KVM: MMU: rewrite audit_mappings_page() function
[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 static struct page *hwpoison_page;
97 static pfn_t hwpoison_pfn;
98
99 static struct page *fault_page;
100 static pfn_t fault_pfn;
101
102 inline int kvm_is_mmio_pfn(pfn_t pfn)
103 {
104         if (pfn_valid(pfn)) {
105                 struct page *page = compound_head(pfn_to_page(pfn));
106                 return PageReserved(page);
107         }
108
109         return true;
110 }
111
112 /*
113  * Switches to specified vcpu, until a matching vcpu_put()
114  */
115 void vcpu_load(struct kvm_vcpu *vcpu)
116 {
117         int cpu;
118
119         mutex_lock(&vcpu->mutex);
120         cpu = get_cpu();
121         preempt_notifier_register(&vcpu->preempt_notifier);
122         kvm_arch_vcpu_load(vcpu, cpu);
123         put_cpu();
124 }
125
126 void vcpu_put(struct kvm_vcpu *vcpu)
127 {
128         preempt_disable();
129         kvm_arch_vcpu_put(vcpu);
130         preempt_notifier_unregister(&vcpu->preempt_notifier);
131         preempt_enable();
132         mutex_unlock(&vcpu->mutex);
133 }
134
135 static void ack_flush(void *_completed)
136 {
137 }
138
139 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
140 {
141         int i, cpu, me;
142         cpumask_var_t cpus;
143         bool called = true;
144         struct kvm_vcpu *vcpu;
145
146         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
147
148         raw_spin_lock(&kvm->requests_lock);
149         me = smp_processor_id();
150         kvm_for_each_vcpu(i, vcpu, kvm) {
151                 if (kvm_make_check_request(req, vcpu))
152                         continue;
153                 cpu = vcpu->cpu;
154                 if (cpus != NULL && cpu != -1 && cpu != me)
155                         cpumask_set_cpu(cpu, cpus);
156         }
157         if (unlikely(cpus == NULL))
158                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
159         else if (!cpumask_empty(cpus))
160                 smp_call_function_many(cpus, ack_flush, NULL, 1);
161         else
162                 called = false;
163         raw_spin_unlock(&kvm->requests_lock);
164         free_cpumask_var(cpus);
165         return called;
166 }
167
168 void kvm_flush_remote_tlbs(struct kvm *kvm)
169 {
170         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
171                 ++kvm->stat.remote_tlb_flush;
172 }
173
174 void kvm_reload_remote_mmus(struct kvm *kvm)
175 {
176         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
177 }
178
179 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
180 {
181         struct page *page;
182         int r;
183
184         mutex_init(&vcpu->mutex);
185         vcpu->cpu = -1;
186         vcpu->kvm = kvm;
187         vcpu->vcpu_id = id;
188         init_waitqueue_head(&vcpu->wq);
189
190         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
191         if (!page) {
192                 r = -ENOMEM;
193                 goto fail;
194         }
195         vcpu->run = page_address(page);
196
197         r = kvm_arch_vcpu_init(vcpu);
198         if (r < 0)
199                 goto fail_free_run;
200         return 0;
201
202 fail_free_run:
203         free_page((unsigned long)vcpu->run);
204 fail:
205         return r;
206 }
207 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
208
209 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
210 {
211         kvm_arch_vcpu_uninit(vcpu);
212         free_page((unsigned long)vcpu->run);
213 }
214 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
215
216 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
217 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
218 {
219         return container_of(mn, struct kvm, mmu_notifier);
220 }
221
222 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
223                                              struct mm_struct *mm,
224                                              unsigned long address)
225 {
226         struct kvm *kvm = mmu_notifier_to_kvm(mn);
227         int need_tlb_flush, idx;
228
229         /*
230          * When ->invalidate_page runs, the linux pte has been zapped
231          * already but the page is still allocated until
232          * ->invalidate_page returns. So if we increase the sequence
233          * here the kvm page fault will notice if the spte can't be
234          * established because the page is going to be freed. If
235          * instead the kvm page fault establishes the spte before
236          * ->invalidate_page runs, kvm_unmap_hva will release it
237          * before returning.
238          *
239          * The sequence increase only need to be seen at spin_unlock
240          * time, and not at spin_lock time.
241          *
242          * Increasing the sequence after the spin_unlock would be
243          * unsafe because the kvm page fault could then establish the
244          * pte after kvm_unmap_hva returned, without noticing the page
245          * is going to be freed.
246          */
247         idx = srcu_read_lock(&kvm->srcu);
248         spin_lock(&kvm->mmu_lock);
249         kvm->mmu_notifier_seq++;
250         need_tlb_flush = kvm_unmap_hva(kvm, address);
251         spin_unlock(&kvm->mmu_lock);
252         srcu_read_unlock(&kvm->srcu, idx);
253
254         /* we've to flush the tlb before the pages can be freed */
255         if (need_tlb_flush)
256                 kvm_flush_remote_tlbs(kvm);
257
258 }
259
260 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
261                                         struct mm_struct *mm,
262                                         unsigned long address,
263                                         pte_t pte)
264 {
265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
266         int idx;
267
268         idx = srcu_read_lock(&kvm->srcu);
269         spin_lock(&kvm->mmu_lock);
270         kvm->mmu_notifier_seq++;
271         kvm_set_spte_hva(kvm, address, pte);
272         spin_unlock(&kvm->mmu_lock);
273         srcu_read_unlock(&kvm->srcu, idx);
274 }
275
276 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
277                                                     struct mm_struct *mm,
278                                                     unsigned long start,
279                                                     unsigned long end)
280 {
281         struct kvm *kvm = mmu_notifier_to_kvm(mn);
282         int need_tlb_flush = 0, idx;
283
284         idx = srcu_read_lock(&kvm->srcu);
285         spin_lock(&kvm->mmu_lock);
286         /*
287          * The count increase must become visible at unlock time as no
288          * spte can be established without taking the mmu_lock and
289          * count is also read inside the mmu_lock critical section.
290          */
291         kvm->mmu_notifier_count++;
292         for (; start < end; start += PAGE_SIZE)
293                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
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 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
303                                                   struct mm_struct *mm,
304                                                   unsigned long start,
305                                                   unsigned long end)
306 {
307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
308
309         spin_lock(&kvm->mmu_lock);
310         /*
311          * This sequence increase will notify the kvm page fault that
312          * the page that is going to be mapped in the spte could have
313          * been freed.
314          */
315         kvm->mmu_notifier_seq++;
316         /*
317          * The above sequence increase must be visible before the
318          * below count decrease but both values are read by the kvm
319          * page fault under mmu_lock spinlock so we don't need to add
320          * a smb_wmb() here in between the two.
