KVM: introduce KVM_ERR_PTR_BAD_PAGE
[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 bool kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (is_error_pfn(pfn))
106                 return false;
107
108         if (pfn_valid(pfn)) {
109                 int reserved;
110                 struct page *tail = pfn_to_page(pfn);
111                 struct page *head = compound_trans_head(tail);
112                 reserved = PageReserved(head);
113                 if (head != tail) {
114                         /*
115                          * "head" is not a dangling pointer
116                          * (compound_trans_head takes care of that)
117                          * but the hugepage may have been splitted
118                          * from under us (and we may not hold a
119                          * reference count on the head page so it can
120                          * be reused before we run PageReferenced), so
121                          * we've to check PageTail before returning
122                          * what we just read.
123                          */
124                         smp_rmb();
125                         if (PageTail(tail))
126                                 return reserved;
127                 }
128                 return PageReserved(tail);
129         }
130
131         return true;
132 }
133
134 /*
135  * Switches to specified vcpu, until a matching vcpu_put()
136  */
137 void vcpu_load(struct kvm_vcpu *vcpu)
138 {
139         int cpu;
140
141         mutex_lock(&vcpu->mutex);
142         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
143                 /* The thread running this VCPU changed. */
144                 struct pid *oldpid = vcpu->pid;
145                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
146                 rcu_assign_pointer(vcpu->pid, newpid);
147                 synchronize_rcu();
148                 put_pid(oldpid);
149         }
150         cpu = get_cpu();
151         preempt_notifier_register(&vcpu->preempt_notifier);
152         kvm_arch_vcpu_load(vcpu, cpu);
153         put_cpu();
154 }
155
156 void vcpu_put(struct kvm_vcpu *vcpu)
157 {
158         preempt_disable();
159         kvm_arch_vcpu_put(vcpu);
160         preempt_notifier_unregister(&vcpu->preempt_notifier);
161         preempt_enable();
162         mutex_unlock(&vcpu->mutex);
163 }
164
165 static void ack_flush(void *_completed)
166 {
167 }
168
169 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
170 {
171         int i, cpu, me;
172         cpumask_var_t cpus;
173         bool called = true;
174         struct kvm_vcpu *vcpu;
175
176         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
177
178         me = get_cpu();
179         kvm_for_each_vcpu(i, vcpu, kvm) {
180                 kvm_make_request(req, vcpu);
181                 cpu = vcpu->cpu;
182
183                 /* Set ->requests bit before we read ->mode */
184                 smp_mb();
185
186                 if (cpus != NULL && cpu != -1 && cpu != me &&
187                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
188                         cpumask_set_cpu(cpu, cpus);
189         }
190         if (unlikely(cpus == NULL))
191                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
192         else if (!cpumask_empty(cpus))
193                 smp_call_function_many(cpus, ack_flush, NULL, 1);
194         else
195                 called = false;
196         put_cpu();
197         free_cpumask_var(cpus);
198         return called;
199 }
200
201 void kvm_flush_remote_tlbs(struct kvm *kvm)
202 {
203         long dirty_count = kvm->tlbs_dirty;
204
205         smp_mb();
206         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
207                 ++kvm->stat.remote_tlb_flush;
208         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
209 }
210
211 void kvm_reload_remote_mmus(struct kvm *kvm)
212 {
213         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
214 }
215
216 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
217 {
218         struct page *page;
219         int r;
220
221         mutex_init(&vcpu->mutex);
222         vcpu->cpu = -1;
223         vcpu->kvm = kvm;
224         vcpu->vcpu_id = id;
225         vcpu->pid = NULL;
226         init_waitqueue_head(&vcpu->wq);
227         kvm_async_pf_vcpu_init(vcpu);
228
229         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
230         if (!page) {
231                 r = -ENOMEM;
232                 goto fail;
233         }
234         vcpu->run = page_address(page);
235
236         kvm_vcpu_set_in_spin_loop(vcpu, false);
237         kvm_vcpu_set_dy_eligible(vcpu, false);
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->dirty_bitmap != dont->dirty_bitmap)
554                 kvm_destroy_dirty_bitmap(free);
555
556         kvm_arch_free_memslot(free, dont);
557
558         free->npages = 0;
559 }
560
561 void kvm_free_physmem(struct kvm *kvm)
562 {
563         struct kvm_memslots *slots = kvm->memslots;
564         struct kvm_memory_slot *memslot;
565
566         kvm_for_each_memslot(memslot, slots)
567                 kvm_free_physmem_slot(memslot, NULL);
568
569         kfree(kvm->memslots);
570 }
571
572 static void kvm_destroy_vm(struct kvm *kvm)
573 {
574         int i;
575         struct mm_struct *mm = kvm->mm;
576
577         kvm_arch_sync_events(kvm);
578         raw_spin_lock(&kvm_lock);
579         list_del(&kvm->vm_list);
580         raw_spin_unlock(&kvm_lock);
581         kvm_free_irq_routing(kvm);
582         for (i = 0; i < KVM_NR_BUSES; i++)
583                 kvm_io_bus_destroy(kvm->buses[i]);
584         kvm_coalesced_mmio_free(kvm);
585 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
586         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
587 #else
588         kvm_arch_flush_shadow(kvm);
589 #endif
590         kvm_arch_destroy_vm(kvm);
591         kvm_free_physmem(kvm);
592         cleanup_srcu_struct(&kvm->srcu);
593         kvm_arch_free_vm(kvm);
594         hardware_disable_all();
595         mmdrop(mm);
596 }
597
598 void kvm_get_kvm(struct kvm *kvm)
599 {
600         atomic_inc(&kvm->users_count);
601 }
602 EXPORT_SYMBOL_GPL(kvm_get_kvm);
603
604 void kvm_put_kvm(struct kvm *kvm)
605 {
606         if (atomic_dec_and_test(&kvm->users_count))
607                 kvm_destroy_vm(kvm);
608 }
609 EXPORT_SYMBOL_GPL(kvm_put_kvm);
610
611
612 static int kvm_vm_release(struct inode *inode, struct file *filp)
613 {
614         struct kvm *kvm = filp->private_data;
615
616         kvm_irqfd_release(kvm);
617
618         kvm_put_kvm(kvm);
619         return 0;
620 }
621
622 /*
623  * Allocation size is twice as large as the actual dirty bitmap size.
624  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
625  */
626 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
627 {
628 #ifndef CONFIG_S390
629         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
630
631         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
632         if (!memslot->dirty_bitmap)
633                 return -ENOMEM;
634
635 #endif /* !CONFIG_S390 */
636         return 0;
637 }
638
639 static int cmp_memslot(const void *slot1, const void *slot2)
640 {
641         struct kvm_memory_slot *s1, *s2;
642
643         s1 = (struct kvm_memory_slot *)slot1;
644         s2 = (struct kvm_memory_slot *)slot2;
645
646         if (s1->npages < s2->npages)
647                 return 1;
648         if (s1->npages > s2->npages)
649                 return -1;
650
651         return 0;
652 }
653
654 /*
655  * Sort the memslots base on its size, so the larger slots
656  * will get better fit.
657  */
658 static void sort_memslots(struct kvm_memslots *slots)
659 {
660         int i;
661
662         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
663               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
664
665         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
666                 slots->id_to_index[slots->memslots[i].id] = i;
667 }
668
669 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
670 {
671         if (new) {
672                 int id = new->id;
673                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
674                 unsigned long npages = old->npages;
675
676                 *old = *new;
677                 if (new->npages != npages)
678                         sort_memslots(slots);
679         }
680
681         slots->generation++;
682 }
683
684 /*
685  * Allocate some memory and give it an address in the guest physical address
686  * space.
687  *
688  * Discontiguous memory is allowed, mostly for framebuffers.
689  *
690  * Must be called holding mmap_sem for write.
