KVM: introduce KVM_PFN_ERR_BAD
[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 int is_error_page(struct page *page)
926 {
927         return IS_ERR(page);
928 }
929 EXPORT_SYMBOL_GPL(is_error_page);
930
931 int is_error_pfn(pfn_t pfn)
932 {
933         return IS_ERR_VALUE(pfn);
934 }
935 EXPORT_SYMBOL_GPL(is_error_pfn);
936
937 int is_noslot_pfn(pfn_t pfn)
938 {
939         return pfn == -ENOENT;
940 }
941 EXPORT_SYMBOL_GPL(is_noslot_pfn);
942
943 int is_invalid_pfn(pfn_t pfn)
944 {
945         return !is_noslot_pfn(pfn) && is_error_pfn(pfn);
946 }
947 EXPORT_SYMBOL_GPL(is_invalid_pfn);
948
949 struct page *get_bad_page(void)
950 {
951         return ERR_PTR(-ENOENT);
952 }
953
954 static inline unsigned long bad_hva(void)
955 {
956         return PAGE_OFFSET;
957 }
958
959 int kvm_is_error_hva(unsigned long addr)
960 {
961         return addr == bad_hva();
962 }
963 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
964
965 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
966 {
967         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
968 }
969 EXPORT_SYMBOL_GPL(gfn_to_memslot);
970
971 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
972 {
973         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
974
975         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
976               memslot->flags & KVM_MEMSLOT_INVALID)
977                 return 0;
978
979         return 1;
980 }
981 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
982
983 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
984 {
985         struct vm_area_struct *vma;
986         unsigned long addr, size;
987
988         size = PAGE_SIZE;
989
990         addr = gfn_to_hva(kvm, gfn);
991         if (kvm_is_error_hva(addr))
992                 return PAGE_SIZE;
993
994         down_read(&current->mm->mmap_sem);
995         vma = find_vma(current->mm, addr);
996         if (!vma)
997                 goto out;
998
999         size = vma_kernel_pagesize(vma);
1000
1001 out:
1002         up_read(&current->mm->mmap_sem);
1003
1004         return size;
1005 }
1006
1007 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1008                                      gfn_t *nr_pages)
1009 {
1010         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1011                 return bad_hva();
1012
1013         if (nr_pages)
1014                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1015
1016         return gfn_to_hva_memslot(slot, gfn);
1017 }
1018
1019 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1020 {
1021         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1022 }
1023 EXPORT_SYMBOL_GPL(gfn_to_hva);
1024
1025 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1026         unsigned long start, int write, struct page **page)
1027 {
1028         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1029
1030         if (write)
1031                 flags |= FOLL_WRITE;
1032
1033         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1034 }
1035
1036 static inline int check_user_page_hwpoison(unsigned long addr)
1037 {
1038         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1039
1040         rc = __get_user_pages(current, current->mm, addr, 1,
1041                               flags, NULL, NULL, NULL);
1042         return rc == -EHWPOISON;
1043 }
1044
1045 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1046                         bool write_fault, bool *writable)
1047 {
1048         struct page *page[1];
1049         int npages = 0;
1050         pfn_t pfn;
1051
1052         /* we can do it either atomically or asynchronously, not both */
1053         BUG_ON(atomic && async);
1054
1055         BUG_ON(!write_fault && !writable);
1056
1057         if (writable)
1058                 *writable = true;
1059
1060         if (atomic || async)
1061                 npages = __get_user_pages_fast(addr, 1, 1, page);
1062
1063         if (unlikely(npages != 1) && !atomic) {
1064                 might_sleep();
1065
1066                 if (writable)
1067                         *writable = write_fault;
1068
1069                 if (async) {
1070                         down_read(&current->mm->mmap_sem);
1071                         npages = get_user_page_nowait(current, current->mm,
1072                                                      addr, write_fault, page);
1073                         up_read(&current->mm->mmap_sem);
1074                 } else
1075                         npages = get_user_pages_fast(addr, 1, write_fault,
1076                                                      page);
1077
1078                 /* map read fault as writable if possible */
1079                 if (unlikely(!write_fault) && npages == 1) {
1080                         struct page *wpage[1];
1081
1082                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1083                         if (npages == 1) {
1084                                 *writable = true;
1085                                 put_page(page[0]);
1086                                 page[0] = wpage[0];
1087                         }
1088                         npages = 1;
1089                 }
1090         }
1091
1092         if (unlikely(npages != 1)) {
1093                 struct vm_area_struct *vma;
1094
1095                 if (atomic)
1096                         return KVM_PFN_ERR_FAULT;
1097
1098                 down_read(&current->mm->mmap_sem);
1099                 if (npages == -EHWPOISON ||
1100                         (!async && check_user_page_hwpoison(addr))) {
1101                         up_read(&current->mm->mmap_sem);
1102                         return KVM_PFN_ERR_HWPOISON;
1103                 }
1104
1105                 vma = find_vma_intersection(current->mm, addr, addr+1);
1106
1107                 if (vma == NULL)
1108                         pfn = KVM_PFN_ERR_FAULT;
1109                 else if ((vma->vm_flags & VM_PFNMAP)) {
1110                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1111                                 vma->vm_pgoff;
1112                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1113                 } else {
1114                         if (async && (vma->vm_flags & VM_WRITE))
1115                                 *async = true;
1116                         pfn = KVM_PFN_ERR_FAULT;
1117                 }
1118                 up_read(&current->mm->mmap_sem);
1119         } else
1120                 pfn = page_to_pfn(page[0]);
1121
1122         return pfn;
1123 }
1124
1125 pfn_t hva_to_pfn_atomic(unsigned long addr)
1126 {
1127         return hva_to_pfn(addr, true, NULL, true, NULL);
1128 }
1129 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1130
1131 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1132                           bool write_fault, bool *writable)
1133 {
1134         unsigned long addr;
1135
1136         if (async)
1137                 *async = false;
1138
1139         addr = gfn_to_hva(kvm, gfn);
1140         if (kvm_is_error_hva(addr))
1141                 return KVM_PFN_ERR_BAD;
1142
1143         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1144 }
1145
1146 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1147 {
1148         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1149 }
1150 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1151
1152 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1153                        bool write_fault, bool *writable)
1154 {
1155         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1156 }
1157 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1158
1159 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1160 {
1161         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1162 }
1163 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1164
1165 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1166                       bool *writable)
1167 {
1168         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1169 }
1170 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1171
1172 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1173 {
1174         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1175         return hva_to_pfn(addr, false, NULL, true, NULL);
1176 }
1177
1178 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1179                                                                   int nr_pages)
1180 {
1181         unsigned long addr;
1182         gfn_t entry;
1183
