KVM: Introduce kvm_memory_slot::arch and move lpage_info into it
[linux-3.10.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  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.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/syscore_ops.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 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 static struct page *hwpoison_page;
104 static pfn_t hwpoison_pfn;
105
106 struct page *fault_page;
107 pfn_t fault_pfn;
108
109 inline int kvm_is_mmio_pfn(pfn_t pfn)
110 {
111         if (pfn_valid(pfn)) {
112                 int reserved;
113                 struct page *tail = pfn_to_page(pfn);
114                 struct page *head = compound_trans_head(tail);
115                 reserved = PageReserved(head);
116                 if (head != tail) {
117                         /*
118                          * "head" is not a dangling pointer
119                          * (compound_trans_head takes care of that)
120                          * but the hugepage may have been splitted
121                          * from under us (and we may not hold a
122                          * reference count on the head page so it can
123                          * be reused before we run PageReferenced), so
124                          * we've to check PageTail before returning
125                          * what we just read.
126                          */
127                         smp_rmb();
128                         if (PageTail(tail))
129                                 return reserved;
130                 }
131                 return PageReserved(tail);
132         }
133
134         return true;
135 }
136
137 /*
138  * Switches to specified vcpu, until a matching vcpu_put()
139  */
140 void vcpu_load(struct kvm_vcpu *vcpu)
141 {
142         int cpu;
143
144         mutex_lock(&vcpu->mutex);
145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
146                 /* The thread running this VCPU changed. */
147                 struct pid *oldpid = vcpu->pid;
148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
149                 rcu_assign_pointer(vcpu->pid, newpid);
150                 synchronize_rcu();
151                 put_pid(oldpid);
152         }
153         cpu = get_cpu();
154         preempt_notifier_register(&vcpu->preempt_notifier);
155         kvm_arch_vcpu_load(vcpu, cpu);
156         put_cpu();
157 }
158
159 void vcpu_put(struct kvm_vcpu *vcpu)
160 {
161         preempt_disable();
162         kvm_arch_vcpu_put(vcpu);
163         preempt_notifier_unregister(&vcpu->preempt_notifier);
164         preempt_enable();
165         mutex_unlock(&vcpu->mutex);
166 }
167
168 static void ack_flush(void *_completed)
169 {
170 }
171
172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 {
174         int i, cpu, me;
175         cpumask_var_t cpus;
176         bool called = true;
177         struct kvm_vcpu *vcpu;
178
179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180
181         me = get_cpu();
182         kvm_for_each_vcpu(i, vcpu, kvm) {
183                 kvm_make_request(req, vcpu);
184                 cpu = vcpu->cpu;
185
186                 /* Set ->requests bit before we read ->mode */
187                 smp_mb();
188
189                 if (cpus != NULL && cpu != -1 && cpu != me &&
190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191                         cpumask_set_cpu(cpu, cpus);
192         }
193         if (unlikely(cpus == NULL))
194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195         else if (!cpumask_empty(cpus))
196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
197         else
198                 called = false;
199         put_cpu();
200         free_cpumask_var(cpus);
201         return called;
202 }
203
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
205 {
206         int dirty_count = kvm->tlbs_dirty;
207
208         smp_mb();
209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
210                 ++kvm->stat.remote_tlb_flush;
211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
212 }
213
214 void kvm_reload_remote_mmus(struct kvm *kvm)
215 {
216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
217 }
218
219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
220 {
221         struct page *page;
222         int r;
223
224         mutex_init(&vcpu->mutex);
225         vcpu->cpu = -1;
226         vcpu->kvm = kvm;
227         vcpu->vcpu_id = id;
228         vcpu->pid = NULL;
229         init_waitqueue_head(&vcpu->wq);
230         kvm_async_pf_vcpu_init(vcpu);
231
232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
233         if (!page) {
234                 r = -ENOMEM;
235                 goto fail;
236         }
237         vcpu->run = page_address(page);
238
239         r = kvm_arch_vcpu_init(vcpu);
240         if (r < 0)
241                 goto fail_free_run;
242         return 0;
243
244 fail_free_run:
245         free_page((unsigned long)vcpu->run);
246 fail:
247         return r;
248 }
249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250
251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 {
253         put_pid(vcpu->pid);
254         kvm_arch_vcpu_uninit(vcpu);
255         free_page((unsigned long)vcpu->run);
256 }
257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258
259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 {
262         return container_of(mn, struct kvm, mmu_notifier);
263 }
264
265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
266                                              struct mm_struct *mm,
267                                              unsigned long address)
268 {
269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
270         int need_tlb_flush, idx;
271
272         /*
273          * When ->invalidate_page runs, the linux pte has been zapped
274          * already but the page is still allocated until
275          * ->invalidate_page returns. So if we increase the sequence
276          * here the kvm page fault will notice if the spte can't be
277          * established because the page is going to be freed. If
278          * instead the kvm page fault establishes the spte before
279          * ->invalidate_page runs, kvm_unmap_hva will release it
280          * before returning.
281          *
282          * The sequence increase only need to be seen at spin_unlock
283          * time, and not at spin_lock time.
284          *
285          * Increasing the sequence after the spin_unlock would be
286          * unsafe because the kvm page fault could then establish the
287          * pte after kvm_unmap_hva returned, without noticing the page
288          * is going to be freed.
289          */
290         idx = srcu_read_lock(&kvm->srcu);
291         spin_lock(&kvm->mmu_lock);
292         kvm->mmu_notifier_seq++;
293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294         spin_unlock(&kvm->mmu_lock);
295         srcu_read_unlock(&kvm->srcu, idx);
296
297         /* we've to flush the tlb before the pages can be freed */
298         if (need_tlb_flush)
299                 kvm_flush_remote_tlbs(kvm);
300
301 }
302
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304                                         struct mm_struct *mm,
305                                         unsigned long address,
306                                         pte_t pte)
307 {
308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
309         int idx;
310
311         idx = srcu_read_lock(&kvm->srcu);
312         spin_lock(&kvm->mmu_lock);
313         kvm->mmu_notifier_seq++;
314         kvm_set_spte_hva(kvm, address, pte);
315         spin_unlock(&kvm->mmu_lock);
316         srcu_read_unlock(&kvm->srcu, idx);
317 }
318
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320                                                     struct mm_struct *mm,
321                                                     unsigned long start,
322                                                     unsigned long end)
323 {
324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
325         int need_tlb_flush = 0, idx;
326
327         idx = srcu_read_lock(&kvm->srcu);
328         spin_lock(&kvm->mmu_lock);
329         /*
330          * The count increase must become visible at unlock time as no
331          * spte can be established without taking the mmu_lock and
332          * count is also read inside the mmu_lock critical section.
333          */
334         kvm->mmu_notifier_count++;
335         for (; start < end; start += PAGE_SIZE)
336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
337         need_tlb_flush |= kvm->tlbs_dirty;
338         spin_unlock(&kvm->mmu_lock);
339         srcu_read_unlock(&kvm->srcu, idx);
340
341         /* we've to flush the tlb before the pages can be freed */
342         if (need_tlb_flush)
343                 kvm_flush_remote_tlbs(kvm);
344 }
345
346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
347                                                   struct mm_struct *mm,
348                                                   unsigned long start,
349                                                   unsigned long end)
350 {
351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
352
353         spin_lock(&kvm->mmu_lock);
354         /*
355          * This sequence increase will notify the kvm page fault that
356          * the page that is going to be mapped in the spte could have
357          * been freed.
358          */
359         kvm->mmu_notifier_seq++;
360         smp_wmb();
361         /*
362          * The above sequence increase must be visible before the
363          * below count decrease, which is ensured by the smp_wmb above
364          * in conjunction with the smp_rmb in mmu_notifier_retry().