691  */
692 int __kvm_set_memory_region(struct kvm *kvm,
693                             struct kvm_userspace_memory_region *mem,
694                             int user_alloc)
695 {
696         int r;
697         gfn_t base_gfn;
698         unsigned long npages;
699         unsigned long i;
700         struct kvm_memory_slot *memslot;
701         struct kvm_memory_slot old, new;
702         struct kvm_memslots *slots, *old_memslots;
703
704         r = -EINVAL;
705         /* General sanity checks */
706         if (mem->memory_size & (PAGE_SIZE - 1))
707                 goto out;
708         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
709                 goto out;
710         /* We can read the guest memory with __xxx_user() later on. */
711         if (user_alloc &&
712             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
713              !access_ok(VERIFY_WRITE,
714                         (void __user *)(unsigned long)mem->userspace_addr,
715                         mem->memory_size)))
716                 goto out;
717         if (mem->slot >= KVM_MEM_SLOTS_NUM)
718                 goto out;
719         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
720                 goto out;
721
722         memslot = id_to_memslot(kvm->memslots, mem->slot);
723         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
724         npages = mem->memory_size >> PAGE_SHIFT;
725
726         r = -EINVAL;
727         if (npages > KVM_MEM_MAX_NR_PAGES)
728                 goto out;
729
730         if (!npages)
731                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
732
733         new = old = *memslot;
734
735         new.id = mem->slot;
736         new.base_gfn = base_gfn;
737         new.npages = npages;
738         new.flags = mem->flags;
739
740         /* Disallow changing a memory slot's size. */
741         r = -EINVAL;
742         if (npages && old.npages && npages != old.npages)
743                 goto out_free;
744
745         /* Check for overlaps */
746         r = -EEXIST;
747         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
748                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
749
750                 if (s == memslot || !s->npages)
751                         continue;
752                 if (!((base_gfn + npages <= s->base_gfn) ||
753                       (base_gfn >= s->base_gfn + s->npages)))
754                         goto out_free;
755         }
756
757         /* Free page dirty bitmap if unneeded */
758         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
759                 new.dirty_bitmap = NULL;
760
761         r = -ENOMEM;
762
763         /* Allocate if a slot is being created */
764         if (npages && !old.npages) {
765                 new.user_alloc = user_alloc;
766                 new.userspace_addr = mem->userspace_addr;
767
768                 if (kvm_arch_create_memslot(&new, npages))
769                         goto out_free;
770         }
771
772         /* Allocate page dirty bitmap if needed */
773         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
774                 if (kvm_create_dirty_bitmap(&new) < 0)
775                         goto out_free;
776                 /* destroy any largepage mappings for dirty tracking */
777         }
778
779         if (!npages) {
780                 struct kvm_memory_slot *slot;
781
782                 r = -ENOMEM;
783                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
784                                 GFP_KERNEL);
785                 if (!slots)
786                         goto out_free;
787                 slot = id_to_memslot(slots, mem->slot);
788                 slot->flags |= KVM_MEMSLOT_INVALID;
789
790                 update_memslots(slots, NULL);
791
792                 old_memslots = kvm->memslots;
793                 rcu_assign_pointer(kvm->memslots, slots);
794                 synchronize_srcu_expedited(&kvm->srcu);
795                 /* From this point no new shadow pages pointing to a deleted
796                  * memslot will be created.
797                  *
798                  * validation of sp->gfn happens in:
799                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
800                  *      - kvm_is_visible_gfn (mmu_check_roots)
801                  */
802                 kvm_arch_flush_shadow(kvm);
803                 kfree(old_memslots);
804         }
805
806         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
807         if (r)
808                 goto out_free;
809
810         /* map/unmap the pages in iommu page table */
811         if (npages) {
812                 r = kvm_iommu_map_pages(kvm, &new);
813                 if (r)
814                         goto out_free;
815         } else
816                 kvm_iommu_unmap_pages(kvm, &old);
817
818         r = -ENOMEM;
819         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
820                         GFP_KERNEL);
821         if (!slots)
822                 goto out_free;
823
824         /* actual memory is freed via old in kvm_free_physmem_slot below */
825         if (!npages) {
826                 new.dirty_bitmap = NULL;
827                 memset(&new.arch, 0, sizeof(new.arch));
828         }
829
830         update_memslots(slots, &new);
831         old_memslots = kvm->memslots;
832         rcu_assign_pointer(kvm->memslots, slots);
833         synchronize_srcu_expedited(&kvm->srcu);
834
835         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
836
837         /*
838          * If the new memory slot is created, we need to clear all
839          * mmio sptes.
840          */
841         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
842                 kvm_arch_flush_shadow(kvm);
843
844         kvm_free_physmem_slot(&old, &new);
845         kfree(old_memslots);
846
847         return 0;
848
849 out_free:
850         kvm_free_physmem_slot(&new, &old);
851 out:
852         return r;
853
854 }
855 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
856
857 int kvm_set_memory_region(struct kvm *kvm,
858                           struct kvm_userspace_memory_region *mem,
859                           int user_alloc)
860 {
861         int r;
862
863         mutex_lock(&kvm->slots_lock);
864         r = __kvm_set_memory_region(kvm, mem, user_alloc);
865         mutex_unlock(&kvm->slots_lock);
866         return r;
867 }
868 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
869
870 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
871                                    struct
872                                    kvm_userspace_memory_region *mem,
873                                    int user_alloc)
874 {
875         if (mem->slot >= KVM_MEMORY_SLOTS)
876                 return -EINVAL;
877         return kvm_set_memory_region(kvm, mem, user_alloc);
878 }
879
880 int kvm_get_dirty_log(struct kvm *kvm,
881                         struct kvm_dirty_log *log, int *is_dirty)
882 {
883         struct kvm_memory_slot *memslot;
884         int r, i;
885         unsigned long n;
886         unsigned long any = 0;
887
888         r = -EINVAL;
889         if (log->slot >= KVM_MEMORY_SLOTS)
890                 goto out;
891
892         memslot = id_to_memslot(kvm->memslots, log->slot);
893         r = -ENOENT;
894         if (!memslot->dirty_bitmap)
895                 goto out;
896
897         n = kvm_dirty_bitmap_bytes(memslot);
898
899         for (i = 0; !any && i < n/sizeof(long); ++i)
900                 any = memslot->dirty_bitmap[i];
901
902         r = -EFAULT;
903         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
904                 goto out;
905
906         if (any)
907                 *is_dirty = 1;
908
909         r = 0;
910 out:
911         return r;
912 }
913
914 bool kvm_largepages_enabled(void)
915 {
916         return largepages_enabled;
917 }
918
919 void kvm_disable_largepages(void)
920 {
921         largepages_enabled = false;
922 }
923 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
924
925 static inline unsigned long bad_hva(void)
926 {
927         return PAGE_OFFSET;
928 }
929
930 int kvm_is_error_hva(unsigned long addr)
931 {
932         return addr == bad_hva();
933 }
934 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
935
936 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
937 {
938         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
939 }
940 EXPORT_SYMBOL_GPL(gfn_to_memslot);
941
942 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
943 {
944         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
945
946         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
947               memslot->flags & KVM_MEMSLOT_INVALID)
948                 return 0;
949
950         return 1;
951 }
952 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
953
954 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
955 {
956         struct vm_area_struct *vma;
957         unsigned long addr, size;
958
959         size = PAGE_SIZE;
960
961         addr = gfn_to_hva(kvm, gfn);
962         if (kvm_is_error_hva(addr))
963                 return PAGE_SIZE;
964
965         down_read(&current->mm->mmap_sem);
966         vma = find_vma(current->mm, addr);
967         if (!vma)
968                 goto out;
969
970         size = vma_kernel_pagesize(vma);
971
972 out:
973         up_read(&current->mm->mmap_sem);
974
975         return size;
976 }
977
978 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
979                                      gfn_t *nr_pages)
980 {
981         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
982                 return bad_hva();
983
984         if (nr_pages)
985                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
986
987         return gfn_to_hva_memslot(slot, gfn);
988 }
989
990 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
991 {
992         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
993 }
994 EXPORT_SYMBOL_GPL(gfn_to_hva);
995
996 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
997         unsigned long start, int write, struct page **page)
998 {
999         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1000
1001         if (write)
1002                 flags |= FOLL_WRITE;
1003
1004         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1005 }
1006
1007 static inline int check_user_page_hwpoison(unsigned long addr)
1008 {
1009         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1010
1011         rc = __get_user_pages(current, current->mm, addr, 1,
1012                               flags, NULL, NULL, NULL);
1013         return rc == -EHWPOISON;
1014 }
1015
1016 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1017                         bool write_fault, bool *writable)
1018 {
1019         struct page *page[1];
1020         int npages = 0;
1021         pfn_t pfn;
1022
1023         /* we can do it either atomically or asynchronously, not both */
