470e30520fe877fcec570caab262e61522fe0b7d
[linux-3.10.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 static struct page *hwpoison_page;
104 static pfn_t hwpoison_pfn;
105
106 struct page *fault_page;
107 pfn_t fault_pfn;
108
109 inline int kvm_is_mmio_pfn(pfn_t pfn)
110 {
111         if (pfn_valid(pfn)) {
112                 int reserved;
113                 struct page *tail = pfn_to_page(pfn);
114                 struct page *head = compound_trans_head(tail);
115                 reserved = PageReserved(head);
116                 if (head != tail) {
117                         /*
118                          * "head" is not a dangling pointer
119                          * (compound_trans_head takes care of that)
120                          * but the hugepage may have been splitted
121                          * from under us (and we may not hold a
122                          * reference count on the head page so it can
123                          * be reused before we run PageReferenced), so
124                          * we've to check PageTail before returning
125                          * what we just read.
126                          */
127                         smp_rmb();
128                         if (PageTail(tail))
129                                 return reserved;
130                 }
131                 return PageReserved(tail);
132         }
133
134         return true;
135 }
136
137 /*
138  * Switches to specified vcpu, until a matching vcpu_put()
139  */
140 void vcpu_load(struct kvm_vcpu *vcpu)
141 {
142         int cpu;
143
144         mutex_lock(&vcpu->mutex);
145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
146                 /* The thread running this VCPU changed. */
147                 struct pid *oldpid = vcpu->pid;
148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
149                 rcu_assign_pointer(vcpu->pid, newpid);
150                 synchronize_rcu();
151                 put_pid(oldpid);
152         }
153         cpu = get_cpu();
154         preempt_notifier_register(&vcpu->preempt_notifier);
155         kvm_arch_vcpu_load(vcpu, cpu);
156         put_cpu();
157 }
158
159 void vcpu_put(struct kvm_vcpu *vcpu)
160 {
161         preempt_disable();
162         kvm_arch_vcpu_put(vcpu);
163         preempt_notifier_unregister(&vcpu->preempt_notifier);
164         preempt_enable();
165         mutex_unlock(&vcpu->mutex);
166 }
167
168 static void ack_flush(void *_completed)
169 {
170 }
171
172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 {
174         int i, cpu, me;
175         cpumask_var_t cpus;
176         bool called = true;
177         struct kvm_vcpu *vcpu;
178
179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180
181         me = get_cpu();
182         kvm_for_each_vcpu(i, vcpu, kvm) {
183                 kvm_make_request(req, vcpu);
184                 cpu = vcpu->cpu;
185
186                 /* Set ->requests bit before we read ->mode */
187                 smp_mb();
188
189                 if (cpus != NULL && cpu != -1 && cpu != me &&
190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191                         cpumask_set_cpu(cpu, cpus);
192         }
193         if (unlikely(cpus == NULL))
194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195         else if (!cpumask_empty(cpus))
196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
197         else
198                 called = false;
199         put_cpu();
200         free_cpumask_var(cpus);
201         return called;
202 }
203
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
205 {
206         int dirty_count = kvm->tlbs_dirty;
207
208         smp_mb();
209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
210                 ++kvm->stat.remote_tlb_flush;
211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
212 }
213
214 void kvm_reload_remote_mmus(struct kvm *kvm)
215 {
216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
217 }
218
219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
220 {
221         struct page *page;
222         int r;
223
224         mutex_init(&vcpu->mutex);
225         vcpu->cpu = -1;
226         vcpu->kvm = kvm;
227         vcpu->vcpu_id = id;
228         vcpu->pid = NULL;
229         init_waitqueue_head(&vcpu->wq);
230         kvm_async_pf_vcpu_init(vcpu);
231
232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
233         if (!page) {
234                 r = -ENOMEM;
235                 goto fail;
236         }
237         vcpu->run = page_address(page);
238
239         r = kvm_arch_vcpu_init(vcpu);
240         if (r < 0)
241                 goto fail_free_run;
242         return 0;
243
244 fail_free_run:
245         free_page((unsigned long)vcpu->run);
246 fail:
247         return r;
248 }
249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250
251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 {
253         put_pid(vcpu->pid);
254         kvm_arch_vcpu_uninit(vcpu);
255         free_page((unsigned long)vcpu->run);
256 }
257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258
259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 {
262         return container_of(mn, struct kvm, mmu_notifier);
263 }
264
265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
266                                              struct mm_struct *mm,
267                                              unsigned long address)
268 {
269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
270         int need_tlb_flush, idx;
271
272         /*
273          * When ->invalidate_page runs, the linux pte has been zapped
274          * already but the page is still allocated until
275          * ->invalidate_page returns. So if we increase the sequence
276          * here the kvm page fault will notice if the spte can't be
277          * established because the page is going to be freed. If
278          * instead the kvm page fault establishes the spte before
279          * ->invalidate_page runs, kvm_unmap_hva will release it
280          * before returning.
281          *
282          * The sequence increase only need to be seen at spin_unlock
283          * time, and not at spin_lock time.
284          *
285          * Increasing the sequence after the spin_unlock would be
286          * unsafe because the kvm page fault could then establish the
287          * pte after kvm_unmap_hva returned, without noticing the page
288          * is going to be freed.
289          */
290         idx = srcu_read_lock(&kvm->srcu);
291         spin_lock(&kvm->mmu_lock);
292         kvm->mmu_notifier_seq++;
293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294         spin_unlock(&kvm->mmu_lock);
295         srcu_read_unlock(&kvm->srcu, idx);
296
297         /* we've to flush the tlb before the pages can be freed */
298         if (need_tlb_flush)
299                 kvm_flush_remote_tlbs(kvm);
300
301 }
302
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304                                         struct mm_struct *mm,
305                                         unsigned long address,
306                                         pte_t pte)
307 {
308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
309         int idx;
310
311         idx = srcu_read_lock(&kvm->srcu);
312         spin_lock(&kvm->mmu_lock);
313         kvm->mmu_notifier_seq++;
314         kvm_set_spte_hva(kvm, address, pte);
315         spin_unlock(&kvm->mmu_lock);
316         srcu_read_unlock(&kvm->srcu, idx);
317 }
318
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320                                                     struct mm_struct *mm,
321                                                     unsigned long start,
322                                                     unsigned long end)
323 {
324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
325         int need_tlb_flush = 0, idx;
326
327         idx = srcu_read_lock(&kvm->srcu);
328         spin_lock(&kvm->mmu_lock);
329         /*
330          * The count increase must become visible at unlock time as no
331          * spte can be established without taking the mmu_lock and
332          * count is also read inside the mmu_lock critical section.
333          */
334         kvm->mmu_notifier_count++;
335         for (; start < end; start += PAGE_SIZE)
336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
337         need_tlb_flush |= kvm->tlbs_dirty;
338         spin_unlock(&kvm->mmu_lock);
339         srcu_read_unlock(&kvm->srcu, idx);
340
341         /* we've to flush the tlb before the pages can be freed */
342         if (need_tlb_flush)
343                 kvm_flush_remote_tlbs(kvm);
344 }
345
346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
347                                                   struct mm_struct *mm,
348                                                   unsigned long start,
349                                                   unsigned long end)
350 {
351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
352
353         spin_lock(&kvm->mmu_lock);
354         /*
355          * This sequence increase will notify the kvm page fault that
356          * the page that is going to be mapped in the spte could have
357          * been freed.
358          */
359         kvm->mmu_notifier_seq++;
360         smp_wmb();
361         /*
362          * The above sequence increase must be visible before the
363          * below count decrease, which is ensured by the smp_wmb above
364          * in conjunction with the smp_rmb in mmu_notifier_retry().
