08bd304f8bc722bcb7d829ba70fc89014089106f
[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 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
931 {
932         struct kvm_memory_slot *slot;
933
934         slot = gfn_to_memslot(kvm, gfn);
935         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
936                 return bad_hva();
937         return gfn_to_hva_memslot(slot, gfn);
938 }
939 EXPORT_SYMBOL_GPL(gfn_to_hva);
940
941 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic)
942 {
943         struct page *page[1];
944         int npages;
945         pfn_t pfn;
946
947         if (atomic)
948                 npages = __get_user_pages_fast(addr, 1, 1, page);
949         else {
950                 might_sleep();
951                 npages = get_user_pages_fast(addr, 1, 1, page);
952         }
953
954         if (unlikely(npages != 1)) {
955                 struct vm_area_struct *vma;
956
957                 if (atomic)
958                         goto return_fault_page;
959
960                 down_read(&current->mm->mmap_sem);
961                 if (is_hwpoison_address(addr)) {
962                         up_read(&current->mm->mmap_sem);
963                         get_page(hwpoison_page);
964                         return page_to_pfn(hwpoison_page);
965                 }
966
967                 vma = find_vma(current->mm, addr);
968
969                 if (vma == NULL || addr < vma->vm_start ||
970                     !(vma->vm_flags & VM_PFNMAP)) {
971                         up_read(&current->mm->mmap_sem);
972 return_fault_page:
973                         get_page(fault_page);
974                         return page_to_pfn(fault_page);
975                 }
976
977                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
978                 up_read(&current->mm->mmap_sem);
979                 BUG_ON(!kvm_is_mmio_pfn(pfn));
980         } else
981                 pfn = page_to_pfn(page[0]);
982
983         return pfn;
984 }
985
986 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
987 {
988         return hva_to_pfn(kvm, addr, true);
989 }
990 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
991
992 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
993 {
994         unsigned long addr;
995
996         addr = gfn_to_hva(kvm, gfn);
997         if (kvm_is_error_hva(addr)) {
998                 get_page(bad_page);
999                 return page_to_pfn(bad_page);
1000         }
1001
1002         return hva_to_pfn(kvm, addr, false);
1003 }
1004 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1005
1006 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1007                          struct kvm_memory_slot *slot, gfn_t gfn)
1008 {
1009         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1010         return hva_to_pfn(kvm, addr, false);
1011 }
1012
1013 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1014 {
1015         pfn_t pfn;
1016
1017         pfn = gfn_to_pfn(kvm, gfn);
1018         if (!kvm_is_mmio_pfn(pfn))
1019                 return pfn_to_page(pfn);
1020
1021         WARN_ON(kvm_is_mmio_pfn(pfn));
1022
1023         get_page(bad_page);
1024         return bad_page;
1025 }
1026
1027 EXPORT_SYMBOL_GPL(gfn_to_page);
1028
1029 void kvm_release_page_clean(struct page *page)
1030 {
1031         kvm_release_pfn_clean(page_to_pfn(page));
1032 }
1033 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1034
1035 void kvm_release_pfn_clean(pfn_t pfn)
1036 {
1037         if (!kvm_is_mmio_pfn(pfn))
1038                 put_page(pfn_to_page(pfn));
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1041
1042 void kvm_release_page_dirty(struct page *page)
1043 {
1044         kvm_release_pfn_dirty(page_to_pfn(page));
1045 }
1046 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1047
1048 void kvm_release_pfn_dirty(pfn_t pfn)
1049 {
1050         kvm_set_pfn_dirty(pfn);
1051         kvm_release_pfn_clean(pfn);
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1054
1055 void kvm_set_page_dirty(struct page *page)
1056 {
1057         kvm_set_pfn_dirty(page_to_pfn(page));
1058 }
1059 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1060
1061 void kvm_set_pfn_dirty(pfn_t pfn)
1062 {
1063         if (!kvm_is_mmio_pfn(pfn)) {
1064                 struct page *page = pfn_to_page(pfn);
1065                 if (!PageReserved(page))
1066                         SetPageDirty(page);
1067         }
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1070
1071 void kvm_set_pfn_accessed(pfn_t pfn)
1072 {
1073         if (!kvm_is_mmio_pfn(pfn))
1074                 mark_page_accessed(pfn_to_page(pfn));
1075 }
1076 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1077
1078 void kvm_get_pfn(pfn_t pfn)
1079 {
1080         if (!kvm_is_mmio_pfn(pfn))
1081                 get_page(pfn_to_page(pfn));
1082 }
1083 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1084
1085 static int next_segment(unsigned long len, int offset)
1086 {
1087         if (len > PAGE_SIZE - offset)
1088                 return PAGE_SIZE - offset;
1089         else
1090                 return len;
1091 }
1092
1093 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1094                         int len)
1095 {
1096         int r;
1097         unsigned long addr;
1098
1099         addr = gfn_to_hva(kvm, gfn);
1100         if (kvm_is_error_hva(addr))
1101                 return -EFAULT;
1102         r = copy_from_user(data, (void __user *)addr + offset, len);
1103         if (r)
1104                 return -EFAULT;
1105         return 0;
1106 }
1107 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1108
1109 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1110 {
1111         gfn_t gfn = gpa >> PAGE_SHIFT;
1112         int seg;
1113         int offset = offset_in_page(gpa);
1114         int ret;
1115
1116         while ((seg = next_segment(len, offset)) != 0) {
1117                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1118                 if (ret < 0)
1119                         return ret;
1120                 offset = 0;
1121                 len -= seg;
