74f731920945c72375450865c448cdbabb1a829e
[linux-2.6.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affilates.
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/sysdev.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
51 #include <asm/processor.h>
52 #include <asm/io.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
56
57 #include "coalesced_mmio.h"
58
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
61
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
64
65 /*
66  * Ordering of locks:
67  *
68  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
69  */
70
71 DEFINE_SPINLOCK(kvm_lock);
72 LIST_HEAD(vm_list);
73
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
77
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
80
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
82
83 struct dentry *kvm_debugfs_dir;
84
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
86                            unsigned long arg);
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
89
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
91
92 static bool kvm_rebooting;
93
94 static bool largepages_enabled = true;
95
96 struct page *hwpoison_page;
97 pfn_t hwpoison_pfn;
98
99 inline int kvm_is_mmio_pfn(pfn_t pfn)
100 {
101         if (pfn_valid(pfn)) {
102                 struct page *page = compound_head(pfn_to_page(pfn));
103                 return PageReserved(page);
104         }
105
106         return true;
107 }
108
109 /*
110  * Switches to specified vcpu, until a matching vcpu_put()
111  */
112 void vcpu_load(struct kvm_vcpu *vcpu)
113 {
114         int cpu;
115
116         mutex_lock(&vcpu->mutex);
117         cpu = get_cpu();
118         preempt_notifier_register(&vcpu->preempt_notifier);
119         kvm_arch_vcpu_load(vcpu, cpu);
120         put_cpu();
121 }
122
123 void vcpu_put(struct kvm_vcpu *vcpu)
124 {
125         preempt_disable();
126         kvm_arch_vcpu_put(vcpu);
127         preempt_notifier_unregister(&vcpu->preempt_notifier);
128         preempt_enable();
129         mutex_unlock(&vcpu->mutex);
130 }
131
132 static void ack_flush(void *_completed)
133 {
134 }
135
136 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
137 {
138         int i, cpu, me;
139         cpumask_var_t cpus;
140         bool called = true;
141         struct kvm_vcpu *vcpu;
142
143         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
144
145         raw_spin_lock(&kvm->requests_lock);
146         me = smp_processor_id();
147         kvm_for_each_vcpu(i, vcpu, kvm) {
148                 if (kvm_make_check_request(req, vcpu))
149                         continue;
150                 cpu = vcpu->cpu;
151                 if (cpus != NULL && cpu != -1 && cpu != me)
152                         cpumask_set_cpu(cpu, cpus);
153         }
154         if (unlikely(cpus == NULL))
155                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
156         else if (!cpumask_empty(cpus))
157                 smp_call_function_many(cpus, ack_flush, NULL, 1);
158         else
159                 called = false;
160         raw_spin_unlock(&kvm->requests_lock);
161         free_cpumask_var(cpus);
162         return called;
163 }
164
165 void kvm_flush_remote_tlbs(struct kvm *kvm)
166 {
167         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
168                 ++kvm->stat.remote_tlb_flush;
169 }
170
171 void kvm_reload_remote_mmus(struct kvm *kvm)
172 {
173         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
174 }
175
176 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
177 {
178         struct page *page;
179         int r;
180
181         mutex_init(&vcpu->mutex);
182         vcpu->cpu = -1;
183         vcpu->kvm = kvm;
184         vcpu->vcpu_id = id;
185         init_waitqueue_head(&vcpu->wq);
186
187         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
188         if (!page) {
189                 r = -ENOMEM;
190                 goto fail;
191         }
192         vcpu->run = page_address(page);
193
194         r = kvm_arch_vcpu_init(vcpu);
195         if (r < 0)
196                 goto fail_free_run;
197         return 0;
198
199 fail_free_run:
200         free_page((unsigned long)vcpu->run);
201 fail:
202         return r;
203 }
204 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
205
206 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
207 {
208         kvm_arch_vcpu_uninit(vcpu);
209         free_page((unsigned long)vcpu->run);
210 }
211 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
212
213 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
214 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
215 {
216         return container_of(mn, struct kvm, mmu_notifier);
217 }
218
219 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
220                                              struct mm_struct *mm,
221                                              unsigned long address)
222 {
223         struct kvm *kvm = mmu_notifier_to_kvm(mn);
224         int need_tlb_flush, idx;
225
226         /*
227          * When ->invalidate_page runs, the linux pte has been zapped
228          * already but the page is still allocated until
229          * ->invalidate_page returns. So if we increase the sequence
230          * here the kvm page fault will notice if the spte can't be
231          * established because the page is going to be freed. If
232          * instead the kvm page fault establishes the spte before
233          * ->invalidate_page runs, kvm_unmap_hva will release it
234          * before returning.
235          *
236          * The sequence increase only need to be seen at spin_unlock
237          * time, and not at spin_lock time.
238          *
239          * Increasing the sequence after the spin_unlock would be
240          * unsafe because the kvm page fault could then establish the
241          * pte after kvm_unmap_hva returned, without noticing the page
242          * is going to be freed.
243          */
244         idx = srcu_read_lock(&kvm->srcu);
245         spin_lock(&kvm->mmu_lock);
246         kvm->mmu_notifier_seq++;
247         need_tlb_flush = kvm_unmap_hva(kvm, address);
248         spin_unlock(&kvm->mmu_lock);
249         srcu_read_unlock(&kvm->srcu, idx);
250
251         /* we've to flush the tlb before the pages can be freed */
252         if (need_tlb_flush)
253                 kvm_flush_remote_tlbs(kvm);
254
255 }
256
257 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
258                                         struct mm_struct *mm,
259                                         unsigned long address,
260                                         pte_t pte)
261 {
262         struct kvm *kvm = mmu_notifier_to_kvm(mn);
263         int idx;
264
265         idx = srcu_read_lock(&kvm->srcu);
266         spin_lock(&kvm->mmu_lock);
267         kvm->mmu_notifier_seq++;
268         kvm_set_spte_hva(kvm, address, pte);
269         spin_unlock(&kvm->mmu_lock);
270         srcu_read_unlock(&kvm->srcu, idx);
271 }
272
273 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
274                                                     struct mm_struct *mm,
275                                                     unsigned long start,
276                                                     unsigned long end)
277 {
278         struct kvm *kvm = mmu_notifier_to_kvm(mn);
279         int need_tlb_flush = 0, idx;
280
281         idx = srcu_read_lock(&kvm->srcu);
282         spin_lock(&kvm->mmu_lock);
283         /*
284          * The count increase must become visible at unlock time as no
285          * spte can be established without taking the mmu_lock and
286          * count is also read inside the mmu_lock critical section.
287          */
288         kvm->mmu_notifier_count++;
289         for (; start < end; start += PAGE_SIZE)
290                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
291         spin_unlock(&kvm->mmu_lock);
292         srcu_read_unlock(&kvm->srcu, idx);
293
294         /* we've to flush the tlb before the pages can be freed */
295         if (need_tlb_flush)
296                 kvm_flush_remote_tlbs(kvm);
297 }
298
299 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
300                                                   struct mm_struct *mm,
301                                                   unsigned long start,
302                                                   unsigned long end)
303 {
304         struct kvm *kvm = mmu_notifier_to_kvm(mn);
305
306         spin_lock(&kvm->mmu_lock);
307         /*
308          * This sequence increase will notify the kvm page fault that
309          * the page that is going to be mapped in the spte could have
310          * been freed.
311          */
312         kvm->mmu_notifier_seq++;
313         /*
314          * The above sequence increase must be visible before the
315          * below count decrease but both values are read by the kvm
316          * page fault under mmu_lock spinlock so we don't need to add
317          * a smb_wmb() here in between the two.