321          */
322         kvm->mmu_notifier_count--;
323         spin_unlock(&kvm->mmu_lock);
324
325         BUG_ON(kvm->mmu_notifier_count < 0);
326 }
327
328 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
329                                               struct mm_struct *mm,
330                                               unsigned long address)
331 {
332         struct kvm *kvm = mmu_notifier_to_kvm(mn);
333         int young, idx;
334
335         idx = srcu_read_lock(&kvm->srcu);
336         spin_lock(&kvm->mmu_lock);
337         young = kvm_age_hva(kvm, address);
338         spin_unlock(&kvm->mmu_lock);
339         srcu_read_unlock(&kvm->srcu, idx);
340
341         if (young)
342                 kvm_flush_remote_tlbs(kvm);
343
344         return young;
345 }
346
347 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
348                                      struct mm_struct *mm)
349 {
350         struct kvm *kvm = mmu_notifier_to_kvm(mn);
351         int idx;
352
353         idx = srcu_read_lock(&kvm->srcu);
354         kvm_arch_flush_shadow(kvm);
355         srcu_read_unlock(&kvm->srcu, idx);
356 }
357
358 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
359         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
360         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
361         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
362         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
363         .change_pte             = kvm_mmu_notifier_change_pte,
364         .release                = kvm_mmu_notifier_release,
365 };
366
367 static int kvm_init_mmu_notifier(struct kvm *kvm)
368 {
369         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
370         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
371 }
372
373 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
374
375 static int kvm_init_mmu_notifier(struct kvm *kvm)
376 {
377         return 0;
378 }
379
380 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
381
382 static struct kvm *kvm_create_vm(void)
383 {
384         int r = 0, i;
385         struct kvm *kvm = kvm_arch_create_vm();
386
387         if (IS_ERR(kvm))
388                 goto out;
389
390         r = hardware_enable_all();
391         if (r)
392                 goto out_err_nodisable;
393
394 #ifdef CONFIG_HAVE_KVM_IRQCHIP
395         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
396         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
397 #endif
398
399         r = -ENOMEM;
400         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
401         if (!kvm->memslots)
402                 goto out_err;
403         if (init_srcu_struct(&kvm->srcu))
404                 goto out_err;
405         for (i = 0; i < KVM_NR_BUSES; i++) {
406                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
407                                         GFP_KERNEL);
408                 if (!kvm->buses[i]) {
409                         cleanup_srcu_struct(&kvm->srcu);
410                         goto out_err;
411                 }
412         }
413
414         r = kvm_init_mmu_notifier(kvm);
415         if (r) {
416                 cleanup_srcu_struct(&kvm->srcu);
417                 goto out_err;
418         }
419
420         kvm->mm = current->mm;
421         atomic_inc(&kvm->mm->mm_count);
422         spin_lock_init(&kvm->mmu_lock);
423         raw_spin_lock_init(&kvm->requests_lock);
424         kvm_eventfd_init(kvm);
425         mutex_init(&kvm->lock);
426         mutex_init(&kvm->irq_lock);
427         mutex_init(&kvm->slots_lock);
428         atomic_set(&kvm->users_count, 1);
429         spin_lock(&kvm_lock);
430         list_add(&kvm->vm_list, &vm_list);
431         spin_unlock(&kvm_lock);
432 out:
433         return kvm;
434
435 out_err:
436         hardware_disable_all();
437 out_err_nodisable:
438         for (i = 0; i < KVM_NR_BUSES; i++)
439                 kfree(kvm->buses[i]);
440         kfree(kvm->memslots);
441         kfree(kvm);
442         return ERR_PTR(r);
443 }
444
445 /*
446  * Free any memory in @free but not in @dont.
447  */
448 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
449                                   struct kvm_memory_slot *dont)
450 {
451         int i;
452
453         if (!dont || free->rmap != dont->rmap)
454                 vfree(free->rmap);
455
456         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
457                 vfree(free->dirty_bitmap);
458
459
460         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
461                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
462                         vfree(free->lpage_info[i]);
463                         free->lpage_info[i] = NULL;
464                 }
465         }
466
467         free->npages = 0;
468         free->dirty_bitmap = NULL;
469         free->rmap = NULL;
470 }
471
472 void kvm_free_physmem(struct kvm *kvm)
473 {
474         int i;
475         struct kvm_memslots *slots = kvm->memslots;
476
477         for (i = 0; i < slots->nmemslots; ++i)
478                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
479
480         kfree(kvm->memslots);
481 }
482
483 static void kvm_destroy_vm(struct kvm *kvm)
484 {
485         int i;
486         struct mm_struct *mm = kvm->mm;
487
488         kvm_arch_sync_events(kvm);
489         spin_lock(&kvm_lock);
490         list_del(&kvm->vm_list);
491         spin_unlock(&kvm_lock);
492         kvm_free_irq_routing(kvm);
493         for (i = 0; i < KVM_NR_BUSES; i++)
494                 kvm_io_bus_destroy(kvm->buses[i]);
495         kvm_coalesced_mmio_free(kvm);
496 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
497         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
498 #else
499         kvm_arch_flush_shadow(kvm);
500 #endif
501         kvm_arch_destroy_vm(kvm);
502         hardware_disable_all();
503         mmdrop(mm);
504 }
505
506 void kvm_get_kvm(struct kvm *kvm)
507 {
508         atomic_inc(&kvm->users_count);
509 }
510 EXPORT_SYMBOL_GPL(kvm_get_kvm);
511
512 void kvm_put_kvm(struct kvm *kvm)
513 {
514         if (atomic_dec_and_test(&kvm->users_count))
515                 kvm_destroy_vm(kvm);
516 }
517 EXPORT_SYMBOL_GPL(kvm_put_kvm);
518
519
520 static int kvm_vm_release(struct inode *inode, struct file *filp)
521 {
522         struct kvm *kvm = filp->private_data;
523
524         kvm_irqfd_release(kvm);
525
526         kvm_put_kvm(kvm);
527         return 0;
528 }
529
530 /*
531  * Allocate some memory and give it an address in the guest physical address
532  * space.
533  *
534  * Discontiguous memory is allowed, mostly for framebuffers.
535  *
536  * Must be called holding mmap_sem for write.
537  */
538 int __kvm_set_memory_region(struct kvm *kvm,
539                             struct kvm_userspace_memory_region *mem,
540                             int user_alloc)
541 {
542         int r, flush_shadow = 0;
543         gfn_t base_gfn;
544         unsigned long npages;
545         unsigned long i;
546         struct kvm_memory_slot *memslot;
547         struct kvm_memory_slot old, new;
548         struct kvm_memslots *slots, *old_memslots;
549
550         r = -EINVAL;
551         /* General sanity checks */
552         if (mem->memory_size & (PAGE_SIZE - 1))
553                 goto out;
554         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
555                 goto out;
556         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
557                 goto out;
558         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
559                 goto out;
560         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
561                 goto out;
562
563         memslot = &kvm->memslots->memslots[mem->slot];
564         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
565         npages = mem->memory_size >> PAGE_SHIFT;
566
567         r = -EINVAL;
568         if (npages > KVM_MEM_MAX_NR_PAGES)
569                 goto out;
570
571         if (!npages)
572                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
573
574         new = old = *memslot;
575
576         new.id = mem->slot;
577         new.base_gfn = base_gfn;
578         new.npages = npages;
579         new.flags = mem->flags;
580
581         /* Disallow changing a memory slot's size. */
582         r = -EINVAL;
583         if (npages && old.npages && npages != old.npages)
584                 goto out_free;
585
586         /* Check for overlaps */
587         r = -EEXIST;
588         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
589                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
590
591                 if (s == memslot || !s->npages)
592                         continue;
593                 if (!((base_gfn + npages <= s->base_gfn) ||
594                       (base_gfn >= s->base_gfn + s->npages)))
595                         goto out_free;
596         }
597
598         /* Free page dirty bitmap if unneeded */
599         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
600                 new.dirty_bitmap = NULL;
601
602         r = -ENOMEM;
603
604         /* Allocate if a slot is being created */
605 #ifndef CONFIG_S390
606         if (npages && !new.rmap) {
607                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
608
609                 if (!new.rmap)
610                         goto out_free;
611
612                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
613
614                 new.user_alloc = user_alloc;
615                 new.userspace_addr = mem->userspace_addr;
616         }
617         if (!npages)
618                 goto skip_lpage;
619
620         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
621                 unsigned long ugfn;
622                 unsigned long j;
623                 int lpages;
624                 int level = i + 2;
625
626                 /* Avoid unused variable warning if no large pages */
627                 (void)level;
628
629                 if (new.lpage_info[i])
630                         continue;
631
632                 lpages = 1 + ((base_gfn + npages - 1)
633                              >> KVM_HPAGE_GFN_SHIFT(level));
634                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
635
636                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
637
638                 if (!new.lpage_info[i])
639                         goto out_free;
640
641                 memset(new.lpage_info[i], 0,
642                        lpages * sizeof(*new.lpage_info[i]));
643
644                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
645                         new.lpage_info[i][0].write_count = 1;
646                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
647                         new.lpage_info[i][lpages - 1].write_count = 1;
648                 ugfn = new.userspace_addr >> PAGE_SHIFT;
649                 /*
650                  * If the gfn and userspace address are not aligned wrt each
651                  * other, or if explicitly asked to, disable large page
652                  * support for this slot
653                  */
654                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
655                     !largepages_enabled)
656                         for (j = 0; j < lpages; ++j)
657                                 new.lpage_info[i][j].write_count = 1;
658         }
659
660 skip_lpage:
661
662         /* Allocate page dirty bitmap if needed */
663         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
664                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
665
666                 new.dirty_bitmap = vmalloc(dirty_bytes);
667                 if (!new.dirty_bitmap)
668                         goto out_free;
669                 memset(new.dirty_bitmap, 0, dirty_bytes);
670                 /* destroy any largepage mappings for dirty tracking */
671                 if (old.npages)
672                         flush_shadow = 1;
673         }
674 #else  /* not defined CONFIG_S390 */
675         new.user_alloc = user_alloc;
676         if (user_alloc)
677                 new.userspace_addr = mem->userspace_addr;
678 #endif /* not defined CONFIG_S390 */
679
680         if (!npages) {
681                 r = -ENOMEM;
682                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
683                 if (!slots)
684                         goto out_free;
685                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
686                 if (mem->slot >= slots->nmemslots)
687                         slots->nmemslots = mem->slot + 1;
688                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
689
690                 old_memslots = kvm->memslots;
691                 rcu_assign_pointer(kvm->memslots, slots);
692                 synchronize_srcu_expedited(&kvm->srcu);
693                 /* From this point no new shadow pages pointing to a deleted
694                  * memslot will be created.