365          */
366         kvm->mmu_notifier_count--;
367         spin_unlock(&kvm->mmu_lock);
368
369         BUG_ON(kvm->mmu_notifier_count < 0);
370 }
371
372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
373                                               struct mm_struct *mm,
374                                               unsigned long address)
375 {
376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
377         int young, idx;
378
379         idx = srcu_read_lock(&kvm->srcu);
380         spin_lock(&kvm->mmu_lock);
381         young = kvm_age_hva(kvm, address);
382         spin_unlock(&kvm->mmu_lock);
383         srcu_read_unlock(&kvm->srcu, idx);
384
385         if (young)
386                 kvm_flush_remote_tlbs(kvm);
387
388         return young;
389 }
390
391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392                                        struct mm_struct *mm,
393                                        unsigned long address)
394 {
395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
396         int young, idx;
397
398         idx = srcu_read_lock(&kvm->srcu);
399         spin_lock(&kvm->mmu_lock);
400         young = kvm_test_age_hva(kvm, address);
401         spin_unlock(&kvm->mmu_lock);
402         srcu_read_unlock(&kvm->srcu, idx);
403
404         return young;
405 }
406
407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408                                      struct mm_struct *mm)
409 {
410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
411         int idx;
412
413         idx = srcu_read_lock(&kvm->srcu);
414         kvm_arch_flush_shadow(kvm);
415         srcu_read_unlock(&kvm->srcu, idx);
416 }
417
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
423         .test_young             = kvm_mmu_notifier_test_young,
424         .change_pte             = kvm_mmu_notifier_change_pte,
425         .release                = kvm_mmu_notifier_release,
426 };
427
428 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 {
430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 }
433
434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435
436 static int kvm_init_mmu_notifier(struct kvm *kvm)
437 {
438         return 0;
439 }
440
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442
443 static void kvm_init_memslots_id(struct kvm *kvm)
444 {
445         int i;
446         struct kvm_memslots *slots = kvm->memslots;
447
448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449                 slots->id_to_index[i] = slots->memslots[i].id = i;
450 }
451
452 static struct kvm *kvm_create_vm(unsigned long type)
453 {
454         int r, i;
455         struct kvm *kvm = kvm_arch_alloc_vm();
456
457         if (!kvm)
458                 return ERR_PTR(-ENOMEM);
459
460         r = kvm_arch_init_vm(kvm, type);
461         if (r)
462                 goto out_err_nodisable;
463
464         r = hardware_enable_all();
465         if (r)
466                 goto out_err_nodisable;
467
468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
469         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 #endif
472
473         r = -ENOMEM;
474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
475         if (!kvm->memslots)
476                 goto out_err_nosrcu;
477         kvm_init_memslots_id(kvm);
478         if (init_srcu_struct(&kvm->srcu))
479                 goto out_err_nosrcu;
480         for (i = 0; i < KVM_NR_BUSES; i++) {
481                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
482                                         GFP_KERNEL);
483                 if (!kvm->buses[i])
484                         goto out_err;
485         }
486
487         spin_lock_init(&kvm->mmu_lock);
488         kvm->mm = current->mm;
489         atomic_inc(&kvm->mm->mm_count);
490         kvm_eventfd_init(kvm);
491         mutex_init(&kvm->lock);
492         mutex_init(&kvm->irq_lock);
493         mutex_init(&kvm->slots_lock);
494         atomic_set(&kvm->users_count, 1);
495
496         r = kvm_init_mmu_notifier(kvm);
497         if (r)
498                 goto out_err;
499
500         raw_spin_lock(&kvm_lock);
501         list_add(&kvm->vm_list, &vm_list);
502         raw_spin_unlock(&kvm_lock);
503
504         return kvm;
505
506 out_err:
507         cleanup_srcu_struct(&kvm->srcu);
508 out_err_nosrcu:
509         hardware_disable_all();
510 out_err_nodisable:
511         for (i = 0; i < KVM_NR_BUSES; i++)
512                 kfree(kvm->buses[i]);
513         kfree(kvm->memslots);
514         kvm_arch_free_vm(kvm);
515         return ERR_PTR(r);
516 }
517
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
519 {
520         if (!memslot->dirty_bitmap)
521                 return;
522
523         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
524                 vfree(memslot->dirty_bitmap_head);
525         else
526                 kfree(memslot->dirty_bitmap_head);
527
528         memslot->dirty_bitmap = NULL;
529         memslot->dirty_bitmap_head = NULL;
530 }
531
532 /*
533  * Free any memory in @free but not in @dont.
534  */
535 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
536                                   struct kvm_memory_slot *dont)
537 {
538         int i;
539
540         if (!dont || free->rmap != dont->rmap)
541                 vfree(free->rmap);
542
543         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
544                 kvm_destroy_dirty_bitmap(free);
545
546
547         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
548                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
549                         vfree(free->lpage_info[i]);
550                         free->lpage_info[i] = NULL;
551                 }
552         }
553
554         free->npages = 0;
555         free->rmap = NULL;
556 }
557
558 void kvm_free_physmem(struct kvm *kvm)
559 {
560         struct kvm_memslots *slots = kvm->memslots;
561         struct kvm_memory_slot *memslot;
562
563         kvm_for_each_memslot(memslot, slots)
564                 kvm_free_physmem_slot(memslot, NULL);
565
566         kfree(kvm->memslots);
567 }
568
569 static void kvm_destroy_vm(struct kvm *kvm)
570 {
571         int i;
572         struct mm_struct *mm = kvm->mm;
573
574         kvm_arch_sync_events(kvm);
575         raw_spin_lock(&kvm_lock);
576         list_del(&kvm->vm_list);
577         raw_spin_unlock(&kvm_lock);
578         kvm_free_irq_routing(kvm);
579         for (i = 0; i < KVM_NR_BUSES; i++)
580                 kvm_io_bus_destroy(kvm->buses[i]);
581         kvm_coalesced_mmio_free(kvm);
582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
584 #else
585         kvm_arch_flush_shadow(kvm);
586 #endif
587         kvm_arch_destroy_vm(kvm);
588         kvm_free_physmem(kvm);
589         cleanup_srcu_struct(&kvm->srcu);
590         kvm_arch_free_vm(kvm);
591         hardware_disable_all();
592         mmdrop(mm);
593 }
594
595 void kvm_get_kvm(struct kvm *kvm)
596 {
597         atomic_inc(&kvm->users_count);
598 }
599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
600
601 void kvm_put_kvm(struct kvm *kvm)
602 {
603         if (atomic_dec_and_test(&kvm->users_count))
604                 kvm_destroy_vm(kvm);
605 }
606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
607
608
609 static int kvm_vm_release(struct inode *inode, struct file *filp)
610 {
611         struct kvm *kvm = filp->private_data;
612
613         kvm_irqfd_release(kvm);
614
615         kvm_put_kvm(kvm);
616         return 0;
617 }
618
619 #ifndef CONFIG_S390
620 /*
621  * Allocation size is twice as large as the actual dirty bitmap size.
622  * This makes it possible to do double buffering: see x86's
623  * kvm_vm_ioctl_get_dirty_log().
624  */
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626 {
627         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
628
629         if (dirty_bytes > PAGE_SIZE)
630                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
631         else
632                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
633
634         if (!memslot->dirty_bitmap)
635                 return -ENOMEM;
636
637         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
638         memslot->nr_dirty_pages = 0;
639         return 0;
640 }
641 #endif /* !CONFIG_S390 */
642
643 static int cmp_memslot(const void *slot1, const void *slot2)
644 {
645         struct kvm_memory_slot *s1, *s2;
646
647         s1 = (struct kvm_memory_slot *)slot1;
648         s2 = (struct kvm_memory_slot *)slot2;
649
650         if (s1->npages < s2->npages)
651                 return 1;
652         if (s1->npages > s2->npages)
653                 return -1;
654
655         return 0;
656 }
657
658 /*
659  * Sort the memslots base on its size, so the larger slots
660  * will get better fit.
661  */
662 static void sort_memslots(struct kvm_memslots *slots)
663 {
664         int i;
665
666         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
667               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
668
669         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
670                 slots->id_to_index[slots->memslots[i].id] = i;
671 }
672
673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
674 {
675         if (new) {
676                 int id = new->id;
677                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
678                 unsigned long npages = old->npages;
679
680                 *old = *new;
681                 if (new->npages != npages)
682                         sort_memslots(slots);
683         }
684
685         slots->generation++;
686 }
687
688 /*
689  * Allocate some memory and give it an address in the guest physical address
690  * space.
691  *
692  * Discontiguous memory is allowed, mostly for framebuffers.
693  *
694  * Must be called holding mmap_sem for write.
695  */
696 int __kvm_set_memory_region(struct kvm *kvm,
697                             struct kvm_userspace_memory_region *mem,
698                             int user_alloc)
699 {
700         int r;
701         gfn_t base_gfn;
702         unsigned long npages;
703         unsigned long i;
704         struct kvm_memory_slot *memslot;
705         struct kvm_memory_slot old, new;
706         struct kvm_memslots *slots, *old_memslots;
707
708         r = -EINVAL;
709         /* General sanity checks */
710         if (mem->memory_size & (PAGE_SIZE - 1))
711                 goto out;
712         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
713                 goto out;
714         /* We can read the guest memory with __xxx_user() later on. */
715         if (user_alloc &&
716             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
717              !access_ok(VERIFY_WRITE,
718                         (void __user *)(unsigned long)mem->userspace_addr,
719                         mem->memory_size)))
720                 goto out;
721         if (mem->slot >= KVM_MEM_SLOTS_NUM)
722                 goto out;
723         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
724                 goto out;
725
726         memslot = id_to_memslot(kvm->memslots, mem->slot);
727         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
728         npages = mem->memory_size >> PAGE_SHIFT;
729
730         r = -EINVAL;
731         if (npages > KVM_MEM_MAX_NR_PAGES)
732                 goto out;
733
734         if (!npages)
735                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
736
737         new = old = *memslot;
738
739         new.id = mem->slot;
740         new.base_gfn = base_gfn;
741         new.npages = npages;
742         new.flags = mem->flags;
743
744         /* Disallow changing a memory slot's size. */
745         r = -EINVAL;
746         if (npages && old.npages && npages != old.npages)
747                 goto out_free;
748
749         /* Check for overlaps */
750         r = -EEXIST;
751         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
752                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
753
754                 if (s == memslot || !s->npages)
755                         continue;
756                 if (!((base_gfn + npages <= s->base_gfn) ||
757                       (base_gfn >= s->base_gfn + s->npages)))
758                         goto out_free;
759         }
760
761         /* Free page dirty bitmap if unneeded */
762         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
763                 new.dirty_bitmap = NULL;
764
765         r = -ENOMEM;
766
767         /* Allocate if a slot is being created */
768 #ifndef CONFIG_S390
769         if (npages && !new.rmap) {
770                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
771
772                 if (!new.rmap)
773                         goto out_free;
774
775                 new.user_alloc = user_alloc;
776                 new.userspace_addr = mem->userspace_addr;
777         }
778         if (!npages)
779                 goto skip_lpage;
780
781         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
782                 unsigned long ugfn;
783                 unsigned long j;
784                 int lpages;
785                 int level = i + 2;
786
787                 /* Avoid unused variable warning if no large pages */
788                 (void)level;
789
790                 if (new.lpage_info[i])
791                         continue;
792
793                 lpages = 1 + ((base_gfn + npages - 1)
794                              >> KVM_HPAGE_GFN_SHIFT(level));
795                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
796
797                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
798
799                 if (!new.lpage_info[i])
800                         goto out_free;
801
802                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
803                         new.lpage_info[i][0].write_count = 1;
804                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
805                         new.lpage_info[i][lpages - 1].write_count = 1;
806                 ugfn = new.userspace_addr >> PAGE_SHIFT;
807                 /*
808                  * If the gfn and userspace address are not aligned wrt each
809                  * other, or if explicitly asked to, disable large page
810                  * support for this slot
811                  */
812                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
813                     !largepages_enabled)
814                         for (j = 0; j < lpages; ++j)
815                                 new.lpage_info[i][j].write_count = 1;
816         }
817
818 skip_lpage:
819
820         /* Allocate page dirty bitmap if needed */
821         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
822                 if (kvm_create_dirty_bitmap(&new) < 0)
823                         goto out_free;
824                 /* destroy any largepage mappings for dirty tracking */
825         }
826 #else  /* not defined CONFIG_S390 */
827         new.user_alloc = user_alloc;
828         if (user_alloc)
829                 new.userspace_addr = mem->userspace_addr;
830 #endif /* not defined CONFIG_S390 */
831
832         if (!npages) {
833                 struct kvm_memory_slot *slot;
834
835                 r = -ENOMEM;
836                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
837                                 GFP_KERNEL);
838                 if (!slots)
839                         goto out_free;
840                 slot = id_to_memslot(slots, mem->slot);
841                 slot->flags |= KVM_MEMSLOT_INVALID;
842
843                 update_memslots(slots, NULL);
844
845                 old_memslots = kvm->memslots;
846                 rcu_assign_pointer(kvm->memslots, slots);
847                 synchronize_srcu_expedited(&kvm->srcu);
848                 /* From this point no new shadow pages pointing to a deleted
849                  * memslot will be created.