318          */
319         kvm->mmu_notifier_count--;
320         spin_unlock(&kvm->mmu_lock);
321
322         BUG_ON(kvm->mmu_notifier_count < 0);
323 }
324
325 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
326                                               struct mm_struct *mm,
327                                               unsigned long address)
328 {
329         struct kvm *kvm = mmu_notifier_to_kvm(mn);
330         int young, idx;
331
332         idx = srcu_read_lock(&kvm->srcu);
333         spin_lock(&kvm->mmu_lock);
334         young = kvm_age_hva(kvm, address);
335         spin_unlock(&kvm->mmu_lock);
336         srcu_read_unlock(&kvm->srcu, idx);
337
338         if (young)
339                 kvm_flush_remote_tlbs(kvm);
340
341         return young;
342 }
343
344 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
345                                      struct mm_struct *mm)
346 {
347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
348         int idx;
349
350         idx = srcu_read_lock(&kvm->srcu);
351         kvm_arch_flush_shadow(kvm);
352         srcu_read_unlock(&kvm->srcu, idx);
353 }
354
355 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
356         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
357         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
358         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
359         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
360         .change_pte             = kvm_mmu_notifier_change_pte,
361         .release                = kvm_mmu_notifier_release,
362 };
363
364 static int kvm_init_mmu_notifier(struct kvm *kvm)
365 {
366         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
367         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
368 }
369
370 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
371
372 static int kvm_init_mmu_notifier(struct kvm *kvm)
373 {
374         return 0;
375 }
376
377 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
378
379 static struct kvm *kvm_create_vm(void)
380 {
381         int r = 0, i;
382         struct kvm *kvm = kvm_arch_create_vm();
383
384         if (IS_ERR(kvm))
385                 goto out;
386
387         r = hardware_enable_all();
388         if (r)
389                 goto out_err_nodisable;
390
391 #ifdef CONFIG_HAVE_KVM_IRQCHIP
392         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
393         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
394 #endif
395
396         r = -ENOMEM;
397         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
398         if (!kvm->memslots)
399                 goto out_err;
400         if (init_srcu_struct(&kvm->srcu))
401                 goto out_err;
402         for (i = 0; i < KVM_NR_BUSES; i++) {
403                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
404                                         GFP_KERNEL);
405                 if (!kvm->buses[i]) {
406                         cleanup_srcu_struct(&kvm->srcu);
407                         goto out_err;
408                 }
409         }
410
411         r = kvm_init_mmu_notifier(kvm);
412         if (r) {
413                 cleanup_srcu_struct(&kvm->srcu);
414                 goto out_err;
415         }
416
417         kvm->mm = current->mm;
418         atomic_inc(&kvm->mm->mm_count);
419         spin_lock_init(&kvm->mmu_lock);
420         raw_spin_lock_init(&kvm->requests_lock);
421         kvm_eventfd_init(kvm);
422         mutex_init(&kvm->lock);
423         mutex_init(&kvm->irq_lock);
424         mutex_init(&kvm->slots_lock);
425         atomic_set(&kvm->users_count, 1);
426         spin_lock(&kvm_lock);
427         list_add(&kvm->vm_list, &vm_list);
428         spin_unlock(&kvm_lock);
429 out:
430         return kvm;
431
432 out_err:
433         hardware_disable_all();
434 out_err_nodisable:
435         for (i = 0; i < KVM_NR_BUSES; i++)
436                 kfree(kvm->buses[i]);
437         kfree(kvm->memslots);
438         kfree(kvm);
439         return ERR_PTR(r);
440 }
441
442 /*
443  * Free any memory in @free but not in @dont.
444  */
445 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
446                                   struct kvm_memory_slot *dont)
447 {
448         int i;
449
450         if (!dont || free->rmap != dont->rmap)
451                 vfree(free->rmap);
452
453         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
454                 vfree(free->dirty_bitmap);
455
456
457         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
458                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
459                         vfree(free->lpage_info[i]);
460                         free->lpage_info[i] = NULL;
461                 }
462         }
463
464         free->npages = 0;
465         free->dirty_bitmap = NULL;
466         free->rmap = NULL;
467 }
468
469 void kvm_free_physmem(struct kvm *kvm)
470 {
471         int i;
472         struct kvm_memslots *slots = kvm->memslots;
473
474         for (i = 0; i < slots->nmemslots; ++i)
475                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
476
477         kfree(kvm->memslots);
478 }
479
480 static void kvm_destroy_vm(struct kvm *kvm)
481 {
482         int i;
483         struct mm_struct *mm = kvm->mm;
484
485         kvm_arch_sync_events(kvm);
486         spin_lock(&kvm_lock);
487         list_del(&kvm->vm_list);
488         spin_unlock(&kvm_lock);
489         kvm_free_irq_routing(kvm);
490         for (i = 0; i < KVM_NR_BUSES; i++)
491                 kvm_io_bus_destroy(kvm->buses[i]);
492         kvm_coalesced_mmio_free(kvm);
493 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
494         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
495 #else
496         kvm_arch_flush_shadow(kvm);
497 #endif
498         kvm_arch_destroy_vm(kvm);
499         hardware_disable_all();
500         mmdrop(mm);
501 }
502
503 void kvm_get_kvm(struct kvm *kvm)
504 {
505         atomic_inc(&kvm->users_count);
506 }
507 EXPORT_SYMBOL_GPL(kvm_get_kvm);
508
509 void kvm_put_kvm(struct kvm *kvm)
510 {
511         if (atomic_dec_and_test(&kvm->users_count))
512                 kvm_destroy_vm(kvm);
513 }
514 EXPORT_SYMBOL_GPL(kvm_put_kvm);
515
516
517 static int kvm_vm_release(struct inode *inode, struct file *filp)
518 {
519         struct kvm *kvm = filp->private_data;
520
521         kvm_irqfd_release(kvm);
522
523         kvm_put_kvm(kvm);
524         return 0;
525 }
526
527 /*
528  * Allocate some memory and give it an address in the guest physical address
529  * space.
530  *
531  * Discontiguous memory is allowed, mostly for framebuffers.
532  *
533  * Must be called holding mmap_sem for write.