695                  *
696                  * validation of sp->gfn happens in:
697                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
698                  *      - kvm_is_visible_gfn (mmu_check_roots)
699                  */
700                 kvm_arch_flush_shadow(kvm);
701                 kfree(old_memslots);
702         }
703
704         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
705         if (r)
706                 goto out_free;
707
708 #ifdef CONFIG_DMAR
709         /* map the pages in iommu page table */
710         if (npages) {
711                 r = kvm_iommu_map_pages(kvm, &new);
712                 if (r)
713                         goto out_free;
714         }
715 #endif
716
717         r = -ENOMEM;
718         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
719         if (!slots)
720                 goto out_free;
721         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
722         if (mem->slot >= slots->nmemslots)
723                 slots->nmemslots = mem->slot + 1;
724
725         /* actual memory is freed via old in kvm_free_physmem_slot below */
726         if (!npages) {
727                 new.rmap = NULL;
728                 new.dirty_bitmap = NULL;
729                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
730                         new.lpage_info[i] = NULL;
731         }
732
733         slots->memslots[mem->slot] = new;
734         old_memslots = kvm->memslots;
735         rcu_assign_pointer(kvm->memslots, slots);
736         synchronize_srcu_expedited(&kvm->srcu);
737
738         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
739
740         kvm_free_physmem_slot(&old, &new);
741         kfree(old_memslots);
742
743         if (flush_shadow)
744                 kvm_arch_flush_shadow(kvm);
745
746         return 0;
747
748 out_free:
749         kvm_free_physmem_slot(&new, &old);
750 out:
751         return r;
752
753 }
754 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
755
756 int kvm_set_memory_region(struct kvm *kvm,
757                           struct kvm_userspace_memory_region *mem,
758                           int user_alloc)
759 {
760         int r;
761
762         mutex_lock(&kvm->slots_lock);
763         r = __kvm_set_memory_region(kvm, mem, user_alloc);
764         mutex_unlock(&kvm->slots_lock);
765         return r;
766 }
767 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
768
769 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
770                                    struct
771                                    kvm_userspace_memory_region *mem,
772                                    int user_alloc)
773 {
774         if (mem->slot >= KVM_MEMORY_SLOTS)
775                 return -EINVAL;
776         return kvm_set_memory_region(kvm, mem, user_alloc);
777 }
778
779 int kvm_get_dirty_log(struct kvm *kvm,
780                         struct kvm_dirty_log *log, int *is_dirty)
781 {
782         struct kvm_memory_slot *memslot;
783         int r, i;
784         unsigned long n;
785         unsigned long any = 0;
786
787         r = -EINVAL;
788         if (log->slot >= KVM_MEMORY_SLOTS)
789                 goto out;
790
791         memslot = &kvm->memslots->memslots[log->slot];
792         r = -ENOENT;
793         if (!memslot->dirty_bitmap)
794                 goto out;
795
796         n = kvm_dirty_bitmap_bytes(memslot);
797
798         for (i = 0; !any && i < n/sizeof(long); ++i)
799                 any = memslot->dirty_bitmap[i];
800
801         r = -EFAULT;
802         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
803                 goto out;
804
805         if (any)
806                 *is_dirty = 1;
807
808         r = 0;
809 out:
810         return r;
811 }
812
813 void kvm_disable_largepages(void)
814 {
815         largepages_enabled = false;
816 }
817 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
818
819 int is_error_page(struct page *page)
820 {
821         return page == bad_page || page == hwpoison_page || page == fault_page;
822 }
823 EXPORT_SYMBOL_GPL(is_error_page);
824
825 int is_error_pfn(pfn_t pfn)
826 {
827         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
828 }
829 EXPORT_SYMBOL_GPL(is_error_pfn);
830
831 int is_hwpoison_pfn(pfn_t pfn)
832 {
833         return pfn == hwpoison_pfn;
834 }
835 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
836
837 int is_fault_pfn(pfn_t pfn)
838 {
839         return pfn == fault_pfn;
840 }
841 EXPORT_SYMBOL_GPL(is_fault_pfn);
842
843 static inline unsigned long bad_hva(void)
844 {
845         return PAGE_OFFSET;
846 }
847
848 int kvm_is_error_hva(unsigned long addr)
849 {
850         return addr == bad_hva();
851 }
852 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
853
854 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
855 {
856         int i;
857         struct kvm_memslots *slots = kvm_memslots(kvm);
858
859         for (i = 0; i < slots->nmemslots; ++i) {
860                 struct kvm_memory_slot *memslot = &slots->memslots[i];
861
862                 if (gfn >= memslot->base_gfn
863                     && gfn < memslot->base_gfn + memslot->npages)
864                         return memslot;
865         }
866         return NULL;
867 }
868 EXPORT_SYMBOL_GPL(gfn_to_memslot);
869
870 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
871 {
872         int i;
873         struct kvm_memslots *slots = kvm_memslots(kvm);
874
875         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
876                 struct kvm_memory_slot *memslot = &slots->memslots[i];
877
878                 if (memslot->flags & KVM_MEMSLOT_INVALID)
879                         continue;
880
881                 if (gfn >= memslot->base_gfn
882                     && gfn < memslot->base_gfn + memslot->npages)
883                         return 1;
884         }
885         return 0;
886 }
887 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
888
889 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
890 {
891         struct vm_area_struct *vma;
892         unsigned long addr, size;
893
894         size = PAGE_SIZE;
895
896         addr = gfn_to_hva(kvm, gfn);
897         if (kvm_is_error_hva(addr))
898                 return PAGE_SIZE;
899
900         down_read(&current->mm->mmap_sem);
901         vma = find_vma(current->mm, addr);
902         if (!vma)
903                 goto out;
904
905         size = vma_kernel_pagesize(vma);
906
907 out:
908         up_read(&current->mm->mmap_sem);
909
910         return size;
911 }
912
913 int memslot_id(struct kvm *kvm, gfn_t gfn)
914 {
915         int i;
916         struct kvm_memslots *slots = kvm_memslots(kvm);
917         struct kvm_memory_slot *memslot = NULL;
918
919         for (i = 0; i < slots->nmemslots; ++i) {
920                 memslot = &slots->memslots[i];
921
922                 if (gfn >= memslot->base_gfn
923                     && gfn < memslot->base_gfn + memslot->npages)
924                         break;
925         }
926
927         return memslot - slots->memslots;
928 }
929
930 static unsigned long gfn_to_hva_many(struct kvm *kvm, gfn_t gfn,
931                                      gfn_t *nr_pages)
932 {
933         struct kvm_memory_slot *slot;
934
935         slot = gfn_to_memslot(kvm, gfn);
936         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
937                 return bad_hva();
938
939         if (nr_pages)
940                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
941
942         return gfn_to_hva_memslot(slot, gfn);
943 }
944
945 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
946 {
947         return gfn_to_hva_many(kvm, gfn, NULL);
948 }
949 EXPORT_SYMBOL_GPL(gfn_to_hva);
950
951 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic)
952 {
953         struct page *page[1];
954         int npages;
955         pfn_t pfn;
956
957         if (atomic)
958                 npages = __get_user_pages_fast(addr, 1, 1, page);
959         else {
960                 might_sleep();
961                 npages = get_user_pages_fast(addr, 1, 1, page);
962         }
963
964         if (unlikely(npages != 1)) {
965                 struct vm_area_struct *vma;
966
967                 if (atomic)
968                         goto return_fault_page;
969
970                 down_read(&current->mm->mmap_sem);
971                 if (is_hwpoison_address(addr)) {
972                         up_read(&current->mm->mmap_sem);
973                         get_page(hwpoison_page);
974                         return page_to_pfn(hwpoison_page);
975                 }
976
977                 vma = find_vma(current->mm, addr);
978
979                 if (vma == NULL || addr < vma->vm_start ||
980                     !(vma->vm_flags & VM_PFNMAP)) {
