KVM: sort memslots by its size and use line search
[linux-2.6.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 static struct page *hwpoison_page;
104 static pfn_t hwpoison_pfn;
105
106 struct page *fault_page;
107 pfn_t fault_pfn;
108
109 inline int kvm_is_mmio_pfn(pfn_t pfn)
110 {
111         if (pfn_valid(pfn)) {
112                 int reserved;
113                 struct page *tail = pfn_to_page(pfn);
114                 struct page *head = compound_trans_head(tail);
115                 reserved = PageReserved(head);
116                 if (head != tail) {
117                         /*
118                          * "head" is not a dangling pointer
119                          * (compound_trans_head takes care of that)
120                          * but the hugepage may have been splitted
121                          * from under us (and we may not hold a
122                          * reference count on the head page so it can
123                          * be reused before we run PageReferenced), so
124                          * we've to check PageTail before returning
125                          * what we just read.
126                          */
127                         smp_rmb();
128                         if (PageTail(tail))
129                                 return reserved;
130                 }
131                 return PageReserved(tail);
132         }
133
134         return true;
135 }
136
137 /*
138  * Switches to specified vcpu, until a matching vcpu_put()
139  */
140 void vcpu_load(struct kvm_vcpu *vcpu)
141 {
142         int cpu;
143
144         mutex_lock(&vcpu->mutex);
145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
146                 /* The thread running this VCPU changed. */
147                 struct pid *oldpid = vcpu->pid;
148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
149                 rcu_assign_pointer(vcpu->pid, newpid);
150                 synchronize_rcu();
151                 put_pid(oldpid);
152         }
153         cpu = get_cpu();
154         preempt_notifier_register(&vcpu->preempt_notifier);
155         kvm_arch_vcpu_load(vcpu, cpu);
156         put_cpu();
157 }
158
159 void vcpu_put(struct kvm_vcpu *vcpu)
160 {
161         preempt_disable();
162         kvm_arch_vcpu_put(vcpu);
163         preempt_notifier_unregister(&vcpu->preempt_notifier);
164         preempt_enable();
165         mutex_unlock(&vcpu->mutex);
166 }
167
168 static void ack_flush(void *_completed)
169 {
170 }
171
172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 {
174         int i, cpu, me;
175         cpumask_var_t cpus;
176         bool called = true;
177         struct kvm_vcpu *vcpu;
178
179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180
181         me = get_cpu();
182         kvm_for_each_vcpu(i, vcpu, kvm) {
183                 kvm_make_request(req, vcpu);
184                 cpu = vcpu->cpu;
185
186                 /* Set ->requests bit before we read ->mode */
187                 smp_mb();
188
189                 if (cpus != NULL && cpu != -1 && cpu != me &&
190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191                         cpumask_set_cpu(cpu, cpus);
192         }
193         if (unlikely(cpus == NULL))
194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195         else if (!cpumask_empty(cpus))
196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
197         else
198                 called = false;
199         put_cpu();
200         free_cpumask_var(cpus);
201         return called;
202 }
203
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
205 {
206         int dirty_count = kvm->tlbs_dirty;
207
208         smp_mb();
209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
210                 ++kvm->stat.remote_tlb_flush;
211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
212 }
213
214 void kvm_reload_remote_mmus(struct kvm *kvm)
215 {
216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
217 }
218
219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
220 {
221         struct page *page;
222         int r;
223
224         mutex_init(&vcpu->mutex);
225         vcpu->cpu = -1;
226         vcpu->kvm = kvm;
227         vcpu->vcpu_id = id;
228         vcpu->pid = NULL;
229         init_waitqueue_head(&vcpu->wq);
230         kvm_async_pf_vcpu_init(vcpu);
231
232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
233         if (!page) {
234                 r = -ENOMEM;
235                 goto fail;
236         }
237         vcpu->run = page_address(page);
238
239         r = kvm_arch_vcpu_init(vcpu);
240         if (r < 0)
241                 goto fail_free_run;
242         return 0;
243
244 fail_free_run:
245         free_page((unsigned long)vcpu->run);
246 fail:
247         return r;
248 }
249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250
251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 {
253         put_pid(vcpu->pid);
254         kvm_arch_vcpu_uninit(vcpu);
255         free_page((unsigned long)vcpu->run);
256 }
257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258
259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 {
262         return container_of(mn, struct kvm, mmu_notifier);
263 }
264
265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
266                                              struct mm_struct *mm,
267                                              unsigned long address)
268 {
269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
270         int need_tlb_flush, idx;
271
272         /*
273          * When ->invalidate_page runs, the linux pte has been zapped
274          * already but the page is still allocated until
275          * ->invalidate_page returns. So if we increase the sequence
276          * here the kvm page fault will notice if the spte can't be
277          * established because the page is going to be freed. If
278          * instead the kvm page fault establishes the spte before
279          * ->invalidate_page runs, kvm_unmap_hva will release it
280          * before returning.
281          *
282          * The sequence increase only need to be seen at spin_unlock
283          * time, and not at spin_lock time.
284          *
285          * Increasing the sequence after the spin_unlock would be
286          * unsafe because the kvm page fault could then establish the
287          * pte after kvm_unmap_hva returned, without noticing the page
288          * is going to be freed.
289          */
290         idx = srcu_read_lock(&kvm->srcu);
291         spin_lock(&kvm->mmu_lock);
292         kvm->mmu_notifier_seq++;
293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294         spin_unlock(&kvm->mmu_lock);
295         srcu_read_unlock(&kvm->srcu, idx);
296
297         /* we've to flush the tlb before the pages can be freed */
298         if (need_tlb_flush)
299                 kvm_flush_remote_tlbs(kvm);
300
301 }
302
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304                                         struct mm_struct *mm,
305                                         unsigned long address,
306                                         pte_t pte)
307 {
308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
309         int idx;
310
311         idx = srcu_read_lock(&kvm->srcu);
312         spin_lock(&kvm->mmu_lock);
313         kvm->mmu_notifier_seq++;
314         kvm_set_spte_hva(kvm, address, pte);
315         spin_unlock(&kvm->mmu_lock);
316         srcu_read_unlock(&kvm->srcu, idx);
317 }
318
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320                                                     struct mm_struct *mm,
321                                                     unsigned long start,
322                                                     unsigned long end)
323 {
324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
325         int need_tlb_flush = 0, idx;
326
327         idx = srcu_read_lock(&kvm->srcu);
328         spin_lock(&kvm->mmu_lock);
329         /*
330          * The count increase must become visible at unlock time as no
331          * spte can be established without taking the mmu_lock and
332          * count is also read inside the mmu_lock critical section.
333          */
334         kvm->mmu_notifier_count++;
335         for (; start < end; start += PAGE_SIZE)
336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
337         need_tlb_flush |= kvm->tlbs_dirty;
338         spin_unlock(&kvm->mmu_lock);
339         srcu_read_unlock(&kvm->srcu, idx);
340
341         /* we've to flush the tlb before the pages can be freed */
342         if (need_tlb_flush)
343                 kvm_flush_remote_tlbs(kvm);
344 }
345
346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
347                                                   struct mm_struct *mm,
348                                                   unsigned long start,
349                                                   unsigned long end)
350 {
351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
352
353         spin_lock(&kvm->mmu_lock);
354         /*
355          * This sequence increase will notify the kvm page fault that
356          * the page that is going to be mapped in the spte could have
357          * been freed.
358          */
359         kvm->mmu_notifier_seq++;
360         /*
361          * The above sequence increase must be visible before the
362          * below count decrease but both values are read by the kvm
363          * page fault under mmu_lock spinlock so we don't need to add
364          * a smb_wmb() here in between the two.
