KVM: Activate Virtualization On Demand
[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 #define CREATE_TRACE_POINTS
57 #include <trace/events/kvm.h>
58
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
61
62 /*
63  * Ordering of locks:
64  *
65  *              kvm->slots_lock --> kvm->lock --> kvm->irq_lock
66  */
67
68 DEFINE_SPINLOCK(kvm_lock);
69 LIST_HEAD(vm_list);
70
71 static cpumask_var_t cpus_hardware_enabled;
72 static int kvm_usage_count = 0;
73 static atomic_t hardware_enable_failed;
74
75 struct kmem_cache *kvm_vcpu_cache;
76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
77
78 static __read_mostly struct preempt_ops kvm_preempt_ops;
79
80 struct dentry *kvm_debugfs_dir;
81
82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
83                            unsigned long arg);
84 static int hardware_enable_all(void);
85 static void hardware_disable_all(void);
86
87 static bool kvm_rebooting;
88
89 static bool largepages_enabled = true;
90
91 inline int kvm_is_mmio_pfn(pfn_t pfn)
92 {
93         if (pfn_valid(pfn)) {
94                 struct page *page = compound_head(pfn_to_page(pfn));
95                 return PageReserved(page);
96         }
97
98         return true;
99 }
100
101 /*
102  * Switches to specified vcpu, until a matching vcpu_put()
103  */
104 void vcpu_load(struct kvm_vcpu *vcpu)
105 {
106         int cpu;
107
108         mutex_lock(&vcpu->mutex);
109         cpu = get_cpu();
110         preempt_notifier_register(&vcpu->preempt_notifier);
111         kvm_arch_vcpu_load(vcpu, cpu);
112         put_cpu();
113 }
114
115 void vcpu_put(struct kvm_vcpu *vcpu)
116 {
117         preempt_disable();
118         kvm_arch_vcpu_put(vcpu);
119         preempt_notifier_unregister(&vcpu->preempt_notifier);
120         preempt_enable();
121         mutex_unlock(&vcpu->mutex);
122 }
123
124 static void ack_flush(void *_completed)
125 {
126 }
127
128 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
129 {
130         int i, cpu, me;
131         cpumask_var_t cpus;
132         bool called = true;
133         struct kvm_vcpu *vcpu;
134
135         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
136
137         spin_lock(&kvm->requests_lock);
138         me = smp_processor_id();
139         kvm_for_each_vcpu(i, vcpu, kvm) {
140                 if (test_and_set_bit(req, &vcpu->requests))
141                         continue;
142                 cpu = vcpu->cpu;
143                 if (cpus != NULL && cpu != -1 && cpu != me)
144                         cpumask_set_cpu(cpu, cpus);
145         }
146         if (unlikely(cpus == NULL))
147                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
148         else if (!cpumask_empty(cpus))
149                 smp_call_function_many(cpus, ack_flush, NULL, 1);
150         else
151                 called = false;
152         spin_unlock(&kvm->requests_lock);
153         free_cpumask_var(cpus);
154         return called;
155 }
156
157 void kvm_flush_remote_tlbs(struct kvm *kvm)
158 {
159         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
160                 ++kvm->stat.remote_tlb_flush;
161 }
162
163 void kvm_reload_remote_mmus(struct kvm *kvm)
164 {
165         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
166 }
167
168 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
169 {
170         struct page *page;
171         int r;
172
173         mutex_init(&vcpu->mutex);
174         vcpu->cpu = -1;
175         vcpu->kvm = kvm;
176         vcpu->vcpu_id = id;
177         init_waitqueue_head(&vcpu->wq);
178
179         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
180         if (!page) {
181                 r = -ENOMEM;
182                 goto fail;
183         }
184         vcpu->run = page_address(page);
185
186         r = kvm_arch_vcpu_init(vcpu);
187         if (r < 0)
188                 goto fail_free_run;
189         return 0;
190
191 fail_free_run:
192         free_page((unsigned long)vcpu->run);
193 fail:
194         return r;
195 }
196 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
197
198 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
199 {
200         kvm_arch_vcpu_uninit(vcpu);
201         free_page((unsigned long)vcpu->run);
202 }
203 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
204
205 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
206 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
207 {
208         return container_of(mn, struct kvm, mmu_notifier);
209 }
210
211 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
212                                              struct mm_struct *mm,
213                                              unsigned long address)
214 {
215         struct kvm *kvm = mmu_notifier_to_kvm(mn);
216         int need_tlb_flush;
217
218         /*
219          * When ->invalidate_page runs, the linux pte has been zapped
220          * already but the page is still allocated until
221          * ->invalidate_page returns. So if we increase the sequence
222          * here the kvm page fault will notice if the spte can't be
223          * established because the page is going to be freed. If
224          * instead the kvm page fault establishes the spte before
225          * ->invalidate_page runs, kvm_unmap_hva will release it
226          * before returning.
227          *
228          * The sequence increase only need to be seen at spin_unlock
229          * time, and not at spin_lock time.
230          *
231          * Increasing the sequence after the spin_unlock would be
232          * unsafe because the kvm page fault could then establish the
233          * pte after kvm_unmap_hva returned, without noticing the page
234          * is going to be freed.
235          */
236         spin_lock(&kvm->mmu_lock);
237         kvm->mmu_notifier_seq++;
238         need_tlb_flush = kvm_unmap_hva(kvm, address);
239         spin_unlock(&kvm->mmu_lock);
240
241         /* we've to flush the tlb before the pages can be freed */
242         if (need_tlb_flush)
243                 kvm_flush_remote_tlbs(kvm);
244
245 }
246
247 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
248                                         struct mm_struct *mm,
249                                         unsigned long address,
250                                         pte_t pte)
251 {
252         struct kvm *kvm = mmu_notifier_to_kvm(mn);
253
254         spin_lock(&kvm->mmu_lock);
255         kvm->mmu_notifier_seq++;
256         kvm_set_spte_hva(kvm, address, pte);
257         spin_unlock(&kvm->mmu_lock);
258 }
259
260 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
261                                                     struct mm_struct *mm,
262                                                     unsigned long start,
263                                                     unsigned long end)
264 {
265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
266         int need_tlb_flush = 0;
267
268         spin_lock(&kvm->mmu_lock);
269         /*
270          * The count increase must become visible at unlock time as no
271          * spte can be established without taking the mmu_lock and
272          * count is also read inside the mmu_lock critical section.
273          */
274         kvm->mmu_notifier_count++;
275         for (; start < end; start += PAGE_SIZE)
276                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
277         spin_unlock(&kvm->mmu_lock);
278
279         /* we've to flush the tlb before the pages can be freed */
280         if (need_tlb_flush)
281                 kvm_flush_remote_tlbs(kvm);
282 }
283
284 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
285                                                   struct mm_struct *mm,
286                                                   unsigned long start,
287                                                   unsigned long end)
288 {
289         struct kvm *kvm = mmu_notifier_to_kvm(mn);
290
291         spin_lock(&kvm->mmu_lock);
292         /*
293          * This sequence increase will notify the kvm page fault that
294          * the page that is going to be mapped in the spte could have
295          * been freed.
