a30447c5eb4af94bb78b4debe390a4bc4936aa7b
[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         int i;
539
540         if (!dont || free->rmap != dont->rmap)
541                 vfree(free->rmap);
542
543         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
544                 kvm_destroy_dirty_bitmap(free);
545
546
547         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
548                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
549                         vfree(free->lpage_info[i]);
550                         free->lpage_info[i] = NULL;
551                 }
552         }
553
554         free->npages = 0;
555         free->rmap = NULL;
556 }
557
558 void kvm_free_physmem(struct kvm *kvm)
559 {
560         struct kvm_memslots *slots = kvm->memslots;
561         struct kvm_memory_slot *memslot;
562
563         kvm_for_each_memslot(memslot, slots)
564                 kvm_free_physmem_slot(memslot, NULL);
565
566         kfree(kvm->memslots);
567 }
568
569 static void kvm_destroy_vm(struct kvm *kvm)
570 {
571         int i;
572         struct mm_struct *mm = kvm->mm;
573
574         kvm_arch_sync_events(kvm);
575         raw_spin_lock(&kvm_lock);
576         list_del(&kvm->vm_list);
577         raw_spin_unlock(&kvm_lock);
578         kvm_free_irq_routing(kvm);
579         for (i = 0; i < KVM_NR_BUSES; i++)
580                 kvm_io_bus_destroy(kvm->buses[i]);
581         kvm_coalesced_mmio_free(kvm);
582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
584 #else
585         kvm_arch_flush_shadow(kvm);
586 #endif
587         kvm_arch_destroy_vm(kvm);
588         kvm_free_physmem(kvm);
589         cleanup_srcu_struct(&kvm->srcu);
590         kvm_arch_free_vm(kvm);
591         hardware_disable_all();
592         mmdrop(mm);
593 }
594
595 void kvm_get_kvm(struct kvm *kvm)
596 {
597         atomic_inc(&kvm->users_count);
598 }
599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
600
601 void kvm_put_kvm(struct kvm *kvm)
602 {
603         if (atomic_dec_and_test(&kvm->users_count))
604                 kvm_destroy_vm(kvm);
605 }
606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
607
608
609 static int kvm_vm_release(struct inode *inode, struct file *filp)
610 {
611         struct kvm *kvm = filp->private_data;
612
613         kvm_irqfd_release(kvm);
614
615         kvm_put_kvm(kvm);
616         return 0;
617 }
618
619 /*
620  * Allocation size is twice as large as the actual dirty bitmap size.
621  * This makes it possible to do double buffering: see x86's
622  * kvm_vm_ioctl_get_dirty_log().
623  */
624 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
625 {
626 #ifndef CONFIG_S390
627         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
628
629         if (dirty_bytes > PAGE_SIZE)
630                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
631         else
632                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
633
634         if (!memslot->dirty_bitmap)
635                 return -ENOMEM;
636
637         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
638         memslot->nr_dirty_pages = 0;
639 #endif /* !CONFIG_S390 */
640         return 0;
641 }
642
643 static int cmp_memslot(const void *slot1, const void *slot2)
644 {
645         struct kvm_memory_slot *s1, *s2;
646
647         s1 = (struct kvm_memory_slot *)slot1;
648         s2 = (struct kvm_memory_slot *)slot2;
649
650         if (s1->npages < s2->npages)
651                 return 1;
652         if (s1->npages > s2->npages)
653                 return -1;
654
655         return 0;
656 }
657
658 /*
659  * Sort the memslots base on its size, so the larger slots
660  * will get better fit.
661  */
662 static void sort_memslots(struct kvm_memslots *slots)
663 {
664         int i;
665
666         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
667               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
668
669         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
670                 slots->id_to_index[slots->memslots[i].id] = i;
671 }
672
673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
674 {
675         if (new) {
676                 int id = new->id;
677                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
678                 unsigned long npages = old->npages;
679
680                 *old = *new;
681                 if (new->npages != npages)
682                         sort_memslots(slots);
683         }
684
685         slots->generation++;
686 }
687
688 #ifndef CONFIG_S390
689 static int create_lpage_info(struct kvm_memory_slot *slot, unsigned long npages)
690 {
691         int i;
692
693         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
694                 unsigned long ugfn;
695                 int lpages;
696                 int level = i + 2;
697
698                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
699                                       slot->base_gfn, level) + 1;
700
701                 slot->lpage_info[i] = vzalloc(lpages * sizeof(*slot->lpage_info[i]));
702                 if (!slot->lpage_info[i])
703                         goto out_free;
704
705                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
706                         slot->lpage_info[i][0].write_count = 1;
707                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
708                         slot->lpage_info[i][lpages - 1].write_count = 1;
709                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
710                 /*
711                  * If the gfn and userspace address are not aligned wrt each
712                  * other, or if explicitly asked to, disable large page
713                  * support for this slot
714                  */
715                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
716                     !largepages_enabled) {
717                         unsigned long j;
718
719                         for (j = 0; j < lpages; ++j)
720                                 slot->lpage_info[i][j].write_count = 1;
721                 }
722         }
723
724         return 0;
725
726 out_free:
727         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
728                 vfree(slot->lpage_info[i]);
729                 slot->lpage_info[i] = NULL;
730         }
731         return -ENOMEM;
732 }
733 #endif /* not defined CONFIG_S390 */
734
735 /*
736  * Allocate some memory and give it an address in the guest physical address
737  * space.
738  *
739  * Discontiguous memory is allowed, mostly for framebuffers.
740  *
741  * Must be called holding mmap_sem for write.
742  */
743 int __kvm_set_memory_region(struct kvm *kvm,
744                             struct kvm_userspace_memory_region *mem,
745                             int user_alloc)
746 {
747         int r;
748         gfn_t base_gfn;
749         unsigned long npages;
750         unsigned long i;
751         struct kvm_memory_slot *memslot;
752         struct kvm_memory_slot old, new;
753         struct kvm_memslots *slots, *old_memslots;
754
755         r = -EINVAL;
756         /* General sanity checks */
757         if (mem->memory_size & (PAGE_SIZE - 1))
758                 goto out;
759         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
760                 goto out;
761         /* We can read the guest memory with __xxx_user() later on. */
762         if (user_alloc &&
763             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
764              !access_ok(VERIFY_WRITE,
765                         (void __user *)(unsigned long)mem->userspace_addr,
766                         mem->memory_size)))
767                 goto out;
768         if (mem->slot >= KVM_MEM_SLOTS_NUM)
769                 goto out;
770         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
771                 goto out;
772
773         memslot = id_to_memslot(kvm->memslots, mem->slot);
774         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
775         npages = mem->memory_size >> PAGE_SHIFT;
776
777         r = -EINVAL;
778         if (npages > KVM_MEM_MAX_NR_PAGES)
779                 goto out;
780
781         if (!npages)
782                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
783
784         new = old = *memslot;
785
786         new.id = mem->slot;
787         new.base_gfn = base_gfn;
788         new.npages = npages;
789         new.flags = mem->flags;
790
791         /* Disallow changing a memory slot's size. */
792         r = -EINVAL;
793         if (npages && old.npages && npages != old.npages)
794                 goto out_free;
795
796         /* Check for overlaps */
797         r = -EEXIST;
798         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
799                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
800
801                 if (s == memslot || !s->npages)
802                         continue;
803                 if (!((base_gfn + npages <= s->base_gfn) ||
804                       (base_gfn >= s->base_gfn + s->npages)))
805                         goto out_free;
806         }
807
808         /* Free page dirty bitmap if unneeded */
809         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
810                 new.dirty_bitmap = NULL;
811
812         r = -ENOMEM;
813
814         /* Allocate if a slot is being created */
815         if (npages && !old.npages) {
816                 new.user_alloc = user_alloc;
817                 new.userspace_addr = mem->userspace_addr;
818 #ifndef CONFIG_S390
819                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
820                 if (!new.rmap)
821                         goto out_free;
822
823                 if (create_lpage_info(&new, npages))
824                         goto out_free;
825 #endif /* not defined CONFIG_S390 */
826         }
827
828         /* Allocate page dirty bitmap if needed */
829         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
830                 if (kvm_create_dirty_bitmap(&new) < 0)
831                         goto out_free;
832                 /* destroy any largepage mappings for dirty tracking */
833         }
834
835         if (!npages) {
836                 struct kvm_memory_slot *slot;
837
838                 r = -ENOMEM;
839                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
840                                 GFP_KERNEL);
841                 if (!slots)
842                         goto out_free;
843                 slot = id_to_memslot(slots, mem->slot);
844                 slot->flags |= KVM_MEMSLOT_INVALID;
845
846                 update_memslots(slots, NULL);
847
848                 old_memslots = kvm->memslots;
849                 rcu_assign_pointer(kvm->memslots, slots);
850                 synchronize_srcu_expedited(&kvm->srcu);
851                 /* From this point no new shadow pages pointing to a deleted
852                  * memslot will be created.