1024         BUG_ON(atomic && async);
1025
1026         BUG_ON(!write_fault && !writable);
1027
1028         if (writable)
1029                 *writable = true;
1030
1031         if (atomic || async)
1032                 npages = __get_user_pages_fast(addr, 1, 1, page);
1033
1034         if (unlikely(npages != 1) && !atomic) {
1035                 might_sleep();
1036
1037                 if (writable)
1038                         *writable = write_fault;
1039
1040                 if (async) {
1041                         down_read(&current->mm->mmap_sem);
1042                         npages = get_user_page_nowait(current, current->mm,
1043                                                      addr, write_fault, page);
1044                         up_read(&current->mm->mmap_sem);
1045                 } else
1046                         npages = get_user_pages_fast(addr, 1, write_fault,
1047                                                      page);
1048
1049                 /* map read fault as writable if possible */
1050                 if (unlikely(!write_fault) && npages == 1) {
1051                         struct page *wpage[1];
1052
1053                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1054                         if (npages == 1) {
1055                                 *writable = true;
1056                                 put_page(page[0]);
1057                                 page[0] = wpage[0];
1058                         }
1059                         npages = 1;
1060                 }
1061         }
1062
1063         if (unlikely(npages != 1)) {
1064                 struct vm_area_struct *vma;
1065
1066                 if (atomic)
1067                         return KVM_PFN_ERR_FAULT;
1068
1069                 down_read(&current->mm->mmap_sem);
1070                 if (npages == -EHWPOISON ||
1071                         (!async && check_user_page_hwpoison(addr))) {
1072                         up_read(&current->mm->mmap_sem);
1073                         return KVM_PFN_ERR_HWPOISON;
1074                 }
1075
1076                 vma = find_vma_intersection(current->mm, addr, addr+1);
1077
1078                 if (vma == NULL)
1079                         pfn = KVM_PFN_ERR_FAULT;
1080                 else if ((vma->vm_flags & VM_PFNMAP)) {
1081                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1082                                 vma->vm_pgoff;
1083                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1084                 } else {
1085                         if (async && (vma->vm_flags & VM_WRITE))
1086                                 *async = true;
1087                         pfn = KVM_PFN_ERR_FAULT;
1088                 }
1089                 up_read(&current->mm->mmap_sem);
1090         } else
1091                 pfn = page_to_pfn(page[0]);
1092
1093         return pfn;
1094 }
1095
1096 pfn_t hva_to_pfn_atomic(unsigned long addr)
1097 {
1098         return hva_to_pfn(addr, true, NULL, true, NULL);
1099 }
1100 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1101
1102 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1103                           bool write_fault, bool *writable)
1104 {
1105         unsigned long addr;
1106
1107         if (async)
1108                 *async = false;
1109
1110         addr = gfn_to_hva(kvm, gfn);
1111         if (kvm_is_error_hva(addr))
1112                 return KVM_PFN_ERR_BAD;
1113
1114         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1115 }
1116
1117 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1118 {
1119         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1120 }
1121 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1122
1123 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1124                        bool write_fault, bool *writable)
1125 {
1126         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1127 }
1128 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1129
1130 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1131 {
1132         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1133 }
1134 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1135
1136 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1137                       bool *writable)
1138 {
1139         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1140 }
1141 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1142
1143 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1144 {
1145         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1146         return hva_to_pfn(addr, false, NULL, true, NULL);
1147 }
1148
1149 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1150                                                                   int nr_pages)
1151 {
1152         unsigned long addr;
1153         gfn_t entry;
1154
1155         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1156         if (kvm_is_error_hva(addr))
1157                 return -1;
1158
1159         if (entry < nr_pages)
1160                 return 0;
1161
1162         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1163 }
1164 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1165
1166 static struct page *kvm_pfn_to_page(pfn_t pfn)
1167 {
1168         WARN_ON(kvm_is_mmio_pfn(pfn));
1169
1170         if (is_error_pfn(pfn) || kvm_is_mmio_pfn(pfn))
1171                 return KVM_ERR_PTR_BAD_PAGE;
1172
1173         return pfn_to_page(pfn);
1174 }
1175
1176 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1177 {
1178         pfn_t pfn;
1179
1180         pfn = gfn_to_pfn(kvm, gfn);
1181
1182         return kvm_pfn_to_page(pfn);
1183 }
1184
1185 EXPORT_SYMBOL_GPL(gfn_to_page);
1186
1187 void kvm_release_page_clean(struct page *page)
1188 {
1189         if (!is_error_page(page))
1190                 kvm_release_pfn_clean(page_to_pfn(page));
1191 }
1192 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1193
1194 void kvm_release_pfn_clean(pfn_t pfn)
1195 {
1196         if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1197                 put_page(pfn_to_page(pfn));
1198 }
1199 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1200
1201 void kvm_release_page_dirty(struct page *page)
1202 {
1203         WARN_ON(is_error_page(page));
1204
1205         kvm_release_pfn_dirty(page_to_pfn(page));
1206 }
1207 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1208
1209 void kvm_release_pfn_dirty(pfn_t pfn)
1210 {
1211         kvm_set_pfn_dirty(pfn);
1212         kvm_release_pfn_clean(pfn);
1213 }
1214 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1215
1216 void kvm_set_page_dirty(struct page *page)
1217 {
1218         kvm_set_pfn_dirty(page_to_pfn(page));
1219 }
1220 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1221
1222 void kvm_set_pfn_dirty(pfn_t pfn)
1223 {
1224         if (!kvm_is_mmio_pfn(pfn)) {
1225                 struct page *page = pfn_to_page(pfn);
1226                 if (!PageReserved(page))
1227                         SetPageDirty(page);
1228         }
1229 }
1230 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1231
1232 void kvm_set_pfn_accessed(pfn_t pfn)
1233 {
1234         if (!kvm_is_mmio_pfn(pfn))
1235                 mark_page_accessed(pfn_to_page(pfn));
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1238
1239 void kvm_get_pfn(pfn_t pfn)
1240 {
1241         if (!kvm_is_mmio_pfn(pfn))
1242                 get_page(pfn_to_page(pfn));
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1245
1246 static int next_segment(unsigned long len, int offset)
1247 {
1248         if (len > PAGE_SIZE - offset)
1249                 return PAGE_SIZE - offset;
1250         else
1251                 return len;
1252 }
1253
1254 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1255                         int len)
1256 {
1257         int r;
1258         unsigned long addr;
1259
1260         addr = gfn_to_hva(kvm, gfn);
1261         if (kvm_is_error_hva(addr))
1262                 return -EFAULT;
1263         r = __copy_from_user(data, (void __user *)addr + offset, len);
1264         if (r)
1265                 return -EFAULT;
1266         return 0;
1267 }
1268 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1269
1270 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1271 {
1272         gfn_t gfn = gpa >> PAGE_SHIFT;
1273         int seg;
1274         int offset = offset_in_page(gpa);
1275         int ret;
1276
1277         while ((seg = next_segment(len, offset)) != 0) {
1278                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1279                 if (ret < 0)
1280                         return ret;
1281                 offset = 0;
1282                 len -= seg;
1283                 data += seg;
1284                 ++gfn;
1285         }
1286         return 0;
1287 }
1288 EXPORT_SYMBOL_GPL(kvm_read_guest);
1289
1290 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1291                           unsigned long len)
1292 {
1293         int r;
1294         unsigned long addr;
1295         gfn_t gfn = gpa >> PAGE_SHIFT;
1296         int offset = offset_in_page(gpa);
1297
1298         addr = gfn_to_hva(kvm, gfn);
1299         if (kvm_is_error_hva(addr))
1300                 return -EFAULT;
1301         pagefault_disable();
1302         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1303         pagefault_enable();
1304         if (r)
1305                 return -EFAULT;
1306         return 0;
1307 }
1308 EXPORT_SYMBOL(kvm_read_guest_atomic);
1309
1310 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1311                          int offset, int len)
1312 {
1313         int r;
1314         unsigned long addr;
1315
1316         addr = gfn_to_hva(kvm, gfn);
1317         if (kvm_is_error_hva(addr))
1318                 return -EFAULT;
1319         r = __copy_to_user((void __user *)addr + offset, data, len);
1320         if (r)
1321                 return -EFAULT;
1322         mark_page_dirty(kvm, gfn);
1323         return 0;
1324 }
1325 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1326
1327 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1328                     unsigned long len)
1329 {
1330         gfn_t gfn = gpa >> PAGE_SHIFT;
1331         int seg;
1332         int offset = offset_in_page(gpa);
1333         int ret;
1334
1335         while ((seg = next_segment(len, offset)) != 0) {
1336                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1337                 if (ret < 0)
1338                         return ret;
1339                 offset = 0;
1340                 len -= seg;
1341                 data += seg;
1342                 ++gfn;
1343         }
1344         return 0;
1345 }
1346
1347 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1348                               gpa_t gpa)
1349 {
1350         struct kvm_memslots *slots = kvm_memslots(kvm);
1351         int offset = offset_in_page(gpa);
1352         gfn_t gfn = gpa >> PAGE_SHIFT;
1353
1354         ghc->gpa = gpa;