365          */
366         kvm->mmu_notifier_count--;
367         spin_unlock(&kvm->mmu_lock);
368
369         BUG_ON(kvm->mmu_notifier_count < 0);
370 }
371
372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
373                                               struct mm_struct *mm,
374                                               unsigned long address)
375 {
376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
377         int young, idx;
378
379         idx = srcu_read_lock(&kvm->srcu);
380         spin_lock(&kvm->mmu_lock);
381         young = kvm_age_hva(kvm, address);
382         spin_unlock(&kvm->mmu_lock);
383         srcu_read_unlock(&kvm->srcu, idx);
384
385         if (young)
386                 kvm_flush_remote_tlbs(kvm);
387
388         return young;
389 }
390
391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392                                        struct mm_struct *mm,
393                                        unsigned long address)
394 {
395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
396         int young, idx;
397
398         idx = srcu_read_lock(&kvm->srcu);
399         spin_lock(&kvm->mmu_lock);
400         young = kvm_test_age_hva(kvm, address);
401         spin_unlock(&kvm->mmu_lock);
402         srcu_read_unlock(&kvm->srcu, idx);
403
404         return young;
405 }
406
407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408                                      struct mm_struct *mm)
409 {
410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
411         int idx;
412
413         idx = srcu_read_lock(&kvm->srcu);
414         kvm_arch_flush_shadow(kvm);
415         srcu_read_unlock(&kvm->srcu, idx);
416 }
417
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
423         .test_young             = kvm_mmu_notifier_test_young,
424         .change_pte             = kvm_mmu_notifier_change_pte,
425         .release                = kvm_mmu_notifier_release,
426 };
427
428 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 {
430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 }
433
434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435
436 static int kvm_init_mmu_notifier(struct kvm *kvm)
437 {
438         return 0;
439 }
440
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442
443 static void kvm_init_memslots_id(struct kvm *kvm)
444 {
445         int i;
446         struct kvm_memslots *slots = kvm->memslots;
447
448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449                 slots->id_to_index[i] = slots->memslots[i].id = i;
450 }
451
452 static struct kvm *kvm_create_vm(unsigned long type)
453 {
454         int r, i;
455         struct kvm *kvm = kvm_arch_alloc_vm();
456
457         if (!kvm)
458                 return ERR_PTR(-ENOMEM);
459
460         r = kvm_arch_init_vm(kvm, type);
461         if (r)
462                 goto out_err_nodisable;
463
464         r = hardware_enable_all();
465         if (r)
466                 goto out_err_nodisable;
467
468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
469         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 #endif
472
473         r = -ENOMEM;
474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
475         if (!kvm->memslots)
476                 goto out_err_nosrcu;
477         kvm_init_memslots_id(kvm);
478         if (init_srcu_struct(&kvm->srcu))
479                 goto out_err_nosrcu;
480         for (i = 0; i < KVM_NR_BUSES; i++) {
481                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
482                                         GFP_KERNEL);
483                 if (!kvm->buses[i])
484                         goto out_err;
485         }
486
487         spin_lock_init(&kvm->mmu_lock);
488         kvm->mm = current->mm;
489         atomic_inc(&kvm->mm->mm_count);
490         kvm_eventfd_init(kvm);
491         mutex_init(&kvm->lock);
492         mutex_init(&kvm->irq_lock);
493         mutex_init(&kvm->slots_lock);
494         atomic_set(&kvm->users_count, 1);
495
496         r = kvm_init_mmu_notifier(kvm);
497         if (r)
498                 goto out_err;
499
500         raw_spin_lock(&kvm_lock);
501         list_add(&kvm->vm_list, &vm_list);
502         raw_spin_unlock(&kvm_lock);
503
504         return kvm;
505
506 out_err:
507         cleanup_srcu_struct(&kvm->srcu);
508 out_err_nosrcu:
509         hardware_disable_all();
510 out_err_nodisable:
511         for (i = 0; i < KVM_NR_BUSES; i++)
512                 kfree(kvm->buses[i]);
513         kfree(kvm->memslots);
514         kvm_arch_free_vm(kvm);
515         return ERR_PTR(r);
516 }
517
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
519 {
520         if (!memslot->dirty_bitmap)
521                 return;
522
523         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
524                 vfree(memslot->dirty_bitmap_head);
525         else
526                 kfree(memslot->dirty_bitmap_head);
527
528         memslot->dirty_bitmap = NULL;
529         memslot->dirty_bitmap_head = NULL;
530 }
531
532 /*
533  * Free any memory in @free but not in @dont.
534  */
535 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
536                                   struct kvm_memory_slot *dont)
537 {
538         if (!dont || free->rmap != dont->rmap)
539                 vfree(free->rmap);
540
541         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
542                 kvm_destroy_dirty_bitmap(free);
543
544         kvm_arch_free_memslot(free, dont);
545
546         free->npages = 0;
547         free->rmap = NULL;
548 }
549
550 void kvm_free_physmem(struct kvm *kvm)
551 {
552         struct kvm_memslots *slots = kvm->memslots;
553         struct kvm_memory_slot *memslot;
554
555         kvm_for_each_memslot(memslot, slots)
556                 kvm_free_physmem_slot(memslot, NULL);
557
558         kfree(kvm->memslots);
559 }
560
561 static void kvm_destroy_vm(struct kvm *kvm)
562 {
563         int i;
564         struct mm_struct *mm = kvm->mm;
565
566         kvm_arch_sync_events(kvm);
567         raw_spin_lock(&kvm_lock);
568         list_del(&kvm->vm_list);
569         raw_spin_unlock(&kvm_lock);
570         kvm_free_irq_routing(kvm);
571         for (i = 0; i < KVM_NR_BUSES; i++)
572                 kvm_io_bus_destroy(kvm->buses[i]);
573         kvm_coalesced_mmio_free(kvm);
574 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
575         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
576 #else
577         kvm_arch_flush_shadow(kvm);
578 #endif
579         kvm_arch_destroy_vm(kvm);
580         kvm_free_physmem(kvm);
581         cleanup_srcu_struct(&kvm->srcu);
582         kvm_arch_free_vm(kvm);
583         hardware_disable_all();
584         mmdrop(mm);
585 }
586
587 void kvm_get_kvm(struct kvm *kvm)
588 {
589         atomic_inc(&kvm->users_count);
590 }
591 EXPORT_SYMBOL_GPL(kvm_get_kvm);
592
593 void kvm_put_kvm(struct kvm *kvm)
594 {
595         if (atomic_dec_and_test(&kvm->users_count))
596                 kvm_destroy_vm(kvm);
597 }
598 EXPORT_SYMBOL_GPL(kvm_put_kvm);
599
600
601 static int kvm_vm_release(struct inode *inode, struct file *filp)
602 {
603         struct kvm *kvm = filp->private_data;
604
605         kvm_irqfd_release(kvm);
606
607         kvm_put_kvm(kvm);
608         return 0;
609 }
610
611 /*
612  * Allocation size is twice as large as the actual dirty bitmap size.
613  * This makes it possible to do double buffering: see x86's
614  * kvm_vm_ioctl_get_dirty_log().
615  */
616 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
617 {
618 #ifndef CONFIG_S390
619         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
620
621         if (dirty_bytes > PAGE_SIZE)
622                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
623         else
624                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
625
626         if (!memslot->dirty_bitmap)
627                 return -ENOMEM;
628
629         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
630         memslot->nr_dirty_pages = 0;
631 #endif /* !CONFIG_S390 */
632         return 0;
633 }
634
635 static int cmp_memslot(const void *slot1, const void *slot2)
636 {
637         struct kvm_memory_slot *s1, *s2;
638
639         s1 = (struct kvm_memory_slot *)slot1;
640         s2 = (struct kvm_memory_slot *)slot2;
641
642         if (s1->npages < s2->npages)
643                 return 1;
644         if (s1->npages > s2->npages)
645                 return -1;
646
647         return 0;
648 }
649
650 /*
651  * Sort the memslots base on its size, so the larger slots
652  * will get better fit.
653  */
654 static void sort_memslots(struct kvm_memslots *slots)
655 {
656         int i;
657
658         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
659               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
660
661         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
662                 slots->id_to_index[slots->memslots[i].id] = i;
663 }
664
665 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
666 {
667         if (new) {
668                 int id = new->id;
669                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
670                 unsigned long npages = old->npages;
671
672                 *old = *new;
673                 if (new->npages != npages)
674                         sort_memslots(slots);
675         }
676
677         slots->generation++;
678 }
679
680 /*
681  * Allocate some memory and give it an address in the guest physical address
682  * space.
683  *
684  * Discontiguous memory is allowed, mostly for framebuffers.
685  *
686  * Must be called holding mmap_sem for write.
687  */
688 int __kvm_set_memory_region(struct kvm *kvm,
689                             struct kvm_userspace_memory_region *mem,
690                             int user_alloc)
691 {
692         int r;
693         gfn_t base_gfn;
694         unsigned long npages;
695         unsigned long i;
696         struct kvm_memory_slot *memslot;
697         struct kvm_memory_slot old, new;
698         struct kvm_memslots *slots, *old_memslots;
699
700         r = -EINVAL;
701         /* General sanity checks */
702         if (mem->memory_size & (PAGE_SIZE - 1))
703                 goto out;
704         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
705                 goto out;
706         /* We can read the guest memory with __xxx_user() later on. */
707         if (user_alloc &&
708             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
709              !access_ok(VERIFY_WRITE,
710                         (void __user *)(unsigned long)mem->userspace_addr,
711                         mem->memory_size)))
712                 goto out;
713         if (mem->slot >= KVM_MEM_SLOTS_NUM)
714                 goto out;
715         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
716                 goto out;
717
718         memslot = id_to_memslot(kvm->memslots, mem->slot);
719         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
720         npages = mem->memory_size >> PAGE_SHIFT;
721
722         r = -EINVAL;
723         if (npages > KVM_MEM_MAX_NR_PAGES)
724                 goto out;
725
726         if (!npages)
727                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
728
729         new = old = *memslot;
730
731         new.id = mem->slot;
732         new.base_gfn = base_gfn;
733         new.npages = npages;
734         new.flags = mem->flags;
735
736         /* Disallow changing a memory slot's size. */
737         r = -EINVAL;
738         if (npages && old.npages && npages != old.npages)
739                 goto out_free;
740
741         /* Check for overlaps */
742         r = -EEXIST;
743         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
744                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
745
746                 if (s == memslot || !s->npages)
747                         continue;
748                 if (!((base_gfn + npages <= s->base_gfn) ||
749                       (base_gfn >= s->base_gfn + s->npages)))
750                         goto out_free;
751         }
752
753         /* Free page dirty bitmap if unneeded */
754         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
755                 new.dirty_bitmap = NULL;
756
757         r = -ENOMEM;
758
759         /* Allocate if a slot is being created */
760         if (npages && !old.npages) {
761                 new.user_alloc = user_alloc;
762                 new.userspace_addr = mem->userspace_addr;
763 #ifndef CONFIG_S390
764                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
765                 if (!new.rmap)
766                         goto out_free;
767 #endif /* not defined CONFIG_S390 */
768                 if (kvm_arch_create_memslot(&new, npages))
769                         goto out_free;
770         }
771
772         /* Allocate page dirty bitmap if needed */
773         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
774                 if (kvm_create_dirty_bitmap(&new) < 0)
775                         goto out_free;
776                 /* destroy any largepage mappings for dirty tracking */
777         }
778
779         if (!npages) {
780                 struct kvm_memory_slot *slot;
781
782                 r = -ENOMEM;
783                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
784                                 GFP_KERNEL);
785                 if (!slots)
786                         goto out_free;
787                 slot = id_to_memslot(slots, mem->slot);
788                 slot->flags |= KVM_MEMSLOT_INVALID;
789
790                 update_memslots(slots, NULL);
791
792                 old_memslots = kvm->memslots;
793                 rcu_assign_pointer(kvm->memslots, slots);
794                 synchronize_srcu_expedited(&kvm->srcu);
795                 /* From this point no new shadow pages pointing to a deleted
796                  * memslot will be created.