1122                 data += seg;
1123                 ++gfn;
1124         }
1125         return 0;
1126 }
1127 EXPORT_SYMBOL_GPL(kvm_read_guest);
1128
1129 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1130                           unsigned long len)
1131 {
1132         int r;
1133         unsigned long addr;
1134         gfn_t gfn = gpa >> PAGE_SHIFT;
1135         int offset = offset_in_page(gpa);
1136
1137         addr = gfn_to_hva(kvm, gfn);
1138         if (kvm_is_error_hva(addr))
1139                 return -EFAULT;
1140         pagefault_disable();
1141         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1142         pagefault_enable();
1143         if (r)
1144                 return -EFAULT;
1145         return 0;
1146 }
1147 EXPORT_SYMBOL(kvm_read_guest_atomic);
1148
1149 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1150                          int offset, int len)
1151 {
1152         int r;
1153         unsigned long addr;
1154
1155         addr = gfn_to_hva(kvm, gfn);
1156         if (kvm_is_error_hva(addr))
1157                 return -EFAULT;
1158         r = copy_to_user((void __user *)addr + offset, data, len);
1159         if (r)
1160                 return -EFAULT;
1161         mark_page_dirty(kvm, gfn);
1162         return 0;
1163 }
1164 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1165
1166 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1167                     unsigned long len)
1168 {
1169         gfn_t gfn = gpa >> PAGE_SHIFT;
1170         int seg;
1171         int offset = offset_in_page(gpa);
1172         int ret;
1173
1174         while ((seg = next_segment(len, offset)) != 0) {
1175                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1176                 if (ret < 0)
1177                         return ret;
1178                 offset = 0;
1179                 len -= seg;
1180                 data += seg;
1181                 ++gfn;
1182         }
1183         return 0;
1184 }
1185
1186 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1187 {
1188         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1191
1192 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1193 {
1194         gfn_t gfn = gpa >> PAGE_SHIFT;
1195         int seg;
1196         int offset = offset_in_page(gpa);
1197         int ret;
1198
1199         while ((seg = next_segment(len, offset)) != 0) {
1200                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1201                 if (ret < 0)
1202                         return ret;
1203                 offset = 0;
1204                 len -= seg;
1205                 ++gfn;
1206         }
1207         return 0;
1208 }
1209 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1210
1211 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1212 {
1213         struct kvm_memory_slot *memslot;
1214
1215         memslot = gfn_to_memslot(kvm, gfn);
1216         if (memslot && memslot->dirty_bitmap) {
1217                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1218
1219                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1220         }
1221 }
1222
1223 /*
1224  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1225  */
1226 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1227 {
1228         DEFINE_WAIT(wait);
1229
1230         for (;;) {
1231                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1232
1233                 if (kvm_arch_vcpu_runnable(vcpu)) {
1234                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1235                         break;
1236                 }
1237                 if (kvm_cpu_has_pending_timer(vcpu))
1238                         break;
1239                 if (signal_pending(current))
1240                         break;
1241
1242                 schedule();
1243         }
1244
1245         finish_wait(&vcpu->wq, &wait);
1246 }
1247
1248 void kvm_resched(struct kvm_vcpu *vcpu)
1249 {
1250         if (!need_resched())
1251                 return;
1252         cond_resched();
1253 }
1254 EXPORT_SYMBOL_GPL(kvm_resched);
1255
1256 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1257 {
1258         ktime_t expires;
1259         DEFINE_WAIT(wait);
1260
1261         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1262
1263         /* Sleep for 100 us, and hope lock-holder got scheduled */
1264         expires = ktime_add_ns(ktime_get(), 100000UL);
1265         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1266
1267         finish_wait(&vcpu->wq, &wait);
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1270
1271 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1272 {
1273         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1274         struct page *page;
1275
1276         if (vmf->pgoff == 0)
1277                 page = virt_to_page(vcpu->run);
1278 #ifdef CONFIG_X86
1279         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1280                 page = virt_to_page(vcpu->arch.pio_data);
1281 #endif
1282 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1283         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1284                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1285 #endif
1286         else
1287                 return VM_FAULT_SIGBUS;
1288         get_page(page);
1289         vmf->page = page;
1290         return 0;
1291 }
1292
1293 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1294         .fault = kvm_vcpu_fault,
1295 };
1296
1297 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1298 {
1299         vma->vm_ops = &kvm_vcpu_vm_ops;
1300         return 0;
1301 }
1302
1303 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1304 {
1305         struct kvm_vcpu *vcpu = filp->private_data;
1306
1307         kvm_put_kvm(vcpu->kvm);
1308         return 0;
1309 }
1310
1311 static struct file_operations kvm_vcpu_fops = {
1312         .release        = kvm_vcpu_release,
1313         .unlocked_ioctl = kvm_vcpu_ioctl,
1314         .compat_ioctl   = kvm_vcpu_ioctl,
1315         .mmap           = kvm_vcpu_mmap,
1316 };
1317
1318 /*
1319  * Allocates an inode for the vcpu.
