KVM: Split lpage_info creation out from __kvm_set_memory_region()
[linux-2.6.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * 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 #ifndef CONFIG_S390
620 /*
621  * Allocation size is twice as large as the actual dirty bitmap size.
622  * This makes it possible to do double buffering: see x86's
623  * kvm_vm_ioctl_get_dirty_log().
624  */
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626 {
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         return 0;
640 }
641 #endif /* !CONFIG_S390 */
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                 if (slot->lpage_info[i])
699                         continue;
700
701                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
702                                       slot->base_gfn, level) + 1;
703
704                 slot->lpage_info[i] = vzalloc(lpages * sizeof(*slot->lpage_info[i]));
705                 if (!slot->lpage_info[i])
706                         goto out_free;
707
708                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
709                         slot->lpage_info[i][0].write_count = 1;
710                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
711                         slot->lpage_info[i][lpages - 1].write_count = 1;
712                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
713                 /*
714                  * If the gfn and userspace address are not aligned wrt each
715                  * other, or if explicitly asked to, disable large page
716                  * support for this slot
717                  */
718                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
719                     !largepages_enabled) {
720                         unsigned long j;
721
722                         for (j = 0; j < lpages; ++j)
723                                 slot->lpage_info[i][j].write_count = 1;
724                 }
725         }
726
727         return 0;
728
729 out_free:
730         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
731                 vfree(slot->lpage_info[i]);
732                 slot->lpage_info[i] = NULL;
733         }
734         return -ENOMEM;
735 }
736 #endif /* not defined CONFIG_S390 */
737
738 /*
739  * Allocate some memory and give it an address in the guest physical address
740  * space.
741  *
742  * Discontiguous memory is allowed, mostly for framebuffers.
743  *
744  * Must be called holding mmap_sem for write.
745  */
746 int __kvm_set_memory_region(struct kvm *kvm,
747                             struct kvm_userspace_memory_region *mem,
748                             int user_alloc)
749 {
750         int r;
751         gfn_t base_gfn;
752         unsigned long npages;
753         unsigned long i;
754         struct kvm_memory_slot *memslot;
755         struct kvm_memory_slot old, new;
756         struct kvm_memslots *slots, *old_memslots;
757
758         r = -EINVAL;
759         /* General sanity checks */
760         if (mem->memory_size & (PAGE_SIZE - 1))
761                 goto out;
762         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
763                 goto out;
764         /* We can read the guest memory with __xxx_user() later on. */
765         if (user_alloc &&
766             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
767              !access_ok(VERIFY_WRITE,
768                         (void __user *)(unsigned long)mem->userspace_addr,
769                         mem->memory_size)))
770                 goto out;
771         if (mem->slot >= KVM_MEM_SLOTS_NUM)
772                 goto out;
773         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
774                 goto out;
775
776         memslot = id_to_memslot(kvm->memslots, mem->slot);
777         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
778         npages = mem->memory_size >> PAGE_SHIFT;
779
780         r = -EINVAL;
781         if (npages > KVM_MEM_MAX_NR_PAGES)
782                 goto out;
783
784         if (!npages)
785                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
786
787         new = old = *memslot;
788
789         new.id = mem->slot;
790         new.base_gfn = base_gfn;
791         new.npages = npages;
792         new.flags = mem->flags;
793
794         /* Disallow changing a memory slot's size. */
795         r = -EINVAL;
796         if (npages && old.npages && npages != old.npages)
797                 goto out_free;
798
799         /* Check for overlaps */
800         r = -EEXIST;
801         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
802                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
803
804                 if (s == memslot || !s->npages)
805                         continue;
806                 if (!((base_gfn + npages <= s->base_gfn) ||
807                       (base_gfn >= s->base_gfn + s->npages)))
808                         goto out_free;
809         }
810
811         /* Free page dirty bitmap if unneeded */
812         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
813                 new.dirty_bitmap = NULL;
814
815         r = -ENOMEM;
816
817         /* Allocate if a slot is being created */
818 #ifndef CONFIG_S390
819         if (npages && !new.rmap) {
820                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
821
822                 if (!new.rmap)
823                         goto out_free;
824
825                 new.user_alloc = user_alloc;
826                 new.userspace_addr = mem->userspace_addr;
827         }
828         if (!npages)
829                 goto skip_lpage;
830
831         if (create_lpage_info(&new, npages))
832                 goto out_free;
833
834 skip_lpage:
835
836         /* Allocate page dirty bitmap if needed */
837         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
838                 if (kvm_create_dirty_bitmap(&new) < 0)
839                         goto out_free;
840                 /* destroy any largepage mappings for dirty tracking */
841         }
842 #else  /* not defined CONFIG_S390 */
843         new.user_alloc = user_alloc;
844         if (user_alloc)
845                 new.userspace_addr = mem->userspace_addr;
846 #endif /* not defined CONFIG_S390 */
847
848         if (!npages) {
849                 struct kvm_memory_slot *slot;
850
851                 r = -ENOMEM;
852                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
853                                 GFP_KERNEL);
854                 if (!slots)
855                         goto out_free;
856                 slot = id_to_memslot(slots, mem->slot);
857                 slot->flags |= KVM_MEMSLOT_INVALID;
858
859                 update_memslots(slots, NULL);
860
861                 old_memslots = kvm->memslots;
862                 rcu_assign_pointer(kvm->memslots, slots);
863                 synchronize_srcu_expedited(&kvm->srcu);
864                 /* From this point no new shadow pages pointing to a deleted
865                  * memslot will be created.
866                  *
867                  * validation of sp->gfn happens in:
868                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
869                  *      - kvm_is_visible_gfn (mmu_check_roots)
870                  */
871                 kvm_arch_flush_shadow(kvm);
872                 kfree(old_memslots);
873         }
874
875         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
876         if (r)
877                 goto out_free;
878
879         /* map the pages in iommu page table */
880         if (npages) {
881                 r = kvm_iommu_map_pages(kvm, &new);
882                 if (r)
883                         goto out_free;
884         }
885
886         r = -ENOMEM;
887         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
888                         GFP_KERNEL);
889         if (!slots)
890                 goto out_free;
891
892         /* actual memory is freed via old in kvm_free_physmem_slot below */
893         if (!npages) {
894                 new.rmap = NULL;
895                 new.dirty_bitmap = NULL;
896                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
897                         new.lpage_info[i] = NULL;
898         }
899
900         update_memslots(slots, &new);
901         old_memslots = kvm->memslots;
902         rcu_assign_pointer(kvm->memslots, slots);
903         synchronize_srcu_expedited(&kvm->srcu);
904
905         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
906
907         /*
908          * If the new memory slot is created, we need to clear all
909          * mmio sptes.