853                  *
854                  * validation of sp->gfn happens in:
855                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
856                  *      - kvm_is_visible_gfn (mmu_check_roots)
857                  */
858                 kvm_arch_flush_shadow(kvm);
859                 kfree(old_memslots);
860         }
861
862         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
863         if (r)
864                 goto out_free;
865
866         /* map the pages in iommu page table */
867         if (npages) {
868                 r = kvm_iommu_map_pages(kvm, &new);
869                 if (r)
870                         goto out_free;
871         }
872
873         r = -ENOMEM;
874         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
875                         GFP_KERNEL);
876         if (!slots)
877                 goto out_free;
878
879         /* actual memory is freed via old in kvm_free_physmem_slot below */
880         if (!npages) {
881                 new.rmap = NULL;
882                 new.dirty_bitmap = NULL;
883                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
884                         new.lpage_info[i] = NULL;
885         }
886
887         update_memslots(slots, &new);
888         old_memslots = kvm->memslots;
889         rcu_assign_pointer(kvm->memslots, slots);
890         synchronize_srcu_expedited(&kvm->srcu);
891
892         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
893
894         /*
895          * If the new memory slot is created, we need to clear all
896          * mmio sptes.
897          */
898         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
899                 kvm_arch_flush_shadow(kvm);
900
901         kvm_free_physmem_slot(&old, &new);
902         kfree(old_memslots);
903
904         return 0;
905
906 out_free:
907         kvm_free_physmem_slot(&new, &old);
908 out:
909         return r;
910
911 }
912 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
913
914 int kvm_set_memory_region(struct kvm *kvm,
915                           struct kvm_userspace_memory_region *mem,
916                           int user_alloc)
917 {
918         int r;
919
920         mutex_lock(&kvm->slots_lock);
921         r = __kvm_set_memory_region(kvm, mem, user_alloc);
922         mutex_unlock(&kvm->slots_lock);
923         return r;
924 }
925 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
926
927 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
928                                    struct
929                                    kvm_userspace_memory_region *mem,
930                                    int user_alloc)
931 {
932         if (mem->slot >= KVM_MEMORY_SLOTS)
933                 return -EINVAL;
934         return kvm_set_memory_region(kvm, mem, user_alloc);
935 }
936
937 int kvm_get_dirty_log(struct kvm *kvm,
938                         struct kvm_dirty_log *log, int *is_dirty)
939 {
940         struct kvm_memory_slot *memslot;
941         int r, i;
942         unsigned long n;
943         unsigned long any = 0;
944
945         r = -EINVAL;
946         if (log->slot >= KVM_MEMORY_SLOTS)
947                 goto out;
948
949         memslot = id_to_memslot(kvm->memslots, log->slot);
950         r = -ENOENT;
951         if (!memslot->dirty_bitmap)
952                 goto out;
953
954         n = kvm_dirty_bitmap_bytes(memslot);
955
956         for (i = 0; !any && i < n/sizeof(long); ++i)
957                 any = memslot->dirty_bitmap[i];
958
959         r = -EFAULT;
960         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
961                 goto out;
962
963         if (any)
964                 *is_dirty = 1;
965
966         r = 0;
967 out:
968         return r;
969 }
970
971 void kvm_disable_largepages(void)
972 {
973         largepages_enabled = false;
974 }
975 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
976
977 int is_error_page(struct page *page)
978 {
979         return page == bad_page || page == hwpoison_page || page == fault_page;
980 }
981 EXPORT_SYMBOL_GPL(is_error_page);
982
983 int is_error_pfn(pfn_t pfn)
984 {
985         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
986 }
987 EXPORT_SYMBOL_GPL(is_error_pfn);
988
989 int is_hwpoison_pfn(pfn_t pfn)
990 {
991         return pfn == hwpoison_pfn;
992 }
993 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
994
995 int is_fault_pfn(pfn_t pfn)
996 {
997         return pfn == fault_pfn;
998 }
999 EXPORT_SYMBOL_GPL(is_fault_pfn);
1000
1001 int is_noslot_pfn(pfn_t pfn)
1002 {
1003         return pfn == bad_pfn;
1004 }
1005 EXPORT_SYMBOL_GPL(is_noslot_pfn);
1006
1007 int is_invalid_pfn(pfn_t pfn)
1008 {
1009         return pfn == hwpoison_pfn || pfn == fault_pfn;
1010 }
1011 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1012
1013 static inline unsigned long bad_hva(void)
1014 {
1015         return PAGE_OFFSET;
1016 }
1017
1018 int kvm_is_error_hva(unsigned long addr)
1019 {
1020         return addr == bad_hva();
1021 }
1022 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1023
1024 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1025 {
1026         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1027 }
1028 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1029
1030 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1031 {
1032         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1033
1034         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1035               memslot->flags & KVM_MEMSLOT_INVALID)
1036                 return 0;
1037
1038         return 1;
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1041
1042 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1043 {
1044         struct vm_area_struct *vma;
1045         unsigned long addr, size;
1046
1047         size = PAGE_SIZE;
1048
1049         addr = gfn_to_hva(kvm, gfn);
1050         if (kvm_is_error_hva(addr))
1051                 return PAGE_SIZE;
1052
1053         down_read(&current->mm->mmap_sem);
1054         vma = find_vma(current->mm, addr);
1055         if (!vma)
1056                 goto out;
1057
1058         size = vma_kernel_pagesize(vma);
1059
1060 out:
1061         up_read(&current->mm->mmap_sem);
1062
1063         return size;
1064 }
1065
1066 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1067                                      gfn_t *nr_pages)
1068 {
1069         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1070                 return bad_hva();
1071
1072         if (nr_pages)
1073                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1074
1075         return gfn_to_hva_memslot(slot, gfn);
1076 }
1077
1078 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1079 {
1080         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1081 }
1082 EXPORT_SYMBOL_GPL(gfn_to_hva);
1083
1084 static pfn_t get_fault_pfn(void)
1085 {
1086         get_page(fault_page);
1087         return fault_pfn;
1088 }
1089
1090 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1091         unsigned long start, int write, struct page **page)
1092 {
1093         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1094
1095         if (write)
1096                 flags |= FOLL_WRITE;
1097
1098         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1099 }
1100
1101 static inline int check_user_page_hwpoison(unsigned long addr)
1102 {
1103         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1104
1105         rc = __get_user_pages(current, current->mm, addr, 1,
1106                               flags, NULL, NULL, NULL);
1107         return rc == -EHWPOISON;
1108 }
1109
1110 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1111                         bool *async, bool write_fault, bool *writable)
1112 {
1113         struct page *page[1];
1114         int npages = 0;
1115         pfn_t pfn;
1116
1117         /* we can do it either atomically or asynchronously, not both */
1118         BUG_ON(atomic && async);
1119
1120         BUG_ON(!write_fault && !writable);
1121
1122         if (writable)
1123                 *writable = true;
1124
1125         if (atomic || async)
1126                 npages = __get_user_pages_fast(addr, 1, 1, page);
1127
1128         if (unlikely(npages != 1) && !atomic) {
1129                 might_sleep();
1130
1131                 if (writable)
1132                         *writable = write_fault;
1133
1134                 if (async) {
1135                         down_read(&current->mm->mmap_sem);
1136                         npages = get_user_page_nowait(current, current->mm,
1137                                                      addr, write_fault, page);
1138                         up_read(&current->mm->mmap_sem);
1139                 } else
1140                         npages = get_user_pages_fast(addr, 1, write_fault,
1141                                                      page);
1142
1143                 /* map read fault as writable if possible */
1144                 if (unlikely(!write_fault) && npages == 1) {
1145                         struct page *wpage[1];
1146
1147                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1148                         if (npages == 1) {
1149                                 *writable = true;
1150                                 put_page(page[0]);
1151                                 page[0] = wpage[0];
1152                         }
1153                         npages = 1;
1154                 }
1155         }
1156
1157         if (unlikely(npages != 1)) {
1158                 struct vm_area_struct *vma;
1159
1160                 if (atomic)
1161                         return get_fault_pfn();
1162
1163                 down_read(&current->mm->mmap_sem);
1164                 if (npages == -EHWPOISON ||
1165                         (!async && check_user_page_hwpoison(addr))) {
1166                         up_read(&current->mm->mmap_sem);
1167                         get_page(hwpoison_page);