1184         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1185         if (kvm_is_error_hva(addr))
1186                 return -1;
1187
1188         if (entry < nr_pages)
1189                 return 0;
1190
1191         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1192 }
1193 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1194
1195 static struct page *kvm_pfn_to_page(pfn_t pfn)
1196 {
1197         WARN_ON(kvm_is_mmio_pfn(pfn));
1198
1199         if (is_error_pfn(pfn) || kvm_is_mmio_pfn(pfn))
1200                 return get_bad_page();
1201
1202         return pfn_to_page(pfn);
1203 }
1204
1205 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1206 {
1207         pfn_t pfn;
1208
1209         pfn = gfn_to_pfn(kvm, gfn);
1210
1211         return kvm_pfn_to_page(pfn);
1212 }
1213
1214 EXPORT_SYMBOL_GPL(gfn_to_page);
1215
1216 void kvm_release_page_clean(struct page *page)
1217 {
1218         if (!is_error_page(page))
1219                 kvm_release_pfn_clean(page_to_pfn(page));
1220 }
1221 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1222
1223 void kvm_release_pfn_clean(pfn_t pfn)
1224 {
1225         if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1226                 put_page(pfn_to_page(pfn));
1227 }
1228 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1229
1230 void kvm_release_page_dirty(struct page *page)
1231 {
1232         WARN_ON(is_error_page(page));
1233
1234         kvm_release_pfn_dirty(page_to_pfn(page));
1235 }
1236 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1237
1238 void kvm_release_pfn_dirty(pfn_t pfn)
1239 {
1240         kvm_set_pfn_dirty(pfn);
1241         kvm_release_pfn_clean(pfn);
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1244
1245 void kvm_set_page_dirty(struct page *page)
1246 {
1247         kvm_set_pfn_dirty(page_to_pfn(page));
1248 }
1249 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1250
1251 void kvm_set_pfn_dirty(pfn_t pfn)
1252 {
1253         if (!kvm_is_mmio_pfn(pfn)) {
1254                 struct page *page = pfn_to_page(pfn);
1255                 if (!PageReserved(page))
1256                         SetPageDirty(page);
1257         }
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1260
1261 void kvm_set_pfn_accessed(pfn_t pfn)
1262 {
1263         if (!kvm_is_mmio_pfn(pfn))
1264                 mark_page_accessed(pfn_to_page(pfn));
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1267
1268 void kvm_get_pfn(pfn_t pfn)
1269 {
1270         if (!kvm_is_mmio_pfn(pfn))
1271                 get_page(pfn_to_page(pfn));
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1274
1275 static int next_segment(unsigned long len, int offset)
1276 {
1277         if (len > PAGE_SIZE - offset)
1278                 return PAGE_SIZE - offset;
1279         else
1280                 return len;
1281 }
1282
1283 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1284                         int len)
1285 {
1286         int r;
1287         unsigned long addr;
1288
1289         addr = gfn_to_hva(kvm, gfn);
1290         if (kvm_is_error_hva(addr))
1291                 return -EFAULT;
1292         r = __copy_from_user(data, (void __user *)addr + offset, len);
1293         if (r)
1294                 return -EFAULT;
1295         return 0;
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1298
1299 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1300 {
1301         gfn_t gfn = gpa >> PAGE_SHIFT;
1302         int seg;
1303         int offset = offset_in_page(gpa);
1304         int ret;
1305
1306         while ((seg = next_segment(len, offset)) != 0) {
1307                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1308                 if (ret < 0)
1309                         return ret;
1310                 offset = 0;
1311                 len -= seg;
1312                 data += seg;
1313                 ++gfn;
1314         }
1315         return 0;
1316 }
1317 EXPORT_SYMBOL_GPL(kvm_read_guest);
1318
1319 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1320                           unsigned long len)
1321 {
1322         int r;
1323         unsigned long addr;
1324         gfn_t gfn = gpa >> PAGE_SHIFT;
1325         int offset = offset_in_page(gpa);
1326
1327         addr = gfn_to_hva(kvm, gfn);
1328         if (kvm_is_error_hva(addr))
1329                 return -EFAULT;
1330         pagefault_disable();
1331         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1332         pagefault_enable();
1333         if (r)
1334                 return -EFAULT;
1335         return 0;
1336 }
1337 EXPORT_SYMBOL(kvm_read_guest_atomic);
1338
1339 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1340                          int offset, int len)
1341 {
1342         int r;
1343         unsigned long addr;
1344
1345         addr = gfn_to_hva(kvm, gfn);
1346         if (kvm_is_error_hva(addr))
1347                 return -EFAULT;
1348         r = __copy_to_user((void __user *)addr + offset, data, len);
1349         if (r)
1350                 return -EFAULT;
1351         mark_page_dirty(kvm, gfn);
1352         return 0;
1353 }
1354 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1355
1356 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1357                     unsigned long len)
1358 {
1359         gfn_t gfn = gpa >> PAGE_SHIFT;
1360         int seg;
1361         int offset = offset_in_page(gpa);
1362         int ret;
1363
1364         while ((seg = next_segment(len, offset)) != 0) {
1365                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1366                 if (ret < 0)
1367                         return ret;
1368                 offset = 0;
1369                 len -= seg;
1370                 data += seg;
1371                 ++gfn;
1372         }
1373         return 0;
1374 }
1375
1376 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1377                               gpa_t gpa)
1378 {
1379         struct kvm_memslots *slots = kvm_memslots(kvm);
1380         int offset = offset_in_page(gpa);
1381         gfn_t gfn = gpa >> PAGE_SHIFT;
1382
1383         ghc->gpa = gpa;
1384         ghc->generation = slots->generation;
1385         ghc->memslot = gfn_to_memslot(kvm, gfn);
1386         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1387         if (!kvm_is_error_hva(ghc->hva))
1388                 ghc->hva += offset;
1389         else
1390                 return -EFAULT;
1391
1392         return 0;
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1395
1396 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1397                            void *data, unsigned long len)
1398 {
1399         struct kvm_memslots *slots = kvm_memslots(kvm);
1400         int r;
1401
1402         if (slots->generation != ghc->generation)
1403                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1404
1405         if (kvm_is_error_hva(ghc->hva))
1406                 return -EFAULT;
1407
1408         r = __copy_to_user((void __user *)ghc->hva, data, len);
1409         if (r)
1410                 return -EFAULT;
1411         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1412
1413         return 0;
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1416
1417 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1418                            void *data, unsigned long len)
1419 {
1420         struct kvm_memslots *slots = kvm_memslots(kvm);
1421         int r;
1422
1423         if (slots->generation != ghc->generation)
1424                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1425
1426         if (kvm_is_error_hva(ghc->hva))
1427                 return -EFAULT;
1428
1429         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1430         if (r)
1431                 return -EFAULT;
1432
1433         return 0;
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1436
1437 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1438 {
1439         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1440                                     offset, len);
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1443
1444 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1445 {
1446         gfn_t gfn = gpa >> PAGE_SHIFT;
1447         int seg;
1448         int offset = offset_in_page(gpa);
1449         int ret;
1450
1451         while ((seg = next_segment(len, offset)) != 0) {
1452                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1453                 if (ret < 0)
1454                         return ret;
1455                 offset = 0;