1355         ghc->generation = slots->generation;
1356         ghc->memslot = gfn_to_memslot(kvm, gfn);
1357         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1358         if (!kvm_is_error_hva(ghc->hva))
1359                 ghc->hva += offset;
1360         else
1361                 return -EFAULT;
1362
1363         return 0;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1366
1367 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1368                            void *data, unsigned long len)
1369 {
1370         struct kvm_memslots *slots = kvm_memslots(kvm);
1371         int r;
1372
1373         if (slots->generation != ghc->generation)
1374                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1375
1376         if (kvm_is_error_hva(ghc->hva))
1377                 return -EFAULT;
1378
1379         r = __copy_to_user((void __user *)ghc->hva, data, len);
1380         if (r)
1381                 return -EFAULT;
1382         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1383
1384         return 0;
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1387
1388 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1389                            void *data, unsigned long len)
1390 {
1391         struct kvm_memslots *slots = kvm_memslots(kvm);
1392         int r;
1393
1394         if (slots->generation != ghc->generation)
1395                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1396
1397         if (kvm_is_error_hva(ghc->hva))
1398                 return -EFAULT;
1399
1400         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1401         if (r)
1402                 return -EFAULT;
1403
1404         return 0;
1405 }
1406 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1407
1408 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1409 {
1410         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1411                                     offset, len);
1412 }
1413 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1414
1415 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1416 {
1417         gfn_t gfn = gpa >> PAGE_SHIFT;
1418         int seg;
1419         int offset = offset_in_page(gpa);
1420         int ret;
1421
1422         while ((seg = next_segment(len, offset)) != 0) {
1423                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1424                 if (ret < 0)
1425                         return ret;
1426                 offset = 0;
1427                 len -= seg;
1428                 ++gfn;
1429         }
1430         return 0;
1431 }
1432 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1433
1434 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1435                              gfn_t gfn)
1436 {
1437         if (memslot && memslot->dirty_bitmap) {
1438                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1439
1440                 /* TODO: introduce set_bit_le() and use it */
1441                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1442         }
1443 }
1444
1445 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1446 {
1447         struct kvm_memory_slot *memslot;
1448
1449         memslot = gfn_to_memslot(kvm, gfn);
1450         mark_page_dirty_in_slot(kvm, memslot, gfn);
1451 }
1452
1453 /*
1454  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1455  */
1456 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1457 {
1458         DEFINE_WAIT(wait);
1459
1460         for (;;) {
1461                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1462
1463                 if (kvm_arch_vcpu_runnable(vcpu)) {
1464                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1465                         break;
1466                 }
1467                 if (kvm_cpu_has_pending_timer(vcpu))
1468                         break;
1469                 if (signal_pending(current))
1470                         break;
1471
1472                 schedule();
1473         }
1474
1475         finish_wait(&vcpu->wq, &wait);
1476 }
1477
1478 #ifndef CONFIG_S390
1479 /*
1480  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1481  */
1482 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1483 {
1484         int me;
1485         int cpu = vcpu->cpu;
1486         wait_queue_head_t *wqp;
1487
1488         wqp = kvm_arch_vcpu_wq(vcpu);
1489         if (waitqueue_active(wqp)) {
1490                 wake_up_interruptible(wqp);
1491                 ++vcpu->stat.halt_wakeup;
1492         }
1493
1494         me = get_cpu();
1495         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1496                 if (kvm_arch_vcpu_should_kick(vcpu))
1497                         smp_send_reschedule(cpu);
1498         put_cpu();
1499 }
1500 #endif /* !CONFIG_S390 */
1501
1502 void kvm_resched(struct kvm_vcpu *vcpu)
1503 {
1504         if (!need_resched())
1505                 return;
1506         cond_resched();
1507 }
1508 EXPORT_SYMBOL_GPL(kvm_resched);
1509
1510 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1511 {
1512         struct pid *pid;
1513         struct task_struct *task = NULL;
1514
1515         rcu_read_lock();
1516         pid = rcu_dereference(target->pid);
1517         if (pid)
1518                 task = get_pid_task(target->pid, PIDTYPE_PID);
1519         rcu_read_unlock();
1520         if (!task)
1521                 return false;
1522         if (task->flags & PF_VCPU) {
1523                 put_task_struct(task);
1524                 return false;
1525         }
1526         if (yield_to(task, 1)) {
1527                 put_task_struct(task);
1528                 return true;
1529         }
1530         put_task_struct(task);
1531         return false;
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1534
1535 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1536 /*
1537  * Helper that checks whether a VCPU is eligible for directed yield.
1538  * Most eligible candidate to yield is decided by following heuristics:
1539  *
1540  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1541  *  (preempted lock holder), indicated by @in_spin_loop.
1542  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1543  *
1544  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1545  *  chance last time (mostly it has become eligible now since we have probably
1546  *  yielded to lockholder in last iteration. This is done by toggling
1547  *  @dy_eligible each time a VCPU checked for eligibility.)
1548  *
1549  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1550  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1551  *  burning. Giving priority for a potential lock-holder increases lock
1552  *  progress.
1553  *
1554  *  Since algorithm is based on heuristics, accessing another VCPU data without
1555  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1556  *  and continue with next VCPU and so on.
1557  */
1558 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1559 {
1560         bool eligible;
1561
1562         eligible = !vcpu->spin_loop.in_spin_loop ||
1563                         (vcpu->spin_loop.in_spin_loop &&
1564                          vcpu->spin_loop.dy_eligible);
1565
1566         if (vcpu->spin_loop.in_spin_loop)
1567                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1568
1569         return eligible;
1570 }
1571 #endif
1572 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1573 {
1574         struct kvm *kvm = me->kvm;
1575         struct kvm_vcpu *vcpu;
1576         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1577         int yielded = 0;
1578         int pass;
1579         int i;
1580
1581         kvm_vcpu_set_in_spin_loop(me, true);
1582         /*
1583          * We boost the priority of a VCPU that is runnable but not
1584          * currently running, because it got preempted by something
1585          * else and called schedule in __vcpu_run.  Hopefully that
1586          * VCPU is holding the lock that we need and will release it.
1587          * We approximate round-robin by starting at the last boosted VCPU.
1588          */
1589         for (pass = 0; pass < 2 && !yielded; pass++) {
1590                 kvm_for_each_vcpu(i, vcpu, kvm) {
1591                         if (!pass && i <= last_boosted_vcpu) {
1592                                 i = last_boosted_vcpu;
1593                                 continue;
1594                         } else if (pass && i > last_boosted_vcpu)
1595                                 break;
1596                         if (vcpu == me)
1597                                 continue;
1598                         if (waitqueue_active(&vcpu->wq))
1599                                 continue;
1600                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1601                                 continue;
1602                         if (kvm_vcpu_yield_to(vcpu)) {
1603                                 kvm->last_boosted_vcpu = i;
1604                                 yielded = 1;
1605                                 break;
1606                         }
1607                 }
1608         }
1609         kvm_vcpu_set_in_spin_loop(me, false);
1610
1611         /* Ensure vcpu is not eligible during next spinloop */
1612         kvm_vcpu_set_dy_eligible(me, false);
1613 }
1614 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1615
1616 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1617 {
1618         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1619         struct page *page;
1620
1621         if (vmf->pgoff == 0)
1622                 page = virt_to_page(vcpu->run);
1623 #ifdef CONFIG_X86
1624         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1625                 page = virt_to_page(vcpu->arch.pio_data);
1626 #endif
1627 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1628         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1629                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1630 #endif
1631         else
1632                 return kvm_arch_vcpu_fault(vcpu, vmf);
1633         get_page(page);
1634         vmf->page = page;
1635         return 0;
1636 }
1637
1638 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1639         .fault = kvm_vcpu_fault,
1640 };
1641
1642 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1643 {
1644         vma->vm_ops = &kvm_vcpu_vm_ops;
1645         return 0;
1646 }
1647
1648 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1649 {
1650         struct kvm_vcpu *vcpu = filp->private_data;
1651
1652         kvm_put_kvm(vcpu->kvm);
1653         return 0;
1654 }
1655
1656 static struct file_operations kvm_vcpu_fops = {
1657         .release        = kvm_vcpu_release,
1658         .unlocked_ioctl = kvm_vcpu_ioctl,
1659 #ifdef CONFIG_COMPAT
1660         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1661 #endif
1662         .mmap           = kvm_vcpu_mmap,
1663         .llseek         = noop_llseek,
1664 };
1665
1666 /*
1667  * Allocates an inode for the vcpu.