1168                         return page_to_pfn(hwpoison_page);
1169                 }
1170
1171                 vma = find_vma_intersection(current->mm, addr, addr+1);
1172
1173                 if (vma == NULL)
1174                         pfn = get_fault_pfn();
1175                 else if ((vma->vm_flags & VM_PFNMAP)) {
1176                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1177                                 vma->vm_pgoff;
1178                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1179                 } else {
1180                         if (async && (vma->vm_flags & VM_WRITE))
1181                                 *async = true;
1182                         pfn = get_fault_pfn();
1183                 }
1184                 up_read(&current->mm->mmap_sem);
1185         } else
1186                 pfn = page_to_pfn(page[0]);
1187
1188         return pfn;
1189 }
1190
1191 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1192 {
1193         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1194 }
1195 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1196
1197 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1198                           bool write_fault, bool *writable)
1199 {
1200         unsigned long addr;
1201
1202         if (async)
1203                 *async = false;
1204
1205         addr = gfn_to_hva(kvm, gfn);
1206         if (kvm_is_error_hva(addr)) {
1207                 get_page(bad_page);
1208                 return page_to_pfn(bad_page);
1209         }
1210
1211         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1212 }
1213
1214 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1215 {
1216         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1217 }
1218 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1219
1220 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1221                        bool write_fault, bool *writable)
1222 {
1223         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1224 }
1225 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1226
1227 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1228 {
1229         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1230 }
1231 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1232
1233 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1234                       bool *writable)
1235 {
1236         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1237 }
1238 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1239
1240 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1241                          struct kvm_memory_slot *slot, gfn_t gfn)
1242 {
1243         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1244         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1245 }
1246
1247 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1248                                                                   int nr_pages)
1249 {
1250         unsigned long addr;
1251         gfn_t entry;
1252
1253         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1254         if (kvm_is_error_hva(addr))
1255                 return -1;
1256
1257         if (entry < nr_pages)
1258                 return 0;
1259
1260         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1261 }
1262 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1263
1264 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1265 {
1266         pfn_t pfn;
1267
1268         pfn = gfn_to_pfn(kvm, gfn);
1269         if (!kvm_is_mmio_pfn(pfn))
1270                 return pfn_to_page(pfn);
1271
1272         WARN_ON(kvm_is_mmio_pfn(pfn));
1273
1274         get_page(bad_page);
1275         return bad_page;
1276 }
1277
1278 EXPORT_SYMBOL_GPL(gfn_to_page);
1279
1280 void kvm_release_page_clean(struct page *page)
1281 {
1282         kvm_release_pfn_clean(page_to_pfn(page));
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1285
1286 void kvm_release_pfn_clean(pfn_t pfn)
1287 {
1288         if (!kvm_is_mmio_pfn(pfn))
1289                 put_page(pfn_to_page(pfn));
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1292
1293 void kvm_release_page_dirty(struct page *page)
1294 {
1295         kvm_release_pfn_dirty(page_to_pfn(page));
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1298
1299 void kvm_release_pfn_dirty(pfn_t pfn)
1300 {
1301         kvm_set_pfn_dirty(pfn);
1302         kvm_release_pfn_clean(pfn);
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1305
1306 void kvm_set_page_dirty(struct page *page)
1307 {
1308         kvm_set_pfn_dirty(page_to_pfn(page));
1309 }
1310 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1311
1312 void kvm_set_pfn_dirty(pfn_t pfn)
1313 {
1314         if (!kvm_is_mmio_pfn(pfn)) {
1315                 struct page *page = pfn_to_page(pfn);
1316                 if (!PageReserved(page))
1317                         SetPageDirty(page);
1318         }
1319 }
1320 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1321
1322 void kvm_set_pfn_accessed(pfn_t pfn)
1323 {
1324         if (!kvm_is_mmio_pfn(pfn))
1325                 mark_page_accessed(pfn_to_page(pfn));
1326 }
1327 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1328
1329 void kvm_get_pfn(pfn_t pfn)
1330 {
1331         if (!kvm_is_mmio_pfn(pfn))
1332                 get_page(pfn_to_page(pfn));
1333 }
1334 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1335
1336 static int next_segment(unsigned long len, int offset)
1337 {
1338         if (len > PAGE_SIZE - offset)
1339                 return PAGE_SIZE - offset;
1340         else
1341                 return len;
1342 }
1343
1344 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1345                         int len)
1346 {
1347         int r;
1348         unsigned long addr;
1349
1350         addr = gfn_to_hva(kvm, gfn);
1351         if (kvm_is_error_hva(addr))
1352                 return -EFAULT;
1353         r = __copy_from_user(data, (void __user *)addr + offset, len);
1354         if (r)
1355                 return -EFAULT;
1356         return 0;
1357 }
1358 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1359
1360 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1361 {
1362         gfn_t gfn = gpa >> PAGE_SHIFT;
1363         int seg;
1364         int offset = offset_in_page(gpa);
1365         int ret;
1366
1367         while ((seg = next_segment(len, offset)) != 0) {
1368                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1369                 if (ret < 0)
1370                         return ret;
1371                 offset = 0;
1372                 len -= seg;
1373                 data += seg;
1374                 ++gfn;
1375         }
1376         return 0;
1377 }
1378 EXPORT_SYMBOL_GPL(kvm_read_guest);
1379
1380 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1381                           unsigned long len)
1382 {
1383         int r;
1384         unsigned long addr;
1385         gfn_t gfn = gpa >> PAGE_SHIFT;
1386         int offset = offset_in_page(gpa);
1387
1388         addr = gfn_to_hva(kvm, gfn);
1389         if (kvm_is_error_hva(addr))
1390                 return -EFAULT;
1391         pagefault_disable();
1392         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1393         pagefault_enable();
1394         if (r)
1395                 return -EFAULT;
1396         return 0;
1397 }
1398 EXPORT_SYMBOL(kvm_read_guest_atomic);
1399
1400 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1401                          int offset, int len)
1402 {
1403         int r;
1404         unsigned long addr;
1405
1406         addr = gfn_to_hva(kvm, gfn);
1407         if (kvm_is_error_hva(addr))
1408                 return -EFAULT;
1409         r = __copy_to_user((void __user *)addr + offset, data, len);
1410         if (r)
1411                 return -EFAULT;
1412         mark_page_dirty(kvm, gfn);
1413         return 0;
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1416
1417 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1418                     unsigned long len)
1419 {
1420         gfn_t gfn = gpa >> PAGE_SHIFT;
1421         int seg;
1422         int offset = offset_in_page(gpa);
1423         int ret;
1424
1425         while ((seg = next_segment(len, offset)) != 0) {
1426                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1427                 if (ret < 0)
1428                         return ret;
1429                 offset = 0;
1430                 len -= seg;
1431                 data += seg;
1432                 ++gfn;
1433         }
1434         return 0;
1435 }
1436
1437 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1438                               gpa_t gpa)
1439 {
1440         struct kvm_memslots *slots = kvm_memslots(kvm);
1441         int offset = offset_in_page(gpa);
1442         gfn_t gfn = gpa >> PAGE_SHIFT;
1443
1444         ghc->gpa = gpa;
1445         ghc->generation = slots->generation;
1446         ghc->memslot = gfn_to_memslot(kvm, gfn);
1447         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1448         if (!kvm_is_error_hva(ghc->hva))
1449                 ghc->hva += offset;
1450         else
1451                 return -EFAULT;
1452
1453         return 0;
1454 }
1455 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1456
1457 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1458                            void *data, unsigned long len)
1459 {
1460         struct kvm_memslots *slots = kvm_memslots(kvm);
1461         int r;
1462
1463         if (slots->generation != ghc->generation)
1464                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1465
1466         if (kvm_is_error_hva(ghc->hva))
1467                 return -EFAULT;
1468
1469         r = __copy_to_user((void __user *)ghc->hva, data, len);
1470         if (r)
1471                 return -EFAULT;
1472         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1473
1474         return 0;