1456                 len -= seg;
1457                 ++gfn;
1458         }
1459         return 0;
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1462
1463 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1464                              gfn_t gfn)
1465 {
1466         if (memslot && memslot->dirty_bitmap) {
1467                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1468
1469                 /* TODO: introduce set_bit_le() and use it */
1470                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1471         }
1472 }
1473
1474 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1475 {
1476         struct kvm_memory_slot *memslot;
1477
1478         memslot = gfn_to_memslot(kvm, gfn);
1479         mark_page_dirty_in_slot(kvm, memslot, gfn);
1480 }
1481
1482 /*
1483  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1484  */
1485 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1486 {
1487         DEFINE_WAIT(wait);
1488
1489         for (;;) {
1490                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1491
1492                 if (kvm_arch_vcpu_runnable(vcpu)) {
1493                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1494                         break;
1495                 }
1496                 if (kvm_cpu_has_pending_timer(vcpu))
1497                         break;
1498                 if (signal_pending(current))
1499                         break;
1500
1501                 schedule();
1502         }
1503
1504         finish_wait(&vcpu->wq, &wait);
1505 }
1506
1507 #ifndef CONFIG_S390
1508 /*
1509  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1510  */
1511 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1512 {
1513         int me;
1514         int cpu = vcpu->cpu;
1515         wait_queue_head_t *wqp;
1516
1517         wqp = kvm_arch_vcpu_wq(vcpu);
1518         if (waitqueue_active(wqp)) {
1519                 wake_up_interruptible(wqp);
1520                 ++vcpu->stat.halt_wakeup;
1521         }
1522
1523         me = get_cpu();
1524         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1525                 if (kvm_arch_vcpu_should_kick(vcpu))
1526                         smp_send_reschedule(cpu);
1527         put_cpu();
1528 }
1529 #endif /* !CONFIG_S390 */
1530
1531 void kvm_resched(struct kvm_vcpu *vcpu)
1532 {
1533         if (!need_resched())
1534                 return;
1535         cond_resched();
1536 }
1537 EXPORT_SYMBOL_GPL(kvm_resched);
1538
1539 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1540 {
1541         struct pid *pid;
1542         struct task_struct *task = NULL;
1543
1544         rcu_read_lock();
1545         pid = rcu_dereference(target->pid);
1546         if (pid)
1547                 task = get_pid_task(target->pid, PIDTYPE_PID);
1548         rcu_read_unlock();
1549         if (!task)
1550                 return false;
1551         if (task->flags & PF_VCPU) {
1552                 put_task_struct(task);
1553                 return false;
1554         }
1555         if (yield_to(task, 1)) {
1556                 put_task_struct(task);
1557                 return true;
1558         }
1559         put_task_struct(task);
1560         return false;
1561 }
1562 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1563
1564 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1565 /*
1566  * Helper that checks whether a VCPU is eligible for directed yield.
1567  * Most eligible candidate to yield is decided by following heuristics:
1568  *
1569  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1570  *  (preempted lock holder), indicated by @in_spin_loop.
1571  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1572  *
1573  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1574  *  chance last time (mostly it has become eligible now since we have probably
1575  *  yielded to lockholder in last iteration. This is done by toggling
1576  *  @dy_eligible each time a VCPU checked for eligibility.)
1577  *
1578  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1579  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1580  *  burning. Giving priority for a potential lock-holder increases lock
1581  *  progress.
1582  *
1583  *  Since algorithm is based on heuristics, accessing another VCPU data without
1584  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1585  *  and continue with next VCPU and so on.
1586  */
1587 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1588 {
1589         bool eligible;
1590
1591         eligible = !vcpu->spin_loop.in_spin_loop ||
1592                         (vcpu->spin_loop.in_spin_loop &&
1593                          vcpu->spin_loop.dy_eligible);
1594
1595         if (vcpu->spin_loop.in_spin_loop)
1596                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1597
1598         return eligible;
1599 }
1600 #endif
1601 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1602 {
1603         struct kvm *kvm = me->kvm;
1604         struct kvm_vcpu *vcpu;
1605         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1606         int yielded = 0;
1607         int pass;
1608         int i;
1609
1610         kvm_vcpu_set_in_spin_loop(me, true);
1611         /*
1612          * We boost the priority of a VCPU that is runnable but not
1613          * currently running, because it got preempted by something
1614          * else and called schedule in __vcpu_run.  Hopefully that
1615          * VCPU is holding the lock that we need and will release it.
1616          * We approximate round-robin by starting at the last boosted VCPU.
1617          */
1618         for (pass = 0; pass < 2 && !yielded; pass++) {
1619                 kvm_for_each_vcpu(i, vcpu, kvm) {
1620                         if (!pass && i <= last_boosted_vcpu) {
1621                                 i = last_boosted_vcpu;
1622                                 continue;
1623                         } else if (pass && i > last_boosted_vcpu)
1624                                 break;
1625                         if (vcpu == me)
1626                                 continue;
1627                         if (waitqueue_active(&vcpu->wq))
1628                                 continue;
1629                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1630                                 continue;
1631                         if (kvm_vcpu_yield_to(vcpu)) {
1632                                 kvm->last_boosted_vcpu = i;
1633                                 yielded = 1;
1634                                 break;
1635                         }
1636                 }
1637         }
1638         kvm_vcpu_set_in_spin_loop(me, false);
1639
1640         /* Ensure vcpu is not eligible during next spinloop */
1641         kvm_vcpu_set_dy_eligible(me, false);
1642 }
1643 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1644
1645 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1646 {
1647         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1648         struct page *page;
1649
1650         if (vmf->pgoff == 0)
1651                 page = virt_to_page(vcpu->run);
1652 #ifdef CONFIG_X86
1653         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1654                 page = virt_to_page(vcpu->arch.pio_data);
1655 #endif
1656 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1657         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1658                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1659 #endif
1660         else
1661                 return kvm_arch_vcpu_fault(vcpu, vmf);
1662         get_page(page);
1663         vmf->page = page;
1664         return 0;
1665 }
1666
1667 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1668         .fault = kvm_vcpu_fault,
1669 };
1670
1671 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1672 {
1673         vma->vm_ops = &kvm_vcpu_vm_ops;
1674         return 0;
1675 }
1676
1677 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1678 {
1679         struct kvm_vcpu *vcpu = filp->private_data;
1680
1681         kvm_put_kvm(vcpu->kvm);
1682         return 0;
1683 }
1684
1685 static struct file_operations kvm_vcpu_fops = {
1686         .release        = kvm_vcpu_release,
1687         .unlocked_ioctl = kvm_vcpu_ioctl,
1688 #ifdef CONFIG_COMPAT
1689         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1690 #endif
1691         .mmap           = kvm_vcpu_mmap,
1692         .llseek         = noop_llseek,
1693 };
1694
1695 /*
1696  * Allocates an inode for the vcpu.