534  */
535 int __kvm_set_memory_region(struct kvm *kvm,
536                             struct kvm_userspace_memory_region *mem,
537                             int user_alloc)
538 {
539         int r, flush_shadow = 0;
540         gfn_t base_gfn;
541         unsigned long npages;
542         unsigned long i;
543         struct kvm_memory_slot *memslot;
544         struct kvm_memory_slot old, new;
545         struct kvm_memslots *slots, *old_memslots;
546
547         r = -EINVAL;
548         /* General sanity checks */
549         if (mem->memory_size & (PAGE_SIZE - 1))
550                 goto out;
551         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
552                 goto out;
553         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
554                 goto out;
555         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
556                 goto out;
557         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
558                 goto out;
559
560         memslot = &kvm->memslots->memslots[mem->slot];
561         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
562         npages = mem->memory_size >> PAGE_SHIFT;
563
564         r = -EINVAL;
565         if (npages > KVM_MEM_MAX_NR_PAGES)
566                 goto out;
567
568         if (!npages)
569                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
570
571         new = old = *memslot;
572
573         new.id = mem->slot;
574         new.base_gfn = base_gfn;
575         new.npages = npages;
576         new.flags = mem->flags;
577
578         /* Disallow changing a memory slot's size. */
579         r = -EINVAL;
580         if (npages && old.npages && npages != old.npages)
581                 goto out_free;
582
583         /* Check for overlaps */
584         r = -EEXIST;
585         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
586                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
587
588                 if (s == memslot || !s->npages)
589                         continue;
590                 if (!((base_gfn + npages <= s->base_gfn) ||
591                       (base_gfn >= s->base_gfn + s->npages)))
592                         goto out_free;
593         }
594
595         /* Free page dirty bitmap if unneeded */
596         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
597                 new.dirty_bitmap = NULL;
598
599         r = -ENOMEM;
600
601         /* Allocate if a slot is being created */
602 #ifndef CONFIG_S390
603         if (npages && !new.rmap) {
604                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
605
606                 if (!new.rmap)
607                         goto out_free;
608
609                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
610
611                 new.user_alloc = user_alloc;
612                 new.userspace_addr = mem->userspace_addr;
613         }
614         if (!npages)
615                 goto skip_lpage;
616
617         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
618                 unsigned long ugfn;
619                 unsigned long j;
620                 int lpages;
621                 int level = i + 2;
622
623                 /* Avoid unused variable warning if no large pages */
624                 (void)level;
625
626                 if (new.lpage_info[i])
627                         continue;
628
629                 lpages = 1 + (base_gfn + npages - 1) /
630                              KVM_PAGES_PER_HPAGE(level);
631                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
632
633                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
634
635                 if (!new.lpage_info[i])
636                         goto out_free;
637
638                 memset(new.lpage_info[i], 0,
639                        lpages * sizeof(*new.lpage_info[i]));
640
641                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
642                         new.lpage_info[i][0].write_count = 1;
643                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
644                         new.lpage_info[i][lpages - 1].write_count = 1;
645                 ugfn = new.userspace_addr >> PAGE_SHIFT;
646                 /*
647                  * If the gfn and userspace address are not aligned wrt each
648                  * other, or if explicitly asked to, disable large page
649                  * support for this slot
650                  */
651                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
652                     !largepages_enabled)
653                         for (j = 0; j < lpages; ++j)
654                                 new.lpage_info[i][j].write_count = 1;
655         }
656
657 skip_lpage:
658
659         /* Allocate page dirty bitmap if needed */
660         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
661                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
662
663                 new.dirty_bitmap = vmalloc(dirty_bytes);
664                 if (!new.dirty_bitmap)
665                         goto out_free;
666                 memset(new.dirty_bitmap, 0, dirty_bytes);
667                 /* destroy any largepage mappings for dirty tracking */
668                 if (old.npages)
669                         flush_shadow = 1;
670         }
671 #else  /* not defined CONFIG_S390 */
672         new.user_alloc = user_alloc;
673         if (user_alloc)
674                 new.userspace_addr = mem->userspace_addr;
675 #endif /* not defined CONFIG_S390 */
676
677         if (!npages) {
678                 r = -ENOMEM;
679                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
680                 if (!slots)
681                         goto out_free;
682                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
683                 if (mem->slot >= slots->nmemslots)
684                         slots->nmemslots = mem->slot + 1;
685                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
686
687                 old_memslots = kvm->memslots;
688                 rcu_assign_pointer(kvm->memslots, slots);
689                 synchronize_srcu_expedited(&kvm->srcu);
690                 /* From this point no new shadow pages pointing to a deleted
691                  * memslot will be created.
692                  *
693                  * validation of sp->gfn happens in:
694                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
695                  *      - kvm_is_visible_gfn (mmu_check_roots)
696                  */
697                 kvm_arch_flush_shadow(kvm);
698                 kfree(old_memslots);
699         }
700
701         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
702         if (r)
703                 goto out_free;
704
705 #ifdef CONFIG_DMAR
706         /* map the pages in iommu page table */
707         if (npages) {
708                 r = kvm_iommu_map_pages(kvm, &new);
709                 if (r)
710                         goto out_free;
711         }
712 #endif
713
714         r = -ENOMEM;
715         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
716         if (!slots)
717                 goto out_free;
718         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
719         if (mem->slot >= slots->nmemslots)
720                 slots->nmemslots = mem->slot + 1;
721
722         /* actual memory is freed via old in kvm_free_physmem_slot below */
723         if (!npages) {
724                 new.rmap = NULL;
725                 new.dirty_bitmap = NULL;
726                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
727                         new.lpage_info[i] = NULL;
728         }
729
730         slots->memslots[mem->slot] = new;
731         old_memslots = kvm->memslots;
732         rcu_assign_pointer(kvm->memslots, slots);
733         synchronize_srcu_expedited(&kvm->srcu);
734
735         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
736
737         kvm_free_physmem_slot(&old, &new);
738         kfree(old_memslots);
739
740         if (flush_shadow)
741                 kvm_arch_flush_shadow(kvm);
742
743         return 0;
744
745 out_free:
746         kvm_free_physmem_slot(&new, &old);
747 out:
748         return r;
749
750 }
751 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
752
753 int kvm_set_memory_region(struct kvm *kvm,
754                           struct kvm_userspace_memory_region *mem,
755                           int user_alloc)
756 {
757         int r;
758
759         mutex_lock(&kvm->slots_lock);
760         r = __kvm_set_memory_region(kvm, mem, user_alloc);
761         mutex_unlock(&kvm->slots_lock);
762         return r;
763 }
764 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
765
766 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
767                                    struct
768                                    kvm_userspace_memory_region *mem,
769                                    int user_alloc)
770 {
771         if (mem->slot >= KVM_MEMORY_SLOTS)
772                 return -EINVAL;
773         return kvm_set_memory_region(kvm, mem, user_alloc);
774 }
775
776 int kvm_get_dirty_log(struct kvm *kvm,
777                         struct kvm_dirty_log *log, int *is_dirty)
778 {
779         struct kvm_memory_slot *memslot;
780         int r, i;
781         unsigned long n;
782         unsigned long any = 0;
783
784         r = -EINVAL;
785         if (log->slot >= KVM_MEMORY_SLOTS)
786                 goto out;
787
788         memslot = &kvm->memslots->memslots[log->slot];
789         r = -ENOENT;
790         if (!memslot->dirty_bitmap)
791                 goto out;
792
793         n = kvm_dirty_bitmap_bytes(memslot);
794