1320  */
1321 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1322 {
1323         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1324 }
1325
1326 /*
1327  * Creates some virtual cpus.  Good luck creating more than one.
1328  */
1329 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1330 {
1331         int r;
1332         struct kvm_vcpu *vcpu, *v;
1333
1334         vcpu = kvm_arch_vcpu_create(kvm, id);
1335         if (IS_ERR(vcpu))
1336                 return PTR_ERR(vcpu);
1337
1338         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1339
1340         r = kvm_arch_vcpu_setup(vcpu);
1341         if (r)
1342                 return r;
1343
1344         mutex_lock(&kvm->lock);
1345         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1346                 r = -EINVAL;
1347                 goto vcpu_destroy;
1348         }
1349
1350         kvm_for_each_vcpu(r, v, kvm)
1351                 if (v->vcpu_id == id) {
1352                         r = -EEXIST;
1353                         goto vcpu_destroy;
1354                 }
1355
1356         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1357
1358         /* Now it's all set up, let userspace reach it */
1359         kvm_get_kvm(kvm);
1360         r = create_vcpu_fd(vcpu);
1361         if (r < 0) {
1362                 kvm_put_kvm(kvm);
1363                 goto vcpu_destroy;
1364         }
1365
1366         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1367         smp_wmb();
1368         atomic_inc(&kvm->online_vcpus);
1369
1370 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1371         if (kvm->bsp_vcpu_id == id)
1372                 kvm->bsp_vcpu = vcpu;
1373 #endif
1374         mutex_unlock(&kvm->lock);
1375         return r;
1376
1377 vcpu_destroy:
1378         mutex_unlock(&kvm->lock);
1379         kvm_arch_vcpu_destroy(vcpu);
1380         return r;
1381 }
1382
1383 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1384 {
1385         if (sigset) {
1386                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1387                 vcpu->sigset_active = 1;
1388                 vcpu->sigset = *sigset;
1389         } else
1390                 vcpu->sigset_active = 0;
1391         return 0;
1392 }
1393
1394 static long kvm_vcpu_ioctl(struct file *filp,
1395                            unsigned int ioctl, unsigned long arg)
1396 {
1397         struct kvm_vcpu *vcpu = filp->private_data;
1398         void __user *argp = (void __user *)arg;
1399         int r;
1400         struct kvm_fpu *fpu = NULL;
1401         struct kvm_sregs *kvm_sregs = NULL;
1402
1403         if (vcpu->kvm->mm != current->mm)
1404                 return -EIO;
1405
1406 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1407         /*
1408          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1409          * so vcpu_load() would break it.
1410          */
1411         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1412                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1413 #endif
1414
1415
1416         vcpu_load(vcpu);
1417         switch (ioctl) {
1418         case KVM_RUN:
1419                 r = -EINVAL;
1420                 if (arg)
1421                         goto out;
1422                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1423                 break;
1424         case KVM_GET_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 = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1432                 if (r)
1433                         goto out_free1;
1434                 r = -EFAULT;
1435                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1436                         goto out_free1;
1437                 r = 0;
1438 out_free1:
1439                 kfree(kvm_regs);
1440                 break;
1441         }
1442         case KVM_SET_REGS: {
1443                 struct kvm_regs *kvm_regs;
1444
1445                 r = -ENOMEM;
1446                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1447                 if (!kvm_regs)
1448                         goto out;
1449                 r = -EFAULT;
1450                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1451                         goto out_free2;
1452                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1453                 if (r)
1454                         goto out_free2;
1455                 r = 0;
1456 out_free2:
1457                 kfree(kvm_regs);
1458                 break;
1459         }
1460         case KVM_GET_SREGS: {
1461                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1462                 r = -ENOMEM;
1463                 if (!kvm_sregs)
1464                         goto out;
1465                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1466                 if (r)
1467                         goto out;
1468                 r = -EFAULT;
1469                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1470                         goto out;
1471                 r = 0;
1472                 break;
1473         }
1474         case KVM_SET_SREGS: {
1475                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1476                 r = -ENOMEM;
1477                 if (!kvm_sregs)
1478                         goto out;
1479                 r = -EFAULT;
1480                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1481                         goto out;
1482                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1483                 if (r)
1484                         goto out;
1485                 r = 0;
1486                 break;
1487         }
1488         case KVM_GET_MP_STATE: {
1489                 struct kvm_mp_state mp_state;
1490
1491                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1492                 if (r)