1697  */
1698 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1699 {
1700         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1701 }
1702
1703 /*
1704  * Creates some virtual cpus.  Good luck creating more than one.
1705  */
1706 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1707 {
1708         int r;
1709         struct kvm_vcpu *vcpu, *v;
1710
1711         vcpu = kvm_arch_vcpu_create(kvm, id);
1712         if (IS_ERR(vcpu))
1713                 return PTR_ERR(vcpu);
1714
1715         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1716
1717         r = kvm_arch_vcpu_setup(vcpu);
1718         if (r)
1719                 goto vcpu_destroy;
1720
1721         mutex_lock(&kvm->lock);
1722         if (!kvm_vcpu_compatible(vcpu)) {
1723                 r = -EINVAL;
1724                 goto unlock_vcpu_destroy;
1725         }
1726         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1727                 r = -EINVAL;
1728                 goto unlock_vcpu_destroy;
1729         }
1730
1731         kvm_for_each_vcpu(r, v, kvm)
1732                 if (v->vcpu_id == id) {
1733                         r = -EEXIST;
1734                         goto unlock_vcpu_destroy;
1735                 }
1736
1737         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1738
1739         /* Now it's all set up, let userspace reach it */
1740         kvm_get_kvm(kvm);
1741         r = create_vcpu_fd(vcpu);
1742         if (r < 0) {
1743                 kvm_put_kvm(kvm);
1744                 goto unlock_vcpu_destroy;
1745         }
1746
1747         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1748         smp_wmb();
1749         atomic_inc(&kvm->online_vcpus);
1750
1751         mutex_unlock(&kvm->lock);
1752         return r;
1753
1754 unlock_vcpu_destroy:
1755         mutex_unlock(&kvm->lock);
1756 vcpu_destroy:
1757         kvm_arch_vcpu_destroy(vcpu);
1758         return r;
1759 }
1760
1761 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1762 {
1763         if (sigset) {
1764                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1765                 vcpu->sigset_active = 1;
1766                 vcpu->sigset = *sigset;
1767         } else
1768                 vcpu->sigset_active = 0;
1769         return 0;
1770 }
1771
1772 static long kvm_vcpu_ioctl(struct file *filp,
1773                            unsigned int ioctl, unsigned long arg)
1774 {
1775         struct kvm_vcpu *vcpu = filp->private_data;
1776         void __user *argp = (void __user *)arg;
1777         int r;
1778         struct kvm_fpu *fpu = NULL;
1779         struct kvm_sregs *kvm_sregs = NULL;
1780
1781         if (vcpu->kvm->mm != current->mm)
1782                 return -EIO;
1783
1784 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1785         /*
1786          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1787          * so vcpu_load() would break it.
1788          */
1789         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1790                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1791 #endif
1792
1793
1794         vcpu_load(vcpu);
1795         switch (ioctl) {
1796         case KVM_RUN:
1797                 r = -EINVAL;
1798                 if (arg)
1799                         goto out;
1800                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1801                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1802                 break;
1803         case KVM_GET_REGS: {
1804                 struct kvm_regs *kvm_regs;
1805
1806                 r = -ENOMEM;
1807                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1808                 if (!kvm_regs)
1809                         goto out;
1810                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1811                 if (r)
1812                         goto out_free1;
1813                 r = -EFAULT;
1814                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1815                         goto out_free1;
1816                 r = 0;
1817 out_free1:
1818                 kfree(kvm_regs);
1819                 break;
1820         }
1821         case KVM_SET_REGS: {
1822                 struct kvm_regs *kvm_regs;
1823
1824                 r = -ENOMEM;
1825                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1826                 if (IS_ERR(kvm_regs)) {
1827                         r = PTR_ERR(kvm_regs);
1828                         goto out;
1829                 }
1830                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1831                 if (r)
1832                         goto out_free2;
1833                 r = 0;
1834 out_free2:
1835                 kfree(kvm_regs);
1836                 break;
1837         }
1838         case KVM_GET_SREGS: {
1839                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1840                 r = -ENOMEM;
1841                 if (!kvm_sregs)
1842                         goto out;
1843                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1844                 if (r)
1845                         goto out;
1846                 r = -EFAULT;
1847                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1848                         goto out;
1849                 r = 0;
1850                 break;
1851         }
1852         case KVM_SET_SREGS: {
1853                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1854                 if (IS_ERR(kvm_sregs)) {
1855                         r = PTR_ERR(kvm_sregs);
1856                         goto out;
1857                 }
1858                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1859                 if (r)
1860                         goto out;
1861                 r = 0;
1862                 break;
1863         }
1864         case KVM_GET_MP_STATE: {
1865                 struct kvm_mp_state mp_state;
1866
1867                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1868                 if (r)
1869                         goto out;
1870                 r = -EFAULT;
1871                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1872                         goto out;
1873                 r = 0;
1874                 break;
1875         }
1876         case KVM_SET_MP_STATE: {
1877                 struct kvm_mp_state mp_state;
1878
1879                 r = -EFAULT;
1880                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1881                         goto out;
1882                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1883                 if (r)
1884                         goto out;
1885                 r = 0;
1886                 break;
1887         }
1888         case KVM_TRANSLATE: {
1889                 struct kvm_translation tr;
1890
1891                 r = -EFAULT;
1892                 if (copy_from_user(&tr, argp, sizeof tr))
1893                         goto out;
1894                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1895                 if (r)
1896                         goto out;
1897                 r = -EFAULT;
1898                 if (copy_to_user(argp, &tr, sizeof tr))
1899                         goto out;
1900                 r = 0;
1901                 break;
1902         }
1903         case KVM_SET_GUEST_DEBUG: {
1904                 struct kvm_guest_debug dbg;
1905
1906                 r = -EFAULT;
1907                 if (copy_from_user(&dbg, argp, sizeof dbg))
1908                         goto out;
1909                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1910                 if (r)
1911                         goto out;
1912                 r = 0;
1913                 break;
1914         }
1915         case KVM_SET_SIGNAL_MASK: {
1916                 struct kvm_signal_mask __user *sigmask_arg = argp;
1917                 struct kvm_signal_mask kvm_sigmask;
1918                 sigset_t sigset, *p;