795         for (i = 0; !any && i < n/sizeof(long); ++i)
796                 any = memslot->dirty_bitmap[i];
797
798         r = -EFAULT;
799         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
800                 goto out;
801
802         if (any)
803                 *is_dirty = 1;
804
805         r = 0;
806 out:
807         return r;
808 }
809
810 void kvm_disable_largepages(void)
811 {
812         largepages_enabled = false;
813 }
814 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
815
816 int is_error_page(struct page *page)
817 {
818         return page == bad_page || page == hwpoison_page;
819 }
820 EXPORT_SYMBOL_GPL(is_error_page);
821
822 int is_error_pfn(pfn_t pfn)
823 {
824         return pfn == bad_pfn || pfn == hwpoison_pfn;
825 }
826 EXPORT_SYMBOL_GPL(is_error_pfn);
827
828 int is_hwpoison_pfn(pfn_t pfn)
829 {
830         return pfn == hwpoison_pfn;
831 }
832 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
833
834 static inline unsigned long bad_hva(void)
835 {
836         return PAGE_OFFSET;
837 }
838
839 int kvm_is_error_hva(unsigned long addr)
840 {
841         return addr == bad_hva();
842 }
843 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
844
845 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
846 {
847         int i;
848         struct kvm_memslots *slots = kvm_memslots(kvm);
849
850         for (i = 0; i < slots->nmemslots; ++i) {
851                 struct kvm_memory_slot *memslot = &slots->memslots[i];
852
853                 if (gfn >= memslot->base_gfn
854                     && gfn < memslot->base_gfn + memslot->npages)
855                         return memslot;
856         }
857         return NULL;
858 }
859 EXPORT_SYMBOL_GPL(gfn_to_memslot);
860
861 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
862 {
863         int i;
864         struct kvm_memslots *slots = kvm_memslots(kvm);
865
866         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
867                 struct kvm_memory_slot *memslot = &slots->memslots[i];
868
869                 if (memslot->flags & KVM_MEMSLOT_INVALID)
870                         continue;
871
872                 if (gfn >= memslot->base_gfn
873                     && gfn < memslot->base_gfn + memslot->npages)
874                         return 1;
875         }
876         return 0;
877 }
878 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
879
880 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
881 {
882         struct vm_area_struct *vma;
883         unsigned long addr, size;
884
885         size = PAGE_SIZE;
886
887         addr = gfn_to_hva(kvm, gfn);
888         if (kvm_is_error_hva(addr))
889                 return PAGE_SIZE;
890
891         down_read(&current->mm->mmap_sem);
892         vma = find_vma(current->mm, addr);
893         if (!vma)
894                 goto out;
895
896         size = vma_kernel_pagesize(vma);
897
898 out:
899         up_read(&current->mm->mmap_sem);
900
901         return size;
902 }
903
904 int memslot_id(struct kvm *kvm, gfn_t gfn)
905 {
906         int i;
907         struct kvm_memslots *slots = kvm_memslots(kvm);
908         struct kvm_memory_slot *memslot = NULL;
909
910         for (i = 0; i < slots->nmemslots; ++i) {
911                 memslot = &slots->memslots[i];
912
913                 if (gfn >= memslot->base_gfn
914                     && gfn < memslot->base_gfn + memslot->npages)
915                         break;
916         }
917
918         return memslot - slots->memslots;
919 }
920
921 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
922 {
923         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
924 }
925
926 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
927 {
928         struct kvm_memory_slot *slot;
929
930         slot = gfn_to_memslot(kvm, gfn);
931         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
932                 return bad_hva();
933         return gfn_to_hva_memslot(slot, gfn);
934 }
935 EXPORT_SYMBOL_GPL(gfn_to_hva);
936
937 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
938 {
939         struct page *page[1];
940         int npages;
941         pfn_t pfn;
942
943         might_sleep();
944
945         npages = get_user_pages_fast(addr, 1, 1, page);
946
947         if (unlikely(npages != 1)) {
948                 struct vm_area_struct *vma;
949
950                 if (is_hwpoison_address(addr)) {
951                         get_page(hwpoison_page);
952                         return page_to_pfn(hwpoison_page);
953                 }
954
955                 down_read(&current->mm->mmap_sem);
956                 vma = find_vma(current->mm, addr);
957
958                 if (vma == NULL || addr < vma->vm_start ||
959                     !(vma->vm_flags & VM_PFNMAP)) {
960                         up_read(&current->mm->mmap_sem);
961                         get_page(bad_page);
962                         return page_to_pfn(bad_page);
963                 }
964
965                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
966                 up_read(&current->mm->mmap_sem);
967                 BUG_ON(!kvm_is_mmio_pfn(pfn));
968         } else
969                 pfn = page_to_pfn(page[0]);
970
971         return pfn;
972 }
973
974 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
975 {
976         unsigned long addr;
977
978         addr = gfn_to_hva(kvm, gfn);
979         if (kvm_is_error_hva(addr)) {
980                 get_page(bad_page);
981                 return page_to_pfn(bad_page);
982         }
983
984         return hva_to_pfn(kvm, addr);
985 }
986 EXPORT_SYMBOL_GPL(gfn_to_pfn);
987
988 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
989                          struct kvm_memory_slot *slot, gfn_t gfn)
990 {
991         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
992         return hva_to_pfn(kvm, addr);
993 }
994
995 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
996 {
997         pfn_t pfn;
998
999         pfn = gfn_to_pfn(kvm, gfn);
1000         if (!kvm_is_mmio_pfn(pfn))
1001                 return pfn_to_page(pfn);
1002
1003         WARN_ON(kvm_is_mmio_pfn(pfn));
1004
1005         get_page(bad_page);
1006         return bad_page;
1007 }
1008
1009 EXPORT_SYMBOL_GPL(gfn_to_page);
1010
1011 void kvm_release_page_clean(struct page *page)
1012 {
1013         kvm_release_pfn_clean(page_to_pfn(page));
1014 }
1015 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1016
1017 void kvm_release_pfn_clean(pfn_t pfn)
1018 {
1019         if (!kvm_is_mmio_pfn(pfn))
1020                 put_page(pfn_to_page(pfn));
1021 }
1022 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1023
1024 void kvm_release_page_dirty(struct page *page)
1025 {
1026         kvm_release_pfn_dirty(page_to_pfn(page));
1027 }
1028 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1029
1030 void kvm_release_pfn_dirty(pfn_t pfn)
1031 {
1032         kvm_set_pfn_dirty(pfn);
1033         kvm_release_pfn_clean(pfn);
1034 }
1035 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1036
1037 void kvm_set_page_dirty(struct page *page)
1038 {
1039         kvm_set_pfn_dirty(page_to_pfn(page));
1040 }
1041 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1042
1043 void kvm_set_pfn_dirty(pfn_t pfn)
1044 {
1045         if (!kvm_is_mmio_pfn(pfn)) {
1046                 struct page *page = pfn_to_page(pfn);
1047                 if (!PageReserved(page))
1048                         SetPageDirty(page);
1049         }
1050 }
1051 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1052
1053 void kvm_set_pfn_accessed(pfn_t pfn)
1054 {
1055         if (!kvm_is_mmio_pfn(pfn))
1056                 mark_page_accessed(pfn_to_page(pfn));
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1059
1060 void kvm_get_pfn(pfn_t pfn)
1061 {
1062         if (!kvm_is_mmio_pfn(pfn))
1063                 get_page(pfn_to_page(pfn));
1064 }
1065 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1066
1067 static int next_segment(unsigned long len, int offset)
1068 {
1069         if (len > PAGE_SIZE - offset)
1070                 return PAGE_SIZE - offset;
1071         else
1072                 return len;
1073 }
1074
1075 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1076                         int len)
1077 {
1078         int r;
1079         unsigned long addr;
1080
1081         addr = gfn_to_hva(kvm, gfn);
1082         if (kvm_is_error_hva(addr))
1083                 return -EFAULT;
1084         r = copy_from_user(data, (void __user *)addr + offset, len);
1085         if (r)
1086                 return -EFAULT;
1087         return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1090
1091 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1092 {
1093         gfn_t gfn = gpa >> PAGE_SHIFT;
1094         int seg;
1095         int offset = offset_in_page(gpa);
1096         int ret;
1097
1098         while ((seg = next_segment(len, offset)) != 0) {
1099                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1100                 if (ret < 0)
1101                         return ret;
1102                 offset = 0;
1103                 len -= seg;
1104                 data += seg;
1105                 ++gfn;
1106         }
1107         return 0;
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_read_guest);
1110
1111 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1112                           unsigned long len)
1113 {
1114         int r;
1115         unsigned long addr;
1116         gfn_t gfn = gpa >> PAGE_SHIFT;
1117         int offset = offset_in_page(gpa);
1118
1119         addr = gfn_to_hva(kvm, gfn);
1120         if (kvm_is_error_hva(addr))
1121                 return -EFAULT;
1122         pagefault_disable();