910          */
911         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
912                 kvm_arch_flush_shadow(kvm);
913
914         kvm_free_physmem_slot(&old, &new);
915         kfree(old_memslots);
916
917         return 0;
918
919 out_free:
920         kvm_free_physmem_slot(&new, &old);
921 out:
922         return r;
923
924 }
925 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
926
927 int kvm_set_memory_region(struct kvm *kvm,
928                           struct kvm_userspace_memory_region *mem,
929                           int user_alloc)
930 {
931         int r;
932
933         mutex_lock(&kvm->slots_lock);
934         r = __kvm_set_memory_region(kvm, mem, user_alloc);
935         mutex_unlock(&kvm->slots_lock);
936         return r;
937 }
938 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
939
940 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
941                                    struct
942                                    kvm_userspace_memory_region *mem,
943                                    int user_alloc)
944 {
945         if (mem->slot >= KVM_MEMORY_SLOTS)
946                 return -EINVAL;
947         return kvm_set_memory_region(kvm, mem, user_alloc);
948 }
949
950 int kvm_get_dirty_log(struct kvm *kvm,
951                         struct kvm_dirty_log *log, int *is_dirty)
952 {
953         struct kvm_memory_slot *memslot;
954         int r, i;
955         unsigned long n;
956         unsigned long any = 0;
957
958         r = -EINVAL;
959         if (log->slot >= KVM_MEMORY_SLOTS)
960                 goto out;
961
962         memslot = id_to_memslot(kvm->memslots, log->slot);
963         r = -ENOENT;
964         if (!memslot->dirty_bitmap)
965                 goto out;
966
967         n = kvm_dirty_bitmap_bytes(memslot);
968
969         for (i = 0; !any && i < n/sizeof(long); ++i)
970                 any = memslot->dirty_bitmap[i];
971
972         r = -EFAULT;
973         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
974                 goto out;
975
976         if (any)
977                 *is_dirty = 1;
978
979         r = 0;
980 out:
981         return r;
982 }
983
984 void kvm_disable_largepages(void)
985 {
986         largepages_enabled = false;
987 }
988 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
989
990 int is_error_page(struct page *page)
991 {
992         return page == bad_page || page == hwpoison_page || page == fault_page;
993 }
994 EXPORT_SYMBOL_GPL(is_error_page);
995
996 int is_error_pfn(pfn_t pfn)
997 {
998         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
999 }
1000 EXPORT_SYMBOL_GPL(is_error_pfn);
1001
1002 int is_hwpoison_pfn(pfn_t pfn)
1003 {
1004         return pfn == hwpoison_pfn;
1005 }
1006 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
1007
1008 int is_fault_pfn(pfn_t pfn)
1009 {
1010         return pfn == fault_pfn;
1011 }
1012 EXPORT_SYMBOL_GPL(is_fault_pfn);
1013
1014 int is_noslot_pfn(pfn_t pfn)
1015 {
1016         return pfn == bad_pfn;
1017 }
1018 EXPORT_SYMBOL_GPL(is_noslot_pfn);
1019
1020 int is_invalid_pfn(pfn_t pfn)
1021 {
1022         return pfn == hwpoison_pfn || pfn == fault_pfn;
1023 }
1024 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1025
1026 static inline unsigned long bad_hva(void)
1027 {
1028         return PAGE_OFFSET;
1029 }
1030
1031 int kvm_is_error_hva(unsigned long addr)
1032 {
1033         return addr == bad_hva();
1034 }
1035 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1036
1037 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1038 {
1039         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1040 }
1041 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1042
1043 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1044 {
1045         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1046
1047         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1048               memslot->flags & KVM_MEMSLOT_INVALID)
1049                 return 0;
1050
1051         return 1;
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1054
1055 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1056 {
1057         struct vm_area_struct *vma;
1058         unsigned long addr, size;
1059
1060         size = PAGE_SIZE;
1061
1062         addr = gfn_to_hva(kvm, gfn);
1063         if (kvm_is_error_hva(addr))
1064                 return PAGE_SIZE;
1065
1066         down_read(&current->mm->mmap_sem);
1067         vma = find_vma(current->mm, addr);
1068         if (!vma)
1069                 goto out;
1070
1071         size = vma_kernel_pagesize(vma);
1072
1073 out:
1074         up_read(&current->mm->mmap_sem);
1075
1076         return size;
1077 }
1078
1079 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1080                                      gfn_t *nr_pages)
1081 {
1082         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1083                 return bad_hva();
1084
1085         if (nr_pages)
1086                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1087
1088         return gfn_to_hva_memslot(slot, gfn);
1089 }
1090
1091 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1092 {
1093         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1094 }
1095 EXPORT_SYMBOL_GPL(gfn_to_hva);
1096
1097 static pfn_t get_fault_pfn(void)
1098 {
1099         get_page(fault_page);
1100         return fault_pfn;
1101 }
1102
1103 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1104         unsigned long start, int write, struct page **page)
1105 {
1106         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1107
1108         if (write)
1109                 flags |= FOLL_WRITE;
1110
1111         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1112 }
1113
1114 static inline int check_user_page_hwpoison(unsigned long addr)
1115 {
1116         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1117
1118         rc = __get_user_pages(current, current->mm, addr, 1,
1119                               flags, NULL, NULL, NULL);
1120         return rc == -EHWPOISON;
1121 }
1122
1123 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1124                         bool *async, bool write_fault, bool *writable)
1125 {
1126         struct page *page[1];
1127         int npages = 0;
1128         pfn_t pfn;
1129
1130         /* we can do it either atomically or asynchronously, not both */
1131         BUG_ON(atomic && async);
1132
1133         BUG_ON(!write_fault && !writable);
1134
1135         if (writable)
1136                 *writable = true;
1137
1138         if (atomic || async)
1139                 npages = __get_user_pages_fast(addr, 1, 1, page);
1140
1141         if (unlikely(npages != 1) && !atomic) {
1142                 might_sleep();
1143
1144                 if (writable)
1145                         *writable = write_fault;
1146
1147                 if (async) {
1148                         down_read(&current->mm->mmap_sem);
1149                         npages = get_user_page_nowait(current, current->mm,
1150                                                      addr, write_fault, page);
1151                         up_read(&current->mm->mmap_sem);
1152                 } else
1153                         npages = get_user_pages_fast(addr, 1, write_fault,
1154                                                      page);
1155
1156                 /* map read fault as writable if possible */
1157                 if (unlikely(!write_fault) && npages == 1) {
1158                         struct page *wpage[1];
1159
1160                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1161                         if (npages == 1) {
1162                                 *writable = true;
1163                                 put_page(page[0]);
1164                                 page[0] = wpage[0];
1165                         }
1166                         npages = 1;
1167                 }
1168         }
1169
1170         if (unlikely(npages != 1)) {
1171                 struct vm_area_struct *vma;
1172
1173                 if (atomic)
1174                         return get_fault_pfn();
1175
1176                 down_read(&current->mm->mmap_sem);
1177                 if (npages == -EHWPOISON ||
1178                         (!async && check_user_page_hwpoison(addr))) {
1179                         up_read(&current->mm->mmap_sem);
1180                         get_page(hwpoison_page);
1181                         return page_to_pfn(hwpoison_page);
1182                 }
1183
1184                 vma = find_vma_intersection(current->mm, addr, addr+1);
1185
1186                 if (vma == NULL)
1187                         pfn = get_fault_pfn();
1188                 else if ((vma->vm_flags & VM_PFNMAP)) {
1189                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1190                                 vma->vm_pgoff;
1191                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1192                 } else {
1193                         if (async && (vma->vm_flags & VM_WRITE))
1194                                 *async = true;
1195                         pfn = get_fault_pfn();
1196                 }
1197                 up_read(&current->mm->mmap_sem);
1198         } else
1199                 pfn = page_to_pfn(page[0]);
1200
1201         return pfn;
1202 }
1203
1204 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1205 {
1206         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1207 }
1208 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1209
1210 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1211                           bool write_fault, bool *writable)
1212 {
1213         unsigned long addr;
1214
1215         if (async)
1216                 *async = false;
1217
1218         addr = gfn_to_hva(kvm, gfn);
1219         if (kvm_is_error_hva(addr)) {
1220                 get_page(bad_page);
1221                 return page_to_pfn(bad_page);
1222         }
1223
1224         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1225 }
1226
1227 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1228 {
1229         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1230 }
1231 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1232
1233 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1234                        bool write_fault, bool *writable)
1235 {
1236         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1237 }
1238 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1239
1240 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1241 {
1242         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1243 }
1244 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1245
1246 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1247                       bool *writable)
1248 {
1249         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1250 }
1251 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1252
1253 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1254                          struct kvm_memory_slot *slot, gfn_t gfn)