1919
1920                 p = NULL;
1921                 if (argp) {
1922                         r = -EFAULT;
1923                         if (copy_from_user(&kvm_sigmask, argp,
1924                                            sizeof kvm_sigmask))
1925                                 goto out;
1926                         r = -EINVAL;
1927                         if (kvm_sigmask.len != sizeof sigset)
1928                                 goto out;
1929                         r = -EFAULT;
1930                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1931                                            sizeof sigset))
1932                                 goto out;
1933                         p = &sigset;
1934                 }
1935                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1936                 break;
1937         }
1938         case KVM_GET_FPU: {
1939                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1940                 r = -ENOMEM;
1941                 if (!fpu)
1942                         goto out;
1943                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1944                 if (r)
1945                         goto out;
1946                 r = -EFAULT;
1947                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1948                         goto out;
1949                 r = 0;
1950                 break;
1951         }
1952         case KVM_SET_FPU: {
1953                 fpu = memdup_user(argp, sizeof(*fpu));
1954                 if (IS_ERR(fpu)) {
1955                         r = PTR_ERR(fpu);
1956                         goto out;
1957                 }
1958                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1959                 if (r)
1960                         goto out;
1961                 r = 0;
1962                 break;
1963         }
1964         default:
1965                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1966         }
1967 out:
1968         vcpu_put(vcpu);
1969         kfree(fpu);
1970         kfree(kvm_sregs);
1971         return r;
1972 }
1973
1974 #ifdef CONFIG_COMPAT
1975 static long kvm_vcpu_compat_ioctl(struct file *filp,
1976                                   unsigned int ioctl, unsigned long arg)
1977 {
1978         struct kvm_vcpu *vcpu = filp->private_data;
1979         void __user *argp = compat_ptr(arg);
1980         int r;
1981
1982         if (vcpu->kvm->mm != current->mm)
1983                 return -EIO;
1984
1985         switch (ioctl) {
1986         case KVM_SET_SIGNAL_MASK: {
1987                 struct kvm_signal_mask __user *sigmask_arg = argp;
1988                 struct kvm_signal_mask kvm_sigmask;
1989                 compat_sigset_t csigset;
1990                 sigset_t sigset;
1991
1992                 if (argp) {
1993                         r = -EFAULT;
1994                         if (copy_from_user(&kvm_sigmask, argp,
1995                                            sizeof kvm_sigmask))
1996                                 goto out;
1997                         r = -EINVAL;
1998                         if (kvm_sigmask.len != sizeof csigset)
1999                                 goto out;
2000                         r = -EFAULT;
2001                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2002                                            sizeof csigset))
2003                                 goto out;
2004                 }
2005                 sigset_from_compat(&sigset, &csigset);
2006                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2007                 break;
2008         }
2009         default:
2010                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2011         }
2012
2013 out:
2014         return r;
2015 }
2016 #endif
2017
2018 static long kvm_vm_ioctl(struct file *filp,
2019                            unsigned int ioctl, unsigned long arg)
2020 {
2021         struct kvm *kvm = filp->private_data;
2022         void __user *argp = (void __user *)arg;
2023         int r;
2024
2025         if (kvm->mm != current->mm)
2026                 return -EIO;
2027         switch (ioctl) {
2028         case KVM_CREATE_VCPU:
2029                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2030                 if (r < 0)
2031                         goto out;
2032                 break;
2033         case KVM_SET_USER_MEMORY_REGION: {
2034                 struct kvm_userspace_memory_region kvm_userspace_mem;
2035
2036                 r = -EFAULT;
2037                 if (copy_from_user(&kvm_userspace_mem, argp,
2038                                                 sizeof kvm_userspace_mem))
2039                         goto out;
2040
2041                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2042                 if (r)
2043                         goto out;
2044                 break;
2045         }
2046         case KVM_GET_DIRTY_LOG: {
2047                 struct kvm_dirty_log log;
2048
2049                 r = -EFAULT;
2050                 if (copy_from_user(&log, argp, sizeof log))
2051                         goto out;
2052                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2053                 if (r)
2054                         goto out;
2055                 break;
2056         }
2057 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2058         case KVM_REGISTER_COALESCED_MMIO: {
2059                 struct kvm_coalesced_mmio_zone zone;
2060                 r = -EFAULT;
2061                 if (copy_from_user(&zone, argp, sizeof zone))
2062                         goto out;
2063                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2064                 if (r)
2065                         goto out;
2066                 r = 0;
2067                 break;
2068         }
2069         case KVM_UNREGISTER_COALESCED_MMIO: {
2070                 struct kvm_coalesced_mmio_zone zone;
2071                 r = -EFAULT;
2072                 if (copy_from_user(&zone, argp, sizeof zone))
2073                         goto out;
2074                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2075                 if (r)
2076                         goto out;
2077                 r = 0;
2078                 break;
2079         }
2080 #endif
2081         case KVM_IRQFD: {
2082                 struct kvm_irqfd data;
2083
2084                 r = -EFAULT;
2085                 if (copy_from_user(&data, argp, sizeof data))
2086                         goto out;
2087                 r = kvm_irqfd(kvm, &data);
2088                 break;
2089         }
2090         case KVM_IOEVENTFD: {
2091                 struct kvm_ioeventfd data;
2092
2093                 r = -EFAULT;
2094                 if (copy_from_user(&data, argp, sizeof data))
2095                         goto out;
2096                 r = kvm_ioeventfd(kvm, &data);
2097                 break;
2098         }
2099 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2100         case KVM_SET_BOOT_CPU_ID:
2101                 r = 0;
2102                 mutex_lock(&kvm->lock);
2103                 if (atomic_read(&kvm->online_vcpus) != 0)
2104                         r = -EBUSY;
2105                 else
2106                         kvm->bsp_vcpu_id = arg;
2107                 mutex_unlock(&kvm->lock);
2108                 break;
2109 #endif
2110 #ifdef CONFIG_HAVE_KVM_MSI
2111         case KVM_SIGNAL_MSI: {
2112                 struct kvm_msi msi;
2113
2114                 r = -EFAULT;
2115                 if (copy_from_user(&msi, argp, sizeof msi))
2116                         goto out;
2117                 r = kvm_send_userspace_msi(kvm, &msi);
2118                 break;
2119         }
2120 #endif
2121 #ifdef __KVM_HAVE_IRQ_LINE
2122         case KVM_IRQ_LINE_STATUS:
2123         case KVM_IRQ_LINE: {
2124                 struct kvm_irq_level irq_event;
2125
