KVM: Introduce kvm_unmap_hva_range() for kvm_mmu_notifier_invalidate_range_start()
[linux-3.10.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 static struct page *hwpoison_page;
104 static pfn_t hwpoison_pfn;
105
106 struct page *fault_page;
107 pfn_t fault_pfn;
108
109 inline int kvm_is_mmio_pfn(pfn_t pfn)
110 {
111         if (pfn_valid(pfn)) {
112                 int reserved;
113                 struct page *tail = pfn_to_page(pfn);
114                 struct page *head = compound_trans_head(tail);
115                 reserved = PageReserved(head);
116                 if (head != tail) {
117                         /*
118                          * "head" is not a dangling pointer
119                          * (compound_trans_head takes care of that)
120                          * but the hugepage may have been splitted
121                          * from under us (and we may not hold a
122                          * reference count on the head page so it can
123                          * be reused before we run PageReferenced), so
124                          * we've to check PageTail before returning
125                          * what we just read.
126                          */
127                         smp_rmb();
128                         if (PageTail(tail))
129                                 return reserved;
130                 }
131                 return PageReserved(tail);
132         }
133
134         return true;
135 }
136
137 /*
138  * Switches to specified vcpu, until a matching vcpu_put()
139  */
140 void vcpu_load(struct kvm_vcpu *vcpu)
141 {
142         int cpu;
143
144         mutex_lock(&vcpu->mutex);
145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
146                 /* The thread running this VCPU changed. */
147                 struct pid *oldpid = vcpu->pid;
148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
149                 rcu_assign_pointer(vcpu->pid, newpid);
150                 synchronize_rcu();
151                 put_pid(oldpid);
152         }
153         cpu = get_cpu();
154         preempt_notifier_register(&vcpu->preempt_notifier);
155         kvm_arch_vcpu_load(vcpu, cpu);
156         put_cpu();
157 }
158
159 void vcpu_put(struct kvm_vcpu *vcpu)
160 {
161         preempt_disable();
162         kvm_arch_vcpu_put(vcpu);
163         preempt_notifier_unregister(&vcpu->preempt_notifier);
164         preempt_enable();
165         mutex_unlock(&vcpu->mutex);
166 }
167
168 static void ack_flush(void *_completed)
169 {
170 }
171
172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 {
174         int i, cpu, me;
175         cpumask_var_t cpus;
176         bool called = true;
177         struct kvm_vcpu *vcpu;
178
179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180
181         me = get_cpu();
182         kvm_for_each_vcpu(i, vcpu, kvm) {
183                 kvm_make_request(req, vcpu);
184                 cpu = vcpu->cpu;
185
186                 /* Set ->requests bit before we read ->mode */
187                 smp_mb();
188
189                 if (cpus != NULL && cpu != -1 && cpu != me &&
190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191                         cpumask_set_cpu(cpu, cpus);
192         }
193         if (unlikely(cpus == NULL))
194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195         else if (!cpumask_empty(cpus))
196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
197         else
198                 called = false;
199         put_cpu();
200         free_cpumask_var(cpus);
201         return called;
202 }
203
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
205 {
206         long 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
293         kvm->mmu_notifier_seq++;
294         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
295         /* we've to flush the tlb before the pages can be freed */
296         if (need_tlb_flush)
297                 kvm_flush_remote_tlbs(kvm);
298
299         spin_unlock(&kvm->mmu_lock);
300         srcu_read_unlock(&kvm->srcu, idx);
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         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
336         need_tlb_flush |= kvm->tlbs_dirty;
337         /* we've to flush the tlb before the pages can be freed */
338         if (need_tlb_flush)
339                 kvm_flush_remote_tlbs(kvm);
340
341         spin_unlock(&kvm->mmu_lock);
342         srcu_read_unlock(&kvm->srcu, idx);
343 }
344
345 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
346                                                   struct mm_struct *mm,
347                                                   unsigned long start,
348                                                   unsigned long end)
349 {
350         struct kvm *kvm = mmu_notifier_to_kvm(mn);
351
352         spin_lock(&kvm->mmu_lock);
353         /*
354          * This sequence increase will notify the kvm page fault that
355          * the page that is going to be mapped in the spte could have
356          * been freed.
357          */
358         kvm->mmu_notifier_seq++;
359         smp_wmb();
360         /*
361          * The above sequence increase must be visible before the
362          * below count decrease, which is ensured by the smp_wmb above
363          * in conjunction with the smp_rmb in mmu_notifier_retry().
364          */
365         kvm->mmu_notifier_count--;
366         spin_unlock(&kvm->mmu_lock);
367
368         BUG_ON(kvm->mmu_notifier_count < 0);
369 }
370
371 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
372                                               struct mm_struct *mm,
373                                               unsigned long address)
374 {
375         struct kvm *kvm = mmu_notifier_to_kvm(mn);
376         int young, idx;
377
378         idx = srcu_read_lock(&kvm->srcu);
379         spin_lock(&kvm->mmu_lock);
380
381         young = kvm_age_hva(kvm, address);
382         if (young)
383                 kvm_flush_remote_tlbs(kvm);
384
385         spin_unlock(&kvm->mmu_lock);
386         srcu_read_unlock(&kvm->srcu, idx);
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 /*
519  * Avoid using vmalloc for a small buffer.
520  * Should not be used when the size is statically known.
521  */
522 void *kvm_kvzalloc(unsigned long size)
523 {
524         if (size > PAGE_SIZE)
525                 return vzalloc(size);
526         else
527                 return kzalloc(size, GFP_KERNEL);
528 }
529
530 void kvm_kvfree(const void *addr)
531 {
532         if (is_vmalloc_addr(addr))
533                 vfree(addr);
534         else
535                 kfree(addr);
536 }
537
538 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
539 {
540         if (!memslot->dirty_bitmap)
541                 return;
542
543         kvm_kvfree(memslot->dirty_bitmap);
544         memslot->dirty_bitmap = NULL;
545 }
546
547 /*
548  * Free any memory in @free but not in @dont.
549  */
550 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
551                                   struct kvm_memory_slot *dont)
552 {
553         if (!dont || free->rmap != dont->rmap)
554                 vfree(free->rmap);
555
556         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
557                 kvm_destroy_dirty_bitmap(free);
558
559         kvm_arch_free_memslot(free, dont);
560
561         free->npages = 0;
562         free->rmap = NULL;
563 }
564
565 void kvm_free_physmem(struct kvm *kvm)
566 {
567         struct kvm_memslots *slots = kvm->memslots;
568         struct kvm_memory_slot *memslot;
569
570         kvm_for_each_memslot(memslot, slots)
571                 kvm_free_physmem_slot(memslot, NULL);
572
573         kfree(kvm->memslots);
574 }
575
576 static void kvm_destroy_vm(struct kvm *kvm)
577 {
578         int i;
579         struct mm_struct *mm = kvm->mm;
580
581         kvm_arch_sync_events(kvm);
582         raw_spin_lock(&kvm_lock);
583         list_del(&kvm->vm_list);
584         raw_spin_unlock(&kvm_lock);
585         kvm_free_irq_routing(kvm);
586         for (i = 0; i < KVM_NR_BUSES; i++)
587                 kvm_io_bus_destroy(kvm->buses[i]);
588         kvm_coalesced_mmio_free(kvm);
589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
590         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
591 #else
592         kvm_arch_flush_shadow(kvm);
593 #endif
594         kvm_arch_destroy_vm(kvm);
595         kvm_free_physmem(kvm);
596         cleanup_srcu_struct(&kvm->srcu);
597         kvm_arch_free_vm(kvm);
598         hardware_disable_all();
599         mmdrop(mm);
600 }
601
602 void kvm_get_kvm(struct kvm *kvm)
603 {
604         atomic_inc(&kvm->users_count);
605 }
606 EXPORT_SYMBOL_GPL(kvm_get_kvm);
607
608 void kvm_put_kvm(struct kvm *kvm)
609 {
610         if (atomic_dec_and_test(&kvm->users_count))
611                 kvm_destroy_vm(kvm);
612 }
613 EXPORT_SYMBOL_GPL(kvm_put_kvm);
614
615
616 static int kvm_vm_release(struct inode *inode, struct file *filp)
617 {
618         struct kvm *kvm = filp->private_data;
619
620         kvm_irqfd_release(kvm);
621
622         kvm_put_kvm(kvm);
623         return 0;
624 }
625
626 /*
627  * Allocation size is twice as large as the actual dirty bitmap size.
