kvm->lock);
693         return r;
694 }
695 #endif
696
697 inline int kvm_is_mmio_pfn(pfn_t pfn)
698 {
699         if (pfn_valid(pfn)) {
700                 struct page *page = compound_head(pfn_to_page(pfn));
701                 return PageReserved(page);
702         }
703
704         return true;
705 }
706
707 /*
708  * Switches to specified vcpu, until a matching vcpu_put()
709  */
710 void vcpu_load(struct kvm_vcpu *vcpu)
711 {
712         int cpu;
713
714         mutex_lock(&vcpu->mutex);
715         cpu = get_cpu();
716         preempt_notifier_register(&vcpu->preempt_notifier);
717         kvm_arch_vcpu_load(vcpu, cpu);
718         put_cpu();
719 }
720
721 void vcpu_put(struct kvm_vcpu *vcpu)
722 {
723         preempt_disable();
724         kvm_arch_vcpu_put(vcpu);
725         preempt_notifier_unregister(&vcpu->preempt_notifier);
726         preempt_enable();
727         mutex_unlock(&vcpu->mutex);
728 }
729
730 static void ack_flush(void *_completed)
731 {
732 }
733
734 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
735 {
736         int i, cpu, me;
737         cpumask_var_t cpus;
738         bool called = true;
739         struct kvm_vcpu *vcpu;
740
741         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
742
743         spin_lock(&kvm->requests_lock);
744         me = smp_processor_id();
745         kvm_for_each_vcpu(i, vcpu, kvm) {
746                 if (test_and_set_bit(req, &vcpu->requests))
747                         continue;
748                 cpu = vcpu->cpu;
749                 if (cpus != NULL && cpu != -1 && cpu != me)
750                         cpumask_set_cpu(cpu, cpus);
751         }
752         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;
758         spin_unlock(&kvm->requests_lock);
759         free_cpumask_var(cpus);
760         return called;
761 }
762
763 void 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;
767 }
768
769 void kvm_reload_remote_mmus(struct kvm *kvm)
770 {
771         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
772 }
773
774 int 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;
781         vcpu->kvm = kvm;
782         vcpu->vcpu_id = id;
783         init_waitqueue_head(&vcpu->wq);
784
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
792         r = kvm_arch_vcpu_init(vcpu);
793         if (r < 0)
794                 goto fail_free_run;
795         return 0;
796
797 fail_free_run:
798         free_page((unsigned long)vcpu->run);
799 fail:
800         return r;
801 }
802 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
803
804 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
805 {
806         kvm_arch_vcpu_uninit(vcpu);
807         free_page((unsigned long)vcpu->run);
808 }
809 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
810
811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
812 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
813 {
814         return container_of(mn, struct kvm, mmu_notifier);
815 }
816
817 static 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
853 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
854                                                     struct mm_struct *mm,
855                                                     unsigned long start,
856                                                     unsigned long end)
857 {
858         struct kvm *kvm = mmu_notifier_to_kvm(mn);
859         int need_tlb_flush = 0;
860
861         spin_lock(&kvm->mmu_lock);
862         /*
863          * The count increase must become visible at unlock time as no
864          * spte can be established without taking the mmu_lock and
865          * count is also read inside the mmu_lock critical section.
866          */
867         kvm->mmu_notifier_count++;
868         for (; start < end; start += PAGE_SIZE)
869                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
870         spin_unlock(&kvm->mmu_lock);
871
872         /* we've to flush the tlb before the pages can be freed */
873         if (need_tlb_flush)
874                 kvm_flush_remote_tlbs(kvm);
875 }
876
877 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
878                                                   struct mm_struct *mm,
879                                                   unsigned long start,
880                                                   unsigned long end)
881 {
882         struct kvm *kvm = mmu_notifier_to_kvm(mn);
883
884         spin_lock(&kvm->mmu_lock);
885         /*
886          * This sequence increase will notify the kvm page fault that
887          * the page that is going to be mapped in the spte could have
888          * been freed.
889          */
890         kvm->mmu_notifier_seq++;
891         /*
892          * The above sequence increase must be visible before the
893          * below count decrease but both values are read by the kvm
894          * page fault under mmu_lock spinlock so we don't need to add
895          * a smb_wmb() here in between the two.
896          */
897         kvm->mmu_notifier_count--;
898         spin_unlock(&kvm->mmu_lock);
899
900         BUG_ON(kvm->mmu_notifier_count < 0);
901 }
902
903 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
904                                               struct mm_struct *mm,
905                                               unsigned long address)
906 {
907         struct kvm *kvm = mmu_notifier_to_kvm(mn);
908         int young;
909
910         spin_lock(&kvm->mmu_lock);
911         young = kvm_age_hva(kvm, address);
912         spin_unlock(&kvm->mmu_lock);
913
914         if (young)
915                 kvm_flush_remote_tlbs(kvm);
916
917         return young;
918 }
919
920 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
921                                      struct mm_struct *mm)
922 {
923         struct kvm *kvm = mmu_notifier_to_kvm(mn);
924         kvm_arch_flush_shadow(kvm);
925 }
926
927 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
928         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
929         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
930         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
931         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
932         .release                = kvm_mmu_notifier_release,
933 };
934 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
935
936 static struct kvm *kvm_create_vm(void)
937 {
938         struct kvm *kvm = kvm_arch_create_vm();
939 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
940         struct page *page;
941 #endif
942
943         if (IS_ERR(kvm))
944                 goto out;
945 #ifdef CONFIG_HAVE_KVM_IRQCHIP
946         INIT_LIST_HEAD(&kvm->irq_routing);
947         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
948 #endif
949
950 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
951         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
952         if (!page) {
953                 kfree(kvm);
954                 return ERR_PTR(-ENOMEM);
955         }
956         kvm->coalesced_mmio_ring =
957                         (struct kvm_coalesced_mmio_ring *)page_address(page);
958 #endif
959
960 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
961         {
962                 int err;
963                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
964                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
965                 if (err) {
966 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
967                         put_page(page);
968 #endif
969                         kfree(kvm);
970                         return ERR_PTR(err);
971                 }
972         }
973 #endif
974
975         kvm->mm = current->mm;
976         atomic_inc(&kvm->mm->mm_count);
977         spin_lock_init(&kvm->mmu_lock);
978         spin_lock_init(&kvm->requests_lock);
979         kvm_io_bus_init(&kvm->pio_bus);
980         kvm_eventfd_init(kvm);
981         mutex_init(&kvm->lock);
982         mutex_init(&kvm->irq_lock);
983         kvm_io_bus_init(&kvm->mmio_bus);
984         init_rwsem(&kvm->slots_lock);
985         atomic_set(&kvm->users_count, 1);
986         spin_lock(&kvm_lock);
987         list_add(&kvm->vm_list, &vm_list);
988         spin_unlock(&kvm_lock);
989 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
990         kvm_coalesced_mmio_init(kvm);
991 #endif
992 out:
993         return kvm;
994 }
995
996 /*
997  * Free any memory in @free but not in @dont.