2126                 r = -EFAULT;
2127                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2128                         goto out;
2129
2130                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2131                 if (r)
2132                         goto out;
2133
2134                 r = -EFAULT;
2135                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2136                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2137                                 goto out;
2138                 }
2139
2140                 r = 0;
2141                 break;
2142         }
2143 #endif
2144         default:
2145                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2146                 if (r == -ENOTTY)
2147                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2148         }
2149 out:
2150         return r;
2151 }
2152
2153 #ifdef CONFIG_COMPAT
2154 struct compat_kvm_dirty_log {
2155         __u32 slot;
2156         __u32 padding1;
2157         union {
2158                 compat_uptr_t dirty_bitmap; /* one bit per page */
2159                 __u64 padding2;
2160         };
2161 };
2162
2163 static long kvm_vm_compat_ioctl(struct file *filp,
2164                            unsigned int ioctl, unsigned long arg)
2165 {
2166         struct kvm *kvm = filp->private_data;
2167         int r;
2168
2169         if (kvm->mm != current->mm)
2170                 return -EIO;
2171         switch (ioctl) {
2172         case KVM_GET_DIRTY_LOG: {
2173                 struct compat_kvm_dirty_log compat_log;
2174                 struct kvm_dirty_log log;
2175
2176                 r = -EFAULT;
2177                 if (copy_from_user(&compat_log, (void __user *)arg,
2178                                    sizeof(compat_log)))
2179                         goto out;
2180                 log.slot         = compat_log.slot;
2181                 log.padding1     = compat_log.padding1;
2182                 log.padding2     = compat_log.padding2;
2183                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2184
2185                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2186                 if (r)
2187                         goto out;
2188                 break;
2189         }
2190         default:
2191                 r = kvm_vm_ioctl(filp, ioctl, arg);
2192         }
2193
2194 out:
2195         return r;
2196 }
2197 #endif
2198
2199 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2200 {
2201         struct page *page[1];
2202         unsigned long addr;
2203         int npages;
2204         gfn_t gfn = vmf->pgoff;
2205         struct kvm *kvm = vma->vm_file->private_data;
2206
2207         addr = gfn_to_hva(kvm, gfn);
2208         if (kvm_is_error_hva(addr))
2209                 return VM_FAULT_SIGBUS;
2210
2211         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2212                                 NULL);
2213         if (unlikely(npages != 1))
2214                 return VM_FAULT_SIGBUS;
2215
2216         vmf->page = page[0];
2217         return 0;
2218 }
2219
2220 static const struct vm_operations_struct kvm_vm_vm_ops = {
2221         .fault = kvm_vm_fault,
2222 };
2223
2224 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2225 {
2226         vma->vm_ops = &kvm_vm_vm_ops;
2227         return 0;
2228 }
2229
2230 static struct file_operations kvm_vm_fops = {
2231         .release        = kvm_vm_release,
2232         .unlocked_ioctl = kvm_vm_ioctl,
2233 #ifdef CONFIG_COMPAT
2234         .compat_ioctl   = kvm_vm_compat_ioctl,
2235 #endif
2236         .mmap           = kvm_vm_mmap,
2237         .llseek         = noop_llseek,
2238 };
2239
2240 static int kvm_dev_ioctl_create_vm(unsigned long type)
2241 {
2242         int r;
2243         struct kvm *kvm;
2244
2245         kvm = kvm_create_vm(type);
2246         if (IS_ERR(kvm))
2247                 return PTR_ERR(kvm);
2248 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2249         r = kvm_coalesced_mmio_init(kvm);
2250         if (r < 0) {
2251                 kvm_put_kvm(kvm);
2252                 return r;
2253         }
2254 #endif
2255         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2256         if (r < 0)
2257                 kvm_put_kvm(kvm);
2258
2259         return r;
2260 }
2261
2262 static long kvm_dev_ioctl_check_extension_generic(long arg)
2263 {
2264         switch (arg) {
2265         case KVM_CAP_USER_MEMORY:
2266         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2267         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2268 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2269         case KVM_CAP_SET_BOOT_CPU_ID:
2270 #endif
2271         case KVM_CAP_INTERNAL_ERROR_DATA:
2272 #ifdef CONFIG_HAVE_KVM_MSI
2273         case KVM_CAP_SIGNAL_MSI:
2274 #endif
2275                 return 1;
2276 #ifdef KVM_CAP_IRQ_ROUTING
2277         case KVM_CAP_IRQ_ROUTING:
2278                 return KVM_MAX_IRQ_ROUTES;
2279 #endif
2280         default:
2281                 break;
2282         }
2283         return kvm_dev_ioctl_check_extension(arg);
2284 }
2285
2286 static long kvm_dev_ioctl(struct file *filp,
2287                           unsigned int ioctl, unsigned long arg)
2288 {
2289         long r = -EINVAL;
2290
2291         switch (ioctl) {
2292         case KVM_GET_API_VERSION:
2293                 r = -EINVAL;
2294                 if (arg)
2295                         goto out;
2296                 r = KVM_API_VERSION;
2297                 break;
2298         case KVM_CREATE_VM:
2299                 r = kvm_dev_ioctl_create_vm(arg);
2300                 break;
2301         case KVM_CHECK_EXTENSION:
2302                 r = kvm_dev_ioctl_check_extension_generic(arg);
2303                 break;
2304         case KVM_GET_VCPU_MMAP_SIZE:
2305                 r = -EINVAL;
2306                 if (arg)
2307                         goto out;
2308                 r = PAGE_SIZE;     /* struct kvm_run */
2309 #ifdef CONFIG_X86
2310                 r += PAGE_SIZE;    /* pio data page */
2311 #endif
2312 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2313                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2314 #endif
2315                 break;
2316         case KVM_TRACE_ENABLE:
2317         case KVM_TRACE_PAUSE:
2318         case KVM_TRACE_DISABLE:
2319                 r = -EOPNOTSUPP;
2320                 break;
2321         default:
2322                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2323         }
2324 out:
2325         return r;
2326 }
2327
2328 static struct file_operations kvm_chardev_ops = {
2329         .unlocked_ioctl = kvm_dev_ioctl,
2330         .compat_ioctl   = kvm_dev_ioctl,
2331         .llseek         = noop_llseek,
2332 };
2333
2334 static struct miscdevice kvm_dev = {
2335         KVM_MINOR,
2336         "kvm",
2337         &kvm_chardev_ops,
2338 };
2339
2340 static void hardware_enable_nolock(void *junk)
2341 {
2342         int cpu = raw_smp_processor_id();
2343         int r;
2344
2345         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2346                 return;
2347
2348         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2349
2350         r = kvm_arch_hardware_enable(NULL);
2351
2352         if (r) {
2353                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2354                 atomic_inc(&hardware_enable_failed);
2355                 printk(KERN_INFO "kvm: enabling virtualization on "
2356                                  "CPU%d failed\n", cpu);
2357         }
2358 }
2359
2360 static void hardware_enable(void *junk)
2361 {
2362         raw_spin_lock(&kvm_lock);
2363         hardware_enable_nolock(junk);
2364         raw_spin_unlock(&kvm_lock);
2365 }
2366
2367 static void hardware_disable_nolock(void *junk)
2368 {
2369         int cpu = raw_smp_processor_id();
2370
2371         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2372                 return;