365          */
366         kvm->mmu_notifier_count--;
367         spin_unlock(&kvm->mmu_lock);
368
369         BUG_ON(kvm->mmu_notifier_count < 0);
370 }
371
372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
373                                               struct mm_struct *mm,
374                                               unsigned long address)
375 {
376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
377         int young, idx;
378
379         idx = srcu_read_lock(&kvm->srcu);
380         spin_lock(&kvm->mmu_lock);
381         young = kvm_age_hva(kvm, address);
382         spin_unlock(&kvm->mmu_lock);
383         srcu_read_unlock(&kvm->srcu, idx);
384
385         if (young)
386                 kvm_flush_remote_tlbs(kvm);
387
388         return young;
389 }
390
391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392                                        struct mm_struct *mm,
393                                        unsigned long address)
394 {
395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
396         int young, idx;
397
398         idx = srcu_read_lock(&kvm->srcu);
399         spin_lock(&kvm->mmu_lock);
400         young = kvm_test_age_hva(kvm, address);
401         spin_unlock(&kvm->mmu_lock);
402         srcu_read_unlock(&kvm->srcu, idx);
403
404         return young;
405 }
406
407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408                                      struct mm_struct *mm)
409 {
410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
411         int idx;
412
413         idx = srcu_read_lock(&kvm->srcu);
414         kvm_arch_flush_shadow(kvm);
415         srcu_read_unlock(&kvm->srcu, idx);
416 }
417
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
423         .test_young             = kvm_mmu_notifier_test_young,
424         .change_pte             = kvm_mmu_notifier_change_pte,
425         .release                = kvm_mmu_notifier_release,
426 };
427
428 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 {
430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 }
433
434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435
436 static int kvm_init_mmu_notifier(struct kvm *kvm)
437 {
438         return 0;
439 }
440
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442
443 static void kvm_init_memslots_id(struct kvm *kvm)
444 {
445         int i;
446         struct kvm_memslots *slots = kvm->memslots;
447
448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449                 slots->memslots[i].id = i;
450 }
451
452 static struct kvm *kvm_create_vm(void)
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);
461         if (r)
462                 goto out_err_nodisable;
463
464         r = hardware_enable_all();
465         if (r)
466                 goto out_err_nodisable;
467
468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
469         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 #endif
472
473         r = -ENOMEM;
474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
475         if (!kvm->memslots)
476                 goto out_err_nosrcu;
477         kvm_init_memslots_id(kvm);
478         if (init_srcu_struct(&kvm->srcu))
479                 goto out_err_nosrcu;
480         for (i = 0; i < KVM_NR_BUSES; i++) {
481                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
482                                         GFP_KERNEL);
483                 if (!kvm->buses[i])
484                         goto out_err;
485         }
486
487         spin_lock_init(&kvm->mmu_lock);
488         kvm->mm = current->mm;
489         atomic_inc(&kvm->mm->mm_count);
490         kvm_eventfd_init(kvm);
491         mutex_init(&kvm->lock);
492         mutex_init(&kvm->irq_lock);
493         mutex_init(&kvm->slots_lock);
494         atomic_set(&kvm->users_count, 1);
495
496         r = kvm_init_mmu_notifier(kvm);
497         if (r)
498                 goto out_err;
499
500         raw_spin_lock(&kvm_lock);
501         list_add(&kvm->vm_list, &vm_list);
502         raw_spin_unlock(&kvm_lock);
503
504         return kvm;
505
506 out_err:
507         cleanup_srcu_struct(&kvm->srcu);
508 out_err_nosrcu:
509         hardware_disable_all();
510 out_err_nodisable:
511         for (i = 0; i < KVM_NR_BUSES; i++)
512                 kfree(kvm->buses[i]);
513         kfree(kvm->memslots);
514         kvm_arch_free_vm(kvm);
515         return ERR_PTR(r);
516 }
517
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
519 {
520         if (!memslot->dirty_bitmap)
521                 return;
522
523         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
524                 vfree(memslot->dirty_bitmap_head);
525         else
526                 kfree(memslot->dirty_bitmap_head);
527
528         memslot->dirty_bitmap = NULL;
529         memslot->dirty_bitmap_head = NULL;
530 }
531
532 /*
533  * Free any memory in @free but not in @dont.
534  */
535 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
536                                   struct kvm_memory_slot *dont)
537 {
538         int i;
539
540         if (!dont || free->rmap != dont->rmap)
541                 vfree(free->rmap);
542
543         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
544                 kvm_destroy_dirty_bitmap(free);
545
546
547         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
548                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
549                         vfree(free->lpage_info[i]);
550                         free->lpage_info[i] = NULL;
551                 }
552         }
553
554         free->npages = 0;
555         free->rmap = NULL;
556 }
557
558 void kvm_free_physmem(struct kvm *kvm)
559 {
560         struct kvm_memslots *slots = kvm->memslots;
561         struct kvm_memory_slot *memslot;
562
563         kvm_for_each_memslot(memslot, slots)
564                 kvm_free_physmem_slot(memslot, NULL);
565
566         kfree(kvm->memslots);
567 }
568
569 static void kvm_destroy_vm(struct kvm *kvm)
570 {
571         int i;
572         struct mm_struct *mm = kvm->mm;
573
574         kvm_arch_sync_events(kvm);
575         raw_spin_lock(&kvm_lock);
576         list_del(&kvm->vm_list);
577         raw_spin_unlock(&kvm_lock);
578         kvm_free_irq_routing(kvm);
579         for (i = 0; i < KVM_NR_BUSES; i++)
580                 kvm_io_bus_destroy(kvm->buses[i]);
581         kvm_coalesced_mmio_free(kvm);
582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
584 #else
585         kvm_arch_flush_shadow(kvm);
586 #endif
587         kvm_arch_destroy_vm(kvm);
588         kvm_free_physmem(kvm);
589         cleanup_srcu_struct(&kvm->srcu);
590         kvm_arch_free_vm(kvm);
591         hardware_disable_all();
592         mmdrop(mm);
593 }
594
595 void kvm_get_kvm(struct kvm *kvm)
596 {
597         atomic_inc(&kvm->users_count);
598 }
599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
600
601 void kvm_put_kvm(struct kvm *kvm)
602 {
603         if (atomic_dec_and_test(&kvm->users_count))
604                 kvm_destroy_vm(kvm);
605 }
606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
607
608
609 static int kvm_vm_release(struct inode *inode, struct file *filp)
610 {
611         struct kvm *kvm = filp->private_data;
612
613         kvm_irqfd_release(kvm);
614
615         kvm_put_kvm(kvm);
616         return 0;
617 }
618
619 #ifndef CONFIG_S390
620 /*
621  * Allocation size is twice as large as the actual dirty bitmap size.
622  * This makes it possible to do double buffering: see x86's
623  * kvm_vm_ioctl_get_dirty_log().
624  */
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626 {
627         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
628
629         if (dirty_bytes > PAGE_SIZE)
630                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
631         else
632                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
633
634         if (!memslot->dirty_bitmap)
635                 return -ENOMEM;
636
637         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
638         memslot->nr_dirty_pages = 0;
639         return 0;
640 }
641 #endif /* !CONFIG_S390 */
642
643 static struct kvm_memory_slot *
644 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
645 {
646         struct kvm_memory_slot *memslot;
647
648         kvm_for_each_memslot(memslot, slots)
649                 if (gfn >= memslot->base_gfn &&
650                       gfn < memslot->base_gfn + memslot->npages)
651                         return memslot;
652
653         return NULL;
654 }
655
656 static int cmp_memslot(const void *slot1, const void *slot2)
657 {
658         struct kvm_memory_slot *s1, *s2;
659
660         s1 = (struct kvm_memory_slot *)slot1;
661         s2 = (struct kvm_memory_slot *)slot2;
662
663         if (s1->npages < s2->npages)
664                 return 1;
665         if (s1->npages > s2->npages)
666                 return -1;
667
668         return 0;
669 }
670
671 /*
672  * Sort the memslots base on its size, so the larger slots
673  * will get better fit.
674  */
675 static void sort_memslots(struct kvm_memslots *slots)
676 {
677         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
678               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
679 }
680
681 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
682 {
683         if (new) {
684                 int id = new->id;
685                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
686                 unsigned long npages = old->npages;
687
688                 *old = *new;
689                 if (new->npages != npages)
690                         sort_memslots(slots);
691         }
692
693         slots->generation++;
694 }
695
696 /*
697  * Allocate some memory and give it an address in the guest physical address
698  * space.
699  *
700  * Discontiguous memory is allowed, mostly for framebuffers.
701  *
702  * Must be called holding mmap_sem for write.
703  */
704 int __kvm_set_memory_region(struct kvm *kvm,
705                             struct kvm_userspace_memory_region *mem,
706                             int user_alloc)
707 {
708         int r;
709         gfn_t base_gfn;
710         unsigned long npages;
711         unsigned long i;
712         struct kvm_memory_slot *memslot;
713         struct kvm_memory_slot old, new;
714         struct kvm_memslots *slots, *old_memslots;
715
716         r = -EINVAL;
717         /* General sanity checks */
718         if (mem->memory_size & (PAGE_SIZE - 1))
719                 goto out;
720         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
721                 goto out;
722         /* We can read the guest memory with __xxx_user() later on. */
723         if (user_alloc &&
724             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
725              !access_ok(VERIFY_WRITE,
726                         (void __user *)(unsigned long)mem->userspace_addr,
727                         mem->memory_size)))
728                 goto out;
729         if (mem->slot >= KVM_MEM_SLOTS_NUM)
730                 goto out;
731         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
732                 goto out;
733
734         memslot = id_to_memslot(kvm->memslots, mem->slot);
735         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
736         npages = mem->memory_size >> PAGE_SHIFT;
737
738         r = -EINVAL;
739         if (npages > KVM_MEM_MAX_NR_PAGES)
740                 goto out;
741
742         if (!npages)
743                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
744
745         new = old = *memslot;
746
747         new.id = mem->slot;
748         new.base_gfn = base_gfn;
749         new.npages = npages;
750         new.flags = mem->flags;
751
752         /* Disallow changing a memory slot's size. */
753         r = -EINVAL;
754         if (npages && old.npages && npages != old.npages)
755                 goto out_free;
756
757         /* Check for overlaps */
758         r = -EEXIST;
759         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
760                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
761
762                 if (s == memslot || !s->npages)
763                         continue;
764                 if (!((base_gfn + npages <= s->base_gfn) ||
765                       (base_gfn >= s->base_gfn + s->npages)))
766                         goto out_free;
767         }
768
769         /* Free page dirty bitmap if unneeded */
770         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
771                 new.dirty_bitmap = NULL;
772
773         r = -ENOMEM;
774
775         /* Allocate if a slot is being created */
776 #ifndef CONFIG_S390
777         if (npages && !new.rmap) {
778                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
779
780                 if (!new.rmap)
781                         goto out_free;
782
783                 new.user_alloc = user_alloc;
784                 new.userspace_addr = mem->userspace_addr;
785         }
786         if (!npages)
787                 goto skip_lpage;
788
789         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
790                 unsigned long ugfn;
791                 unsigned long j;
792                 int lpages;
793                 int level = i + 2;
794
795                 /* Avoid unused variable warning if no large pages */
796                 (void)level;
797
798                 if (new.lpage_info[i])
799                         continue;
800
801                 lpages = 1 + ((base_gfn + npages - 1)
802                              >> KVM_HPAGE_GFN_SHIFT(level));
803                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
804
805                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
806
807                 if (!new.lpage_info[i])
808                         goto out_free;
809
810                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
811                         new.lpage_info[i][0].write_count = 1;
812                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
813                         new.lpage_info[i][lpages - 1].write_count = 1;
814                 ugfn = new.userspace_addr >> PAGE_SHIFT;
815                 /*
816                  * If the gfn and userspace address are not aligned wrt each
817                  * other, or if explicitly asked to, disable large page
818                  * support for this slot
819                  */
820                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
821                     !largepages_enabled)
822                         for (j = 0; j < lpages; ++j)
823                                 new.lpage_info[i][j].write_count = 1;
824         }
825
826 skip_lpage:
827
828         /* Allocate page dirty bitmap if needed */
829         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
830                 if (kvm_create_dirty_bitmap(&new) < 0)
831                         goto out_free;
832                 /* destroy any largepage mappings for dirty tracking */
833         }
834 #else  /* not defined CONFIG_S390 */
835         new.user_alloc = user_alloc;
836         if (user_alloc)
837                 new.userspace_addr = mem->userspace_addr;
838 #endif /* not defined CONFIG_S390 */
839
840         if (!npages) {
841                 struct kvm_memory_slot *slot;
842
843                 r = -ENOMEM;
844                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
845                                 GFP_KERNEL);
846                 if (!slots)
847                         goto out_free;
848                 slot = id_to_memslot(slots, mem->slot);
849                 slot->flags |= KVM_MEMSLOT_INVALID;
850
851                 update_memslots(slots, NULL);
852
853                 old_memslots = kvm->memslots;
854                 rcu_assign_pointer(kvm->memslots, slots);
855                 synchronize_srcu_expedited(&kvm->srcu);
856                 /* From this point no new shadow pages pointing to a deleted
857                  * memslot will be created.