1123         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1124         pagefault_enable();
1125         if (r)
1126                 return -EFAULT;
1127         return 0;
1128 }
1129 EXPORT_SYMBOL(kvm_read_guest_atomic);
1130
1131 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1132                          int offset, int len)
1133 {
1134         int r;
1135         unsigned long addr;
1136
1137         addr = gfn_to_hva(kvm, gfn);
1138         if (kvm_is_error_hva(addr))
1139                 return -EFAULT;
1140         r = copy_to_user((void __user *)addr + offset, data, len);
1141         if (r)
1142                 return -EFAULT;
1143         mark_page_dirty(kvm, gfn);
1144         return 0;
1145 }
1146 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1147
1148 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1149                     unsigned long len)
1150 {
1151         gfn_t gfn = gpa >> PAGE_SHIFT;
1152         int seg;
1153         int offset = offset_in_page(gpa);
1154         int ret;
1155
1156         while ((seg = next_segment(len, offset)) != 0) {
1157                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1158                 if (ret < 0)
1159                         return ret;
1160                 offset = 0;
1161                 len -= seg;
1162                 data += seg;
1163                 ++gfn;
1164         }
1165         return 0;
1166 }
1167
1168 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1169 {
1170         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1171 }
1172 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1173
1174 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1175 {
1176         gfn_t gfn = gpa >> PAGE_SHIFT;
1177         int seg;
1178         int offset = offset_in_page(gpa);
1179         int ret;
1180
1181         while ((seg = next_segment(len, offset)) != 0) {
1182                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1183                 if (ret < 0)
1184                         return ret;
1185                 offset = 0;
1186                 len -= seg;
1187                 ++gfn;
1188         }
1189         return 0;
1190 }
1191 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1192
1193 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1194 {
1195         struct kvm_memory_slot *memslot;
1196
1197         memslot = gfn_to_memslot(kvm, gfn);
1198         if (memslot && memslot->dirty_bitmap) {
1199                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1200
1201                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1202         }
1203 }
1204
1205 /*
1206  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1207  */
1208 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1209 {
1210         DEFINE_WAIT(wait);
1211
1212         for (;;) {
1213                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1214
1215                 if (kvm_arch_vcpu_runnable(vcpu)) {
1216                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1217                         break;
1218                 }
1219                 if (kvm_cpu_has_pending_timer(vcpu))
1220                         break;
1221                 if (signal_pending(current))
1222                         break;
1223
1224                 schedule();
1225         }
1226
1227         finish_wait(&vcpu->wq, &wait);
1228 }
1229
1230 void kvm_resched(struct kvm_vcpu *vcpu)
1231 {
1232         if (!need_resched())
1233                 return;
1234         cond_resched();
1235 }
1236 EXPORT_SYMBOL_GPL(kvm_resched);
1237
1238 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1239 {
1240         ktime_t expires;
1241         DEFINE_WAIT(wait);
1242
1243         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1244
1245         /* Sleep for 100 us, and hope lock-holder got scheduled */
1246         expires = ktime_add_ns(ktime_get(), 100000UL);
1247         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1248
1249         finish_wait(&vcpu->wq, &wait);
1250 }
1251 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1252
1253 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1254 {
1255         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1256         struct page *page;
1257
1258         if (vmf->pgoff == 0)
1259                 page = virt_to_page(vcpu->run);
1260 #ifdef CONFIG_X86
1261         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1262                 page = virt_to_page(vcpu->arch.pio_data);
1263 #endif
1264 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1265         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1266                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1267 #endif
1268         else
1269                 return VM_FAULT_SIGBUS;
1270         get_page(page);
1271         vmf->page = page;
1272         return 0;
1273 }
1274
1275 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1276         .fault = kvm_vcpu_fault,
1277 };
1278
1279 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1280 {
1281         vma->vm_ops = &kvm_vcpu_vm_ops;
1282         return 0;
1283 }
1284
1285 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1286 {
1287         struct kvm_vcpu *vcpu = filp->private_data;
1288
1289         kvm_put_kvm(vcpu->kvm);
1290         return 0;
1291 }
1292
1293 static struct file_operations kvm_vcpu_fops = {
1294         .release        = kvm_vcpu_release,
1295         .unlocked_ioctl = kvm_vcpu_ioctl,
1296         .compat_ioctl   = kvm_vcpu_ioctl,
1297         .mmap           = kvm_vcpu_mmap,
1298 };
1299
1300 /*
1301  * Allocates an inode for the vcpu.
1302  */
1303 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1304 {
1305         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1306 }
1307
1308 /*
1309  * Creates some virtual cpus.  Good luck creating more than one.
1310  */
1311 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1312 {
1313         int r;
1314         struct kvm_vcpu *vcpu, *v;
1315
1316         vcpu = kvm_arch_vcpu_create(kvm, id);
1317         if (IS_ERR(vcpu))
1318                 return PTR_ERR(vcpu);
1319
1320         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1321
1322         r = kvm_arch_vcpu_setup(vcpu);
1323         if (r)
1324                 return r;
1325
1326         mutex_lock(&kvm->lock);
1327         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1328                 r = -EINVAL;
1329                 goto vcpu_destroy;
1330         }
1331
1332         kvm_for_each_vcpu(r, v, kvm)
1333                 if (v->vcpu_id == id) {
1334                         r = -EEXIST;
1335                         goto vcpu_destroy;
1336                 }
1337
1338         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1339
1340         /* Now it's all set up, let userspace reach it */
1341         kvm_get_kvm(kvm);
1342         r = create_vcpu_fd(vcpu);
1343         if (r < 0) {
1344                 kvm_put_kvm(kvm);
1345                 goto vcpu_destroy;
1346         }
1347
1348         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1349         smp_wmb();
1350         atomic_inc(&kvm->online_vcpus);
1351
1352 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1353         if (kvm->bsp_vcpu_id == id)
1354                 kvm->bsp_vcpu = vcpu;
1355 #endif
1356         mutex_unlock(&kvm->lock);
1357         return r;
1358
1359 vcpu_destroy:
1360         mutex_unlock(&kvm->lock);
1361         kvm_arch_vcpu_destroy(vcpu);
1362         return r;
1363 }
1364
1365 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1366 {
1367         if (sigset) {
1368                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1369                 vcpu->sigset_active = 1;
1370                 vcpu->sigset = *sigset;
1371         } else
1372                 vcpu->sigset_active = 0;
1373         return 0;
1374 }
1375
1376 static long kvm_vcpu_ioctl(struct file *filp,
1377                            unsigned int ioctl, unsigned long arg)
1378 {
1379         struct kvm_vcpu *vcpu = filp->private_data;
1380         void __user *argp = (void __user *)arg;
1381         int r;
1382         struct kvm_fpu *fpu = NULL;
1383         struct kvm_sregs *kvm_sregs = NULL;
1384
1385         if (vcpu->kvm->mm != current->mm)
1386                 return -EIO;
1387
1388 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1389         /*
1390          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1391          * so vcpu_load() would break it.
1392          */
1393         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1394                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1395 #endif
1396
1397
1398         vcpu_load(vcpu);
1399         switch (ioctl) {
1400         case KVM_RUN:
1401                 r = -EINVAL;
1402                 if (arg)
1403                         goto out;
1404                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1405                 break;
1406         case KVM_GET_REGS: {
1407                 struct kvm_regs *kvm_regs;
1408
1409                 r = -ENOMEM;
1410                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1411                 if (!kvm_regs)
1412                         goto out;
1413                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1414                 if (r)
1415                         goto out_free1;
1416                 r = -EFAULT;
1417                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1418                         goto out_free1;
1419                 r = 0;
1420 out_free1:
1421                 kfree(kvm_regs);
1422                 break;
1423         }
1424         case KVM_SET_REGS: {