628  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
629  */
630 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
631 {
632 #ifndef CONFIG_S390
633         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
634
635         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
636         if (!memslot->dirty_bitmap)
637                 return -ENOMEM;
638
639 #endif /* !CONFIG_S390 */
640         return 0;
641 }
642
643 static int cmp_memslot(const void *slot1, const void *slot2)
644 {
645         struct kvm_memory_slot *s1, *s2;
646
647         s1 = (struct kvm_memory_slot *)slot1;
648         s2 = (struct kvm_memory_slot *)slot2;
649
650         if (s1->npages < s2->npages)
651                 return 1;
652         if (s1->npages > s2->npages)
653                 return -1;
654
655         return 0;
656 }
657
658 /*
659  * Sort the memslots base on its size, so the larger slots
660  * will get better fit.
661  */
662 static void sort_memslots(struct kvm_memslots *slots)
663 {
664         int i;
665
666         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
667               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
668
669         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
670                 slots->id_to_index[slots->memslots[i].id] = i;
671 }
672
673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
674 {
675         if (new) {
676                 int id = new->id;
677                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
678                 unsigned long npages = old->npages;
679
680                 *old = *new;
681                 if (new->npages != npages)
682                         sort_memslots(slots);
683         }
684
685         slots->generation++;
686 }
687
688 /*
689  * Allocate some memory and give it an address in the guest physical address
690  * space.
691  *
692  * Discontiguous memory is allowed, mostly for framebuffers.
693  *
694  * Must be called holding mmap_sem for write.
695  */
696 int __kvm_set_memory_region(struct kvm *kvm,
697                             struct kvm_userspace_memory_region *mem,
698                             int user_alloc)
699 {
700         int r;
701         gfn_t base_gfn;
702         unsigned long npages;
703         unsigned long i;
704         struct kvm_memory_slot *memslot;
705         struct kvm_memory_slot old, new;
706         struct kvm_memslots *slots, *old_memslots;
707
708         r = -EINVAL;
709         /* General sanity checks */
710         if (mem->memory_size & (PAGE_SIZE - 1))
711                 goto out;
712         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
713                 goto out;
714         /* We can read the guest memory with __xxx_user() later on. */
715         if (user_alloc &&
716             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
717              !access_ok(VERIFY_WRITE,
718                         (void __user *)(unsigned long)mem->userspace_addr,
719                         mem->memory_size)))
720                 goto out;
721         if (mem->slot >= KVM_MEM_SLOTS_NUM)
722                 goto out;
723         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
724                 goto out;
725
726         memslot = id_to_memslot(kvm->memslots, mem->slot);
727         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
728         npages = mem->memory_size >> PAGE_SHIFT;
729
730         r = -EINVAL;
731         if (npages > KVM_MEM_MAX_NR_PAGES)
732                 goto out;
733
734         if (!npages)
735                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
736
737         new = old = *memslot;
738
739         new.id = mem->slot;
740         new.base_gfn = base_gfn;
741         new.npages = npages;
742         new.flags = mem->flags;
743
744         /* Disallow changing a memory slot's size. */
745         r = -EINVAL;
746         if (npages && old.npages && npages != old.npages)
747                 goto out_free;
748
749         /* Check for overlaps */
750         r = -EEXIST;
751         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
752                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
753
754                 if (s == memslot || !s->npages)
755                         continue;
756                 if (!((base_gfn + npages <= s->base_gfn) ||
757                       (base_gfn >= s->base_gfn + s->npages)))
758                         goto out_free;
759         }
760
761         /* Free page dirty bitmap if unneeded */
762         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
763                 new.dirty_bitmap = NULL;
764
765         r = -ENOMEM;
766
767         /* Allocate if a slot is being created */
768         if (npages && !old.npages) {
769                 new.user_alloc = user_alloc;
770                 new.userspace_addr = mem->userspace_addr;
771 #ifndef CONFIG_S390
772                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
773                 if (!new.rmap)
774                         goto out_free;
775 #endif /* not defined CONFIG_S390 */
776                 if (kvm_arch_create_memslot(&new, npages))
777                         goto out_free;
778         }
779
780         /* Allocate page dirty bitmap if needed */
781         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
782                 if (kvm_create_dirty_bitmap(&new) < 0)
783                         goto out_free;
784                 /* destroy any largepage mappings for dirty tracking */
785         }
786
787         if (!npages) {
788                 struct kvm_memory_slot *slot;
789
790                 r = -ENOMEM;
791                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
792                                 GFP_KERNEL);
793                 if (!slots)
794                         goto out_free;
795                 slot = id_to_memslot(slots, mem->slot);
796                 slot->flags |= KVM_MEMSLOT_INVALID;
797
798                 update_memslots(slots, NULL);
799
800                 old_memslots = kvm->memslots;
801                 rcu_assign_pointer(kvm->memslots, slots);
802                 synchronize_srcu_expedited(&kvm->srcu);
803                 /* From this point no new shadow pages pointing to a deleted
804                  * memslot will be created.
805                  *
806                  * validation of sp->gfn happens in:
807                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
808                  *      - kvm_is_visible_gfn (mmu_check_roots)
809                  */
810                 kvm_arch_flush_shadow(kvm);
811                 kfree(old_memslots);
812         }
813
814         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
815         if (r)
816                 goto out_free;
817
818         /* map/unmap the pages in iommu page table */
819         if (npages) {
820                 r = kvm_iommu_map_pages(kvm, &new);
821                 if (r)
822                         goto out_free;
823         } else
824                 kvm_iommu_unmap_pages(kvm, &old);
825
826         r = -ENOMEM;
827         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
828                         GFP_KERNEL);
829         if (!slots)
830                 goto out_free;
831
832         /* actual memory is freed via old in kvm_free_physmem_slot below */
833         if (!npages) {
834                 new.rmap = NULL;
835                 new.dirty_bitmap = NULL;
836                 memset(&new.arch, 0, sizeof(new.arch));
837         }
838
839         update_memslots(slots, &new);
840         old_memslots = kvm->memslots;
841         rcu_assign_pointer(kvm->memslots, slots);
842         synchronize_srcu_expedited(&kvm->srcu);
843
844         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
845
846         /*
847          * If the new memory slot is created, we need to clear all
848          * mmio sptes.
849          */
850         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
851                 kvm_arch_flush_shadow(kvm);
852
853         kvm_free_physmem_slot(&old, &new);
854         kfree(old_memslots);
855
856         return 0;
857
858 out_free:
859         kvm_free_physmem_slot(&new, &old);
860 out:
861         return r;
862
863 }
864 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
865
866 int kvm_set_memory_region(struct kvm *kvm,
867                           struct kvm_userspace_memory_region *mem,
868                           int user_alloc)
869 {
870         int r;
871
872         mutex_lock(&kvm->slots_lock);
873         r = __kvm_set_memory_region(kvm, mem, user_alloc);
874         mutex_unlock(&kvm->slots_lock);
875         return r;
876 }
877 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
878
879 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
880                                    struct
881                                    kvm_userspace_memory_region *mem,
882                                    int user_alloc)
883 {
884         if (mem->slot >= KVM_MEMORY_SLOTS)
885                 return -EINVAL;
886         return kvm_set_memory_region(kvm, mem, user_alloc);
887 }
888
889 int kvm_get_dirty_log(struct kvm *kvm,
890                         struct kvm_dirty_log *log, int *is_dirty)
891 {
892         struct kvm_memory_slot *memslot;
893         int r, i;
894         unsigned long n;
895         unsigned long any = 0;
896
897         r = -EINVAL;
898         if (log->slot >= KVM_MEMORY_SLOTS)
899                 goto out;
900
901         memslot = id_to_memslot(kvm->memslots, log->slot);
902         r = -ENOENT;
903         if (!memslot->dirty_bitmap)
904                 goto out;
905
906         n = kvm_dirty_bitmap_bytes(memslot);
907
908         for (i = 0; !any && i < n/sizeof(long); ++i)
909                 any = memslot->dirty_bitmap[i];
910
911         r = -EFAULT;
912         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
913                 goto out;
914
915         if (any)
916                 *is_dirty = 1;
917
918         r = 0;
919 out:
920         return r;
921 }
922
923 bool kvm_largepages_enabled(void)
924 {
925         return largepages_enabled;
926 }
927
928 void kvm_disable_largepages(void)
929 {
930         largepages_enabled = false;
931 }
932 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
933
934 int is_error_page(struct page *page)
935 {
936         return page == bad_page || page == hwpoison_page || page == fault_page;
937 }
938 EXPORT_SYMBOL_GPL(is_error_page);
939