797                  *
798                  * validation of sp->gfn happens in:
799                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
800                  *      - kvm_is_visible_gfn (mmu_check_roots)
801                  */
802                 kvm_arch_flush_shadow(kvm);
803                 kfree(old_memslots);
804         }
805
806         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
807         if (r)
808                 goto out_free;
809
810         /* map the pages in iommu page table */
811         if (npages) {
812                 r = kvm_iommu_map_pages(kvm, &new);
813                 if (r)
814                         goto out_free;
815         }
816
817         r = -ENOMEM;
818         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
819                         GFP_KERNEL);
820         if (!slots)
821                 goto out_free;
822
823         /* actual memory is freed via old in kvm_free_physmem_slot below */
824         if (!npages) {
825                 new.rmap = NULL;
826                 new.dirty_bitmap = NULL;
827                 memset(&new.arch, 0, sizeof(new.arch));
828         }
829
830         update_memslots(slots, &new);
831         old_memslots = kvm->memslots;
832         rcu_assign_pointer(kvm->memslots, slots);
833         synchronize_srcu_expedited(&kvm->srcu);
834
835         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
836
837         /*
838          * If the new memory slot is created, we need to clear all
839          * mmio sptes.
840          */
841         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
842                 kvm_arch_flush_shadow(kvm);
843
844         kvm_free_physmem_slot(&old, &new);
845         kfree(old_memslots);
846
847         return 0;
848
849 out_free:
850         kvm_free_physmem_slot(&new, &old);
851 out:
852         return r;
853
854 }
855 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
856
857 int kvm_set_memory_region(struct kvm *kvm,
858                           struct kvm_userspace_memory_region *mem,
859                           int user_alloc)
860 {
861         int r;
862
863         mutex_lock(&kvm->slots_lock);
864         r = __kvm_set_memory_region(kvm, mem, user_alloc);
865         mutex_unlock(&kvm->slots_lock);
866         return r;
867 }
868 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
869
870 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
871                                    struct
872                                    kvm_userspace_memory_region *mem,
873                                    int user_alloc)
874 {
875         if (mem->slot >= KVM_MEMORY_SLOTS)
876                 return -EINVAL;
877         return kvm_set_memory_region(kvm, mem, user_alloc);
878 }
879
880 int kvm_get_dirty_log(struct kvm *kvm,
881                         struct kvm_dirty_log *log, int *is_dirty)
882 {
883         struct kvm_memory_slot *memslot;
884         int r, i;
885         unsigned long n;
886         unsigned long any = 0;
887
888         r = -EINVAL;
889         if (log->slot >= KVM_MEMORY_SLOTS)
890                 goto out;
891
892         memslot = id_to_memslot(kvm->memslots, log->slot);
893         r = -ENOENT;
894         if (!memslot->dirty_bitmap)
895                 goto out;
896
897         n = kvm_dirty_bitmap_bytes(memslot);
898
899         for (i = 0; !any && i < n/sizeof(long); ++i)
900                 any = memslot->dirty_bitmap[i];
901
902         r = -EFAULT;
903         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
904                 goto out;
905
906         if (any)
907                 *is_dirty = 1;
908
909         r = 0;
910 out:
911         return r;
912 }
913
914 bool kvm_largepages_enabled(void)
915 {
916         return largepages_enabled;
917 }
918
919 void kvm_disable_largepages(void)
920 {
921         largepages_enabled = false;
922 }
923 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
924
925 int is_error_page(struct page *page)
926 {
927         return page == bad_page || page == hwpoison_page || page == fault_page;
928 }
929 EXPORT_SYMBOL_GPL(is_error_page);
930
931 int is_error_pfn(pfn_t pfn)
932 {
933         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
934 }
935 EXPORT_SYMBOL_GPL(is_error_pfn);
936
937 int is_hwpoison_pfn(pfn_t pfn)
938 {
939         return pfn == hwpoison_pfn;
940 }
941 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
942
943 int is_fault_pfn(pfn_t pfn)
944 {
945         return pfn == fault_pfn;
946 }
947 EXPORT_SYMBOL_GPL(is_fault_pfn);
948
949 int is_noslot_pfn(pfn_t pfn)
950 {
951         return pfn == bad_pfn;
952 }
953 EXPORT_SYMBOL_GPL(is_noslot_pfn);
954
955 int is_invalid_pfn(pfn_t pfn)
956 {
957         return pfn == hwpoison_pfn || pfn == fault_pfn;
958 }
959 EXPORT_SYMBOL_GPL(is_invalid_pfn);
960
961 static inline unsigned long bad_hva(void)
962 {
963         return PAGE_OFFSET;
964 }
965
966 int kvm_is_error_hva(unsigned long addr)
967 {
968         return addr == bad_hva();
969 }
970 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
971
972 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
973 {
974         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
975 }
976 EXPORT_SYMBOL_GPL(gfn_to_memslot);
977
978 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
979 {
980         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
981
982         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
983               memslot->flags & KVM_MEMSLOT_INVALID)
984                 return 0;
985
986         return 1;
987 }
988 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
989
990 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
991 {
992         struct vm_area_struct *vma;
993         unsigned long addr, size;
994
995         size = PAGE_SIZE;
996
997         addr = gfn_to_hva(kvm, gfn);
998         if (kvm_is_error_hva(addr))
999                 return PAGE_SIZE;
1000
1001         down_read(&current->mm->mmap_sem);
1002         vma = find_vma(current->mm, addr);
1003         if (!vma)
1004                 goto out;
1005
1006         size = vma_kernel_pagesize(vma);
1007
1008 out:
1009         up_read(&current->mm->mmap_sem);
1010
1011         return size;
1012 }
1013
1014 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1015                                      gfn_t *nr_pages)
1016 {
1017         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1018                 return bad_hva();
1019
1020         if (nr_pages)
1021                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1022
1023         return gfn_to_hva_memslot(slot, gfn);
1024 }
1025
1026 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1027 {
1028         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1029 }
1030 EXPORT_SYMBOL_GPL(gfn_to_hva);
1031
1032 static pfn_t get_fault_pfn(void)
1033 {
1034         get_page(fault_page);
1035         return fault_pfn;
1036 }
1037
1038 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1039         unsigned long start, int write, struct page **page)
1040 {
1041         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1042
1043         if (write)
1044                 flags |= FOLL_WRITE;
1045
1046         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1047 }
1048
1049 static inline int check_user_page_hwpoison(unsigned long addr)
1050 {
1051         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1052
1053         rc = __get_user_pages(current, current->mm, addr, 1,
1054                               flags, NULL, NULL, NULL);
1055         return rc == -EHWPOISON;
1056 }
1057
1058 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1059                         bool *async, bool write_fault, bool *writable)
1060 {
1061         struct page *page[1];
1062         int npages = 0;
1063         pfn_t pfn;
1064
1065         /* we can do it either atomically or asynchronously, not both */
1066         BUG_ON(atomic && async);
1067
1068         BUG_ON(!write_fault && !writable);
1069
1070         if (writable)
1071                 *writable = true;
1072
1073         if (atomic || async)
1074                 npages = __get_user_pages_fast(addr, 1, 1, page);
1075
1076         if (unlikely(npages != 1) && !atomic) {
1077                 might_sleep();
1078
1079                 if (writable)
1080                         *writable = write_fault;
1081
1082                 if (async) {
1083                         down_read(&current->mm->mmap_sem);
1084                         npages = get_user_page_nowait(current, current->mm,
1085                                                      addr, write_fault, page);
1086                         up_read(&current->mm->mmap_sem);
1087                 } else
1088                         npages = get_user_pages_fast(addr, 1, write_fault,
1089                                                      page);
1090
1091                 /* map read fault as writable if possible */
1092                 if (unlikely(!write_fault) && npages == 1) {
1093                         struct page *wpage[1];
1094
1095                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1096                         if (npages == 1) {
1097                                 *writable = true;
1098                                 put_page(page[0]);
1099                                 page[0] = wpage[0];
1100                         }
1101                         npages = 1;
1102                 }
1103         }
1104
1105         if (unlikely(npages != 1)) {
1106                 struct vm_area_struct *vma;
1107
1108                 if (atomic)
1109                         return get_fault_pfn();
1110
1111                 down_read(&current->mm->mmap_sem);
1112                 if (npages == -EHWPOISON ||
1113                         (!async && check_user_page_hwpoison(addr))) {
1114                         up_read(&current->mm->mmap_sem);
1115                         get_page(hwpoison_page);
1116                         return page_to_pfn(hwpoison_page);
1117                 }
1118
1119                 vma = find_vma_intersection(current->mm, addr, addr+1);
1120
1121                 if (vma == NULL)
1122                         pfn = get_fault_pfn();
1123                 else if ((vma->vm_flags & VM_PFNMAP)) {
1124                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1125                                 vma->vm_pgoff;
1126                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1127                 } else {
1128                         if (async && (vma->vm_flags & VM_WRITE))
1129                                 *async = true;
1130                         pfn = get_fault_pfn();
1131                 }
1132                 up_read(&current->mm->mmap_sem);
1133         } else
1134                 pfn = page_to_pfn(page[0]);
1135
1136         return pfn;
1137 }
1138
1139 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1140 {
1141         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1142 }
1143 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1144
1145 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1146                           bool write_fault, bool *writable)
1147 {
1148         unsigned long addr;
1149
1150         if (async)
1151                 *async = false;
1152
1153         addr = gfn_to_hva(kvm, gfn);
1154         if (kvm_is_error_hva(addr)) {
1155                 get_page(bad_page);
1156                 return page_to_pfn(bad_page);
1157         }
1158
1159         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1160 }
1161
1162 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1163 {
1164         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1165 }
1166 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1167
1168 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1169                        bool write_fault, bool *writable)