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1477
1478 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1479                            void *data, unsigned long len)
1480 {
1481         struct kvm_memslots *slots = kvm_memslots(kvm);
1482         int r;
1483
1484         if (slots->generation != ghc->generation)
1485                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1486
1487         if (kvm_is_error_hva(ghc->hva))
1488                 return -EFAULT;
1489
1490         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1491         if (r)
1492                 return -EFAULT;
1493
1494         return 0;
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1497
1498 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1499 {
1500         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1501                                     offset, len);
1502 }
1503 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1504
1505 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1506 {
1507         gfn_t gfn = gpa >> PAGE_SHIFT;
1508         int seg;
1509         int offset = offset_in_page(gpa);
1510         int ret;
1511
1512         while ((seg = next_segment(len, offset)) != 0) {
1513                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1514                 if (ret < 0)
1515                         return ret;
1516                 offset = 0;
1517                 len -= seg;
1518                 ++gfn;
1519         }
1520         return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1523
1524 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1525                              gfn_t gfn)
1526 {
1527         if (memslot && memslot->dirty_bitmap) {
1528                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1529
1530                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1531                         memslot->nr_dirty_pages++;
1532         }
1533 }
1534
1535 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1536 {
1537         struct kvm_memory_slot *memslot;
1538
1539         memslot = gfn_to_memslot(kvm, gfn);
1540         mark_page_dirty_in_slot(kvm, memslot, gfn);
1541 }
1542
1543 /*
1544  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1545  */
1546 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1547 {
1548         DEFINE_WAIT(wait);
1549
1550         for (;;) {
1551                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1552
1553                 if (kvm_arch_vcpu_runnable(vcpu)) {
1554                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1555                         break;
1556                 }
1557                 if (kvm_cpu_has_pending_timer(vcpu))
1558                         break;
1559                 if (signal_pending(current))
1560                         break;
1561
1562                 schedule();
1563         }
1564
1565         finish_wait(&vcpu->wq, &wait);
1566 }
1567
1568 void kvm_resched(struct kvm_vcpu *vcpu)
1569 {
1570         if (!need_resched())
1571                 return;
1572         cond_resched();
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_resched);
1575
1576 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1577 {
1578         struct kvm *kvm = me->kvm;
1579         struct kvm_vcpu *vcpu;
1580         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1581         int yielded = 0;
1582         int pass;
1583         int i;
1584
1585         /*
1586          * We boost the priority of a VCPU that is runnable but not
1587          * currently running, because it got preempted by something
1588          * else and called schedule in __vcpu_run.  Hopefully that
1589          * VCPU is holding the lock that we need and will release it.
1590          * We approximate round-robin by starting at the last boosted VCPU.
1591          */
1592         for (pass = 0; pass < 2 && !yielded; pass++) {
1593                 kvm_for_each_vcpu(i, vcpu, kvm) {
1594                         struct task_struct *task = NULL;
1595                         struct pid *pid;
1596                         if (!pass && i < last_boosted_vcpu) {
1597                                 i = last_boosted_vcpu;
1598                                 continue;
1599                         } else if (pass && i > last_boosted_vcpu)
1600                                 break;
1601                         if (vcpu == me)
1602                                 continue;
1603                         if (waitqueue_active(&vcpu->wq))
1604                                 continue;
1605                         rcu_read_lock();
1606                         pid = rcu_dereference(vcpu->pid);
1607                         if (pid)
1608                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1609                         rcu_read_unlock();
1610                         if (!task)
1611                                 continue;
1612                         if (task->flags & PF_VCPU) {
1613                                 put_task_struct(task);
1614                                 continue;
1615                         }
1616                         if (yield_to(task, 1)) {
1617                                 put_task_struct(task);
1618                                 kvm->last_boosted_vcpu = i;
1619                                 yielded = 1;
1620                                 break;
1621                         }
1622                         put_task_struct(task);
1623                 }
1624         }
1625 }
1626 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1627
1628 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1629 {
1630         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1631         struct page *page;
1632
1633         if (vmf->pgoff == 0)
1634                 page = virt_to_page(vcpu->run);
1635 #ifdef CONFIG_X86
1636         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1637                 page = virt_to_page(vcpu->arch.pio_data);
1638 #endif
1639 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1640         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1641                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1642 #endif
1643         else
1644                 return kvm_arch_vcpu_fault(vcpu, vmf);
1645         get_page(page);
1646         vmf->page = page;
1647         return 0;
1648 }
1649
1650 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1651         .fault = kvm_vcpu_fault,
1652 };
1653
1654 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1655 {
1656         vma->vm_ops = &kvm_vcpu_vm_ops;
1657         return 0;
1658 }
1659
1660 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1661 {
1662         struct kvm_vcpu *vcpu = filp->private_data;
1663
1664         kvm_put_kvm(vcpu->kvm);
1665         return 0;
1666 }
1667
1668 static struct file_operations kvm_vcpu_fops = {
1669         .release        = kvm_vcpu_release,
1670         .unlocked_ioctl = kvm_vcpu_ioctl,
1671 #ifdef CONFIG_COMPAT
1672         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1673 #endif
1674         .mmap           = kvm_vcpu_mmap,
1675         .llseek         = noop_llseek,
1676 };
1677
1678 /*
1679  * Allocates an inode for the vcpu.
1680  */
1681 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1682 {
1683         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1684 }
1685
1686 /*
1687  * Creates some virtual cpus.  Good luck creating more than one.
1688  */
1689 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1690 {
1691         int r;
1692         struct kvm_vcpu *vcpu, *v;
1693
1694         vcpu = kvm_arch_vcpu_create(kvm, id);
1695         if (IS_ERR(vcpu))
1696                 return PTR_ERR(vcpu);
1697
1698         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1699
1700         r = kvm_arch_vcpu_setup(vcpu);
1701         if (r)
1702                 goto vcpu_destroy;
1703
1704         mutex_lock(&kvm->lock);
1705         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1706                 r = -EINVAL;
1707                 goto unlock_vcpu_destroy;
1708         }
1709
1710         kvm_for_each_vcpu(r, v, kvm)
1711                 if (v->vcpu_id == id) {
1712                         r = -EEXIST;
1713                         goto unlock_vcpu_destroy;
1714                 }
1715
1716         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1717
1718         /* Now it's all set up, let userspace reach it */
1719         kvm_get_kvm(kvm);
1720         r = create_vcpu_fd(vcpu);
1721         if (r < 0) {
1722                 kvm_put_kvm(kvm);
1723                 goto unlock_vcpu_destroy;
1724         }
1725
1726         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1727         smp_wmb();
1728         atomic_inc(&kvm->online_vcpus);
1729
1730         mutex_unlock(&kvm->lock);
1731         return r;
1732
1733 unlock_vcpu_destroy:
1734         mutex_unlock(&kvm->lock);
1735 vcpu_destroy:
1736         kvm_arch_vcpu_destroy(vcpu);
1737         return r;
1738 }
1739
1740 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1741 {
1742         if (sigset) {
1743                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1744                 vcpu->sigset_active = 1;
1745                 vcpu->sigset = *sigset;
1746         } else
1747                 vcpu->sigset_active = 0;
1748         return 0;
1749 }
1750
1751 static long kvm_vcpu_ioctl(struct file *filp,
1752                            unsigned int ioctl, unsigned long arg)
1753 {
1754         struct kvm_vcpu *vcpu = filp->private_data;
1755         void __user *argp = (void __user *)arg;
1756         int r;
1757         struct kvm_fpu *fpu = NULL;
1758         struct kvm_sregs *kvm_sregs = NULL;
1759
1760         if (vcpu->kvm->mm != current->mm)
1761                 return -EIO;
1762
1763 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1764         /*
1765          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1766          * so vcpu_load() would break it.