940 int is_error_pfn(pfn_t pfn)
941 {
942         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
943 }
944 EXPORT_SYMBOL_GPL(is_error_pfn);
945
946 int is_hwpoison_pfn(pfn_t pfn)
947 {
948         return pfn == hwpoison_pfn;
949 }
950 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
951
952 int is_fault_pfn(pfn_t pfn)
953 {
954         return pfn == fault_pfn;
955 }
956 EXPORT_SYMBOL_GPL(is_fault_pfn);
957
958 int is_noslot_pfn(pfn_t pfn)
959 {
960         return pfn == bad_pfn;
961 }
962 EXPORT_SYMBOL_GPL(is_noslot_pfn);
963
964 int is_invalid_pfn(pfn_t pfn)
965 {
966         return pfn == hwpoison_pfn || pfn == fault_pfn;
967 }
968 EXPORT_SYMBOL_GPL(is_invalid_pfn);
969
970 static inline unsigned long bad_hva(void)
971 {
972         return PAGE_OFFSET;
973 }
974
975 int kvm_is_error_hva(unsigned long addr)
976 {
977         return addr == bad_hva();
978 }
979 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
980
981 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
982 {
983         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
984 }
985 EXPORT_SYMBOL_GPL(gfn_to_memslot);
986
987 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
988 {
989         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
990
991         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
992               memslot->flags & KVM_MEMSLOT_INVALID)
993                 return 0;
994
995         return 1;
996 }
997 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
998
999 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1000 {
1001         struct vm_area_struct *vma;
1002         unsigned long addr, size;
1003
1004         size = PAGE_SIZE;
1005
1006         addr = gfn_to_hva(kvm, gfn);
1007         if (kvm_is_error_hva(addr))
1008                 return PAGE_SIZE;
1009
1010         down_read(&current->mm->mmap_sem);
1011         vma = find_vma(current->mm, addr);
1012         if (!vma)
1013                 goto out;
1014
1015         size = vma_kernel_pagesize(vma);
1016
1017 out:
1018         up_read(&current->mm->mmap_sem);
1019
1020         return size;
1021 }
1022
1023 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1024                                      gfn_t *nr_pages)
1025 {
1026         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1027                 return bad_hva();
1028
1029         if (nr_pages)
1030                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1031
1032         return gfn_to_hva_memslot(slot, gfn);
1033 }
1034
1035 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1036 {
1037         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1038 }
1039 EXPORT_SYMBOL_GPL(gfn_to_hva);
1040
1041 static pfn_t get_fault_pfn(void)
1042 {
1043         get_page(fault_page);
1044         return fault_pfn;
1045 }
1046
1047 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1048         unsigned long start, int write, struct page **page)
1049 {
1050         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1051
1052         if (write)
1053                 flags |= FOLL_WRITE;
1054
1055         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1056 }
1057
1058 static inline int check_user_page_hwpoison(unsigned long addr)
1059 {
1060         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1061
1062         rc = __get_user_pages(current, current->mm, addr, 1,
1063                               flags, NULL, NULL, NULL);
1064         return rc == -EHWPOISON;
1065 }
1066
1067 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1068                         bool *async, bool write_fault, bool *writable)
1069 {
1070         struct page *page[1];
1071         int npages = 0;
1072         pfn_t pfn;
1073
1074         /* we can do it either atomically or asynchronously, not both */
1075         BUG_ON(atomic && async);
1076
1077         BUG_ON(!write_fault && !writable);
1078
1079         if (writable)
1080                 *writable = true;
1081
1082         if (atomic || async)
1083                 npages = __get_user_pages_fast(addr, 1, 1, page);
1084
1085         if (unlikely(npages != 1) && !atomic) {
1086                 might_sleep();
1087
1088                 if (writable)
1089                         *writable = write_fault;
1090
1091                 if (async) {
1092                         down_read(&current->mm->mmap_sem);
1093                         npages = get_user_page_nowait(current, current->mm,
1094                                                      addr, write_fault, page);
1095                         up_read(&current->mm->mmap_sem);
1096                 } else
1097                         npages = get_user_pages_fast(addr, 1, write_fault,
1098                                                      page);
1099
1100                 /* map read fault as writable if possible */
1101                 if (unlikely(!write_fault) && npages == 1) {
1102                         struct page *wpage[1];
1103
1104                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1105                         if (npages == 1) {
1106                                 *writable = true;
1107                                 put_page(page[0]);
1108                                 page[0] = wpage[0];
1109                         }
1110                         npages = 1;
1111                 }
1112         }
1113
1114         if (unlikely(npages != 1)) {
1115                 struct vm_area_struct *vma;
1116
1117                 if (atomic)
1118                         return get_fault_pfn();
1119
1120                 down_read(&current->mm->mmap_sem);
1121                 if (npages == -EHWPOISON ||
1122                         (!async && check_user_page_hwpoison(addr))) {
1123                         up_read(&current->mm->mmap_sem);
1124                         get_page(hwpoison_page);
1125                         return page_to_pfn(hwpoison_page);
1126                 }
1127
1128                 vma = find_vma_intersection(current->mm, addr, addr+1);
1129
1130                 if (vma == NULL)
1131                         pfn = get_fault_pfn();
1132                 else if ((vma->vm_flags & VM_PFNMAP)) {
1133                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1134                                 vma->vm_pgoff;
1135                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1136                 } else {
1137                         if (async && (vma->vm_flags & VM_WRITE))
1138                                 *async = true;
1139                         pfn = get_fault_pfn();
1140                 }
1141                 up_read(&current->mm->mmap_sem);
1142         } else
1143                 pfn = page_to_pfn(page[0]);
1144
1145         return pfn;
1146 }
1147
1148 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1149 {
1150         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1151 }
1152 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1153
1154 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1155                           bool write_fault, bool *writable)
1156 {
1157         unsigned long addr;
1158
1159         if (async)
1160                 *async = false;
1161
1162         addr = gfn_to_hva(kvm, gfn);
1163         if (kvm_is_error_hva(addr)) {
1164                 get_page(bad_page);
1165                 return page_to_pfn(bad_page);
1166         }
1167
1168         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1169 }
1170
1171 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1172 {
1173         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1174 }
1175 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1176
1177 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1178                        bool write_fault, bool *writable)
1179 {
1180         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1181 }
1182 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1183
1184 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1185 {
1186         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1187 }
1188 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1189
1190 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1191                       bool *writable)
1192 {
1193         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1194 }
1195 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1196
1197 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1198                          struct kvm_memory_slot *slot, gfn_t gfn)
1199 {
1200         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1201         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1202 }
1203
1204 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1205                                                                   int nr_pages)
1206 {
1207         unsigned long addr;
1208         gfn_t entry;
1209
1210         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1211         if (kvm_is_error_hva(addr))
1212                 return -1;
1213
1214         if (entry < nr_pages)
1215                 return 0;
1216
1217         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1218 }
1219 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1220
1221 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1222 {
1223         pfn_t pfn;
1224
1225         pfn = gfn_to_pfn(kvm, gfn);
1226         if (!kvm_is_mmio_pfn(pfn))
1227                 return pfn_to_page(pfn);
1228
1229         WARN_ON(kvm_is_mmio_pfn(pfn));
1230
1231         get_page(bad_page);
1232         return bad_page;
1233 }
1234
1235 EXPORT_SYMBOL_GPL(gfn_to_page);
1236
1237 void kvm_release_page_clean(struct page *page)
1238 {
1239         kvm_release_pfn_clean(page_to_pfn(page));
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1242
1243 void kvm_release_pfn_clean(pfn_t pfn)
1244 {
1245         if (!kvm_is_mmio_pfn(pfn))
1246                 put_page(pfn_to_page(pfn));
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1249
1250 void kvm_release_page_dirty(struct page *page)
1251 {
1252         kvm_release_pfn_dirty(page_to_pfn(page));
1253 }
1254 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1255
1256 void kvm_release_pfn_dirty(pfn_t pfn)
1257 {
1258         kvm_set_pfn_dirty(pfn);
1259         kvm_release_pfn_clean(pfn);
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1262
1263 void kvm_set_page_dirty(struct page *page)
1264 {
1265         kvm_set_pfn_dirty(page_to_pfn(page));
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1268