1170 {
1171         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1172 }
1173 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1174
1175 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1176 {
1177         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1178 }
1179 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1180
1181 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1182                       bool *writable)
1183 {
1184         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1185 }
1186 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1187
1188 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1189                          struct kvm_memory_slot *slot, gfn_t gfn)
1190 {
1191         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1192         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1193 }
1194
1195 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1196                                                                   int nr_pages)
1197 {
1198         unsigned long addr;
1199         gfn_t entry;
1200
1201         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1202         if (kvm_is_error_hva(addr))
1203                 return -1;
1204
1205         if (entry < nr_pages)
1206                 return 0;
1207
1208         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1209 }
1210 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1211
1212 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1213 {
1214         pfn_t pfn;
1215
1216         pfn = gfn_to_pfn(kvm, gfn);
1217         if (!kvm_is_mmio_pfn(pfn))
1218                 return pfn_to_page(pfn);
1219
1220         WARN_ON(kvm_is_mmio_pfn(pfn));
1221
1222         get_page(bad_page);
1223         return bad_page;
1224 }
1225
1226 EXPORT_SYMBOL_GPL(gfn_to_page);
1227
1228 void kvm_release_page_clean(struct page *page)
1229 {
1230         kvm_release_pfn_clean(page_to_pfn(page));
1231 }
1232 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1233
1234 void kvm_release_pfn_clean(pfn_t pfn)
1235 {
1236         if (!kvm_is_mmio_pfn(pfn))
1237                 put_page(pfn_to_page(pfn));
1238 }
1239 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1240
1241 void kvm_release_page_dirty(struct page *page)
1242 {
1243         kvm_release_pfn_dirty(page_to_pfn(page));
1244 }
1245 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1246
1247 void kvm_release_pfn_dirty(pfn_t pfn)
1248 {
1249         kvm_set_pfn_dirty(pfn);
1250         kvm_release_pfn_clean(pfn);
1251 }
1252 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1253
1254 void kvm_set_page_dirty(struct page *page)
1255 {
1256         kvm_set_pfn_dirty(page_to_pfn(page));
1257 }
1258 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1259
1260 void kvm_set_pfn_dirty(pfn_t pfn)
1261 {
1262         if (!kvm_is_mmio_pfn(pfn)) {
1263                 struct page *page = pfn_to_page(pfn);
1264                 if (!PageReserved(page))
1265                         SetPageDirty(page);
1266         }
1267 }
1268 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1269
1270 void kvm_set_pfn_accessed(pfn_t pfn)
1271 {
1272         if (!kvm_is_mmio_pfn(pfn))
1273                 mark_page_accessed(pfn_to_page(pfn));
1274 }
1275 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1276
1277 void kvm_get_pfn(pfn_t pfn)
1278 {
1279         if (!kvm_is_mmio_pfn(pfn))
1280                 get_page(pfn_to_page(pfn));
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1283
1284 static int next_segment(unsigned long len, int offset)
1285 {
1286         if (len > PAGE_SIZE - offset)
1287                 return PAGE_SIZE - offset;
1288         else
1289                 return len;
1290 }
1291
1292 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1293                         int len)
1294 {
1295         int r;
1296         unsigned long addr;
1297
1298         addr = gfn_to_hva(kvm, gfn);
1299         if (kvm_is_error_hva(addr))
1300                 return -EFAULT;
1301         r = __copy_from_user(data, (void __user *)addr + offset, len);
1302         if (r)
1303                 return -EFAULT;
1304         return 0;
1305 }
1306 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1307
1308 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1309 {
1310         gfn_t gfn = gpa >> PAGE_SHIFT;
1311         int seg;
1312         int offset = offset_in_page(gpa);
1313         int ret;
1314
1315         while ((seg = next_segment(len, offset)) != 0) {
1316                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1317                 if (ret < 0)
1318                         return ret;
1319                 offset = 0;
1320                 len -= seg;
1321                 data += seg;
1322                 ++gfn;
1323         }
1324         return 0;
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_read_guest);
1327
1328 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1329                           unsigned long len)
1330 {
1331         int r;
1332         unsigned long addr;
1333         gfn_t gfn = gpa >> PAGE_SHIFT;
1334         int offset = offset_in_page(gpa);
1335
1336         addr = gfn_to_hva(kvm, gfn);
1337         if (kvm_is_error_hva(addr))
1338                 return -EFAULT;
1339         pagefault_disable();
1340         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1341         pagefault_enable();
1342         if (r)
1343                 return -EFAULT;
1344         return 0;
1345 }
1346 EXPORT_SYMBOL(kvm_read_guest_atomic);
1347
1348 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1349                          int offset, int len)
1350 {
1351         int r;
1352         unsigned long addr;
1353
1354         addr = gfn_to_hva(kvm, gfn);
1355         if (kvm_is_error_hva(addr))
1356                 return -EFAULT;
1357         r = __copy_to_user((void __user *)addr + offset, data, len);
1358         if (r)
1359                 return -EFAULT;
1360         mark_page_dirty(kvm, gfn);
1361         return 0;
1362 }
1363 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1364
1365 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1366                     unsigned long len)
1367 {
1368         gfn_t gfn = gpa >> PAGE_SHIFT;
1369         int seg;
1370         int offset = offset_in_page(gpa);
1371         int ret;
1372
1373         while ((seg = next_segment(len, offset)) != 0) {
1374                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1375                 if (ret < 0)
1376                         return ret;
1377                 offset = 0;
1378                 len -= seg;
1379                 data += seg;
1380                 ++gfn;
1381         }
1382         return 0;
1383 }
1384
1385 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1386                               gpa_t gpa)
1387 {
1388         struct kvm_memslots *slots = kvm_memslots(kvm);
1389         int offset = offset_in_page(gpa);
1390         gfn_t gfn = gpa >> PAGE_SHIFT;
1391
1392         ghc->gpa = gpa;
1393         ghc->generation = slots->generation;
1394         ghc->memslot = gfn_to_memslot(kvm, gfn);
1395         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1396         if (!kvm_is_error_hva(ghc->hva))
1397                 ghc->hva += offset;
1398         else
1399                 return -EFAULT;
1400
1401         return 0;
1402 }
1403 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1404
1405 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1406                            void *data, unsigned long len)
1407 {
1408         struct kvm_memslots *slots = kvm_memslots(kvm);
1409         int r;
1410
1411         if (slots->generation != ghc->generation)
1412                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1413
1414         if (kvm_is_error_hva(ghc->hva))
1415                 return -EFAULT;
1416
1417         r = __copy_to_user((void __user *)ghc->hva, data, len);
1418         if (r)
1419                 return -EFAULT;
1420         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1421
1422         return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1425
1426 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1427                            void *data, unsigned long len)
1428 {
1429         struct kvm_memslots *slots = kvm_memslots(kvm);
1430         int r;
1431
1432         if (slots->generation != ghc->generation)
1433                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1434
1435         if (kvm_is_error_hva(ghc->hva))
1436                 return -EFAULT;
1437
1438         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1439         if (r)
1440                 return -EFAULT;
1441
1442         return 0;
1443 }
1444 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1445
1446 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1447 {
1448         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1449                                     offset, len);
1450 }
1451 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1452
1453 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1454 {
1455         gfn_t gfn = gpa >> PAGE_SHIFT;
1456         int seg;
1457         int offset = offset_in_page(gpa);
1458         int ret;
1459
1460         while ((seg = next_segment(len, offset)) != 0) {
1461                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1462                 if (ret < 0)
1463                         return ret;
1464                 offset = 0;
1465                 len -= seg;
1466                 ++gfn;
1467         }
1468         return 0;
1469 }
1470 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1471
1472 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1473                              gfn_t gfn)
1474 {
1475         if (memslot && memslot->dirty_bitmap) {
1476                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1477
1478                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1479                         memslot->nr_dirty_pages++;
1480         }
1481 }
1482
1483 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1484 {
1485         struct kvm_memory_slot *memslot;
1486
1487         memslot = gfn_to_memslot(kvm, gfn);
1488         mark_page_dirty_in_slot(kvm, memslot, gfn);
1489 }
1490
1491 /*
1492  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1493  */
1494 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1495 {
1496         DEFINE_WAIT(wait);
1497
1498         for (;;) {
1499                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1500
1501                 if (kvm_arch_vcpu_runnable(vcpu)) {
1502                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1503                         break;
1504                 }
1505                 if (kvm_cpu_has_pending_timer(vcpu))
1506                         break;
1507                 if (signal_pending(current))
1508                         break;
1509
1510                 schedule();
1511         }
1512
1513         finish_wait(&vcpu->wq, &wait);
1514 }
1515
1516 void kvm_resched(struct kvm_vcpu *vcpu)
1517 {
1518         if (!need_resched())
1519                 return;
1520         cond_resched();
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_resched);
1523
1524 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1525 {
1526         struct kvm *kvm = me->kvm;
1527         struct kvm_vcpu *vcpu;
1528         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1529         int yielded = 0;
1530         int pass;
1531         int i;
1532
1533         /*
1534          * We boost the priority of a VCPU that is runnable but not
1535          * currently running, because it got preempted by something
1536          * else and called schedule in __vcpu_run.  Hopefully that
1537          * VCPU is holding the lock that we need and will release it.