1425                 struct kvm_regs *kvm_regs;
1426
1427                 r = -ENOMEM;
1428                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1429                 if (!kvm_regs)
1430                         goto out;
1431                 r = -EFAULT;
1432                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1433                         goto out_free2;
1434                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1435                 if (r)
1436                         goto out_free2;
1437                 r = 0;
1438 out_free2:
1439                 kfree(kvm_regs);
1440                 break;
1441         }
1442         case KVM_GET_SREGS: {
1443                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1444                 r = -ENOMEM;
1445                 if (!kvm_sregs)
1446                         goto out;
1447                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1448                 if (r)
1449                         goto out;
1450                 r = -EFAULT;
1451                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1452                         goto out;
1453                 r = 0;
1454                 break;
1455         }
1456         case KVM_SET_SREGS: {
1457                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1458                 r = -ENOMEM;
1459                 if (!kvm_sregs)
1460                         goto out;
1461                 r = -EFAULT;
1462                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1463                         goto out;
1464                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1465                 if (r)
1466                         goto out;
1467                 r = 0;
1468                 break;
1469         }
1470         case KVM_GET_MP_STATE: {
1471                 struct kvm_mp_state mp_state;
1472
1473                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1474                 if (r)
1475                         goto out;
1476                 r = -EFAULT;
1477                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1478                         goto out;
1479                 r = 0;
1480                 break;
1481         }
1482         case KVM_SET_MP_STATE: {
1483                 struct kvm_mp_state mp_state;
1484
1485                 r = -EFAULT;
1486                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1487                         goto out;
1488                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1489                 if (r)
1490                         goto out;
1491                 r = 0;
1492                 break;
1493         }
1494         case KVM_TRANSLATE: {
1495                 struct kvm_translation tr;
1496
1497                 r = -EFAULT;
1498                 if (copy_from_user(&tr, argp, sizeof tr))
1499                         goto out;
1500                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1501                 if (r)
1502                         goto out;
1503                 r = -EFAULT;
1504                 if (copy_to_user(argp, &tr, sizeof tr))
1505                         goto out;
1506                 r = 0;
1507                 break;
1508         }
1509         case KVM_SET_GUEST_DEBUG: {
1510                 struct kvm_guest_debug dbg;
1511
1512                 r = -EFAULT;
1513                 if (copy_from_user(&dbg, argp, sizeof dbg))
1514                         goto out;
1515                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1516                 if (r)
1517                         goto out;
1518                 r = 0;
1519                 break;
1520         }
1521         case KVM_SET_SIGNAL_MASK: {
1522                 struct kvm_signal_mask __user *sigmask_arg = argp;
1523                 struct kvm_signal_mask kvm_sigmask;
1524                 sigset_t sigset, *p;
1525
1526                 p = NULL;
1527                 if (argp) {
1528                         r = -EFAULT;
1529                         if (copy_from_user(&kvm_sigmask, argp,
1530                                            sizeof kvm_sigmask))
1531                                 goto out;
1532                         r = -EINVAL;
1533                         if (kvm_sigmask.len != sizeof sigset)
1534                                 goto out;
1535                         r = -EFAULT;
1536                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1537                                            sizeof sigset))
1538                                 goto out;
1539                         p = &sigset;
1540                 }
1541                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1542                 break;
1543         }
1544         case KVM_GET_FPU: {
1545                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1546                 r = -ENOMEM;
1547                 if (!fpu)
1548                         goto out;
1549                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1550                 if (r)
1551                         goto out;
1552                 r = -EFAULT;
1553                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1554                         goto out;
1555                 r = 0;
1556                 break;
1557         }
1558         case KVM_SET_FPU: {
1559                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1560                 r = -ENOMEM;
1561                 if (!fpu)
1562                         goto out;
1563                 r = -EFAULT;
1564                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1565                         goto out;
1566                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1567                 if (r)
1568                         goto out;
1569                 r = 0;
1570                 break;
1571         }
1572         default:
1573                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1574         }
1575 out:
1576         vcpu_put(vcpu);
1577         kfree(fpu);
1578         kfree(kvm_sregs);
1579         return r;
1580 }
1581
1582 static long kvm_vm_ioctl(struct file *filp,
1583                            unsigned int ioctl, unsigned long arg)
1584 {
1585         struct kvm *kvm = filp->private_data;
1586         void __user *argp = (void __user *)arg;
1587         int r;
1588
1589         if (kvm->mm != current->mm)
1590                 return -EIO;
1591         switch (ioctl) {
1592         case KVM_CREATE_VCPU:
1593                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1594                 if (r < 0)
1595                         goto out;
1596                 break;
1597         case KVM_SET_USER_MEMORY_REGION: {
1598                 struct kvm_userspace_memory_region kvm_userspace_mem;
1599
1600                 r = -EFAULT;
1601                 if (copy_from_user(&kvm_userspace_mem, argp,
1602                                                 sizeof kvm_userspace_mem))
1603                         goto out;
1604
1605                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1606                 if (r)
1607                         goto out;
1608                 break;
1609         }
1610         case KVM_GET_DIRTY_LOG: {
1611                 struct kvm_dirty_log log;
1612
1613                 r = -EFAULT;
1614                 if (copy_from_user(&log, argp, sizeof log))
1615                         goto out;
1616                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1617                 if (r)
1618                         goto out;
1619                 break;
1620         }
1621 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1622         case KVM_REGISTER_COALESCED_MMIO: {
1623                 struct kvm_coalesced_mmio_zone zone;
1624                 r = -EFAULT;
1625                 if (copy_from_user(&zone, argp, sizeof zone))
1626                         goto out;
1627                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1628                 if (r)
1629                         goto out;
1630                 r = 0;
1631                 break;
1632         }
1633         case KVM_UNREGISTER_COALESCED_MMIO: {
1634                 struct kvm_coalesced_mmio_zone zone;
1635                 r = -EFAULT;
1636                 if (copy_from_user(&zone, argp, sizeof zone))
1637                         goto out;
1638                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1639                 if (r)
1640                         goto out;
1641                 r = 0;
1642                 break;
1643         }
1644 #endif
1645         case KVM_IRQFD: {
1646                 struct kvm_irqfd data;
1647
1648                 r = -EFAULT;
1649                 if (copy_from_user(&data, argp, sizeof data))
1650                         goto out;
1651                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1652                 break;
1653         }
1654         case KVM_IOEVENTFD: {
1655                 struct kvm_ioeventfd data;
1656
1657                 r = -EFAULT;
1658                 if (copy_from_user(&data, argp, sizeof data))
1659                         goto out;
1660                 r = kvm_ioeventfd(kvm, &data);
1661                 break;
1662         }
1663 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1664         case KVM_SET_BOOT_CPU_ID:
1665                 r = 0;
1666                 mutex_lock(&kvm->lock);
1667                 if (atomic_read(&kvm->online_vcpus) != 0)
1668                         r = -EBUSY;
1669                 else
1670                         kvm->bsp_vcpu_id = arg;
1671                 mutex_unlock(&kvm->lock);
1672                 break;