981                         up_read(&current->mm->mmap_sem);
982 return_fault_page:
983                         get_page(fault_page);
984                         return page_to_pfn(fault_page);
985                 }
986
987                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
988                 up_read(&current->mm->mmap_sem);
989                 BUG_ON(!kvm_is_mmio_pfn(pfn));
990         } else
991                 pfn = page_to_pfn(page[0]);
992
993         return pfn;
994 }
995
996 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
997 {
998         return hva_to_pfn(kvm, addr, true);
999 }
1000 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1001
1002 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic)
1003 {
1004         unsigned long addr;
1005
1006         addr = gfn_to_hva(kvm, gfn);
1007         if (kvm_is_error_hva(addr)) {
1008                 get_page(bad_page);
1009                 return page_to_pfn(bad_page);
1010         }
1011
1012         return hva_to_pfn(kvm, addr, atomic);
1013 }
1014
1015 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1016 {
1017         return __gfn_to_pfn(kvm, gfn, true);
1018 }
1019 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1020
1021 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1022 {
1023         return __gfn_to_pfn(kvm, gfn, false);
1024 }
1025 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1026
1027 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1028                          struct kvm_memory_slot *slot, gfn_t gfn)
1029 {
1030         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1031         return hva_to_pfn(kvm, addr, false);
1032 }
1033
1034 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1035                                                                   int nr_pages)
1036 {
1037         unsigned long addr;
1038         gfn_t entry;
1039
1040         addr = gfn_to_hva_many(kvm, gfn, &entry);
1041         if (kvm_is_error_hva(addr))
1042                 return -1;
1043
1044         if (entry < nr_pages)
1045                 return 0;
1046
1047         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1048 }
1049 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1050
1051 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1052 {
1053         pfn_t pfn;
1054
1055         pfn = gfn_to_pfn(kvm, gfn);
1056         if (!kvm_is_mmio_pfn(pfn))
1057                 return pfn_to_page(pfn);
1058
1059         WARN_ON(kvm_is_mmio_pfn(pfn));
1060
1061         get_page(bad_page);
1062         return bad_page;
1063 }
1064
1065 EXPORT_SYMBOL_GPL(gfn_to_page);
1066
1067 void kvm_release_page_clean(struct page *page)
1068 {
1069         kvm_release_pfn_clean(page_to_pfn(page));
1070 }
1071 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1072
1073 void kvm_release_pfn_clean(pfn_t pfn)
1074 {
1075         if (!kvm_is_mmio_pfn(pfn))
1076                 put_page(pfn_to_page(pfn));
1077 }
1078 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1079
1080 void kvm_release_page_dirty(struct page *page)
1081 {
1082         kvm_release_pfn_dirty(page_to_pfn(page));
1083 }
1084 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1085
1086 void kvm_release_pfn_dirty(pfn_t pfn)
1087 {
1088         kvm_set_pfn_dirty(pfn);
1089         kvm_release_pfn_clean(pfn);
1090 }
1091 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1092
1093 void kvm_set_page_dirty(struct page *page)
1094 {
1095         kvm_set_pfn_dirty(page_to_pfn(page));
1096 }
1097 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1098
1099 void kvm_set_pfn_dirty(pfn_t pfn)
1100 {
1101         if (!kvm_is_mmio_pfn(pfn)) {
1102                 struct page *page = pfn_to_page(pfn);
1103                 if (!PageReserved(page))
1104                         SetPageDirty(page);
1105         }
1106 }
1107 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1108
1109 void kvm_set_pfn_accessed(pfn_t pfn)
1110 {
1111         if (!kvm_is_mmio_pfn(pfn))
1112                 mark_page_accessed(pfn_to_page(pfn));
1113 }
1114 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1115
1116 void kvm_get_pfn(pfn_t pfn)
1117 {
1118         if (!kvm_is_mmio_pfn(pfn))
1119                 get_page(pfn_to_page(pfn));
1120 }
1121 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1122
1123 static int next_segment(unsigned long len, int offset)
1124 {
1125         if (len > PAGE_SIZE - offset)
1126                 return PAGE_SIZE - offset;
1127         else
1128                 return len;
1129 }
1130
1131 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1132                         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_from_user(data, (void __user *)addr + offset, len);
1141         if (r)
1142                 return -EFAULT;
1143         return 0;
1144 }
1145 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1146
1147 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1148 {
1149         gfn_t gfn = gpa >> PAGE_SHIFT;
1150         int seg;
1151         int offset = offset_in_page(gpa);
1152         int ret;
1153
1154         while ((seg = next_segment(len, offset)) != 0) {
1155                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1156                 if (ret < 0)
1157                         return ret;
1158                 offset = 0;
1159                 len -= seg;
1160                 data += seg;
1161                 ++gfn;
1162         }
1163         return 0;
1164 }
1165 EXPORT_SYMBOL_GPL(kvm_read_guest);
1166
1167 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1168                           unsigned long len)
1169 {
1170         int r;
1171         unsigned long addr;
1172         gfn_t gfn = gpa >> PAGE_SHIFT;
1173         int offset = offset_in_page(gpa);
1174
1175         addr = gfn_to_hva(kvm, gfn);
1176         if (kvm_is_error_hva(addr))
1177                 return -EFAULT;
1178         pagefault_disable();
1179         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1180         pagefault_enable();
1181         if (r)
1182                 return -EFAULT;
1183         return 0;
1184 }
1185 EXPORT_SYMBOL(kvm_read_guest_atomic);
1186
1187 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1188                          int offset, int len)
1189 {
1190         int r;
1191         unsigned long addr;
1192
1193         addr = gfn_to_hva(kvm, gfn);
1194         if (kvm_is_error_hva(addr))
1195                 return -EFAULT;
1196         r = copy_to_user((void __user *)addr + offset, data, len);
1197         if (r)
1198                 return -EFAULT;
1199         mark_page_dirty(kvm, gfn);
1200         return 0;
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1203
1204 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1205                     unsigned long len)
1206 {
1207         gfn_t gfn = gpa >> PAGE_SHIFT;
1208         int seg;
1209         int offset = offset_in_page(gpa);
1210         int ret;
1211
1212         while ((seg = next_segment(len, offset)) != 0) {
1213                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1214                 if (ret < 0)
1215                         return ret;
1216                 offset = 0;
1217                 len -= seg;
1218                 data += seg;
1219                 ++gfn;
1220         }
1221         return 0;
1222 }
1223
1224 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1225 {
1226         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1227 }
1228 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1229
1230 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1231 {
1232         gfn_t gfn = gpa >> PAGE_SHIFT;
1233         int seg;
1234         int offset = offset_in_page(gpa);
1235         int ret;
1236
1237         while ((seg = next_segment(len, offset)) != 0) {
1238                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1239                 if (ret < 0)
1240                         return ret;
1241                 offset = 0;
1242                 len -= seg;
1243                 ++gfn;
1244         }
1245         return 0;
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1248
1249 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1250 {
1251         struct kvm_memory_slot *memslot;
1252
1253         memslot = gfn_to_memslot(kvm, gfn);
1254         if (memslot && memslot->dirty_bitmap) {
1255                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1256
1257                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1258         }
1259 }
1260
1261 /*
1262  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1263  */
1264 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1265 {
1266         DEFINE_WAIT(wait);
1267
1268         for (;;) {
1269                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1270