1255 {
1256         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1257         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1258 }
1259
1260 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1261                                                                   int nr_pages)
1262 {
1263         unsigned long addr;
1264         gfn_t entry;
1265
1266         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1267         if (kvm_is_error_hva(addr))
1268                 return -1;
1269
1270         if (entry < nr_pages)
1271                 return 0;
1272
1273         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1274 }
1275 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1276
1277 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1278 {
1279         pfn_t pfn;
1280
1281         pfn = gfn_to_pfn(kvm, gfn);
1282         if (!kvm_is_mmio_pfn(pfn))
1283                 return pfn_to_page(pfn);
1284
1285         WARN_ON(kvm_is_mmio_pfn(pfn));
1286
1287         get_page(bad_page);
1288         return bad_page;
1289 }
1290
1291 EXPORT_SYMBOL_GPL(gfn_to_page);
1292
1293 void kvm_release_page_clean(struct page *page)
1294 {
1295         kvm_release_pfn_clean(page_to_pfn(page));
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1298
1299 void kvm_release_pfn_clean(pfn_t pfn)
1300 {
1301         if (!kvm_is_mmio_pfn(pfn))
1302                 put_page(pfn_to_page(pfn));
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1305
1306 void kvm_release_page_dirty(struct page *page)
1307 {
1308         kvm_release_pfn_dirty(page_to_pfn(page));
1309 }
1310 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1311
1312 void kvm_release_pfn_dirty(pfn_t pfn)
1313 {
1314         kvm_set_pfn_dirty(pfn);
1315         kvm_release_pfn_clean(pfn);
1316 }
1317 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1318
1319 void kvm_set_page_dirty(struct page *page)
1320 {
1321         kvm_set_pfn_dirty(page_to_pfn(page));
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1324
1325 void kvm_set_pfn_dirty(pfn_t pfn)
1326 {
1327         if (!kvm_is_mmio_pfn(pfn)) {
1328                 struct page *page = pfn_to_page(pfn);
1329                 if (!PageReserved(page))
1330                         SetPageDirty(page);
1331         }
1332 }
1333 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1334
1335 void kvm_set_pfn_accessed(pfn_t pfn)
1336 {
1337         if (!kvm_is_mmio_pfn(pfn))
1338                 mark_page_accessed(pfn_to_page(pfn));
1339 }
1340 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1341
1342 void kvm_get_pfn(pfn_t pfn)
1343 {
1344         if (!kvm_is_mmio_pfn(pfn))
1345                 get_page(pfn_to_page(pfn));
1346 }
1347 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1348
1349 static int next_segment(unsigned long len, int offset)
1350 {
1351         if (len > PAGE_SIZE - offset)
1352                 return PAGE_SIZE - offset;
1353         else
1354                 return len;
1355 }
1356
1357 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1358                         int len)
1359 {
1360         int r;
1361         unsigned long addr;
1362
1363         addr = gfn_to_hva(kvm, gfn);
1364         if (kvm_is_error_hva(addr))
1365                 return -EFAULT;
1366         r = __copy_from_user(data, (void __user *)addr + offset, len);
1367         if (r)
1368                 return -EFAULT;
1369         return 0;
1370 }
1371 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1372
1373 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1374 {
1375         gfn_t gfn = gpa >> PAGE_SHIFT;
1376         int seg;
1377         int offset = offset_in_page(gpa);
1378         int ret;
1379
1380         while ((seg = next_segment(len, offset)) != 0) {
1381                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1382                 if (ret < 0)
1383                         return ret;
1384                 offset = 0;
1385                 len -= seg;
1386                 data += seg;
1387                 ++gfn;
1388         }
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_read_guest);
1392
1393 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1394                           unsigned long len)
1395 {
1396         int r;
1397         unsigned long addr;
1398         gfn_t gfn = gpa >> PAGE_SHIFT;
1399         int offset = offset_in_page(gpa);
1400
1401         addr = gfn_to_hva(kvm, gfn);
1402         if (kvm_is_error_hva(addr))
1403                 return -EFAULT;
1404         pagefault_disable();
1405         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1406         pagefault_enable();
1407         if (r)
1408                 return -EFAULT;
1409         return 0;
1410 }
1411 EXPORT_SYMBOL(kvm_read_guest_atomic);
1412
1413 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1414                          int offset, int len)
1415 {
1416         int r;
1417         unsigned long addr;
1418
1419         addr = gfn_to_hva(kvm, gfn);
1420         if (kvm_is_error_hva(addr))
1421                 return -EFAULT;
1422         r = __copy_to_user((void __user *)addr + offset, data, len);
1423         if (r)
1424                 return -EFAULT;
1425         mark_page_dirty(kvm, gfn);
1426         return 0;
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1429
1430 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1431                     unsigned long len)
1432 {
1433         gfn_t gfn = gpa >> PAGE_SHIFT;
1434         int seg;
1435         int offset = offset_in_page(gpa);
1436         int ret;
1437
1438         while ((seg = next_segment(len, offset)) != 0) {
1439                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1440                 if (ret < 0)
1441                         return ret;
1442                 offset = 0;
1443                 len -= seg;
1444                 data += seg;
1445                 ++gfn;
1446         }
1447         return 0;
1448 }
1449
1450 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1451                               gpa_t gpa)
1452 {
1453         struct kvm_memslots *slots = kvm_memslots(kvm);
1454         int offset = offset_in_page(gpa);
1455         gfn_t gfn = gpa >> PAGE_SHIFT;
1456
1457         ghc->gpa = gpa;
1458         ghc->generation = slots->generation;
1459         ghc->memslot = gfn_to_memslot(kvm, gfn);
1460         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1461         if (!kvm_is_error_hva(ghc->hva))
1462                 ghc->hva += offset;
1463         else
1464                 return -EFAULT;
1465
1466         return 0;
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1469
1470 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1471                            void *data, unsigned long len)
1472 {
1473         struct kvm_memslots *slots = kvm_memslots(kvm);
1474         int r;
1475
1476         if (slots->generation != ghc->generation)
1477                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1478
1479         if (kvm_is_error_hva(ghc->hva))
1480                 return -EFAULT;
1481
1482         r = __copy_to_user((void __user *)ghc->hva, data, len);
1483         if (r)
1484                 return -EFAULT;
1485         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1486
1487         return 0;
1488 }
1489 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1490
1491 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1492                            void *data, unsigned long len)
1493 {
1494         struct kvm_memslots *slots = kvm_memslots(kvm);
1495         int r;
1496
1497         if (slots->generation != ghc->generation)
1498                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1499
1500         if (kvm_is_error_hva(ghc->hva))
1501                 return -EFAULT;
1502
1503         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1504         if (r)
1505                 return -EFAULT;
1506
1507         return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1510
1511 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1512 {
1513         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1514                                     offset, len);
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1517
1518 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1519 {
1520         gfn_t gfn = gpa >> PAGE_SHIFT;
1521         int seg;
1522         int offset = offset_in_page(gpa);
1523         int ret;
1524
1525         while ((seg = next_segment(len, offset)) != 0) {
1526                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1527                 if (ret < 0)
1528                         return ret;
1529                 offset = 0;
1530                 len -= seg;
1531                 ++gfn;
1532         }
1533         return 0;
1534 }
1535 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1536
1537 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1538                              gfn_t gfn)
1539 {
1540         if (memslot && memslot->dirty_bitmap) {
1541                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1542
1543                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1544                         memslot->nr_dirty_pages++;
1545         }
1546 }
1547
1548 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1549 {
1550         struct kvm_memory_slot *memslot;
1551
1552         memslot = gfn_to_memslot(kvm, gfn);
1553         mark_page_dirty_in_slot(kvm, memslot, gfn);
1554 }
1555
1556 /*
1557  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1558  */
1559 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1560 {
1561         DEFINE_WAIT(wait);
1562
1563         for (;;) {
1564                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1565
1566                 if (kvm_arch_vcpu_runnable(vcpu)) {
1567                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1568                         break;
1569                 }
1570                 if (kvm_cpu_has_pending_timer(vcpu))
1571                         break;
1572                 if (signal_pending(current))
1573                         break;
1574
1575                 schedule();
1576         }
1577
1578         finish_wait(&vcpu->wq, &wait);
1579 }
1580
1581 void kvm_resched(struct kvm_vcpu *vcpu)
1582 {
1583         if (!need_resched())
1584                 return;
1585         cond_resched();
1586 }
1587 EXPORT_SYMBOL_GPL(kvm_resched);
1588
1589 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1590 {
1591         struct kvm *kvm = me->kvm;
1592         struct kvm_vcpu *vcpu;
1593         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1594         int yielded = 0;
1595         int pass;
1596         int i;
1597
1598         /*
1599          * We boost the priority of a VCPU that is runnable but not
1600          * currently running, because it got preempted by something
1601          * else and called schedule in __vcpu_run.  Hopefully that
1602          * VCPU is holding the lock that we need and will release it.