858                  *
859                  * validation of sp->gfn happens in:
860                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
861                  *      - kvm_is_visible_gfn (mmu_check_roots)
862                  */
863                 kvm_arch_flush_shadow(kvm);
864                 kfree(old_memslots);
865         }
866
867         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
868         if (r)
869                 goto out_free;
870
871         /* map the pages in iommu page table */
872         if (npages) {
873                 r = kvm_iommu_map_pages(kvm, &new);
874                 if (r)
875                         goto out_free;
876         }
877
878         r = -ENOMEM;
879         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
880                         GFP_KERNEL);
881         if (!slots)
882                 goto out_free;
883
884         /* actual memory is freed via old in kvm_free_physmem_slot below */
885         if (!npages) {
886                 new.rmap = NULL;
887                 new.dirty_bitmap = NULL;
888                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
889                         new.lpage_info[i] = NULL;
890         }
891
892         update_memslots(slots, &new);
893         old_memslots = kvm->memslots;
894         rcu_assign_pointer(kvm->memslots, slots);
895         synchronize_srcu_expedited(&kvm->srcu);
896
897         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
898
899         /*
900          * If the new memory slot is created, we need to clear all
901          * mmio sptes.
902          */
903         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
904                 kvm_arch_flush_shadow(kvm);
905
906         kvm_free_physmem_slot(&old, &new);
907         kfree(old_memslots);
908
909         return 0;
910
911 out_free:
912         kvm_free_physmem_slot(&new, &old);
913 out:
914         return r;
915
916 }
917 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
918
919 int kvm_set_memory_region(struct kvm *kvm,
920                           struct kvm_userspace_memory_region *mem,
921                           int user_alloc)
922 {
923         int r;
924
925         mutex_lock(&kvm->slots_lock);
926         r = __kvm_set_memory_region(kvm, mem, user_alloc);
927         mutex_unlock(&kvm->slots_lock);
928         return r;
929 }
930 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
931
932 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
933                                    struct
934                                    kvm_userspace_memory_region *mem,
935                                    int user_alloc)
936 {
937         if (mem->slot >= KVM_MEMORY_SLOTS)
938                 return -EINVAL;
939         return kvm_set_memory_region(kvm, mem, user_alloc);
940 }
941
942 int kvm_get_dirty_log(struct kvm *kvm,
943                         struct kvm_dirty_log *log, int *is_dirty)
944 {
945         struct kvm_memory_slot *memslot;
946         int r, i;
947         unsigned long n;
948         unsigned long any = 0;
949
950         r = -EINVAL;
951         if (log->slot >= KVM_MEMORY_SLOTS)
952                 goto out;
953
954         memslot = id_to_memslot(kvm->memslots, log->slot);
955         r = -ENOENT;
956         if (!memslot->dirty_bitmap)
957                 goto out;
958
959         n = kvm_dirty_bitmap_bytes(memslot);
960
961         for (i = 0; !any && i < n/sizeof(long); ++i)
962                 any = memslot->dirty_bitmap[i];
963
964         r = -EFAULT;
965         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
966                 goto out;
967
968         if (any)
969                 *is_dirty = 1;
970
971         r = 0;
972 out:
973         return r;
974 }
975
976 void kvm_disable_largepages(void)
977 {
978         largepages_enabled = false;
979 }
980 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
981
982 int is_error_page(struct page *page)
983 {
984         return page == bad_page || page == hwpoison_page || page == fault_page;
985 }
986 EXPORT_SYMBOL_GPL(is_error_page);
987
988 int is_error_pfn(pfn_t pfn)
989 {
990         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
991 }
992 EXPORT_SYMBOL_GPL(is_error_pfn);
993
994 int is_hwpoison_pfn(pfn_t pfn)
995 {
996         return pfn == hwpoison_pfn;
997 }
998 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
999
1000 int is_fault_pfn(pfn_t pfn)
1001 {
1002         return pfn == fault_pfn;
1003 }
1004 EXPORT_SYMBOL_GPL(is_fault_pfn);
1005
1006 int is_noslot_pfn(pfn_t pfn)
1007 {
1008         return pfn == bad_pfn;
1009 }
1010 EXPORT_SYMBOL_GPL(is_noslot_pfn);
1011
1012 int is_invalid_pfn(pfn_t pfn)
1013 {
1014         return pfn == hwpoison_pfn || pfn == fault_pfn;
1015 }
1016 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1017
1018 static inline unsigned long bad_hva(void)
1019 {
1020         return PAGE_OFFSET;
1021 }
1022
1023 int kvm_is_error_hva(unsigned long addr)
1024 {
1025         return addr == bad_hva();
1026 }
1027 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1028
1029 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
1030                                                 gfn_t gfn)
1031 {
1032         return search_memslots(slots, gfn);
1033 }
1034
1035 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1036 {
1037         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1038 }
1039 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1040
1041 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1042 {
1043         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1044
1045         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1046               memslot->flags & KVM_MEMSLOT_INVALID)
1047                 return 0;
1048
1049         return 1;
1050 }
1051 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1052
1053 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1054 {
1055         struct vm_area_struct *vma;
1056         unsigned long addr, size;
1057
1058         size = PAGE_SIZE;
1059
1060         addr = gfn_to_hva(kvm, gfn);
1061         if (kvm_is_error_hva(addr))
1062                 return PAGE_SIZE;
1063
1064         down_read(&current->mm->mmap_sem);
1065         vma = find_vma(current->mm, addr);
1066         if (!vma)
1067                 goto out;
1068
1069         size = vma_kernel_pagesize(vma);
1070
1071 out:
1072         up_read(&current->mm->mmap_sem);
1073
1074         return size;
1075 }
1076
1077 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1078                                      gfn_t *nr_pages)
1079 {
1080         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1081                 return bad_hva();
1082
1083         if (nr_pages)
1084                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1085
1086         return gfn_to_hva_memslot(slot, gfn);
1087 }
1088
1089 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1090 {
1091         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1092 }
1093 EXPORT_SYMBOL_GPL(gfn_to_hva);
1094
1095 static pfn_t get_fault_pfn(void)
1096 {
1097         get_page(fault_page);
1098         return fault_pfn;
1099 }
1100
1101 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1102         unsigned long start, int write, struct page **page)
1103 {
1104         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1105
1106         if (write)
1107                 flags |= FOLL_WRITE;
1108
1109         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1110 }
1111
1112 static inline int check_user_page_hwpoison(unsigned long addr)
1113 {
1114         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1115
1116         rc = __get_user_pages(current, current->mm, addr, 1,
1117                               flags, NULL, NULL, NULL);
1118         return rc == -EHWPOISON;
1119 }
1120
1121 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1122                         bool *async, bool write_fault, bool *writable)
1123 {
1124         struct page *page[1];
1125         int npages = 0;
1126         pfn_t pfn;
1127
1128         /* we can do it either atomically or asynchronously, not both */
1129         BUG_ON(atomic && async);
1130
1131         BUG_ON(!write_fault && !writable);
1132
1133         if (writable)
1134                 *writable = true;
1135
1136         if (atomic || async)
1137                 npages = __get_user_pages_fast(addr, 1, 1, page);
1138
1139         if (unlikely(npages != 1) && !atomic) {
1140                 might_sleep();
1141
1142                 if (writable)
1143                         *writable = write_fault;
1144
1145                 if (async) {
1146                         down_read(&current->mm->mmap_sem);
1147                         npages = get_user_page_nowait(current, current->mm,
1148                                                      addr, write_fault, page);
1149                         up_read(&current->mm->mmap_sem);
1150                 } else
1151                         npages = get_user_pages_fast(addr, 1, write_fault,
1152                                                      page);
1153
1154                 /* map read fault as writable if possible */
1155                 if (unlikely(!write_fault) && npages == 1) {
1156                         struct page *wpage[1];
1157
1158                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1159                         if (npages == 1) {
1160                                 *writable = true;
1161                                 put_page(page[0]);
1162                                 page[0] = wpage[0];
1163                         }
1164                         npages = 1;
1165                 }
1166         }
1167
1168         if (unlikely(npages != 1)) {
1169                 struct vm_area_struct *vma;
1170
1171                 if (atomic)
1172                         return get_fault_pfn();
1173
1174                 down_read(&current->mm->mmap_sem);
1175                 if (npages == -EHWPOISON ||
1176                         (!async && check_user_page_hwpoison(addr))) {
1177                         up_read(&current->mm->mmap_sem);
1178                         get_page(hwpoison_page);
1179                         return page_to_pfn(hwpoison_page);
1180                 }
1181
1182                 vma = find_vma_intersection(current->mm, addr, addr+1);
1183
1184                 if (vma == NULL)
1185                         pfn = get_fault_pfn();
1186                 else if ((vma->vm_flags & VM_PFNMAP)) {
1187                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1188                                 vma->vm_pgoff;
1189                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1190                 } else {
1191                         if (async && (vma->vm_flags & VM_WRITE))
1192                                 *async = true;
1193                         pfn = get_fault_pfn();
1194                 }
1195                 up_read(&current->mm->mmap_sem);
1196         } else
1197                 pfn = page_to_pfn(page[0]);
1198
1199         return pfn;
1200 }
1201
1202 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1203 {
1204         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1205 }
1206 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1207
1208 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1209                           bool write_fault, bool *writable)
1210 {
1211         unsigned long addr;
1212
1213         if (async)
1214                 *async = false;
1215
1216         addr = gfn_to_hva(kvm, gfn);
1217         if (kvm_is_error_hva(addr)) {
1218                 get_page(bad_page);
1219                 return page_to_pfn(bad_page);
1220         }
1221
1222         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1223 }
1224
1225 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1226 {
1227         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1228 }