1668  */
1669 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1670 {
1671         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1672 }
1673
1674 /*
1675  * Creates some virtual cpus.  Good luck creating more than one.
1676  */
1677 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1678 {
1679         int r;
1680         struct kvm_vcpu *vcpu, *v;
1681
1682         vcpu = kvm_arch_vcpu_create(kvm, id);
1683         if (IS_ERR(vcpu))
1684                 return PTR_ERR(vcpu);
1685
1686         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1687
1688         r = kvm_arch_vcpu_setup(vcpu);
1689         if (r)
1690                 goto vcpu_destroy;
1691
1692         mutex_lock(&kvm->lock);
1693         if (!kvm_vcpu_compatible(vcpu)) {
1694                 r = -EINVAL;
1695                 goto unlock_vcpu_destroy;
1696         }
1697         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1698                 r = -EINVAL;
1699                 goto unlock_vcpu_destroy;
1700         }
1701
1702         kvm_for_each_vcpu(r, v, kvm)
1703                 if (v->vcpu_id == id) {
1704                         r = -EEXIST;
1705                         goto unlock_vcpu_destroy;
1706                 }
1707
1708         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1709
1710         /* Now it's all set up, let userspace reach it */
1711         kvm_get_kvm(kvm);
1712         r = create_vcpu_fd(vcpu);
1713         if (r < 0) {
1714                 kvm_put_kvm(kvm);
1715                 goto unlock_vcpu_destroy;
1716         }
1717
1718         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1719         smp_wmb();
1720         atomic_inc(&kvm->online_vcpus);
1721
1722         mutex_unlock(&kvm->lock);
1723         return r;
1724
1725 unlock_vcpu_destroy:
1726         mutex_unlock(&kvm->lock);
1727 vcpu_destroy:
1728         kvm_arch_vcpu_destroy(vcpu);
1729         return r;
1730 }
1731
1732 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1733 {
1734         if (sigset) {
1735                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1736                 vcpu->sigset_active = 1;
1737                 vcpu->sigset = *sigset;
1738         } else
1739                 vcpu->sigset_active = 0;
1740         return 0;
1741 }
1742
1743 static long kvm_vcpu_ioctl(struct file *filp,
1744                            unsigned int ioctl, unsigned long arg)
1745 {
1746         struct kvm_vcpu *vcpu = filp->private_data;
1747         void __user *argp = (void __user *)arg;
1748         int r;
1749         struct kvm_fpu *fpu = NULL;
1750         struct kvm_sregs *kvm_sregs = NULL;
1751
1752         if (vcpu->kvm->mm != current->mm)
1753                 return -EIO;
1754
1755 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1756         /*
1757          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1758          * so vcpu_load() would break it.
1759          */
1760         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1761                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1762 #endif
1763
1764
1765         vcpu_load(vcpu);
1766         switch (ioctl) {
1767         case KVM_RUN:
1768                 r = -EINVAL;
1769                 if (arg)
1770                         goto out;
1771                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1772                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1773                 break;
1774         case KVM_GET_REGS: {
1775                 struct kvm_regs *kvm_regs;
1776
1777                 r = -ENOMEM;
1778                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1779                 if (!kvm_regs)
1780                         goto out;
1781                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1782                 if (r)
1783                         goto out_free1;
1784                 r = -EFAULT;
1785                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1786                         goto out_free1;
1787                 r = 0;
1788 out_free1:
1789                 kfree(kvm_regs);
1790                 break;
1791         }
1792         case KVM_SET_REGS: {
1793                 struct kvm_regs *kvm_regs;
1794
1795                 r = -ENOMEM;
1796                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1797                 if (IS_ERR(kvm_regs)) {
1798                         r = PTR_ERR(kvm_regs);
1799                         goto out;
1800                 }
1801                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1802                 if (r)
1803                         goto out_free2;
1804                 r = 0;
1805 out_free2:
1806                 kfree(kvm_regs);
1807                 break;
1808         }
1809         case KVM_GET_SREGS: {
1810                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1811                 r = -ENOMEM;
1812                 if (!kvm_sregs)
1813                         goto out;
1814                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1815                 if (r)
1816                         goto out;
1817                 r = -EFAULT;
1818                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1819                         goto out;
1820                 r = 0;
1821                 break;
1822         }
1823         case KVM_SET_SREGS: {
1824                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1825                 if (IS_ERR(kvm_sregs)) {
1826                         r = PTR_ERR(kvm_sregs);
1827                         goto out;
1828                 }
1829                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1830                 if (r)
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_GET_MP_STATE: {
1836                 struct kvm_mp_state mp_state;
1837
1838                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1839                 if (r)
1840                         goto out;
1841                 r = -EFAULT;
1842                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_SET_MP_STATE: {
1848                 struct kvm_mp_state mp_state;
1849
1850                 r = -EFAULT;
1851                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1852                         goto out;
1853                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1854                 if (r)
1855                         goto out;
1856                 r = 0;
1857                 break;
1858         }
1859         case KVM_TRANSLATE: {
1860                 struct kvm_translation tr;
1861
1862                 r = -EFAULT;
1863                 if (copy_from_user(&tr, argp, sizeof tr))
1864                         goto out;
1865                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1866                 if (r)
1867                         goto out;
1868                 r = -EFAULT;
1869                 if (copy_to_user(argp, &tr, sizeof tr))
1870                         goto out;
1871                 r = 0;
1872                 break;
1873         }
1874         case KVM_SET_GUEST_DEBUG: {
1875                 struct kvm_guest_debug dbg;
1876
1877                 r = -EFAULT;
1878                 if (copy_from_user(&dbg, argp, sizeof dbg))
1879                         goto out;
1880                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1881                 if (r)
1882                         goto out;
1883                 r = 0;
1884                 break;
1885         }
1886         case KVM_SET_SIGNAL_MASK: {
1887                 struct kvm_signal_mask __user *sigmask_arg = argp;
1888                 struct kvm_signal_mask kvm_sigmask;
1889                 sigset_t sigset, *p;
1890
1891                 p = NULL;
1892                 if (argp) {
1893                         r = -EFAULT;
1894                         if (copy_from_user(&kvm_sigmask, argp,
1895                                            sizeof kvm_sigmask))
1896                                 goto out;
1897                         r = -EINVAL;
1898                         if (kvm_sigmask.len != sizeof sigset)
1899                                 goto out;
1900                         r = -EFAULT;
1901                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1902                                            sizeof sigset))
1903                                 goto out;
1904                         p = &sigset;
1905                 }
1906                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1907                 break;
1908         }
1909         case KVM_GET_FPU: {
1910                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1911                 r = -ENOMEM;
1912                 if (!fpu)
1913                         goto out;
1914                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1915                 if (r)
1916                         goto out;
1917                 r = -EFAULT;
1918                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1919                         goto out;
1920                 r = 0;
1921                 break;
1922         }
1923         case KVM_SET_FPU: {
1924                 fpu = memdup_user(argp, sizeof(*fpu));
1925                 if (IS_ERR(fpu)) {
1926                         r = PTR_ERR(fpu);
1927                         goto out;
1928                 }
1929                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1930                 if (r)
1931                         goto out;
1932                 r = 0;
1933                 break;
1934         }
1935         default:
1936                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1937         }
1938 out:
1939         vcpu_put(vcpu);
1940         kfree(fpu);
1941         kfree(kvm_sregs);
1942         return r;
1943 }
1944