1603          * We approximate round-robin by starting at the last boosted VCPU.
1604          */
1605         for (pass = 0; pass < 2 && !yielded; pass++) {
1606                 kvm_for_each_vcpu(i, vcpu, kvm) {
1607                         struct task_struct *task = NULL;
1608                         struct pid *pid;
1609                         if (!pass && i < last_boosted_vcpu) {
1610                                 i = last_boosted_vcpu;
1611                                 continue;
1612                         } else if (pass && i > last_boosted_vcpu)
1613                                 break;
1614                         if (vcpu == me)
1615                                 continue;
1616                         if (waitqueue_active(&vcpu->wq))
1617                                 continue;
1618                         rcu_read_lock();
1619                         pid = rcu_dereference(vcpu->pid);
1620                         if (pid)
1621                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1622                         rcu_read_unlock();
1623                         if (!task)
1624                                 continue;
1625                         if (task->flags & PF_VCPU) {
1626                                 put_task_struct(task);
1627                                 continue;
1628                         }
1629                         if (yield_to(task, 1)) {
1630                                 put_task_struct(task);
1631                                 kvm->last_boosted_vcpu = i;
1632                                 yielded = 1;
1633                                 break;
1634                         }
1635                         put_task_struct(task);
1636                 }
1637         }
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1640
1641 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1642 {
1643         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1644         struct page *page;
1645
1646         if (vmf->pgoff == 0)
1647                 page = virt_to_page(vcpu->run);
1648 #ifdef CONFIG_X86
1649         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1650                 page = virt_to_page(vcpu->arch.pio_data);
1651 #endif
1652 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1653         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1654                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1655 #endif
1656         else
1657                 return kvm_arch_vcpu_fault(vcpu, vmf);
1658         get_page(page);
1659         vmf->page = page;
1660         return 0;
1661 }
1662
1663 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1664         .fault = kvm_vcpu_fault,
1665 };
1666
1667 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1668 {
1669         vma->vm_ops = &kvm_vcpu_vm_ops;
1670         return 0;
1671 }
1672
1673 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1674 {
1675         struct kvm_vcpu *vcpu = filp->private_data;
1676
1677         kvm_put_kvm(vcpu->kvm);
1678         return 0;
1679 }
1680
1681 static struct file_operations kvm_vcpu_fops = {
1682         .release        = kvm_vcpu_release,
1683         .unlocked_ioctl = kvm_vcpu_ioctl,
1684 #ifdef CONFIG_COMPAT
1685         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1686 #endif
1687         .mmap           = kvm_vcpu_mmap,
1688         .llseek         = noop_llseek,
1689 };
1690
1691 /*
1692  * Allocates an inode for the vcpu.
1693  */
1694 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1695 {
1696         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1697 }
1698
1699 /*
1700  * Creates some virtual cpus.  Good luck creating more than one.
1701  */
1702 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1703 {
1704         int r;
1705         struct kvm_vcpu *vcpu, *v;
1706
1707         vcpu = kvm_arch_vcpu_create(kvm, id);
1708         if (IS_ERR(vcpu))
1709                 return PTR_ERR(vcpu);
1710
1711         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1712
1713         r = kvm_arch_vcpu_setup(vcpu);
1714         if (r)
1715                 goto vcpu_destroy;
1716
1717         mutex_lock(&kvm->lock);
1718         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1719                 r = -EINVAL;
1720                 goto unlock_vcpu_destroy;
1721         }
1722
1723         kvm_for_each_vcpu(r, v, kvm)
1724                 if (v->vcpu_id == id) {
1725                         r = -EEXIST;
1726                         goto unlock_vcpu_destroy;
1727                 }
1728
1729         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1730
1731         /* Now it's all set up, let userspace reach it */
1732         kvm_get_kvm(kvm);
1733         r = create_vcpu_fd(vcpu);
1734         if (r < 0) {
1735                 kvm_put_kvm(kvm);
1736                 goto unlock_vcpu_destroy;
1737         }
1738
1739         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1740         smp_wmb();
1741         atomic_inc(&kvm->online_vcpus);
1742
1743         mutex_unlock(&kvm->lock);
1744         return r;
1745
1746 unlock_vcpu_destroy:
1747         mutex_unlock(&kvm->lock);
1748 vcpu_destroy:
1749         kvm_arch_vcpu_destroy(vcpu);
1750         return r;
1751 }
1752
1753 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1754 {
1755         if (sigset) {
1756                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1757                 vcpu->sigset_active = 1;
1758                 vcpu->sigset = *sigset;
1759         } else
1760                 vcpu->sigset_active = 0;
1761         return 0;
1762 }
1763
1764 static long kvm_vcpu_ioctl(struct file *filp,
1765                            unsigned int ioctl, unsigned long arg)
1766 {
1767         struct kvm_vcpu *vcpu = filp->private_data;
1768         void __user *argp = (void __user *)arg;
1769         int r;
1770         struct kvm_fpu *fpu = NULL;
1771         struct kvm_sregs *kvm_sregs = NULL;
1772
1773         if (vcpu->kvm->mm != current->mm)
1774                 return -EIO;
1775
1776 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1777         /*
1778          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1779          * so vcpu_load() would break it.