1229 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1230
1231 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1232                        bool write_fault, bool *writable)
1233 {
1234         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1235 }
1236 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1237
1238 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1239 {
1240         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1241 }
1242 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1243
1244 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1245                       bool *writable)
1246 {
1247         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1248 }
1249 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1250
1251 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1252                          struct kvm_memory_slot *slot, gfn_t gfn)
1253 {
1254         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1255         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1256 }
1257
1258 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1259                                                                   int nr_pages)
1260 {
1261         unsigned long addr;
1262         gfn_t entry;
1263
1264         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1265         if (kvm_is_error_hva(addr))
1266                 return -1;
1267
1268         if (entry < nr_pages)
1269                 return 0;
1270
1271         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1272 }
1273 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1274
1275 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1276 {
1277         pfn_t pfn;
1278
1279         pfn = gfn_to_pfn(kvm, gfn);
1280         if (!kvm_is_mmio_pfn(pfn))
1281                 return pfn_to_page(pfn);
1282
1283         WARN_ON(kvm_is_mmio_pfn(pfn));
1284
1285         get_page(bad_page);
1286         return bad_page;
1287 }
1288
1289 EXPORT_SYMBOL_GPL(gfn_to_page);
1290
1291 void kvm_release_page_clean(struct page *page)
1292 {
1293         kvm_release_pfn_clean(page_to_pfn(page));
1294 }
1295 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1296
1297 void kvm_release_pfn_clean(pfn_t pfn)
1298 {
1299         if (!kvm_is_mmio_pfn(pfn))
1300                 put_page(pfn_to_page(pfn));
1301 }
1302 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1303
1304 void kvm_release_page_dirty(struct page *page)
1305 {
1306         kvm_release_pfn_dirty(page_to_pfn(page));
1307 }
1308 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1309
1310 void kvm_release_pfn_dirty(pfn_t pfn)
1311 {
1312         kvm_set_pfn_dirty(pfn);
1313         kvm_release_pfn_clean(pfn);
1314 }
1315 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1316
1317 void kvm_set_page_dirty(struct page *page)
1318 {
1319         kvm_set_pfn_dirty(page_to_pfn(page));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1322
1323 void kvm_set_pfn_dirty(pfn_t pfn)
1324 {
1325         if (!kvm_is_mmio_pfn(pfn)) {
1326                 struct page *page = pfn_to_page(pfn);
1327                 if (!PageReserved(page))
1328                         SetPageDirty(page);
1329         }
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1332
1333 void kvm_set_pfn_accessed(pfn_t pfn)
1334 {
1335         if (!kvm_is_mmio_pfn(pfn))
1336                 mark_page_accessed(pfn_to_page(pfn));
1337 }
1338 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1339
1340 void kvm_get_pfn(pfn_t pfn)
1341 {
1342         if (!kvm_is_mmio_pfn(pfn))
1343                 get_page(pfn_to_page(pfn));
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1346
1347 static int next_segment(unsigned long len, int offset)
1348 {
1349         if (len > PAGE_SIZE - offset)
1350                 return PAGE_SIZE - offset;
1351         else
1352                 return len;
1353 }
1354
1355 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1356                         int len)
1357 {
1358         int r;
1359         unsigned long addr;
1360
1361         addr = gfn_to_hva(kvm, gfn);
1362         if (kvm_is_error_hva(addr))
1363                 return -EFAULT;
1364         r = __copy_from_user(data, (void __user *)addr + offset, len);
1365         if (r)
1366                 return -EFAULT;
1367         return 0;
1368 }
1369 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1370
1371 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1372 {
1373         gfn_t gfn = gpa >> PAGE_SHIFT;
1374         int seg;
1375         int offset = offset_in_page(gpa);
1376         int ret;
1377
1378         while ((seg = next_segment(len, offset)) != 0) {
1379                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1380                 if (ret < 0)
1381                         return ret;
1382                 offset = 0;
1383                 len -= seg;
1384                 data += seg;
1385                 ++gfn;
1386         }
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_read_guest);
1390
1391 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1392                           unsigned long len)
1393 {
1394         int r;
1395         unsigned long addr;
1396         gfn_t gfn = gpa >> PAGE_SHIFT;
1397         int offset = offset_in_page(gpa);
1398
1399         addr = gfn_to_hva(kvm, gfn);
1400         if (kvm_is_error_hva(addr))
1401                 return -EFAULT;
1402         pagefault_disable();
1403         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1404         pagefault_enable();
1405         if (r)
1406                 return -EFAULT;
1407         return 0;
1408 }
1409 EXPORT_SYMBOL(kvm_read_guest_atomic);
1410
1411 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1412                          int offset, int len)
1413 {
1414         int r;
1415         unsigned long addr;
1416
1417         addr = gfn_to_hva(kvm, gfn);
1418         if (kvm_is_error_hva(addr))
1419                 return -EFAULT;
1420         r = __copy_to_user((void __user *)addr + offset, data, len);
1421         if (r)
1422                 return -EFAULT;
1423         mark_page_dirty(kvm, gfn);
1424         return 0;
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1427
1428 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1429                     unsigned long len)
1430 {
1431         gfn_t gfn = gpa >> PAGE_SHIFT;
1432         int seg;
1433         int offset = offset_in_page(gpa);
1434         int ret;
1435
1436         while ((seg = next_segment(len, offset)) != 0) {
1437                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1438                 if (ret < 0)
1439                         return ret;
1440                 offset = 0;
1441                 len -= seg;
1442                 data += seg;
1443                 ++gfn;
1444         }
1445         return 0;
1446 }
1447
1448 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1449                               gpa_t gpa)
1450 {
1451         struct kvm_memslots *slots = kvm_memslots(kvm);
1452         int offset = offset_in_page(gpa);
1453         gfn_t gfn = gpa >> PAGE_SHIFT;
1454
1455         ghc->gpa = gpa;
1456         ghc->generation = slots->generation;
1457         ghc->memslot = __gfn_to_memslot(slots, gfn);
1458         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1459         if (!kvm_is_error_hva(ghc->hva))
1460                 ghc->hva += offset;
1461         else
1462                 return -EFAULT;
1463
1464         return 0;
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1467
1468 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1469                            void *data, unsigned long len)
1470 {
1471         struct kvm_memslots *slots = kvm_memslots(kvm);
1472         int r;
1473
1474         if (slots->generation != ghc->generation)
1475                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1476
1477         if (kvm_is_error_hva(ghc->hva))
1478                 return -EFAULT;
1479
1480         r = __copy_to_user((void __user *)ghc->hva, data, len);
1481         if (r)
1482                 return -EFAULT;
1483         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1484
1485         return 0;
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1488
1489 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1490                            void *data, unsigned long len)
1491 {
1492         struct kvm_memslots *slots = kvm_memslots(kvm);
1493         int r;
1494
1495         if (slots->generation != ghc->generation)
1496                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1497
1498         if (kvm_is_error_hva(ghc->hva))
1499                 return -EFAULT;
1500
1501         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1502         if (r)
1503                 return -EFAULT;
1504
1505         return 0;
1506 }
1507 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1508
1509 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1510 {
1511         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1512                                     offset, len);
1513 }
1514 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1515
1516 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1517 {
1518         gfn_t gfn = gpa >> PAGE_SHIFT;
1519         int seg;
1520         int offset = offset_in_page(gpa);
1521         int ret;
1522
1523         while ((seg = next_segment(len, offset)) != 0) {
1524                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1525                 if (ret < 0)
1526                         return ret;
1527                 offset = 0;
1528                 len -= seg;
1529                 ++gfn;
1530         }
1531         return 0;
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1534
1535 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1536                              gfn_t gfn)
1537 {
1538         if (memslot && memslot->dirty_bitmap) {
1539                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1540
1541                 if (!__test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1542                         memslot->nr_dirty_pages++;
1543         }
1544 }
1545
1546 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1547 {
1548         struct kvm_memory_slot *memslot;
1549
1550         memslot = gfn_to_memslot(kvm, gfn);
1551         mark_page_dirty_in_slot(kvm, memslot, gfn);
1552 }
1553
1554 /*
1555  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1556  */
1557 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1558 {
1559         DEFINE_WAIT(wait);
1560
1561         for (;;) {
1562                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1563
1564                 if (kvm_arch_vcpu_runnable(vcpu)) {
1565                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1566                         break;
1567                 }
1568                 if (kvm_cpu_has_pending_timer(vcpu))
1569                         break;
1570                 if (signal_pending(current))
1571                         break;
1572
1573                 schedule();
1574         }
1575
1576         finish_wait(&vcpu->wq, &wait);
1577 }
1578
1579 void kvm_resched(struct kvm_vcpu *vcpu)
1580 {
1581         if (!need_resched())
1582                 return;
1583         cond_resched();
1584 }
1585 EXPORT_SYMBOL_GPL(kvm_resched);
1586
1587 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1588 {
1589         struct kvm *kvm = me->kvm;
1590         struct kvm_vcpu *vcpu;
1591         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1592         int yielded = 0;
1593         int pass;
1594         int i;
1595
1596         /*
1597          * We boost the priority of a VCPU that is runnable but not
1598          * currently running, because it got preempted by something
1599          * else and called schedule in __vcpu_run.  Hopefully that
1600          * VCPU is holding the lock that we need and will release it.