296          */
297         kvm->mmu_notifier_seq++;
298         /*
299          * The above sequence increase must be visible before the
300          * below count decrease but both values are read by the kvm
301          * page fault under mmu_lock spinlock so we don't need to add
302          * a smb_wmb() here in between the two.
303          */
304         kvm->mmu_notifier_count--;
305         spin_unlock(&kvm->mmu_lock);
306
307         BUG_ON(kvm->mmu_notifier_count < 0);
308 }
309
310 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
311                                               struct mm_struct *mm,
312                                               unsigned long address)
313 {
314         struct kvm *kvm = mmu_notifier_to_kvm(mn);
315         int young;
316
317         spin_lock(&kvm->mmu_lock);
318         young = kvm_age_hva(kvm, address);
319         spin_unlock(&kvm->mmu_lock);
320
321         if (young)
322                 kvm_flush_remote_tlbs(kvm);
323
324         return young;
325 }
326
327 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
328                                      struct mm_struct *mm)
329 {
330         struct kvm *kvm = mmu_notifier_to_kvm(mn);
331         kvm_arch_flush_shadow(kvm);
332 }
333
334 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
335         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
336         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
337         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
338         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
339         .change_pte             = kvm_mmu_notifier_change_pte,
340         .release                = kvm_mmu_notifier_release,
341 };
342 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
343
344 static struct kvm *kvm_create_vm(void)
345 {
346         int r = 0;
347         struct kvm *kvm = kvm_arch_create_vm();
348 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
349         struct page *page;
350 #endif
351
352         if (IS_ERR(kvm))
353                 goto out;
354
355         r = hardware_enable_all();
356         if (r)
357                 goto out_err_nodisable;
358
359 #ifdef CONFIG_HAVE_KVM_IRQCHIP
360         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
361         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
362 #endif
363
364 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
365         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
366         if (!page) {
367                 r = -ENOMEM;
368                 goto out_err;
369         }
370         kvm->coalesced_mmio_ring =
371                         (struct kvm_coalesced_mmio_ring *)page_address(page);
372 #endif
373
374 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
375         {
376                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
377                 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
378                 if (r) {
379 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
380                         put_page(page);
381 #endif
382                         goto out_err;
383                 }
384         }
385 #endif
386
387         kvm->mm = current->mm;
388         atomic_inc(&kvm->mm->mm_count);
389         spin_lock_init(&kvm->mmu_lock);
390         spin_lock_init(&kvm->requests_lock);
391         kvm_io_bus_init(&kvm->pio_bus);
392         kvm_eventfd_init(kvm);
393         mutex_init(&kvm->lock);
394         mutex_init(&kvm->irq_lock);
395         kvm_io_bus_init(&kvm->mmio_bus);
396         init_rwsem(&kvm->slots_lock);
397         atomic_set(&kvm->users_count, 1);
398         spin_lock(&kvm_lock);
399         list_add(&kvm->vm_list, &vm_list);
400         spin_unlock(&kvm_lock);
401 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
402         kvm_coalesced_mmio_init(kvm);
403 #endif
404 out:
405         return kvm;
406
407 out_err:
408         hardware_disable_all();
409 out_err_nodisable:
410         kfree(kvm);
411         return ERR_PTR(r);
412 }
413
414 /*
415  * Free any memory in @free but not in @dont.
416  */
417 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
418                                   struct kvm_memory_slot *dont)
419 {
420         int i;
421
422         if (!dont || free->rmap != dont->rmap)
423                 vfree(free->rmap);
424
425         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
426                 vfree(free->dirty_bitmap);
427
428
429         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
430                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
431                         vfree(free->lpage_info[i]);
432                         free->lpage_info[i] = NULL;
433                 }
434         }
435
436         free->npages = 0;
437         free->dirty_bitmap = NULL;
438         free->rmap = NULL;
439 }
440
441 void kvm_free_physmem(struct kvm *kvm)
442 {
443         int i;
444
445         for (i = 0; i < kvm->nmemslots; ++i)
446                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
447 }
448
449 static void kvm_destroy_vm(struct kvm *kvm)
450 {
451         struct mm_struct *mm = kvm->mm;
452
453         kvm_arch_sync_events(kvm);
454         spin_lock(&kvm_lock);
455         list_del(&kvm->vm_list);
456         spin_unlock(&kvm_lock);
457         kvm_free_irq_routing(kvm);
458         kvm_io_bus_destroy(&kvm->pio_bus);
459         kvm_io_bus_destroy(&kvm->mmio_bus);
460 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
461         if (kvm->coalesced_mmio_ring != NULL)
462                 free_page((unsigned long)kvm->coalesced_mmio_ring);
463 #endif
464 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
465         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
466 #else
467         kvm_arch_flush_shadow(kvm);
468 #endif
469         kvm_arch_destroy_vm(kvm);
470         hardware_disable_all();
471         mmdrop(mm);
472 }
473
474 void kvm_get_kvm(struct kvm *kvm)
475 {
476         atomic_inc(&kvm->users_count);
477 }
478 EXPORT_SYMBOL_GPL(kvm_get_kvm);
479
480 void kvm_put_kvm(struct kvm *kvm)
481 {
482         if (atomic_dec_and_test(&kvm->users_count))
483                 kvm_destroy_vm(kvm);
484 }
485 EXPORT_SYMBOL_GPL(kvm_put_kvm);
486
487
488 static int kvm_vm_release(struct inode *inode, struct file *filp)
489 {
490         struct kvm *kvm = filp->private_data;
491
492         kvm_irqfd_release(kvm);
493
494         kvm_put_kvm(kvm);
495         return 0;
496 }
497
498 /*
499  * Allocate some memory and give it an address in the guest physical address
500  * space.
501  *
502  * Discontiguous memory is allowed, mostly for framebuffers.
503  *
504  * Must be called holding mmap_sem for write.
505  */
506 int __kvm_set_memory_region(struct kvm *kvm,
507                             struct kvm_userspace_memory_region *mem,
508                             int user_alloc)
509 {
510         int r;
511         gfn_t base_gfn;
512         unsigned long npages;
513         unsigned long i;
514         struct kvm_memory_slot *memslot;
515         struct kvm_memory_slot old, new;
516
517         r = -EINVAL;
518         /* General sanity checks */
519         if (mem->memory_size & (PAGE_SIZE - 1))
520                 goto out;
521         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
522                 goto out;
523         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
524                 goto out;
525         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
526                 goto out;
527         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
528                 goto out;
529
530         memslot = &kvm->memslots[mem->slot];
531         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
532         npages = mem->memory_size >> PAGE_SHIFT;
533
534         if (!npages)
535                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
536
537         new = old = *memslot;
538
539         new.base_gfn = base_gfn;
540         new.npages = npages;
541         new.flags = mem->flags;
542
543         /* Disallow changing a memory slot's size. */
544         r = -EINVAL;
545         if (npages && old.npages && npages != old.npages)
546                 goto out_free;
547
548         /* Check for overlaps */
549         r = -EEXIST;
550         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
551                 struct kvm_memory_slot *s = &kvm->memslots[i];
552
553                 if (s == memslot || !s->npages)
554                         continue;
555                 if (!((base_gfn + npages <= s->base_gfn) ||
556                       (base_gfn >= s->base_gfn + s->npages)))
557                         goto out_free;
558         }
559
560         /* Free page dirty bitmap if unneeded */
561         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
562                 new.dirty_bitmap = NULL;
563
564         r = -ENOMEM;
565
566         /* Allocate if a slot is being created */
567 #ifndef CONFIG_S390
568         if (npages && !new.rmap) {
569                 new.rmap = vmalloc(npages * sizeof(struct page *));
570
571                 if (!new.rmap)
572                         goto out_free;
573
574                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
575
576                 new.user_alloc = user_alloc;
577                 /*
578                  * hva_to_rmmap() serialzies with the mmu_lock and to be
579                  * safe it has to ignore memslots with !user_alloc &&
580                  * !userspace_addr.