998  */
999 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1000                                   struct kvm_memory_slot *dont)
1001 {
1002         int i;
1003
1004         if (!dont || free->rmap != dont->rmap)
1005                 vfree(free->rmap);
1006
1007         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1008                 vfree(free->dirty_bitmap);
1009
1010
1011         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1012                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1013                         vfree(free->lpage_info[i]);
1014                         free->lpage_info[i] = NULL;
1015                 }
1016         }
1017
1018         free->npages = 0;
1019         free->dirty_bitmap = NULL;
1020         free->rmap = NULL;
1021 }
1022
1023 void kvm_free_physmem(struct kvm *kvm)
1024 {
1025         int i;
1026
1027         for (i = 0; i < kvm->nmemslots; ++i)
1028                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1029 }
1030
1031 static void kvm_destroy_vm(struct kvm *kvm)
1032 {
1033         struct mm_struct *mm = kvm->mm;
1034
1035         kvm_arch_sync_events(kvm);
1036         spin_lock(&kvm_lock);
1037         list_del(&kvm->vm_list);
1038         spin_unlock(&kvm_lock);
1039         kvm_free_irq_routing(kvm);
1040         kvm_io_bus_destroy(&kvm->pio_bus);
1041         kvm_io_bus_destroy(&kvm->mmio_bus);
1042 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1043         if (kvm->coalesced_mmio_ring != NULL)
1044                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1045 #endif
1046 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1047         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1048 #else
1049         kvm_arch_flush_shadow(kvm);
1050 #endif
1051         kvm_arch_destroy_vm(kvm);
1052         mmdrop(mm);
1053 }
1054
1055 void kvm_get_kvm(struct kvm *kvm)
1056 {
1057         atomic_inc(&kvm->users_count);
1058 }
1059 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1060
1061 void kvm_put_kvm(struct kvm *kvm)
1062 {
1063         if (atomic_dec_and_test(&kvm->users_count))
1064                 kvm_destroy_vm(kvm);
1065 }
1066 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1067
1068
1069 static int kvm_vm_release(struct inode *inode, struct file *filp)
1070 {
1071         struct kvm *kvm = filp->private_data;
1072
1073         kvm_irqfd_release(kvm);
1074
1075         kvm_put_kvm(kvm);
1076         return 0;
1077 }
1078
1079 /*
1080  * Allocate some memory and give it an address in the guest physical address
1081  * space.
1082  *
1083  * Discontiguous memory is allowed, mostly for framebuffers.
1084  *
1085  * Must be called holding mmap_sem for write.
1086  */
1087 int __kvm_set_memory_region(struct kvm *kvm,
1088                             struct kvm_userspace_memory_region *mem,
1089                             int user_alloc)
1090 {
1091         int r;
1092         gfn_t base_gfn;
1093         unsigned long npages;
1094         unsigned long i;
1095         struct kvm_memory_slot *memslot;
1096         struct kvm_memory_slot old, new;
1097
1098         r = -EINVAL;
1099         /* General sanity checks */
1100         if (mem->memory_size & (PAGE_SIZE - 1))
1101                 goto out;
1102         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1103                 goto out;
1104         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1105                 goto out;
1106         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1107                 goto out;
1108         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1109                 goto out;
1110
1111         memslot = &kvm->memslots[mem->slot];
1112         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1113         npages = mem->memory_size >> PAGE_SHIFT;
1114
1115         if (!npages)
1116                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1117
1118         new = old = *memslot;
1119
1120         new.base_gfn = base_gfn;
1121         new.npages = npages;
1122         new.flags = mem->flags;
1123
1124         /* Disallow changing a memory slot's size. */
1125         r = -EINVAL;
1126         if (npages && old.npages && npages != old.npages)
1127                 goto out_free;
1128
1129         /* Check for overlaps */
1130         r = -EEXIST;
1131         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1132                 struct kvm_memory_slot *s = &kvm->memslots[i];
1133
1134                 if (s == memslot || !s->npages)
1135                         continue;
1136                 if (!((base_gfn + npages <= s->base_gfn) ||
1137                       (base_gfn >= s->base_gfn + s->npages)))
1138                         goto out_free;
1139         }
1140
1141         /* Free page dirty bitmap if unneeded */
1142         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1143                 new.dirty_bitmap = NULL;
1144
1145         r = -ENOMEM;
1146
1147         /* Allocate if a slot is being created */
1148 #ifndef CONFIG_S390
1149         if (npages && !new.rmap) {
1150                 new.rmap = vmalloc(npages * sizeof(struct page *));
1151
1152                 if (!new.rmap)
1153                         goto out_free;
1154
1155                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1156
1157                 new.user_alloc = user_alloc;
1158                 /*
1159                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1160                  * safe it has to ignore memslots with !user_alloc &&
1161                  * !userspace_addr.
1162                  */
1163                 if (user_alloc)
1164                         new.userspace_addr = mem->userspace_addr;
1165                 else
1166                         new.userspace_addr = 0;
1167         }
1168         if (!npages)
1169                 goto skip_lpage;
1170
1171         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1172                 unsigned long ugfn;
1173                 unsigned long j;
1174                 int lpages;
1175                 int level = i + 2;
1176
1177                 /* Avoid unused variable warning if no large pages */
1178                 (void)level;
1179
1180                 if (new.lpage_info[i])
1181                         continue;
1182
1183                 lpages = 1 + (base_gfn + npages - 1) /
1184                              KVM_PAGES_PER_HPAGE(level);
1185                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1186
1187                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1188
1189                 if (!new.lpage_info[i])
1190                         goto out_free;
1191
1192                 memset(new.lpage_info[i], 0,
1193                        lpages * sizeof(*new.lpage_info[i]));
1194
1195                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1196                         new.lpage_info[i][0].write_count = 1;
1197                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1198                         new.lpage_info[i][lpages - 1].write_count = 1;
1199                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1200                 /*
1201                  * If the gfn and userspace address are not aligned wrt each
1202                  * other, or if explicitly asked to, disable large page
1203                  * support for this slot
1204                  */
1205                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1206                     !largepages_enabled)
1207                         for (j = 0; j < lpages; ++j)
1208                                 new.lpage_info[i][j].write_count = 1;
1209         }
1210
1211 skip_lpage:
1212
1213         /* Allocate page dirty bitmap if needed */
1214         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1215                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1216
1217                 new.dirty_bitmap = vmalloc(dirty_bytes);
1218                 if (!new.dirty_bitmap)
1219                         goto out_free;
1220                 memset(new.dirty_bitmap, 0, dirty_bytes);
1221                 if (old.npages)