2373         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2374         kvm_arch_hardware_disable(NULL);
2375 }
2376
2377 static void hardware_disable(void *junk)
2378 {
2379         raw_spin_lock(&kvm_lock);
2380         hardware_disable_nolock(junk);
2381         raw_spin_unlock(&kvm_lock);
2382 }
2383
2384 static void hardware_disable_all_nolock(void)
2385 {
2386         BUG_ON(!kvm_usage_count);
2387
2388         kvm_usage_count--;
2389         if (!kvm_usage_count)
2390                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2391 }
2392
2393 static void hardware_disable_all(void)
2394 {
2395         raw_spin_lock(&kvm_lock);
2396         hardware_disable_all_nolock();
2397         raw_spin_unlock(&kvm_lock);
2398 }
2399
2400 static int hardware_enable_all(void)
2401 {
2402         int r = 0;
2403
2404         raw_spin_lock(&kvm_lock);
2405
2406         kvm_usage_count++;
2407         if (kvm_usage_count == 1) {
2408                 atomic_set(&hardware_enable_failed, 0);
2409                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2410
2411                 if (atomic_read(&hardware_enable_failed)) {
2412                         hardware_disable_all_nolock();
2413                         r = -EBUSY;
2414                 }
2415         }
2416
2417         raw_spin_unlock(&kvm_lock);
2418
2419         return r;
2420 }
2421
2422 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2423                            void *v)
2424 {
2425         int cpu = (long)v;
2426
2427         if (!kvm_usage_count)
2428                 return NOTIFY_OK;
2429
2430         val &= ~CPU_TASKS_FROZEN;
2431         switch (val) {
2432         case CPU_DYING:
2433                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2434                        cpu);
2435                 hardware_disable(NULL);
2436                 break;
2437         case CPU_STARTING:
2438                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2439                        cpu);
2440                 hardware_enable(NULL);
2441                 break;
2442         }
2443         return NOTIFY_OK;
2444 }
2445
2446
2447 asmlinkage void kvm_spurious_fault(void)
2448 {
2449         /* Fault while not rebooting.  We want the trace. */
2450         BUG();
2451 }
2452 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2453
2454 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2455                       void *v)
2456 {
2457         /*
2458          * Some (well, at least mine) BIOSes hang on reboot if
2459          * in vmx root mode.
2460          *
2461          * And Intel TXT required VMX off for all cpu when system shutdown.
2462          */
2463         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2464         kvm_rebooting = true;
2465         on_each_cpu(hardware_disable_nolock, NULL, 1);
2466         return NOTIFY_OK;
2467 }
2468
2469 static struct notifier_block kvm_reboot_notifier = {
2470         .notifier_call = kvm_reboot,
2471         .priority = 0,
2472 };
2473
2474 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2475 {
2476         int i;
2477
2478         for (i = 0; i < bus->dev_count; i++) {
2479                 struct kvm_io_device *pos = bus->range[i].dev;
2480
2481                 kvm_iodevice_destructor(pos);
2482         }
2483         kfree(bus);
2484 }
2485
2486 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2487 {
2488         const struct kvm_io_range *r1 = p1;
2489         const struct kvm_io_range *r2 = p2;
2490
2491         if (r1->addr < r2->addr)
2492                 return -1;
2493         if (r1->addr + r1->len > r2->addr + r2->len)
2494                 return 1;
2495         return 0;
2496 }
2497
2498 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2499                           gpa_t addr, int len)
2500 {
2501         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2502                 .addr = addr,
2503                 .len = len,
2504                 .dev = dev,
2505         };
2506
2507         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2508                 kvm_io_bus_sort_cmp, NULL);
2509
2510         return 0;
2511 }
2512
2513 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2514                              gpa_t addr, int len)
2515 {
2516         struct kvm_io_range *range, key;
2517         int off;
2518
2519         key = (struct kvm_io_range) {
2520                 .addr = addr,
2521                 .len = len,
2522         };
2523
2524         range = bsearch(&key, bus->range, bus->dev_count,
2525                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2526         if (range == NULL)
2527                 return -ENOENT;
2528
2529         off = range - bus->range;
2530
2531         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2532                 off--;
2533
2534         return off;
2535 }
2536
2537 /* kvm_io_bus_write - called under kvm->slots_lock */
2538 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2539                      int len, const void *val)
2540 {
2541         int idx;
2542         struct kvm_io_bus *bus;
2543         struct kvm_io_range range;
2544
2545         range = (struct kvm_io_range) {
2546                 .addr = addr,
2547                 .len = len,
2548         };
2549
2550         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2551         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2552         if (idx < 0)
2553                 return -EOPNOTSUPP;
2554
2555         while (idx < bus->dev_count &&
2556                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2557                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2558                         return 0;
2559                 idx++;
2560         }
2561
2562         return -EOPNOTSUPP;
2563 }
2564
2565 /* kvm_io_bus_read - called under kvm->slots_lock */
2566 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2567                     int len, void *val)
2568 {
2569         int idx;
2570         struct kvm_io_bus *bus;
2571         struct kvm_io_range range;
2572
2573         range = (struct kvm_io_range) {
2574                 .addr = addr,
2575                 .len = len,
2576         };
2577
2578         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2579         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2580         if (idx < 0)
2581                 return -EOPNOTSUPP;
2582
2583         while (idx < bus->dev_count &&
2584                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2585                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2586                         return 0;
2587                 idx++;
2588         }
2589
2590         return -EOPNOTSUPP;
2591 }
2592
2593 /* Caller must hold slots_lock. */
2594 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2595                             int len, struct kvm_io_device *dev)
2596 {
2597         struct kvm_io_bus *new_bus, *bus;
2598
2599         bus = kvm->buses[bus_idx];
2600         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2601                 return -ENOSPC;
2602
2603         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2604                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2605         if (!new_bus)
2606                 return -ENOMEM;
2607         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2608                sizeof(struct kvm_io_range)));
2609         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2610         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2611         synchronize_srcu_expedited(&kvm->srcu);
2612         kfree(bus);
2613
2614         return 0;
2615 }
2616
2617 /* Caller must hold slots_lock. */
2618 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2619                               struct kvm_io_device *dev)
2620 {
2621         int i, r;
2622         struct kvm_io_bus *new_bus, *bus;