1767          */
1768         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1769                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1770 #endif
1771
1772
1773         vcpu_load(vcpu);
1774         switch (ioctl) {
1775         case KVM_RUN:
1776                 r = -EINVAL;
1777                 if (arg)
1778                         goto out;
1779                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1780                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1781                 break;
1782         case KVM_GET_REGS: {
1783                 struct kvm_regs *kvm_regs;
1784
1785                 r = -ENOMEM;
1786                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1787                 if (!kvm_regs)
1788                         goto out;
1789                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1790                 if (r)
1791                         goto out_free1;
1792                 r = -EFAULT;
1793                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1794                         goto out_free1;
1795                 r = 0;
1796 out_free1:
1797                 kfree(kvm_regs);
1798                 break;
1799         }
1800         case KVM_SET_REGS: {
1801                 struct kvm_regs *kvm_regs;
1802
1803                 r = -ENOMEM;
1804                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1805                 if (IS_ERR(kvm_regs)) {
1806                         r = PTR_ERR(kvm_regs);
1807                         goto out;
1808                 }
1809                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1810                 if (r)
1811                         goto out_free2;
1812                 r = 0;
1813 out_free2:
1814                 kfree(kvm_regs);
1815                 break;
1816         }
1817         case KVM_GET_SREGS: {
1818                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1819                 r = -ENOMEM;
1820                 if (!kvm_sregs)
1821                         goto out;
1822                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1823                 if (r)
1824                         goto out;
1825                 r = -EFAULT;
1826                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1827                         goto out;
1828                 r = 0;
1829                 break;
1830         }
1831         case KVM_SET_SREGS: {
1832                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1833                 if (IS_ERR(kvm_sregs)) {
1834                         r = PTR_ERR(kvm_sregs);
1835                         goto out;
1836                 }
1837                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1838                 if (r)
1839                         goto out;
1840                 r = 0;
1841                 break;
1842         }
1843         case KVM_GET_MP_STATE: {
1844                 struct kvm_mp_state mp_state;
1845
1846                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1847                 if (r)
1848                         goto out;
1849                 r = -EFAULT;
1850                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1851                         goto out;
1852                 r = 0;
1853                 break;
1854         }
1855         case KVM_SET_MP_STATE: {
1856                 struct kvm_mp_state mp_state;
1857
1858                 r = -EFAULT;
1859                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1860                         goto out;
1861                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1862                 if (r)
1863                         goto out;
1864                 r = 0;
1865                 break;
1866         }
1867         case KVM_TRANSLATE: {
1868                 struct kvm_translation tr;
1869
1870                 r = -EFAULT;
1871                 if (copy_from_user(&tr, argp, sizeof tr))
1872                         goto out;
1873                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1874                 if (r)
1875                         goto out;
1876                 r = -EFAULT;
1877                 if (copy_to_user(argp, &tr, sizeof tr))
1878                         goto out;
1879                 r = 0;
1880                 break;
1881         }
1882         case KVM_SET_GUEST_DEBUG: {
1883                 struct kvm_guest_debug dbg;
1884
1885                 r = -EFAULT;
1886                 if (copy_from_user(&dbg, argp, sizeof dbg))
1887                         goto out;
1888                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1889                 if (r)
1890                         goto out;
1891                 r = 0;
1892                 break;
1893         }
1894         case KVM_SET_SIGNAL_MASK: {
1895                 struct kvm_signal_mask __user *sigmask_arg = argp;
1896                 struct kvm_signal_mask kvm_sigmask;
1897                 sigset_t sigset, *p;
1898
1899                 p = NULL;
1900                 if (argp) {
1901                         r = -EFAULT;
1902                         if (copy_from_user(&kvm_sigmask, argp,
1903                                            sizeof kvm_sigmask))
1904                                 goto out;
1905                         r = -EINVAL;
1906                         if (kvm_sigmask.len != sizeof sigset)
1907                                 goto out;
1908                         r = -EFAULT;
1909                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1910                                            sizeof sigset))
1911                                 goto out;
1912                         p = &sigset;
1913                 }
1914                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1915                 break;
1916         }
1917         case KVM_GET_FPU: {
1918                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1919                 r = -ENOMEM;
1920                 if (!fpu)
1921                         goto out;
1922                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1923                 if (r)
1924                         goto out;
1925                 r = -EFAULT;
1926                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1927                         goto out;
1928                 r = 0;
1929                 break;
1930         }
1931         case KVM_SET_FPU: {
1932                 fpu = memdup_user(argp, sizeof(*fpu));
1933                 if (IS_ERR(fpu)) {
1934                         r = PTR_ERR(fpu);
1935                         goto out;
1936                 }
1937                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1938                 if (r)
1939                         goto out;
1940                 r = 0;
1941                 break;
1942         }
1943         default:
1944                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1945         }
1946 out:
1947         vcpu_put(vcpu);
1948         kfree(fpu);
1949         kfree(kvm_sregs);
1950         return r;
1951 }
1952
1953 #ifdef CONFIG_COMPAT
1954 static long kvm_vcpu_compat_ioctl(struct file *filp,
1955                                   unsigned int ioctl, unsigned long arg)
1956 {
1957         struct kvm_vcpu *vcpu = filp->private_data;
1958         void __user *argp = compat_ptr(arg);
1959         int r;
1960
1961         if (vcpu->kvm->mm != current->mm)
1962                 return -EIO;
1963
1964         switch (ioctl) {
1965         case KVM_SET_SIGNAL_MASK: {
1966                 struct kvm_signal_mask __user *sigmask_arg = argp;
1967                 struct kvm_signal_mask kvm_sigmask;
1968                 compat_sigset_t csigset;
1969                 sigset_t sigset;
1970
1971                 if (argp) {
1972                         r = -EFAULT;
1973                         if (copy_from_user(&kvm_sigmask, argp,
1974                                            sizeof kvm_sigmask))
1975                                 goto out;
1976                         r = -EINVAL;
1977                         if (kvm_sigmask.len != sizeof csigset)
1978                                 goto out;
1979                         r = -EFAULT;
1980                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1981                                            sizeof csigset))
1982                                 goto out;
1983                 }
1984                 sigset_from_compat(&sigset, &csigset);
1985                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1986                 break;
1987         }
1988         default:
1989                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1990         }
1991
1992 out:
1993         return r;
1994 }
1995 #endif
1996
1997 static long kvm_vm_ioctl(struct file *filp,
1998                            unsigned int ioctl, unsigned long arg)
1999 {
2000         struct kvm *kvm = filp->private_data;
2001         void __user *argp = (void __user *)arg;
2002         int r;
2003
2004         if (kvm->mm != current->mm)
2005                 return -EIO;