1222                         kvm_arch_flush_shadow(kvm);
1223         }
1224 #else  /* not defined CONFIG_S390 */
1225         new.user_alloc = user_alloc;
1226         if (user_alloc)
1227                 new.userspace_addr = mem->userspace_addr;
1228 #endif /* not defined CONFIG_S390 */
1229
1230         if (!npages)
1231                 kvm_arch_flush_shadow(kvm);
1232
1233         spin_lock(&kvm->mmu_lock);
1234         if (mem->slot >= kvm->nmemslots)
1235                 kvm->nmemslots = mem->slot + 1;
1236
1237         *memslot = new;
1238         spin_unlock(&kvm->mmu_lock);
1239
1240         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1241         if (r) {
1242                 spin_lock(&kvm->mmu_lock);
1243                 *memslot = old;
1244                 spin_unlock(&kvm->mmu_lock);
1245                 goto out_free;
1246         }
1247
1248         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1249         /* Slot deletion case: we have to update the current slot */
1250         spin_lock(&kvm->mmu_lock);
1251         if (!npages)
1252                 *memslot = old;
1253         spin_unlock(&kvm->mmu_lock);
1254 #ifdef CONFIG_DMAR
1255         /* map the pages in iommu page table */
1256         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1257         if (r)
1258                 goto out;
1259 #endif
1260         return 0;
1261
1262 out_free:
1263         kvm_free_physmem_slot(&new, &old);
1264 out:
1265         return r;
1266
1267 }
1268 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1269
1270 int kvm_set_memory_region(struct kvm *kvm,
1271                           struct kvm_userspace_memory_region *mem,
1272                           int user_alloc)
1273 {
1274         int r;
1275
1276         down_write(&kvm->slots_lock);
1277         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1278         up_write(&kvm->slots_lock);
1279         return r;
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1282
1283 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1284                                    struct
1285                                    kvm_userspace_memory_region *mem,
1286                                    int user_alloc)
1287 {
1288         if (mem->slot >= KVM_MEMORY_SLOTS)
1289                 return -EINVAL;
1290         return kvm_set_memory_region(kvm, mem, user_alloc);
1291 }
1292
1293 int kvm_get_dirty_log(struct kvm *kvm,
1294                         struct kvm_dirty_log *log, int *is_dirty)
1295 {
1296         struct kvm_memory_slot *memslot;
1297         int r, i;
1298         int n;
1299         unsigned long any = 0;
1300
1301         r = -EINVAL;
1302         if (log->slot >= KVM_MEMORY_SLOTS)
1303                 goto out;
1304
1305         memslot = &kvm->memslots[log->slot];
1306         r = -ENOENT;
1307         if (!memslot->dirty_bitmap)
1308                 goto out;
1309
1310         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1311
1312         for (i = 0; !any && i < n/sizeof(long); ++i)
1313                 any = memslot->dirty_bitmap[i];
1314
1315         r = -EFAULT;
1316         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1317                 goto out;
1318
1319         if (any)
1320                 *is_dirty = 1;
1321
1322         r = 0;
1323 out:
1324         return r;
1325 }
1326
1327 void kvm_disable_largepages(void)
1328 {
1329         largepages_enabled = false;
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1332
1333 int is_error_page(struct page *page)
1334 {
1335         return page == bad_page;
1336 }
1337 EXPORT_SYMBOL_GPL(is_error_page);
1338
1339 int is_error_pfn(pfn_t pfn)
1340 {
1341         return pfn == bad_pfn;
1342 }
1343 EXPORT_SYMBOL_GPL(is_error_pfn);
1344
1345 static inline unsigned long bad_hva(void)
1346 {
1347         return PAGE_OFFSET;
1348 }
1349
1350 int kvm_is_error_hva(unsigned long addr)
1351 {
1352         return addr == bad_hva();
1353 }
1354 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1355
1356 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1357 {
1358         int i;
1359
1360         for (i = 0; i < kvm->nmemslots; ++i) {
1361                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1362
1363                 if (gfn >= memslot->base_gfn
1364                     && gfn < memslot->base_gfn + memslot->npages)
1365                         return memslot;
1366         }
1367         return NULL;
1368 }
1369 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1370
1371 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1372 {
1373         gfn = unalias_gfn(kvm, gfn);
1374         return gfn_to_memslot_unaliased(kvm, gfn);
1375 }
1376
1377 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1378 {
1379         int i;
1380
1381         gfn = unalias_gfn(kvm, gfn);
1382         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1383                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1384
1385                 if (gfn >= memslot->base_gfn
1386                     && gfn < memslot->base_gfn + memslot->npages)
1387                         return 1;
1388         }
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1392
1393 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1394 {
1395         struct kvm_memory_slot *slot;
1396
1397         gfn = unalias_gfn(kvm, gfn);
1398         slot = gfn_to_memslot_unaliased(kvm, gfn);
1399         if (!slot)
1400                 return bad_hva();
1401         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1402 }
1403 EXPORT_SYMBOL_GPL(gfn_to_hva);
1404
1405 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1406 {
1407         struct page *page[1];
1408         unsigned long addr;
1409         int npages;
1410         pfn_t pfn;
1411
1412         might_sleep();
1413
1414         addr = gfn_to_hva(kvm, gfn);
1415         if (kvm_is_error_hva(addr)) {
1416                 get_page(bad_page);
1417                 return page_to_pfn(bad_page);
1418         }
1419
1420         npages = get_user_pages_fast(addr, 1, 1, page);
1421
1422         if (unlikely(npages != 1)) {
1423                 struct vm_area_struct *vma;
1424
1425                 down_read(&current->mm->mmap_sem);
1426                 vma = find_vma(current->mm, addr);
1427
1428                 if (vma == NULL || addr < vma->vm_start ||
1429                     !(vma->vm_flags & VM_PFNMAP)) {
1430                         up_read(&current->mm->mmap_sem);
1431                         get_page(bad_page);
1432                         return page_to_pfn(bad_page);
1433                 }
1434
1435                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1436                 up_read(&current->mm->mmap_sem);
1437                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1438         } else
1439                 pfn = page_to_pfn(page[0]);
1440
1441         return pfn;
1442 }
1443
1444 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1445
1446 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1447 {
1448         pfn_t pfn;
1449
1450         pfn = gfn_to_pfn(kvm, gfn);
1451         if (!kvm_is_mmio_pfn(pfn))
1452                 return pfn_to_page(pfn);
1453
1454         WARN_ON(kvm_is_mmio_pfn(pfn));
1455
1456         get_page(bad_page);
1457         return bad_page;
1458 }
1459
1460 EXPORT_SYMBOL_GPL(gfn_to_page);
1461
1462 void kvm_release_page_clean(struct page *page)
1463 {
1464         kvm_release_pfn_clean(page_to_pfn(page));
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1467
1468 void kvm_release_pfn_clean(pfn_t pfn)
1469 {
1470         if (!kvm_is_mmio_pfn(pfn))
1471                 put_page(pfn_to_page(pfn));
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1474
1475 void kvm_release_page_dirty(struct page *page)
1476 {
1477         kvm_release_pfn_dirty(page_to_pfn(page));
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1480
1481 void kvm_release_pfn_dirty(pfn_t pfn)
1482 {
1483         kvm_set_pfn_dirty(pfn);
1484         kvm_release_pfn_clean(pfn);
1485 }
1486 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1487
1488 void kvm_set_page_dirty(struct page *page)
1489 {