1538          * We approximate round-robin by starting at the last boosted VCPU.
1539          */
1540         for (pass = 0; pass < 2 && !yielded; pass++) {
1541                 kvm_for_each_vcpu(i, vcpu, kvm) {
1542                         struct task_struct *task = NULL;
1543                         struct pid *pid;
1544                         if (!pass && i < last_boosted_vcpu) {
1545                                 i = last_boosted_vcpu;
1546                                 continue;
1547                         } else if (pass && i > last_boosted_vcpu)
1548                                 break;
1549                         if (vcpu == me)
1550                                 continue;
1551                         if (waitqueue_active(&vcpu->wq))
1552                                 continue;
1553                         rcu_read_lock();
1554                         pid = rcu_dereference(vcpu->pid);
1555                         if (pid)
1556                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1557                         rcu_read_unlock();
1558                         if (!task)
1559                                 continue;
1560                         if (task->flags & PF_VCPU) {
1561                                 put_task_struct(task);
1562                                 continue;
1563                         }
1564                         if (yield_to(task, 1)) {
1565                                 put_task_struct(task);
1566                                 kvm->last_boosted_vcpu = i;
1567                                 yielded = 1;
1568                                 break;
1569                         }
1570                         put_task_struct(task);
1571                 }
1572         }
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1575
1576 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1577 {
1578         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1579         struct page *page;
1580
1581         if (vmf->pgoff == 0)
1582                 page = virt_to_page(vcpu->run);
1583 #ifdef CONFIG_X86
1584         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1585                 page = virt_to_page(vcpu->arch.pio_data);
1586 #endif
1587 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1588         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1589                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1590 #endif
1591         else
1592                 return kvm_arch_vcpu_fault(vcpu, vmf);
1593         get_page(page);
1594         vmf->page = page;
1595         return 0;
1596 }
1597
1598 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1599         .fault = kvm_vcpu_fault,
1600 };
1601
1602 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1603 {
1604         vma->vm_ops = &kvm_vcpu_vm_ops;
1605         return 0;
1606 }
1607
1608 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1609 {
1610         struct kvm_vcpu *vcpu = filp->private_data;
1611
1612         kvm_put_kvm(vcpu->kvm);
1613         return 0;
1614 }
1615
1616 static struct file_operations kvm_vcpu_fops = {
1617         .release        = kvm_vcpu_release,
1618         .unlocked_ioctl = kvm_vcpu_ioctl,
1619 #ifdef CONFIG_COMPAT
1620         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1621 #endif
1622         .mmap           = kvm_vcpu_mmap,
1623         .llseek         = noop_llseek,
1624 };
1625
1626 /*
1627  * Allocates an inode for the vcpu.
1628  */
1629 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1630 {
1631         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1632 }
1633
1634 /*
1635  * Creates some virtual cpus.  Good luck creating more than one.
1636  */
1637 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1638 {
1639         int r;
1640         struct kvm_vcpu *vcpu, *v;
1641
1642         vcpu = kvm_arch_vcpu_create(kvm, id);
1643         if (IS_ERR(vcpu))
1644                 return PTR_ERR(vcpu);
1645
1646         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1647
1648         r = kvm_arch_vcpu_setup(vcpu);
1649         if (r)
1650                 goto vcpu_destroy;
1651
1652         mutex_lock(&kvm->lock);
1653         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1654                 r = -EINVAL;
1655                 goto unlock_vcpu_destroy;
1656         }
1657
1658         kvm_for_each_vcpu(r, v, kvm)
1659                 if (v->vcpu_id == id) {
1660                         r = -EEXIST;
1661                         goto unlock_vcpu_destroy;
1662                 }
1663
1664         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1665
1666         /* Now it's all set up, let userspace reach it */
1667         kvm_get_kvm(kvm);
1668         r = create_vcpu_fd(vcpu);
1669         if (r < 0) {
1670                 kvm_put_kvm(kvm);
1671                 goto unlock_vcpu_destroy;
1672         }
1673
1674         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1675         smp_wmb();
1676         atomic_inc(&kvm->online_vcpus);
1677
1678         mutex_unlock(&kvm->lock);
1679         return r;
1680
1681 unlock_vcpu_destroy:
1682         mutex_unlock(&kvm->lock);
1683 vcpu_destroy:
1684         kvm_arch_vcpu_destroy(vcpu);
1685         return r;
1686 }
1687
1688 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1689 {
1690         if (sigset) {
1691                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1692                 vcpu->sigset_active = 1;
1693                 vcpu->sigset = *sigset;
1694         } else
1695                 vcpu->sigset_active = 0;
1696         return 0;
1697 }
1698
1699 static long kvm_vcpu_ioctl(struct file *filp,
1700                            unsigned int ioctl, unsigned long arg)
1701 {
1702         struct kvm_vcpu *vcpu = filp->private_data;
1703         void __user *argp = (void __user *)arg;
1704         int r;
1705         struct kvm_fpu *fpu = NULL;
1706         struct kvm_sregs *kvm_sregs = NULL;
1707
1708         if (vcpu->kvm->mm != current->mm)
1709                 return -EIO;
1710
1711 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1712         /*
1713          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1714          * so vcpu_load() would break it.
1715          */
1716         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1717                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1718 #endif
1719
1720
1721         vcpu_load(vcpu);
1722         switch (ioctl) {
1723         case KVM_RUN:
1724                 r = -EINVAL;
1725                 if (arg)
1726                         goto out;
1727                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1728                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1729                 break;
1730         case KVM_GET_REGS: {
1731                 struct kvm_regs *kvm_regs;
1732
1733                 r = -ENOMEM;
1734                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1735                 if (!kvm_regs)
1736                         goto out;
1737                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1738                 if (r)
1739                         goto out_free1;
1740                 r = -EFAULT;
1741                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1742                         goto out_free1;
1743                 r = 0;
1744 out_free1:
1745                 kfree(kvm_regs);
1746                 break;
1747         }
1748         case KVM_SET_REGS: {
1749                 struct kvm_regs *kvm_regs;
1750
1751                 r = -ENOMEM;
1752                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1753                 if (IS_ERR(kvm_regs)) {
1754                         r = PTR_ERR(kvm_regs);
1755                         goto out;
1756                 }
1757                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1758                 if (r)
1759                         goto out_free2;
1760                 r = 0;
1761 out_free2:
1762                 kfree(kvm_regs);
1763                 break;
1764         }
1765         case KVM_GET_SREGS: {
1766                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1767                 r = -ENOMEM;
1768                 if (!kvm_sregs)
1769                         goto out;
1770                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1771                 if (r)
1772                         goto out;
1773                 r = -EFAULT;
1774                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1775                         goto out;
1776                 r = 0;
1777                 break;
1778         }
1779         case KVM_SET_SREGS: {
1780                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1781                 if (IS_ERR(kvm_sregs)) {
1782                         r = PTR_ERR(kvm_sregs);
1783                         goto out;
1784                 }
1785                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1786                 if (r)
1787                         goto out;
1788                 r = 0;
1789                 break;
1790         }
1791         case KVM_GET_MP_STATE: {
1792                 struct kvm_mp_state mp_state;
1793
1794                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1795                 if (r)
1796                         goto out;
1797                 r = -EFAULT;
1798                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1799                         goto out;
1800                 r = 0;
1801                 break;
1802         }
1803         case KVM_SET_MP_STATE: {
1804                 struct kvm_mp_state mp_state;
1805
1806                 r = -EFAULT;
1807                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1808                         goto out;
1809                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1810                 if (r)
1811                         goto out;
1812                 r = 0;
1813                 break;
1814         }
1815         case KVM_TRANSLATE: {
1816                 struct kvm_translation tr;
1817
1818                 r = -EFAULT;
1819                 if (copy_from_user(&tr, argp, sizeof tr))
1820                         goto out;
1821                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1822                 if (r)
1823                         goto out;
1824                 r = -EFAULT;
1825                 if (copy_to_user(argp, &tr, sizeof tr))
1826                         goto out;
1827                 r = 0;
1828                 break;
1829         }
1830         case KVM_SET_GUEST_DEBUG: {
1831                 struct kvm_guest_debug dbg;