1269 void kvm_set_pfn_dirty(pfn_t pfn)
1270 {
1271         if (!kvm_is_mmio_pfn(pfn)) {
1272                 struct page *page = pfn_to_page(pfn);
1273                 if (!PageReserved(page))
1274                         SetPageDirty(page);
1275         }
1276 }
1277 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1278
1279 void kvm_set_pfn_accessed(pfn_t pfn)
1280 {
1281         if (!kvm_is_mmio_pfn(pfn))
1282                 mark_page_accessed(pfn_to_page(pfn));
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1285
1286 void kvm_get_pfn(pfn_t pfn)
1287 {
1288         if (!kvm_is_mmio_pfn(pfn))
1289                 get_page(pfn_to_page(pfn));
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1292
1293 static int next_segment(unsigned long len, int offset)
1294 {
1295         if (len > PAGE_SIZE - offset)
1296                 return PAGE_SIZE - offset;
1297         else
1298                 return len;
1299 }
1300
1301 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1302                         int len)
1303 {
1304         int r;
1305         unsigned long addr;
1306
1307         addr = gfn_to_hva(kvm, gfn);
1308         if (kvm_is_error_hva(addr))
1309                 return -EFAULT;
1310         r = __copy_from_user(data, (void __user *)addr + offset, len);
1311         if (r)
1312                 return -EFAULT;
1313         return 0;
1314 }
1315 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1316
1317 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1318 {
1319         gfn_t gfn = gpa >> PAGE_SHIFT;
1320         int seg;
1321         int offset = offset_in_page(gpa);
1322         int ret;
1323
1324         while ((seg = next_segment(len, offset)) != 0) {
1325                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1326                 if (ret < 0)
1327                         return ret;
1328                 offset = 0;
1329                 len -= seg;
1330                 data += seg;
1331                 ++gfn;
1332         }
1333         return 0;
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_read_guest);
1336
1337 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1338                           unsigned long len)
1339 {
1340         int r;
1341         unsigned long addr;
1342         gfn_t gfn = gpa >> PAGE_SHIFT;
1343         int offset = offset_in_page(gpa);
1344
1345         addr = gfn_to_hva(kvm, gfn);
1346         if (kvm_is_error_hva(addr))
1347                 return -EFAULT;
1348         pagefault_disable();
1349         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1350         pagefault_enable();
1351         if (r)
1352                 return -EFAULT;
1353         return 0;
1354 }
1355 EXPORT_SYMBOL(kvm_read_guest_atomic);
1356
1357 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1358                          int offset, 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_to_user((void __user *)addr + offset, data, len);
1367         if (r)
1368                 return -EFAULT;
1369         mark_page_dirty(kvm, gfn);
1370         return 0;
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1373
1374 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1375                     unsigned long len)
1376 {
1377         gfn_t gfn = gpa >> PAGE_SHIFT;
1378         int seg;
1379         int offset = offset_in_page(gpa);
1380         int ret;
1381
1382         while ((seg = next_segment(len, offset)) != 0) {
1383                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1384                 if (ret < 0)
1385                         return ret;
1386                 offset = 0;
1387                 len -= seg;
1388                 data += seg;
1389                 ++gfn;
1390         }
1391         return 0;
1392 }
1393
1394 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1395                               gpa_t gpa)
1396 {
1397         struct kvm_memslots *slots = kvm_memslots(kvm);
1398         int offset = offset_in_page(gpa);
1399         gfn_t gfn = gpa >> PAGE_SHIFT;
1400
1401         ghc->gpa = gpa;
1402         ghc->generation = slots->generation;
1403         ghc->memslot = gfn_to_memslot(kvm, gfn);
1404         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1405         if (!kvm_is_error_hva(ghc->hva))
1406                 ghc->hva += offset;
1407         else
1408                 return -EFAULT;
1409
1410         return 0;
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1413
1414 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1415                            void *data, unsigned long len)
1416 {
1417         struct kvm_memslots *slots = kvm_memslots(kvm);
1418         int r;
1419
1420         if (slots->generation != ghc->generation)
1421                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1422
1423         if (kvm_is_error_hva(ghc->hva))
1424                 return -EFAULT;
1425
1426         r = __copy_to_user((void __user *)ghc->hva, data, len);
1427         if (r)
1428                 return -EFAULT;
1429         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1430
1431         return 0;
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1434
1435 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1436                            void *data, unsigned long len)
1437 {
1438         struct kvm_memslots *slots = kvm_memslots(kvm);
1439         int r;
1440
1441         if (slots->generation != ghc->generation)
1442                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1443
1444         if (kvm_is_error_hva(ghc->hva))
1445                 return -EFAULT;
1446
1447         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1448         if (r)
1449                 return -EFAULT;
1450
1451         return 0;
1452 }
1453 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1454
1455 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1456 {
1457         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1458                                     offset, len);
1459 }
1460 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1461
1462 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1463 {
1464         gfn_t gfn = gpa >> PAGE_SHIFT;
1465         int seg;
1466         int offset = offset_in_page(gpa);
1467         int ret;
1468
1469         while ((seg = next_segment(len, offset)) != 0) {
1470                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1471                 if (ret < 0)
1472                         return ret;
1473                 offset = 0;
1474                 len -= seg;
1475                 ++gfn;
1476         }
1477         return 0;
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1480
1481 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1482                              gfn_t gfn)
1483 {
1484         if (memslot && memslot->dirty_bitmap) {
1485                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1486
1487                 /* TODO: introduce set_bit_le() and use it */
1488                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1489         }
1490 }
1491
1492 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1493 {
1494         struct kvm_memory_slot *memslot;
1495
1496         memslot = gfn_to_memslot(kvm, gfn);
1497         mark_page_dirty_in_slot(kvm, memslot, gfn);
1498 }
1499
1500 /*
1501  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1502  */
1503 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1504 {
1505         DEFINE_WAIT(wait);
1506
1507         for (;;) {
1508                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1509
1510                 if (kvm_arch_vcpu_runnable(vcpu)) {
1511                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1512                         break;
1513                 }
1514                 if (kvm_cpu_has_pending_timer(vcpu))
1515                         break;
1516                 if (signal_pending(current))
1517                         break;
1518
1519                 schedule();
1520         }
1521
1522         finish_wait(&vcpu->wq, &wait);
1523 }
1524
1525 #ifndef CONFIG_S390
1526 /*
1527  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1528  */
1529 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1530 {
1531         int me;
1532         int cpu = vcpu->cpu;
1533         wait_queue_head_t *wqp;
1534
1535         wqp = kvm_arch_vcpu_wq(vcpu);
1536         if (waitqueue_active(wqp)) {
1537                 wake_up_interruptible(wqp);
1538                 ++vcpu->stat.halt_wakeup;
1539         }
1540
1541         me = get_cpu();
1542         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1543                 if (kvm_arch_vcpu_should_kick(vcpu))
1544                         smp_send_reschedule(cpu);
1545         put_cpu();
1546 }
1547 #endif /* !CONFIG_S390 */
1548
1549 void kvm_resched(struct kvm_vcpu *vcpu)
1550 {
1551         if (!need_resched())
1552                 return;
1553         cond_resched();
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_resched);
1556
1557 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1558 {
1559         struct pid *pid;
1560         struct task_struct *task = NULL;
1561
1562         rcu_read_lock();
1563         pid = rcu_dereference(target->pid);
1564         if (pid)
1565                 task = get_pid_task(target->pid, PIDTYPE_PID);
1566         rcu_read_unlock();
1567         if (!task)
1568                 return false;
1569         if (task->flags & PF_VCPU) {
1570                 put_task_struct(task);
1571                 return false;
1572         }
1573         if (yield_to(task, 1)) {
1574                 put_task_struct(task);
1575                 return true;
1576         }
1577         put_task_struct(task);
1578         return false;
1579 }
1580 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1581
1582 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1583 {
1584         struct kvm *kvm = me->kvm;
1585         struct kvm_vcpu *vcpu;
1586         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1587         int yielded = 0;
1588         int pass;
1589         int i;
1590
1591         /*
1592          * We boost the priority of a VCPU that is runnable but not
1593          * currently running, because it got preempted by something
1594          * else and called schedule in __vcpu_run.  Hopefully that
1595          * VCPU is holding the lock that we need and will release it.
1596          * We approximate round-robin by starting at the last boosted VCPU.