1490         kvm_set_pfn_dirty(page_to_pfn(page));
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1493
1494 void kvm_set_pfn_dirty(pfn_t pfn)
1495 {
1496         if (!kvm_is_mmio_pfn(pfn)) {
1497                 struct page *page = pfn_to_page(pfn);
1498                 if (!PageReserved(page))
1499                         SetPageDirty(page);
1500         }
1501 }
1502 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1503
1504 void kvm_set_pfn_accessed(pfn_t pfn)
1505 {
1506         if (!kvm_is_mmio_pfn(pfn))
1507                 mark_page_accessed(pfn_to_page(pfn));
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1510
1511 void kvm_get_pfn(pfn_t pfn)
1512 {
1513         if (!kvm_is_mmio_pfn(pfn))
1514                 get_page(pfn_to_page(pfn));
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1517
1518 static int next_segment(unsigned long len, int offset)
1519 {
1520         if (len > PAGE_SIZE - offset)
1521                 return PAGE_SIZE - offset;
1522         else
1523                 return len;
1524 }
1525
1526 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1527                         int len)
1528 {
1529         int r;
1530         unsigned long addr;
1531
1532         addr = gfn_to_hva(kvm, gfn);
1533         if (kvm_is_error_hva(addr))
1534                 return -EFAULT;
1535         r = copy_from_user(data, (void __user *)addr + offset, len);
1536         if (r)
1537                 return -EFAULT;
1538         return 0;
1539 }
1540 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1541
1542 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1543 {
1544         gfn_t gfn = gpa >> PAGE_SHIFT;
1545         int seg;
1546         int offset = offset_in_page(gpa);
1547         int ret;
1548
1549         while ((seg = next_segment(len, offset)) != 0) {
1550                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1551                 if (ret < 0)
1552                         return ret;
1553                 offset = 0;
1554                 len -= seg;
1555                 data += seg;
1556                 ++gfn;
1557         }
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(kvm_read_guest);
1561
1562 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1563                           unsigned long len)
1564 {
1565         int r;
1566         unsigned long addr;
1567         gfn_t gfn = gpa >> PAGE_SHIFT;
1568         int offset = offset_in_page(gpa);
1569
1570         addr = gfn_to_hva(kvm, gfn);
1571         if (kvm_is_error_hva(addr))
1572                 return -EFAULT;
1573         pagefault_disable();
1574         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1575         pagefault_enable();
1576         if (r)
1577                 return -EFAULT;
1578         return 0;
1579 }
1580 EXPORT_SYMBOL(kvm_read_guest_atomic);
1581
1582 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1583                          int offset, int len)
1584 {
1585         int r;
1586         unsigned long addr;
1587
1588         addr = gfn_to_hva(kvm, gfn);
1589         if (kvm_is_error_hva(addr))
1590                 return -EFAULT;
1591         r = copy_to_user((void __user *)addr + offset, data, len);
1592         if (r)
1593                 return -EFAULT;
1594         mark_page_dirty(kvm, gfn);
1595         return 0;
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1598
1599 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1600                     unsigned long len)
1601 {
1602         gfn_t gfn = gpa >> PAGE_SHIFT;
1603         int seg;
1604         int offset = offset_in_page(gpa);
1605         int ret;
1606
1607         while ((seg = next_segment(len, offset)) != 0) {
1608                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1609                 if (ret < 0)
1610                         return ret;
1611                 offset = 0;
1612                 len -= seg;
1613                 data += seg;
1614                 ++gfn;
1615         }
1616         return 0;
1617 }
1618
1619 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1620 {
1621         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1622 }
1623 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1624
1625 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1626 {
1627         gfn_t gfn = gpa >> PAGE_SHIFT;
1628         int seg;
1629         int offset = offset_in_page(gpa);
1630         int ret;
1631
1632         while ((seg = next_segment(len, offset)) != 0) {
1633                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1634                 if (ret < 0)
1635                         return ret;
1636                 offset = 0;
1637                 len -= seg;
1638                 ++gfn;
1639         }
1640         return 0;
1641 }
1642 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1643
1644 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1645 {
1646         struct kvm_memory_slot *memslot;
1647
1648         gfn = unalias_gfn(kvm, gfn);
1649         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1650         if (memslot && memslot->dirty_bitmap) {
1651                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1652
1653                 /* avoid RMW */
1654                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1655                         set_bit(rel_gfn, memslot->dirty_bitmap);
1656         }
1657 }
1658
1659 /*
1660  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1661  */
1662 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1663 {
1664         DEFINE_WAIT(wait);
1665
1666         for (;;) {
1667                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1668
1669                 if (kvm_arch_vcpu_runnable(vcpu)) {
1670                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1671                         break;
1672                 }
1673                 if (kvm_cpu_has_pending_timer(vcpu))
1674                         break;
1675                 if (signal_pending(current))
1676                         break;
1677
1678                 vcpu_put(vcpu);
1679                 schedule();
1680                 vcpu_load(vcpu);
1681         }
1682
1683         finish_wait(&vcpu->wq, &wait);
1684 }
1685
1686 void kvm_resched(struct kvm_vcpu *vcpu)
1687 {
1688         if (!need_resched())
1689                 return;
1690         cond_resched();
1691 }
1692 EXPORT_SYMBOL_GPL(kvm_resched);
1693
1694 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1695 {
1696         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1697         struct page *page;
1698
1699         if (vmf->pgoff == 0)
1700                 page = virt_to_page(vcpu->run);
1701 #ifdef CONFIG_X86
1702         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1703                 page = virt_to_page(vcpu->arch.pio_data);
1704 #endif
1705 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1706         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1707                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1708 #endif
1709         else
1710                 return VM_FAULT_SIGBUS;
1711         get_page(page);
1712         vmf->page = page;
1713         return 0;
1714 }
1715
1716 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1717         .fault = kvm_vcpu_fault,
1718 };
1719
1720 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1721 {
1722         vma->vm_ops = &kvm_vcpu_vm_ops;
1723         return 0;
1724 }
1725
1726 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1727 {
1728         struct kvm_vcpu *vcpu = filp->private_data;
1729
1730         kvm_put_kvm(vcpu->kvm);
1731         return 0;
1732 }
1733
1734 static struct file_operations kvm_vcpu_fops = {
1735         .release        = kvm_vcpu_release,
1736         .unlocked_ioctl = kvm_vcpu_ioctl,
1737         .compat_ioctl   = kvm_vcpu_ioctl,
1738         .mmap           = kvm_vcpu_mmap,
1739 };
1740
1741 /*
1742  * Allocates an inode for the vcpu.
1743  */
1744 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1745 {
1746         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1747 }
1748
1749 /*
1750  * Creates some virtual cpus.  Good luck creating more than one.