1271                 if (kvm_arch_vcpu_runnable(vcpu)) {
1272                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1273                         break;
1274                 }
1275                 if (kvm_cpu_has_pending_timer(vcpu))
1276                         break;
1277                 if (signal_pending(current))
1278                         break;
1279
1280                 schedule();
1281         }
1282
1283         finish_wait(&vcpu->wq, &wait);
1284 }
1285
1286 void kvm_resched(struct kvm_vcpu *vcpu)
1287 {
1288         if (!need_resched())
1289                 return;
1290         cond_resched();
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_resched);
1293
1294 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1295 {
1296         ktime_t expires;
1297         DEFINE_WAIT(wait);
1298
1299         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1300
1301         /* Sleep for 100 us, and hope lock-holder got scheduled */
1302         expires = ktime_add_ns(ktime_get(), 100000UL);
1303         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1304
1305         finish_wait(&vcpu->wq, &wait);
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1308
1309 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1310 {
1311         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1312         struct page *page;
1313
1314         if (vmf->pgoff == 0)
1315                 page = virt_to_page(vcpu->run);
1316 #ifdef CONFIG_X86
1317         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1318                 page = virt_to_page(vcpu->arch.pio_data);
1319 #endif
1320 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1321         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1322                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1323 #endif
1324         else
1325                 return VM_FAULT_SIGBUS;
1326         get_page(page);
1327         vmf->page = page;
1328         return 0;
1329 }
1330
1331 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1332         .fault = kvm_vcpu_fault,
1333 };
1334
1335 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1336 {
1337         vma->vm_ops = &kvm_vcpu_vm_ops;
1338         return 0;
1339 }
1340
1341 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1342 {
1343         struct kvm_vcpu *vcpu = filp->private_data;
1344
1345         kvm_put_kvm(vcpu->kvm);
1346         return 0;
1347 }
1348
1349 static struct file_operations kvm_vcpu_fops = {
1350         .release        = kvm_vcpu_release,
1351         .unlocked_ioctl = kvm_vcpu_ioctl,
1352         .compat_ioctl   = kvm_vcpu_ioctl,
1353         .mmap           = kvm_vcpu_mmap,
1354 };
1355
1356 /*
1357  * Allocates an inode for the vcpu.
1358  */
1359 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1360 {
1361         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1362 }
1363
1364 /*
1365  * Creates some virtual cpus.  Good luck creating more than one.
1366  */
1367 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1368 {
1369         int r;
1370         struct kvm_vcpu *vcpu, *v;
1371
1372         vcpu = kvm_arch_vcpu_create(kvm, id);
1373         if (IS_ERR(vcpu))
1374                 return PTR_ERR(vcpu);
1375
1376         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1377
1378         r = kvm_arch_vcpu_setup(vcpu);
1379         if (r)
1380                 return r;
1381
1382         mutex_lock(&kvm->lock);
1383         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1384                 r = -EINVAL;
1385                 goto vcpu_destroy;
1386         }
1387
1388         kvm_for_each_vcpu(r, v, kvm)
1389                 if (v->vcpu_id == id) {
1390                         r = -EEXIST;
1391                         goto vcpu_destroy;
1392                 }
1393
1394         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1395
1396         /* Now it's all set up, let userspace reach it */
1397         kvm_get_kvm(kvm);
1398         r = create_vcpu_fd(vcpu);
1399         if (r < 0) {
1400                 kvm_put_kvm(kvm);
1401                 goto vcpu_destroy;
1402         }
1403
1404         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1405         smp_wmb();
1406         atomic_inc(&kvm->online_vcpus);
1407
1408 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1409         if (kvm->bsp_vcpu_id == id)
1410                 kvm->bsp_vcpu = vcpu;
1411 #endif
1412         mutex_unlock(&kvm->lock);
1413         return r;
1414
1415 vcpu_destroy:
1416         mutex_unlock(&kvm->lock);
1417         kvm_arch_vcpu_destroy(vcpu);
1418         return r;
1419 }
1420
1421 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1422 {
1423         if (sigset) {
1424                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1425                 vcpu->sigset_active = 1;
1426                 vcpu->sigset = *sigset;
1427         } else
1428                 vcpu->sigset_active = 0;
1429         return 0;
1430 }
1431
1432 static long kvm_vcpu_ioctl(struct file *filp,
1433                            unsigned int ioctl, unsigned long arg)
1434 {
1435         struct kvm_vcpu *vcpu = filp->private_data;
1436         void __user *argp = (void __user *)arg;
1437         int r;
1438         struct kvm_fpu *fpu = NULL;
1439         struct kvm_sregs *kvm_sregs = NULL;
1440
1441         if (vcpu->kvm->mm != current->mm)
1442                 return -EIO;
1443
1444 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1445         /*
1446          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1447          * so vcpu_load() would break it.
1448          */
1449         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1450                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1451 #endif
1452
1453
1454         vcpu_load(vcpu);
1455         switch (ioctl) {
1456         case KVM_RUN:
1457                 r = -EINVAL;
1458                 if (arg)
1459                         goto out;
1460                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1461                 break;
1462         case KVM_GET_REGS: {
1463                 struct kvm_regs *kvm_regs;
1464
1465                 r = -ENOMEM;
1466                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1467                 if (!kvm_regs)
1468                         goto out;
1469                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1470                 if (r)
1471                         goto out_free1;
1472                 r = -EFAULT;
1473                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1474                         goto out_free1;
1475                 r = 0;
1476 out_free1:
1477                 kfree(kvm_regs);
1478                 break;
1479         }
1480         case KVM_SET_REGS: {
1481                 struct kvm_regs *kvm_regs;
1482
1483                 r = -ENOMEM;
1484                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1485                 if (!kvm_regs)
1486                         goto out;
1487                 r = -EFAULT;
1488                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1489                         goto out_free2;
1490                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1491                 if (r)
1492                         goto out_free2;
1493                 r = 0;
1494 out_free2:
1495                 kfree(kvm_regs);
1496                 break;
1497         }
1498         case KVM_GET_SREGS: {
1499                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1500                 r = -ENOMEM;
1501                 if (!kvm_sregs)
1502                         goto out;
1503                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1504                 if (r)
1505                         goto out;
1506                 r = -EFAULT;
1507                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1508                         goto out;
1509                 r = 0;
1510                 break;
1511         }
1512         case KVM_SET_SREGS: {
1513                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1514                 r = -ENOMEM;
1515                 if (!kvm_sregs)
1516                         goto out;
1517                 r = -EFAULT;
1518                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1519                         goto out;
1520                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1521                 if (r)
1522                         goto out;
1523                 r = 0;
1524                 break;
1525         }
1526         case KVM_GET_MP_STATE: {
1527                 struct kvm_mp_state mp_state;
1528
1529                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1530                 if (r)