581                  */
582                 if (user_alloc)
583                         new.userspace_addr = mem->userspace_addr;
584                 else
585                         new.userspace_addr = 0;
586         }
587         if (!npages)
588                 goto skip_lpage;
589
590         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
591                 unsigned long ugfn;
592                 unsigned long j;
593                 int lpages;
594                 int level = i + 2;
595
596                 /* Avoid unused variable warning if no large pages */
597                 (void)level;
598
599                 if (new.lpage_info[i])
600                         continue;
601
602                 lpages = 1 + (base_gfn + npages - 1) /
603                              KVM_PAGES_PER_HPAGE(level);
604                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
605
606                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
607
608                 if (!new.lpage_info[i])
609                         goto out_free;
610
611                 memset(new.lpage_info[i], 0,
612                        lpages * sizeof(*new.lpage_info[i]));
613
614                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
615                         new.lpage_info[i][0].write_count = 1;
616                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
617                         new.lpage_info[i][lpages - 1].write_count = 1;
618                 ugfn = new.userspace_addr >> PAGE_SHIFT;
619                 /*
620                  * If the gfn and userspace address are not aligned wrt each
621                  * other, or if explicitly asked to, disable large page
622                  * support for this slot
623                  */
624                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
625                     !largepages_enabled)
626                         for (j = 0; j < lpages; ++j)
627                                 new.lpage_info[i][j].write_count = 1;
628         }
629
630 skip_lpage:
631
632         /* Allocate page dirty bitmap if needed */
633         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
634                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
635
636                 new.dirty_bitmap = vmalloc(dirty_bytes);
637                 if (!new.dirty_bitmap)
638                         goto out_free;
639                 memset(new.dirty_bitmap, 0, dirty_bytes);
640                 if (old.npages)
641                         kvm_arch_flush_shadow(kvm);
642         }
643 #else  /* not defined CONFIG_S390 */
644         new.user_alloc = user_alloc;
645         if (user_alloc)
646                 new.userspace_addr = mem->userspace_addr;
647 #endif /* not defined CONFIG_S390 */
648
649         if (!npages)
650                 kvm_arch_flush_shadow(kvm);
651
652         spin_lock(&kvm->mmu_lock);
653         if (mem->slot >= kvm->nmemslots)
654                 kvm->nmemslots = mem->slot + 1;
655
656         *memslot = new;
657         spin_unlock(&kvm->mmu_lock);
658
659         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
660         if (r) {
661                 spin_lock(&kvm->mmu_lock);
662                 *memslot = old;
663                 spin_unlock(&kvm->mmu_lock);
664                 goto out_free;
665         }
666
667         kvm_free_physmem_slot(&old, npages ? &new : NULL);
668         /* Slot deletion case: we have to update the current slot */
669         spin_lock(&kvm->mmu_lock);
670         if (!npages)
671                 *memslot = old;
672         spin_unlock(&kvm->mmu_lock);
673 #ifdef CONFIG_DMAR
674         /* map the pages in iommu page table */
675         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
676         if (r)
677                 goto out;
678 #endif
679         return 0;
680
681 out_free:
682         kvm_free_physmem_slot(&new, &old);
683 out:
684         return r;
685
686 }
687 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
688
689 int kvm_set_memory_region(struct kvm *kvm,
690                           struct kvm_userspace_memory_region *mem,
691                           int user_alloc)
692 {
693         int r;
694
695         down_write(&kvm->slots_lock);
696         r = __kvm_set_memory_region(kvm, mem, user_alloc);
697         up_write(&kvm->slots_lock);
698         return r;
699 }
700 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
701
702 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
703                                    struct
704                                    kvm_userspace_memory_region *mem,
705                                    int user_alloc)
706 {
707         if (mem->slot >= KVM_MEMORY_SLOTS)
708                 return -EINVAL;
709         return kvm_set_memory_region(kvm, mem, user_alloc);
710 }
711
712 int kvm_get_dirty_log(struct kvm *kvm,
713                         struct kvm_dirty_log *log, int *is_dirty)
714 {
715         struct kvm_memory_slot *memslot;
716         int r, i;
717         int n;
718         unsigned long any = 0;
719
720         r = -EINVAL;
721         if (log->slot >= KVM_MEMORY_SLOTS)
722                 goto out;
723
724         memslot = &kvm->memslots[log->slot];
725         r = -ENOENT;
726         if (!memslot->dirty_bitmap)
727                 goto out;
728
729         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
730
731         for (i = 0; !any && i < n/sizeof(long); ++i)
732                 any = memslot->dirty_bitmap[i];
733
734         r = -EFAULT;
735         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
736                 goto out;
737
738         if (any)
739                 *is_dirty = 1;
740
741         r = 0;
742 out:
743         return r;
744 }
745
746 void kvm_disable_largepages(void)
747 {
748         largepages_enabled = false;
749 }
750 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
751
752 int is_error_page(struct page *page)
753 {
754         return page == bad_page;
755 }
756 EXPORT_SYMBOL_GPL(is_error_page);
757
758 int is_error_pfn(pfn_t pfn)
759 {
760         return pfn == bad_pfn;
761 }
762 EXPORT_SYMBOL_GPL(is_error_pfn);
763
764 static inline unsigned long bad_hva(void)
765 {
766         return PAGE_OFFSET;
767 }
768
769 int kvm_is_error_hva(unsigned long addr)
770 {
771         return addr == bad_hva();
772 }
773 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
774
775 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
776 {
777         int i;
778
779         for (i = 0; i < kvm->nmemslots; ++i) {
780                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
781
782                 if (gfn >= memslot->base_gfn
783                     && gfn < memslot->base_gfn + memslot->npages)
784                         return memslot;
785         }
786         return NULL;
787 }
788 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
789
790 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
791 {
792         gfn = unalias_gfn(kvm, gfn);
793         return gfn_to_memslot_unaliased(kvm, gfn);
794 }
795
796 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
797 {
798         int i;
799
800         gfn = unalias_gfn(kvm, gfn);
801         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
802                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
803
804                 if (gfn >= memslot->base_gfn
805                     && gfn < memslot->base_gfn + memslot->npages)
806                         return 1;
807         }
808         return 0;
809 }
810 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
811
812 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
813 {
814         struct kvm_memory_slot *slot;
815
816         gfn = unalias_gfn(kvm, gfn);
817         slot = gfn_to_memslot_unaliased(kvm, gfn);
818         if (!slot)
819                 return bad_hva();
820         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
821 }
822 EXPORT_SYMBOL_GPL(gfn_to_hva);
823
824 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
825 {
826         struct page *page[1];
827         unsigned long addr;
828         int npages;
829         pfn_t pfn;
830
831         might_sleep();
832
833         addr = gfn_to_hva(kvm, gfn);
834         if (kvm_is_error_hva(addr)) {
835                 get_page(bad_page);
836                 return page_to_pfn(bad_page);
837         }
838
839         npages = get_user_pages_fast(addr, 1, 1, page);
840
841         if (unlikely(npages != 1)) {
842                 struct vm_area_struct *vma;
843
844                 down_read(&current->mm->mmap_sem);
845                 vma = find_vma(current->mm, addr);
846
847                 if (vma == NULL || addr < vma->vm_start ||
848                     !(vma->vm_flags & VM_PFNMAP)) {
849                         up_read(&current->mm->mmap_sem);
850                         get_page(bad_page);
851                         return page_to_pfn(bad_page);
852                 }
853
854                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
855                 up_read(&current->mm->mmap_sem);
856                 BUG_ON(!kvm_is_mmio_pfn(pfn));
857         } else