1832
1833                 r = -EFAULT;
1834                 if (copy_from_user(&dbg, argp, sizeof dbg))
1835                         goto out;
1836                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1837                 if (r)
1838                         goto out;
1839                 r = 0;
1840                 break;
1841         }
1842         case KVM_SET_SIGNAL_MASK: {
1843                 struct kvm_signal_mask __user *sigmask_arg = argp;
1844                 struct kvm_signal_mask kvm_sigmask;
1845                 sigset_t sigset, *p;
1846
1847                 p = NULL;
1848                 if (argp) {
1849                         r = -EFAULT;
1850                         if (copy_from_user(&kvm_sigmask, argp,
1851                                            sizeof kvm_sigmask))
1852                                 goto out;
1853                         r = -EINVAL;
1854                         if (kvm_sigmask.len != sizeof sigset)
1855                                 goto out;
1856                         r = -EFAULT;
1857                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1858                                            sizeof sigset))
1859                                 goto out;
1860                         p = &sigset;
1861                 }
1862                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1863                 break;
1864         }
1865         case KVM_GET_FPU: {
1866                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1867                 r = -ENOMEM;
1868                 if (!fpu)
1869                         goto out;
1870                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1871                 if (r)
1872                         goto out;
1873                 r = -EFAULT;
1874                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1875                         goto out;
1876                 r = 0;
1877                 break;
1878         }
1879         case KVM_SET_FPU: {
1880                 fpu = memdup_user(argp, sizeof(*fpu));
1881                 if (IS_ERR(fpu)) {
1882                         r = PTR_ERR(fpu);
1883                         goto out;
1884                 }
1885                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1886                 if (r)
1887                         goto out;
1888                 r = 0;
1889                 break;
1890         }
1891         default:
1892                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1893         }
1894 out:
1895         vcpu_put(vcpu);
1896         kfree(fpu);
1897         kfree(kvm_sregs);
1898         return r;
1899 }
1900
1901 #ifdef CONFIG_COMPAT
1902 static long kvm_vcpu_compat_ioctl(struct file *filp,
1903                                   unsigned int ioctl, unsigned long arg)
1904 {
1905         struct kvm_vcpu *vcpu = filp->private_data;
1906         void __user *argp = compat_ptr(arg);
1907         int r;
1908
1909         if (vcpu->kvm->mm != current->mm)
1910                 return -EIO;
1911
1912         switch (ioctl) {
1913         case KVM_SET_SIGNAL_MASK: {
1914                 struct kvm_signal_mask __user *sigmask_arg = argp;
1915                 struct kvm_signal_mask kvm_sigmask;
1916                 compat_sigset_t csigset;
1917                 sigset_t sigset;
1918
1919                 if (argp) {
1920                         r = -EFAULT;
1921                         if (copy_from_user(&kvm_sigmask, argp,
1922                                            sizeof kvm_sigmask))
1923                                 goto out;
1924                         r = -EINVAL;
1925                         if (kvm_sigmask.len != sizeof csigset)
1926                                 goto out;
1927                         r = -EFAULT;
1928                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1929                                            sizeof csigset))
1930                                 goto out;
1931                 }
1932                 sigset_from_compat(&sigset, &csigset);
1933                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1934                 break;
1935         }
1936         default:
1937                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1938         }
1939
1940 out:
1941         return r;
1942 }
1943 #endif
1944
1945 static long kvm_vm_ioctl(struct file *filp,
1946                            unsigned int ioctl, unsigned long arg)
1947 {
1948         struct kvm *kvm = filp->private_data;
1949         void __user *argp = (void __user *)arg;
1950         int r;
1951
1952         if (kvm->mm != current->mm)
1953                 return -EIO;
1954         switch (ioctl) {
1955         case KVM_CREATE_VCPU:
1956                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1957                 if (r < 0)
1958                         goto out;
1959                 break;
1960         case KVM_SET_USER_MEMORY_REGION: {
1961                 struct kvm_userspace_memory_region kvm_userspace_mem;
1962
1963                 r = -EFAULT;
1964                 if (copy_from_user(&kvm_userspace_mem, argp,
1965                                                 sizeof kvm_userspace_mem))
1966                         goto out;
1967
1968                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1969                 if (r)
1970                         goto out;
1971                 break;
1972         }
1973         case KVM_GET_DIRTY_LOG: {
1974                 struct kvm_dirty_log log;
1975
1976                 r = -EFAULT;
1977                 if (copy_from_user(&log, argp, sizeof log))
1978                         goto out;
1979                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1980                 if (r)
1981                         goto out;
1982                 break;
1983         }
1984 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1985         case KVM_REGISTER_COALESCED_MMIO: {
1986                 struct kvm_coalesced_mmio_zone zone;
1987                 r = -EFAULT;
1988                 if (copy_from_user(&zone, argp, sizeof zone))
1989                         goto out;
1990                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1991                 if (r)
1992                         goto out;
1993                 r = 0;
1994                 break;
1995         }
1996         case KVM_UNREGISTER_COALESCED_MMIO: {
1997                 struct kvm_coalesced_mmio_zone zone;
1998                 r = -EFAULT;
1999                 if (copy_from_user(&zone, argp, sizeof zone))
2000                         goto out;
2001                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2002                 if (r)
2003                         goto out;
2004                 r = 0;
2005                 break;
2006         }
2007 #endif
2008         case KVM_IRQFD: {
2009                 struct kvm_irqfd data;
2010
2011                 r = -EFAULT;
2012                 if (copy_from_user(&data, argp, sizeof data))
2013                         goto out;
2014                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2015                 break;
2016         }
2017         case KVM_IOEVENTFD: {
2018                 struct kvm_ioeventfd data;
2019
2020                 r = -EFAULT;
2021                 if (copy_from_user(&data, argp, sizeof data))
2022                         goto out;
2023                 r = kvm_ioeventfd(kvm, &data);
2024                 break;
2025         }
2026 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2027         case KVM_SET_BOOT_CPU_ID:
2028                 r = 0;
2029                 mutex_lock(&kvm->lock);
2030                 if (atomic_read(&kvm->online_vcpus) != 0)
2031                         r = -EBUSY;
2032                 else
2033                         kvm->bsp_vcpu_id = arg;
2034                 mutex_unlock(&kvm->lock);
2035                 break;
2036 #endif
2037         default:
2038                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2039                 if (r == -ENOTTY)
2040                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2041         }
2042 out:
2043         return r;
2044 }
2045
2046 #ifdef CONFIG_COMPAT
2047 struct compat_kvm_dirty_log {
2048         __u32 slot;
2049         __u32 padding1;
2050         union {
2051                 compat_uptr_t dirty_bitmap; /* one bit per page */
2052                 __u64 padding2;
2053         };
2054 };
2055
2056 static long kvm_vm_compat_ioctl(struct file *filp,
2057                            unsigned int ioctl, unsigned long arg)
2058 {
2059         struct kvm *kvm = filp->private_data;
2060         int r;
2061
2062         if (kvm->mm != current->mm)
2063                 return -EIO;
2064         switch (ioctl) {
2065         case KVM_GET_DIRTY_LOG: {
2066                 struct compat_kvm_dirty_log compat_log;
2067                 struct kvm_dirty_log log;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&compat_log, (void __user *)arg,
2071                                    sizeof(compat_log)))
2072                         goto out;
2073                 log.slot         = compat_log.slot;
2074                 log.padding1     = compat_log.padding1;
2075                 log.padding2     = compat_log.padding2;
2076                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2077
2078                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2079                 if (r)
2080                         goto out;
2081                 break;
2082         }
2083         default:
2084                 r = kvm_vm_ioctl(filp, ioctl, arg);
2085         }
2086
2087 out:
2088         return r;
2089 }
2090 #endif
2091
2092 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2093 {
2094         struct page *page[1];
2095         unsigned long addr;
2096         int npages;
2097         gfn_t gfn = vmf->pgoff;
2098         struct kvm *kvm = vma->vm_file->private_data;
2099
2100         addr = gfn_to_hva(kvm, gfn);
2101         if (kvm_is_error_hva(addr))
2102                 return VM_FAULT_SIGBUS;
2103
2104         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2105                                 NULL);
2106         if (unlikely(npages != 1))
2107                 return VM_FAULT_SIGBUS;
2108
2109         vmf->page = page[0];
2110         return 0;
2111 }
2112
2113 static const struct vm_operations_struct kvm_vm_vm_ops = {
2114         .fault = kvm_vm_fault,
2115 };
2116
2117 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2118 {
2119         vma->vm_ops = &kvm_vm_vm_ops;
2120         return 0;
2121 }
2122
2123 static struct file_operations kvm_vm_fops = {