850                  *
851                  * validation of sp->gfn happens in:
852                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
853                  *      - kvm_is_visible_gfn (mmu_check_roots)
854                  */
855                 kvm_arch_flush_shadow(kvm);
856                 kfree(old_memslots);
857         }
858
859         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
860         if (r)
861                 goto out_free;
862
863         /* map the pages in iommu page table */
864         if (npages) {
865                 r = kvm_iommu_map_pages(kvm, &new);
866                 if (r)
867                         goto out_free;
868         }
869
870         r = -ENOMEM;
871         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
872                         GFP_KERNEL);
873         if (!slots)
874                 goto out_free;
875
876         /* actual memory is freed via old in kvm_free_physmem_slot below */
877         if (!npages) {
878                 new.rmap = NULL;
879                 new.dirty_bitmap = NULL;
880                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
881                         new.lpage_info[i] = NULL;
882         }
883
884         update_memslots(slots, &new);
885         old_memslots = kvm->memslots;
886         rcu_assign_pointer(kvm->memslots, slots);
887         synchronize_srcu_expedited(&kvm->srcu);
888
889         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
890
891         /*
892          * If the new memory slot is created, we need to clear all
893          * mmio sptes.
894          */
895         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
896                 kvm_arch_flush_shadow(kvm);
897
898         kvm_free_physmem_slot(&old, &new);
899         kfree(old_memslots);
900
901         return 0;
902
903 out_free:
904         kvm_free_physmem_slot(&new, &old);
905 out:
906         return r;
907
908 }
909 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
910
911 int kvm_set_memory_region(struct kvm *kvm,
912                           struct kvm_userspace_memory_region *mem,
913                           int user_alloc)
914 {
915         int r;
916
917         mutex_lock(&kvm->slots_lock);
918         r = __kvm_set_memory_region(kvm, mem, user_alloc);
919         mutex_unlock(&kvm->slots_lock);
920         return r;
921 }
922 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
923
924 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
925                                    struct
926                                    kvm_userspace_memory_region *mem,
927                                    int user_alloc)
928 {
929         if (mem->slot >= KVM_MEMORY_SLOTS)
930                 return -EINVAL;
931         return kvm_set_memory_region(kvm, mem, user_alloc);
932 }
933
934 int kvm_get_dirty_log(struct kvm *kvm,
935                         struct kvm_dirty_log *log, int *is_dirty)
936 {
937         struct kvm_memory_slot *memslot;
938         int r, i;
939         unsigned long n;
940         unsigned long any = 0;
941
942         r = -EINVAL;
943         if (log->slot >= KVM_MEMORY_SLOTS)
944                 goto out;
945
946         memslot = id_to_memslot(kvm->memslots, log->slot);
947         r = -ENOENT;
948         if (!memslot->dirty_bitmap)
949                 goto out;
950
951         n = kvm_dirty_bitmap_bytes(memslot);
952
953         for (i = 0; !any && i < n/sizeof(long); ++i)
954                 any = memslot->dirty_bitmap[i];
955
956         r = -EFAULT;
957         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
958                 goto out;
959
960         if (any)
961                 *is_dirty = 1;
962
963         r = 0;
964 out:
965         return r;
966 }
967
968 void kvm_disable_largepages(void)
969 {
970         largepages_enabled = false;
971 }
972 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
973
974 int is_error_page(struct page *page)
975 {
976         return page == bad_page || page == hwpoison_page || page == fault_page;
977 }
978 EXPORT_SYMBOL_GPL(is_error_page);
979
980 int is_error_pfn(pfn_t pfn)
981 {
982         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
983 }
984 EXPORT_SYMBOL_GPL(is_error_pfn);
985
986 int is_hwpoison_pfn(pfn_t pfn)
987 {
988         return pfn == hwpoison_pfn;
989 }
990 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
991
992 int is_fault_pfn(pfn_t pfn)
993 {
994         return pfn == fault_pfn;
995 }
996 EXPORT_SYMBOL_GPL(is_fault_pfn);
997
998 int is_noslot_pfn(pfn_t pfn)
999 {
1000         return pfn == bad_pfn;
1001 }
1002 EXPORT_SYMBOL_GPL(is_noslot_pfn);
1003
1004 int is_invalid_pfn(pfn_t pfn)
1005 {
1006         return pfn == hwpoison_pfn || pfn == fault_pfn;
1007 }
1008 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1009
1010 static inline unsigned long bad_hva(void)
1011 {
1012         return PAGE_OFFSET;
1013 }
1014
1015 int kvm_is_error_hva(unsigned long addr)
1016 {
1017         return addr == bad_hva();
1018 }
1019 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1020
1021 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1022 {
1023         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1024 }
1025 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1026
1027 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1028 {
1029         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1030
1031         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1032               memslot->flags & KVM_MEMSLOT_INVALID)
1033                 return 0;
1034
1035         return 1;
1036 }
1037 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1038
1039 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1040 {
1041         struct vm_area_struct *vma;
1042         unsigned long addr, size;
1043
1044         size = PAGE_SIZE;
1045
1046         addr = gfn_to_hva(kvm, gfn);
1047         if (kvm_is_error_hva(addr))
1048                 return PAGE_SIZE;
1049
1050         down_read(&current->mm->mmap_sem);
1051         vma = find_vma(current->mm, addr);
1052         if (!vma)
1053                 goto out;
1054
1055         size = vma_kernel_pagesize(vma);
1056
1057 out:
1058         up_read(&current->mm->mmap_sem);
1059
1060         return size;
1061 }
1062
1063 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1064                                      gfn_t *nr_pages)
1065 {
1066         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1067                 return bad_hva();
1068
1069         if (nr_pages)
1070                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1071
1072         return gfn_to_hva_memslot(slot, gfn);
1073 }
1074
1075 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1076 {
1077         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1078 }
1079 EXPORT_SYMBOL_GPL(gfn_to_hva);
1080
1081 static pfn_t get_fault_pfn(void)
1082 {
1083         get_page(fault_page);
1084         return fault_pfn;
1085 }
1086
1087 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1088         unsigned long start, int write, struct page **page)
1089 {
1090         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1091
1092         if (write)
1093                 flags |= FOLL_WRITE;
1094
1095         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1096 }
1097
1098 static inline int check_user_page_hwpoison(unsigned long addr)
1099 {
1100         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1101
1102         rc = __get_user_pages(current, current->mm, addr, 1,
1103                               flags, NULL, NULL, NULL);
1104         return rc == -EHWPOISON;
1105 }
1106
1107 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1108                         bool *async, bool write_fault, bool *writable)
1109 {
1110         struct page *page[1];
1111         int npages = 0;
1112         pfn_t pfn;
1113
1114         /* we can do it either atomically or asynchronously, not both */
1115         BUG_ON(atomic && async);
1116
1117         BUG_ON(!write_fault && !writable);
1118
1119         if (writable)
1120                 *writable = true;
1121
1122         if (atomic || async)
1123                 npages = __get_user_pages_fast(addr, 1, 1, page);
1124
1125         if (unlikely(npages != 1) && !atomic) {
1126                 might_sleep();
1127
1128                 if (writable)
1129                         *writable = write_fault;
1130
1131                 if (async) {
1132                         down_read(&current->mm->mmap_sem);
1133                         npages = get_user_page_nowait(current, current->mm,
1134                                                      addr, write_fault, page);
1135                         up_read(&current->mm->mmap_sem);
1136                 } else
1137                         npages = get_user_pages_fast(addr, 1, write_fault,
1138                                                      page);
1139
1140                 /* map read fault as writable if possible */
1141                 if (unlikely(!write_fault) && npages == 1) {
1142                         struct page *wpage[1];
1143
1144                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1145                         if (npages == 1) {
1146                                 *writable = true;
1147                                 put_page(page[0]);
1148                                 page[0] = wpage[0];
1149                         }
1150                         npages = 1;
1151                 }
1152         }
1153
1154         if (unlikely(npages != 1)) {
1155                 struct vm_area_struct *vma;
1156
1157                 if (atomic)
1158                         return get_fault_pfn();
1159
1160                 down_read(&current->mm->mmap_sem);
1161                 if (npages == -EHWPOISON ||
1162                         (!async && check_user_page_hwpoison(addr))) {
1163                         up_read(&current->mm->mmap_sem);
1164                         get_page(hwpoison_page);
1165                         return page_to_pfn(hwpoison_page);
1166                 }
1167
1168                 vma = find_vma_intersection(current->mm, addr, addr+1);
1169
1170                 if (vma == NULL)
1171                         pfn = get_fault_pfn();
1172                 else if ((vma->vm_flags & VM_PFNMAP)) {
1173                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1174                                 vma->vm_pgoff;
1175                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1176                 } else {
1177                         if (async && (vma->vm_flags & VM_WRITE))
1178                                 *async = true;
1179                         pfn = get_fault_pfn();
1180                 }
1181                 up_read(&current->mm->mmap_sem);
1182         } else
1183                 pfn = page_to_pfn(page[0]);
1184
1185         return pfn;
1186 }
1187
1188 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1189 {
1190         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1191 }
1192 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1193
1194 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1195                           bool write_fault, bool *writable)
1196 {
1197         unsigned long addr;
1198
1199         if (async)
1200                 *async = false;
1201
1202         addr = gfn_to_hva(kvm, gfn);
1203         if (kvm_is_error_hva(addr)) {
1204                 get_page(bad_page);
1205                 return page_to_pfn(bad_page);
1206         }
1207
1208         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1209 }
1210
1211 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1212 {
1213         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1214 }
1215 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1216