1751  */
1752 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1753 {
1754         int r;
1755         struct kvm_vcpu *vcpu, *v;
1756
1757         vcpu = kvm_arch_vcpu_create(kvm, id);
1758         if (IS_ERR(vcpu))
1759                 return PTR_ERR(vcpu);
1760
1761         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1762
1763         r = kvm_arch_vcpu_setup(vcpu);
1764         if (r)
1765                 return r;
1766
1767         mutex_lock(&kvm->lock);
1768         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1769                 r = -EINVAL;
1770                 goto vcpu_destroy;
1771         }
1772
1773         kvm_for_each_vcpu(r, v, kvm)
1774                 if (v->vcpu_id == id) {
1775                         r = -EEXIST;
1776                         goto vcpu_destroy;
1777                 }
1778
1779         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1780
1781         /* Now it's all set up, let userspace reach it */
1782         kvm_get_kvm(kvm);
1783         r = create_vcpu_fd(vcpu);
1784         if (r < 0) {
1785                 kvm_put_kvm(kvm);
1786                 goto vcpu_destroy;
1787         }
1788
1789         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1790         smp_wmb();
1791         atomic_inc(&kvm->online_vcpus);
1792
1793 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1794         if (kvm->bsp_vcpu_id == id)
1795                 kvm->bsp_vcpu = vcpu;
1796 #endif
1797         mutex_unlock(&kvm->lock);
1798         return r;
1799
1800 vcpu_destroy:
1801         mutex_unlock(&kvm->lock);
1802         kvm_arch_vcpu_destroy(vcpu);
1803         return r;
1804 }
1805
1806 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1807 {
1808         if (sigset) {
1809                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1810                 vcpu->sigset_active = 1;
1811                 vcpu->sigset = *sigset;
1812         } else
1813                 vcpu->sigset_active = 0;
1814         return 0;
1815 }
1816
1817 #ifdef __KVM_HAVE_MSIX
1818 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1819                                     struct kvm_assigned_msix_nr *entry_nr)
1820 {
1821         int r = 0;
1822         struct kvm_assigned_dev_kernel *adev;
1823
1824         mutex_lock(&kvm->lock);
1825
1826         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1827                                       entry_nr->assigned_dev_id);
1828         if (!adev) {
1829                 r = -EINVAL;
1830                 goto msix_nr_out;
1831         }
1832
1833         if (adev->entries_nr == 0) {
1834                 adev->entries_nr = entry_nr->entry_nr;
1835                 if (adev->entries_nr == 0 ||
1836                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1837                         r = -EINVAL;
1838                         goto msix_nr_out;
1839                 }
1840
1841                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1842                                                 entry_nr->entry_nr,
1843                                                 GFP_KERNEL);
1844                 if (!adev->host_msix_entries) {
1845                         r = -ENOMEM;
1846                         goto msix_nr_out;
1847                 }
1848                 adev->guest_msix_entries = kzalloc(
1849                                 sizeof(struct kvm_guest_msix_entry) *
1850                                 entry_nr->entry_nr, GFP_KERNEL);
1851                 if (!adev->guest_msix_entries) {
1852                         kfree(adev->host_msix_entries);
1853                         r = -ENOMEM;
1854                         goto msix_nr_out;
1855                 }
1856         } else /* Not allowed set MSI-X number twice */
1857                 r = -EINVAL;
1858 msix_nr_out:
1859         mutex_unlock(&kvm->lock);
1860         return r;
1861 }
1862
1863 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1864                                        struct kvm_assigned_msix_entry *entry)
1865 {
1866         int r = 0, i;
1867         struct kvm_assigned_dev_kernel *adev;
1868
1869         mutex_lock(&kvm->lock);
1870
1871         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1872                                       entry->assigned_dev_id);
1873
1874         if (!adev) {
1875                 r = -EINVAL;
1876                 goto msix_entry_out;
1877         }
1878
1879         for (i = 0; i < adev->entries_nr; i++)
1880                 if (adev->guest_msix_entries[i].vector == 0 ||
1881                     adev->guest_msix_entries[i].entry == entry->entry) {
1882                         adev->guest_msix_entries[i].entry = entry->entry;
1883                         adev->guest_msix_entries[i].vector = entry->gsi;
1884                         adev->host_msix_entries[i].entry = entry->entry;
1885                         break;
1886                 }
1887         if (i == adev->entries_nr) {
1888                 r = -ENOSPC;
1889                 goto msix_entry_out;
1890         }
1891
1892 msix_entry_out:
1893         mutex_unlock(&kvm->lock);
1894
1895         return r;
1896 }
1897 #endif
1898
1899 static long kvm_vcpu_ioctl(struct file *filp,
1900                            unsigned int ioctl, unsigned long arg)
1901 {
1902         struct kvm_vcpu *vcpu = filp->private_data;
1903         void __user *argp = (void __user *)arg;
1904         int r;
1905         struct kvm_fpu *fpu = NULL;
1906         struct kvm_sregs *kvm_sregs = NULL;
1907
1908         if (vcpu->kvm->mm != current->mm)
1909                 return -EIO;
1910         switch (ioctl) {
1911         case KVM_RUN:
1912                 r = -EINVAL;
1913                 if (arg)
1914                         goto out;
1915                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1916                 break;
1917         case KVM_GET_REGS: {
1918                 struct kvm_regs *kvm_regs;
1919
1920                 r = -ENOMEM;
1921                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1922                 if (!kvm_regs)
1923                         goto out;
1924                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1925                 if (r)
1926                         goto out_free1;
1927                 r = -EFAULT;
1928                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1929                         goto out_free1;
1930                 r = 0;
1931 out_free1:
1932                 kfree(kvm_regs);
1933                 break;
1934         }
1935         case KVM_SET_REGS: {
1936                 struct kvm_regs *kvm_regs;
1937
1938                 r = -ENOMEM;
1939                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1940                 if (!kvm_regs)
1941                         goto out;
1942                 r = -EFAULT;
1943                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1944                         goto out_free2;
1945                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1946                 if (r)
1947                         goto out_free2;
1948                 r = 0;
1949 out_free2:
1950                 kfree(kvm_regs);
1951                 break;
1952         }
1953         case KVM_GET_SREGS: {
1954                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1955                 r = -ENOMEM;
1956                 if (!kvm_sregs)
1957                         goto out;
1958                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1959                 if (r)
1960                         goto out;
1961                 r = -EFAULT;
1962                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1963                         goto out;
1964                 r = 0;
1965                 break;
1966         }
1967         case KVM_SET_SREGS: {
1968                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1969                 r = -ENOMEM;
1970                 if (!kvm_sregs)
1971                         goto out;
1972                 r = -EFAULT;
1973                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1974                         goto out;
1975                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1976                 if (r)
1977                         goto out;
1978                 r = 0;
1979                 break;
1980         }
1981         case KVM_GET_MP_STATE: {
1982                 struct kvm_mp_state mp_state;
1983
1984                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1985                 if (r)
1986                         goto out;
1987                 r = -EFAULT;
1988                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1989                         goto out;
1990                 r = 0;
1991                 break;
1992         }
1993         case KVM_SET_MP_STATE: {
1994                 struct kvm_mp_state mp_state;
1995
1996                 r = -EFAULT;
1997                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1998                         goto out;
1999                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2000                 if (r)
2001                         goto out;
2002                 r = 0;
2003                 break;
2004         }
2005         case KVM_TRANSLATE: {
2006                 struct kvm_translation tr;