1601          * We approximate round-robin by starting at the last boosted VCPU.
1602          */
1603         for (pass = 0; pass < 2 && !yielded; pass++) {
1604                 kvm_for_each_vcpu(i, vcpu, kvm) {
1605                         struct task_struct *task = NULL;
1606                         struct pid *pid;
1607                         if (!pass && i < last_boosted_vcpu) {
1608                                 i = last_boosted_vcpu;
1609                                 continue;
1610                         } else if (pass && i > last_boosted_vcpu)
1611                                 break;
1612                         if (vcpu == me)
1613                                 continue;
1614                         if (waitqueue_active(&vcpu->wq))
1615                                 continue;
1616                         rcu_read_lock();
1617                         pid = rcu_dereference(vcpu->pid);
1618                         if (pid)
1619                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1620                         rcu_read_unlock();
1621                         if (!task)
1622                                 continue;
1623                         if (task->flags & PF_VCPU) {
1624                                 put_task_struct(task);
1625                                 continue;
1626                         }
1627                         if (yield_to(task, 1)) {
1628                                 put_task_struct(task);
1629                                 kvm->last_boosted_vcpu = i;
1630                                 yielded = 1;
1631                                 break;
1632                         }
1633                         put_task_struct(task);
1634                 }
1635         }
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1638
1639 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1640 {
1641         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1642         struct page *page;
1643
1644         if (vmf->pgoff == 0)
1645                 page = virt_to_page(vcpu->run);
1646 #ifdef CONFIG_X86
1647         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1648                 page = virt_to_page(vcpu->arch.pio_data);
1649 #endif
1650 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1651         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1652                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1653 #endif
1654         else
1655                 return VM_FAULT_SIGBUS;
1656         get_page(page);
1657         vmf->page = page;
1658         return 0;
1659 }
1660
1661 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1662         .fault = kvm_vcpu_fault,
1663 };
1664
1665 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1666 {
1667         vma->vm_ops = &kvm_vcpu_vm_ops;
1668         return 0;
1669 }
1670
1671 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1672 {
1673         struct kvm_vcpu *vcpu = filp->private_data;
1674
1675         kvm_put_kvm(vcpu->kvm);
1676         return 0;
1677 }
1678
1679 static struct file_operations kvm_vcpu_fops = {
1680         .release        = kvm_vcpu_release,
1681         .unlocked_ioctl = kvm_vcpu_ioctl,
1682 #ifdef CONFIG_COMPAT
1683         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1684 #endif
1685         .mmap           = kvm_vcpu_mmap,
1686         .llseek         = noop_llseek,
1687 };
1688
1689 /*
1690  * Allocates an inode for the vcpu.
1691  */
1692 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1693 {
1694         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1695 }
1696
1697 /*
1698  * Creates some virtual cpus.  Good luck creating more than one.
1699  */
1700 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1701 {
1702         int r;
1703         struct kvm_vcpu *vcpu, *v;
1704
1705         vcpu = kvm_arch_vcpu_create(kvm, id);
1706         if (IS_ERR(vcpu))
1707                 return PTR_ERR(vcpu);
1708
1709         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1710
1711         r = kvm_arch_vcpu_setup(vcpu);
1712         if (r)
1713                 goto vcpu_destroy;
1714
1715         mutex_lock(&kvm->lock);
1716         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1717                 r = -EINVAL;
1718                 goto unlock_vcpu_destroy;
1719         }
1720
1721         kvm_for_each_vcpu(r, v, kvm)
1722                 if (v->vcpu_id == id) {
1723                         r = -EEXIST;
1724                         goto unlock_vcpu_destroy;
1725                 }
1726
1727         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1728
1729         /* Now it's all set up, let userspace reach it */
1730         kvm_get_kvm(kvm);
1731         r = create_vcpu_fd(vcpu);
1732         if (r < 0) {
1733                 kvm_put_kvm(kvm);
1734                 goto unlock_vcpu_destroy;
1735         }
1736
1737         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1738         smp_wmb();
1739         atomic_inc(&kvm->online_vcpus);
1740
1741 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1742         if (kvm->bsp_vcpu_id == id)
1743                 kvm->bsp_vcpu = vcpu;
1744 #endif
1745         mutex_unlock(&kvm->lock);
1746         return r;
1747
1748 unlock_vcpu_destroy:
1749         mutex_unlock(&kvm->lock);
1750 vcpu_destroy:
1751         kvm_arch_vcpu_destroy(vcpu);
1752         return r;
1753 }
1754
1755 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1756 {
1757         if (sigset) {
1758                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1759                 vcpu->sigset_active = 1;
1760                 vcpu->sigset = *sigset;
1761         } else
1762                 vcpu->sigset_active = 0;
1763         return 0;
1764 }
1765
1766 static long kvm_vcpu_ioctl(struct file *filp,
1767                            unsigned int ioctl, unsigned long arg)
1768 {
1769         struct kvm_vcpu *vcpu = filp->private_data;
1770         void __user *argp = (void __user *)arg;
1771         int r;
1772         struct kvm_fpu *fpu = NULL;
1773         struct kvm_sregs *kvm_sregs = NULL;
1774
1775         if (vcpu->kvm->mm != current->mm)
1776                 return -EIO;
1777
1778 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1779         /*
1780          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1781          * so vcpu_load() would break it.