858                 pfn = page_to_pfn(page[0]);
859
860         return pfn;
861 }
862
863 EXPORT_SYMBOL_GPL(gfn_to_pfn);
864
865 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
866 {
867         pfn_t pfn;
868
869         pfn = gfn_to_pfn(kvm, gfn);
870         if (!kvm_is_mmio_pfn(pfn))
871                 return pfn_to_page(pfn);
872
873         WARN_ON(kvm_is_mmio_pfn(pfn));
874
875         get_page(bad_page);
876         return bad_page;
877 }
878
879 EXPORT_SYMBOL_GPL(gfn_to_page);
880
881 void kvm_release_page_clean(struct page *page)
882 {
883         kvm_release_pfn_clean(page_to_pfn(page));
884 }
885 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
886
887 void kvm_release_pfn_clean(pfn_t pfn)
888 {
889         if (!kvm_is_mmio_pfn(pfn))
890                 put_page(pfn_to_page(pfn));
891 }
892 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
893
894 void kvm_release_page_dirty(struct page *page)
895 {
896         kvm_release_pfn_dirty(page_to_pfn(page));
897 }
898 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
899
900 void kvm_release_pfn_dirty(pfn_t pfn)
901 {
902         kvm_set_pfn_dirty(pfn);
903         kvm_release_pfn_clean(pfn);
904 }
905 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
906
907 void kvm_set_page_dirty(struct page *page)
908 {
909         kvm_set_pfn_dirty(page_to_pfn(page));
910 }
911 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
912
913 void kvm_set_pfn_dirty(pfn_t pfn)
914 {
915         if (!kvm_is_mmio_pfn(pfn)) {
916                 struct page *page = pfn_to_page(pfn);
917                 if (!PageReserved(page))
918                         SetPageDirty(page);
919         }
920 }
921 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
922
923 void kvm_set_pfn_accessed(pfn_t pfn)
924 {
925         if (!kvm_is_mmio_pfn(pfn))
926                 mark_page_accessed(pfn_to_page(pfn));
927 }
928 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
929
930 void kvm_get_pfn(pfn_t pfn)
931 {
932         if (!kvm_is_mmio_pfn(pfn))
933                 get_page(pfn_to_page(pfn));
934 }
935 EXPORT_SYMBOL_GPL(kvm_get_pfn);
936
937 static int next_segment(unsigned long len, int offset)
938 {
939         if (len > PAGE_SIZE - offset)
940                 return PAGE_SIZE - offset;
941         else
942                 return len;
943 }
944
945 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
946                         int len)
947 {
948         int r;
949         unsigned long addr;
950
951         addr = gfn_to_hva(kvm, gfn);
952         if (kvm_is_error_hva(addr))
953                 return -EFAULT;
954         r = copy_from_user(data, (void __user *)addr + offset, len);
955         if (r)
956                 return -EFAULT;
957         return 0;
958 }
959 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
960
961 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
962 {
963         gfn_t gfn = gpa >> PAGE_SHIFT;
964         int seg;
965         int offset = offset_in_page(gpa);
966         int ret;
967
968         while ((seg = next_segment(len, offset)) != 0) {
969                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
970                 if (ret < 0)
971                         return ret;
972                 offset = 0;
973                 len -= seg;
974                 data += seg;
975                 ++gfn;
976         }
977         return 0;
978 }
979 EXPORT_SYMBOL_GPL(kvm_read_guest);
980
981 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
982                           unsigned long len)
983 {
984         int r;
985         unsigned long addr;
986         gfn_t gfn = gpa >> PAGE_SHIFT;
987         int offset = offset_in_page(gpa);
988
989         addr = gfn_to_hva(kvm, gfn);
990         if (kvm_is_error_hva(addr))
991                 return -EFAULT;
992         pagefault_disable();
993         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
994         pagefault_enable();
995         if (r)
996                 return -EFAULT;
997         return 0;
998 }
999 EXPORT_SYMBOL(kvm_read_guest_atomic);
1000
1001 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1002                          int offset, int len)
1003 {
1004         int r;
1005         unsigned long addr;
1006
1007         addr = gfn_to_hva(kvm, gfn);
1008         if (kvm_is_error_hva(addr))
1009                 return -EFAULT;
1010         r = copy_to_user((void __user *)addr + offset, data, len);
1011         if (r)
1012                 return -EFAULT;
1013         mark_page_dirty(kvm, gfn);
1014         return 0;
1015 }
1016 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1017
1018 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1019                     unsigned long len)
1020 {
1021         gfn_t gfn = gpa >> PAGE_SHIFT;
1022         int seg;
1023         int offset = offset_in_page(gpa);
1024         int ret;
1025
1026         while ((seg = next_segment(len, offset)) != 0) {
1027                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1028                 if (ret < 0)
1029                         return ret;
1030                 offset = 0;
1031                 len -= seg;
1032                 data += seg;
1033                 ++gfn;
1034         }
1035         return 0;
1036 }
1037
1038 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1039 {
1040         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1041 }
1042 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1043
1044 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1045 {
1046         gfn_t gfn = gpa >> PAGE_SHIFT;
1047         int seg;
1048         int offset = offset_in_page(gpa);
1049         int ret;
1050
1051         while ((seg = next_segment(len, offset)) != 0) {
1052                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1053                 if (ret < 0)
1054                         return ret;
1055                 offset = 0;
1056                 len -= seg;
1057                 ++gfn;
1058         }
1059         return 0;
1060 }
1061 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1062
1063 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1064 {
1065         struct kvm_memory_slot *memslot;
1066
1067         gfn = unalias_gfn(kvm, gfn);
1068         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1069         if (memslot && memslot->dirty_bitmap) {
1070                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1071
1072                 /* avoid RMW */
1073                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1074                         set_bit(rel_gfn, memslot->dirty_bitmap);
1075         }
1076 }
1077
1078 /*
1079  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1080  */
1081 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1082 {
1083         DEFINE_WAIT(wait);
1084
1085         for (;;) {
1086                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1087
1088                 if (kvm_arch_vcpu_runnable(vcpu)) {
1089                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1090                         break;
1091                 }
1092                 if (kvm_cpu_has_pending_timer(vcpu))
1093                         break;
1094                 if (signal_pending(current))
1095                         break;
1096
1097                 schedule();
1098         }
1099
1100         finish_wait(&vcpu->wq, &wait);
1101 }
1102
1103 void kvm_resched(struct kvm_vcpu *vcpu)
1104 {
1105         if (!need_resched())
1106                 return;
1107         cond_resched();
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_resched);
1110
1111 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1112 {
1113         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1114         struct page *page;
1115
1116         if (vmf->pgoff == 0)
1117                 page = virt_to_page(vcpu->run);
1118 #ifdef CONFIG_X86
1119         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1120                 page = virt_to_page(vcpu->arch.pio_data);
1121 #endif
1122 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1123         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1124                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1125 #endif
1126         else
1127                 return VM_FAULT_SIGBUS;
1128         get_page(page);
1129         vmf->page = page;
1130         return 0;
1131 }
1132
1133 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1134         .fault = kvm_vcpu_fault,
1135 };
1136
1137 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1138 {
1139         vma->vm_ops = &kvm_vcpu_vm_ops;
1140         return 0;
1141 }
1142
1143 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1144 {
1145         struct kvm_vcpu *vcpu = filp->private_data;
1146
1147         kvm_put_kvm(vcpu->kvm);
1148         return 0;
1149 }
1150
1151 static struct file_operations kvm_vcpu_fops = {
1152         .release        = kvm_vcpu_release,
1153         .unlocked_ioctl = kvm_vcpu_ioctl,
1154         .compat_ioctl   = kvm_vcpu_ioctl,
1155         .mmap           = kvm_vcpu_mmap,
1156 };
1157
1158 /*
1159  * Allocates an inode for the vcpu.