1531                         goto out;
1532                 r = -EFAULT;
1533                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1534                         goto out;
1535                 r = 0;
1536                 break;
1537         }
1538         case KVM_SET_MP_STATE: {
1539                 struct kvm_mp_state mp_state;
1540
1541                 r = -EFAULT;
1542                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1543                         goto out;
1544                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1545                 if (r)
1546                         goto out;
1547                 r = 0;
1548                 break;
1549         }
1550         case KVM_TRANSLATE: {
1551                 struct kvm_translation tr;
1552
1553                 r = -EFAULT;
1554                 if (copy_from_user(&tr, argp, sizeof tr))
1555                         goto out;
1556                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1557                 if (r)
1558                         goto out;
1559                 r = -EFAULT;
1560                 if (copy_to_user(argp, &tr, sizeof tr))
1561                         goto out;
1562                 r = 0;
1563                 break;
1564         }
1565         case KVM_SET_GUEST_DEBUG: {
1566                 struct kvm_guest_debug dbg;
1567
1568                 r = -EFAULT;
1569                 if (copy_from_user(&dbg, argp, sizeof dbg))
1570                         goto out;
1571                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1572                 if (r)
1573                         goto out;
1574                 r = 0;
1575                 break;
1576         }
1577         case KVM_SET_SIGNAL_MASK: {
1578                 struct kvm_signal_mask __user *sigmask_arg = argp;
1579                 struct kvm_signal_mask kvm_sigmask;
1580                 sigset_t sigset, *p;
1581
1582                 p = NULL;
1583                 if (argp) {
1584                         r = -EFAULT;
1585                         if (copy_from_user(&kvm_sigmask, argp,
1586                                            sizeof kvm_sigmask))
1587                                 goto out;
1588                         r = -EINVAL;
1589                         if (kvm_sigmask.len != sizeof sigset)
1590                                 goto out;
1591                         r = -EFAULT;
1592                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1593                                            sizeof sigset))
1594                                 goto out;
1595                         p = &sigset;
1596                 }
1597                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1598                 break;
1599         }
1600         case KVM_GET_FPU: {
1601                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1602                 r = -ENOMEM;
1603                 if (!fpu)
1604                         goto out;
1605                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1606                 if (r)
1607                         goto out;
1608                 r = -EFAULT;
1609                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1610                         goto out;
1611                 r = 0;
1612                 break;
1613         }
1614         case KVM_SET_FPU: {
1615                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1616                 r = -ENOMEM;
1617                 if (!fpu)
1618                         goto out;
1619                 r = -EFAULT;
1620                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1621                         goto out;
1622                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1623                 if (r)
1624                         goto out;
1625                 r = 0;
1626                 break;
1627         }
1628         default:
1629                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1630         }
1631 out:
1632         vcpu_put(vcpu);
1633         kfree(fpu);
1634         kfree(kvm_sregs);
1635         return r;
1636 }
1637
1638 static long kvm_vm_ioctl(struct file *filp,
1639                            unsigned int ioctl, unsigned long arg)
1640 {
1641         struct kvm *kvm = filp->private_data;
1642         void __user *argp = (void __user *)arg;
1643         int r;
1644
1645         if (kvm->mm != current->mm)
1646                 return -EIO;
1647         switch (ioctl) {
1648         case KVM_CREATE_VCPU:
1649                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1650                 if (r < 0)
1651                         goto out;
1652                 break;
1653         case KVM_SET_USER_MEMORY_REGION: {
1654                 struct kvm_userspace_memory_region kvm_userspace_mem;
1655
1656                 r = -EFAULT;
1657                 if (copy_from_user(&kvm_userspace_mem, argp,
1658                                                 sizeof kvm_userspace_mem))
1659                         goto out;
1660
1661                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1662                 if (r)
1663                         goto out;
1664                 break;
1665         }
1666         case KVM_GET_DIRTY_LOG: {
1667                 struct kvm_dirty_log log;
1668
1669                 r = -EFAULT;
1670                 if (copy_from_user(&log, argp, sizeof log))
1671                         goto out;
1672                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1673                 if (r)
1674                         goto out;
1675                 break;
1676         }
1677 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1678         case KVM_REGISTER_COALESCED_MMIO: {
1679                 struct kvm_coalesced_mmio_zone zone;
1680                 r = -EFAULT;
1681                 if (copy_from_user(&zone, argp, sizeof zone))
1682                         goto out;
1683                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1684                 if (r)
1685                         goto out;
1686                 r = 0;
1687                 break;
1688         }
1689         case KVM_UNREGISTER_COALESCED_MMIO: {
1690                 struct kvm_coalesced_mmio_zone zone;
1691                 r = -EFAULT;
1692                 if (copy_from_user(&zone, argp, sizeof zone))
1693                         goto out;
1694                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1695                 if (r)
1696                         goto out;
1697                 r = 0;
1698                 break;
1699         }
1700 #endif
1701         case KVM_IRQFD: {
1702                 struct kvm_irqfd data;
1703
1704                 r = -EFAULT;
1705                 if (copy_from_user(&data, argp, sizeof data))
1706                         goto out;
1707                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1708                 break;
1709         }
1710         case KVM_IOEVENTFD: {
1711                 struct kvm_ioeventfd data;
1712
1713                 r = -EFAULT;
1714                 if (copy_from_user(&data, argp, sizeof data))
1715                         goto out;
1716                 r = kvm_ioeventfd(kvm, &data);
1717                 break;
1718         }
1719 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1720         case KVM_SET_BOOT_CPU_ID:
1721                 r = 0;
1722                 mutex_lock(&kvm->lock);
1723                 if (atomic_read(&kvm->online_vcpus) != 0)
1724                         r = -EBUSY;
1725                 else
1726                         kvm->bsp_vcpu_id = arg;
1727                 mutex_unlock(&kvm->lock);
1728                 break;
1729 #endif
1730         default:
1731                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1732                 if (r == -ENOTTY)
1733                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1734         }
1735 out:
1736         return r;
1737 }
1738
1739 #ifdef CONFIG_COMPAT
1740 struct compat_kvm_dirty_log {
1741         __u32 slot;
1742         __u32 padding1;
1743         union {
1744                 compat_uptr_t dirty_bitmap; /* one bit per page */
1745                 __u64 padding2;
1746         };
1747 };
1748
1749 static long kvm_vm_compat_ioctl(struct file *filp,
1750                            unsigned int ioctl, unsigned long arg)
1751 {
1752         struct kvm *kvm = filp->private_data;
1753         int r;
1754
1755         if (kvm->mm != current->mm)
1756                 return -EIO;
1757         switch (ioctl) {
1758         case KVM_GET_DIRTY_LOG: {
1759                 struct compat_kvm_dirty_log compat_log;
1760                 struct kvm_dirty_log log;
1761
1762                 r = -EFAULT;