1782          */
1783         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1784                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1785 #endif
1786
1787
1788         vcpu_load(vcpu);
1789         switch (ioctl) {
1790         case KVM_RUN:
1791                 r = -EINVAL;
1792                 if (arg)
1793                         goto out;
1794                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1795                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1796                 break;
1797         case KVM_GET_REGS: {
1798                 struct kvm_regs *kvm_regs;
1799
1800                 r = -ENOMEM;
1801                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1802                 if (!kvm_regs)
1803                         goto out;
1804                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1805                 if (r)
1806                         goto out_free1;
1807                 r = -EFAULT;
1808                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1809                         goto out_free1;
1810                 r = 0;
1811 out_free1:
1812                 kfree(kvm_regs);
1813                 break;
1814         }
1815         case KVM_SET_REGS: {
1816                 struct kvm_regs *kvm_regs;
1817
1818                 r = -ENOMEM;
1819                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1820                 if (!kvm_regs)
1821                         goto out;
1822                 r = -EFAULT;
1823                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1824                         goto out_free2;
1825                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1826                 if (r)
1827                         goto out_free2;
1828                 r = 0;
1829 out_free2:
1830                 kfree(kvm_regs);
1831                 break;
1832         }
1833         case KVM_GET_SREGS: {
1834                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1835                 r = -ENOMEM;
1836                 if (!kvm_sregs)
1837                         goto out;
1838                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1839                 if (r)
1840                         goto out;
1841                 r = -EFAULT;
1842                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_SET_SREGS: {
1848                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1849                 r = -ENOMEM;
1850                 if (!kvm_sregs)
1851                         goto out;
1852                 r = -EFAULT;
1853                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1854                         goto out;
1855                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1856                 if (r)
1857                         goto out;
1858                 r = 0;
1859                 break;
1860         }
1861         case KVM_GET_MP_STATE: {
1862                 struct kvm_mp_state mp_state;
1863
1864                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1865                 if (r)
1866                         goto out;
1867                 r = -EFAULT;
1868                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1869                         goto out;
1870                 r = 0;
1871                 break;
1872         }
1873         case KVM_SET_MP_STATE: {
1874                 struct kvm_mp_state mp_state;
1875
1876                 r = -EFAULT;
1877                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1878                         goto out;
1879                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1880                 if (r)
1881                         goto out;
1882                 r = 0;
1883                 break;
1884         }
1885         case KVM_TRANSLATE: {
1886                 struct kvm_translation tr;
1887
1888                 r = -EFAULT;
1889                 if (copy_from_user(&tr, argp, sizeof tr))
1890                         goto out;
1891                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1892                 if (r)
1893                         goto out;
1894                 r = -EFAULT;
1895                 if (copy_to_user(argp, &tr, sizeof tr))
1896                         goto out;
1897                 r = 0;
1898                 break;
1899         }
1900         case KVM_SET_GUEST_DEBUG: {
1901                 struct kvm_guest_debug dbg;
1902
1903                 r = -EFAULT;
1904                 if (copy_from_user(&dbg, argp, sizeof dbg))
1905                         goto out;
1906                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1907                 if (r)
1908                         goto out;
1909                 r = 0;
1910                 break;
1911         }
1912         case KVM_SET_SIGNAL_MASK: {
1913                 struct kvm_signal_mask __user *sigmask_arg = argp;
1914                 struct kvm_signal_mask kvm_sigmask;
1915                 sigset_t sigset, *p;
1916
1917                 p = NULL;
1918                 if (argp) {
1919                         r = -EFAULT;
1920                         if (copy_from_user(&kvm_sigmask, argp,
1921                                            sizeof kvm_sigmask))
1922                                 goto out;
1923                         r = -EINVAL;
1924                         if (kvm_sigmask.len != sizeof sigset)
1925                                 goto out;
1926                         r = -EFAULT;
1927                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1928                                            sizeof sigset))
1929                                 goto out;
1930                         p = &sigset;
1931                 }
1932                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1933                 break;
1934         }
1935         case KVM_GET_FPU: {
1936                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1937                 r = -ENOMEM;
1938                 if (!fpu)
1939                         goto out;
1940                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1941                 if (r)
1942                         goto out;
1943                 r = -EFAULT;
1944                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1945                         goto out;
1946                 r = 0;
1947                 break;
1948         }
1949         case KVM_SET_FPU: {
1950                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1951                 r = -ENOMEM;
1952                 if (!fpu)
1953                         goto out;
1954                 r = -EFAULT;
1955                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1956                         goto out;
1957                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1958                 if (r)
1959                         goto out;
1960                 r = 0;
1961                 break;
1962         }
1963         default:
1964                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1965         }
1966 out:
1967         vcpu_put(vcpu);
1968         kfree(fpu);
1969         kfree(kvm_sregs);
1970         return r;
1971 }
1972
1973 #ifdef CONFIG_COMPAT
1974 static long kvm_vcpu_compat_ioctl(struct file *filp,
1975                                   unsigned int ioctl, unsigned long arg)
1976 {
1977         struct kvm_vcpu *vcpu = filp->private_data;
1978         void __user *argp = compat_ptr(arg);
1979         int r;
1980
1981         if (vcpu->kvm->mm != current->mm)
1982                 return -EIO;
1983
1984         switch (ioctl) {
1985         case KVM_SET_SIGNAL_MASK: {
1986                 struct kvm_signal_mask __user *sigmask_arg = argp;
1987                 struct kvm_signal_mask kvm_sigmask;
1988                 compat_sigset_t csigset;
1989                 sigset_t sigset;
1990
1991                 if (argp) {
1992                         r = -EFAULT;
1993                         if (copy_from_user(&kvm_sigmask, argp,
1994                                            sizeof kvm_sigmask))
1995                                 goto out;
1996                         r = -EINVAL;
1997                         if (kvm_sigmask.len != sizeof csigset)
1998                                 goto out;
1999                         r = -EFAULT;
2000                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2001                                            sizeof csigset))
2002                                 goto out;
2003                 }
2004                 sigset_from_compat(&sigset, &csigset);
2005                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2006                 break;
2007         }
2008         default:
2009                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2010         }
2011
2012 out:
2013         return r;
2014 }
2015 #endif
2016
2017 static long kvm_vm_ioctl(struct file *filp,
2018                            unsigned int ioctl, unsigned long arg)
2019 {
2020         struct kvm *kvm = filp->private_data;
2021         void __user *argp = (void __user *)arg;
2022         int r;
2023
2024         if (kvm->mm != current->mm)
2025                 return -EIO;
2026         switch (ioctl) {
2027         case KVM_CREATE_VCPU:
2028                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2029                 if (r < 0)
2030                         goto out;
2031                 break;
2032         case KVM_SET_USER_MEMORY_REGION: {
2033                 struct kvm_userspace_memory_region kvm_userspace_mem;
2034
2035                 r = -EFAULT;
2036                 if (copy_from_user(&kvm_userspace_mem, argp,
2037                                                 sizeof kvm_userspace_mem))
2038                         goto out;
2039
2040                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2041                 if (r)
2042                         goto out;
2043                 break;
2044         }
2045         case KVM_GET_DIRTY_LOG: {
2046                 struct kvm_dirty_log log;
2047
2048                 r = -EFAULT;
2049                 if (copy_from_user(&log, argp, sizeof log))
2050                         goto out;
2051                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2052                 if (r)
2053                         goto out;
2054                 break;
2055         }
2056 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2057         case KVM_REGISTER_COALESCED_MMIO: {
2058                 struct kvm_coalesced_mmio_zone zone;
2059                 r = -EFAULT;
2060                 if (copy_from_user(&zone, argp, sizeof zone))
2061                         goto out;
2062                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2063                 if (r)
2064                         goto out;
2065                 r = 0;
2066                 break;
2067         }
2068         case KVM_UNREGISTER_COALESCED_MMIO: {
2069                 struct kvm_coalesced_mmio_zone zone;
2070                 r = -EFAULT;
2071                 if (copy_from_user(&zone, argp, sizeof zone))
2072                         goto out;
2073                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2074                 if (r)