2623
2624         bus = kvm->buses[bus_idx];
2625         r = -ENOENT;
2626         for (i = 0; i < bus->dev_count; i++)
2627                 if (bus->range[i].dev == dev) {
2628                         r = 0;
2629                         break;
2630                 }
2631
2632         if (r)
2633                 return r;
2634
2635         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2636                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2637         if (!new_bus)
2638                 return -ENOMEM;
2639
2640         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2641         new_bus->dev_count--;
2642         memcpy(new_bus->range + i, bus->range + i + 1,
2643                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2644
2645         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2646         synchronize_srcu_expedited(&kvm->srcu);
2647         kfree(bus);
2648         return r;
2649 }
2650
2651 static struct notifier_block kvm_cpu_notifier = {
2652         .notifier_call = kvm_cpu_hotplug,
2653 };
2654
2655 static int vm_stat_get(void *_offset, u64 *val)
2656 {
2657         unsigned offset = (long)_offset;
2658         struct kvm *kvm;
2659
2660         *val = 0;
2661         raw_spin_lock(&kvm_lock);
2662         list_for_each_entry(kvm, &vm_list, vm_list)
2663                 *val += *(u32 *)((void *)kvm + offset);
2664         raw_spin_unlock(&kvm_lock);
2665         return 0;
2666 }
2667
2668 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2669
2670 static int vcpu_stat_get(void *_offset, u64 *val)
2671 {
2672         unsigned offset = (long)_offset;
2673         struct kvm *kvm;
2674         struct kvm_vcpu *vcpu;
2675         int i;
2676
2677         *val = 0;
2678         raw_spin_lock(&kvm_lock);
2679         list_for_each_entry(kvm, &vm_list, vm_list)
2680                 kvm_for_each_vcpu(i, vcpu, kvm)
2681                         *val += *(u32 *)((void *)vcpu + offset);
2682
2683         raw_spin_unlock(&kvm_lock);
2684         return 0;
2685 }
2686
2687 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2688
2689 static const struct file_operations *stat_fops[] = {
2690         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2691         [KVM_STAT_VM]   = &vm_stat_fops,
2692 };
2693
2694 static int kvm_init_debug(void)
2695 {
2696         int r = -EFAULT;
2697         struct kvm_stats_debugfs_item *p;
2698
2699         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2700         if (kvm_debugfs_dir == NULL)
2701                 goto out;
2702
2703         for (p = debugfs_entries; p->name; ++p) {
2704                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2705                                                 (void *)(long)p->offset,
2706                                                 stat_fops[p->kind]);
2707                 if (p->dentry == NULL)
2708                         goto out_dir;
2709         }
2710
2711         return 0;
2712
2713 out_dir:
2714         debugfs_remove_recursive(kvm_debugfs_dir);
2715 out:
2716         return r;
2717 }
2718
2719 static void kvm_exit_debug(void)
2720 {
2721         struct kvm_stats_debugfs_item *p;
2722
2723         for (p = debugfs_entries; p->name; ++p)
2724                 debugfs_remove(p->dentry);
2725         debugfs_remove(kvm_debugfs_dir);
2726 }
2727
2728 static int kvm_suspend(void)
2729 {
2730         if (kvm_usage_count)
2731                 hardware_disable_nolock(NULL);
2732         return 0;
2733 }
2734
2735 static void kvm_resume(void)
2736 {
2737         if (kvm_usage_count) {
2738                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2739                 hardware_enable_nolock(NULL);
2740         }
2741 }
2742
2743 static struct syscore_ops kvm_syscore_ops = {
2744         .suspend = kvm_suspend,
2745         .resume = kvm_resume,
2746 };
2747
2748 static inline
2749 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2750 {
2751         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2752 }
2753
2754 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2755 {
2756         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2757
2758         kvm_arch_vcpu_load(vcpu, cpu);
2759 }
2760
2761 static void kvm_sched_out(struct preempt_notifier *pn,
2762                           struct task_struct *next)
2763 {
2764         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2765
2766         kvm_arch_vcpu_put(vcpu);
2767 }
2768
2769 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2770                   struct module *module)
2771 {
2772         int r;
2773         int cpu;
2774
2775         r = kvm_arch_init(opaque);
2776         if (r)
2777                 goto out_fail;
2778
2779         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2780                 r = -ENOMEM;
2781                 goto out_free_0;
2782         }
2783
2784         r = kvm_arch_hardware_setup();
2785         if (r < 0)
2786                 goto out_free_0a;
2787
2788         for_each_online_cpu(cpu) {
2789                 smp_call_function_single(cpu,
2790                                 kvm_arch_check_processor_compat,
2791                                 &r, 1);
2792                 if (r < 0)
2793                         goto out_free_1;
2794         }
2795
2796         r = register_cpu_notifier(&kvm_cpu_notifier);
2797         if (r)
2798                 goto out_free_2;
2799         register_reboot_notifier(&kvm_reboot_notifier);
2800
2801         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2802         if (!vcpu_align)
2803                 vcpu_align = __alignof__(struct kvm_vcpu);
2804         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2805                                            0, NULL);
2806         if (!kvm_vcpu_cache) {
2807                 r = -ENOMEM;
2808                 goto out_free_3;
2809         }
2810
2811         r = kvm_async_pf_init();
2812         if (r)
2813                 goto out_free;
2814
2815         kvm_chardev_ops.owner = module;
2816         kvm_vm_fops.owner = module;
2817         kvm_vcpu_fops.owner = module;
2818
2819         r = misc_register(&kvm_dev);
2820         if (r) {
2821                 printk(KERN_ERR "kvm: misc device register failed\n");
2822                 goto out_unreg;
2823         }
2824
2825         register_syscore_ops(&kvm_syscore_ops);
2826
2827         kvm_preempt_ops.sched_in = kvm_sched_in;
2828         kvm_preempt_ops.sched_out = kvm_sched_out;
2829
2830         r = kvm_init_debug();
2831         if (r) {
2832                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2833                 goto out_undebugfs;
2834         }
2835
2836         return 0;
2837
2838 out_undebugfs:
2839         unregister_syscore_ops(&kvm_syscore_ops);
2840 out_unreg:
2841         kvm_async_pf_deinit();
2842 out_free:
2843         kmem_cache_destroy(kvm_vcpu_cache);
2844 out_free_3:
2845         unregister_reboot_notifier(&kvm_reboot_notifier);
2846         unregister_cpu_notifier(&kvm_cpu_notifier);
2847 out_free_2:
2848 out_free_1:
2849         kvm_arch_hardware_unsetup();
2850 out_free_0a:
2851         free_cpumask_var(cpus_hardware_enabled);
2852 out_free_0:
2853         kvm_arch_exit();
2854 out_fail:
2855         return r;
2856 }
2857 EXPORT_SYMBOL_GPL(kvm_init);
2858
2859 void kvm_exit(void)
2860 {
2861         kvm_exit_debug();
2862         misc_deregister(&kvm_dev);
2863         kmem_cache_destroy(kvm_vcpu_cache);
2864         kvm_async_pf_deinit();
2865         unregister_syscore_ops(&kvm_syscore_ops);
2866         unregister_reboot_notifier(&kvm_reboot_notifier);
2867         unregister_cpu_notifier(&kvm_cpu_notifier);
2868         on_each_cpu(hardware_disable_nolock, NULL, 1);
2869         kvm_arch_hardware_unsetup();
2870         kvm_arch_exit();
2871         free_cpumask_var(cpus_hardware_enabled);
2872 }
2873 EXPORT_SYMBOL_GPL(kvm_exit);