1763                 if (copy_from_user(&compat_log, (void __user *)arg,
1764                                    sizeof(compat_log)))
1765                         goto out;
1766                 log.slot         = compat_log.slot;
1767                 log.padding1     = compat_log.padding1;
1768                 log.padding2     = compat_log.padding2;
1769                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1770
1771                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1772                 if (r)
1773                         goto out;
1774                 break;
1775         }
1776         default:
1777                 r = kvm_vm_ioctl(filp, ioctl, arg);
1778         }
1779
1780 out:
1781         return r;
1782 }
1783 #endif
1784
1785 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1786 {
1787         struct page *page[1];
1788         unsigned long addr;
1789         int npages;
1790         gfn_t gfn = vmf->pgoff;
1791         struct kvm *kvm = vma->vm_file->private_data;
1792
1793         addr = gfn_to_hva(kvm, gfn);
1794         if (kvm_is_error_hva(addr))
1795                 return VM_FAULT_SIGBUS;
1796
1797         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1798                                 NULL);
1799         if (unlikely(npages != 1))
1800                 return VM_FAULT_SIGBUS;
1801
1802         vmf->page = page[0];
1803         return 0;
1804 }
1805
1806 static const struct vm_operations_struct kvm_vm_vm_ops = {
1807         .fault = kvm_vm_fault,
1808 };
1809
1810 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1811 {
1812         vma->vm_ops = &kvm_vm_vm_ops;
1813         return 0;
1814 }
1815
1816 static struct file_operations kvm_vm_fops = {
1817         .release        = kvm_vm_release,
1818         .unlocked_ioctl = kvm_vm_ioctl,
1819 #ifdef CONFIG_COMPAT
1820         .compat_ioctl   = kvm_vm_compat_ioctl,
1821 #endif
1822         .mmap           = kvm_vm_mmap,
1823 };
1824
1825 static int kvm_dev_ioctl_create_vm(void)
1826 {
1827         int fd, r;
1828         struct kvm *kvm;
1829
1830         kvm = kvm_create_vm();
1831         if (IS_ERR(kvm))
1832                 return PTR_ERR(kvm);
1833 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1834         r = kvm_coalesced_mmio_init(kvm);
1835         if (r < 0) {
1836                 kvm_put_kvm(kvm);
1837                 return r;
1838         }
1839 #endif
1840         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1841         if (fd < 0)
1842                 kvm_put_kvm(kvm);
1843
1844         return fd;
1845 }
1846
1847 static long kvm_dev_ioctl_check_extension_generic(long arg)
1848 {
1849         switch (arg) {
1850         case KVM_CAP_USER_MEMORY:
1851         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1852         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1853 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1854         case KVM_CAP_SET_BOOT_CPU_ID:
1855 #endif
1856         case KVM_CAP_INTERNAL_ERROR_DATA:
1857                 return 1;
1858 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1859         case KVM_CAP_IRQ_ROUTING:
1860                 return KVM_MAX_IRQ_ROUTES;
1861 #endif
1862         default:
1863                 break;
1864         }
1865         return kvm_dev_ioctl_check_extension(arg);
1866 }
1867
1868 static long kvm_dev_ioctl(struct file *filp,
1869                           unsigned int ioctl, unsigned long arg)
1870 {
1871         long r = -EINVAL;
1872
1873         switch (ioctl) {
1874         case KVM_GET_API_VERSION:
1875                 r = -EINVAL;
1876                 if (arg)
1877                         goto out;
1878                 r = KVM_API_VERSION;
1879                 break;
1880         case KVM_CREATE_VM:
1881                 r = -EINVAL;
1882                 if (arg)
1883                         goto out;
1884                 r = kvm_dev_ioctl_create_vm();
1885                 break;
1886         case KVM_CHECK_EXTENSION:
1887                 r = kvm_dev_ioctl_check_extension_generic(arg);
1888                 break;
1889         case KVM_GET_VCPU_MMAP_SIZE:
1890                 r = -EINVAL;
1891                 if (arg)
1892                         goto out;
1893                 r = PAGE_SIZE;     /* struct kvm_run */
1894 #ifdef CONFIG_X86
1895                 r += PAGE_SIZE;    /* pio data page */
1896 #endif
1897 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1898                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1899 #endif
1900                 break;
1901         case KVM_TRACE_ENABLE:
1902         case KVM_TRACE_PAUSE:
1903         case KVM_TRACE_DISABLE:
1904                 r = -EOPNOTSUPP;
1905                 break;
1906         default:
1907                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1908         }
1909 out:
1910         return r;
1911 }
1912
1913 static struct file_operations kvm_chardev_ops = {
1914         .unlocked_ioctl = kvm_dev_ioctl,
1915         .compat_ioctl   = kvm_dev_ioctl,
1916 };
1917
1918 static struct miscdevice kvm_dev = {
1919         KVM_MINOR,
1920         "kvm",
1921         &kvm_chardev_ops,
1922 };
1923
1924 static void hardware_enable(void *junk)
1925 {
1926         int cpu = raw_smp_processor_id();
1927         int r;
1928
1929         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1930                 return;
1931
1932         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1933
1934         r = kvm_arch_hardware_enable(NULL);
1935
1936         if (r) {
1937                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1938                 atomic_inc(&hardware_enable_failed);
1939                 printk(KERN_INFO "kvm: enabling virtualization on "
1940                                  "CPU%d failed\n", cpu);
1941         }
1942 }
1943
1944 static void hardware_disable(void *junk)
1945 {
1946         int cpu = raw_smp_processor_id();
1947
1948         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1949                 return;
1950         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1951         kvm_arch_hardware_disable(NULL);
1952 }
1953
1954 static void hardware_disable_all_nolock(void)
1955 {
1956         BUG_ON(!kvm_usage_count);
1957
1958         kvm_usage_count--;
1959         if (!kvm_usage_count)
1960                 on_each_cpu(hardware_disable, NULL, 1);
1961 }
1962
1963 static void hardware_disable_all(void)
1964 {
1965         spin_lock(&kvm_lock);
1966         hardware_disable_all_nolock();
1967         spin_unlock(&kvm_lock);
1968 }
1969
1970 static int hardware_enable_all(void)
1971 {
1972         int r = 0;
1973
1974         spin_lock(&kvm_lock);
1975
1976         kvm_usage_count++;
1977         if (kvm_usage_count == 1) {
1978                 atomic_set(&hardware_enable_failed, 0);
1979                 on_each_cpu(hardware_enable, NULL, 1);
1980
1981                 if (atomic_read(&hardware_enable_failed)) {
1982                         hardware_disable_all_nolock();
1983                         r = -EBUSY;
1984                 }
1985         }
1986
1987         spin_unlock(&kvm_lock);
1988
1989         return r;
1990 }
1991
1992 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1993                            void *v)
1994 {
1995         int cpu = (long)v;
1996
1997         if (!kvm_usage_count)
1998                 return NOTIFY_OK;
1999
2000         val &= ~CPU_TASKS_FROZEN;
2001         switch (val) {
2002         case CPU_DYING:
2003                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2004                        cpu);
2005                 hardware_disable(NULL);
2006                 break;
2007         case CPU_STARTING:
2008                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2009                        cpu);
2010                 spin_lock(&kvm_lock);
2011                 hardware_enable(NULL);
2012                 spin_unlock(&kvm_lock);
2013                 break;
2014         }
2015         return NOTIFY_OK;
2016 }
2017
2018
2019 asmlinkage void kvm_handle_fault_on_reboot(void)
2020 {
2021         if (kvm_rebooting) {
2022                 /* spin while reset goes on */
2023                 local_irq_enable();
2024                 while (true)
2025                         ;
2026         }
2027         /* Fault while not rebooting.  We want the trace. */
2028         BUG();
2029 }
2030 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2031
2032 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2033                       void *v)
2034 {
2035         /*
2036          * Some (well, at least mine) BIOSes hang on reboot if
2037          * in vmx root mode.
2038          *
2039          * And Intel TXT required VMX off for all cpu when system shutdown.