1493                         goto out;
1494                 r = -EFAULT;
1495                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1496                         goto out;
1497                 r = 0;
1498                 break;
1499         }
1500         case KVM_SET_MP_STATE: {
1501                 struct kvm_mp_state mp_state;
1502
1503                 r = -EFAULT;
1504                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1505                         goto out;
1506                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1507                 if (r)
1508                         goto out;
1509                 r = 0;
1510                 break;
1511         }
1512         case KVM_TRANSLATE: {
1513                 struct kvm_translation tr;
1514
1515                 r = -EFAULT;
1516                 if (copy_from_user(&tr, argp, sizeof tr))
1517                         goto out;
1518                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1519                 if (r)
1520                         goto out;
1521                 r = -EFAULT;
1522                 if (copy_to_user(argp, &tr, sizeof tr))
1523                         goto out;
1524                 r = 0;
1525                 break;
1526         }
1527         case KVM_SET_GUEST_DEBUG: {
1528                 struct kvm_guest_debug dbg;
1529
1530                 r = -EFAULT;
1531                 if (copy_from_user(&dbg, argp, sizeof dbg))
1532                         goto out;
1533                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1534                 if (r)
1535                         goto out;
1536                 r = 0;
1537                 break;
1538         }
1539         case KVM_SET_SIGNAL_MASK: {
1540                 struct kvm_signal_mask __user *sigmask_arg = argp;
1541                 struct kvm_signal_mask kvm_sigmask;
1542                 sigset_t sigset, *p;
1543
1544                 p = NULL;
1545                 if (argp) {
1546                         r = -EFAULT;
1547                         if (copy_from_user(&kvm_sigmask, argp,
1548                                            sizeof kvm_sigmask))
1549                                 goto out;
1550                         r = -EINVAL;
1551                         if (kvm_sigmask.len != sizeof sigset)
1552                                 goto out;
1553                         r = -EFAULT;
1554                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1555                                            sizeof sigset))
1556                                 goto out;
1557                         p = &sigset;
1558                 }
1559                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1560                 break;
1561         }
1562         case KVM_GET_FPU: {
1563                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1564                 r = -ENOMEM;
1565                 if (!fpu)
1566                         goto out;
1567                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1568                 if (r)
1569                         goto out;
1570                 r = -EFAULT;
1571                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1572                         goto out;
1573                 r = 0;
1574                 break;
1575         }
1576         case KVM_SET_FPU: {
1577                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1578                 r = -ENOMEM;
1579                 if (!fpu)
1580                         goto out;
1581                 r = -EFAULT;
1582                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1583                         goto out;
1584                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1585                 if (r)
1586                         goto out;
1587                 r = 0;
1588                 break;
1589         }
1590         default:
1591                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1592         }
1593 out:
1594         vcpu_put(vcpu);
1595         kfree(fpu);
1596         kfree(kvm_sregs);
1597         return r;
1598 }
1599
1600 static long kvm_vm_ioctl(struct file *filp,
1601                            unsigned int ioctl, unsigned long arg)
1602 {
1603         struct kvm *kvm = filp->private_data;
1604         void __user *argp = (void __user *)arg;
1605         int r;
1606
1607         if (kvm->mm != current->mm)
1608                 return -EIO;
1609         switch (ioctl) {
1610         case KVM_CREATE_VCPU:
1611                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1612                 if (r < 0)
1613                         goto out;
1614                 break;
1615         case KVM_SET_USER_MEMORY_REGION: {
1616                 struct kvm_userspace_memory_region kvm_userspace_mem;
1617
1618                 r = -EFAULT;
1619                 if (copy_from_user(&kvm_userspace_mem, argp,
1620                                                 sizeof kvm_userspace_mem))
1621                         goto out;
1622
1623                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1624                 if (r)
1625                         goto out;
1626                 break;
1627         }
1628         case KVM_GET_DIRTY_LOG: {
1629                 struct kvm_dirty_log log;
1630
1631                 r = -EFAULT;
1632                 if (copy_from_user(&log, argp, sizeof log))
1633                         goto out;
1634                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1635                 if (r)
1636                         goto out;
1637                 break;
1638         }
1639 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1640         case KVM_REGISTER_COALESCED_MMIO: {
1641                 struct kvm_coalesced_mmio_zone zone;
1642                 r = -EFAULT;