2124         .release        = kvm_vm_release,
2125         .unlocked_ioctl = kvm_vm_ioctl,
2126 #ifdef CONFIG_COMPAT
2127         .compat_ioctl   = kvm_vm_compat_ioctl,
2128 #endif
2129         .mmap           = kvm_vm_mmap,
2130         .llseek         = noop_llseek,
2131 };
2132
2133 static int kvm_dev_ioctl_create_vm(unsigned long type)
2134 {
2135         int r;
2136         struct kvm *kvm;
2137
2138         kvm = kvm_create_vm(type);
2139         if (IS_ERR(kvm))
2140                 return PTR_ERR(kvm);
2141 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2142         r = kvm_coalesced_mmio_init(kvm);
2143         if (r < 0) {
2144                 kvm_put_kvm(kvm);
2145                 return r;
2146         }
2147 #endif
2148         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2149         if (r < 0)
2150                 kvm_put_kvm(kvm);
2151
2152         return r;
2153 }
2154
2155 static long kvm_dev_ioctl_check_extension_generic(long arg)
2156 {
2157         switch (arg) {
2158         case KVM_CAP_USER_MEMORY:
2159         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2160         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2161 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2162         case KVM_CAP_SET_BOOT_CPU_ID:
2163 #endif
2164         case KVM_CAP_INTERNAL_ERROR_DATA:
2165                 return 1;
2166 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2167         case KVM_CAP_IRQ_ROUTING:
2168                 return KVM_MAX_IRQ_ROUTES;
2169 #endif
2170         default:
2171                 break;
2172         }
2173         return kvm_dev_ioctl_check_extension(arg);
2174 }
2175
2176 static long kvm_dev_ioctl(struct file *filp,
2177                           unsigned int ioctl, unsigned long arg)
2178 {
2179         long r = -EINVAL;
2180
2181         switch (ioctl) {
2182         case KVM_GET_API_VERSION:
2183                 r = -EINVAL;
2184                 if (arg)
2185                         goto out;
2186                 r = KVM_API_VERSION;
2187                 break;
2188         case KVM_CREATE_VM:
2189                 r = kvm_dev_ioctl_create_vm(arg);
2190                 break;
2191         case KVM_CHECK_EXTENSION:
2192                 r = kvm_dev_ioctl_check_extension_generic(arg);
2193                 break;
2194         case KVM_GET_VCPU_MMAP_SIZE:
2195                 r = -EINVAL;
2196                 if (arg)
2197                         goto out;
2198                 r = PAGE_SIZE;     /* struct kvm_run */
2199 #ifdef CONFIG_X86
2200                 r += PAGE_SIZE;    /* pio data page */
2201 #endif
2202 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2203                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2204 #endif
2205                 break;
2206         case KVM_TRACE_ENABLE:
2207         case KVM_TRACE_PAUSE:
2208         case KVM_TRACE_DISABLE:
2209                 r = -EOPNOTSUPP;
2210                 break;
2211         default:
2212                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2213         }
2214 out:
2215         return r;
2216 }
2217
2218 static struct file_operations kvm_chardev_ops = {
2219         .unlocked_ioctl = kvm_dev_ioctl,
2220         .compat_ioctl   = kvm_dev_ioctl,
2221         .llseek         = noop_llseek,
2222 };
2223
2224 static struct miscdevice kvm_dev = {
2225         KVM_MINOR,
2226         "kvm",
2227         &kvm_chardev_ops,
2228 };
2229
2230 static void hardware_enable_nolock(void *junk)
2231 {
2232         int cpu = raw_smp_processor_id();
2233         int r;
2234
2235         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2236                 return;
2237
2238         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2239
2240         r = kvm_arch_hardware_enable(NULL);
2241
2242         if (r) {
2243                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2244                 atomic_inc(&hardware_enable_failed);
2245                 printk(KERN_INFO "kvm: enabling virtualization on "
2246                                  "CPU%d failed\n", cpu);
2247         }
2248 }
2249
2250 static void hardware_enable(void *junk)
2251 {
2252         raw_spin_lock(&kvm_lock);
2253         hardware_enable_nolock(junk);
2254         raw_spin_unlock(&kvm_lock);
2255 }
2256
2257 static void hardware_disable_nolock(void *junk)
2258 {
2259         int cpu = raw_smp_processor_id();
2260
2261         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2262                 return;
2263         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2264         kvm_arch_hardware_disable(NULL);
2265 }
2266
2267 static void hardware_disable(void *junk)
2268 {
2269         raw_spin_lock(&kvm_lock);
2270         hardware_disable_nolock(junk);
2271         raw_spin_unlock(&kvm_lock);
2272 }
2273
2274 static void hardware_disable_all_nolock(void)
2275 {
2276         BUG_ON(!kvm_usage_count);
2277
2278         kvm_usage_count--;
2279         if (!kvm_usage_count)
2280                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2281 }
2282
2283 static void hardware_disable_all(void)
2284 {
2285         raw_spin_lock(&kvm_lock);
2286         hardware_disable_all_nolock();
2287         raw_spin_unlock(&kvm_lock);
2288 }
2289
2290 static int hardware_enable_all(void)
2291 {
2292         int r = 0;
2293
2294         raw_spin_lock(&kvm_lock);
2295
2296         kvm_usage_count++;
2297         if (kvm_usage_count == 1) {
2298                 atomic_set(&hardware_enable_failed, 0);
2299                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2300
2301                 if (atomic_read(&hardware_enable_failed)) {
2302                         hardware_disable_all_nolock();
2303                         r = -EBUSY;
2304                 }
2305         }
2306
2307         raw_spin_unlock(&kvm_lock);
2308
2309         return r;
2310 }
2311
2312 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2313                            void *v)
2314 {
2315         int cpu = (long)v;
2316
2317         if (!kvm_usage_count)
2318                 return NOTIFY_OK;
2319
2320         val &= ~CPU_TASKS_FROZEN;
2321         switch (val) {
2322         case CPU_DYING:
2323                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2324                        cpu);
2325                 hardware_disable(NULL);
2326                 break;
2327         case CPU_STARTING:
2328                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2329                        cpu);
2330                 hardware_enable(NULL);
2331                 break;
2332         }
2333         return NOTIFY_OK;
2334 }
2335
2336
2337 asmlinkage void kvm_spurious_fault(void)
2338 {
2339         /* Fault while not rebooting.  We want the trace. */
2340         BUG();
2341 }
2342 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2343
2344 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2345                       void *v)
2346 {
2347         /*
2348          * Some (well, at least mine) BIOSes hang on reboot if
2349          * in vmx root mode.
2350          *
2351          * And Intel TXT required VMX off for all cpu when system shutdown.
2352          */
2353         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2354         kvm_rebooting = true;
2355         on_each_cpu(hardware_disable_nolock, NULL, 1);
2356         return NOTIFY_OK;
2357 }
2358
2359 static struct notifier_block kvm_reboot_notifier = {
2360         .notifier_call = kvm_reboot,
2361         .priority = 0,
2362 };
2363
2364 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2365 {
2366         int i;
2367
2368         for (i = 0; i < bus->dev_count; i++) {
2369                 struct kvm_io_device *pos = bus->range[i].dev;
2370
2371                 kvm_iodevice_destructor(pos);
2372         }
2373         kfree(bus);
2374 }
2375
2376 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2377 {
2378         const struct kvm_io_range *r1 = p1;
2379         const struct kvm_io_range *r2 = p2;
2380
2381         if (r1->addr < r2->addr)
2382                 return -1;
2383         if (r1->addr + r1->len > r2->addr + r2->len)
2384                 return 1;
2385         return 0;
2386 }
2387
2388 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2389                           gpa_t addr, int len)
2390 {
2391         if (bus->dev_count == NR_IOBUS_DEVS)
2392                 return -ENOSPC;
2393
2394         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2395                 .addr = addr,
2396                 .len = len,
2397                 .dev = dev,
2398         };
2399
2400         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2401                 kvm_io_bus_sort_cmp, NULL);
2402
2403         return 0;
2404 }
2405
2406 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2407                              gpa_t addr, int len)
2408 {
2409         struct kvm_io_range *range, key;
2410         int off;
2411
2412         key = (struct kvm_io_range) {
2413                 .addr = addr,
2414                 .len = len,
2415         };
2416
2417         range = bsearch(&key, bus->range, bus->dev_count,
2418                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2419         if (range == NULL)
2420                 return -ENOENT;
2421
2422         off = range - bus->range;
2423
2424         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2425                 off--;
2426
2427         return off;
2428 }
2429
2430 /* kvm_io_bus_write - called under kvm->slots_lock */
2431 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2432                      int len, const void *val)
2433 {
2434         int idx;
2435         struct kvm_io_bus *bus;
2436         struct kvm_io_range range;
2437
2438         range = (struct kvm_io_range) {
2439                 .addr = addr,
2440                 .len = len,
2441         };
2442
2443         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2444         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2445         if (idx < 0)
2446                 return -EOPNOTSUPP;
2447
2448         while (idx < bus->dev_count &&
2449                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2450                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2451                         return 0;
2452                 idx++;
2453         }
2454
2455         return -EOPNOTSUPP;
2456 }
2457
2458 /* kvm_io_bus_read - called under kvm->slots_lock */
2459 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2460                     int len, void *val)
2461 {
2462         int idx;
2463         struct kvm_io_bus *bus;
2464         struct kvm_io_range range;
2465