1597          */
1598         for (pass = 0; pass < 2 && !yielded; pass++) {
1599                 kvm_for_each_vcpu(i, vcpu, kvm) {
1600                         if (!pass && i <= last_boosted_vcpu) {
1601                                 i = last_boosted_vcpu;
1602                                 continue;
1603                         } else if (pass && i > last_boosted_vcpu)
1604                                 break;
1605                         if (vcpu == me)
1606                                 continue;
1607                         if (waitqueue_active(&vcpu->wq))
1608                                 continue;
1609                         if (kvm_vcpu_yield_to(vcpu)) {
1610                                 kvm->last_boosted_vcpu = i;
1611                                 yielded = 1;
1612                                 break;
1613                         }
1614                 }
1615         }
1616 }
1617 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1618
1619 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1620 {
1621         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1622         struct page *page;
1623
1624         if (vmf->pgoff == 0)
1625                 page = virt_to_page(vcpu->run);
1626 #ifdef CONFIG_X86
1627         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1628                 page = virt_to_page(vcpu->arch.pio_data);
1629 #endif
1630 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1631         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1632                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1633 #endif
1634         else
1635                 return kvm_arch_vcpu_fault(vcpu, vmf);
1636         get_page(page);
1637         vmf->page = page;
1638         return 0;
1639 }
1640
1641 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1642         .fault = kvm_vcpu_fault,
1643 };
1644
1645 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1646 {
1647         vma->vm_ops = &kvm_vcpu_vm_ops;
1648         return 0;
1649 }
1650
1651 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1652 {
1653         struct kvm_vcpu *vcpu = filp->private_data;
1654
1655         kvm_put_kvm(vcpu->kvm);
1656         return 0;
1657 }
1658
1659 static struct file_operations kvm_vcpu_fops = {
1660         .release        = kvm_vcpu_release,
1661         .unlocked_ioctl = kvm_vcpu_ioctl,
1662 #ifdef CONFIG_COMPAT
1663         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1664 #endif
1665         .mmap           = kvm_vcpu_mmap,
1666         .llseek         = noop_llseek,
1667 };
1668
1669 /*
1670  * Allocates an inode for the vcpu.
1671  */
1672 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1673 {
1674         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1675 }
1676
1677 /*
1678  * Creates some virtual cpus.  Good luck creating more than one.
1679  */
1680 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1681 {
1682         int r;
1683         struct kvm_vcpu *vcpu, *v;
1684
1685         vcpu = kvm_arch_vcpu_create(kvm, id);
1686         if (IS_ERR(vcpu))
1687                 return PTR_ERR(vcpu);
1688
1689         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1690
1691         r = kvm_arch_vcpu_setup(vcpu);
1692         if (r)
1693                 goto vcpu_destroy;
1694
1695         mutex_lock(&kvm->lock);
1696         if (!kvm_vcpu_compatible(vcpu)) {
1697                 r = -EINVAL;
1698                 goto unlock_vcpu_destroy;
1699         }
1700         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1701                 r = -EINVAL;
1702                 goto unlock_vcpu_destroy;
1703         }
1704
1705         kvm_for_each_vcpu(r, v, kvm)
1706                 if (v->vcpu_id == id) {
1707                         r = -EEXIST;
1708                         goto unlock_vcpu_destroy;
1709                 }
1710
1711         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1712
1713         /* Now it's all set up, let userspace reach it */
1714         kvm_get_kvm(kvm);
1715         r = create_vcpu_fd(vcpu);
1716         if (r < 0) {
1717                 kvm_put_kvm(kvm);
1718                 goto unlock_vcpu_destroy;
1719         }
1720
1721         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1722         smp_wmb();
1723         atomic_inc(&kvm->online_vcpus);
1724
1725         mutex_unlock(&kvm->lock);
1726         return r;
1727
1728 unlock_vcpu_destroy:
1729         mutex_unlock(&kvm->lock);
1730 vcpu_destroy:
1731         kvm_arch_vcpu_destroy(vcpu);
1732         return r;
1733 }
1734
1735 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1736 {
1737         if (sigset) {
1738                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1739                 vcpu->sigset_active = 1;
1740                 vcpu->sigset = *sigset;
1741         } else
1742                 vcpu->sigset_active = 0;
1743         return 0;
1744 }
1745
1746 static long kvm_vcpu_ioctl(struct file *filp,
1747                            unsigned int ioctl, unsigned long arg)
1748 {
1749         struct kvm_vcpu *vcpu = filp->private_data;
1750         void __user *argp = (void __user *)arg;
1751         int r;
1752         struct kvm_fpu *fpu = NULL;
1753         struct kvm_sregs *kvm_sregs = NULL;
1754
1755         if (vcpu->kvm->mm != current->mm)
1756                 return -EIO;
1757
1758 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1759         /*
1760          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1761          * so vcpu_load() would break it.
1762          */
1763         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1764                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1765 #endif
1766
1767
1768         vcpu_load(vcpu);
1769         switch (ioctl) {
1770         case KVM_RUN:
1771                 r = -EINVAL;
1772                 if (arg)
1773                         goto out;
1774                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1775                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1776                 break;
1777         case KVM_GET_REGS: {
1778                 struct kvm_regs *kvm_regs;
1779
1780                 r = -ENOMEM;
1781                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1782                 if (!kvm_regs)
1783                         goto out;
1784                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1785                 if (r)
1786                         goto out_free1;
1787                 r = -EFAULT;
1788                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1789                         goto out_free1;
1790                 r = 0;
1791 out_free1:
1792                 kfree(kvm_regs);
1793                 break;
1794         }
1795         case KVM_SET_REGS: {
1796                 struct kvm_regs *kvm_regs;
1797
1798                 r = -ENOMEM;
1799                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1800                 if (IS_ERR(kvm_regs)) {
1801                         r = PTR_ERR(kvm_regs);
1802                         goto out;
1803                 }
1804                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1805                 if (r)
1806                         goto out_free2;
1807                 r = 0;
1808 out_free2:
1809                 kfree(kvm_regs);
1810                 break;
1811         }
1812         case KVM_GET_SREGS: {
1813                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1814                 r = -ENOMEM;
1815                 if (!kvm_sregs)
1816                         goto out;
1817                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1818                 if (r)
1819                         goto out;
1820                 r = -EFAULT;
1821                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1822                         goto out;
1823                 r = 0;
1824                 break;
1825         }
1826         case KVM_SET_SREGS: {
1827                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1828                 if (IS_ERR(kvm_sregs)) {
1829                         r = PTR_ERR(kvm_sregs);
1830                         goto out;
1831                 }
1832                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1833                 if (r)
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_GET_MP_STATE: {
1839                 struct kvm_mp_state mp_state;
1840
1841                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1842                 if (r)
1843                         goto out;
1844                 r = -EFAULT;
1845                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_SET_MP_STATE: {
1851                 struct kvm_mp_state mp_state;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1855                         goto out;
1856                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1857                 if (r)
1858                         goto out;
1859                 r = 0;
1860                 break;
1861         }
1862         case KVM_TRANSLATE: {
1863                 struct kvm_translation tr;
1864
1865                 r = -EFAULT;
1866                 if (copy_from_user(&tr, argp, sizeof tr))
1867                         goto out;
1868                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1869                 if (r)
1870                         goto out;
1871                 r = -EFAULT;
1872                 if (copy_to_user(argp, &tr, sizeof tr))
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_SET_GUEST_DEBUG: {
1878                 struct kvm_guest_debug dbg;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&dbg, argp, sizeof dbg))
1882                         goto out;
1883                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1884                 if (r)
1885                         goto out;
1886                 r = 0;
1887                 break;
1888         }
1889         case KVM_SET_SIGNAL_MASK: {
1890                 struct kvm_signal_mask __user *sigmask_arg = argp;
1891                 struct kvm_signal_mask kvm_sigmask;
1892                 sigset_t sigset, *p;
1893
1894                 p = NULL;
1895                 if (argp) {
1896                         r = -EFAULT;
1897                         if (copy_from_user(&kvm_sigmask, argp,
1898                                            sizeof kvm_sigmask))
1899                                 goto out;
1900                         r = -EINVAL;