2007
2008                 r = -EFAULT;
2009                 if (copy_from_user(&tr, argp, sizeof tr))
2010                         goto out;
2011                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2012                 if (r)
2013                         goto out;
2014                 r = -EFAULT;
2015                 if (copy_to_user(argp, &tr, sizeof tr))
2016                         goto out;
2017                 r = 0;
2018                 break;
2019         }
2020         case KVM_SET_GUEST_DEBUG: {
2021                 struct kvm_guest_debug dbg;
2022
2023                 r = -EFAULT;
2024                 if (copy_from_user(&dbg, argp, sizeof dbg))
2025                         goto out;
2026                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2027                 if (r)
2028                         goto out;
2029                 r = 0;
2030                 break;
2031         }
2032         case KVM_SET_SIGNAL_MASK: {
2033                 struct kvm_signal_mask __user *sigmask_arg = argp;
2034                 struct kvm_signal_mask kvm_sigmask;
2035                 sigset_t sigset, *p;
2036
2037                 p = NULL;
2038                 if (argp) {
2039                         r = -EFAULT;
2040                         if (copy_from_user(&kvm_sigmask, argp,
2041                                            sizeof kvm_sigmask))
2042                                 goto out;
2043                         r = -EINVAL;
2044                         if (kvm_sigmask.len != sizeof sigset)
2045                                 goto out;
2046                         r = -EFAULT;
2047                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2048                                            sizeof sigset))
2049                                 goto out;
2050                         p = &sigset;
2051                 }
2052                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2053                 break;
2054         }
2055         case KVM_GET_FPU: {
2056                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2057                 r = -ENOMEM;
2058                 if (!fpu)
2059                         goto out;
2060                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2061                 if (r)
2062                         goto out;
2063                 r = -EFAULT;
2064                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2065                         goto out;
2066                 r = 0;
2067                 break;
2068         }
2069         case KVM_SET_FPU: {
2070                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2071                 r = -ENOMEM;
2072                 if (!fpu)
2073                         goto out;
2074                 r = -EFAULT;
2075                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2076                         goto out;
2077                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2078                 if (r)
2079                         goto out;
2080                 r = 0;
2081                 break;
2082         }
2083         default:
2084                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2085         }
2086 out:
2087         kfree(fpu);
2088         kfree(kvm_sregs);
2089         return r;
2090 }
2091
2092 static long kvm_vm_ioctl(struct file *filp,
2093                            unsigned int ioctl, unsigned long arg)
2094 {
2095         struct kvm *kvm = filp->private_data;
2096         void __user *argp = (void __user *)arg;
2097         int r;
2098
2099         if (kvm->mm != current->mm)
2100                 return -EIO;
2101         switch (ioctl) {
2102         case KVM_CREATE_VCPU:
2103                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2104                 if (r < 0)
2105                         goto out;
2106                 break;
2107         case KVM_SET_USER_MEMORY_REGION: {
2108                 struct kvm_userspace_memory_region kvm_userspace_mem;
2109
2110                 r = -EFAULT;
2111                 if (copy_from_user(&kvm_userspace_mem, argp,
2112                                                 sizeof kvm_userspace_mem))
2113                         goto out;
2114
2115                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2116                 if (r)
2117                         goto out;
2118                 break;
2119         }
2120         case KVM_GET_DIRTY_LOG: {
2121                 struct kvm_dirty_log log;
2122
2123                 r = -EFAULT;
2124                 if (copy_from_user(&log, argp, sizeof log))
2125                         goto out;
2126                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2127                 if (r)
2128                         goto out;
2129                 break;
2130         }
2131 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2132         case KVM_REGISTER_COALESCED_MMIO: {
2133                 struct kvm_coalesced_mmio_zone zone;
2134                 r = -EFAULT;
2135                 if (copy_from_user(&zone, argp, sizeof zone))
2136                         goto out;
2137                 r = -ENXIO;
2138                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2139                 if (r)
2140                         goto out;
2141                 r = 0;
2142                 break;
2143         }
2144         case KVM_UNREGISTER_COALESCED_MMIO: {
2145                 struct kvm_coalesced_mmio_zone zone;
2146                 r = -EFAULT;
2147                 if (copy_from_user(&zone, argp, sizeof zone))
2148                         goto out;
2149                 r = -ENXIO;
2150                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2151                 if (r)
2152                         goto out;
2153                 r = 0;
2154                 break;
2155         }
2156 #endif
2157 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2158         case KVM_ASSIGN_PCI_DEVICE: {
2159                 struct kvm_assigned_pci_dev assigned_dev;
2160
2161                 r = -EFAULT;
2162                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2163                         goto out;
2164                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2165                 if (r)
2166                         goto out;
2167                 break;
2168         }
2169         case KVM_ASSIGN_IRQ: {
2170                 r = -EOPNOTSUPP;
2171                 break;
2172         }
2173 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2174         case KVM_ASSIGN_DEV_IRQ: {
2175                 struct kvm_assigned_irq assigned_irq;
2176
2177                 r = -EFAULT;
2178                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2179                         goto out;
2180                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2181                 if (r)
2182                         goto out;
2183                 break;
2184         }
2185         case KVM_DEASSIGN_DEV_IRQ: {
2186                 struct kvm_assigned_irq assigned_irq;
2187
2188                 r = -EFAULT;
2189                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2190                         goto out;
2191                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2192                 if (r)
2193                         goto out;
2194                 break;
2195         }
2196 #endif
2197 #endif
2198 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2199         case KVM_DEASSIGN_PCI_DEVICE: {
2200                 struct kvm_assigned_pci_dev assigned_dev;
2201
2202                 r = -EFAULT;
2203                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2204                         goto out;
2205                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2206                 if (r)
2207                         goto out;
2208                 break;
2209         }
2210 #endif
2211 #ifdef KVM_CAP_IRQ_ROUTING
2212         case KVM_SET_GSI_ROUTING: {
2213                 struct kvm_irq_routing routing;
2214                 struct kvm_irq_routing __user *urouting;
2215                 struct kvm_irq_routing_entry *entries;
2216
2217                 r = -EFAULT;
2218                 if (copy_from_user(&routing, argp, sizeof(routing)))
2219                         goto out;
2220                 r = -EINVAL;
2221                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2222                         goto out;
2223                 if (routing.flags)
2224                         goto out;
2225                 r = -ENOMEM;
2226                 entries = vmalloc(routing.nr * sizeof(*entries));
2227                 if (!entries)
2228                         goto out;
2229                 r = -EFAULT;
2230                 urouting = argp;
2231                 if (copy_from_user(entries, urouting->entries,
2232                                    routing.nr * sizeof(*entries)))
2233                         goto out_free_irq_routing;
2234                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2235                                         routing.flags);
2236         out_free_irq_routing:
2237                 vfree(entries);
2238                 break;
2239         }
2240 #endif /* KVM_CAP_IRQ_ROUTING */