1643                 if (copy_from_user(&zone, argp, sizeof zone))
1644                         goto out;
1645                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1646                 if (r)
1647                         goto out;
1648                 r = 0;
1649                 break;
1650         }
1651         case KVM_UNREGISTER_COALESCED_MMIO: {
1652                 struct kvm_coalesced_mmio_zone zone;
1653                 r = -EFAULT;
1654                 if (copy_from_user(&zone, argp, sizeof zone))
1655                         goto out;
1656                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1657                 if (r)
1658                         goto out;
1659                 r = 0;
1660                 break;
1661         }
1662 #endif
1663         case KVM_IRQFD: {
1664                 struct kvm_irqfd data;
1665
1666                 r = -EFAULT;
1667                 if (copy_from_user(&data, argp, sizeof data))
1668                         goto out;
1669                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1670                 break;
1671         }
1672         case KVM_IOEVENTFD: {
1673                 struct kvm_ioeventfd data;
1674
1675                 r = -EFAULT;
1676                 if (copy_from_user(&data, argp, sizeof data))
1677                         goto out;
1678                 r = kvm_ioeventfd(kvm, &data);
1679                 break;
1680         }
1681 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1682         case KVM_SET_BOOT_CPU_ID:
1683                 r = 0;
1684                 mutex_lock(&kvm->lock);
1685                 if (atomic_read(&kvm->online_vcpus) != 0)
1686                         r = -EBUSY;
1687                 else
1688                         kvm->bsp_vcpu_id = arg;
1689                 mutex_unlock(&kvm->lock);
1690                 break;
1691 #endif
1692         default:
1693                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1694                 if (r == -ENOTTY)
1695                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1696         }
1697 out:
1698         return r;
1699 }
1700
1701 #ifdef CONFIG_COMPAT
1702 struct compat_kvm_dirty_log {
1703         __u32 slot;
1704         __u32 padding1;
1705         union {
1706                 compat_uptr_t dirty_bitmap; /* one bit per page */
1707                 __u64 padding2;
1708         };
1709 };
1710
1711 static long kvm_vm_compat_ioctl(struct file *filp,
1712                            unsigned int ioctl, unsigned long arg)
1713 {
1714         struct kvm *kvm = filp->private_data;
1715         int r;
1716
1717         if (kvm->mm != current->mm)
1718                 return -EIO;
1719         switch (ioctl) {
1720         case KVM_GET_DIRTY_LOG: {
1721                 struct compat_kvm_dirty_log compat_log;
1722                 struct kvm_dirty_log log;
1723
1724                 r = -EFAULT;
1725                 if (copy_from_user(&compat_log, (void __user *)arg,
1726                                    sizeof(compat_log)))
1727                         goto out;
1728                 log.slot         = compat_log.slot;
1729                 log.padding1     = compat_log.padding1;
1730                 log.padding2     = compat_log.padding2;
1731                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1732
1733                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1734                 if (r)
1735                         goto out;
1736                 break;
1737         }
1738         default:
1739                 r = kvm_vm_ioctl(filp, ioctl, arg);
1740         }
1741
1742 out:
1743         return r;
1744 }
1745 #endif
1746
1747 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1748 {
1749         struct page *page[1];
1750         unsigned long addr;
1751         int npages;
1752         gfn_t gfn = vmf->pgoff;
1753         struct kvm *kvm = vma->vm_file->private_data;
1754
1755         addr = gfn_to_hva(kvm, gfn);
1756         if (kvm_is_error_hva(addr))
1757                 return VM_FAULT_SIGBUS;
1758
1759         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1760                                 NULL);
1761         if (unlikely(npages != 1))
1762                 return VM_FAULT_SIGBUS;
1763
1764         vmf->page = page[0];
1765         return 0;
1766 }
1767
1768 static const struct vm_operations_struct kvm_vm_vm_ops = {
1769         .fault = kvm_vm_fault,
1770 };
1771
1772 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1773 {
1774         vma->vm_ops = &kvm_vm_vm_ops;
1775         return 0;
1776 }
1777
1778 static struct file_operations kvm_vm_fops = {
1779         .release        = kvm_vm_release,
1780         .unlocked_ioctl = kvm_vm_ioctl,
1781 #ifdef CONFIG_COMPAT
1782         .compat_ioctl   = kvm_vm_compat_ioctl,
1783 #endif
1784         .mmap           = kvm_vm_mmap,
1785 };
1786
1787 static int kvm_dev_ioctl_create_vm(void)
1788 {
1789         int fd, r;
1790         struct kvm *kvm;
1791
1792         kvm = kvm_create_vm();
1793         if (IS_ERR(kvm))
1794                 return PTR_ERR(kvm);
1795 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1796         r = kvm_coalesced_mmio_init(kvm);
1797         if (r < 0) {
1798                 kvm_put_kvm(kvm);
1799                 return r;
1800         }
1801 #endif
1802         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1803         if (fd < 0)
1804                 kvm_put_kvm(kvm);
1805
1806         return fd;
1807 }
1808
1809 static long kvm_dev_ioctl_check_extension_generic(long arg)
1810 {
1811         switch (arg) {
1812         case KVM_CAP_USER_MEMORY:
1813         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1814         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1815 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1816         case KVM_CAP_SET_BOOT_CPU_ID:
1817 #endif
1818         case KVM_CAP_INTERNAL_ERROR_DATA:
1819                 return 1;
1820 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1821         case KVM_CAP_IRQ_ROUTING:
1822                 return KVM_MAX_IRQ_ROUTES;
1823 #endif
1824         default:
1825                 break;
1826         }
1827         return kvm_dev_ioctl_check_extension(arg);
1828 }
1829
1830 static long kvm_dev_ioctl(struct file *filp,
1831                           unsigned int ioctl, unsigned long arg)
1832 {
1833         long r = -EINVAL;
1834
1835         switch (ioctl) {
1836         case KVM_GET_API_VERSION:
1837                 r = -EINVAL;
1838                 if (arg)
1839                         goto out;
1840                 r = KVM_API_VERSION;
1841                 break;
1842         case KVM_CREATE_VM:
1843                 r = -EINVAL;
1844                 if (arg)
1845                         goto out;
1846                 r = kvm_dev_ioctl_create_vm();
1847                 break;
1848         case KVM_CHECK_EXTENSION:
1849                 r = kvm_dev_ioctl_check_extension_generic(arg);
1850                 break;
1851         case KVM_GET_VCPU_MMAP_SIZE:
1852                 r = -EINVAL;
1853                 if (arg)
1854                         goto out;
1855                 r = PAGE_SIZE;     /* struct kvm_run */
1856 #ifdef CONFIG_X86
1857                 r += PAGE_SIZE;    /* pio data page */
1858 #endif
1859 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1860                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1861 #endif
1862                 break;
1863         case KVM_TRACE_ENABLE:
1864         case KVM_TRACE_PAUSE:
1865         case KVM_TRACE_DISABLE:
1866                 r = -EOPNOTSUPP;
1867                 break;
1868         default:
1869                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1870         }
1871 out:
1872         return r;
1873 }
1874
1875 static struct file_operations kvm_chardev_ops = {
1876         .unlocked_ioctl = kvm_dev_ioctl,
1877         .compat_ioctl   = kvm_dev_ioctl,
1878 };
1879
1880 static struct miscdevice kvm_dev = {
1881         KVM_MINOR,
1882         "kvm",
1883         &kvm_chardev_ops,
1884 };
1885
1886 static void hardware_enable(void *junk)
1887 {
1888         int cpu = raw_smp_processor_id();
1889         int r;
1890
1891         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1892                 return;
1893
1894         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1895
1896         r = kvm_arch_hardware_enable(NULL);
1897
1898         if (r) {
1899                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1900                 atomic_inc(&hardware_enable_failed);
1901                 printk(KERN_INFO "kvm: enabling virtualization on "
1902                                  "CPU%d failed\n", cpu);
1903         }
1904 }
1905
1906 static void hardware_disable(void *junk)
1907 {
1908         int cpu = raw_smp_processor_id();
1909
1910         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1911                 return;
1912         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1913         kvm_arch_hardware_disable(NULL);
1914 }
1915
1916 static void hardware_disable_all_nolock(void)
1917 {
1918         BUG_ON(!kvm_usage_count);
1919
1920         kvm_usage_count--;
1921         if (!kvm_usage_count)
1922                 on_each_cpu(hardware_disable, NULL, 1);
1923 }
1924
1925 static void hardware_disable_all(void)
1926 {
1927         spin_lock(&kvm_lock);
1928         hardware_disable_all_nolock();
1929         spin_unlock(&kvm_lock);
1930 }
1931
1932 static int hardware_enable_all(void)
1933 {
1934         int r = 0;
1935
1936         spin_lock(&kvm_lock);
1937
1938         kvm_usage_count++;
1939         if (kvm_usage_count == 1) {
1940                 atomic_set(&hardware_enable_failed, 0);
1941                 on_each_cpu(hardware_enable, NULL, 1);
1942
1943                 if (atomic_read(&hardware_enable_failed)) {
1944                         hardware_disable_all_nolock();
1945                         r = -EBUSY;
1946                 }
1947         }
1948
1949         spin_unlock(&kvm_lock);
1950
1951         return r;
1952 }
1953
1954 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1955                            void *v)
1956 {
1957         int cpu = (long)v;
1958
1959         if (!kvm_usage_count)
1960                 return NOTIFY_OK;
1961
1962         val &= ~CPU_TASKS_FROZEN;
1963         switch (val) {
1964         case CPU_DYING:
1965                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1966                        cpu);
1967                 hardware_disable(NULL);
1968                 break;
1969         case CPU_STARTING:
1970                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1971                        cpu);
1972                 spin_lock(&kvm_lock);
1973                 hardware_enable(NULL);
1974                 spin_unlock(&kvm_lock);
1975                 break;
1976         }
1977         return NOTIFY_OK;
1978 }
1979
1980
1981 asmlinkage void kvm_handle_fault_on_reboot(void)
1982 {
1983         if (kvm_rebooting) {
1984                 /* spin while reset goes on */
1985                 local_irq_enable();
1986                 while (true)
1987                         ;
1988         }
1989         /* Fault while not rebooting.  We want the trace. */
1990         BUG();
1991 }
1992 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1993
1994 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1995                       void *v)
1996 {
1997         /*
1998          * Some (well, at least mine) BIOSes hang on reboot if
1999          * in vmx root mode.
2000          *
2001          * And Intel TXT required VMX off for all cpu when system shutdown.