1901                         if (kvm_sigmask.len != sizeof sigset)
1902                                 goto out;
1903                         r = -EFAULT;
1904                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1905                                            sizeof sigset))
1906                                 goto out;
1907                         p = &sigset;
1908                 }
1909                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1910                 break;
1911         }
1912         case KVM_GET_FPU: {
1913                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1914                 r = -ENOMEM;
1915                 if (!fpu)
1916                         goto out;
1917                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1918                 if (r)
1919                         goto out;
1920                 r = -EFAULT;
1921                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1922                         goto out;
1923                 r = 0;
1924                 break;
1925         }
1926         case KVM_SET_FPU: {
1927                 fpu = memdup_user(argp, sizeof(*fpu));
1928                 if (IS_ERR(fpu)) {
1929                         r = PTR_ERR(fpu);
1930                         goto out;
1931                 }
1932                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1933                 if (r)
1934                         goto out;
1935                 r = 0;
1936                 break;
1937         }
1938         default:
1939                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1940         }
1941 out:
1942         vcpu_put(vcpu);
1943         kfree(fpu);
1944         kfree(kvm_sregs);
1945         return r;
1946 }
1947
1948 #ifdef CONFIG_COMPAT
1949 static long kvm_vcpu_compat_ioctl(struct file *filp,
1950                                   unsigned int ioctl, unsigned long arg)
1951 {
1952         struct kvm_vcpu *vcpu = filp->private_data;
1953         void __user *argp = compat_ptr(arg);
1954         int r;
1955
1956         if (vcpu->kvm->mm != current->mm)
1957                 return -EIO;
1958
1959         switch (ioctl) {
1960         case KVM_SET_SIGNAL_MASK: {
1961                 struct kvm_signal_mask __user *sigmask_arg = argp;
1962                 struct kvm_signal_mask kvm_sigmask;
1963                 compat_sigset_t csigset;
1964                 sigset_t sigset;
1965
1966                 if (argp) {
1967                         r = -EFAULT;
1968                         if (copy_from_user(&kvm_sigmask, argp,
1969                                            sizeof kvm_sigmask))
1970                                 goto out;
1971                         r = -EINVAL;
1972                         if (kvm_sigmask.len != sizeof csigset)
1973                                 goto out;
1974                         r = -EFAULT;
1975                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1976                                            sizeof csigset))
1977                                 goto out;
1978                 }
1979                 sigset_from_compat(&sigset, &csigset);
1980                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1981                 break;
1982         }
1983         default:
1984                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1985         }
1986
1987 out:
1988         return r;
1989 }
1990 #endif
1991
1992 static long kvm_vm_ioctl(struct file *filp,
1993                            unsigned int ioctl, unsigned long arg)
1994 {
1995         struct kvm *kvm = filp->private_data;
1996         void __user *argp = (void __user *)arg;
1997         int r;
1998
1999         if (kvm->mm != current->mm)
2000                 return -EIO;
2001         switch (ioctl) {
2002         case KVM_CREATE_VCPU:
2003                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2004                 if (r < 0)
2005                         goto out;
2006                 break;
2007         case KVM_SET_USER_MEMORY_REGION: {
2008                 struct kvm_userspace_memory_region kvm_userspace_mem;
2009
2010                 r = -EFAULT;
2011                 if (copy_from_user(&kvm_userspace_mem, argp,
2012                                                 sizeof kvm_userspace_mem))
2013                         goto out;
2014
2015                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2016                 if (r)
2017                         goto out;
2018                 break;
2019         }
2020         case KVM_GET_DIRTY_LOG: {
2021                 struct kvm_dirty_log log;
2022
2023                 r = -EFAULT;
2024                 if (copy_from_user(&log, argp, sizeof log))
2025                         goto out;
2026                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2027                 if (r)
2028                         goto out;
2029                 break;
2030         }
2031 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2032         case KVM_REGISTER_COALESCED_MMIO: {
2033                 struct kvm_coalesced_mmio_zone zone;
2034                 r = -EFAULT;
2035                 if (copy_from_user(&zone, argp, sizeof zone))
2036                         goto out;
2037                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2038                 if (r)
2039                         goto out;
2040                 r = 0;
2041                 break;
2042         }
2043         case KVM_UNREGISTER_COALESCED_MMIO: {
2044                 struct kvm_coalesced_mmio_zone zone;
2045                 r = -EFAULT;
2046                 if (copy_from_user(&zone, argp, sizeof zone))
2047                         goto out;
2048                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2049                 if (r)
2050                         goto out;
2051                 r = 0;
2052                 break;
2053         }
2054 #endif
2055         case KVM_IRQFD: {
2056                 struct kvm_irqfd data;
2057
2058                 r = -EFAULT;
2059                 if (copy_from_user(&data, argp, sizeof data))
2060                         goto out;
2061                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2062                 break;
2063         }
2064         case KVM_IOEVENTFD: {
2065                 struct kvm_ioeventfd data;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&data, argp, sizeof data))
2069                         goto out;
2070                 r = kvm_ioeventfd(kvm, &data);
2071                 break;
2072         }
2073 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2074         case KVM_SET_BOOT_CPU_ID:
2075                 r = 0;
2076                 mutex_lock(&kvm->lock);
2077                 if (atomic_read(&kvm->online_vcpus) != 0)
2078                         r = -EBUSY;
2079                 else
2080                         kvm->bsp_vcpu_id = arg;
2081                 mutex_unlock(&kvm->lock);
2082                 break;
2083 #endif
2084 #ifdef CONFIG_HAVE_KVM_MSI
2085         case KVM_SIGNAL_MSI: {
2086                 struct kvm_msi msi;
2087
2088                 r = -EFAULT;
2089                 if (copy_from_user(&msi, argp, sizeof msi))
2090                         goto out;
2091                 r = kvm_send_userspace_msi(kvm, &msi);
2092                 break;
2093         }
2094 #endif
2095         default:
2096                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2097                 if (r == -ENOTTY)
2098                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2099         }
2100 out:
2101         return r;
2102 }
2103
2104 #ifdef CONFIG_COMPAT
2105 struct compat_kvm_dirty_log {
2106         __u32 slot;
2107         __u32 padding1;
2108         union {
2109                 compat_uptr_t dirty_bitmap; /* one bit per page */
2110                 __u64 padding2;
2111         };
2112 };
2113
2114 static long kvm_vm_compat_ioctl(struct file *filp,
2115                            unsigned int ioctl, unsigned long arg)
2116 {
2117         struct kvm *kvm = filp->private_data;
2118         int r;
2119
2120         if (kvm->mm != current->mm)
2121                 return -EIO;
2122         switch (ioctl) {
2123         case KVM_GET_DIRTY_LOG: {
2124                 struct compat_kvm_dirty_log compat_log;
2125                 struct kvm_dirty_log log;
2126
2127                 r = -EFAULT;
2128                 if (copy_from_user(&compat_log, (void __user *)arg,
2129                                    sizeof(compat_log)))
2130                         goto out;
2131                 log.slot         = compat_log.slot;
2132                 log.padding1     = compat_log.padding1;
2133                 log.padding2     = compat_log.padding2;
2134                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2135
2136                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2137                 if (r)
2138                         goto out;
2139                 break;
2140         }
2141         default:
2142                 r = kvm_vm_ioctl(filp, ioctl, arg);
2143         }
2144
2145 out:
2146         return r;
2147 }
2148 #endif
2149
2150 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2151 {
2152         struct page *page[1];
2153         unsigned long addr;
2154         int npages;
2155         gfn_t gfn = vmf->pgoff;
2156         struct kvm *kvm = vma->vm_file->private_data;
2157
2158         addr = gfn_to_hva(kvm, gfn);
2159         if (kvm_is_error_hva(addr))
2160                 return VM_FAULT_SIGBUS;
2161
2162         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2163                                 NULL);
2164         if (unlikely(npages != 1))
2165                 return VM_FAULT_SIGBUS;
2166
2167         vmf->page = page[0];
2168         return 0;
2169 }
2170
2171 static const struct vm_operations_struct kvm_vm_vm_ops = {
2172         .fault = kvm_vm_fault,
2173 };
2174
2175 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2176 {
2177         vma->vm_ops = &kvm_vm_vm_ops;
2178         return 0;
2179 }
2180
2181 static struct file_operations kvm_vm_fops = {
2182         .release        = kvm_vm_release,
2183         .unlocked_ioctl = kvm_vm_ioctl,
2184 #ifdef CONFIG_COMPAT
2185         .compat_ioctl   = kvm_vm_compat_ioctl,
2186 #endif
2187         .mmap           = kvm_vm_mmap,
2188         .llseek         = noop_llseek,
2189 };
2190
2191 static int kvm_dev_ioctl_create_vm(unsigned long type)
2192 {
2193         int r;
2194         struct kvm *kvm;