1160  */
1161 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1162 {
1163         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1164 }
1165
1166 /*
1167  * Creates some virtual cpus.  Good luck creating more than one.
1168  */
1169 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1170 {
1171         int r;
1172         struct kvm_vcpu *vcpu, *v;
1173
1174         vcpu = kvm_arch_vcpu_create(kvm, id);
1175         if (IS_ERR(vcpu))
1176                 return PTR_ERR(vcpu);
1177
1178         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1179
1180         r = kvm_arch_vcpu_setup(vcpu);
1181         if (r)
1182                 return r;
1183
1184         mutex_lock(&kvm->lock);
1185         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1186                 r = -EINVAL;
1187                 goto vcpu_destroy;
1188         }
1189
1190         kvm_for_each_vcpu(r, v, kvm)
1191                 if (v->vcpu_id == id) {
1192                         r = -EEXIST;
1193                         goto vcpu_destroy;
1194                 }
1195
1196         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1197
1198         /* Now it's all set up, let userspace reach it */
1199         kvm_get_kvm(kvm);
1200         r = create_vcpu_fd(vcpu);
1201         if (r < 0) {
1202                 kvm_put_kvm(kvm);
1203                 goto vcpu_destroy;
1204         }
1205
1206         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1207         smp_wmb();
1208         atomic_inc(&kvm->online_vcpus);
1209
1210 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1211         if (kvm->bsp_vcpu_id == id)
1212                 kvm->bsp_vcpu = vcpu;
1213 #endif
1214         mutex_unlock(&kvm->lock);
1215         return r;
1216
1217 vcpu_destroy:
1218         mutex_unlock(&kvm->lock);
1219         kvm_arch_vcpu_destroy(vcpu);
1220         return r;
1221 }
1222
1223 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1224 {
1225         if (sigset) {
1226                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1227                 vcpu->sigset_active = 1;
1228                 vcpu->sigset = *sigset;
1229         } else
1230                 vcpu->sigset_active = 0;
1231         return 0;
1232 }
1233
1234 static long kvm_vcpu_ioctl(struct file *filp,
1235                            unsigned int ioctl, unsigned long arg)
1236 {
1237         struct kvm_vcpu *vcpu = filp->private_data;
1238         void __user *argp = (void __user *)arg;
1239         int r;
1240         struct kvm_fpu *fpu = NULL;
1241         struct kvm_sregs *kvm_sregs = NULL;
1242
1243         if (vcpu->kvm->mm != current->mm)
1244                 return -EIO;
1245         switch (ioctl) {
1246         case KVM_RUN:
1247                 r = -EINVAL;
1248                 if (arg)
1249                         goto out;
1250                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1251                 break;
1252         case KVM_GET_REGS: {
1253                 struct kvm_regs *kvm_regs;
1254
1255                 r = -ENOMEM;
1256                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1257                 if (!kvm_regs)
1258                         goto out;
1259                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1260                 if (r)
1261                         goto out_free1;
1262                 r = -EFAULT;
1263                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1264                         goto out_free1;
1265                 r = 0;
1266 out_free1:
1267                 kfree(kvm_regs);
1268                 break;
1269         }
1270         case KVM_SET_REGS: {
1271                 struct kvm_regs *kvm_regs;
1272
1273                 r = -ENOMEM;
1274                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1275                 if (!kvm_regs)
1276                         goto out;
1277                 r = -EFAULT;
1278                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1279                         goto out_free2;
1280                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1281                 if (r)
1282                         goto out_free2;
1283                 r = 0;
1284 out_free2:
1285                 kfree(kvm_regs);
1286                 break;
1287         }
1288         case KVM_GET_SREGS: {
1289                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1290                 r = -ENOMEM;
1291                 if (!kvm_sregs)
1292                         goto out;
1293                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1294                 if (r)
1295                         goto out;
1296                 r = -EFAULT;
1297                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1298                         goto out;
1299                 r = 0;
1300                 break;
1301         }
1302         case KVM_SET_SREGS: {
1303                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1304                 r = -ENOMEM;
1305                 if (!kvm_sregs)
1306                         goto out;
1307                 r = -EFAULT;
1308                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1309                         goto out;
1310                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1311                 if (r)
1312                         goto out;
1313                 r = 0;
1314                 break;
1315         }
1316         case KVM_GET_MP_STATE: {
1317                 struct kvm_mp_state mp_state;
1318
1319                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1320                 if (r)
1321                         goto out;
1322                 r = -EFAULT;
1323                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1324                         goto out;
1325                 r = 0;
1326                 break;
1327         }
1328         case KVM_SET_MP_STATE: {
1329                 struct kvm_mp_state mp_state;
1330
1331                 r = -EFAULT;
1332                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1333                         goto out;
1334                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1335                 if (r)
1336                         goto out;
1337                 r = 0;
1338                 break;
1339         }
1340         case KVM_TRANSLATE: {
1341                 struct kvm_translation tr;
1342
1343                 r = -EFAULT;
1344                 if (copy_from_user(&tr, argp, sizeof tr))
1345                         goto out;
1346                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1347                 if (r)
1348                         goto out;
1349                 r = -EFAULT;
1350                 if (copy_to_user(argp, &tr, sizeof tr))
1351                         goto out;
1352                 r = 0;
1353                 break;
1354         }
1355         case KVM_SET_GUEST_DEBUG: {
1356                 struct kvm_guest_debug dbg;
1357
1358                 r = -EFAULT;
1359                 if (copy_from_user(&dbg, argp, sizeof dbg))
1360                         goto out;
1361                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1362                 if (r)
1363                         goto out;
1364                 r = 0;
1365                 break;
1366         }
1367         case KVM_SET_SIGNAL_MASK: {
1368                 struct kvm_signal_mask __user *sigmask_arg = argp;
1369                 struct kvm_signal_mask kvm_sigmask;
1370                 sigset_t sigset, *p;
1371
1372                 p = NULL;
1373                 if (argp) {
1374                         r = -EFAULT;
1375                         if (copy_from_user(&kvm_sigmask, argp,
1376                                            sizeof kvm_sigmask))
1377                                 goto out;
1378                         r = -EINVAL;