1217 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1218                        bool write_fault, bool *writable)
1219 {
1220         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1221 }
1222 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1223
1224 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1225 {
1226         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1227 }
1228 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1229
1230 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1231                       bool *writable)
1232 {
1233         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1234 }
1235 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1236
1237 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1238                          struct kvm_memory_slot *slot, gfn_t gfn)
1239 {
1240         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1241         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1242 }
1243
1244 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1245                                                                   int nr_pages)
1246 {
1247         unsigned long addr;
1248         gfn_t entry;
1249
1250         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1251         if (kvm_is_error_hva(addr))
1252                 return -1;
1253
1254         if (entry < nr_pages)
1255                 return 0;
1256
1257         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1258 }
1259 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1260
1261 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1262 {
1263         pfn_t pfn;
1264
1265         pfn = gfn_to_pfn(kvm, gfn);
1266         if (!kvm_is_mmio_pfn(pfn))
1267                 return pfn_to_page(pfn);
1268
1269         WARN_ON(kvm_is_mmio_pfn(pfn));
1270
1271         get_page(bad_page);
1272         return bad_page;
1273 }
1274
1275 EXPORT_SYMBOL_GPL(gfn_to_page);
1276
1277 void kvm_release_page_clean(struct page *page)
1278 {
1279         kvm_release_pfn_clean(page_to_pfn(page));
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1282
1283 void kvm_release_pfn_clean(pfn_t pfn)
1284 {
1285         if (!kvm_is_mmio_pfn(pfn))
1286                 put_page(pfn_to_page(pfn));
1287 }
1288 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1289
1290 void kvm_release_page_dirty(struct page *page)
1291 {
1292         kvm_release_pfn_dirty(page_to_pfn(page));
1293 }
1294 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1295
1296 void kvm_release_pfn_dirty(pfn_t pfn)
1297 {
1298         kvm_set_pfn_dirty(pfn);
1299         kvm_release_pfn_clean(pfn);
1300 }
1301 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1302
1303 void kvm_set_page_dirty(struct page *page)
1304 {
1305         kvm_set_pfn_dirty(page_to_pfn(page));
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1308
1309 void kvm_set_pfn_dirty(pfn_t pfn)
1310 {
1311         if (!kvm_is_mmio_pfn(pfn)) {
1312                 struct page *page = pfn_to_page(pfn);
1313                 if (!PageReserved(page))
1314                         SetPageDirty(page);
1315         }
1316 }
1317 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1318
1319 void kvm_set_pfn_accessed(pfn_t pfn)
1320 {
1321         if (!kvm_is_mmio_pfn(pfn))
1322                 mark_page_accessed(pfn_to_page(pfn));
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1325
1326 void kvm_get_pfn(pfn_t pfn)
1327 {
1328         if (!kvm_is_mmio_pfn(pfn))
1329                 get_page(pfn_to_page(pfn));
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1332
1333 static int next_segment(unsigned long len, int offset)
1334 {
1335         if (len > PAGE_SIZE - offset)
1336                 return PAGE_SIZE - offset;
1337         else
1338                 return len;
1339 }
1340
1341 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1342                         int len)
1343 {
1344         int r;
1345         unsigned long addr;
1346
1347         addr = gfn_to_hva(kvm, gfn);
1348         if (kvm_is_error_hva(addr))
1349                 return -EFAULT;
1350         r = __copy_from_user(data, (void __user *)addr + offset, len);
1351         if (r)
1352                 return -EFAULT;
1353         return 0;
1354 }
1355 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1356
1357 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1358 {
1359         gfn_t gfn = gpa >> PAGE_SHIFT;
1360         int seg;
1361         int offset = offset_in_page(gpa);
1362         int ret;
1363
1364         while ((seg = next_segment(len, offset)) != 0) {
1365                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1366                 if (ret < 0)
1367                         return ret;
1368                 offset = 0;
1369                 len -= seg;
1370                 data += seg;
1371                 ++gfn;
1372         }
1373         return 0;
1374 }
1375 EXPORT_SYMBOL_GPL(kvm_read_guest);
1376
1377 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1378                           unsigned long len)
1379 {
1380         int r;
1381         unsigned long addr;
1382         gfn_t gfn = gpa >> PAGE_SHIFT;
1383         int offset = offset_in_page(gpa);
1384
1385         addr = gfn_to_hva(kvm, gfn);
1386         if (kvm_is_error_hva(addr))
1387                 return -EFAULT;
1388         pagefault_disable();
1389         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1390         pagefault_enable();
1391         if (r)
1392                 return -EFAULT;
1393         return 0;
1394 }
1395 EXPORT_SYMBOL(kvm_read_guest_atomic);
1396
1397 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1398                          int offset, int len)
1399 {
1400         int r;
1401         unsigned long addr;
1402
1403         addr = gfn_to_hva(kvm, gfn);
1404         if (kvm_is_error_hva(addr))
1405                 return -EFAULT;
1406         r = __copy_to_user((void __user *)addr + offset, data, len);
1407         if (r)
1408                 return -EFAULT;
1409         mark_page_dirty(kvm, gfn);
1410         return 0;
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1413
1414 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1415                     unsigned long len)
1416 {
1417         gfn_t gfn = gpa >> PAGE_SHIFT;
1418         int seg;
1419         int offset = offset_in_page(gpa);
1420         int ret;
1421
1422         while ((seg = next_segment(len, offset)) != 0) {
1423                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1424                 if (ret < 0)
1425                         return ret;
1426                 offset = 0;
1427                 len -= seg;
1428                 data += seg;
1429                 ++gfn;
1430         }
1431         return 0;
1432 }
1433
1434 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1435                               gpa_t gpa)
1436 {
1437         struct kvm_memslots *slots = kvm_memslots(kvm);
1438         int offset = offset_in_page(gpa);
1439         gfn_t gfn = gpa >> PAGE_SHIFT;
1440
1441         ghc->gpa = gpa;
1442         ghc->generation = slots->generation;
1443         ghc->memslot = gfn_to_memslot(kvm, gfn);
1444         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1445         if (!kvm_is_error_hva(ghc->hva))
1446                 ghc->hva += offset;
1447         else
1448                 return -EFAULT;
1449
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1453
1454 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1455                            void *data, unsigned long len)
1456 {
1457         struct kvm_memslots *slots = kvm_memslots(kvm);
1458         int r;
1459
1460         if (slots->generation != ghc->generation)
1461                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1462
1463         if (kvm_is_error_hva(ghc->hva))
1464                 return -EFAULT;
1465
1466         r = __copy_to_user((void __user *)ghc->hva, data, len);
1467         if (r)
1468                 return -EFAULT;
1469         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1470
1471         return 0;
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1474
1475 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1476                            void *data, unsigned long len)
1477 {
1478         struct kvm_memslots *slots = kvm_memslots(kvm);
1479         int r;
1480
1481         if (slots->generation != ghc->generation)
1482                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1483
1484         if (kvm_is_error_hva(ghc->hva))
1485                 return -EFAULT;
1486
1487         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1488         if (r)
1489                 return -EFAULT;
1490
1491         return 0;
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1494
1495 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1496 {
1497         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1498                                     offset, len);
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1501
1502 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1503 {
1504         gfn_t gfn = gpa >> PAGE_SHIFT;
1505         int seg;
1506         int offset = offset_in_page(gpa);
1507         int ret;
1508
1509         while ((seg = next_segment(len, offset)) != 0) {
1510                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1511                 if (ret < 0)
1512                         return ret;
1513                 offset = 0;
1514                 len -= seg;
1515                 ++gfn;
1516         }
1517         return 0;
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1520
1521 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1522                              gfn_t gfn)
1523 {
1524         if (memslot && memslot->dirty_bitmap) {
1525                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1526
1527                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1528                         memslot->nr_dirty_pages++;
1529         }
1530 }
1531
1532 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1533 {
1534         struct kvm_memory_slot *memslot;
1535
1536         memslot = gfn_to_memslot(kvm, gfn);
1537         mark_page_dirty_in_slot(kvm, memslot, gfn);
1538 }
1539
1540 /*
1541  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1542  */
1543 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1544 {
1545         DEFINE_WAIT(wait);
1546
1547         for (;;) {
1548                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1549
1550                 if (kvm_arch_vcpu_runnable(vcpu)) {
1551                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1552                         break;
1553                 }
1554                 if (kvm_cpu_has_pending_timer(vcpu))
1555                         break;
1556                 if (signal_pending(current))
1557                         break;
1558
1559                 schedule();
1560         }
1561
1562         finish_wait(&vcpu->wq, &wait);
1563 }
1564
1565 void kvm_resched(struct kvm_vcpu *vcpu)
1566 {
1567         if (!need_resched())
1568                 return;
1569         cond_resched();
1570 }
1571 EXPORT_SYMBOL_GPL(kvm_resched);
1572
1573 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1574 {
1575         struct kvm *kvm = me->kvm;
1576         struct kvm_vcpu *vcpu;
1577         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1578         int yielded = 0;
1579         int pass;
1580         int i;
1581
1582         /*
1583          * We boost the priority of a VCPU that is runnable but not
1584          * currently running, because it got preempted by something
1585          * else and called schedule in __vcpu_run.  Hopefully that
1586          * VCPU is holding the lock that we need and will release it.