2241 #ifdef __KVM_HAVE_MSIX
2242         case KVM_ASSIGN_SET_MSIX_NR: {
2243                 struct kvm_assigned_msix_nr entry_nr;
2244                 r = -EFAULT;
2245                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2246                         goto out;
2247                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2248                 if (r)
2249                         goto out;
2250                 break;
2251         }
2252         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2253                 struct kvm_assigned_msix_entry entry;
2254                 r = -EFAULT;
2255                 if (copy_from_user(&entry, argp, sizeof entry))
2256                         goto out;
2257                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2258                 if (r)
2259                         goto out;
2260                 break;
2261         }
2262 #endif
2263         case KVM_IRQFD: {
2264                 struct kvm_irqfd data;
2265
2266                 r = -EFAULT;
2267                 if (copy_from_user(&data, argp, sizeof data))
2268                         goto out;
2269                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2270                 break;
2271         }
2272         case KVM_IOEVENTFD: {
2273                 struct kvm_ioeventfd data;
2274
2275                 r = -EFAULT;
2276                 if (copy_from_user(&data, argp, sizeof data))
2277                         goto out;
2278                 r = kvm_ioeventfd(kvm, &data);
2279                 break;
2280         }
2281 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2282         case KVM_SET_BOOT_CPU_ID:
2283                 r = 0;
2284                 mutex_lock(&kvm->lock);
2285                 if (atomic_read(&kvm->online_vcpus) != 0)
2286                         r = -EBUSY;
2287                 else
2288                         kvm->bsp_vcpu_id = arg;
2289                 mutex_unlock(&kvm->lock);
2290                 break;
2291 #endif
2292         default:
2293                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2294         }
2295 out:
2296         return r;
2297 }
2298
2299 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2300 {
2301         struct page *page[1];
2302         unsigned long addr;
2303         int npages;
2304         gfn_t gfn = vmf->pgoff;
2305         struct kvm *kvm = vma->vm_file->private_data;
2306
2307         addr = gfn_to_hva(kvm, gfn);
2308         if (kvm_is_error_hva(addr))
2309                 return VM_FAULT_SIGBUS;
2310
2311         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2312                                 NULL);
2313         if (unlikely(npages != 1))
2314                 return VM_FAULT_SIGBUS;
2315
2316         vmf->page = page[0];
2317         return 0;
2318 }
2319
2320 static const struct vm_operations_struct kvm_vm_vm_ops = {
2321         .fault = kvm_vm_fault,
2322 };
2323
2324 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2325 {
2326         vma->vm_ops = &kvm_vm_vm_ops;
2327         return 0;
2328 }
2329
2330 static struct file_operations kvm_vm_fops = {
2331         .release        = kvm_vm_release,
2332         .unlocked_ioctl = kvm_vm_ioctl,
2333         .compat_ioctl   = kvm_vm_ioctl,
2334         .mmap           = kvm_vm_mmap,
2335 };
2336
2337 static int kvm_dev_ioctl_create_vm(void)
2338 {
2339         int fd;
2340         struct kvm *kvm;
2341
2342         kvm = kvm_create_vm();
2343         if (IS_ERR(kvm))
2344                 return PTR_ERR(kvm);
2345         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2346         if (fd < 0)
2347                 kvm_put_kvm(kvm);
2348
2349         return fd;
2350 }
2351
2352 static long kvm_dev_ioctl_check_extension_generic(long arg)
2353 {
2354         switch (arg) {
2355         case KVM_CAP_USER_MEMORY:
2356         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2357         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2358 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2359         case KVM_CAP_SET_BOOT_CPU_ID:
2360 #endif
2361                 return 1;
2362 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2363         case KVM_CAP_IRQ_ROUTING:
2364                 return KVM_MAX_IRQ_ROUTES;
2365 #endif
2366         default:
2367                 break;
2368         }
2369         return kvm_dev_ioctl_check_extension(arg);
2370 }
2371
2372 static long kvm_dev_ioctl(struct file *filp,
2373                           unsigned int ioctl, unsigned long arg)
2374 {
2375         long r = -EINVAL;
2376
2377         switch (ioctl) {
2378         case KVM_GET_API_VERSION:
2379                 r = -EINVAL;
2380                 if (arg)
2381                         goto out;
2382                 r = KVM_API_VERSION;
2383                 break;
2384         case KVM_CREATE_VM:
2385                 r = -EINVAL;
2386                 if (arg)
2387                         goto out;
2388                 r = kvm_dev_ioctl_create_vm();
2389                 break;
2390         case KVM_CHECK_EXTENSION:
2391                 r = kvm_dev_ioctl_check_extension_generic(arg);
2392                 break;
2393         case KVM_GET_VCPU_MMAP_SIZE:
2394                 r = -EINVAL;
2395                 if (arg)
2396                         goto out;
2397                 r = PAGE_SIZE;     /* struct kvm_run */
2398 #ifdef CONFIG_X86
2399                 r += PAGE_SIZE;    /* pio data page */
2400 #endif
2401 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2402                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2403 #endif
2404                 break;
2405         case KVM_TRACE_ENABLE:
2406         case KVM_TRACE_PAUSE:
2407         case KVM_TRACE_DISABLE:
2408                 r = -EOPNOTSUPP;
2409                 break;
2410         default:
2411                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2412         }
2413 out:
2414         return r;
2415 }
2416
2417 static struct file_operations kvm_chardev_ops = {
2418         .unlocked_ioctl = kvm_dev_ioctl,
2419         .compat_ioctl   = kvm_dev_ioctl,
2420 };
2421
2422 static struct miscdevice kvm_dev = {
2423         KVM_MINOR,
2424         "kvm",
2425         &kvm_chardev_ops,
2426 };
2427
2428 static void hardware_enable(void *junk)
2429 {
2430         int cpu = raw_smp_processor_id();
2431
2432         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2433                 return;
2434         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2435         kvm_arch_hardware_enable(NULL);
2436 }
2437
2438 static void hardware_disable(void *junk)
2439 {
2440         int cpu = raw_smp_processor_id();
2441
2442         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2443                 return;
2444         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2445         kvm_arch_hardware_disable(NULL);
2446 }
2447
2448 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2449                            void *v)
2450 {
2451         int cpu = (long)v;
2452
2453         val &= ~CPU_TASKS_FROZEN;
2454         switch (val) {
2455         case CPU_DYING:
2456                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2457                        cpu);
2458                 hardware_disable(NULL);
2459                 break;
2460         case CPU_UP_CANCELED:
2461                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2462                        cpu);
2463                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2464                 break;
2465         case CPU_ONLINE:
2466                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2467                        cpu);
2468                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2469                 break;
2470         }
2471         return NOTIFY_OK;
2472 }
2473
2474
2475 asmlinkage void kvm_handle_fault_on_reboot(void)
2476 {
2477         if (kvm_rebooting)
2478                 /* spin while reset goes on */
2479                 while (true)
2480                         ;
2481         /* Fault while not rebooting.  We want the trace. */
2482         BUG();
2483 }
2484 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2485
2486 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2487                       void *v)
2488 {
2489         /*
2490          * Some (well, at least mine) BIOSes hang on reboot if
2491          * in vmx root mode.
2492          *
2493          * And Intel TXT required VMX off for all cpu when system shutdown.
2494          */
2495         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2496         kvm_rebooting = true;
2497         on_each_cpu(hardware_disable, NULL, 1);
2498         return NOTIFY_OK;
2499 }
2500
2501 static struct notifier_block kvm_reboot_notifier = {
2502         .notifier_call = kvm_reboot,
2503         .priority = 0,
2504 };
2505
2506 void kvm_io_bus_init(struct kvm_io_bus *bus)
2507 {
2508         memset(bus, 0, sizeof(*bus));
2509 }
2510
2511 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2512 {
2513         int i;
2514
2515         for (i = 0; i < bus->dev_count; i++) {
2516                 struct kvm_io_device *pos = bus->devs[i];
2517
2518                 kvm_iodevice_destructor(pos);
2519         }
2520 }
2521
2522 /* kvm_io_bus_write - called under kvm->slots_lock */