1379                         if (kvm_sigmask.len != sizeof sigset)
1380                                 goto out;
1381                         r = -EFAULT;
1382                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1383                                            sizeof sigset))
1384                                 goto out;
1385                         p = &sigset;
1386                 }
1387                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1388                 break;
1389         }
1390         case KVM_GET_FPU: {
1391                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1392                 r = -ENOMEM;
1393                 if (!fpu)
1394                         goto out;
1395                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1396                 if (r)
1397                         goto out;
1398                 r = -EFAULT;
1399                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1400                         goto out;
1401                 r = 0;
1402                 break;
1403         }
1404         case KVM_SET_FPU: {
1405                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1406                 r = -ENOMEM;
1407                 if (!fpu)
1408                         goto out;
1409                 r = -EFAULT;
1410                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1411                         goto out;
1412                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1413                 if (r)
1414                         goto out;
1415                 r = 0;
1416                 break;
1417         }
1418         default:
1419                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1420         }
1421 out:
1422         kfree(fpu);
1423         kfree(kvm_sregs);
1424         return r;
1425 }
1426
1427 static long kvm_vm_ioctl(struct file *filp,
1428                            unsigned int ioctl, unsigned long arg)
1429 {
1430         struct kvm *kvm = filp->private_data;
1431         void __user *argp = (void __user *)arg;
1432         int r;
1433
1434         if (kvm->mm != current->mm)
1435                 return -EIO;
1436         switch (ioctl) {
1437         case KVM_CREATE_VCPU:
1438                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1439                 if (r < 0)
1440                         goto out;
1441                 break;
1442         case KVM_SET_USER_MEMORY_REGION: {
1443                 struct kvm_userspace_memory_region kvm_userspace_mem;
1444
1445                 r = -EFAULT;
1446                 if (copy_from_user(&kvm_userspace_mem, argp,
1447                                                 sizeof kvm_userspace_mem))
1448                         goto out;
1449
1450                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1451                 if (r)
1452                         goto out;
1453                 break;
1454         }
1455         case KVM_GET_DIRTY_LOG: {
1456                 struct kvm_dirty_log log;
1457
1458                 r = -EFAULT;
1459                 if (copy_from_user(&log, argp, sizeof log))
1460                         goto out;
1461                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1462                 if (r)
1463                         goto out;
1464                 break;
1465         }
1466 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1467         case KVM_REGISTER_COALESCED_MMIO: {
1468                 struct kvm_coalesced_mmio_zone zone;
1469                 r = -EFAULT;
1470                 if (copy_from_user(&zone, argp, sizeof zone))
1471                         goto out;
1472                 r = -ENXIO;
1473                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1474                 if (r)
1475                         goto out;
1476                 r = 0;
1477                 break;
1478         }
1479         case KVM_UNREGISTER_COALESCED_MMIO: {
1480                 struct kvm_coalesced_mmio_zone zone;
1481                 r = -EFAULT;
1482                 if (copy_from_user(&zone, argp, sizeof zone))
1483                         goto out;
1484                 r = -ENXIO;
1485                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1486                 if (r)
1487                         goto out;
1488                 r = 0;
1489                 break;
1490         }
1491 #endif
1492         case KVM_IRQFD: {
1493                 struct kvm_irqfd data;
1494
1495                 r = -EFAULT;
1496                 if (copy_from_user(&data, argp, sizeof data))
1497                         goto out;
1498                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1499                 break;
1500         }
1501         case KVM_IOEVENTFD: {
1502                 struct kvm_ioeventfd data;
1503
1504                 r = -EFAULT;
1505                 if (copy_from_user(&data, argp, sizeof data))
1506                         goto out;
1507                 r = kvm_ioeventfd(kvm, &data);
1508                 break;
1509         }
1510 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1511         case KVM_SET_BOOT_CPU_ID:
1512                 r = 0;
1513                 mutex_lock(&kvm->lock);
1514                 if (atomic_read(&kvm->online_vcpus) != 0)
1515                         r = -EBUSY;
1516                 else
1517                         kvm->bsp_vcpu_id = arg;
1518                 mutex_unlock(&kvm->lock);
1519                 break;
1520 #endif
1521         default:
1522                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1523                 if (r == -ENOTTY)
1524                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1525         }
1526 out:
1527         return r;
1528 }
1529
1530 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1531 {
1532         struct page *page[1];
1533         unsigned long addr;
1534         int npages;
1535         gfn_t gfn = vmf->pgoff;
1536         struct kvm *kvm = vma->vm_file->private_data;
1537
1538         addr = gfn_to_hva(kvm, gfn);
1539         if (kvm_is_error_hva(addr))
1540                 return VM_FAULT_SIGBUS;
1541
1542         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1543                                 NULL);
1544         if (unlikely(npages != 1))
1545                 return VM_FAULT_SIGBUS;
1546
1547         vmf->page = page[0];
1548         return 0;
1549 }
1550
1551 static const struct vm_operations_struct kvm_vm_vm_ops = {
1552         .fault = kvm_vm_fault,
1553 };
1554
1555 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1556 {
1557         vma->vm_ops = &kvm_vm_vm_ops;
1558         return 0;
1559 }
1560
1561 static struct file_operations kvm_vm_fops = {
1562         .release        = kvm_vm_release,
1563         .unlocked_ioctl = kvm_vm_ioctl,
1564         .compat_ioctl   = kvm_vm_ioctl,
1565         .mmap           = kvm_vm_mmap,
1566 };
1567
1568 static int kvm_dev_ioctl_create_vm(void)
1569 {
1570         int fd;
1571         struct kvm *kvm;
1572
1573         kvm = kvm_create_vm();
1574         if (IS_ERR(kvm))
1575                 return PTR_ERR(kvm);
1576         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1577         if (fd < 0)
1578                 kvm_put_kvm(kvm);
1579
1580         return fd;
1581 }
1582
1583 static long kvm_dev_ioctl_check_extension_generic(long arg)
1584 {
1585         switch (arg) {
1586         case KVM_CAP_USER_MEMORY:
1587         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1588         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1589 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1590         case KVM_CAP_SET_BOOT_CPU_ID:
1591 #endif
1592                 return 1;
1593 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1594         case KVM_CAP_IRQ_ROUTING:
1595                 return KVM_MAX_IRQ_ROUTES;
1596 #endif