2040          */
2041         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2042         kvm_rebooting = true;
2043         on_each_cpu(hardware_disable, NULL, 1);
2044         return NOTIFY_OK;
2045 }
2046
2047 static struct notifier_block kvm_reboot_notifier = {
2048         .notifier_call = kvm_reboot,
2049         .priority = 0,
2050 };
2051
2052 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2053 {
2054         int i;
2055
2056         for (i = 0; i < bus->dev_count; i++) {
2057                 struct kvm_io_device *pos = bus->devs[i];
2058
2059                 kvm_iodevice_destructor(pos);
2060         }
2061         kfree(bus);
2062 }
2063
2064 /* kvm_io_bus_write - called under kvm->slots_lock */
2065 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2066                      int len, const void *val)
2067 {
2068         int i;
2069         struct kvm_io_bus *bus;
2070
2071         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2072         for (i = 0; i < bus->dev_count; i++)
2073                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2074                         return 0;
2075         return -EOPNOTSUPP;
2076 }
2077
2078 /* kvm_io_bus_read - called under kvm->slots_lock */
2079 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2080                     int len, void *val)
2081 {
2082         int i;
2083         struct kvm_io_bus *bus;
2084
2085         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2086         for (i = 0; i < bus->dev_count; i++)
2087                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2088                         return 0;
2089         return -EOPNOTSUPP;
2090 }
2091
2092 /* Caller must hold slots_lock. */
2093 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2094                             struct kvm_io_device *dev)
2095 {
2096         struct kvm_io_bus *new_bus, *bus;
2097
2098         bus = kvm->buses[bus_idx];
2099         if (bus->dev_count > NR_IOBUS_DEVS-1)
2100                 return -ENOSPC;
2101
2102         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2103         if (!new_bus)
2104                 return -ENOMEM;
2105         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2106         new_bus->devs[new_bus->dev_count++] = dev;
2107         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2108         synchronize_srcu_expedited(&kvm->srcu);
2109         kfree(bus);
2110
2111         return 0;
2112 }
2113
2114 /* Caller must hold slots_lock. */
2115 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2116                               struct kvm_io_device *dev)
2117 {
2118         int i, r;
2119         struct kvm_io_bus *new_bus, *bus;
2120
2121         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2122         if (!new_bus)
2123                 return -ENOMEM;
2124
2125         bus = kvm->buses[bus_idx];
2126         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2127
2128         r = -ENOENT;
2129         for (i = 0; i < new_bus->dev_count; i++)
2130                 if (new_bus->devs[i] == dev) {
2131                         r = 0;
2132                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2133                         break;
2134                 }
2135
2136         if (r) {
2137                 kfree(new_bus);
2138                 return r;
2139         }
2140
2141         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2142         synchronize_srcu_expedited(&kvm->srcu);
2143         kfree(bus);
2144         return r;
2145 }
2146
2147 static struct notifier_block kvm_cpu_notifier = {
2148         .notifier_call = kvm_cpu_hotplug,
2149 };
2150
2151 static int vm_stat_get(void *_offset, u64 *val)
2152 {
2153         unsigned offset = (long)_offset;
2154         struct kvm *kvm;
2155
2156         *val = 0;
2157         spin_lock(&kvm_lock);
2158         list_for_each_entry(kvm, &vm_list, vm_list)
2159                 *val += *(u32 *)((void *)kvm + offset);
2160         spin_unlock(&kvm_lock);
2161         return 0;
2162 }
2163
2164 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2165
2166 static int vcpu_stat_get(void *_offset, u64 *val)
2167 {
2168         unsigned offset = (long)_offset;
2169         struct kvm *kvm;
2170         struct kvm_vcpu *vcpu;
2171         int i;
2172
2173         *val = 0;
2174         spin_lock(&kvm_lock);
2175         list_for_each_entry(kvm, &vm_list, vm_list)
2176                 kvm_for_each_vcpu(i, vcpu, kvm)
2177                         *val += *(u32 *)((void *)vcpu + offset);
2178
2179         spin_unlock(&kvm_lock);
2180         return 0;
2181 }
2182
2183 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2184
2185 static const struct file_operations *stat_fops[] = {
2186         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2187         [KVM_STAT_VM]   = &vm_stat_fops,
2188 };
2189
2190 static void kvm_init_debug(void)
2191 {
2192         struct kvm_stats_debugfs_item *p;
2193
2194         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2195         for (p = debugfs_entries; p->name; ++p)
2196                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2197                                                 (void *)(long)p->offset,
2198                                                 stat_fops[p->kind]);
2199 }
2200
2201 static void kvm_exit_debug(void)
2202 {
2203         struct kvm_stats_debugfs_item *p;
2204
2205         for (p = debugfs_entries; p->name; ++p)
2206                 debugfs_remove(p->dentry);
2207         debugfs_remove(kvm_debugfs_dir);
2208 }
2209
2210 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2211 {
2212         if (kvm_usage_count)
2213                 hardware_disable(NULL);
2214         return 0;
2215 }
2216
2217 static int kvm_resume(struct sys_device *dev)
2218 {
2219         if (kvm_usage_count) {
2220                 WARN_ON(spin_is_locked(&kvm_lock));
2221                 hardware_enable(NULL);
2222         }
2223         return 0;
2224 }
2225
2226 static struct sysdev_class kvm_sysdev_class = {
2227         .name = "kvm",
2228         .suspend = kvm_suspend,
2229         .resume = kvm_resume,
2230 };
2231
2232 static struct sys_device kvm_sysdev = {
2233         .id = 0,
2234         .cls = &kvm_sysdev_class,
2235 };
2236
2237 struct page *bad_page;
2238 pfn_t bad_pfn;
2239
2240 static inline
2241 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2242 {
2243         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2244 }
2245
2246 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2247 {
2248         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2249
2250         kvm_arch_vcpu_load(vcpu, cpu);
2251 }
2252
2253 static void kvm_sched_out(struct preempt_notifier *pn,
2254                           struct task_struct *next)
2255 {
2256         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2257
2258         kvm_arch_vcpu_put(vcpu);
2259 }
2260
2261 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2262                   struct module *module)
2263 {
2264         int r;
2265         int cpu;
2266
2267         r = kvm_arch_init(opaque);
2268         if (r)
2269                 goto out_fail;
2270
2271         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2272
2273         if (bad_page == NULL) {
2274                 r = -ENOMEM;
2275                 goto out;
2276         }
2277
2278         bad_pfn = page_to_pfn(bad_page);
2279
2280         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2281
2282         if (hwpoison_page == NULL) {
2283                 r = -ENOMEM;
2284                 goto out_free_0;
2285         }
2286
2287         hwpoison_pfn = page_to_pfn(hwpoison_page);
2288
2289         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2290
2291         if (fault_page == NULL) {
2292                 r = -ENOMEM;
2293                 goto out_free_0;
2294         }
2295
2296         fault_pfn = page_to_pfn(fault_page);
2297
2298         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2299                 r = -ENOMEM;
2300                 goto out_free_0;
2301         }
2302
2303         r = kvm_arch_hardware_setup();
2304         if (r < 0)
2305                 goto out_free_0a;
2306
2307         for_each_online_cpu(cpu) {
2308                 smp_call_function_single(cpu,
2309                                 kvm_arch_check_processor_compat,
2310                                 &r, 1);
2311                 if (r < 0)
2312                         goto out_free_1;
2313         }
2314
2315         r = register_cpu_notifier(&kvm_cpu_notifier);
2316         if (r)
2317                 goto out_free_2;
2318         register_reboot_notifier(&kvm_reboot_notifier);
2319
2320         r = sysdev_class_register(&kvm_sysdev_class);
2321         if (r)
2322                 goto out_free_3;
2323
2324         r = sysdev_register(&kvm_sysdev);
2325         if (r)
2326                 goto out_free_4;
2327
2328         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2329         if (!vcpu_align)
2330                 vcpu_align = __alignof__(struct kvm_vcpu);
2331         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2332                                            0, NULL);
2333         if (!kvm_vcpu_cache) {
2334                 r = -ENOMEM;
2335                 goto out_free_5;
2336         }
2337
2338         kvm_chardev_ops.owner = module;
2339         kvm_vm_fops.owner = module;
2340         kvm_vcpu_fops.owner = module;
2341
2342         r = misc_register(&kvm_dev);
2343         if (r) {
2344                 printk(KERN_ERR "kvm: misc device register failed\n");
2345                 goto out_free;
2346         }
2347
2348         kvm_preempt_ops.sched_in = kvm_sched_in;
2349         kvm_preempt_ops.sched_out = kvm_sched_out;
2350
2351         kvm_init_debug();
2352
2353         return 0;
2354
2355 out_free:
2356         kmem_cache_destroy(kvm_vcpu_cache);
2357 out_free_5:
2358         sysdev_unregister(&kvm_sysdev);
2359 out_free_4:
2360         sysdev_class_unregister(&kvm_sysdev_class);
2361 out_free_3:
2362         unregister_reboot_notifier(&kvm_reboot_notifier);
2363         unregister_cpu_notifier(&kvm_cpu_notifier);
2364 out_free_2:
2365 out_free_1:
2366         kvm_arch_hardware_unsetup();
2367 out_free_0a:
2368         free_cpumask_var(cpus_hardware_enabled);
2369 out_free_0:
2370         if (fault_page)
2371                 __free_page(fault_page);
2372         if (hwpoison_page)
2373                 __free_page(hwpoison_page);
2374         __free_page(bad_page);
2375 out:
2376         kvm_arch_exit();
2377 out_fail:
2378         return r;
2379 }
2380 EXPORT_SYMBOL_GPL(kvm_init);
2381
2382 void kvm_exit(void)
2383 {
2384         kvm_exit_debug();
2385         misc_deregister(&kvm_dev);
2386         kmem_cache_destroy(kvm_vcpu_cache);
2387         sysdev_unregister(&kvm_sysdev);
2388         sysdev_class_unregister(&kvm_sysdev_class);
2389         unregister_reboot_notifier(&kvm_reboot_notifier);
2390         unregister_cpu_notifier(&kvm_cpu_notifier);
2391         on_each_cpu(hardware_disable, NULL, 1);
2392         kvm_arch_hardware_unsetup();
2393         kvm_arch_exit();
2394         free_cpumask_var(cpus_hardware_enabled);
2395         __free_page(hwpoison_page);
2396         __free_page(bad_page);
2397 }
2398 EXPORT_SYMBOL_GPL(kvm_exit);