1780          */
1781         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1782                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1783 #endif
1784
1785
1786         vcpu_load(vcpu);
1787         switch (ioctl) {
1788         case KVM_RUN:
1789                 r = -EINVAL;
1790                 if (arg)
1791                         goto out;
1792                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1793                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1794                 break;
1795         case KVM_GET_REGS: {
1796                 struct kvm_regs *kvm_regs;
1797
1798                 r = -ENOMEM;
1799                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1800                 if (!kvm_regs)
1801                         goto out;
1802                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1803                 if (r)
1804                         goto out_free1;
1805                 r = -EFAULT;
1806                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1807                         goto out_free1;
1808                 r = 0;
1809 out_free1:
1810                 kfree(kvm_regs);
1811                 break;
1812         }
1813         case KVM_SET_REGS: {
1814                 struct kvm_regs *kvm_regs;
1815
1816                 r = -ENOMEM;
1817                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1818                 if (IS_ERR(kvm_regs)) {
1819                         r = PTR_ERR(kvm_regs);
1820                         goto out;
1821                 }
1822                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1823                 if (r)
1824                         goto out_free2;
1825                 r = 0;
1826 out_free2:
1827                 kfree(kvm_regs);
1828                 break;
1829         }
1830         case KVM_GET_SREGS: {
1831                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1832                 r = -ENOMEM;
1833                 if (!kvm_sregs)
1834                         goto out;
1835                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1836                 if (r)
1837                         goto out;
1838                 r = -EFAULT;
1839                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1840                         goto out;
1841                 r = 0;
1842                 break;
1843         }
1844         case KVM_SET_SREGS: {
1845                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1846                 if (IS_ERR(kvm_sregs)) {
1847                         r = PTR_ERR(kvm_sregs);
1848                         goto out;
1849                 }
1850                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1851                 if (r)
1852                         goto out;
1853                 r = 0;
1854                 break;
1855         }
1856         case KVM_GET_MP_STATE: {
1857                 struct kvm_mp_state mp_state;
1858
1859                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1860                 if (r)
1861                         goto out;
1862                 r = -EFAULT;
1863                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1864                         goto out;
1865                 r = 0;
1866                 break;
1867         }
1868         case KVM_SET_MP_STATE: {
1869                 struct kvm_mp_state mp_state;
1870
1871                 r = -EFAULT;
1872                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1873                         goto out;
1874                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1875                 if (r)
1876                         goto out;
1877                 r = 0;
1878                 break;
1879         }
1880         case KVM_TRANSLATE: {
1881                 struct kvm_translation tr;
1882
1883                 r = -EFAULT;
1884                 if (copy_from_user(&tr, argp, sizeof tr))
1885                         goto out;
1886                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1887                 if (r)
1888                         goto out;
1889                 r = -EFAULT;
1890                 if (copy_to_user(argp, &tr, sizeof tr))
1891                         goto out;
1892                 r = 0;
1893                 break;
1894         }
1895         case KVM_SET_GUEST_DEBUG: {
1896                 struct kvm_guest_debug dbg;
1897
1898                 r = -EFAULT;
1899                 if (copy_from_user(&dbg, argp, sizeof dbg))
1900                         goto out;
1901                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1902                 if (r)
1903                         goto out;
1904                 r = 0;
1905                 break;
1906         }
1907         case KVM_SET_SIGNAL_MASK: {
1908                 struct kvm_signal_mask __user *sigmask_arg = argp;
1909                 struct kvm_signal_mask kvm_sigmask;
1910                 sigset_t sigset, *p;
1911
1912                 p = NULL;
1913                 if (argp) {
1914                         r = -EFAULT;
1915                         if (copy_from_user(&kvm_sigmask, argp,
1916                                            sizeof kvm_sigmask))
1917                                 goto out;
1918                         r = -EINVAL;
1919                         if (kvm_sigmask.len != sizeof sigset)
1920                                 goto out;
1921                         r = -EFAULT;
1922                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1923                                            sizeof sigset))
1924                                 goto out;
1925                         p = &sigset;
1926                 }
1927                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1928                 break;
1929         }
1930         case KVM_GET_FPU: {
1931                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1932                 r = -ENOMEM;
1933                 if (!fpu)
1934                         goto out;
1935                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1936                 if (r)
1937                         goto out;
1938                 r = -EFAULT;
1939                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1940                         goto out;
1941                 r = 0;
1942                 break;
1943         }
1944         case KVM_SET_FPU: {
1945                 fpu = memdup_user(argp, sizeof(*fpu));
1946                 if (IS_ERR(fpu)) {
1947                         r = PTR_ERR(fpu);
1948                         goto out;
1949                 }
1950                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1951                 if (r)
1952                         goto out;
1953                 r = 0;
1954                 break;
1955         }
1956         default:
1957                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1958         }
1959 out:
1960         vcpu_put(vcpu);
1961         kfree(fpu);
1962         kfree(kvm_sregs);
1963         return r;
1964 }
1965
1966 #ifdef CONFIG_COMPAT
1967 static long kvm_vcpu_compat_ioctl(struct file *filp,
1968                                   unsigned int ioctl, unsigned long arg)
1969 {
1970         struct kvm_vcpu *vcpu = filp->private_data;
1971         void __user *argp = compat_ptr(arg);
1972         int r;
1973
1974         if (vcpu->kvm->mm != current->mm)
1975                 return -EIO;
1976
1977         switch (ioctl) {
1978         case KVM_SET_SIGNAL_MASK: {
1979                 struct kvm_signal_mask __user *sigmask_arg = argp;
1980                 struct kvm_signal_mask kvm_sigmask;
1981                 compat_sigset_t csigset;
1982                 sigset_t sigset;
1983
1984                 if (argp) {
1985                         r = -EFAULT;
1986                         if (copy_from_user(&kvm_sigmask, argp,
1987                                            sizeof kvm_sigmask))
1988                                 goto out;
1989                         r = -EINVAL;
1990                         if (kvm_sigmask.len != sizeof csigset)
1991                                 goto out;
1992                         r = -EFAULT;
1993                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1994                                            sizeof csigset))
1995                                 goto out;
1996                 }
1997                 sigset_from_compat(&sigset, &csigset);
1998                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1999                 break;
2000         }
2001         default:
2002                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2003         }
2004
2005 out:
2006         return r;
2007 }
2008 #endif
2009
2010 static long kvm_vm_ioctl(struct file *filp,
2011                            unsigned int ioctl, unsigned long arg)
2012 {
2013         struct kvm *kvm = filp->private_data;
2014         void __user *argp = (void __user *)arg;
2015         int r;
2016
2017         if (kvm->mm != current->mm)
2018                 return -EIO;