1673 #endif
1674         default:
1675                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1676                 if (r == -ENOTTY)
1677                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1678         }
1679 out:
1680         return r;
1681 }
1682
1683 #ifdef CONFIG_COMPAT
1684 struct compat_kvm_dirty_log {
1685         __u32 slot;
1686         __u32 padding1;
1687         union {
1688                 compat_uptr_t dirty_bitmap; /* one bit per page */
1689                 __u64 padding2;
1690         };
1691 };
1692
1693 static long kvm_vm_compat_ioctl(struct file *filp,
1694                            unsigned int ioctl, unsigned long arg)
1695 {
1696         struct kvm *kvm = filp->private_data;
1697         int r;
1698
1699         if (kvm->mm != current->mm)
1700                 return -EIO;
1701         switch (ioctl) {
1702         case KVM_GET_DIRTY_LOG: {
1703                 struct compat_kvm_dirty_log compat_log;
1704                 struct kvm_dirty_log log;
1705
1706                 r = -EFAULT;
1707                 if (copy_from_user(&compat_log, (void __user *)arg,
1708                                    sizeof(compat_log)))
1709                         goto out;
1710                 log.slot         = compat_log.slot;
1711                 log.padding1     = compat_log.padding1;
1712                 log.padding2     = compat_log.padding2;
1713                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1714
1715                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1716                 if (r)
1717                         goto out;
1718                 break;
1719         }
1720         default:
1721                 r = kvm_vm_ioctl(filp, ioctl, arg);
1722         }
1723
1724 out:
1725         return r;
1726 }
1727 #endif
1728
1729 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1730 {
1731         struct page *page[1];
1732         unsigned long addr;
1733         int npages;
1734         gfn_t gfn = vmf->pgoff;
1735         struct kvm *kvm = vma->vm_file->private_data;
1736
1737         addr = gfn_to_hva(kvm, gfn);
1738         if (kvm_is_error_hva(addr))
1739                 return VM_FAULT_SIGBUS;
1740
1741         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1742                                 NULL);
1743         if (unlikely(npages != 1))
1744                 return VM_FAULT_SIGBUS;
1745
1746         vmf->page = page[0];
1747         return 0;
1748 }
1749
1750 static const struct vm_operations_struct kvm_vm_vm_ops = {
1751         .fault = kvm_vm_fault,
1752 };
1753
1754 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1755 {
1756         vma->vm_ops = &kvm_vm_vm_ops;
1757         return 0;
1758 }
1759
1760 static struct file_operations kvm_vm_fops = {
1761         .release        = kvm_vm_release,
1762         .unlocked_ioctl = kvm_vm_ioctl,
1763 #ifdef CONFIG_COMPAT
1764         .compat_ioctl   = kvm_vm_compat_ioctl,
1765 #endif
1766         .mmap           = kvm_vm_mmap,
1767 };
1768
1769 static int kvm_dev_ioctl_create_vm(void)
1770 {
1771         int fd, r;
1772         struct kvm *kvm;
1773
1774         kvm = kvm_create_vm();
1775         if (IS_ERR(kvm))
1776                 return PTR_ERR(kvm);
1777 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1778         r = kvm_coalesced_mmio_init(kvm);
1779         if (r < 0) {
1780                 kvm_put_kvm(kvm);
1781                 return r;
1782         }
1783 #endif
1784         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1785         if (fd < 0)
1786                 kvm_put_kvm(kvm);
1787
1788         return fd;
1789 }
1790
1791 static long kvm_dev_ioctl_check_extension_generic(long arg)
1792 {
1793         switch (arg) {
1794         case KVM_CAP_USER_MEMORY:
1795         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1796         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1797 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1798         case KVM_CAP_SET_BOOT_CPU_ID:
1799 #endif
1800         case KVM_CAP_INTERNAL_ERROR_DATA:
1801                 return 1;
1802 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1803         case KVM_CAP_IRQ_ROUTING:
1804                 return KVM_MAX_IRQ_ROUTES;
1805 #endif
1806         default:
1807                 break;
1808         }
1809         return kvm_dev_ioctl_check_extension(arg);
1810 }
1811
1812 static long kvm_dev_ioctl(struct file *filp,
1813                           unsigned int ioctl, unsigned long arg)
1814 {
1815         long r = -EINVAL;
1816
1817         switch (ioctl) {
1818         case KVM_GET_API_VERSION:
1819                 r = -EINVAL;
1820                 if (arg)
1821                         goto out;
1822                 r = KVM_API_VERSION;
1823                 break;
1824         case KVM_CREATE_VM:
1825                 r = -EINVAL;
1826                 if (arg)
1827                         goto out;
1828                 r = kvm_dev_ioctl_create_vm();
1829                 break;
1830         case KVM_CHECK_EXTENSION:
1831                 r = kvm_dev_ioctl_check_extension_generic(arg);
1832                 break;
1833         case KVM_GET_VCPU_MMAP_SIZE:
1834                 r = -EINVAL;
1835                 if (arg)
1836                         goto out;
1837                 r = PAGE_SIZE;     /* struct kvm_run */
1838 #ifdef CONFIG_X86
1839                 r += PAGE_SIZE;    /* pio data page */
1840 #endif
1841 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1842                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1843 #endif
1844                 break;
1845         case KVM_TRACE_ENABLE:
1846         case KVM_TRACE_PAUSE:
1847         case KVM_TRACE_DISABLE:
1848                 r = -EOPNOTSUPP;
1849                 break;
1850         default:
1851                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1852         }
1853 out:
1854         return r;
1855 }
1856
1857 static struct file_operations kvm_chardev_ops = {
1858         .unlocked_ioctl = kvm_dev_ioctl,
1859         .compat_ioctl   = kvm_dev_ioctl,
1860 };
1861
1862 static struct miscdevice kvm_dev = {
1863         KVM_MINOR,
1864         "kvm",
1865         &kvm_chardev_ops,
1866 };
1867
1868 static void hardware_enable(void *junk)
1869 {
1870         int cpu = raw_smp_processor_id();
1871         int r;
1872
1873         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1874                 return;
1875
1876         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1877
1878         r = kvm_arch_hardware_enable(NULL);
1879
1880         if (r) {
1881                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1882                 atomic_inc(&hardware_enable_failed);
1883                 printk(KERN_INFO "kvm: enabling virtualization on "
1884                                  "CPU%d failed\n", cpu);
1885         }
1886 }
1887
1888 static void hardware_disable(void *junk)
1889 {
1890         int cpu = raw_smp_processor_id();
1891
1892         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1893                 return;
1894         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1895         kvm_arch_hardware_disable(NULL);
1896 }
1897
1898 static void hardware_disable_all_nolock(void)
1899 {
1900         BUG_ON(!kvm_usage_count);
1901
1902         kvm_usage_count--;
1903         if (!kvm_usage_count)
1904                 on_each_cpu(hardware_disable, NULL, 1);
1905 }
1906
1907 static void hardware_disable_all(void)
1908 {
1909         spin_lock(&kvm_lock);
1910         hardware_disable_all_nolock();
1911         spin_unlock(&kvm_lock);
1912 }
1913
1914 static int hardware_enable_all(void)
1915 {
1916         int r = 0;
1917
1918         spin_lock(&kvm_lock);
1919
1920         kvm_usage_count++;
1921         if (kvm_usage_count == 1) {
1922                 atomic_set(&hardware_enable_failed, 0);
1923                 on_each_cpu(hardware_enable, NULL, 1);
1924
1925                 if (atomic_read(&hardware_enable_failed)) {
1926                         hardware_disable_all_nolock();
1927                         r = -EBUSY;
1928                 }
1929         }
1930
1931         spin_unlock(&kvm_lock);
1932
1933         return r;
1934 }
1935
1936 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1937                            void *v)
1938 {
1939         int cpu = (long)v;
1940
1941         if (!kvm_usage_count)
1942                 return NOTIFY_OK;
1943
1944         val &= ~CPU_TASKS_FROZEN;
1945         switch (val) {
1946         case CPU_DYING:
1947                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1948                        cpu);
1949                 hardware_disable(NULL);
1950                 break;
1951         case CPU_ONLINE:
1952                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1953                        cpu);
1954                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1955                 break;
1956         }
1957         return NOTIFY_OK;
1958 }
1959
1960
1961 asmlinkage void kvm_handle_fault_on_reboot(void)
1962 {
1963         if (kvm_rebooting)
1964                 /* spin while reset goes on */
1965                 while (true)
1966                         ;
1967         /* Fault while not rebooting.  We want the trace. */
1968         BUG();
1969 }
1970 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1971
1972 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1973                       void *v)
1974 {
1975         /*
1976          * Some (well, at least mine) BIOSes hang on reboot if
1977          * in vmx root mode.
1978          *
1979          * And Intel TXT required VMX off for all cpu when system shutdown.