2466         range = (struct kvm_io_range) {
2467                 .addr = addr,
2468                 .len = len,
2469         };
2470
2471         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2472         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2473         if (idx < 0)
2474                 return -EOPNOTSUPP;
2475
2476         while (idx < bus->dev_count &&
2477                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2478                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2479                         return 0;
2480                 idx++;
2481         }
2482
2483         return -EOPNOTSUPP;
2484 }
2485
2486 /* Caller must hold slots_lock. */
2487 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2488                             int len, struct kvm_io_device *dev)
2489 {
2490         struct kvm_io_bus *new_bus, *bus;
2491
2492         bus = kvm->buses[bus_idx];
2493         if (bus->dev_count > NR_IOBUS_DEVS-1)
2494                 return -ENOSPC;
2495
2496         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2497         if (!new_bus)
2498                 return -ENOMEM;
2499         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2500         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2501         synchronize_srcu_expedited(&kvm->srcu);
2502         kfree(bus);
2503
2504         return 0;
2505 }
2506
2507 /* Caller must hold slots_lock. */
2508 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2509                               struct kvm_io_device *dev)
2510 {
2511         int i, r;
2512         struct kvm_io_bus *new_bus, *bus;
2513
2514         bus = kvm->buses[bus_idx];
2515
2516         new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2517         if (!new_bus)
2518                 return -ENOMEM;
2519
2520         r = -ENOENT;
2521         for (i = 0; i < new_bus->dev_count; i++)
2522                 if (new_bus->range[i].dev == dev) {
2523                         r = 0;
2524                         new_bus->dev_count--;
2525                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2526                         sort(new_bus->range, new_bus->dev_count,
2527                              sizeof(struct kvm_io_range),
2528                              kvm_io_bus_sort_cmp, NULL);
2529                         break;
2530                 }
2531
2532         if (r) {
2533                 kfree(new_bus);
2534                 return r;
2535         }
2536
2537         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2538         synchronize_srcu_expedited(&kvm->srcu);
2539         kfree(bus);
2540         return r;
2541 }
2542
2543 static struct notifier_block kvm_cpu_notifier = {
2544         .notifier_call = kvm_cpu_hotplug,
2545 };
2546
2547 static int vm_stat_get(void *_offset, u64 *val)
2548 {
2549         unsigned offset = (long)_offset;
2550         struct kvm *kvm;
2551
2552         *val = 0;
2553         raw_spin_lock(&kvm_lock);
2554         list_for_each_entry(kvm, &vm_list, vm_list)
2555                 *val += *(u32 *)((void *)kvm + offset);
2556         raw_spin_unlock(&kvm_lock);
2557         return 0;
2558 }
2559
2560 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2561
2562 static int vcpu_stat_get(void *_offset, u64 *val)
2563 {
2564         unsigned offset = (long)_offset;
2565         struct kvm *kvm;
2566         struct kvm_vcpu *vcpu;
2567         int i;
2568
2569         *val = 0;
2570         raw_spin_lock(&kvm_lock);
2571         list_for_each_entry(kvm, &vm_list, vm_list)
2572                 kvm_for_each_vcpu(i, vcpu, kvm)
2573                         *val += *(u32 *)((void *)vcpu + offset);
2574
2575         raw_spin_unlock(&kvm_lock);
2576         return 0;
2577 }
2578
2579 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2580
2581 static const struct file_operations *stat_fops[] = {
2582         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2583         [KVM_STAT_VM]   = &vm_stat_fops,
2584 };
2585
2586 static int kvm_init_debug(void)
2587 {
2588         int r = -EFAULT;
2589         struct kvm_stats_debugfs_item *p;
2590
2591         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2592         if (kvm_debugfs_dir == NULL)
2593                 goto out;
2594
2595         for (p = debugfs_entries; p->name; ++p) {
2596                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2597                                                 (void *)(long)p->offset,
2598                                                 stat_fops[p->kind]);
2599                 if (p->dentry == NULL)
2600                         goto out_dir;
2601         }
2602
2603         return 0;
2604
2605 out_dir:
2606         debugfs_remove_recursive(kvm_debugfs_dir);
2607 out:
2608         return r;
2609 }
2610
2611 static void kvm_exit_debug(void)
2612 {
2613         struct kvm_stats_debugfs_item *p;
2614
2615         for (p = debugfs_entries; p->name; ++p)
2616                 debugfs_remove(p->dentry);
2617         debugfs_remove(kvm_debugfs_dir);
2618 }
2619
2620 static int kvm_suspend(void)
2621 {
2622         if (kvm_usage_count)
2623                 hardware_disable_nolock(NULL);
2624         return 0;
2625 }
2626
2627 static void kvm_resume(void)
2628 {
2629         if (kvm_usage_count) {
2630                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2631                 hardware_enable_nolock(NULL);
2632         }
2633 }
2634
2635 static struct syscore_ops kvm_syscore_ops = {
2636         .suspend = kvm_suspend,
2637         .resume = kvm_resume,
2638 };
2639
2640 struct page *bad_page;
2641 pfn_t bad_pfn;
2642
2643 static inline
2644 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2645 {
2646         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2647 }
2648
2649 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2650 {
2651         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2652
2653         kvm_arch_vcpu_load(vcpu, cpu);
2654 }
2655
2656 static void kvm_sched_out(struct preempt_notifier *pn,
2657                           struct task_struct *next)
2658 {
2659         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2660
2661         kvm_arch_vcpu_put(vcpu);
2662 }
2663
2664 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2665                   struct module *module)
2666 {
2667         int r;
2668         int cpu;
2669
2670         r = kvm_arch_init(opaque);
2671         if (r)
2672                 goto out_fail;
2673
2674         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2675
2676         if (bad_page == NULL) {
2677                 r = -ENOMEM;
2678                 goto out;
2679         }
2680
2681         bad_pfn = page_to_pfn(bad_page);
2682
2683         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2684
2685         if (hwpoison_page == NULL) {
2686                 r = -ENOMEM;
2687                 goto out_free_0;
2688         }
2689
2690         hwpoison_pfn = page_to_pfn(hwpoison_page);
2691
2692         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2693
2694         if (fault_page == NULL) {
2695                 r = -ENOMEM;
2696                 goto out_free_0;
2697         }
2698
2699         fault_pfn = page_to_pfn(fault_page);
2700
2701         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2702                 r = -ENOMEM;
2703                 goto out_free_0;
2704         }
2705
2706         r = kvm_arch_hardware_setup();
2707         if (r < 0)
2708                 goto out_free_0a;
2709
2710         for_each_online_cpu(cpu) {
2711                 smp_call_function_single(cpu,
2712                                 kvm_arch_check_processor_compat,
2713                                 &r, 1);
2714                 if (r < 0)
2715                         goto out_free_1;
2716         }
2717
2718         r = register_cpu_notifier(&kvm_cpu_notifier);
2719         if (r)
2720                 goto out_free_2;
2721         register_reboot_notifier(&kvm_reboot_notifier);
2722
2723         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2724         if (!vcpu_align)
2725                 vcpu_align = __alignof__(struct kvm_vcpu);
2726         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2727                                            0, NULL);
2728         if (!kvm_vcpu_cache) {
2729                 r = -ENOMEM;
2730                 goto out_free_3;
2731         }
2732
2733         r = kvm_async_pf_init();
2734         if (r)
2735                 goto out_free;
2736
2737         kvm_chardev_ops.owner = module;
2738         kvm_vm_fops.owner = module;
2739         kvm_vcpu_fops.owner = module;
2740
2741         r = misc_register(&kvm_dev);
2742         if (r) {
2743                 printk(KERN_ERR "kvm: misc device register failed\n");
2744                 goto out_unreg;
2745         }
2746
2747         register_syscore_ops(&kvm_syscore_ops);
2748
2749         kvm_preempt_ops.sched_in = kvm_sched_in;
2750         kvm_preempt_ops.sched_out = kvm_sched_out;
2751
2752         r = kvm_init_debug();
2753         if (r) {
2754                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2755                 goto out_undebugfs;
2756         }
2757
2758         return 0;
2759
2760 out_undebugfs:
2761         unregister_syscore_ops(&kvm_syscore_ops);
2762 out_unreg:
2763         kvm_async_pf_deinit();
2764 out_free:
2765         kmem_cache_destroy(kvm_vcpu_cache);
2766 out_free_3:
2767         unregister_reboot_notifier(&kvm_reboot_notifier);
2768         unregister_cpu_notifier(&kvm_cpu_notifier);
2769 out_free_2:
2770 out_free_1:
2771         kvm_arch_hardware_unsetup();
2772 out_free_0a:
2773         free_cpumask_var(cpus_hardware_enabled);
2774 out_free_0:
2775         if (fault_page)
2776                 __free_page(fault_page);
2777         if (hwpoison_page)
2778                 __free_page(hwpoison_page);
2779         __free_page(bad_page);
2780 out:
2781         kvm_arch_exit();
2782 out_fail:
2783         return r;
2784 }
2785 EXPORT_SYMBOL_GPL(kvm_init);
2786
2787 void kvm_exit(void)
2788 {
2789         kvm_exit_debug();
2790         misc_deregister(&kvm_dev);
2791         kmem_cache_destroy(kvm_vcpu_cache);
2792         kvm_async_pf_deinit();
2793         unregister_syscore_ops(&kvm_syscore_ops);
2794         unregister_reboot_notifier(&kvm_reboot_notifier);
2795         unregister_cpu_notifier(&kvm_cpu_notifier);
2796         on_each_cpu(hardware_disable_nolock, NULL, 1);
2797         kvm_arch_hardware_unsetup();
2798         kvm_arch_exit();
2799         free_cpumask_var(cpus_hardware_enabled);
2800         __free_page(hwpoison_page);
2801         __free_page(bad_page);
2802 }
2803 EXPORT_SYMBOL_GPL(kvm_exit);