1587          * We approximate round-robin by starting at the last boosted VCPU.
1588          */
1589         for (pass = 0; pass < 2 && !yielded; pass++) {
1590                 kvm_for_each_vcpu(i, vcpu, kvm) {
1591                         struct task_struct *task = NULL;
1592                         struct pid *pid;
1593                         if (!pass && i < last_boosted_vcpu) {
1594                                 i = last_boosted_vcpu;
1595                                 continue;
1596                         } else if (pass && i > last_boosted_vcpu)
1597                                 break;
1598                         if (vcpu == me)
1599                                 continue;
1600                         if (waitqueue_active(&vcpu->wq))
1601                                 continue;
1602                         rcu_read_lock();
1603                         pid = rcu_dereference(vcpu->pid);
1604                         if (pid)
1605                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1606                         rcu_read_unlock();
1607                         if (!task)
1608                                 continue;
1609                         if (task->flags & PF_VCPU) {
1610                                 put_task_struct(task);
1611                                 continue;
1612                         }
1613                         if (yield_to(task, 1)) {
1614                                 put_task_struct(task);
1615                                 kvm->last_boosted_vcpu = i;
1616                                 yielded = 1;
1617                                 break;
1618                         }
1619                         put_task_struct(task);
1620                 }
1621         }
1622 }
1623 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1624
1625 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1626 {
1627         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1628         struct page *page;
1629
1630         if (vmf->pgoff == 0)
1631                 page = virt_to_page(vcpu->run);
1632 #ifdef CONFIG_X86
1633         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1634                 page = virt_to_page(vcpu->arch.pio_data);
1635 #endif
1636 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1637         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1638                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1639 #endif
1640         else
1641                 return kvm_arch_vcpu_fault(vcpu, vmf);
1642         get_page(page);
1643         vmf->page = page;
1644         return 0;
1645 }
1646
1647 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1648         .fault = kvm_vcpu_fault,
1649 };
1650
1651 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1652 {
1653         vma->vm_ops = &kvm_vcpu_vm_ops;
1654         return 0;
1655 }
1656
1657 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1658 {
1659         struct kvm_vcpu *vcpu = filp->private_data;
1660
1661         kvm_put_kvm(vcpu->kvm);
1662         return 0;
1663 }
1664
1665 static struct file_operations kvm_vcpu_fops = {
1666         .release        = kvm_vcpu_release,
1667         .unlocked_ioctl = kvm_vcpu_ioctl,
1668 #ifdef CONFIG_COMPAT
1669         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1670 #endif
1671         .mmap           = kvm_vcpu_mmap,
1672         .llseek         = noop_llseek,
1673 };
1674
1675 /*
1676  * Allocates an inode for the vcpu.
1677  */
1678 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1679 {
1680         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1681 }
1682
1683 /*
1684  * Creates some virtual cpus.  Good luck creating more than one.
1685  */
1686 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1687 {
1688         int r;
1689         struct kvm_vcpu *vcpu, *v;
1690
1691         vcpu = kvm_arch_vcpu_create(kvm, id);
1692         if (IS_ERR(vcpu))
1693                 return PTR_ERR(vcpu);
1694
1695         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1696
1697         r = kvm_arch_vcpu_setup(vcpu);
1698         if (r)
1699                 goto vcpu_destroy;
1700
1701         mutex_lock(&kvm->lock);
1702         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1703                 r = -EINVAL;
1704                 goto unlock_vcpu_destroy;
1705         }
1706
1707         kvm_for_each_vcpu(r, v, kvm)
1708                 if (v->vcpu_id == id) {
1709                         r = -EEXIST;
1710                         goto unlock_vcpu_destroy;
1711                 }
1712
1713         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1714
1715         /* Now it's all set up, let userspace reach it */
1716         kvm_get_kvm(kvm);
1717         r = create_vcpu_fd(vcpu);
1718         if (r < 0) {
1719                 kvm_put_kvm(kvm);
1720                 goto unlock_vcpu_destroy;
1721         }
1722
1723         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1724         smp_wmb();
1725         atomic_inc(&kvm->online_vcpus);
1726
1727         mutex_unlock(&kvm->lock);
1728         return r;
1729
1730 unlock_vcpu_destroy:
1731         mutex_unlock(&kvm->lock);
1732 vcpu_destroy:
1733         kvm_arch_vcpu_destroy(vcpu);
1734         return r;
1735 }
1736
1737 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1738 {
1739         if (sigset) {
1740                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1741                 vcpu->sigset_active = 1;
1742                 vcpu->sigset = *sigset;
1743         } else
1744                 vcpu->sigset_active = 0;
1745         return 0;
1746 }
1747
1748 static long kvm_vcpu_ioctl(struct file *filp,
1749                            unsigned int ioctl, unsigned long arg)
1750 {
1751         struct kvm_vcpu *vcpu = filp->private_data;
1752         void __user *argp = (void __user *)arg;
1753         int r;
1754         struct kvm_fpu *fpu = NULL;
1755         struct kvm_sregs *kvm_sregs = NULL;
1756
1757         if (vcpu->kvm->mm != current->mm)
1758                 return -EIO;
1759
1760 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1761         /*
1762          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1763          * so vcpu_load() would break it.