2019         switch (ioctl) {
2020         case KVM_CREATE_VCPU:
2021                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2022                 if (r < 0)
2023                         goto out;
2024                 break;
2025         case KVM_SET_USER_MEMORY_REGION: {
2026                 struct kvm_userspace_memory_region kvm_userspace_mem;
2027
2028                 r = -EFAULT;
2029                 if (copy_from_user(&kvm_userspace_mem, argp,
2030                                                 sizeof kvm_userspace_mem))
2031                         goto out;
2032
2033                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2034                 if (r)
2035                         goto out;
2036                 break;
2037         }
2038         case KVM_GET_DIRTY_LOG: {
2039                 struct kvm_dirty_log log;
2040
2041                 r = -EFAULT;
2042                 if (copy_from_user(&log, argp, sizeof log))
2043                         goto out;
2044                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2045                 if (r)
2046                         goto out;
2047                 break;
2048         }
2049 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2050         case KVM_REGISTER_COALESCED_MMIO: {
2051                 struct kvm_coalesced_mmio_zone zone;
2052                 r = -EFAULT;
2053                 if (copy_from_user(&zone, argp, sizeof zone))
2054                         goto out;
2055                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2056                 if (r)
2057                         goto out;
2058                 r = 0;
2059                 break;
2060         }
2061         case KVM_UNREGISTER_COALESCED_MMIO: {
2062                 struct kvm_coalesced_mmio_zone zone;
2063                 r = -EFAULT;
2064                 if (copy_from_user(&zone, argp, sizeof zone))
2065                         goto out;
2066                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2067                 if (r)
2068                         goto out;
2069                 r = 0;
2070                 break;
2071         }
2072 #endif
2073         case KVM_IRQFD: {
2074                 struct kvm_irqfd data;
2075
2076                 r = -EFAULT;
2077                 if (copy_from_user(&data, argp, sizeof data))
2078                         goto out;
2079                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2080                 break;
2081         }
2082         case KVM_IOEVENTFD: {
2083                 struct kvm_ioeventfd data;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&data, argp, sizeof data))
2087                         goto out;
2088                 r = kvm_ioeventfd(kvm, &data);
2089                 break;
2090         }
2091 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2092         case KVM_SET_BOOT_CPU_ID:
2093                 r = 0;
2094                 mutex_lock(&kvm->lock);
2095                 if (atomic_read(&kvm->online_vcpus) != 0)
2096                         r = -EBUSY;
2097                 else
2098                         kvm->bsp_vcpu_id = arg;
2099                 mutex_unlock(&kvm->lock);
2100                 break;
2101 #endif
2102         default:
2103                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2104                 if (r == -ENOTTY)
2105                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2106         }
2107 out:
2108         return r;
2109 }
2110
2111 #ifdef CONFIG_COMPAT
2112 struct compat_kvm_dirty_log {
2113         __u32 slot;
2114         __u32 padding1;
2115         union {
2116                 compat_uptr_t dirty_bitmap; /* one bit per page */
2117                 __u64 padding2;
2118         };
2119 };
2120
2121 static long kvm_vm_compat_ioctl(struct file *filp,
2122                            unsigned int ioctl, unsigned long arg)
2123 {
2124         struct kvm *kvm = filp->private_data;
2125         int r;
2126
2127         if (kvm->mm != current->mm)
2128                 return -EIO;
2129         switch (ioctl) {
2130         case KVM_GET_DIRTY_LOG: {
2131                 struct compat_kvm_dirty_log compat_log;
2132                 struct kvm_dirty_log log;
2133
2134                 r = -EFAULT;
2135                 if (copy_from_user(&compat_log, (void __user *)arg,
2136                                    sizeof(compat_log)))
2137                         goto out;
2138                 log.slot         = compat_log.slot;
2139                 log.padding1     = compat_log.padding1;
2140                 log.padding2     = compat_log.padding2;
2141                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2142
2143                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2144                 if (r)
2145                         goto out;
2146                 break;
2147         }
2148         default:
2149                 r = kvm_vm_ioctl(filp, ioctl, arg);
2150         }
2151
2152 out:
2153         return r;
2154 }
2155 #endif
2156
2157 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2158 {
2159         struct page *page[1];
2160         unsigned long addr;
2161         int npages;
2162         gfn_t gfn = vmf->pgoff;
2163         struct kvm *kvm = vma->vm_file->private_data;
2164
2165         addr = gfn_to_hva(kvm, gfn);
2166         if (kvm_is_error_hva(addr))
2167                 return VM_FAULT_SIGBUS;
2168
2169         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2170                                 NULL);
2171         if (unlikely(npages != 1))
2172                 return VM_FAULT_SIGBUS;
2173
2174         vmf->page = page[0];
2175         return 0;
2176 }
2177
2178 static const struct vm_operations_struct kvm_vm_vm_ops = {
2179         .fault = kvm_vm_fault,
2180 };
2181
2182 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2183 {
2184         vma->vm_ops = &kvm_vm_vm_ops;
2185         return 0;
2186 }
2187
2188 static struct file_operations kvm_vm_fops = {
2189         .release        = kvm_vm_release,
2190         .unlocked_ioctl = kvm_vm_ioctl,
2191 #ifdef CONFIG_COMPAT
2192         .compat_ioctl   = kvm_vm_compat_ioctl,
2193 #endif
2194         .mmap           = kvm_vm_mmap,
2195         .llseek         = noop_llseek,
2196 };
2197
2198 static int kvm_dev_ioctl_create_vm(unsigned long type)
2199 {
2200         int r;
2201         struct kvm *kvm;
2202
2203         kvm = kvm_create_vm(type);
2204         if (IS_ERR(kvm))
2205                 return PTR_ERR(kvm);
2206 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2207         r = kvm_coalesced_mmio_init(kvm);
2208         if (r < 0) {
2209                 kvm_put_kvm(kvm);
2210                 return r;
2211         }
2212 #endif
2213         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2214         if (r < 0)
2215                 kvm_put_kvm(kvm);
2216
2217         return r;
2218 }
2219
2220 static long kvm_dev_ioctl_check_extension_generic(long arg)
2221 {
2222         switch (arg) {
2223         case KVM_CAP_USER_MEMORY:
2224         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2225         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2226 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2227         case KVM_CAP_SET_BOOT_CPU_ID:
2228 #endif
2229         case KVM_CAP_INTERNAL_ERROR_DATA:
2230                 return 1;
2231 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2232         case KVM_CAP_IRQ_ROUTING:
2233                 return KVM_MAX_IRQ_ROUTES;
2234 #endif
2235         default:
2236                 break;
2237         }
2238         return kvm_dev_ioctl_check_extension(arg);
2239 }
2240
2241 static long kvm_dev_ioctl(struct file *filp,
2242                           unsigned int ioctl, unsigned long arg)
2243 {
2244         long r = -EINVAL;
2245
2246         switch (ioctl) {
2247         case KVM_GET_API_VERSION:
2248                 r = -EINVAL;
2249                 if (arg)
2250                         goto out;
2251                 r = KVM_API_VERSION;
2252                 break;
2253         case KVM_CREATE_VM:
2254                 r = kvm_dev_ioctl_create_vm(arg);
2255                 break;
2256         case KVM_CHECK_EXTENSION:
2257                 r = kvm_dev_ioctl_check_extension_generic(arg);
2258                 break;
2259         case KVM_GET_VCPU_MMAP_SIZE:
2260                 r = -EINVAL;
2261                 if (arg)
2262                         goto out;
2263                 r = PAGE_SIZE;     /* struct kvm_run */
2264 #ifdef CONFIG_X86
2265                 r += PAGE_SIZE;    /* pio data page */
2266 #endif
2267 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2268                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2269 #endif
2270                 break;
2271         case KVM_TRACE_ENABLE:
2272         case KVM_TRACE_PAUSE:
2273         case KVM_TRACE_DISABLE:
2274                 r = -EOPNOTSUPP;
2275                 break;
2276         default:
2277                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2278         }
2279 out:
2280         return r;
2281 }
2282