1980          */
1981         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1982         kvm_rebooting = true;
1983         on_each_cpu(hardware_disable, NULL, 1);
1984         return NOTIFY_OK;
1985 }
1986
1987 static struct notifier_block kvm_reboot_notifier = {
1988         .notifier_call = kvm_reboot,
1989         .priority = 0,
1990 };
1991
1992 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1993 {
1994         int i;
1995
1996         for (i = 0; i < bus->dev_count; i++) {
1997                 struct kvm_io_device *pos = bus->devs[i];
1998
1999                 kvm_iodevice_destructor(pos);
2000         }
2001         kfree(bus);
2002 }
2003
2004 /* kvm_io_bus_write - called under kvm->slots_lock */
2005 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2006                      int len, const void *val)
2007 {
2008         int i;
2009         struct kvm_io_bus *bus;
2010
2011         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2012         for (i = 0; i < bus->dev_count; i++)
2013                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2014                         return 0;
2015         return -EOPNOTSUPP;
2016 }
2017
2018 /* kvm_io_bus_read - called under kvm->slots_lock */
2019 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2020                     int len, void *val)
2021 {
2022         int i;
2023         struct kvm_io_bus *bus;
2024
2025         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2026         for (i = 0; i < bus->dev_count; i++)
2027                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2028                         return 0;
2029         return -EOPNOTSUPP;
2030 }
2031
2032 /* Caller must hold slots_lock. */
2033 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2034                             struct kvm_io_device *dev)
2035 {
2036         struct kvm_io_bus *new_bus, *bus;
2037
2038         bus = kvm->buses[bus_idx];
2039         if (bus->dev_count > NR_IOBUS_DEVS-1)
2040                 return -ENOSPC;
2041
2042         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2043         if (!new_bus)
2044                 return -ENOMEM;
2045         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2046         new_bus->devs[new_bus->dev_count++] = dev;
2047         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2048         synchronize_srcu_expedited(&kvm->srcu);
2049         kfree(bus);
2050
2051         return 0;
2052 }
2053
2054 /* Caller must hold slots_lock. */
2055 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2056                               struct kvm_io_device *dev)
2057 {
2058         int i, r;
2059         struct kvm_io_bus *new_bus, *bus;
2060
2061         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2062         if (!new_bus)
2063                 return -ENOMEM;
2064
2065         bus = kvm->buses[bus_idx];
2066         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2067
2068         r = -ENOENT;
2069         for (i = 0; i < new_bus->dev_count; i++)
2070                 if (new_bus->devs[i] == dev) {
2071                         r = 0;
2072                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2073                         break;
2074                 }
2075
2076         if (r) {
2077                 kfree(new_bus);
2078                 return r;
2079         }
2080
2081         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2082         synchronize_srcu_expedited(&kvm->srcu);
2083         kfree(bus);
2084         return r;
2085 }
2086
2087 static struct notifier_block kvm_cpu_notifier = {
2088         .notifier_call = kvm_cpu_hotplug,
2089         .priority = 20, /* must be > scheduler priority */
2090 };
2091
2092 static int vm_stat_get(void *_offset, u64 *val)
2093 {
2094         unsigned offset = (long)_offset;
2095         struct kvm *kvm;
2096
2097         *val = 0;
2098         spin_lock(&kvm_lock);
2099         list_for_each_entry(kvm, &vm_list, vm_list)
2100                 *val += *(u32 *)((void *)kvm + offset);
2101         spin_unlock(&kvm_lock);
2102         return 0;
2103 }
2104
2105 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2106
2107 static int vcpu_stat_get(void *_offset, u64 *val)
2108 {
2109         unsigned offset = (long)_offset;
2110         struct kvm *kvm;
2111         struct kvm_vcpu *vcpu;
2112         int i;
2113
2114         *val = 0;
2115         spin_lock(&kvm_lock);
2116         list_for_each_entry(kvm, &vm_list, vm_list)
2117                 kvm_for_each_vcpu(i, vcpu, kvm)
2118                         *val += *(u32 *)((void *)vcpu + offset);
2119
2120         spin_unlock(&kvm_lock);
2121         return 0;
2122 }
2123
2124 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2125
2126 static const struct file_operations *stat_fops[] = {
2127         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2128         [KVM_STAT_VM]   = &vm_stat_fops,
2129 };
2130
2131 static void kvm_init_debug(void)
2132 {
2133         struct kvm_stats_debugfs_item *p;
2134
2135         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2136         for (p = debugfs_entries; p->name; ++p)
2137                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2138                                                 (void *)(long)p->offset,
2139                                                 stat_fops[p->kind]);
2140 }
2141
2142 static void kvm_exit_debug(void)
2143 {
2144         struct kvm_stats_debugfs_item *p;
2145
2146         for (p = debugfs_entries; p->name; ++p)
2147                 debugfs_remove(p->dentry);
2148         debugfs_remove(kvm_debugfs_dir);
2149 }
2150
2151 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2152 {
2153         if (kvm_usage_count)
2154                 hardware_disable(NULL);
2155         return 0;
2156 }
2157
2158 static int kvm_resume(struct sys_device *dev)
2159 {
2160         if (kvm_usage_count)
2161                 hardware_enable(NULL);
2162         return 0;
2163 }
2164
2165 static struct sysdev_class kvm_sysdev_class = {
2166         .name = "kvm",
2167         .suspend = kvm_suspend,
2168         .resume = kvm_resume,
2169 };
2170
2171 static struct sys_device kvm_sysdev = {
2172         .id = 0,
2173         .cls = &kvm_sysdev_class,
2174 };
2175
2176 struct page *bad_page;
2177 pfn_t bad_pfn;
2178
2179 static inline
2180 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2181 {
2182         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2183 }
2184
2185 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2186 {
2187         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2188
2189         kvm_arch_vcpu_load(vcpu, cpu);
2190 }
2191
2192 static void kvm_sched_out(struct preempt_notifier *pn,
2193                           struct task_struct *next)
2194 {
2195         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2196
2197         kvm_arch_vcpu_put(vcpu);
2198 }
2199
2200 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2201                   struct module *module)
2202 {
2203         int r;
2204         int cpu;
2205
2206         r = kvm_arch_init(opaque);
2207         if (r)
2208                 goto out_fail;
2209
2210         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2211
2212         if (bad_page == NULL) {
2213                 r = -ENOMEM;
2214                 goto out;
2215         }
2216
2217         bad_pfn = page_to_pfn(bad_page);
2218
2219         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2220
2221         if (hwpoison_page == NULL) {
2222                 r = -ENOMEM;
2223                 goto out_free_0;
2224         }
2225
2226         hwpoison_pfn = page_to_pfn(hwpoison_page);
2227
2228         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2229                 r = -ENOMEM;
2230                 goto out_free_0;
2231         }
2232
2233         r = kvm_arch_hardware_setup();
2234         if (r < 0)
2235                 goto out_free_0a;
2236
2237         for_each_online_cpu(cpu) {
2238                 smp_call_function_single(cpu,
2239                                 kvm_arch_check_processor_compat,
2240                                 &r, 1);
2241                 if (r < 0)
2242                         goto out_free_1;
2243         }
2244
2245         r = register_cpu_notifier(&kvm_cpu_notifier);
2246         if (r)
2247                 goto out_free_2;
2248         register_reboot_notifier(&kvm_reboot_notifier);
2249
2250         r = sysdev_class_register(&kvm_sysdev_class);
2251         if (r)
2252                 goto out_free_3;
2253
2254         r = sysdev_register(&kvm_sysdev);
2255         if (r)
2256                 goto out_free_4;
2257
2258         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2259         if (!vcpu_align)
2260                 vcpu_align = __alignof__(struct kvm_vcpu);
2261         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2262                                            0, NULL);
2263         if (!kvm_vcpu_cache) {
2264                 r = -ENOMEM;
2265                 goto out_free_5;
2266         }
2267
2268         kvm_chardev_ops.owner = module;
2269         kvm_vm_fops.owner = module;
2270         kvm_vcpu_fops.owner = module;
2271
2272         r = misc_register(&kvm_dev);
2273         if (r) {
2274                 printk(KERN_ERR "kvm: misc device register failed\n");
2275                 goto out_free;
2276         }
2277
2278         kvm_preempt_ops.sched_in = kvm_sched_in;
2279         kvm_preempt_ops.sched_out = kvm_sched_out;
2280
2281         kvm_init_debug();
2282
2283         return 0;
2284
2285 out_free:
2286         kmem_cache_destroy(kvm_vcpu_cache);
2287 out_free_5:
2288         sysdev_unregister(&kvm_sysdev);
2289 out_free_4:
2290         sysdev_class_unregister(&kvm_sysdev_class);
2291 out_free_3:
2292         unregister_reboot_notifier(&kvm_reboot_notifier);
2293         unregister_cpu_notifier(&kvm_cpu_notifier);
2294 out_free_2:
2295 out_free_1:
2296         kvm_arch_hardware_unsetup();
2297 out_free_0a:
2298         free_cpumask_var(cpus_hardware_enabled);
2299 out_free_0:
2300         if (hwpoison_page)
2301                 __free_page(hwpoison_page);
2302         __free_page(bad_page);
2303 out:
2304         kvm_arch_exit();
2305 out_fail:
2306         return r;
2307 }
2308 EXPORT_SYMBOL_GPL(kvm_init);
2309
2310 void kvm_exit(void)
2311 {
2312         kvm_exit_debug();
2313         misc_deregister(&kvm_dev);
2314         kmem_cache_destroy(kvm_vcpu_cache);
2315         sysdev_unregister(&kvm_sysdev);
2316         sysdev_class_unregister(&kvm_sysdev_class);
2317         unregister_reboot_notifier(&kvm_reboot_notifier);
2318         unregister_cpu_notifier(&kvm_cpu_notifier);
2319         on_each_cpu(hardware_disable, NULL, 1);
2320         kvm_arch_hardware_unsetup();
2321         kvm_arch_exit();
2322         free_cpumask_var(cpus_hardware_enabled);
2323         __free_page(hwpoison_page);
2324         __free_page(bad_page);
2325 }
2326 EXPORT_SYMBOL_GPL(kvm_exit);