2075                         goto out;
2076                 r = 0;
2077                 break;
2078         }
2079 #endif
2080         case KVM_IRQFD: {
2081                 struct kvm_irqfd data;
2082
2083                 r = -EFAULT;
2084                 if (copy_from_user(&data, argp, sizeof data))
2085                         goto out;
2086                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2087                 break;
2088         }
2089         case KVM_IOEVENTFD: {
2090                 struct kvm_ioeventfd data;
2091
2092                 r = -EFAULT;
2093                 if (copy_from_user(&data, argp, sizeof data))
2094                         goto out;
2095                 r = kvm_ioeventfd(kvm, &data);
2096                 break;
2097         }
2098 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2099         case KVM_SET_BOOT_CPU_ID:
2100                 r = 0;
2101                 mutex_lock(&kvm->lock);
2102                 if (atomic_read(&kvm->online_vcpus) != 0)
2103                         r = -EBUSY;
2104                 else
2105                         kvm->bsp_vcpu_id = arg;
2106                 mutex_unlock(&kvm->lock);
2107                 break;
2108 #endif
2109         default:
2110                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2111                 if (r == -ENOTTY)
2112                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2113         }
2114 out:
2115         return r;
2116 }
2117
2118 #ifdef CONFIG_COMPAT
2119 struct compat_kvm_dirty_log {
2120         __u32 slot;
2121         __u32 padding1;
2122         union {
2123                 compat_uptr_t dirty_bitmap; /* one bit per page */
2124                 __u64 padding2;
2125         };
2126 };
2127
2128 static long kvm_vm_compat_ioctl(struct file *filp,
2129                            unsigned int ioctl, unsigned long arg)
2130 {
2131         struct kvm *kvm = filp->private_data;
2132         int r;
2133
2134         if (kvm->mm != current->mm)
2135                 return -EIO;
2136         switch (ioctl) {
2137         case KVM_GET_DIRTY_LOG: {
2138                 struct compat_kvm_dirty_log compat_log;
2139                 struct kvm_dirty_log log;
2140
2141                 r = -EFAULT;
2142                 if (copy_from_user(&compat_log, (void __user *)arg,
2143                                    sizeof(compat_log)))
2144                         goto out;
2145                 log.slot         = compat_log.slot;
2146                 log.padding1     = compat_log.padding1;
2147                 log.padding2     = compat_log.padding2;
2148                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2149
2150                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2151                 if (r)
2152                         goto out;
2153                 break;
2154         }
2155         default:
2156                 r = kvm_vm_ioctl(filp, ioctl, arg);
2157         }
2158
2159 out:
2160         return r;
2161 }
2162 #endif
2163
2164 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2165 {
2166         struct page *page[1];
2167         unsigned long addr;
2168         int npages;
2169         gfn_t gfn = vmf->pgoff;
2170         struct kvm *kvm = vma->vm_file->private_data;
2171
2172         addr = gfn_to_hva(kvm, gfn);
2173         if (kvm_is_error_hva(addr))
2174                 return VM_FAULT_SIGBUS;
2175
2176         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2177                                 NULL);
2178         if (unlikely(npages != 1))
2179                 return VM_FAULT_SIGBUS;
2180
2181         vmf->page = page[0];
2182         return 0;
2183 }
2184
2185 static const struct vm_operations_struct kvm_vm_vm_ops = {
2186         .fault = kvm_vm_fault,
2187 };
2188
2189 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2190 {
2191         vma->vm_ops = &kvm_vm_vm_ops;
2192         return 0;
2193 }
2194
2195 static struct file_operations kvm_vm_fops = {
2196         .release        = kvm_vm_release,
2197         .unlocked_ioctl = kvm_vm_ioctl,
2198 #ifdef CONFIG_COMPAT
2199         .compat_ioctl   = kvm_vm_compat_ioctl,
2200 #endif
2201         .mmap           = kvm_vm_mmap,
2202         .llseek         = noop_llseek,
2203 };
2204
2205 static int kvm_dev_ioctl_create_vm(void)
2206 {
2207         int r;
2208         struct kvm *kvm;
2209
2210         kvm = kvm_create_vm();
2211         if (IS_ERR(kvm))
2212                 return PTR_ERR(kvm);
2213 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2214         r = kvm_coalesced_mmio_init(kvm);
2215         if (r < 0) {
2216                 kvm_put_kvm(kvm);
2217                 return r;
2218         }
2219 #endif
2220         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2221         if (r < 0)
2222                 kvm_put_kvm(kvm);
2223
2224         return r;
2225 }
2226
2227 static long kvm_dev_ioctl_check_extension_generic(long arg)
2228 {
2229         switch (arg) {
2230         case KVM_CAP_USER_MEMORY:
2231         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2232         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2233 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2234         case KVM_CAP_SET_BOOT_CPU_ID:
2235 #endif
2236         case KVM_CAP_INTERNAL_ERROR_DATA:
2237                 return 1;
2238 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2239         case KVM_CAP_IRQ_ROUTING:
2240                 return KVM_MAX_IRQ_ROUTES;
2241 #endif
2242         default:
2243                 break;
2244         }
2245         return kvm_dev_ioctl_check_extension(arg);
2246 }
2247
2248 static long kvm_dev_ioctl(struct file *filp,
2249                           unsigned int ioctl, unsigned long arg)
2250 {
2251         long r = -EINVAL;
2252
2253         switch (ioctl) {
2254         case KVM_GET_API_VERSION:
2255                 r = -EINVAL;
2256                 if (arg)
2257                         goto out;
2258                 r = KVM_API_VERSION;
2259                 break;
2260         case KVM_CREATE_VM:
2261                 r = -EINVAL;
2262                 if (arg)
2263                         goto out;
2264                 r = kvm_dev_ioctl_create_vm();
2265                 break;
2266         case KVM_CHECK_EXTENSION:
2267                 r = kvm_dev_ioctl_check_extension_generic(arg);
2268                 break;
2269         case KVM_GET_VCPU_MMAP_SIZE:
2270                 r = -EINVAL;
2271                 if (arg)
2272                         goto out;
2273                 r = PAGE_SIZE;     /* struct kvm_run */
2274 #ifdef CONFIG_X86
2275                 r += PAGE_SIZE;    /* pio data page */
2276 #endif
2277 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2278                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2279 #endif
2280                 break;
2281         case KVM_TRACE_ENABLE:
2282         case KVM_TRACE_PAUSE:
2283         case KVM_TRACE_DISABLE:
2284                 r = -EOPNOTSUPP;
2285                 break;
2286         default:
2287                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2288         }
2289 out:
2290         return r;
2291 }
2292
2293 static struct file_operations kvm_chardev_ops = {
2294         .unlocked_ioctl = kvm_dev_ioctl,
2295         .compat_ioctl   = kvm_dev_ioctl,
2296         .llseek         = noop_llseek,
2297 };
2298
2299 static struct miscdevice kvm_dev = {
2300         KVM_MINOR,
2301         "kvm",
2302         &kvm_chardev_ops,
2303 };
2304
2305 static void hardware_enable_nolock(void *junk)
2306 {
2307         int cpu = raw_smp_processor_id();
2308         int r;
2309
2310         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2311                 return;
2312
2313         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2314
2315         r = kvm_arch_hardware_enable(NULL);
2316
2317         if (r) {
2318                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2319                 atomic_inc(&hardware_enable_failed);
2320                 printk(KERN_INFO "kvm: enabling virtualization on "
2321                                  "CPU%d failed\n", cpu);
2322         }
2323 }
2324
2325 static void hardware_enable(void *junk)
2326 {
2327         raw_spin_lock(&kvm_lock);
2328         hardware_enable_nolock(junk);
2329         raw_spin_unlock(&kvm_lock);
2330 }
2331
2332 static void hardware_disable_nolock(void *junk)
2333 {
2334         int cpu = raw_smp_processor_id();
2335
2336         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2337                 return;
2338         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2339         kvm_arch_hardware_disable(NULL);
2340 }
2341
2342 static void hardware_disable(void *junk)
2343 {
2344         raw_spin_lock(&kvm_lock);
2345         hardware_disable_nolock(junk);
2346         raw_spin_unlock(&kvm_lock);
2347 }
2348
2349 static void hardware_disable_all_nolock(void)
2350 {
2351         BUG_ON(!kvm_usage_count);
2352
2353         kvm_usage_count--;
2354         if (!kvm_usage_count)
2355                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2356 }
2357
2358 static void hardware_disable_all(void)
2359 {
2360         raw_spin_lock(&kvm_lock);
2361         hardware_disable_all_nolock();
2362         raw_spin_unlock(&kvm_lock);
2363 }
2364
2365 static int hardware_enable_all(void)
2366 {
2367         int r = 0;
2368
2369         raw_spin_lock(&kvm_lock);
2370
2371         kvm_usage_count++;
2372         if (kvm_usage_count == 1) {
2373                 atomic_set(&hardware_enable_failed, 0);
2374                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2375
2376                 if (atomic_read(&hardware_enable_failed)) {
2377                         hardware_disable_all_nolock();
2378                         r = -EBUSY;
2379                 }
2380         }
2381
2382         raw_spin_unlock(&kvm_lock);
2383
2384         return r;
2385 }
2386
2387 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2388                            void *v)
2389 {
2390         int cpu = (long)v;
2391
2392         if (!kvm_usage_count)
2393                 return NOTIFY_OK;
2394
2395         val &= ~CPU_TASKS_FROZEN;
2396         switch (val) {
2397         case CPU_DYING:
2398                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2399                        cpu);
2400                 hardware_disable(NULL);
2401                 break;
2402         case CPU_STARTING:
2403                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2404                        cpu);
2405                 hardware_enable(NULL);
2406                 break;
2407         }
2408         return NOTIFY_OK;
2409 }
2410
2411
2412 asmlinkage void kvm_spurious_fault(void)
2413 {
2414         /* Fault while not rebooting.  We want the trace. */
2415         BUG();
2416 }
2417 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2418
2419 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2420                       void *v)
2421 {
2422         /*
2423          * Some (well, at least mine) BIOSes hang on reboot if
2424          * in vmx root mode.
2425          *
2426          * And Intel TXT required VMX off for all cpu when system shutdown.