2002          */
2003         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2004         kvm_rebooting = true;
2005         on_each_cpu(hardware_disable, NULL, 1);
2006         return NOTIFY_OK;
2007 }
2008
2009 static struct notifier_block kvm_reboot_notifier = {
2010         .notifier_call = kvm_reboot,
2011         .priority = 0,
2012 };
2013
2014 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2015 {
2016         int i;
2017
2018         for (i = 0; i < bus->dev_count; i++) {
2019                 struct kvm_io_device *pos = bus->devs[i];
2020
2021                 kvm_iodevice_destructor(pos);
2022         }
2023         kfree(bus);
2024 }
2025
2026 /* kvm_io_bus_write - called under kvm->slots_lock */
2027 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2028                      int len, const void *val)
2029 {
2030         int i;
2031         struct kvm_io_bus *bus;
2032
2033         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2034         for (i = 0; i < bus->dev_count; i++)
2035                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2036                         return 0;
2037         return -EOPNOTSUPP;
2038 }
2039
2040 /* kvm_io_bus_read - called under kvm->slots_lock */
2041 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2042                     int len, void *val)
2043 {
2044         int i;
2045         struct kvm_io_bus *bus;
2046
2047         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2048         for (i = 0; i < bus->dev_count; i++)
2049                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2050                         return 0;
2051         return -EOPNOTSUPP;
2052 }
2053
2054 /* Caller must hold slots_lock. */
2055 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2056                             struct kvm_io_device *dev)
2057 {
2058         struct kvm_io_bus *new_bus, *bus;
2059
2060         bus = kvm->buses[bus_idx];
2061         if (bus->dev_count > NR_IOBUS_DEVS-1)
2062                 return -ENOSPC;
2063
2064         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2065         if (!new_bus)
2066                 return -ENOMEM;
2067         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2068         new_bus->devs[new_bus->dev_count++] = dev;
2069         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2070         synchronize_srcu_expedited(&kvm->srcu);
2071         kfree(bus);
2072
2073         return 0;
2074 }
2075
2076 /* Caller must hold slots_lock. */
2077 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2078                               struct kvm_io_device *dev)
2079 {
2080         int i, r;
2081         struct kvm_io_bus *new_bus, *bus;
2082
2083         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2084         if (!new_bus)
2085                 return -ENOMEM;
2086
2087         bus = kvm->buses[bus_idx];
2088         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2089
2090         r = -ENOENT;
2091         for (i = 0; i < new_bus->dev_count; i++)
2092                 if (new_bus->devs[i] == dev) {
2093                         r = 0;
2094                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2095                         break;
2096                 }
2097
2098         if (r) {
2099                 kfree(new_bus);
2100                 return r;
2101         }
2102
2103         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2104         synchronize_srcu_expedited(&kvm->srcu);
2105         kfree(bus);
2106         return r;
2107 }
2108
2109 static struct notifier_block kvm_cpu_notifier = {
2110         .notifier_call = kvm_cpu_hotplug,
2111 };
2112
2113 static int vm_stat_get(void *_offset, u64 *val)
2114 {
2115         unsigned offset = (long)_offset;
2116         struct kvm *kvm;
2117
2118         *val = 0;
2119         spin_lock(&kvm_lock);
2120         list_for_each_entry(kvm, &vm_list, vm_list)
2121                 *val += *(u32 *)((void *)kvm + offset);
2122         spin_unlock(&kvm_lock);
2123         return 0;
2124 }
2125
2126 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2127
2128 static int vcpu_stat_get(void *_offset, u64 *val)
2129 {
2130         unsigned offset = (long)_offset;
2131         struct kvm *kvm;
2132         struct kvm_vcpu *vcpu;
2133         int i;
2134
2135         *val = 0;
2136         spin_lock(&kvm_lock);
2137         list_for_each_entry(kvm, &vm_list, vm_list)
2138                 kvm_for_each_vcpu(i, vcpu, kvm)
2139                         *val += *(u32 *)((void *)vcpu + offset);
2140
2141         spin_unlock(&kvm_lock);
2142         return 0;
2143 }
2144
2145 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2146
2147 static const struct file_operations *stat_fops[] = {
2148         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2149         [KVM_STAT_VM]   = &vm_stat_fops,
2150 };
2151
2152 static void kvm_init_debug(void)
2153 {
2154         struct kvm_stats_debugfs_item *p;
2155
2156         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2157         for (p = debugfs_entries; p->name; ++p)
2158                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2159                                                 (void *)(long)p->offset,
2160                                                 stat_fops[p->kind]);
2161 }
2162
2163 static void kvm_exit_debug(void)
2164 {
2165         struct kvm_stats_debugfs_item *p;
2166
2167         for (p = debugfs_entries; p->name; ++p)
2168                 debugfs_remove(p->dentry);
2169         debugfs_remove(kvm_debugfs_dir);
2170 }
2171
2172 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2173 {
2174         if (kvm_usage_count)