1764          */
1765         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1766                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1767 #endif
1768
1769
1770         vcpu_load(vcpu);
1771         switch (ioctl) {
1772         case KVM_RUN:
1773                 r = -EINVAL;
1774                 if (arg)
1775                         goto out;
1776                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1777                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1778                 break;
1779         case KVM_GET_REGS: {
1780                 struct kvm_regs *kvm_regs;
1781
1782                 r = -ENOMEM;
1783                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1784                 if (!kvm_regs)
1785                         goto out;
1786                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1787                 if (r)
1788                         goto out_free1;
1789                 r = -EFAULT;
1790                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1791                         goto out_free1;
1792                 r = 0;
1793 out_free1:
1794                 kfree(kvm_regs);
1795                 break;
1796         }
1797         case KVM_SET_REGS: {
1798                 struct kvm_regs *kvm_regs;
1799
1800                 r = -ENOMEM;
1801                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1802                 if (IS_ERR(kvm_regs)) {
1803                         r = PTR_ERR(kvm_regs);
1804                         goto out;
1805                 }
1806                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1807                 if (r)
1808                         goto out_free2;
1809                 r = 0;
1810 out_free2:
1811                 kfree(kvm_regs);
1812                 break;
1813         }
1814         case KVM_GET_SREGS: {
1815                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1816                 r = -ENOMEM;
1817                 if (!kvm_sregs)
1818                         goto out;
1819                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1820                 if (r)
1821                         goto out;
1822                 r = -EFAULT;
1823                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1824                         goto out;
1825                 r = 0;
1826                 break;
1827         }
1828         case KVM_SET_SREGS: {
1829                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1830                 if (IS_ERR(kvm_sregs)) {
1831                         r = PTR_ERR(kvm_sregs);
1832                         goto out;
1833                 }
1834                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1835                 if (r)
1836                         goto out;
1837                 r = 0;
1838                 break;
1839         }
1840         case KVM_GET_MP_STATE: {
1841                 struct kvm_mp_state mp_state;
1842
1843                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1844                 if (r)
1845                         goto out;
1846                 r = -EFAULT;
1847                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1848                         goto out;
1849                 r = 0;
1850                 break;
1851         }
1852         case KVM_SET_MP_STATE: {
1853                 struct kvm_mp_state mp_state;
1854
1855                 r = -EFAULT;
1856                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1857                         goto out;
1858                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1859                 if (r)
1860                         goto out;
1861                 r = 0;
1862                 break;
1863         }
1864         case KVM_TRANSLATE: {
1865                 struct kvm_translation tr;
1866
1867                 r = -EFAULT;
1868                 if (copy_from_user(&tr, argp, sizeof tr))
1869                         goto out;
1870                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1871                 if (r)
1872                         goto out;
1873                 r = -EFAULT;
1874                 if (copy_to_user(argp, &tr, sizeof tr))
1875                         goto out;
1876                 r = 0;
1877                 break;
1878         }
1879         case KVM_SET_GUEST_DEBUG: {
1880                 struct kvm_guest_debug dbg;
1881
1882                 r = -EFAULT;
1883                 if (copy_from_user(&dbg, argp, sizeof dbg))
1884                         goto out;
1885                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1886                 if (r)
1887                         goto out;
1888                 r = 0;
1889                 break;
1890         }
1891         case KVM_SET_SIGNAL_MASK: {
1892                 struct kvm_signal_mask __user *sigmask_arg = argp;
1893                 struct kvm_signal_mask kvm_sigmask;
1894                 sigset_t sigset, *p;
1895
1896                 p = NULL;
1897                 if (argp) {
1898                         r = -EFAULT;
1899                         if (copy_from_user(&kvm_sigmask, argp,
1900                                            sizeof kvm_sigmask))
1901                                 goto out;
1902                         r = -EINVAL;
1903                         if (kvm_sigmask.len != sizeof sigset)
1904                                 goto out;
1905                         r = -EFAULT;
1906                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1907                                            sizeof sigset))
1908                                 goto out;
1909                         p = &sigset;
1910                 }
1911                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1912                 break;
1913         }
1914         case KVM_GET_FPU: {
1915                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1916                 r = -ENOMEM;
1917                 if (!fpu)
1918                         goto out;
1919                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1920                 if (r)
1921                         goto out;
1922                 r = -EFAULT;
1923                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1924                         goto out;
1925                 r = 0;
1926                 break;
1927         }
1928         case KVM_SET_FPU: {
1929                 fpu = memdup_user(argp, sizeof(*fpu));
1930                 if (IS_ERR(fpu)) {
1931                         r = PTR_ERR(fpu);
1932                         goto out;
1933                 }
1934                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1935                 if (r)
1936                         goto out;
1937                 r = 0;
1938                 break;
1939         }
1940         default:
1941                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1942         }
1943 out:
1944         vcpu_put(vcpu);
1945         kfree(fpu);
1946         kfree(kvm_sregs);
1947         return r;
1948 }
1949
1950 #ifdef CONFIG_COMPAT
1951 static long kvm_vcpu_compat_ioctl(struct file *filp,
1952                                   unsigned int ioctl, unsigned long arg)
1953 {
1954         struct kvm_vcpu *vcpu = filp->private_data;
1955         void __user *argp = compat_ptr(arg);
1956         int r;
1957
1958         if (vcpu->kvm->mm != current->mm)
1959                 return -EIO;
1960
1961         switch (ioctl) {
1962         case KVM_SET_SIGNAL_MASK: {
1963                 struct kvm_signal_mask __user *sigmask_arg = argp;
1964                 struct kvm_signal_mask kvm_sigmask;
1965                 compat_sigset_t csigset;
1966                 sigset_t sigset;
1967
1968                 if (argp) {
1969                         r = -EFAULT;
1970                         if (copy_from_user(&kvm_sigmask, argp,
1971                                            sizeof kvm_sigmask))
1972                                 goto out;
1973                         r = -EINVAL;
1974                         if (kvm_sigmask.len != sizeof csigset)
1975                                 goto out;
1976                         r = -EFAULT;
1977                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1978                                            sizeof csigset))
1979                                 goto out;
1980                 }
1981                 sigset_from_compat(&sigset, &csigset);
1982                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1983                 break;
1984         }
1985         default:
1986                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1987         }
1988
1989 out:
1990         return r;
1991 }
1992 #endif
1993
1994 static long kvm_vm_ioctl(struct file *filp,
1995                            unsigned int ioctl, unsigned long arg)
1996 {
1997         struct kvm *kvm = filp->private_data;
1998         void __user *argp = (void __user *)arg;
1999         int r;
2000
2001         if (kvm->mm != current->mm)
2002                 return -EIO;
2003         switch (ioctl) {
2004         case KVM_CREATE_VCPU:
2005                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2006                 if (r < 0)
2007                         goto out;
2008                 break;
2009         case KVM_SET_USER_MEMORY_REGION: {
2010                 struct kvm_userspace_memory_region kvm_userspace_mem;
2011
2012                 r = -EFAULT;
2013                 if (copy_from_user(&kvm_userspace_mem, argp,
2014                                                 sizeof kvm_userspace_mem))
2015                         goto out;
2016
2017                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2018                 if (r)
2019                         goto out;
2020                 break;
2021         }
2022         case KVM_GET_DIRTY_LOG: {
2023                 struct kvm_dirty_log log;
2024
2025                 r = -EFAULT;
2026                 if (copy_from_user(&log, argp, sizeof log))
2027                         goto out;
2028                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2029                 if (r)
2030                         goto out;
2031                 break;
2032         }
2033 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2034         case KVM_REGISTER_COALESCED_MMIO: {
2035                 struct kvm_coalesced_mmio_zone zone;
2036                 r = -EFAULT;
2037                 if (copy_from_user(&zone, argp, sizeof zone))
2038                         goto out;
2039                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2040                 if (r)
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         case KVM_UNREGISTER_COALESCED_MMIO: {
2046                 struct kvm_coalesced_mmio_zone zone;
2047                 r = -EFAULT;
2048                 if (copy_from_user(&zone, argp, sizeof zone))
2049                         goto out;
2050                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2051                 if (r)
2052                         goto out;
2053                 r = 0;
2054                 break;
2055         }
2056 #endif
2057         case KVM_IRQFD: {
2058                 struct kvm_irqfd data;
2059
2060                 r = -EFAULT;
2061                 if (copy_from_user(&data, argp, sizeof data))
2062                         goto out;
2063                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2064                 break;
2065         }
2066         case KVM_IOEVENTFD: {
2067                 struct kvm_ioeventfd data;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&data, argp, sizeof data))
2071                         goto out;
2072                 r = kvm_ioeventfd(kvm, &data);
2073                 break;
2074         }
2075 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2076         case KVM_SET_BOOT_CPU_ID:
2077                 r = 0;
2078                 mutex_lock(&kvm->lock);
2079                 if (atomic_read(&kvm->online_vcpus) != 0)
2080                         r = -EBUSY;