2195
2196         kvm = kvm_create_vm(type);
2197         if (IS_ERR(kvm))
2198                 return PTR_ERR(kvm);
2199 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2200         r = kvm_coalesced_mmio_init(kvm);
2201         if (r < 0) {
2202                 kvm_put_kvm(kvm);
2203                 return r;
2204         }
2205 #endif
2206         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2207         if (r < 0)
2208                 kvm_put_kvm(kvm);
2209
2210         return r;
2211 }
2212
2213 static long kvm_dev_ioctl_check_extension_generic(long arg)
2214 {
2215         switch (arg) {
2216         case KVM_CAP_USER_MEMORY:
2217         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2218         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2219 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2220         case KVM_CAP_SET_BOOT_CPU_ID:
2221 #endif
2222         case KVM_CAP_INTERNAL_ERROR_DATA:
2223 #ifdef CONFIG_HAVE_KVM_MSI
2224         case KVM_CAP_SIGNAL_MSI:
2225 #endif
2226                 return 1;
2227 #ifdef KVM_CAP_IRQ_ROUTING
2228         case KVM_CAP_IRQ_ROUTING:
2229                 return KVM_MAX_IRQ_ROUTES;
2230 #endif
2231         default:
2232                 break;
2233         }
2234         return kvm_dev_ioctl_check_extension(arg);
2235 }
2236
2237 static long kvm_dev_ioctl(struct file *filp,
2238                           unsigned int ioctl, unsigned long arg)
2239 {
2240         long r = -EINVAL;
2241
2242         switch (ioctl) {
2243         case KVM_GET_API_VERSION:
2244                 r = -EINVAL;
2245                 if (arg)
2246                         goto out;
2247                 r = KVM_API_VERSION;
2248                 break;
2249         case KVM_CREATE_VM:
2250                 r = kvm_dev_ioctl_create_vm(arg);
2251                 break;
2252         case KVM_CHECK_EXTENSION:
2253                 r = kvm_dev_ioctl_check_extension_generic(arg);
2254                 break;
2255         case KVM_GET_VCPU_MMAP_SIZE:
2256                 r = -EINVAL;
2257                 if (arg)
2258                         goto out;
2259                 r = PAGE_SIZE;     /* struct kvm_run */
2260 #ifdef CONFIG_X86
2261                 r += PAGE_SIZE;    /* pio data page */
2262 #endif
2263 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2264                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2265 #endif
2266                 break;
2267         case KVM_TRACE_ENABLE:
2268         case KVM_TRACE_PAUSE:
2269         case KVM_TRACE_DISABLE:
2270                 r = -EOPNOTSUPP;
2271                 break;
2272         default:
2273                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2274         }
2275 out:
2276         return r;
2277 }
2278
2279 static struct file_operations kvm_chardev_ops = {
2280         .unlocked_ioctl = kvm_dev_ioctl,
2281         .compat_ioctl   = kvm_dev_ioctl,
2282         .llseek         = noop_llseek,
2283 };
2284
2285 static struct miscdevice kvm_dev = {
2286         KVM_MINOR,
2287         "kvm",
2288         &kvm_chardev_ops,
2289 };
2290
2291 static void hardware_enable_nolock(void *junk)
2292 {
2293         int cpu = raw_smp_processor_id();
2294         int r;
2295
2296         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2297                 return;
2298
2299         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2300
2301         r = kvm_arch_hardware_enable(NULL);
2302
2303         if (r) {
2304                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2305                 atomic_inc(&hardware_enable_failed);
2306                 printk(KERN_INFO "kvm: enabling virtualization on "
2307                                  "CPU%d failed\n", cpu);
2308         }
2309 }
2310
2311 static void hardware_enable(void *junk)
2312 {
2313         raw_spin_lock(&kvm_lock);
2314         hardware_enable_nolock(junk);
2315         raw_spin_unlock(&kvm_lock);
2316 }
2317
2318 static void hardware_disable_nolock(void *junk)
2319 {
2320         int cpu = raw_smp_processor_id();
2321
2322         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2323                 return;
2324         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2325         kvm_arch_hardware_disable(NULL);
2326 }
2327
2328 static void hardware_disable(void *junk)
2329 {
2330         raw_spin_lock(&kvm_lock);
2331         hardware_disable_nolock(junk);
2332         raw_spin_unlock(&kvm_lock);
2333 }
2334
2335 static void hardware_disable_all_nolock(void)
2336 {
2337         BUG_ON(!kvm_usage_count);
2338
2339         kvm_usage_count--;
2340         if (!kvm_usage_count)
2341                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2342 }
2343
2344 static void hardware_disable_all(void)
2345 {
2346         raw_spin_lock(&kvm_lock);
2347         hardware_disable_all_nolock();
2348         raw_spin_unlock(&kvm_lock);
2349 }
2350
2351 static int hardware_enable_all(void)
2352 {
2353         int r = 0;
2354
2355         raw_spin_lock(&kvm_lock);
2356
2357         kvm_usage_count++;
2358         if (kvm_usage_count == 1) {
2359                 atomic_set(&hardware_enable_failed, 0);
2360                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2361
2362                 if (atomic_read(&hardware_enable_failed)) {
2363                         hardware_disable_all_nolock();
2364                         r = -EBUSY;
2365                 }
2366         }
2367
2368         raw_spin_unlock(&kvm_lock);
2369
2370         return r;
2371 }
2372
2373 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2374                            void *v)
2375 {
2376         int cpu = (long)v;
2377
2378         if (!kvm_usage_count)
2379                 return NOTIFY_OK;
2380
2381         val &= ~CPU_TASKS_FROZEN;
2382         switch (val) {
2383         case CPU_DYING:
2384                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2385                        cpu);
2386                 hardware_disable(NULL);
2387                 break;
2388         case CPU_STARTING:
2389                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2390                        cpu);
2391                 hardware_enable(NULL);
2392                 break;
2393         }
2394         return NOTIFY_OK;
2395 }
2396
2397
2398 asmlinkage void kvm_spurious_fault(void)
2399 {
2400         /* Fault while not rebooting.  We want the trace. */
2401         BUG();
2402 }
2403 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2404
2405 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2406                       void *v)
2407 {
2408         /*
2409          * Some (well, at least mine) BIOSes hang on reboot if
2410          * in vmx root mode.
2411          *
2412          * And Intel TXT required VMX off for all cpu when system shutdown.
2413          */
2414         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2415         kvm_rebooting = true;
2416         on_each_cpu(hardware_disable_nolock, NULL, 1);
2417         return NOTIFY_OK;
2418 }
2419
2420 static struct notifier_block kvm_reboot_notifier = {
2421         .notifier_call = kvm_reboot,
2422         .priority = 0,
2423 };
2424
2425 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2426 {
2427         int i;
2428
2429         for (i = 0; i < bus->dev_count; i++) {
2430                 struct kvm_io_device *pos = bus->range[i].dev;
2431
2432                 kvm_iodevice_destructor(pos);
2433         }
2434         kfree(bus);
2435 }
2436
2437 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2438 {
2439         const struct kvm_io_range *r1 = p1;
2440         const struct kvm_io_range *r2 = p2;
2441
2442         if (r1->addr < r2->addr)
2443                 return -1;
2444         if (r1->addr + r1->len > r2->addr + r2->len)
2445                 return 1;
2446         return 0;
2447 }
2448
2449 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2450                           gpa_t addr, int len)
2451 {
2452         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2453                 .addr = addr,
2454                 .len = len,
2455                 .dev = dev,
2456         };
2457
2458         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2459                 kvm_io_bus_sort_cmp, NULL);
2460
2461         return 0;
2462 }
2463
2464 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2465                              gpa_t addr, int len)
2466 {
2467         struct kvm_io_range *range, key;
2468         int off;
2469
2470         key = (struct kvm_io_range) {
2471                 .addr = addr,
2472                 .len = len,
2473         };
2474
2475         range = bsearch(&key, bus->range, bus->dev_count,
2476                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2477         if (range == NULL)
2478                 return -ENOENT;
2479
2480         off = range - bus->range;
2481
2482         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2483                 off--;
2484
2485         return off;
2486 }
2487
2488 /* kvm_io_bus_write - called under kvm->slots_lock */
2489 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2490                      int len, const void *val)
2491 {
2492         int idx;
2493         struct kvm_io_bus *bus;
2494         struct kvm_io_range range;
2495
2496         range = (struct kvm_io_range) {
2497                 .addr = addr,
2498                 .len = len,
2499         };
2500
2501         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2502         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2503         if (idx < 0)
2504                 return -EOPNOTSUPP;
2505
2506         while (idx < bus->dev_count &&
2507                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2508                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2509                         return 0;
2510                 idx++;
2511         }
2512
2513         return -EOPNOTSUPP;
2514 }
2515
2516 /* kvm_io_bus_read - called under kvm->slots_lock */
2517 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2518                     int len, void *val)
2519 {
2520         int idx;
2521         struct kvm_io_bus *bus;
2522         struct kvm_io_range range;
2523
2524         range = (struct kvm_io_range) {
2525                 .addr = addr,
2526                 .len = len,
2527         };
2528
2529         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2530         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2531         if (idx < 0)