1597         default:
1598                 break;
1599         }
1600         return kvm_dev_ioctl_check_extension(arg);
1601 }
1602
1603 static long kvm_dev_ioctl(struct file *filp,
1604                           unsigned int ioctl, unsigned long arg)
1605 {
1606         long r = -EINVAL;
1607
1608         switch (ioctl) {
1609         case KVM_GET_API_VERSION:
1610                 r = -EINVAL;
1611                 if (arg)
1612                         goto out;
1613                 r = KVM_API_VERSION;
1614                 break;
1615         case KVM_CREATE_VM:
1616                 r = -EINVAL;
1617                 if (arg)
1618                         goto out;
1619                 r = kvm_dev_ioctl_create_vm();
1620                 break;
1621         case KVM_CHECK_EXTENSION:
1622                 r = kvm_dev_ioctl_check_extension_generic(arg);
1623                 break;
1624         case KVM_GET_VCPU_MMAP_SIZE:
1625                 r = -EINVAL;
1626                 if (arg)
1627                         goto out;
1628                 r = PAGE_SIZE;     /* struct kvm_run */
1629 #ifdef CONFIG_X86
1630                 r += PAGE_SIZE;    /* pio data page */
1631 #endif
1632 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1633                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1634 #endif
1635                 break;
1636         case KVM_TRACE_ENABLE:
1637         case KVM_TRACE_PAUSE:
1638         case KVM_TRACE_DISABLE:
1639                 r = -EOPNOTSUPP;
1640                 break;
1641         default:
1642                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1643         }
1644 out:
1645         return r;
1646 }
1647
1648 static struct file_operations kvm_chardev_ops = {
1649         .unlocked_ioctl = kvm_dev_ioctl,
1650         .compat_ioctl   = kvm_dev_ioctl,
1651 };
1652
1653 static struct miscdevice kvm_dev = {
1654         KVM_MINOR,
1655         "kvm",
1656         &kvm_chardev_ops,
1657 };
1658
1659 static void hardware_enable(void *junk)
1660 {
1661         int cpu = raw_smp_processor_id();
1662         int r;
1663
1664         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1665                 return;
1666
1667         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1668
1669         r = kvm_arch_hardware_enable(NULL);
1670
1671         if (r) {
1672                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1673                 atomic_inc(&hardware_enable_failed);
1674                 printk(KERN_INFO "kvm: enabling virtualization on "
1675                                  "CPU%d failed\n", cpu);
1676         }
1677 }
1678
1679 static void hardware_disable(void *junk)
1680 {
1681         int cpu = raw_smp_processor_id();
1682
1683         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1684                 return;
1685         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1686         kvm_arch_hardware_disable(NULL);
1687 }
1688
1689 static void hardware_disable_all_nolock(void)
1690 {
1691         BUG_ON(!kvm_usage_count);
1692
1693         kvm_usage_count--;
1694         if (!kvm_usage_count)
1695                 on_each_cpu(hardware_disable, NULL, 1);
1696 }
1697
1698 static void hardware_disable_all(void)
1699 {
1700         spin_lock(&kvm_lock);
1701         hardware_disable_all_nolock();
1702         spin_unlock(&kvm_lock);
1703 }
1704
1705 static int hardware_enable_all(void)
1706 {
1707         int r = 0;
1708
1709         spin_lock(&kvm_lock);
1710
1711         kvm_usage_count++;
1712         if (kvm_usage_count == 1) {
1713                 atomic_set(&hardware_enable_failed, 0);
1714                 on_each_cpu(hardware_enable, NULL, 1);
1715
1716                 if (atomic_read(&hardware_enable_failed)) {
1717                         hardware_disable_all_nolock();
1718                         r = -EBUSY;
1719                 }
1720         }
1721
1722         spin_unlock(&kvm_lock);
1723
1724         return r;
1725 }
1726
1727 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1728                            void *v)
1729 {
1730         int cpu = (long)v;
1731
1732         if (!kvm_usage_count)
1733                 return NOTIFY_OK;
1734
1735         val &= ~CPU_TASKS_FROZEN;
1736         switch (val) {
1737         case CPU_DYING:
1738                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1739                        cpu);
1740                 hardware_disable(NULL);
1741                 break;
1742         case CPU_UP_CANCELED:
1743                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1744                        cpu);
1745                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1746                 break;
1747         case CPU_ONLINE:
1748                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1749                        cpu);
1750                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1751                 break;
1752         }
1753         return NOTIFY_OK;
1754 }
1755
1756
1757 asmlinkage void kvm_handle_fault_on_reboot(void)
1758 {
1759         if (kvm_rebooting)
1760                 /* spin while reset goes on */
1761                 while (true)
1762                         ;
1763         /* Fault while not rebooting.  We want the trace. */
1764         BUG();
1765 }
1766 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1767
1768 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1769                       void *v)
1770 {
1771         /*
1772          * Some (well, at least mine) BIOSes hang on reboot if
1773          * in vmx root mode.
1774          *
1775          * And Intel TXT required VMX off for all cpu when system shutdown.
1776          */
1777         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1778         kvm_rebooting = true;
1779         on_each_cpu(hardware_disable, NULL, 1);
1780         return NOTIFY_OK;
1781 }
1782
1783 static struct notifier_block kvm_reboot_notifier = {
1784         .notifier_call = kvm_reboot,
1785         .priority = 0,
1786 };
1787
1788 void kvm_io_bus_init(struct kvm_io_bus *bus)
1789 {
1790         memset(bus, 0, sizeof(*bus));
1791 }
1792
1793 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1794 {
1795         int i;
1796
1797         for (i = 0; i < bus->dev_count; i++) {
1798                 struct kvm_io_device *pos = bus->devs[i];
1799
1800                 kvm_iodevice_destructor(pos);
1801         }
1802 }
1803
1804 /* kvm_io_bus_write - called under kvm->slots_lock */
1805 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1806                      int len, const void *val)
1807 {
1808         int i;
1809         for (i = 0; i < bus->dev_count; i++)
1810                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1811                         return 0;
1812         return -EOPNOTSUPP;
1813 }
1814
1815 /* kvm_io_bus_read - called under kvm->slots_lock */
1816 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1817 {
1818         int i;
1819         for (i = 0; i < bus->dev_count; i++)
1820                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1821                         return 0;
1822         return -EOPNOTSUPP;
1823 }
1824
1825 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1826                              struct kvm_io_device *dev)
1827 {
1828         int ret;
1829
1830         down_write(&kvm->slots_lock);
1831         ret = __kvm_io_bus_register_dev(bus, dev);
1832         up_write(&kvm->slots_lock);
1833
1834         return ret;
1835 }
1836
1837 /* An unlocked version. Caller must have write lock on slots_lock. */
1838 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1839                               struct kvm_io_device *dev)
1840 {
1841         if (bus->dev_count > NR_IOBUS_DEVS-1)
1842                 return -ENOSPC;
1843
1844         bus->devs[bus->dev_count++] = dev;
1845
1846         return 0;
1847 }