2081                 else
2082                         kvm->bsp_vcpu_id = arg;
2083                 mutex_unlock(&kvm->lock);
2084                 break;
2085 #endif
2086         default:
2087                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2088                 if (r == -ENOTTY)
2089                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2090         }
2091 out:
2092         return r;
2093 }
2094
2095 #ifdef CONFIG_COMPAT
2096 struct compat_kvm_dirty_log {
2097         __u32 slot;
2098         __u32 padding1;
2099         union {
2100                 compat_uptr_t dirty_bitmap; /* one bit per page */
2101                 __u64 padding2;
2102         };
2103 };
2104
2105 static long kvm_vm_compat_ioctl(struct file *filp,
2106                            unsigned int ioctl, unsigned long arg)
2107 {
2108         struct kvm *kvm = filp->private_data;
2109         int r;
2110
2111         if (kvm->mm != current->mm)
2112                 return -EIO;
2113         switch (ioctl) {
2114         case KVM_GET_DIRTY_LOG: {
2115                 struct compat_kvm_dirty_log compat_log;
2116                 struct kvm_dirty_log log;
2117
2118                 r = -EFAULT;
2119                 if (copy_from_user(&compat_log, (void __user *)arg,
2120                                    sizeof(compat_log)))
2121                         goto out;
2122                 log.slot         = compat_log.slot;
2123                 log.padding1     = compat_log.padding1;
2124                 log.padding2     = compat_log.padding2;
2125                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2126
2127                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2128                 if (r)
2129                         goto out;
2130                 break;
2131         }
2132         default:
2133                 r = kvm_vm_ioctl(filp, ioctl, arg);
2134         }
2135
2136 out:
2137         return r;
2138 }
2139 #endif
2140
2141 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2142 {
2143         struct page *page[1];
2144         unsigned long addr;
2145         int npages;
2146         gfn_t gfn = vmf->pgoff;
2147         struct kvm *kvm = vma->vm_file->private_data;
2148
2149         addr = gfn_to_hva(kvm, gfn);
2150         if (kvm_is_error_hva(addr))
2151                 return VM_FAULT_SIGBUS;
2152
2153         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2154                                 NULL);
2155         if (unlikely(npages != 1))
2156                 return VM_FAULT_SIGBUS;
2157
2158         vmf->page = page[0];
2159         return 0;
2160 }
2161
2162 static const struct vm_operations_struct kvm_vm_vm_ops = {
2163         .fault = kvm_vm_fault,
2164 };
2165
2166 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2167 {
2168         vma->vm_ops = &kvm_vm_vm_ops;
2169         return 0;
2170 }
2171
2172 static struct file_operations kvm_vm_fops = {
2173         .release        = kvm_vm_release,
2174         .unlocked_ioctl = kvm_vm_ioctl,
2175 #ifdef CONFIG_COMPAT
2176         .compat_ioctl   = kvm_vm_compat_ioctl,
2177 #endif
2178         .mmap           = kvm_vm_mmap,
2179         .llseek         = noop_llseek,
2180 };
2181
2182 static int kvm_dev_ioctl_create_vm(unsigned long type)
2183 {
2184         int r;
2185         struct kvm *kvm;
2186
2187         kvm = kvm_create_vm(type);
2188         if (IS_ERR(kvm))
2189                 return PTR_ERR(kvm);
2190 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2191         r = kvm_coalesced_mmio_init(kvm);
2192         if (r < 0) {
2193                 kvm_put_kvm(kvm);
2194                 return r;
2195         }
2196 #endif
2197         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2198         if (r < 0)
2199                 kvm_put_kvm(kvm);
2200
2201         return r;
2202 }
2203
2204 static long kvm_dev_ioctl_check_extension_generic(long arg)
2205 {
2206         switch (arg) {
2207         case KVM_CAP_USER_MEMORY:
2208         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2209         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2210 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2211         case KVM_CAP_SET_BOOT_CPU_ID:
2212 #endif
2213         case KVM_CAP_INTERNAL_ERROR_DATA:
2214                 return 1;
2215 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2216         case KVM_CAP_IRQ_ROUTING:
2217                 return KVM_MAX_IRQ_ROUTES;
2218 #endif
2219         default:
2220                 break;
2221         }
2222         return kvm_dev_ioctl_check_extension(arg);
2223 }
2224
2225 static long kvm_dev_ioctl(struct file *filp,
2226                           unsigned int ioctl, unsigned long arg)
2227 {
2228         long r = -EINVAL;
2229
2230         switch (ioctl) {
2231         case KVM_GET_API_VERSION:
2232                 r = -EINVAL;
2233                 if (arg)
2234                         goto out;
2235                 r = KVM_API_VERSION;
2236                 break;
2237         case KVM_CREATE_VM:
2238                 r = kvm_dev_ioctl_create_vm(arg);
2239                 break;
2240         case KVM_CHECK_EXTENSION:
2241                 r = kvm_dev_ioctl_check_extension_generic(arg);
2242                 break;
2243         case KVM_GET_VCPU_MMAP_SIZE:
2244                 r = -EINVAL;
2245                 if (arg)
2246                         goto out;
2247                 r = PAGE_SIZE;     /* struct kvm_run */
2248 #ifdef CONFIG_X86
2249                 r += PAGE_SIZE;    /* pio data page */
2250 #endif
2251 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2252                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2253 #endif
2254                 break;
2255         case KVM_TRACE_ENABLE:
2256         case KVM_TRACE_PAUSE:
2257         case KVM_TRACE_DISABLE:
2258                 r = -EOPNOTSUPP;
2259                 break;
2260         default:
2261                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2262         }
2263 out:
2264         return r;
2265 }
2266
2267 static struct file_operations kvm_chardev_ops = {
2268         .unlocked_ioctl = kvm_dev_ioctl,
2269         .compat_ioctl   = kvm_dev_ioctl,
2270         .llseek         = noop_llseek,
2271 };
2272
2273 static struct miscdevice kvm_dev = {
2274         KVM_MINOR,
2275         "kvm",
2276         &kvm_chardev_ops,
2277 };
2278
2279 static void hardware_enable_nolock(void *junk)
2280 {
2281         int cpu = raw_smp_processor_id();
2282         int r;
2283
2284         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2285                 return;
2286
2287         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2288
2289         r = kvm_arch_hardware_enable(NULL);
2290
2291         if (r) {
2292                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2293                 atomic_inc(&hardware_enable_failed);
2294                 printk(KERN_INFO "kvm: enabling virtualization on "
2295                                  "CPU%d failed\n", cpu);
2296         }
2297 }
2298
2299 static void hardware_enable(void *junk)
2300 {
2301         raw_spin_lock(&kvm_lock);
2302         hardware_enable_nolock(junk);
2303         raw_spin_unlock(&kvm_lock);
2304 }
2305
2306 static void hardware_disable_nolock(void *junk)
2307 {
2308         int cpu = raw_smp_processor_id();
2309
2310         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2311                 return;
2312         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2313         kvm_arch_hardware_disable(NULL);
2314 }
2315
2316 static void hardware_disable(void *junk)
2317 {
2318         raw_spin_lock(&kvm_lock);
2319         hardware_disable_nolock(junk);
2320         raw_spin_unlock(&kvm_lock);
2321 }
2322
2323 static void hardware_disable_all_nolock(void)
2324 {
2325         BUG_ON(!kvm_usage_count);
2326
2327         kvm_usage_count--;
2328         if (!kvm_usage_count)
2329                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2330 }
2331
2332 static void hardware_disable_all(void)
2333 {
2334         raw_spin_lock(&kvm_lock);
2335         hardware_disable_all_nolock();
2336         raw_spin_unlock(&kvm_lock);
2337 }
2338
2339 static int hardware_enable_all(void)
2340 {
2341         int r = 0;
2342
2343         raw_spin_lock(&kvm_lock);
2344
2345         kvm_usage_count++;
2346         if (kvm_usage_count == 1) {
2347                 atomic_set(&hardware_enable_failed, 0);
2348                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2349
2350                 if (atomic_read(&hardware_enable_failed)) {
2351                         hardware_disable_all_nolock();
2352                         r = -EBUSY;
2353                 }
2354         }
2355
2356         raw_spin_unlock(&kvm_lock);
2357
2358         return r;
2359 }
2360
2361 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2362                            void *v)
2363 {
2364         int cpu = (long)v;
2365
2366         if (!kvm_usage_count)
2367                 return NOTIFY_OK;
2368
2369         val &= ~CPU_TASKS_FROZEN;
2370         switch (val) {
2371         case CPU_DYING:
2372                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2373                        cpu);
2374                 hardware_disable(NULL);
2375                 break;
2376         case CPU_STARTING:
2377                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2378                        cpu);
2379                 hardware_enable(NULL);
2380                 break;
2381         }
2382         return NOTIFY_OK;
2383 }
2384
2385
2386 asmlinkage void kvm_spurious_fault(void)
2387 {
2388         /* Fault while not rebooting.  We want the trace. */
2389         BUG();
2390 }
2391 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2392
2393 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2394                       void *v)
2395 {
2396         /*
2397          * Some (well, at least mine) BIOSes hang on reboot if
2398          * in vmx root mode.
2399          *
2400          * And Intel TXT required VMX off for all cpu when system shutdown.
2401          */
2402         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2403         kvm_rebooting = true;
2404         on_each_cpu(hardware_disable_nolock, NULL, 1);
2405         return NOTIFY_OK;
2406 }
2407
2408 static struct notifier_block kvm_reboot_notifier = {
2409         .notifier_call = kvm_reboot,
2410         .priority = 0,
2411 };
2412
2413 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2414 {
2415         int i;
2416
2417         for (i = 0; i < bus->dev_count; i++) {
2418                 struct kvm_io_device *pos = bus->range[i].dev;
2419
2420                 kvm_iodevice_destructor(pos);
2421         }
2422         kfree(bus);
2423 }
2424
2425 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2426 {
2427         const struct kvm_io_range *r1 = p1;
2428         const struct kvm_io_range *r2 = p2;
2429
2430         if (r1->addr < r2->addr)
2431                 return -1;
2432         if (r1->addr + r1->len > r2->addr + r2->len)
2433                 return 1;
2434         return 0;
2435 }
2436
2437 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2438                           gpa_t addr, int len)
2439 {
2440         if (bus->dev_count == NR_IOBUS_DEVS)
2441                 return -ENOSPC;
2442
2443         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2444                 .addr = addr,
2445                 .len = len,
2446                 .dev = dev,
2447         };
2448
2449         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2450                 kvm_io_bus_sort_cmp, NULL);
2451
2452         return 0;
2453 }
2454
2455 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2456                              gpa_t addr, int len)
2457 {
2458         struct kvm_io_range *range, key;
2459         int off;
2460
2461         key = (struct kvm_io_range) {
2462                 .addr = addr,
2463                 .len = len,
2464         };
2465
2466         range = bsearch(&key, bus->range, bus->dev_count,
2467                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2468         if (range == NULL)