2532                 return -EOPNOTSUPP;
2533
2534         while (idx < bus->dev_count &&
2535                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2536                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2537                         return 0;
2538                 idx++;
2539         }
2540
2541         return -EOPNOTSUPP;
2542 }
2543
2544 /* Caller must hold slots_lock. */
2545 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2546                             int len, struct kvm_io_device *dev)
2547 {
2548         struct kvm_io_bus *new_bus, *bus;
2549
2550         bus = kvm->buses[bus_idx];
2551         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2552                 return -ENOSPC;
2553
2554         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2555                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2556         if (!new_bus)
2557                 return -ENOMEM;
2558         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2559                sizeof(struct kvm_io_range)));
2560         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2561         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2562         synchronize_srcu_expedited(&kvm->srcu);
2563         kfree(bus);
2564
2565         return 0;
2566 }
2567
2568 /* Caller must hold slots_lock. */
2569 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2570                               struct kvm_io_device *dev)
2571 {
2572         int i, r;
2573         struct kvm_io_bus *new_bus, *bus;
2574
2575         bus = kvm->buses[bus_idx];
2576         r = -ENOENT;
2577         for (i = 0; i < bus->dev_count; i++)
2578                 if (bus->range[i].dev == dev) {
2579                         r = 0;
2580                         break;
2581                 }
2582
2583         if (r)
2584                 return r;
2585
2586         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2587                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2588         if (!new_bus)
2589                 return -ENOMEM;
2590
2591         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2592         new_bus->dev_count--;
2593         memcpy(new_bus->range + i, bus->range + i + 1,
2594                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2595
2596         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2597         synchronize_srcu_expedited(&kvm->srcu);
2598         kfree(bus);
2599         return r;
2600 }
2601
2602 static struct notifier_block kvm_cpu_notifier = {
2603         .notifier_call = kvm_cpu_hotplug,
2604 };
2605
2606 static int vm_stat_get(void *_offset, u64 *val)
2607 {
2608         unsigned offset = (long)_offset;
2609         struct kvm *kvm;
2610
2611         *val = 0;
2612         raw_spin_lock(&kvm_lock);
2613         list_for_each_entry(kvm, &vm_list, vm_list)
2614                 *val += *(u32 *)((void *)kvm + offset);
2615         raw_spin_unlock(&kvm_lock);
2616         return 0;
2617 }
2618
2619 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2620
2621 static int vcpu_stat_get(void *_offset, u64 *val)
2622 {
2623         unsigned offset = (long)_offset;
2624         struct kvm *kvm;
2625         struct kvm_vcpu *vcpu;
2626         int i;
2627
2628         *val = 0;
2629         raw_spin_lock(&kvm_lock);
2630         list_for_each_entry(kvm, &vm_list, vm_list)
2631                 kvm_for_each_vcpu(i, vcpu, kvm)
2632                         *val += *(u32 *)((void *)vcpu + offset);
2633
2634         raw_spin_unlock(&kvm_lock);
2635         return 0;
2636 }
2637
2638 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2639
2640 static const struct file_operations *stat_fops[] = {
2641         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2642         [KVM_STAT_VM]   = &vm_stat_fops,
2643 };
2644
2645 static int kvm_init_debug(void)
2646 {
2647         int r = -EFAULT;
2648         struct kvm_stats_debugfs_item *p;
2649
2650         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2651         if (kvm_debugfs_dir == NULL)
2652                 goto out;
2653
2654         for (p = debugfs_entries; p->name; ++p) {
2655                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2656                                                 (void *)(long)p->offset,
2657                                                 stat_fops[p->kind]);
2658                 if (p->dentry == NULL)
2659                         goto out_dir;
2660         }
2661
2662         return 0;
2663
2664 out_dir:
2665         debugfs_remove_recursive(kvm_debugfs_dir);
2666 out:
2667         return r;
2668 }
2669
2670 static void kvm_exit_debug(void)
2671 {
2672         struct kvm_stats_debugfs_item *p;
2673
2674         for (p = debugfs_entries; p->name; ++p)
2675                 debugfs_remove(p->dentry);
2676         debugfs_remove(kvm_debugfs_dir);
2677 }
2678
2679 static int kvm_suspend(void)
2680 {
2681         if (kvm_usage_count)
2682                 hardware_disable_nolock(NULL);
2683         return 0;
2684 }
2685
2686 static void kvm_resume(void)
2687 {
2688         if (kvm_usage_count) {
2689                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2690                 hardware_enable_nolock(NULL);
2691         }
2692 }
2693
2694 static struct syscore_ops kvm_syscore_ops = {
2695         .suspend = kvm_suspend,
2696         .resume = kvm_resume,
2697 };
2698
2699 struct page *bad_page;
2700 pfn_t bad_pfn;
2701
2702 static inline
2703 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2704 {
2705         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2706 }
2707
2708 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2709 {
2710         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2711
2712         kvm_arch_vcpu_load(vcpu, cpu);
2713 }
2714
2715 static void kvm_sched_out(struct preempt_notifier *pn,
2716                           struct task_struct *next)
2717 {
2718         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2719
2720         kvm_arch_vcpu_put(vcpu);
2721 }
2722
2723 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2724                   struct module *module)
2725 {
2726         int r;
2727         int cpu;
2728
2729         r = kvm_arch_init(opaque);
2730         if (r)
2731                 goto out_fail;
2732
2733         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2734
2735         if (bad_page == NULL) {
2736                 r = -ENOMEM;
2737                 goto out;
2738         }
2739
2740         bad_pfn = page_to_pfn(bad_page);
2741
2742         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2743
2744         if (hwpoison_page == NULL) {
2745                 r = -ENOMEM;
2746                 goto out_free_0;
2747         }
2748
2749         hwpoison_pfn = page_to_pfn(hwpoison_page);
2750
2751         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2752
2753         if (fault_page == NULL) {
2754                 r = -ENOMEM;
2755                 goto out_free_0;
2756         }
2757
2758         fault_pfn = page_to_pfn(fault_page);
2759
2760         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2761                 r = -ENOMEM;
2762                 goto out_free_0;
2763         }
2764
2765         r = kvm_arch_hardware_setup();
2766         if (r < 0)
2767                 goto out_free_0a;
2768
2769         for_each_online_cpu(cpu) {
2770                 smp_call_function_single(cpu,
2771                                 kvm_arch_check_processor_compat,
2772                                 &r, 1);
2773                 if (r < 0)
2774                         goto out_free_1;
2775         }
2776
2777         r = register_cpu_notifier(&kvm_cpu_notifier);
2778         if (r)
2779                 goto out_free_2;
2780         register_reboot_notifier(&kvm_reboot_notifier);
2781
2782         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2783         if (!vcpu_align)
2784                 vcpu_align = __alignof__(struct kvm_vcpu);
2785         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2786                                            0, NULL);
2787         if (!kvm_vcpu_cache) {
2788                 r = -ENOMEM;
2789                 goto out_free_3;
2790         }
2791
2792         r = kvm_async_pf_init();
2793         if (r)
2794                 goto out_free;
2795
2796         kvm_chardev_ops.owner = module;
2797         kvm_vm_fops.owner = module;
2798         kvm_vcpu_fops.owner = module;
2799
2800         r = misc_register(&kvm_dev);
2801         if (r) {
2802                 printk(KERN_ERR "kvm: misc device register failed\n");
2803                 goto out_unreg;
2804         }
2805
2806         register_syscore_ops(&kvm_syscore_ops);
2807
2808         kvm_preempt_ops.sched_in = kvm_sched_in;
2809         kvm_preempt_ops.sched_out = kvm_sched_out;
2810
2811         r = kvm_init_debug();
2812         if (r) {
2813                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2814                 goto out_undebugfs;
2815         }
2816
2817         return 0;
2818
2819 out_undebugfs:
2820         unregister_syscore_ops(&kvm_syscore_ops);
2821 out_unreg:
2822         kvm_async_pf_deinit();
2823 out_free:
2824         kmem_cache_destroy(kvm_vcpu_cache);
2825 out_free_3:
2826         unregister_reboot_notifier(&kvm_reboot_notifier);
2827         unregister_cpu_notifier(&kvm_cpu_notifier);
2828 out_free_2:
2829 out_free_1:
2830         kvm_arch_hardware_unsetup();
2831 out_free_0a:
2832         free_cpumask_var(cpus_hardware_enabled);
2833 out_free_0:
2834         if (fault_page)
2835                 __free_page(fault_page);
2836         if (hwpoison_page)
2837                 __free_page(hwpoison_page);
2838         __free_page(bad_page);
2839 out:
2840         kvm_arch_exit();
2841 out_fail:
2842         return r;
2843 }
2844 EXPORT_SYMBOL_GPL(kvm_init);
2845
2846 void kvm_exit(void)
2847 {
2848         kvm_exit_debug();
2849         misc_deregister(&kvm_dev);
2850         kmem_cache_destroy(kvm_vcpu_cache);
2851         kvm_async_pf_deinit();
2852         unregister_syscore_ops(&kvm_syscore_ops);
2853         unregister_reboot_notifier(&kvm_reboot_notifier);
2854         unregister_cpu_notifier(&kvm_cpu_notifier);
2855         on_each_cpu(hardware_disable_nolock, NULL, 1);
2856         kvm_arch_hardware_unsetup();
2857         kvm_arch_exit();
2858         free_cpumask_var(cpus_hardware_enabled);
2859         __free_page(hwpoison_page);
2860         __free_page(bad_page);
2861 }
2862 EXPORT_SYMBOL_GPL(kvm_exit);