1945 #ifdef CONFIG_COMPAT
1946 static long kvm_vcpu_compat_ioctl(struct file *filp,
1947                                   unsigned int ioctl, unsigned long arg)
1948 {
1949         struct kvm_vcpu *vcpu = filp->private_data;
1950         void __user *argp = compat_ptr(arg);
1951         int r;
1952
1953         if (vcpu->kvm->mm != current->mm)
1954                 return -EIO;
1955
1956         switch (ioctl) {
1957         case KVM_SET_SIGNAL_MASK: {
1958                 struct kvm_signal_mask __user *sigmask_arg = argp;
1959                 struct kvm_signal_mask kvm_sigmask;
1960                 compat_sigset_t csigset;
1961                 sigset_t sigset;
1962
1963                 if (argp) {
1964                         r = -EFAULT;
1965                         if (copy_from_user(&kvm_sigmask, argp,
1966                                            sizeof kvm_sigmask))
1967                                 goto out;
1968                         r = -EINVAL;
1969                         if (kvm_sigmask.len != sizeof csigset)
1970                                 goto out;
1971                         r = -EFAULT;
1972                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1973                                            sizeof csigset))
1974                                 goto out;
1975                 }
1976                 sigset_from_compat(&sigset, &csigset);
1977                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1978                 break;
1979         }
1980         default:
1981                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1982         }
1983
1984 out:
1985         return r;
1986 }
1987 #endif
1988
1989 static long kvm_vm_ioctl(struct file *filp,
1990                            unsigned int ioctl, unsigned long arg)
1991 {
1992         struct kvm *kvm = filp->private_data;
1993         void __user *argp = (void __user *)arg;
1994         int r;
1995
1996         if (kvm->mm != current->mm)
1997                 return -EIO;
1998         switch (ioctl) {
1999         case KVM_CREATE_VCPU:
2000                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2001                 if (r < 0)
2002                         goto out;
2003                 break;
2004         case KVM_SET_USER_MEMORY_REGION: {
2005                 struct kvm_userspace_memory_region kvm_userspace_mem;
2006
2007                 r = -EFAULT;
2008                 if (copy_from_user(&kvm_userspace_mem, argp,
2009                                                 sizeof kvm_userspace_mem))
2010                         goto out;
2011
2012                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2013                 if (r)
2014                         goto out;
2015                 break;
2016         }
2017         case KVM_GET_DIRTY_LOG: {
2018                 struct kvm_dirty_log log;
2019
2020                 r = -EFAULT;
2021                 if (copy_from_user(&log, argp, sizeof log))
2022                         goto out;
2023                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2024                 if (r)
2025                         goto out;
2026                 break;
2027         }
2028 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2029         case KVM_REGISTER_COALESCED_MMIO: {
2030                 struct kvm_coalesced_mmio_zone zone;
2031                 r = -EFAULT;
2032                 if (copy_from_user(&zone, argp, sizeof zone))
2033                         goto out;
2034                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2035                 if (r)
2036                         goto out;
2037                 r = 0;
2038                 break;
2039         }
2040         case KVM_UNREGISTER_COALESCED_MMIO: {
2041                 struct kvm_coalesced_mmio_zone zone;
2042                 r = -EFAULT;
2043                 if (copy_from_user(&zone, argp, sizeof zone))
2044                         goto out;
2045                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2046                 if (r)
2047                         goto out;
2048                 r = 0;
2049                 break;
2050         }
2051 #endif
2052         case KVM_IRQFD: {
2053                 struct kvm_irqfd data;
2054
2055                 r = -EFAULT;
2056                 if (copy_from_user(&data, argp, sizeof data))
2057                         goto out;
2058                 r = kvm_irqfd(kvm, &data);
2059                 break;
2060         }
2061         case KVM_IOEVENTFD: {
2062                 struct kvm_ioeventfd data;
2063
2064                 r = -EFAULT;
2065                 if (copy_from_user(&data, argp, sizeof data))
2066                         goto out;
2067                 r = kvm_ioeventfd(kvm, &data);
2068                 break;
2069         }
2070 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2071         case KVM_SET_BOOT_CPU_ID:
2072                 r = 0;
2073                 mutex_lock(&kvm->lock);
2074                 if (atomic_read(&kvm->online_vcpus) != 0)
2075                         r = -EBUSY;
2076                 else
2077                         kvm->bsp_vcpu_id = arg;
2078                 mutex_unlock(&kvm->lock);
2079                 break;
2080 #endif
2081 #ifdef CONFIG_HAVE_KVM_MSI
2082         case KVM_SIGNAL_MSI: {
2083                 struct kvm_msi msi;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&msi, argp, sizeof msi))
2087                         goto out;
2088                 r = kvm_send_userspace_msi(kvm, &msi);
2089                 break;
2090         }
2091 #endif
2092 #ifdef __KVM_HAVE_IRQ_LINE
2093         case KVM_IRQ_LINE_STATUS:
2094         case KVM_IRQ_LINE: {
2095                 struct kvm_irq_level irq_event;
2096
2097                 r = -EFAULT;
2098                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2099                         goto out;
2100
2101                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2102                 if (r)
2103                         goto out;
2104
2105                 r = -EFAULT;
2106                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2107                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2108                                 goto out;
2109                 }
2110
2111                 r = 0;
2112                 break;
2113         }
2114 #endif
2115         default:
2116                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2117                 if (r == -ENOTTY)
2118                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2119         }
2120 out:
2121         return r;
2122 }
2123
2124 #ifdef CONFIG_COMPAT
2125 struct compat_kvm_dirty_log {
2126         __u32 slot;
2127         __u32 padding1;
2128         union {
2129                 compat_uptr_t dirty_bitmap; /* one bit per page */
2130                 __u64 padding2;
2131         };
2132 };
2133
2134 static long kvm_vm_compat_ioctl(struct file *filp,
2135                            unsigned int ioctl, unsigned long arg)
2136 {
2137         struct kvm *kvm = filp->private_data;
2138         int r;
2139
2140         if (kvm->mm != current->mm)
2141                 return -EIO;
2142         switch (ioctl) {
2143         case KVM_GET_DIRTY_LOG: {
2144                 struct compat_kvm_dirty_log compat_log;
2145                 struct kvm_dirty_log log;
2146
2147                 r = -EFAULT;
2148                 if (copy_from_user(&compat_log, (void __user *)arg,
2149                                    sizeof(compat_log)))
2150                         goto out;
2151                 log.slot         = compat_log.slot;
2152                 log.padding1     = compat_log.padding1;
2153                 log.padding2     = compat_log.padding2;
2154                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2155
2156                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2157                 if (r)
2158                         goto out;
2159                 break;
2160         }
2161         default:
2162                 r = kvm_vm_ioctl(filp, ioctl, arg);
2163         }
2164
2165 out:
2166         return r;
2167 }
2168 #endif
2169
2170 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2171 {
2172         struct page *page[1];
2173         unsigned long addr;
2174         int npages;
2175         gfn_t gfn = vmf->pgoff;
2176         struct kvm *kvm = vma->vm_file->private_data;
2177
2178         addr = gfn_to_hva(kvm, gfn);
2179         if (kvm_is_error_hva(addr))
2180                 return VM_FAULT_SIGBUS;
2181
2182         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2183                                 NULL);
2184         if (unlikely(npages != 1))
2185                 return VM_FAULT_SIGBUS;
2186
2187         vmf->page = page[0];
2188         return 0;
2189 }
2190
2191 static const struct vm_operations_struct kvm_vm_vm_ops = {
2192         .fault = kvm_vm_fault,
2193 };
2194
2195 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2196 {
2197         vma->vm_ops = &kvm_vm_vm_ops;
2198         return 0;
2199 }
2200
2201 static struct file_operations kvm_vm_fops = {
2202         .release        = kvm_vm_release,
2203         .unlocked_ioctl = kvm_vm_ioctl,
2204 #ifdef CONFIG_COMPAT
2205         .compat_ioctl   = kvm_vm_compat_ioctl,
2206 #endif
2207         .mmap           = kvm_vm_mmap,
2208         .llseek         = noop_llseek,
2209 };
2210
2211 static int kvm_dev_ioctl_create_vm(unsigned long type)
2212 {
2213         int r;
2214         struct kvm *kvm;
2215
2216         kvm = kvm_create_vm(type);
2217         if (IS_ERR(kvm))
2218                 return PTR_ERR(kvm);
2219 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2220         r = kvm_coalesced_mmio_init(kvm);
2221         if (r < 0) {
2222                 kvm_put_kvm(kvm);