1848
1849 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1850                                struct kvm_io_bus *bus,
1851                                struct kvm_io_device *dev)
1852 {
1853         down_write(&kvm->slots_lock);
1854         __kvm_io_bus_unregister_dev(bus, dev);
1855         up_write(&kvm->slots_lock);
1856 }
1857
1858 /* An unlocked version. Caller must have write lock on slots_lock. */
1859 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1860                                  struct kvm_io_device *dev)
1861 {
1862         int i;
1863
1864         for (i = 0; i < bus->dev_count; i++)
1865                 if (bus->devs[i] == dev) {
1866                         bus->devs[i] = bus->devs[--bus->dev_count];
1867                         break;
1868                 }
1869 }
1870
1871 static struct notifier_block kvm_cpu_notifier = {
1872         .notifier_call = kvm_cpu_hotplug,
1873         .priority = 20, /* must be > scheduler priority */
1874 };
1875
1876 static int vm_stat_get(void *_offset, u64 *val)
1877 {
1878         unsigned offset = (long)_offset;
1879         struct kvm *kvm;
1880
1881         *val = 0;
1882         spin_lock(&kvm_lock);
1883         list_for_each_entry(kvm, &vm_list, vm_list)
1884                 *val += *(u32 *)((void *)kvm + offset);
1885         spin_unlock(&kvm_lock);
1886         return 0;
1887 }
1888
1889 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1890
1891 static int vcpu_stat_get(void *_offset, u64 *val)
1892 {
1893         unsigned offset = (long)_offset;
1894         struct kvm *kvm;
1895         struct kvm_vcpu *vcpu;
1896         int i;
1897
1898         *val = 0;
1899         spin_lock(&kvm_lock);
1900         list_for_each_entry(kvm, &vm_list, vm_list)
1901                 kvm_for_each_vcpu(i, vcpu, kvm)
1902                         *val += *(u32 *)((void *)vcpu + offset);
1903
1904         spin_unlock(&kvm_lock);
1905         return 0;
1906 }
1907
1908 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1909
1910 static const struct file_operations *stat_fops[] = {
1911         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1912         [KVM_STAT_VM]   = &vm_stat_fops,
1913 };
1914
1915 static void kvm_init_debug(void)
1916 {
1917         struct kvm_stats_debugfs_item *p;
1918
1919         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1920         for (p = debugfs_entries; p->name; ++p)
1921                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1922                                                 (void *)(long)p->offset,
1923                                                 stat_fops[p->kind]);
1924 }
1925
1926 static void kvm_exit_debug(void)
1927 {
1928         struct kvm_stats_debugfs_item *p;
1929
1930         for (p = debugfs_entries; p->name; ++p)
1931                 debugfs_remove(p->dentry);
1932         debugfs_remove(kvm_debugfs_dir);
1933 }
1934
1935 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1936 {
1937         if (kvm_usage_count)
1938                 hardware_disable(NULL);
1939         return 0;
1940 }
1941
1942 static int kvm_resume(struct sys_device *dev)
1943 {
1944         if (kvm_usage_count)
1945                 hardware_enable(NULL);
1946         return 0;
1947 }
1948
1949 static struct sysdev_class kvm_sysdev_class = {
1950         .name = "kvm",
1951         .suspend = kvm_suspend,
1952         .resume = kvm_resume,
1953 };
1954
1955 static struct sys_device kvm_sysdev = {
1956         .id = 0,
1957         .cls = &kvm_sysdev_class,
1958 };
1959
1960 struct page *bad_page;
1961 pfn_t bad_pfn;
1962
1963 static inline
1964 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1965 {
1966         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1967 }
1968
1969 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1970 {
1971         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1972
1973         kvm_arch_vcpu_load(vcpu, cpu);
1974 }
1975
1976 static void kvm_sched_out(struct preempt_notifier *pn,
1977                           struct task_struct *next)
1978 {
1979         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1980
1981         kvm_arch_vcpu_put(vcpu);
1982 }
1983
1984 int kvm_init(void *opaque, unsigned int vcpu_size,
1985                   struct module *module)
1986 {
1987         int r;
1988         int cpu;
1989
1990         r = kvm_arch_init(opaque);
1991         if (r)
1992                 goto out_fail;
1993
1994         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1995
1996         if (bad_page == NULL) {
1997                 r = -ENOMEM;
1998                 goto out;
1999         }
2000
2001         bad_pfn = page_to_pfn(bad_page);
2002
2003         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2004                 r = -ENOMEM;
2005                 goto out_free_0;
2006         }
2007
2008         r = kvm_arch_hardware_setup();
2009         if (r < 0)
2010                 goto out_free_0a;
2011
2012         for_each_online_cpu(cpu) {
2013                 smp_call_function_single(cpu,
2014                                 kvm_arch_check_processor_compat,
2015                                 &r, 1);
2016                 if (r < 0)
2017                         goto out_free_1;
2018         }
2019
2020         r = register_cpu_notifier(&kvm_cpu_notifier);
2021         if (r)
2022                 goto out_free_2;
2023         register_reboot_notifier(&kvm_reboot_notifier);
2024
2025         r = sysdev_class_register(&kvm_sysdev_class);
2026         if (r)
2027                 goto out_free_3;
2028
2029         r = sysdev_register(&kvm_sysdev);
2030         if (r)
2031                 goto out_free_4;
2032
2033         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2034         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2035                                            __alignof__(struct kvm_vcpu),
2036                                            0, NULL);
2037         if (!kvm_vcpu_cache) {
2038                 r = -ENOMEM;
2039                 goto out_free_5;
2040         }
2041
2042         kvm_chardev_ops.owner = module;
2043         kvm_vm_fops.owner = module;
2044         kvm_vcpu_fops.owner = module;
2045
2046         r = misc_register(&kvm_dev);
2047         if (r) {
2048                 printk(KERN_ERR "kvm: misc device register failed\n");
2049                 goto out_free;
2050         }
2051
2052         kvm_preempt_ops.sched_in = kvm_sched_in;
2053         kvm_preempt_ops.sched_out = kvm_sched_out;
2054
2055         kvm_init_debug();
2056
2057         return 0;
2058
2059 out_free:
2060         kmem_cache_destroy(kvm_vcpu_cache);
2061 out_free_5:
2062         sysdev_unregister(&kvm_sysdev);
2063 out_free_4:
2064         sysdev_class_unregister(&kvm_sysdev_class);
2065 out_free_3:
2066         unregister_reboot_notifier(&kvm_reboot_notifier);
2067         unregister_cpu_notifier(&kvm_cpu_notifier);
2068 out_free_2:
2069 out_free_1:
2070         kvm_arch_hardware_unsetup();
2071 out_free_0a:
2072         free_cpumask_var(cpus_hardware_enabled);
2073 out_free_0:
2074         __free_page(bad_page);
2075 out:
2076         kvm_arch_exit();
2077 out_fail:
2078         return r;
2079 }
2080 EXPORT_SYMBOL_GPL(kvm_init);
2081
2082 void kvm_exit(void)
2083 {
2084         tracepoint_synchronize_unregister();
2085         kvm_exit_debug();
2086         misc_deregister(&kvm_dev);
2087         kmem_cache_destroy(kvm_vcpu_cache);
2088         sysdev_unregister(&kvm_sysdev);
2089         sysdev_class_unregister(&kvm_sysdev_class);
2090         unregister_reboot_notifier(&kvm_reboot_notifier);
2091         unregister_cpu_notifier(&kvm_cpu_notifier);
2092         on_each_cpu(hardware_disable, NULL, 1);
2093         kvm_arch_hardware_unsetup();
2094         kvm_arch_exit();
2095         free_cpumask_var(cpus_hardware_enabled);
2096         __free_page(bad_page);
2097 }
2098 EXPORT_SYMBOL_GPL(kvm_exit);