2006         switch (ioctl) {
2007         case KVM_CREATE_VCPU:
2008                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2009                 if (r < 0)
2010                         goto out;
2011                 break;
2012         case KVM_SET_USER_MEMORY_REGION: {
2013                 struct kvm_userspace_memory_region kvm_userspace_mem;
2014
2015                 r = -EFAULT;
2016                 if (copy_from_user(&kvm_userspace_mem, argp,
2017                                                 sizeof kvm_userspace_mem))
2018                         goto out;
2019
2020                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2021                 if (r)
2022                         goto out;
2023                 break;
2024         }
2025         case KVM_GET_DIRTY_LOG: {
2026                 struct kvm_dirty_log log;
2027
2028                 r = -EFAULT;
2029                 if (copy_from_user(&log, argp, sizeof log))
2030                         goto out;
2031                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2032                 if (r)
2033                         goto out;
2034                 break;
2035         }
2036 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2037         case KVM_REGISTER_COALESCED_MMIO: {
2038                 struct kvm_coalesced_mmio_zone zone;
2039                 r = -EFAULT;
2040                 if (copy_from_user(&zone, argp, sizeof zone))
2041                         goto out;
2042                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2043                 if (r)
2044                         goto out;
2045                 r = 0;
2046                 break;
2047         }
2048         case KVM_UNREGISTER_COALESCED_MMIO: {
2049                 struct kvm_coalesced_mmio_zone zone;
2050                 r = -EFAULT;
2051                 if (copy_from_user(&zone, argp, sizeof zone))
2052                         goto out;
2053                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2054                 if (r)
2055                         goto out;
2056                 r = 0;
2057                 break;
2058         }
2059 #endif
2060         case KVM_IRQFD: {
2061                 struct kvm_irqfd data;
2062
2063                 r = -EFAULT;
2064                 if (copy_from_user(&data, argp, sizeof data))
2065                         goto out;
2066                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2067                 break;
2068         }
2069         case KVM_IOEVENTFD: {
2070                 struct kvm_ioeventfd data;
2071
2072                 r = -EFAULT;
2073                 if (copy_from_user(&data, argp, sizeof data))
2074                         goto out;
2075                 r = kvm_ioeventfd(kvm, &data);
2076                 break;
2077         }
2078 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2079         case KVM_SET_BOOT_CPU_ID:
2080                 r = 0;
2081                 mutex_lock(&kvm->lock);
2082                 if (atomic_read(&kvm->online_vcpus) != 0)
2083                         r = -EBUSY;
2084                 else
2085                         kvm->bsp_vcpu_id = arg;
2086                 mutex_unlock(&kvm->lock);
2087                 break;
2088 #endif
2089         default:
2090                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2091                 if (r == -ENOTTY)
2092                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2093         }
2094 out:
2095         return r;
2096 }
2097
2098 #ifdef CONFIG_COMPAT
2099 struct compat_kvm_dirty_log {
2100         __u32 slot;
2101         __u32 padding1;
2102         union {
2103                 compat_uptr_t dirty_bitmap; /* one bit per page */
2104                 __u64 padding2;
2105         };
2106 };
2107
2108 static long kvm_vm_compat_ioctl(struct file *filp,
2109                            unsigned int ioctl, unsigned long arg)
2110 {
2111         struct kvm *kvm = filp->private_data;
2112         int r;
2113
2114         if (kvm->mm != current->mm)
2115                 return -EIO;
2116         switch (ioctl) {
2117         case KVM_GET_DIRTY_LOG: {
2118                 struct compat_kvm_dirty_log compat_log;
2119                 struct kvm_dirty_log log;
2120
2121                 r = -EFAULT;
2122                 if (copy_from_user(&compat_log, (void __user *)arg,
2123                                    sizeof(compat_log)))
2124                         goto out;
2125                 log.slot         = compat_log.slot;
2126                 log.padding1     = compat_log.padding1;
2127                 log.padding2     = compat_log.padding2;
2128                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2129
2130                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2131                 if (r)
2132                         goto out;
2133                 break;
2134         }
2135         default:
2136                 r = kvm_vm_ioctl(filp, ioctl, arg);
2137         }
2138
2139 out:
2140         return r;
2141 }
2142 #endif
2143
2144 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2145 {
2146         struct page *page[1];
2147         unsigned long addr;
2148         int npages;
2149         gfn_t gfn = vmf->pgoff;
2150         struct kvm *kvm = vma->vm_file->private_data;
2151
2152         addr = gfn_to_hva(kvm, gfn);
2153         if (kvm_is_error_hva(addr))
2154                 return VM_FAULT_SIGBUS;
2155
2156         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2157                                 NULL);
2158         if (unlikely(npages != 1))
2159                 return VM_FAULT_SIGBUS;
2160
2161         vmf->page = page[0];
2162         return 0;
2163 }
2164
2165 static const struct vm_operations_struct kvm_vm_vm_ops = {
2166         .fault = kvm_vm_fault,
2167 };
2168
2169 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2170 {
2171         vma->vm_ops = &kvm_vm_vm_ops;
2172         return 0;
2173 }
2174
2175 static struct file_operations kvm_vm_fops = {
2176         .release        = kvm_vm_release,
2177         .unlocked_ioctl = kvm_vm_ioctl,
2178 #ifdef CONFIG_COMPAT
2179         .compat_ioctl   = kvm_vm_compat_ioctl,
2180 #endif
2181         .mmap           = kvm_vm_mmap,
2182         .llseek         = noop_llseek,
2183 };
2184
2185 static int kvm_dev_ioctl_create_vm(unsigned long type)
2186 {
2187         int r;
2188         struct kvm *kvm;
2189
2190         kvm = kvm_create_vm(type);
2191         if (IS_ERR(kvm))
2192                 return PTR_ERR(kvm);
2193 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2194         r = kvm_coalesced_mmio_init(kvm);
2195         if (r < 0) {
2196                 kvm_put_kvm(kvm);
2197                 return r;
2198         }
2199 #endif
2200         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2201         if (r < 0)
2202                 kvm_put_kvm(kvm);
2203
2204         return r;
2205 }
2206
2207 static long kvm_dev_ioctl_check_extension_generic(long arg)
2208 {
2209         switch (arg) {
2210         case KVM_CAP_USER_MEMORY:
2211         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2212         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2213 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2214         case KVM_CAP_SET_BOOT_CPU_ID:
2215 #endif
2216         case KVM_CAP_INTERNAL_ERROR_DATA:
2217                 return 1;
2218 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2219         case KVM_CAP_IRQ_ROUTING:
2220                 return KVM_MAX_IRQ_ROUTES;
2221 #endif
2222         default:
2223                 break;
2224         }
2225         return kvm_dev_ioctl_check_extension(arg);
2226 }
2227
2228 static long kvm_dev_ioctl(struct file *filp,
2229                           unsigned int ioctl, unsigned long arg)
2230 {
2231         long r = -EINVAL;
2232
2233         switch (ioctl) {
2234         case KVM_GET_API_VERSION:
2235                 r = -EINVAL;
2236                 if (arg)
2237                         goto out;
2238                 r = KVM_API_VERSION;
2239                 break;
2240         case KVM_CREATE_VM:
2241                 r = kvm_dev_ioctl_create_vm(arg);
2242                 break;
2243         case KVM_CHECK_EXTENSION:
2244                 r = kvm_dev_ioctl_check_extension_generic(arg);
2245                 break;
2246         case KVM_GET_VCPU_MMAP_SIZE:
2247                 r = -EINVAL;
2248                 if (arg)
2249                         goto out;
2250                 r = PAGE_SIZE;     /* struct kvm_run */
2251 #ifdef CONFIG_X86
2252                 r += PAGE_SIZE;    /* pio data page */
2253 #endif
2254 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2255                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2256 #endif
2257                 break;
2258         case KVM_TRACE_ENABLE:
2259         case KVM_TRACE_PAUSE:
2260         case KVM_TRACE_DISABLE:
2261                 r = -EOPNOTSUPP;
2262                 break;
2263         default:
2264                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2265         }
2266 out:
2267         return r;