2175                 hardware_disable(NULL);
2176         return 0;
2177 }
2178
2179 static int kvm_resume(struct sys_device *dev)
2180 {
2181         if (kvm_usage_count) {
2182                 WARN_ON(spin_is_locked(&kvm_lock));
2183                 hardware_enable(NULL);
2184         }
2185         return 0;
2186 }
2187
2188 static struct sysdev_class kvm_sysdev_class = {
2189         .name = "kvm",
2190         .suspend = kvm_suspend,
2191         .resume = kvm_resume,
2192 };
2193
2194 static struct sys_device kvm_sysdev = {
2195         .id = 0,
2196         .cls = &kvm_sysdev_class,
2197 };
2198
2199 struct page *bad_page;
2200 pfn_t bad_pfn;
2201
2202 static inline
2203 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2204 {
2205         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2206 }
2207
2208 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2209 {
2210         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2211
2212         kvm_arch_vcpu_load(vcpu, cpu);
2213 }
2214
2215 static void kvm_sched_out(struct preempt_notifier *pn,
2216                           struct task_struct *next)
2217 {
2218         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2219
2220         kvm_arch_vcpu_put(vcpu);
2221 }
2222
2223 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2224                   struct module *module)
2225 {
2226         int r;
2227         int cpu;
2228
2229         r = kvm_arch_init(opaque);
2230         if (r)
2231                 goto out_fail;
2232
2233         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2234
2235         if (bad_page == NULL) {
2236                 r = -ENOMEM;
2237                 goto out;
2238         }
2239
2240         bad_pfn = page_to_pfn(bad_page);
2241
2242         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2243
2244         if (hwpoison_page == NULL) {
2245                 r = -ENOMEM;
2246                 goto out_free_0;
2247         }
2248
2249         hwpoison_pfn = page_to_pfn(hwpoison_page);
2250
2251         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2252
2253         if (fault_page == NULL) {
2254                 r = -ENOMEM;
2255                 goto out_free_0;
2256         }
2257
2258         fault_pfn = page_to_pfn(fault_page);
2259
2260         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2261                 r = -ENOMEM;
2262                 goto out_free_0;
2263         }
2264
2265         r = kvm_arch_hardware_setup();
2266         if (r < 0)
2267                 goto out_free_0a;
2268
2269         for_each_online_cpu(cpu) {
2270                 smp_call_function_single(cpu,
2271                                 kvm_arch_check_processor_compat,
2272                                 &r, 1);
2273                 if (r < 0)
2274                         goto out_free_1;
2275         }
2276
2277         r = register_cpu_notifier(&kvm_cpu_notifier);
2278         if (r)
2279                 goto out_free_2;
2280         register_reboot_notifier(&kvm_reboot_notifier);
2281
2282         r = sysdev_class_register(&kvm_sysdev_class);
2283         if (r)
2284                 goto out_free_3;
2285
2286         r = sysdev_register(&kvm_sysdev);
2287         if (r)
2288                 goto out_free_4;
2289
2290         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2291         if (!vcpu_align)
2292                 vcpu_align = __alignof__(struct kvm_vcpu);
2293         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2294                                            0, NULL);
2295         if (!kvm_vcpu_cache) {
2296                 r = -ENOMEM;
2297                 goto out_free_5;
2298         }
2299
2300         kvm_chardev_ops.owner = module;
2301         kvm_vm_fops.owner = module;
2302         kvm_vcpu_fops.owner = module;
2303
2304         r = misc_register(&kvm_dev);
2305         if (r) {
2306                 printk(KERN_ERR "kvm: misc device register failed\n");
2307                 goto out_free;
2308         }
2309
2310         kvm_preempt_ops.sched_in = kvm_sched_in;
2311         kvm_preempt_ops.sched_out = kvm_sched_out;
2312
2313         kvm_init_debug();
2314
2315         return 0;
2316
2317 out_free:
2318         kmem_cache_destroy(kvm_vcpu_cache);
2319 out_free_5:
2320         sysdev_unregister(&kvm_sysdev);
2321 out_free_4:
2322         sysdev_class_unregister(&kvm_sysdev_class);
2323 out_free_3:
2324         unregister_reboot_notifier(&kvm_reboot_notifier);
2325         unregister_cpu_notifier(&kvm_cpu_notifier);
2326 out_free_2:
2327 out_free_1:
2328         kvm_arch_hardware_unsetup();
2329 out_free_0a:
2330         free_cpumask_var(cpus_hardware_enabled);
2331 out_free_0:
2332         if (fault_page)
2333                 __free_page(fault_page);
2334         if (hwpoison_page)
2335                 __free_page(hwpoison_page);
2336         __free_page(bad_page);
2337 out:
2338         kvm_arch_exit();
2339 out_fail:
2340         return r;
2341 }
2342 EXPORT_SYMBOL_GPL(kvm_init);
2343
2344 void kvm_exit(void)
2345 {
2346         kvm_exit_debug();
2347         misc_deregister(&kvm_dev);
2348         kmem_cache_destroy(kvm_vcpu_cache);
2349         sysdev_unregister(&kvm_sysdev);
2350         sysdev_class_unregister(&kvm_sysdev_class);
2351         unregister_reboot_notifier(&kvm_reboot_notifier);
2352         unregister_cpu_notifier(&kvm_cpu_notifier);
2353         on_each_cpu(hardware_disable, NULL, 1);
2354         kvm_arch_hardware_unsetup();
2355         kvm_arch_exit();
2356         free_cpumask_var(cpus_hardware_enabled);
2357         __free_page(hwpoison_page);
2358         __free_page(bad_page);
2359 }
2360 EXPORT_SYMBOL_GPL(kvm_exit);