2283 static struct file_operations kvm_chardev_ops = {
2284         .unlocked_ioctl = kvm_dev_ioctl,
2285         .compat_ioctl   = kvm_dev_ioctl,
2286         .llseek         = noop_llseek,
2287 };
2288
2289 static struct miscdevice kvm_dev = {
2290         KVM_MINOR,
2291         "kvm",
2292         &kvm_chardev_ops,
2293 };
2294
2295 static void hardware_enable_nolock(void *junk)
2296 {
2297         int cpu = raw_smp_processor_id();
2298         int r;
2299
2300         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2301                 return;
2302
2303         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2304
2305         r = kvm_arch_hardware_enable(NULL);
2306
2307         if (r) {
2308                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2309                 atomic_inc(&hardware_enable_failed);
2310                 printk(KERN_INFO "kvm: enabling virtualization on "
2311                                  "CPU%d failed\n", cpu);
2312         }
2313 }
2314
2315 static void hardware_enable(void *junk)
2316 {
2317         raw_spin_lock(&kvm_lock);
2318         hardware_enable_nolock(junk);
2319         raw_spin_unlock(&kvm_lock);
2320 }
2321
2322 static void hardware_disable_nolock(void *junk)
2323 {
2324         int cpu = raw_smp_processor_id();
2325
2326         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2327                 return;
2328         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2329         kvm_arch_hardware_disable(NULL);
2330 }
2331
2332 static void hardware_disable(void *junk)
2333 {
2334         raw_spin_lock(&kvm_lock);
2335         hardware_disable_nolock(junk);
2336         raw_spin_unlock(&kvm_lock);
2337 }
2338
2339 static void hardware_disable_all_nolock(void)
2340 {
2341         BUG_ON(!kvm_usage_count);
2342
2343         kvm_usage_count--;
2344         if (!kvm_usage_count)
2345                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2346 }
2347
2348 static void hardware_disable_all(void)
2349 {
2350         raw_spin_lock(&kvm_lock);
2351         hardware_disable_all_nolock();
2352         raw_spin_unlock(&kvm_lock);
2353 }
2354
2355 static int hardware_enable_all(void)
2356 {
2357         int r = 0;
2358
2359         raw_spin_lock(&kvm_lock);
2360
2361         kvm_usage_count++;
2362         if (kvm_usage_count == 1) {
2363                 atomic_set(&hardware_enable_failed, 0);
2364                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2365
2366                 if (atomic_read(&hardware_enable_failed)) {
2367                         hardware_disable_all_nolock();
2368                         r = -EBUSY;
2369                 }
2370         }
2371
2372         raw_spin_unlock(&kvm_lock);
2373
2374         return r;
2375 }
2376
2377 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2378                            void *v)
2379 {
2380         int cpu = (long)v;
2381
2382         if (!kvm_usage_count)
2383                 return NOTIFY_OK;
2384
2385         val &= ~CPU_TASKS_FROZEN;
2386         switch (val) {
2387         case CPU_DYING:
2388                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2389                        cpu);
2390                 hardware_disable(NULL);
2391                 break;
2392         case CPU_STARTING:
2393                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2394                        cpu);
2395                 hardware_enable(NULL);
2396                 break;
2397         }
2398         return NOTIFY_OK;
2399 }
2400
2401
2402 asmlinkage void kvm_spurious_fault(void)
2403 {
2404         /* Fault while not rebooting.  We want the trace. */
2405         BUG();
2406 }
2407 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2408
2409 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2410                       void *v)
2411 {
2412         /*
2413          * Some (well, at least mine) BIOSes hang on reboot if
2414          * in vmx root mode.
2415          *
2416          * And Intel TXT required VMX off for all cpu when system shutdown.
2417          */
2418         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2419         kvm_rebooting = true;
2420         on_each_cpu(hardware_disable_nolock, NULL, 1);
2421         return NOTIFY_OK;
2422 }
2423
2424 static struct notifier_block kvm_reboot_notifier = {
2425         .notifier_call = kvm_reboot,
2426         .priority = 0,
2427 };
2428
2429 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2430 {
2431         int i;
2432
2433         for (i = 0; i < bus->dev_count; i++) {
2434                 struct kvm_io_device *pos = bus->range[i].dev;
2435
2436                 kvm_iodevice_destructor(pos);
2437         }
2438         kfree(bus);
2439 }
2440
2441 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2442 {
2443         const struct kvm_io_range *r1 = p1;
2444         const struct kvm_io_range *r2 = p2;
2445
2446         if (r1->addr < r2->addr)
2447                 return -1;
2448         if (r1->addr + r1->len > r2->addr + r2->len)
2449                 return 1;
2450         return 0;
2451 }
2452
2453 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2454                           gpa_t addr, int len)
2455 {
2456         if (bus->dev_count == NR_IOBUS_DEVS)
2457                 return -ENOSPC;
2458
2459         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2460                 .addr = addr,
2461                 .len = len,
2462                 .dev = dev,
2463         };
2464
2465         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2466                 kvm_io_bus_sort_cmp, NULL);
2467
2468         return 0;
2469 }
2470
2471 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2472                              gpa_t addr, int len)
2473 {
2474         struct kvm_io_range *range, key;
2475         int off;
2476
2477         key = (struct kvm_io_range) {
2478                 .addr = addr,
2479                 .len = len,
2480         };
2481
2482         range = bsearch(&key, bus->range, bus->dev_count,
2483                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2484         if (range == NULL)
2485                 return -ENOENT;
2486
2487         off = range - bus->range;
2488
2489         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2490                 off--;
2491
2492         return off;
2493 }
2494
2495 /* kvm_io_bus_write - called under kvm->slots_lock */
2496 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2497                      int len, const void *val)
2498 {
2499         int idx;
2500         struct kvm_io_bus *bus;
2501         struct kvm_io_range range;
2502
2503         range = (struct kvm_io_range) {
2504                 .addr = addr,
2505                 .len = len,
2506         };
2507
2508         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2509         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2510         if (idx < 0)
2511                 return -EOPNOTSUPP;
2512
2513         while (idx < bus->dev_count &&
2514                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2515                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2516                         return 0;
2517                 idx++;
2518         }
2519
2520         return -EOPNOTSUPP;
2521 }
2522
2523 /* kvm_io_bus_read - called under kvm->slots_lock */
2524 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2525                     int len, void *val)
2526 {
2527         int idx;
2528         struct kvm_io_bus *bus;
2529         struct kvm_io_range range;
2530
2531         range = (struct kvm_io_range) {
2532                 .addr = addr,
2533                 .len = len,
2534         };
2535
2536         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2537         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2538         if (idx < 0)
2539                 return -EOPNOTSUPP;
2540
2541         while (idx < bus->dev_count &&
2542                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2543                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2544                         return 0;
2545                 idx++;
2546         }
2547
2548         return -EOPNOTSUPP;
2549 }
2550
2551 /* Caller must hold slots_lock. */
2552 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2553                             int len, struct kvm_io_device *dev)
2554 {
2555         struct kvm_io_bus *new_bus, *bus;
2556
2557         bus = kvm->buses[bus_idx];
2558         if (bus->dev_count > NR_IOBUS_DEVS-1)
2559                 return -ENOSPC;
2560
2561         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2562         if (!new_bus)
2563                 return -ENOMEM;
2564         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2565         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2566         synchronize_srcu_expedited(&kvm->srcu);
2567         kfree(bus);
2568
2569         return 0;
2570 }
2571
2572 /* Caller must hold slots_lock. */
2573 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2574                               struct kvm_io_device *dev)
2575 {