2268 }
2269
2270 static struct file_operations kvm_chardev_ops = {
2271         .unlocked_ioctl = kvm_dev_ioctl,
2272         .compat_ioctl   = kvm_dev_ioctl,
2273         .llseek         = noop_llseek,
2274 };
2275
2276 static struct miscdevice kvm_dev = {
2277         KVM_MINOR,
2278         "kvm",
2279         &kvm_chardev_ops,
2280 };
2281
2282 static void hardware_enable_nolock(void *junk)
2283 {
2284         int cpu = raw_smp_processor_id();
2285         int r;
2286
2287         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2288                 return;
2289
2290         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2291
2292         r = kvm_arch_hardware_enable(NULL);
2293
2294         if (r) {
2295                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2296                 atomic_inc(&hardware_enable_failed);
2297                 printk(KERN_INFO "kvm: enabling virtualization on "
2298                                  "CPU%d failed\n", cpu);
2299         }
2300 }
2301
2302 static void hardware_enable(void *junk)
2303 {
2304         raw_spin_lock(&kvm_lock);
2305         hardware_enable_nolock(junk);
2306         raw_spin_unlock(&kvm_lock);
2307 }
2308
2309 static void hardware_disable_nolock(void *junk)
2310 {
2311         int cpu = raw_smp_processor_id();
2312
2313         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2314                 return;
2315         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2316         kvm_arch_hardware_disable(NULL);
2317 }
2318
2319 static void hardware_disable(void *junk)
2320 {
2321         raw_spin_lock(&kvm_lock);
2322         hardware_disable_nolock(junk);
2323         raw_spin_unlock(&kvm_lock);
2324 }
2325
2326 static void hardware_disable_all_nolock(void)
2327 {
2328         BUG_ON(!kvm_usage_count);
2329
2330         kvm_usage_count--;
2331         if (!kvm_usage_count)
2332                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2333 }
2334
2335 static void hardware_disable_all(void)
2336 {
2337         raw_spin_lock(&kvm_lock);
2338         hardware_disable_all_nolock();
2339         raw_spin_unlock(&kvm_lock);
2340 }
2341
2342 static int hardware_enable_all(void)
2343 {
2344         int r = 0;
2345
2346         raw_spin_lock(&kvm_lock);
2347
2348         kvm_usage_count++;
2349         if (kvm_usage_count == 1) {
2350                 atomic_set(&hardware_enable_failed, 0);
2351                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2352
2353                 if (atomic_read(&hardware_enable_failed)) {
2354                         hardware_disable_all_nolock();
2355                         r = -EBUSY;
2356                 }
2357         }
2358
2359         raw_spin_unlock(&kvm_lock);
2360
2361         return r;
2362 }
2363
2364 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2365                            void *v)
2366 {
2367         int cpu = (long)v;
2368
2369         if (!kvm_usage_count)
2370                 return NOTIFY_OK;
2371
2372         val &= ~CPU_TASKS_FROZEN;
2373         switch (val) {
2374         case CPU_DYING:
2375                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2376                        cpu);
2377                 hardware_disable(NULL);
2378                 break;
2379         case CPU_STARTING:
2380                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2381                        cpu);
2382                 hardware_enable(NULL);
2383                 break;
2384         }
2385         return NOTIFY_OK;
2386 }
2387
2388
2389 asmlinkage void kvm_spurious_fault(void)
2390 {
2391         /* Fault while not rebooting.  We want the trace. */
2392         BUG();
2393 }
2394 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2395
2396 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2397                       void *v)
2398 {
2399         /*
2400          * Some (well, at least mine) BIOSes hang on reboot if
2401          * in vmx root mode.
2402          *
2403          * And Intel TXT required VMX off for all cpu when system shutdown.
2404          */
2405         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2406         kvm_rebooting = true;
2407         on_each_cpu(hardware_disable_nolock, NULL, 1);
2408         return NOTIFY_OK;
2409 }
2410
2411 static struct notifier_block kvm_reboot_notifier = {
2412         .notifier_call = kvm_reboot,
2413         .priority = 0,
2414 };
2415
2416 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2417 {
2418         int i;
2419
2420         for (i = 0; i < bus->dev_count; i++) {
2421                 struct kvm_io_device *pos = bus->range[i].dev;
2422
2423                 kvm_iodevice_destructor(pos);
2424         }
2425         kfree(bus);
2426 }
2427
2428 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2429 {
2430         const struct kvm_io_range *r1 = p1;
2431         const struct kvm_io_range *r2 = p2;
2432
2433         if (r1->addr < r2->addr)
2434                 return -1;
2435         if (r1->addr + r1->len > r2->addr + r2->len)
2436                 return 1;
2437         return 0;
2438 }
2439
2440 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2441                           gpa_t addr, int len)
2442 {
2443         if (bus->dev_count == NR_IOBUS_DEVS)
2444                 return -ENOSPC;
2445
2446         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2447                 .addr = addr,
2448                 .len = len,
2449                 .dev = dev,
2450         };
2451
2452         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2453                 kvm_io_bus_sort_cmp, NULL);
2454
2455         return 0;
2456 }
2457
2458 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2459                              gpa_t addr, int len)
2460 {
2461         struct kvm_io_range *range, key;
2462         int off;
2463
2464         key = (struct kvm_io_range) {
2465                 .addr = addr,
2466                 .len = len,
2467         };
2468
2469         range = bsearch(&key, bus->range, bus->dev_count,
2470                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2471         if (range == NULL)
2472                 return -ENOENT;
2473
2474         off = range - bus->range;
2475
2476         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2477                 off--;
2478
2479         return off;
2480 }
2481
2482 /* kvm_io_bus_write - called under kvm->slots_lock */
2483 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2484                      int len, const void *val)
2485 {
2486         int idx;
2487         struct kvm_io_bus *bus;
2488         struct kvm_io_range range;
2489
2490         range = (struct kvm_io_range) {
2491                 .addr = addr,
2492                 .len = len,
2493         };
2494
2495         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2496         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2497         if (idx < 0)
2498                 return -EOPNOTSUPP;
2499
2500         while (idx < bus->dev_count &&
2501                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2502                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2503                         return 0;
2504                 idx++;
2505         }
2506
2507         return -EOPNOTSUPP;
2508 }
2509
2510 /* kvm_io_bus_read - called under kvm->slots_lock */
2511 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2512                     int len, void *val)
2513 {
2514         int idx;
2515         struct kvm_io_bus *bus;
2516         struct kvm_io_range range;
2517
2518         range = (struct kvm_io_range) {
2519                 .addr = addr,
2520                 .len = len,
2521         };
2522
2523         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2524         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2525         if (idx < 0)
2526                 return -EOPNOTSUPP;
2527
2528         while (idx < bus->dev_count &&
2529                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2530                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2531                         return 0;
2532                 idx++;
2533         }
2534
2535         return -EOPNOTSUPP;
2536 }
2537
2538 /* Caller must hold slots_lock. */
2539 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2540                             int len, struct kvm_io_device *dev)
2541 {
2542         struct kvm_io_bus *new_bus, *bus;
2543
2544         bus = kvm->buses[bus_idx];
2545         if (bus->dev_count > NR_IOBUS_DEVS-1)
2546                 return -ENOSPC;
2547
2548         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2549         if (!new_bus)
2550                 return -ENOMEM;
2551         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2552         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2553         synchronize_srcu_expedited(&kvm->srcu);
2554         kfree(bus);
2555
2556         return 0;
2557 }
2558
2559 /* Caller must hold slots_lock. */
2560 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2561                               struct kvm_io_device *dev)
2562 {