2469                 return -ENOENT;
2470
2471         off = range - bus->range;
2472
2473         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2474                 off--;
2475
2476         return off;
2477 }
2478
2479 /* kvm_io_bus_write - called under kvm->slots_lock */
2480 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2481                      int len, const void *val)
2482 {
2483         int idx;
2484         struct kvm_io_bus *bus;
2485         struct kvm_io_range range;
2486
2487         range = (struct kvm_io_range) {
2488                 .addr = addr,
2489                 .len = len,
2490         };
2491
2492         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2493         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2494         if (idx < 0)
2495                 return -EOPNOTSUPP;
2496
2497         while (idx < bus->dev_count &&
2498                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2499                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2500                         return 0;
2501                 idx++;
2502         }
2503
2504         return -EOPNOTSUPP;
2505 }
2506
2507 /* kvm_io_bus_read - called under kvm->slots_lock */
2508 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2509                     int len, void *val)
2510 {
2511         int idx;
2512         struct kvm_io_bus *bus;
2513         struct kvm_io_range range;
2514
2515         range = (struct kvm_io_range) {
2516                 .addr = addr,
2517                 .len = len,
2518         };
2519
2520         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2521         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2522         if (idx < 0)
2523                 return -EOPNOTSUPP;
2524
2525         while (idx < bus->dev_count &&
2526                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2527                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2528                         return 0;
2529                 idx++;
2530         }
2531
2532         return -EOPNOTSUPP;
2533 }
2534
2535 /* Caller must hold slots_lock. */
2536 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2537                             int len, struct kvm_io_device *dev)
2538 {
2539         struct kvm_io_bus *new_bus, *bus;
2540
2541         bus = kvm->buses[bus_idx];
2542         if (bus->dev_count > NR_IOBUS_DEVS-1)
2543                 return -ENOSPC;
2544
2545         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2546         if (!new_bus)
2547                 return -ENOMEM;
2548         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2549         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2550         synchronize_srcu_expedited(&kvm->srcu);
2551         kfree(bus);
2552
2553         return 0;
2554 }
2555
2556 /* Caller must hold slots_lock. */
2557 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2558                               struct kvm_io_device *dev)
2559 {
2560         int i, r;
2561         struct kvm_io_bus *new_bus, *bus;
2562
2563         bus = kvm->buses[bus_idx];
2564
2565         new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2566         if (!new_bus)
2567                 return -ENOMEM;
2568
2569         r = -ENOENT;
2570         for (i = 0; i < new_bus->dev_count; i++)
2571                 if (new_bus->range[i].dev == dev) {
2572                         r = 0;
2573                         new_bus->dev_count--;
2574                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2575                         sort(new_bus->range, new_bus->dev_count,
2576                              sizeof(struct kvm_io_range),
2577                              kvm_io_bus_sort_cmp, NULL);
2578                         break;
2579                 }
2580
2581         if (r) {
2582                 kfree(new_bus);
2583                 return r;
2584         }
2585
2586         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2587         synchronize_srcu_expedited(&kvm->srcu);
2588         kfree(bus);
2589         return r;
2590 }
2591
2592 static struct notifier_block kvm_cpu_notifier = {
2593         .notifier_call = kvm_cpu_hotplug,
2594 };
2595
2596 static int vm_stat_get(void *_offset, u64 *val)
2597 {
2598         unsigned offset = (long)_offset;
2599         struct kvm *kvm;
2600
2601         *val = 0;
2602         raw_spin_lock(&kvm_lock);
2603         list_for_each_entry(kvm, &vm_list, vm_list)
2604                 *val += *(u32 *)((void *)kvm + offset);
2605         raw_spin_unlock(&kvm_lock);
2606         return 0;
2607 }
2608
2609 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2610
2611 static int vcpu_stat_get(void *_offset, u64 *val)
2612 {
2613         unsigned offset = (long)_offset;
2614         struct kvm *kvm;
2615         struct kvm_vcpu *vcpu;
2616         int i;
2617
2618         *val = 0;
2619         raw_spin_lock(&kvm_lock);
2620         list_for_each_entry(kvm, &vm_list, vm_list)
2621                 kvm_for_each_vcpu(i, vcpu, kvm)
2622                         *val += *(u32 *)((void *)vcpu + offset);
2623
2624         raw_spin_unlock(&kvm_lock);
2625         return 0;
2626 }
2627
2628 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2629
2630 static const struct file_operations *stat_fops[] = {
2631         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2632         [KVM_STAT_VM]   = &vm_stat_fops,
2633 };
2634
2635 static int kvm_init_debug(void)
2636 {
2637         int r = -EFAULT;
2638         struct kvm_stats_debugfs_item *p;
2639
2640         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2641         if (kvm_debugfs_dir == NULL)
2642                 goto out;
2643
2644         for (p = debugfs_entries; p->name; ++p) {
2645                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2646                                                 (void *)(long)p->offset,
2647                                                 stat_fops[p->kind]);
2648                 if (p->dentry == NULL)
2649                         goto out_dir;
2650         }
2651
2652         return 0;
2653
2654 out_dir:
2655         debugfs_remove_recursive(kvm_debugfs_dir);
2656 out:
2657         return r;
2658 }
2659
2660 static void kvm_exit_debug(void)
2661 {
2662         struct kvm_stats_debugfs_item *p;
2663
2664         for (p = debugfs_entries; p->name; ++p)
2665                 debugfs_remove(p->dentry);
2666         debugfs_remove(kvm_debugfs_dir);
2667 }
2668
2669 static int kvm_suspend(void)
2670 {
2671         if (kvm_usage_count)
2672                 hardware_disable_nolock(NULL);
2673         return 0;
2674 }
2675
2676 static void kvm_resume(void)
2677 {
2678         if (kvm_usage_count) {
2679                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2680                 hardware_enable_nolock(NULL);
2681         }
2682 }
2683
2684 static struct syscore_ops kvm_syscore_ops = {
2685         .suspend = kvm_suspend,
2686         .resume = kvm_resume,
2687 };
2688
2689 struct page *bad_page;
2690 pfn_t bad_pfn;
2691
2692 static inline
2693 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2694 {
2695         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2696 }
2697
2698 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2699 {
2700         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2701
2702         kvm_arch_vcpu_load(vcpu, cpu);
2703 }
2704
2705 static void kvm_sched_out(struct preempt_notifier *pn,
2706                           struct task_struct *next)
2707 {
2708         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2709
2710         kvm_arch_vcpu_put(vcpu);
2711 }
2712
2713 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2714                   struct module *module)
2715 {
2716         int r;
2717         int cpu;
2718
2719         r = kvm_arch_init(opaque);
2720         if (r)
2721                 goto out_fail;
2722
2723         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2724
2725         if (bad_page == NULL) {
2726                 r = -ENOMEM;
2727                 goto out;
2728         }
2729
2730         bad_pfn = page_to_pfn(bad_page);
2731
2732         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2733
2734         if (hwpoison_page == NULL) {
2735                 r = -ENOMEM;
2736                 goto out_free_0;
2737         }
2738
2739         hwpoison_pfn = page_to_pfn(hwpoison_page);
2740
2741         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2742
2743         if (fault_page == NULL) {
2744                 r = -ENOMEM;
2745                 goto out_free_0;
2746         }
2747
2748         fault_pfn = page_to_pfn(fault_page);
2749
2750         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2751                 r = -ENOMEM;
2752                 goto out_free_0;
2753         }
2754
2755         r = kvm_arch_hardware_setup();
2756         if (r < 0)
2757                 goto out_free_0a;
2758
2759         for_each_online_cpu(cpu) {
2760                 smp_call_function_single(cpu,
2761                                 kvm_arch_check_processor_compat,
2762                                 &r, 1);
2763                 if (r < 0)
2764                         goto out_free_1;
2765         }
2766
2767         r = register_cpu_notifier(&kvm_cpu_notifier);
2768         if (r)
2769                 goto out_free_2;
2770         register_reboot_notifier(&kvm_reboot_notifier);
2771
2772         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2773         if (!vcpu_align)
2774                 vcpu_align = __alignof__(struct kvm_vcpu);
2775         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2776                                            0, NULL);
2777         if (!kvm_vcpu_cache) {
2778                 r = -ENOMEM;
2779                 goto out_free_3;
2780         }
2781
2782         r = kvm_async_pf_init();
2783         if (r)
2784                 goto out_free;
2785
2786         kvm_chardev_ops.owner = module;
2787         kvm_vm_fops.owner = module;
2788         kvm_vcpu_fops.owner = module;
2789
2790         r = misc_register(&kvm_dev);
2791         if (r) {
2792                 printk(KERN_ERR "kvm: misc device register failed\n");
2793                 goto out_unreg;
2794         }
2795
2796         register_syscore_ops(&kvm_syscore_ops);
2797
2798         kvm_preempt_ops.sched_in = kvm_sched_in;
2799         kvm_preempt_ops.sched_out = kvm_sched_out;
2800
2801         r = kvm_init_debug();
2802         if (r) {
2803                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2804                 goto out_undebugfs;
2805         }
2806
2807         return 0;
2808
2809 out_undebugfs:
2810         unregister_syscore_ops(&kvm_syscore_ops);
2811 out_unreg:
2812         kvm_async_pf_deinit();
2813 out_free:
2814         kmem_cache_destroy(kvm_vcpu_cache);
2815 out_free_3:
2816         unregister_reboot_notifier(&kvm_reboot_notifier);
2817         unregister_cpu_notifier(&kvm_cpu_notifier);
2818 out_free_2:
2819 out_free_1:
2820         kvm_arch_hardware_unsetup();
2821 out_free_0a:
2822         free_cpumask_var(cpus_hardware_enabled);
2823 out_free_0:
2824         if (fault_page)
2825                 __free_page(fault_page);
2826         if (hwpoison_page)
2827                 __free_page(hwpoison_page);
2828         __free_page(bad_page);
2829 out:
2830         kvm_arch_exit();
2831 out_fail:
2832         return r;
2833 }
2834 EXPORT_SYMBOL_GPL(kvm_init);
2835
2836 void kvm_exit(void)
2837 {
2838         kvm_exit_debug();
2839         misc_deregister(&kvm_dev);
2840         kmem_cache_destroy(kvm_vcpu_cache);
2841         kvm_async_pf_deinit();
2842         unregister_syscore_ops(&kvm_syscore_ops);
2843         unregister_reboot_notifier(&kvm_reboot_notifier);
2844         unregister_cpu_notifier(&kvm_cpu_notifier);
2845         on_each_cpu(hardware_disable_nolock, NULL, 1);
2846         kvm_arch_hardware_unsetup();
2847         kvm_arch_exit();
2848         free_cpumask_var(cpus_hardware_enabled);
2849         __free_page(hwpoison_page);
2850         __free_page(bad_page);
2851 }
2852 EXPORT_SYMBOL_GPL(kvm_exit);