2223                 return r;
2224         }
2225 #endif
2226         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2227         if (r < 0)
2228                 kvm_put_kvm(kvm);
2229
2230         return r;
2231 }
2232
2233 static long kvm_dev_ioctl_check_extension_generic(long arg)
2234 {
2235         switch (arg) {
2236         case KVM_CAP_USER_MEMORY:
2237         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2238         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2239 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2240         case KVM_CAP_SET_BOOT_CPU_ID:
2241 #endif
2242         case KVM_CAP_INTERNAL_ERROR_DATA:
2243 #ifdef CONFIG_HAVE_KVM_MSI
2244         case KVM_CAP_SIGNAL_MSI:
2245 #endif
2246                 return 1;
2247 #ifdef KVM_CAP_IRQ_ROUTING
2248         case KVM_CAP_IRQ_ROUTING:
2249                 return KVM_MAX_IRQ_ROUTES;
2250 #endif
2251         default:
2252                 break;
2253         }
2254         return kvm_dev_ioctl_check_extension(arg);
2255 }
2256
2257 static long kvm_dev_ioctl(struct file *filp,
2258                           unsigned int ioctl, unsigned long arg)
2259 {
2260         long r = -EINVAL;
2261
2262         switch (ioctl) {
2263         case KVM_GET_API_VERSION:
2264                 r = -EINVAL;
2265                 if (arg)
2266                         goto out;
2267                 r = KVM_API_VERSION;
2268                 break;
2269         case KVM_CREATE_VM:
2270                 r = kvm_dev_ioctl_create_vm(arg);
2271                 break;
2272         case KVM_CHECK_EXTENSION:
2273                 r = kvm_dev_ioctl_check_extension_generic(arg);
2274                 break;
2275         case KVM_GET_VCPU_MMAP_SIZE:
2276                 r = -EINVAL;
2277                 if (arg)
2278                         goto out;
2279                 r = PAGE_SIZE;     /* struct kvm_run */
2280 #ifdef CONFIG_X86
2281                 r += PAGE_SIZE;    /* pio data page */
2282 #endif
2283 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2284                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2285 #endif
2286                 break;
2287         case KVM_TRACE_ENABLE:
2288         case KVM_TRACE_PAUSE:
2289         case KVM_TRACE_DISABLE:
2290                 r = -EOPNOTSUPP;
2291                 break;
2292         default:
2293                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2294         }
2295 out:
2296         return r;
2297 }
2298
2299 static struct file_operations kvm_chardev_ops = {
2300         .unlocked_ioctl = kvm_dev_ioctl,
2301         .compat_ioctl   = kvm_dev_ioctl,
2302         .llseek         = noop_llseek,
2303 };
2304
2305 static struct miscdevice kvm_dev = {
2306         KVM_MINOR,
2307         "kvm",
2308         &kvm_chardev_ops,
2309 };
2310
2311 static void hardware_enable_nolock(void *junk)
2312 {
2313         int cpu = raw_smp_processor_id();
2314         int r;
2315
2316         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2317                 return;
2318
2319         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2320
2321         r = kvm_arch_hardware_enable(NULL);
2322
2323         if (r) {
2324                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2325                 atomic_inc(&hardware_enable_failed);
2326                 printk(KERN_INFO "kvm: enabling virtualization on "
2327                                  "CPU%d failed\n", cpu);
2328         }
2329 }
2330
2331 static void hardware_enable(void *junk)
2332 {
2333         raw_spin_lock(&kvm_lock);
2334         hardware_enable_nolock(junk);
2335         raw_spin_unlock(&kvm_lock);
2336 }
2337
2338 static void hardware_disable_nolock(void *junk)
2339 {
2340         int cpu = raw_smp_processor_id();
2341
2342         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2343                 return;
2344         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2345         kvm_arch_hardware_disable(NULL);
2346 }
2347
2348 static void hardware_disable(void *junk)
2349 {
2350         raw_spin_lock(&kvm_lock);
2351         hardware_disable_nolock(junk);
2352         raw_spin_unlock(&kvm_lock);
2353 }
2354
2355 static void hardware_disable_all_nolock(void)
2356 {
2357         BUG_ON(!kvm_usage_count);
2358
2359         kvm_usage_count--;
2360         if (!kvm_usage_count)
2361                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2362 }
2363
2364 static void hardware_disable_all(void)
2365 {
2366         raw_spin_lock(&kvm_lock);
2367         hardware_disable_all_nolock();
2368         raw_spin_unlock(&kvm_lock);
2369 }
2370
2371 static int hardware_enable_all(void)
2372 {
2373         int r = 0;
2374
2375         raw_spin_lock(&kvm_lock);
2376
2377         kvm_usage_count++;
2378         if (kvm_usage_count == 1) {
2379                 atomic_set(&hardware_enable_failed, 0);
2380                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2381
2382                 if (atomic_read(&hardware_enable_failed)) {
2383                         hardware_disable_all_nolock();
2384                         r = -EBUSY;
2385                 }
2386         }
2387
2388         raw_spin_unlock(&kvm_lock);
2389
2390         return r;
2391 }
2392
2393 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2394                            void *v)
2395 {
2396         int cpu = (long)v;
2397
2398         if (!kvm_usage_count)
2399                 return NOTIFY_OK;
2400
2401         val &= ~CPU_TASKS_FROZEN;
2402         switch (val) {
2403         case CPU_DYING:
2404                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2405                        cpu);
2406                 hardware_disable(NULL);
2407                 break;
2408         case CPU_STARTING:
2409                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2410                        cpu);
2411                 hardware_enable(NULL);
2412                 break;
2413         }
2414         return NOTIFY_OK;
2415 }
2416
2417
2418 asmlinkage void kvm_spurious_fault(void)
2419 {
2420         /* Fault while not rebooting.  We want the trace. */
2421         BUG();
2422 }
2423 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2424
2425 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2426                       void *v)
2427 {
2428         /*
2429          * Some (well, at least mine) BIOSes hang on reboot if
2430          * in vmx root mode.
2431          *
2432          * And Intel TXT required VMX off for all cpu when system shutdown.
2433          */
2434         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2435         kvm_rebooting = true;
2436         on_each_cpu(hardware_disable_nolock, NULL, 1);
2437         return NOTIFY_OK;
2438 }
2439
2440 static struct notifier_block kvm_reboot_notifier = {
2441         .notifier_call = kvm_reboot,
2442         .priority = 0,
2443 };
2444
2445 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2446 {
2447         int i;
2448
2449         for (i = 0; i < bus->dev_count; i++) {
2450                 struct kvm_io_device *pos = bus->range[i].dev;
2451
2452                 kvm_iodevice_destructor(pos);
2453         }
2454         kfree(bus);
2455 }
2456
2457 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2458 {
2459         const struct kvm_io_range *r1 = p1;
2460         const struct kvm_io_range *r2 = p2;
2461
2462         if (r1->addr < r2->addr)
2463                 return -1;
2464         if (r1->addr + r1->len > r2->addr + r2->len)
2465                 return 1;
2466         return 0;
2467 }
2468
2469 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2470                           gpa_t addr, int len)
2471 {
2472         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2473                 .addr = addr,
2474                 .len = len,
2475                 .dev = dev,
2476         };
2477
2478         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2479                 kvm_io_bus_sort_cmp, NULL);
2480
2481         return 0;
2482 }
2483
2484 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2485                              gpa_t addr, int len)
2486 {
2487         struct kvm_io_range *range, key;
2488         int off;
2489
2490         key = (struct kvm_io_range) {
2491                 .addr = addr,
2492                 .len = len,
2493         };
2494
2495         range = bsearch(&key, bus->range, bus->dev_count,
2496                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2497         if (range == NULL)
2498                 return -ENOENT;
2499
2500         off = range - bus->range;
2501
2502         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2503                 off--;
2504
2505         return off;
2506 }
2507
2508 /* kvm_io_bus_write - called under kvm->slots_lock */
2509 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2510                      int len, const void *val)
2511 {
2512         int idx;
2513         struct kvm_io_bus *bus;
2514         struct kvm_io_range range;
2515
2516         range = (struct kvm_io_range) {
2517                 .addr = addr,
2518                 .len = len,
2519         };
2520
2521         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2522         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2523         if (idx < 0)
2524                 return -EOPNOTSUPP;
2525
2526         while (idx < bus->dev_count &&
2527                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2528                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2529                         return 0;
2530                 idx++;
2531         }
2532
2533         return -EOPNOTSUPP;
2534 }
2535
2536 /* kvm_io_bus_read - called under kvm->slots_lock */
2537 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2538                     int len, void *val)