2576         int i, r;
2577         struct kvm_io_bus *new_bus, *bus;
2578
2579         bus = kvm->buses[bus_idx];
2580
2581         new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2582         if (!new_bus)
2583                 return -ENOMEM;
2584
2585         r = -ENOENT;
2586         for (i = 0; i < new_bus->dev_count; i++)
2587                 if (new_bus->range[i].dev == dev) {
2588                         r = 0;
2589                         new_bus->dev_count--;
2590                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2591                         sort(new_bus->range, new_bus->dev_count,
2592                              sizeof(struct kvm_io_range),
2593                              kvm_io_bus_sort_cmp, NULL);
2594                         break;
2595                 }
2596
2597         if (r) {
2598                 kfree(new_bus);
2599                 return r;
2600         }
2601
2602         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2603         synchronize_srcu_expedited(&kvm->srcu);
2604         kfree(bus);
2605         return r;
2606 }
2607
2608 static struct notifier_block kvm_cpu_notifier = {
2609         .notifier_call = kvm_cpu_hotplug,
2610 };
2611
2612 static int vm_stat_get(void *_offset, u64 *val)
2613 {
2614         unsigned offset = (long)_offset;
2615         struct kvm *kvm;
2616
2617         *val = 0;
2618         raw_spin_lock(&kvm_lock);
2619         list_for_each_entry(kvm, &vm_list, vm_list)
2620                 *val += *(u32 *)((void *)kvm + offset);
2621         raw_spin_unlock(&kvm_lock);
2622         return 0;
2623 }
2624
2625 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2626
2627 static int vcpu_stat_get(void *_offset, u64 *val)
2628 {
2629         unsigned offset = (long)_offset;
2630         struct kvm *kvm;
2631         struct kvm_vcpu *vcpu;
2632         int i;
2633
2634         *val = 0;
2635         raw_spin_lock(&kvm_lock);
2636         list_for_each_entry(kvm, &vm_list, vm_list)
2637                 kvm_for_each_vcpu(i, vcpu, kvm)
2638                         *val += *(u32 *)((void *)vcpu + offset);
2639
2640         raw_spin_unlock(&kvm_lock);
2641         return 0;
2642 }
2643
2644 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2645
2646 static const struct file_operations *stat_fops[] = {
2647         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2648         [KVM_STAT_VM]   = &vm_stat_fops,
2649 };
2650
2651 static int kvm_init_debug(void)
2652 {
2653         int r = -EFAULT;
2654         struct kvm_stats_debugfs_item *p;
2655
2656         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2657         if (kvm_debugfs_dir == NULL)
2658                 goto out;
2659
2660         for (p = debugfs_entries; p->name; ++p) {
2661                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2662                                                 (void *)(long)p->offset,
2663                                                 stat_fops[p->kind]);
2664                 if (p->dentry == NULL)
2665                         goto out_dir;
2666         }
2667
2668         return 0;
2669
2670 out_dir:
2671         debugfs_remove_recursive(kvm_debugfs_dir);
2672 out:
2673         return r;
2674 }
2675
2676 static void kvm_exit_debug(void)
2677 {
2678         struct kvm_stats_debugfs_item *p;
2679
2680         for (p = debugfs_entries; p->name; ++p)
2681                 debugfs_remove(p->dentry);
2682         debugfs_remove(kvm_debugfs_dir);
2683 }
2684
2685 static int kvm_suspend(void)
2686 {
2687         if (kvm_usage_count)
2688                 hardware_disable_nolock(NULL);
2689         return 0;
2690 }
2691
2692 static void kvm_resume(void)
2693 {
2694         if (kvm_usage_count) {
2695                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2696                 hardware_enable_nolock(NULL);
2697         }
2698 }
2699
2700 static struct syscore_ops kvm_syscore_ops = {
2701         .suspend = kvm_suspend,
2702         .resume = kvm_resume,
2703 };
2704
2705 struct page *bad_page;
2706 pfn_t bad_pfn;
2707
2708 static inline
2709 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2710 {
2711         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2712 }
2713
2714 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2715 {
2716         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2717
2718         kvm_arch_vcpu_load(vcpu, cpu);
2719 }
2720
2721 static void kvm_sched_out(struct preempt_notifier *pn,
2722                           struct task_struct *next)
2723 {
2724         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2725
2726         kvm_arch_vcpu_put(vcpu);
2727 }
2728
2729 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2730                   struct module *module)
2731 {
2732         int r;
2733         int cpu;
2734
2735         r = kvm_arch_init(opaque);
2736         if (r)
2737                 goto out_fail;
2738
2739         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2740
2741         if (bad_page == NULL) {
2742                 r = -ENOMEM;
2743                 goto out;
2744         }
2745
2746         bad_pfn = page_to_pfn(bad_page);
2747
2748         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2749
2750         if (hwpoison_page == NULL) {
2751                 r = -ENOMEM;
2752                 goto out_free_0;
2753         }
2754
2755         hwpoison_pfn = page_to_pfn(hwpoison_page);
2756
2757         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2758
2759         if (fault_page == NULL) {
2760                 r = -ENOMEM;
2761                 goto out_free_0;
2762         }
2763
2764         fault_pfn = page_to_pfn(fault_page);
2765
2766         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2767                 r = -ENOMEM;
2768                 goto out_free_0;
2769         }
2770
2771         r = kvm_arch_hardware_setup();
2772         if (r < 0)
2773                 goto out_free_0a;
2774
2775         for_each_online_cpu(cpu) {
2776                 smp_call_function_single(cpu,
2777                                 kvm_arch_check_processor_compat,
2778                                 &r, 1);
2779                 if (r < 0)
2780                         goto out_free_1;
2781         }
2782
2783         r = register_cpu_notifier(&kvm_cpu_notifier);
2784         if (r)
2785                 goto out_free_2;
2786         register_reboot_notifier(&kvm_reboot_notifier);
2787
2788         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2789         if (!vcpu_align)
2790                 vcpu_align = __alignof__(struct kvm_vcpu);
2791         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2792                                            0, NULL);
2793         if (!kvm_vcpu_cache) {
2794                 r = -ENOMEM;
2795                 goto out_free_3;
2796         }
2797
2798         r = kvm_async_pf_init();
2799         if (r)
2800                 goto out_free;
2801
2802         kvm_chardev_ops.owner = module;
2803         kvm_vm_fops.owner = module;
2804         kvm_vcpu_fops.owner = module;
2805
2806         r = misc_register(&kvm_dev);
2807         if (r) {
2808                 printk(KERN_ERR "kvm: misc device register failed\n");
2809                 goto out_unreg;
2810         }
2811
2812         register_syscore_ops(&kvm_syscore_ops);
2813
2814         kvm_preempt_ops.sched_in = kvm_sched_in;
2815         kvm_preempt_ops.sched_out = kvm_sched_out;
2816
2817         r = kvm_init_debug();
2818         if (r) {
2819                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2820                 goto out_undebugfs;
2821         }
2822
2823         return 0;
2824
2825 out_undebugfs:
2826         unregister_syscore_ops(&kvm_syscore_ops);
2827 out_unreg:
2828         kvm_async_pf_deinit();
2829 out_free:
2830         kmem_cache_destroy(kvm_vcpu_cache);
2831 out_free_3:
2832         unregister_reboot_notifier(&kvm_reboot_notifier);
2833         unregister_cpu_notifier(&kvm_cpu_notifier);
2834 out_free_2:
2835 out_free_1:
2836         kvm_arch_hardware_unsetup();
2837 out_free_0a:
2838         free_cpumask_var(cpus_hardware_enabled);
2839 out_free_0:
2840         if (fault_page)
2841                 __free_page(fault_page);
2842         if (hwpoison_page)
2843                 __free_page(hwpoison_page);
2844         __free_page(bad_page);
2845 out:
2846         kvm_arch_exit();
2847 out_fail:
2848         return r;
2849 }
2850 EXPORT_SYMBOL_GPL(kvm_init);
2851
2852 void kvm_exit(void)
2853 {
2854         kvm_exit_debug();
2855         misc_deregister(&kvm_dev);
2856         kmem_cache_destroy(kvm_vcpu_cache);
2857         kvm_async_pf_deinit();
2858         unregister_syscore_ops(&kvm_syscore_ops);
2859         unregister_reboot_notifier(&kvm_reboot_notifier);
2860         unregister_cpu_notifier(&kvm_cpu_notifier);
2861         on_each_cpu(hardware_disable_nolock, NULL, 1);
2862         kvm_arch_hardware_unsetup();
2863         kvm_arch_exit();
2864         free_cpumask_var(cpus_hardware_enabled);
2865         __free_page(hwpoison_page);
2866         __free_page(bad_page);
2867 }
2868 EXPORT_SYMBOL_GPL(kvm_exit);