2523 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2524                      int len, const void *val)
2525 {
2526         int i;
2527         for (i = 0; i < bus->dev_count; i++)
2528                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2529                         return 0;
2530         return -EOPNOTSUPP;
2531 }
2532
2533 /* kvm_io_bus_read - called under kvm->slots_lock */
2534 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2535 {
2536         int i;
2537         for (i = 0; i < bus->dev_count; i++)
2538                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2539                         return 0;
2540         return -EOPNOTSUPP;
2541 }
2542
2543 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2544                              struct kvm_io_device *dev)
2545 {
2546         int ret;
2547
2548         down_write(&kvm->slots_lock);
2549         ret = __kvm_io_bus_register_dev(bus, dev);
2550         up_write(&kvm->slots_lock);
2551
2552         return ret;
2553 }
2554
2555 /* An unlocked version. Caller must have write lock on slots_lock. */
2556 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2557                               struct kvm_io_device *dev)
2558 {
2559         if (bus->dev_count > NR_IOBUS_DEVS-1)
2560                 return -ENOSPC;
2561
2562         bus->devs[bus->dev_count++] = dev;
2563
2564         return 0;
2565 }
2566
2567 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2568                                struct kvm_io_bus *bus,
2569                                struct kvm_io_device *dev)
2570 {
2571         down_write(&kvm->slots_lock);
2572         __kvm_io_bus_unregister_dev(bus, dev);
2573         up_write(&kvm->slots_lock);
2574 }
2575
2576 /* An unlocked version. Caller must have write lock on slots_lock. */
2577 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2578                                  struct kvm_io_device *dev)
2579 {
2580         int i;
2581
2582         for (i = 0; i < bus->dev_count; i++)
2583                 if (bus->devs[i] == dev) {
2584                         bus->devs[i] = bus->devs[--bus->dev_count];
2585                         break;
2586                 }
2587 }
2588
2589 static struct notifier_block kvm_cpu_notifier = {
2590         .notifier_call = kvm_cpu_hotplug,
2591         .priority = 20, /* must be > scheduler priority */
2592 };
2593
2594 static int vm_stat_get(void *_offset, u64 *val)
2595 {
2596         unsigned offset = (long)_offset;
2597         struct kvm *kvm;
2598
2599         *val = 0;
2600         spin_lock(&kvm_lock);
2601         list_for_each_entry(kvm, &vm_list, vm_list)
2602                 *val += *(u32 *)((void *)kvm + offset);
2603         spin_unlock(&kvm_lock);
2604         return 0;
2605 }
2606
2607 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2608
2609 static int vcpu_stat_get(void *_offset, u64 *val)
2610 {
2611         unsigned offset = (long)_offset;
2612         struct kvm *kvm;
2613         struct kvm_vcpu *vcpu;
2614         int i;
2615
2616         *val = 0;
2617         spin_lock(&kvm_lock);
2618         list_for_each_entry(kvm, &vm_list, vm_list)
2619                 kvm_for_each_vcpu(i, vcpu, kvm)
2620                         *val += *(u32 *)((void *)vcpu + offset);
2621
2622         spin_unlock(&kvm_lock);
2623         return 0;
2624 }
2625
2626 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2627
2628 static const struct file_operations *stat_fops[] = {
2629         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2630         [KVM_STAT_VM]   = &vm_stat_fops,
2631 };
2632
2633 static void kvm_init_debug(void)
2634 {
2635         struct kvm_stats_debugfs_item *p;
2636
2637         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2638         for (p = debugfs_entries; p->name; ++p)
2639                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2640                                                 (void *)(long)p->offset,
2641                                                 stat_fops[p->kind]);
2642 }
2643
2644 static void kvm_exit_debug(void)
2645 {
2646         struct kvm_stats_debugfs_item *p;
2647
2648         for (p = debugfs_entries; p->name; ++p)
2649                 debugfs_remove(p->dentry);
2650         debugfs_remove(kvm_debugfs_dir);
2651 }
2652
2653 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2654 {
2655         hardware_disable(NULL);
2656         return 0;
2657 }
2658
2659 static int kvm_resume(struct sys_device *dev)
2660 {
2661         hardware_enable(NULL);
2662         return 0;
2663 }
2664
2665 static struct sysdev_class kvm_sysdev_class = {
2666         .name = "kvm",
2667         .suspend = kvm_suspend,
2668         .resume = kvm_resume,
2669 };
2670
2671 static struct sys_device kvm_sysdev = {
2672         .id = 0,
2673         .cls = &kvm_sysdev_class,
2674 };
2675
2676 struct page *bad_page;
2677 pfn_t bad_pfn;
2678
2679 static inline
2680 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2681 {
2682         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2683 }
2684
2685 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2686 {
2687         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2688
2689         kvm_arch_vcpu_load(vcpu, cpu);
2690 }
2691
2692 static void kvm_sched_out(struct preempt_notifier *pn,
2693                           struct task_struct *next)
2694 {
2695         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2696
2697         kvm_arch_vcpu_put(vcpu);
2698 }
2699
2700 int kvm_init(void *opaque, unsigned int vcpu_size,
2701                   struct module *module)
2702 {
2703         int r;
2704         int cpu;
2705
2706         kvm_init_debug();
2707
2708         r = kvm_arch_init(opaque);
2709         if (r)
2710                 goto out_fail;
2711
2712         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2713
2714         if (bad_page == NULL) {
2715                 r = -ENOMEM;
2716                 goto out;
2717         }
2718
2719         bad_pfn = page_to_pfn(bad_page);
2720
2721         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2722                 r = -ENOMEM;
2723                 goto out_free_0;
2724         }
2725
2726         r = kvm_arch_hardware_setup();
2727         if (r < 0)
2728                 goto out_free_0a;
2729
2730         for_each_online_cpu(cpu) {
2731                 smp_call_function_single(cpu,
2732                                 kvm_arch_check_processor_compat,
2733                                 &r, 1);
2734                 if (r < 0)
2735                         goto out_free_1;
2736         }
2737
2738         on_each_cpu(hardware_enable, NULL, 1);
2739         r = register_cpu_notifier(&kvm_cpu_notifier);
2740         if (r)
2741                 goto out_free_2;
2742         register_reboot_notifier(&kvm_reboot_notifier);
2743
2744         r = sysdev_class_register(&kvm_sysdev_class);
2745         if (r)
2746                 goto out_free_3;
2747
2748         r = sysdev_register(&kvm_sysdev);
2749         if (r)
2750                 goto out_free_4;
2751
2752         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2753         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2754                                            __alignof__(struct kvm_vcpu),
2755                                            0, NULL);
2756         if (!kvm_vcpu_cache) {
2757                 r = -ENOMEM;
2758                 goto out_free_5;
2759         }
2760
2761         kvm_chardev_ops.owner = module;
2762         kvm_vm_fops.owner = module;
2763         kvm_vcpu_fops.owner = module;
2764
2765         r = misc_register(&kvm_dev);
2766         if (r) {
2767                 printk(KERN_ERR "kvm: misc device register failed\n");
2768                 goto out_free;
2769         }
2770
2771         kvm_preempt_ops.sched_in = kvm_sched_in;
2772         kvm_preempt_ops.sched_out = kvm_sched_out;
2773
2774         return 0;
2775
2776 out_free:
2777         kmem_cache_destroy(kvm_vcpu_cache);
2778 out_free_5:
2779         sysdev_unregister(&kvm_sysdev);
2780 out_free_4:
2781         sysdev_class_unregister(&kvm_sysdev_class);
2782 out_free_3:
2783         unregister_reboot_notifier(&kvm_reboot_notifier);
2784         unregister_cpu_notifier(&kvm_cpu_notifier);
2785 out_free_2:
2786         on_each_cpu(hardware_disable, NULL, 1);
2787 out_free_1:
2788         kvm_arch_hardware_unsetup();
2789 out_free_0a:
2790         free_cpumask_var(cpus_hardware_enabled);
2791 out_free_0:
2792         __free_page(bad_page);
2793 out:
2794         kvm_arch_exit();
2795 out_fail:
2796         kvm_exit_debug();
2797         return r;
2798 }
2799 EXPORT_SYMBOL_GPL(kvm_init);
2800
2801 void kvm_exit(void)
2802 {
2803         tracepoint_synchronize_unregister();
2804         misc_deregister(&kvm_dev);
2805         kmem_cache_destroy(kvm_vcpu_cache);
2806         sysdev_unregister(&kvm_sysdev);
2807         sysdev_class_unregister(&kvm_sysdev_class);
2808         unregister_reboot_notifier(&kvm_reboot_notifier);
2809         unregister_cpu_notifier(&kvm_cpu_notifier);
2810         on_each_cpu(hardware_disable, NULL, 1);
2811         kvm_arch_hardware_unsetup();
2812         kvm_arch_exit();
2813         kvm_exit_debug();
2814         free_cpumask_var(cpus_hardware_enabled);
2815         __free_page(bad_page);
2816 }
2817 EXPORT_SYMBOL_GPL(kvm_exit);