2539 {
2540         int idx;
2541         struct kvm_io_bus *bus;
2542         struct kvm_io_range range;
2543
2544         range = (struct kvm_io_range) {
2545                 .addr = addr,
2546                 .len = len,
2547         };
2548
2549         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2550         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2551         if (idx < 0)
2552                 return -EOPNOTSUPP;
2553
2554         while (idx < bus->dev_count &&
2555                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2556                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2557                         return 0;
2558                 idx++;
2559         }
2560
2561         return -EOPNOTSUPP;
2562 }
2563
2564 /* Caller must hold slots_lock. */
2565 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2566                             int len, struct kvm_io_device *dev)
2567 {
2568         struct kvm_io_bus *new_bus, *bus;
2569
2570         bus = kvm->buses[bus_idx];
2571         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2572                 return -ENOSPC;
2573
2574         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2575                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2576         if (!new_bus)
2577                 return -ENOMEM;
2578         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2579                sizeof(struct kvm_io_range)));
2580         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2581         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2582         synchronize_srcu_expedited(&kvm->srcu);
2583         kfree(bus);
2584
2585         return 0;
2586 }
2587
2588 /* Caller must hold slots_lock. */
2589 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2590                               struct kvm_io_device *dev)
2591 {
2592         int i, r;
2593         struct kvm_io_bus *new_bus, *bus;
2594
2595         bus = kvm->buses[bus_idx];
2596         r = -ENOENT;
2597         for (i = 0; i < bus->dev_count; i++)
2598                 if (bus->range[i].dev == dev) {
2599                         r = 0;
2600                         break;
2601                 }
2602
2603         if (r)
2604                 return r;
2605
2606         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2607                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2608         if (!new_bus)
2609                 return -ENOMEM;
2610
2611         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2612         new_bus->dev_count--;
2613         memcpy(new_bus->range + i, bus->range + i + 1,
2614                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2615
2616         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2617         synchronize_srcu_expedited(&kvm->srcu);
2618         kfree(bus);
2619         return r;
2620 }
2621
2622 static struct notifier_block kvm_cpu_notifier = {
2623         .notifier_call = kvm_cpu_hotplug,
2624 };
2625
2626 static int vm_stat_get(void *_offset, u64 *val)
2627 {
2628         unsigned offset = (long)_offset;
2629         struct kvm *kvm;
2630
2631         *val = 0;
2632         raw_spin_lock(&kvm_lock);
2633         list_for_each_entry(kvm, &vm_list, vm_list)
2634                 *val += *(u32 *)((void *)kvm + offset);
2635         raw_spin_unlock(&kvm_lock);
2636         return 0;
2637 }
2638
2639 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2640
2641 static int vcpu_stat_get(void *_offset, u64 *val)
2642 {
2643         unsigned offset = (long)_offset;
2644         struct kvm *kvm;
2645         struct kvm_vcpu *vcpu;
2646         int i;
2647
2648         *val = 0;
2649         raw_spin_lock(&kvm_lock);
2650         list_for_each_entry(kvm, &vm_list, vm_list)
2651                 kvm_for_each_vcpu(i, vcpu, kvm)
2652                         *val += *(u32 *)((void *)vcpu + offset);
2653
2654         raw_spin_unlock(&kvm_lock);
2655         return 0;
2656 }
2657
2658 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2659
2660 static const struct file_operations *stat_fops[] = {
2661         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2662         [KVM_STAT_VM]   = &vm_stat_fops,
2663 };
2664
2665 static int kvm_init_debug(void)
2666 {
2667         int r = -EFAULT;
2668         struct kvm_stats_debugfs_item *p;
2669
2670         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2671         if (kvm_debugfs_dir == NULL)
2672                 goto out;
2673
2674         for (p = debugfs_entries; p->name; ++p) {
2675                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2676                                                 (void *)(long)p->offset,
2677                                                 stat_fops[p->kind]);
2678                 if (p->dentry == NULL)
2679                         goto out_dir;
2680         }
2681
2682         return 0;
2683
2684 out_dir:
2685         debugfs_remove_recursive(kvm_debugfs_dir);
2686 out:
2687         return r;
2688 }
2689
2690 static void kvm_exit_debug(void)
2691 {
2692         struct kvm_stats_debugfs_item *p;
2693
2694         for (p = debugfs_entries; p->name; ++p)
2695                 debugfs_remove(p->dentry);
2696         debugfs_remove(kvm_debugfs_dir);
2697 }
2698
2699 static int kvm_suspend(void)
2700 {
2701         if (kvm_usage_count)
2702                 hardware_disable_nolock(NULL);
2703         return 0;
2704 }
2705
2706 static void kvm_resume(void)
2707 {
2708         if (kvm_usage_count) {
2709                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2710                 hardware_enable_nolock(NULL);
2711         }
2712 }
2713
2714 static struct syscore_ops kvm_syscore_ops = {
2715         .suspend = kvm_suspend,
2716         .resume = kvm_resume,
2717 };
2718
2719 static inline
2720 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2721 {
2722         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2723 }
2724
2725 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2726 {
2727         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2728
2729         kvm_arch_vcpu_load(vcpu, cpu);
2730 }
2731
2732 static void kvm_sched_out(struct preempt_notifier *pn,
2733                           struct task_struct *next)
2734 {
2735         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2736
2737         kvm_arch_vcpu_put(vcpu);
2738 }
2739
2740 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2741                   struct module *module)
2742 {
2743         int r;
2744         int cpu;
2745
2746         r = kvm_arch_init(opaque);
2747         if (r)
2748                 goto out_fail;
2749
2750         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2751                 r = -ENOMEM;
2752                 goto out_free_0;
2753         }
2754
2755         r = kvm_arch_hardware_setup();
2756         if (r < 0)
2757                 goto out_free_0a;
2758
2759         for_each_online_cpu(cpu) {
2760                 smp_call_function_single(cpu,
2761                                 kvm_arch_check_processor_compat,
2762                                 &r, 1);
2763                 if (r < 0)
2764                         goto out_free_1;
2765         }
2766
2767         r = register_cpu_notifier(&kvm_cpu_notifier);
2768         if (r)
2769                 goto out_free_2;
2770         register_reboot_notifier(&kvm_reboot_notifier);
2771
2772         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2773         if (!vcpu_align)
2774                 vcpu_align = __alignof__(struct kvm_vcpu);
2775         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2776                                            0, NULL);
2777         if (!kvm_vcpu_cache) {
2778                 r = -ENOMEM;
2779                 goto out_free_3;
2780         }
2781
2782         r = kvm_async_pf_init();
2783         if (r)
2784                 goto out_free;
2785
2786         kvm_chardev_ops.owner = module;
2787         kvm_vm_fops.owner = module;
2788         kvm_vcpu_fops.owner = module;
2789
2790         r = misc_register(&kvm_dev);
2791         if (r) {
2792                 printk(KERN_ERR "kvm: misc device register failed\n");
2793                 goto out_unreg;
2794         }
2795
2796         register_syscore_ops(&kvm_syscore_ops);
2797
2798         kvm_preempt_ops.sched_in = kvm_sched_in;
2799         kvm_preempt_ops.sched_out = kvm_sched_out;
2800
2801         r = kvm_init_debug();
2802         if (r) {
2803                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2804                 goto out_undebugfs;
2805         }
2806
2807         return 0;
2808
2809 out_undebugfs:
2810         unregister_syscore_ops(&kvm_syscore_ops);
2811 out_unreg:
2812         kvm_async_pf_deinit();
2813 out_free:
2814         kmem_cache_destroy(kvm_vcpu_cache);
2815 out_free_3:
2816         unregister_reboot_notifier(&kvm_reboot_notifier);
2817         unregister_cpu_notifier(&kvm_cpu_notifier);
2818 out_free_2:
2819 out_free_1:
2820         kvm_arch_hardware_unsetup();
2821 out_free_0a:
2822         free_cpumask_var(cpus_hardware_enabled);
2823 out_free_0:
2824         kvm_arch_exit();
2825 out_fail:
2826         return r;
2827 }
2828 EXPORT_SYMBOL_GPL(kvm_init);
2829
2830 void kvm_exit(void)
2831 {
2832         kvm_exit_debug();
2833         misc_deregister(&kvm_dev);
2834         kmem_cache_destroy(kvm_vcpu_cache);
2835         kvm_async_pf_deinit();
2836         unregister_syscore_ops(&kvm_syscore_ops);
2837         unregister_reboot_notifier(&kvm_reboot_notifier);
2838         unregister_cpu_notifier(&kvm_cpu_notifier);
2839         on_each_cpu(hardware_disable_nolock, NULL, 1);
2840         kvm_arch_hardware_unsetup();
2841         kvm_arch_exit();
2842         free_cpumask_var(cpus_hardware_enabled);
2843 }
2844 EXPORT_SYMBOL_GPL(kvm_exit);