2427          */
2428         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2429         kvm_rebooting = true;
2430         on_each_cpu(hardware_disable_nolock, NULL, 1);
2431         return NOTIFY_OK;
2432 }
2433
2434 static struct notifier_block kvm_reboot_notifier = {
2435         .notifier_call = kvm_reboot,
2436         .priority = 0,
2437 };
2438
2439 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2440 {
2441         int i;
2442
2443         for (i = 0; i < bus->dev_count; i++) {
2444                 struct kvm_io_device *pos = bus->range[i].dev;
2445
2446                 kvm_iodevice_destructor(pos);
2447         }
2448         kfree(bus);
2449 }
2450
2451 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2452 {
2453         const struct kvm_io_range *r1 = p1;
2454         const struct kvm_io_range *r2 = p2;
2455
2456         if (r1->addr < r2->addr)
2457                 return -1;
2458         if (r1->addr + r1->len > r2->addr + r2->len)
2459                 return 1;
2460         return 0;
2461 }
2462
2463 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2464                           gpa_t addr, int len)
2465 {
2466         if (bus->dev_count == NR_IOBUS_DEVS)
2467                 return -ENOSPC;
2468
2469         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2470                 .addr = addr,
2471                 .len = len,
2472                 .dev = dev,
2473         };
2474
2475         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2476                 kvm_io_bus_sort_cmp, NULL);
2477
2478         return 0;
2479 }
2480
2481 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2482                              gpa_t addr, int len)
2483 {
2484         struct kvm_io_range *range, key;
2485         int off;
2486
2487         key = (struct kvm_io_range) {
2488                 .addr = addr,
2489                 .len = len,
2490         };
2491
2492         range = bsearch(&key, bus->range, bus->dev_count,
2493                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2494         if (range == NULL)
2495                 return -ENOENT;
2496
2497         off = range - bus->range;
2498
2499         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2500                 off--;
2501
2502         return off;
2503 }
2504
2505 /* kvm_io_bus_write - called under kvm->slots_lock */
2506 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2507                      int len, const void *val)
2508 {
2509         int idx;
2510         struct kvm_io_bus *bus;
2511         struct kvm_io_range range;
2512
2513         range = (struct kvm_io_range) {
2514                 .addr = addr,
2515                 .len = len,
2516         };
2517
2518         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2519         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2520         if (idx < 0)
2521                 return -EOPNOTSUPP;
2522
2523         while (idx < bus->dev_count &&
2524                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2525                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2526                         return 0;
2527                 idx++;
2528         }
2529
2530         return -EOPNOTSUPP;
2531 }
2532
2533 /* kvm_io_bus_read - called under kvm->slots_lock */
2534 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2535                     int len, void *val)
2536 {
2537         int idx;
2538         struct kvm_io_bus *bus;
2539         struct kvm_io_range range;
2540
2541         range = (struct kvm_io_range) {
2542                 .addr = addr,
2543                 .len = len,
2544         };
2545
2546         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2547         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2548         if (idx < 0)
2549                 return -EOPNOTSUPP;
2550
2551         while (idx < bus->dev_count &&
2552                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2553                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2554                         return 0;
2555                 idx++;
2556         }
2557
2558         return -EOPNOTSUPP;
2559 }
2560
2561 /* Caller must hold slots_lock. */
2562 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2563                             int len, struct kvm_io_device *dev)
2564 {
2565         struct kvm_io_bus *new_bus, *bus;
2566
2567         bus = kvm->buses[bus_idx];
2568         if (bus->dev_count > NR_IOBUS_DEVS-1)
2569                 return -ENOSPC;
2570
2571         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2572         if (!new_bus)
2573                 return -ENOMEM;
2574         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2575         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2576         synchronize_srcu_expedited(&kvm->srcu);
2577         kfree(bus);
2578
2579         return 0;
2580 }
2581
2582 /* Caller must hold slots_lock. */
2583 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2584                               struct kvm_io_device *dev)
2585 {
2586         int i, r;
2587         struct kvm_io_bus *new_bus, *bus;
2588
2589         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2590         if (!new_bus)
2591                 return -ENOMEM;
2592
2593         bus = kvm->buses[bus_idx];
2594         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2595
2596         r = -ENOENT;
2597         for (i = 0; i < new_bus->dev_count; i++)
2598                 if (new_bus->range[i].dev == dev) {
2599                         r = 0;
2600                         new_bus->dev_count--;
2601                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2602                         sort(new_bus->range, new_bus->dev_count,
2603                              sizeof(struct kvm_io_range),
2604                              kvm_io_bus_sort_cmp, NULL);
2605                         break;
2606                 }
2607
2608         if (r) {
2609                 kfree(new_bus);
2610                 return r;
2611         }
2612
2613         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2614         synchronize_srcu_expedited(&kvm->srcu);
2615         kfree(bus);
2616         return r;
2617 }
2618
2619 static struct notifier_block kvm_cpu_notifier = {
2620         .notifier_call = kvm_cpu_hotplug,
2621 };
2622
2623 static int vm_stat_get(void *_offset, u64 *val)
2624 {
2625         unsigned offset = (long)_offset;
2626         struct kvm *kvm;
2627
2628         *val = 0;
2629         raw_spin_lock(&kvm_lock);
2630         list_for_each_entry(kvm, &vm_list, vm_list)
2631                 *val += *(u32 *)((void *)kvm + offset);
2632         raw_spin_unlock(&kvm_lock);
2633         return 0;
2634 }
2635
2636 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2637
2638 static int vcpu_stat_get(void *_offset, u64 *val)
2639 {
2640         unsigned offset = (long)_offset;
2641         struct kvm *kvm;
2642         struct kvm_vcpu *vcpu;
2643         int i;
2644
2645         *val = 0;
2646         raw_spin_lock(&kvm_lock);
2647         list_for_each_entry(kvm, &vm_list, vm_list)
2648                 kvm_for_each_vcpu(i, vcpu, kvm)
2649                         *val += *(u32 *)((void *)vcpu + offset);
2650
2651         raw_spin_unlock(&kvm_lock);
2652         return 0;
2653 }
2654
2655 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2656
2657 static const struct file_operations *stat_fops[] = {
2658         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2659         [KVM_STAT_VM]   = &vm_stat_fops,
2660 };
2661
2662 static void kvm_init_debug(void)
2663 {
2664         struct kvm_stats_debugfs_item *p;
2665
2666         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2667         for (p = debugfs_entries; p->name; ++p)
2668                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2669                                                 (void *)(long)p->offset,
2670                                                 stat_fops[p->kind]);
2671 }
2672
2673 static void kvm_exit_debug(void)
2674 {
2675         struct kvm_stats_debugfs_item *p;
2676
2677         for (p = debugfs_entries; p->name; ++p)
2678                 debugfs_remove(p->dentry);
2679         debugfs_remove(kvm_debugfs_dir);
2680 }
2681
2682 static int kvm_suspend(void)
2683 {
2684         if (kvm_usage_count)
2685                 hardware_disable_nolock(NULL);
2686         return 0;
2687 }
2688
2689 static void kvm_resume(void)
2690 {
2691         if (kvm_usage_count) {
2692                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2693                 hardware_enable_nolock(NULL);
2694         }
2695 }
2696
2697 static struct syscore_ops kvm_syscore_ops = {
2698         .suspend = kvm_suspend,
2699         .resume = kvm_resume,
2700 };
2701
2702 struct page *bad_page;
2703 pfn_t bad_pfn;
2704
2705 static inline
2706 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2707 {
2708         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2709 }
2710
2711 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2712 {
2713         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2714
2715         kvm_arch_vcpu_load(vcpu, cpu);
2716 }
2717
2718 static void kvm_sched_out(struct preempt_notifier *pn,
2719                           struct task_struct *next)
2720 {
2721         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2722
2723         kvm_arch_vcpu_put(vcpu);
2724 }
2725
2726 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2727                   struct module *module)
2728 {
2729         int r;
2730         int cpu;
2731
2732         r = kvm_arch_init(opaque);
2733         if (r)
2734                 goto out_fail;
2735
2736         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2737
2738         if (bad_page == NULL) {
2739                 r = -ENOMEM;
2740                 goto out;
2741         }
2742
2743         bad_pfn = page_to_pfn(bad_page);
2744
2745         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2746
2747         if (hwpoison_page == NULL) {
2748                 r = -ENOMEM;
2749                 goto out_free_0;
2750         }
2751
2752         hwpoison_pfn = page_to_pfn(hwpoison_page);
2753
2754         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2755
2756         if (fault_page == NULL) {
2757                 r = -ENOMEM;
2758                 goto out_free_0;
2759         }
2760
2761         fault_pfn = page_to_pfn(fault_page);
2762
2763         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2764                 r = -ENOMEM;
2765                 goto out_free_0;
2766         }
2767
2768         r = kvm_arch_hardware_setup();
2769         if (r < 0)
2770                 goto out_free_0a;
2771
2772         for_each_online_cpu(cpu) {
2773                 smp_call_function_single(cpu,
2774                                 kvm_arch_check_processor_compat,
2775                                 &r, 1);
2776                 if (r < 0)
2777                         goto out_free_1;
2778         }
2779
2780         r = register_cpu_notifier(&kvm_cpu_notifier);
2781         if (r)
2782                 goto out_free_2;
2783         register_reboot_notifier(&kvm_reboot_notifier);
2784
2785         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2786         if (!vcpu_align)
2787                 vcpu_align = __alignof__(struct kvm_vcpu);
2788         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2789                                            0, NULL);
2790         if (!kvm_vcpu_cache) {
2791                 r = -ENOMEM;
2792                 goto out_free_3;
2793         }
2794
2795         r = kvm_async_pf_init();
2796         if (r)
2797                 goto out_free;
2798
2799         kvm_chardev_ops.owner = module;
2800         kvm_vm_fops.owner = module;
2801         kvm_vcpu_fops.owner = module;
2802
2803         r = misc_register(&kvm_dev);
2804         if (r) {
2805                 printk(KERN_ERR "kvm: misc device register failed\n");
2806                 goto out_unreg;
2807         }
2808
2809         register_syscore_ops(&kvm_syscore_ops);
2810
2811         kvm_preempt_ops.sched_in = kvm_sched_in;
2812         kvm_preempt_ops.sched_out = kvm_sched_out;
2813
2814         kvm_init_debug();
2815
2816         return 0;
2817
2818 out_unreg:
2819         kvm_async_pf_deinit();
2820 out_free:
2821         kmem_cache_destroy(kvm_vcpu_cache);
2822 out_free_3:
2823         unregister_reboot_notifier(&kvm_reboot_notifier);
2824         unregister_cpu_notifier(&kvm_cpu_notifier);
2825 out_free_2:
2826 out_free_1:
2827         kvm_arch_hardware_unsetup();
2828 out_free_0a:
2829         free_cpumask_var(cpus_hardware_enabled);
2830 out_free_0:
2831         if (fault_page)
2832                 __free_page(fault_page);
2833         if (hwpoison_page)
2834                 __free_page(hwpoison_page);
2835         __free_page(bad_page);
2836 out:
2837         kvm_arch_exit();
2838 out_fail:
2839         return r;
2840 }
2841 EXPORT_SYMBOL_GPL(kvm_init);
2842
2843 void kvm_exit(void)
2844 {
2845         kvm_exit_debug();
2846         misc_deregister(&kvm_dev);
2847         kmem_cache_destroy(kvm_vcpu_cache);
2848         kvm_async_pf_deinit();
2849         unregister_syscore_ops(&kvm_syscore_ops);
2850         unregister_reboot_notifier(&kvm_reboot_notifier);
2851         unregister_cpu_notifier(&kvm_cpu_notifier);
2852         on_each_cpu(hardware_disable_nolock, NULL, 1);
2853         kvm_arch_hardware_unsetup();
2854         kvm_arch_exit();
2855         free_cpumask_var(cpus_hardware_enabled);
2856         __free_page(hwpoison_page);
2857         __free_page(bad_page);
2858 }
2859 EXPORT_SYMBOL_GPL(kvm_exit);