2563         int i, r;
2564         struct kvm_io_bus *new_bus, *bus;
2565
2566         bus = kvm->buses[bus_idx];
2567
2568         new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2569         if (!new_bus)
2570                 return -ENOMEM;
2571
2572         r = -ENOENT;
2573         for (i = 0; i < new_bus->dev_count; i++)
2574                 if (new_bus->range[i].dev == dev) {
2575                         r = 0;
2576                         new_bus->dev_count--;
2577                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2578                         sort(new_bus->range, new_bus->dev_count,
2579                              sizeof(struct kvm_io_range),
2580                              kvm_io_bus_sort_cmp, NULL);
2581                         break;
2582                 }
2583
2584         if (r) {
2585                 kfree(new_bus);
2586                 return r;
2587         }
2588
2589         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2590         synchronize_srcu_expedited(&kvm->srcu);
2591         kfree(bus);
2592         return r;
2593 }
2594
2595 static struct notifier_block kvm_cpu_notifier = {
2596         .notifier_call = kvm_cpu_hotplug,
2597 };
2598
2599 static int vm_stat_get(void *_offset, u64 *val)
2600 {
2601         unsigned offset = (long)_offset;
2602         struct kvm *kvm;
2603
2604         *val = 0;
2605         raw_spin_lock(&kvm_lock);
2606         list_for_each_entry(kvm, &vm_list, vm_list)
2607                 *val += *(u32 *)((void *)kvm + offset);
2608         raw_spin_unlock(&kvm_lock);
2609         return 0;
2610 }
2611
2612 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2613
2614 static int vcpu_stat_get(void *_offset, u64 *val)
2615 {
2616         unsigned offset = (long)_offset;
2617         struct kvm *kvm;
2618         struct kvm_vcpu *vcpu;
2619         int i;
2620
2621         *val = 0;
2622         raw_spin_lock(&kvm_lock);
2623         list_for_each_entry(kvm, &vm_list, vm_list)
2624                 kvm_for_each_vcpu(i, vcpu, kvm)
2625                         *val += *(u32 *)((void *)vcpu + offset);
2626
2627         raw_spin_unlock(&kvm_lock);
2628         return 0;
2629 }
2630
2631 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2632
2633 static const struct file_operations *stat_fops[] = {
2634         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2635         [KVM_STAT_VM]   = &vm_stat_fops,
2636 };
2637
2638 static int kvm_init_debug(void)
2639 {
2640         int r = -EFAULT;
2641         struct kvm_stats_debugfs_item *p;
2642
2643         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2644         if (kvm_debugfs_dir == NULL)
2645                 goto out;
2646
2647         for (p = debugfs_entries; p->name; ++p) {
2648                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2649                                                 (void *)(long)p->offset,
2650                                                 stat_fops[p->kind]);
2651                 if (p->dentry == NULL)
2652                         goto out_dir;
2653         }
2654
2655         return 0;
2656
2657 out_dir:
2658         debugfs_remove_recursive(kvm_debugfs_dir);
2659 out:
2660         return r;
2661 }
2662
2663 static void kvm_exit_debug(void)
2664 {
2665         struct kvm_stats_debugfs_item *p;
2666
2667         for (p = debugfs_entries; p->name; ++p)
2668                 debugfs_remove(p->dentry);
2669         debugfs_remove(kvm_debugfs_dir);
2670 }
2671
2672 static int kvm_suspend(void)
2673 {
2674         if (kvm_usage_count)
2675                 hardware_disable_nolock(NULL);
2676         return 0;
2677 }
2678
2679 static void kvm_resume(void)
2680 {
2681         if (kvm_usage_count) {
2682                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2683                 hardware_enable_nolock(NULL);
2684         }
2685 }
2686
2687 static struct syscore_ops kvm_syscore_ops = {
2688         .suspend = kvm_suspend,
2689         .resume = kvm_resume,
2690 };
2691
2692 struct page *bad_page;
2693 pfn_t bad_pfn;
2694
2695 static inline
2696 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2697 {
2698         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2699 }
2700
2701 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2702 {
2703         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2704
2705         kvm_arch_vcpu_load(vcpu, cpu);
2706 }
2707
2708 static void kvm_sched_out(struct preempt_notifier *pn,
2709                           struct task_struct *next)
2710 {
2711         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2712
2713         kvm_arch_vcpu_put(vcpu);
2714 }
2715
2716 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2717                   struct module *module)
2718 {
2719         int r;
2720         int cpu;
2721
2722         r = kvm_arch_init(opaque);
2723         if (r)
2724                 goto out_fail;
2725
2726         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2727
2728         if (bad_page == NULL) {
2729                 r = -ENOMEM;
2730                 goto out;
2731         }
2732
2733         bad_pfn = page_to_pfn(bad_page);
2734
2735         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2736
2737         if (hwpoison_page == NULL) {
2738                 r = -ENOMEM;
2739                 goto out_free_0;
2740         }
2741
2742         hwpoison_pfn = page_to_pfn(hwpoison_page);
2743
2744         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2745
2746         if (fault_page == NULL) {
2747                 r = -ENOMEM;
2748                 goto out_free_0;
2749         }
2750
2751         fault_pfn = page_to_pfn(fault_page);
2752
2753         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2754                 r = -ENOMEM;
2755                 goto out_free_0;
2756         }
2757
2758         r = kvm_arch_hardware_setup();
2759         if (r < 0)
2760                 goto out_free_0a;
2761
2762         for_each_online_cpu(cpu) {
2763                 smp_call_function_single(cpu,
2764                                 kvm_arch_check_processor_compat,
2765                                 &r, 1);
2766                 if (r < 0)
2767                         goto out_free_1;
2768         }
2769
2770         r = register_cpu_notifier(&kvm_cpu_notifier);
2771         if (r)
2772                 goto out_free_2;
2773         register_reboot_notifier(&kvm_reboot_notifier);
2774
2775         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2776         if (!vcpu_align)
2777                 vcpu_align = __alignof__(struct kvm_vcpu);
2778         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2779                                            0, NULL);
2780         if (!kvm_vcpu_cache) {
2781                 r = -ENOMEM;
2782                 goto out_free_3;
2783         }
2784
2785         r = kvm_async_pf_init();
2786         if (r)
2787                 goto out_free;
2788
2789         kvm_chardev_ops.owner = module;
2790         kvm_vm_fops.owner = module;
2791         kvm_vcpu_fops.owner = module;
2792
2793         r = misc_register(&kvm_dev);
2794         if (r) {
2795                 printk(KERN_ERR "kvm: misc device register failed\n");
2796                 goto out_unreg;
2797         }
2798
2799         register_syscore_ops(&kvm_syscore_ops);
2800
2801         kvm_preempt_ops.sched_in = kvm_sched_in;
2802         kvm_preempt_ops.sched_out = kvm_sched_out;
2803
2804         r = kvm_init_debug();
2805         if (r) {
2806                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2807                 goto out_undebugfs;
2808         }
2809
2810         return 0;
2811
2812 out_undebugfs:
2813         unregister_syscore_ops(&kvm_syscore_ops);
2814 out_unreg:
2815         kvm_async_pf_deinit();
2816 out_free:
2817         kmem_cache_destroy(kvm_vcpu_cache);
2818 out_free_3:
2819         unregister_reboot_notifier(&kvm_reboot_notifier);
2820         unregister_cpu_notifier(&kvm_cpu_notifier);
2821 out_free_2:
2822 out_free_1:
2823         kvm_arch_hardware_unsetup();
2824 out_free_0a:
2825         free_cpumask_var(cpus_hardware_enabled);
2826 out_free_0:
2827         if (fault_page)
2828                 __free_page(fault_page);
2829         if (hwpoison_page)
2830                 __free_page(hwpoison_page);
2831         __free_page(bad_page);
2832 out:
2833         kvm_arch_exit();
2834 out_fail:
2835         return r;
2836 }
2837 EXPORT_SYMBOL_GPL(kvm_init);
2838
2839 void kvm_exit(void)
2840 {
2841         kvm_exit_debug();
2842         misc_deregister(&kvm_dev);
2843         kmem_cache_destroy(kvm_vcpu_cache);
2844         kvm_async_pf_deinit();
2845         unregister_syscore_ops(&kvm_syscore_ops);
2846         unregister_reboot_notifier(&kvm_reboot_notifier);
2847         unregister_cpu_notifier(&kvm_cpu_notifier);
2848         on_each_cpu(hardware_disable_nolock, NULL, 1);
2849         kvm_arch_hardware_unsetup();
2850         kvm_arch_exit();
2851         free_cpumask_var(cpus_hardware_enabled);
2852         __free_page(hwpoison_page);
2853         __free_page(bad_page);
2854 }
2855 EXPORT_SYMBOL_GPL(kvm_exit);