KVM: Refactor srcu struct release on early errors
[linux-2.6.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/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 #include "async_pf.h"
59
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/kvm.h>
62
63 MODULE_AUTHOR("Qumranet");
64 MODULE_LICENSE("GPL");
65
66 /*
67  * Ordering of locks:
68  *
69  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
70  */
71
72 DEFINE_SPINLOCK(kvm_lock);
73 LIST_HEAD(vm_list);
74
75 static cpumask_var_t cpus_hardware_enabled;
76 static int kvm_usage_count = 0;
77 static atomic_t hardware_enable_failed;
78
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
83
84 struct dentry *kvm_debugfs_dir;
85
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87                            unsigned long arg);
88 static int hardware_enable_all(void);
89 static void hardware_disable_all(void);
90
91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
92
93 static bool kvm_rebooting;
94
95 static bool largepages_enabled = true;
96
97 static struct page *hwpoison_page;
98 static pfn_t hwpoison_pfn;
99
100 static struct page *fault_page;
101 static pfn_t fault_pfn;
102
103 inline int kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 struct page *page = compound_head(pfn_to_page(pfn));
107                 return PageReserved(page);
108         }
109
110         return true;
111 }
112
113 /*
114  * Switches to specified vcpu, until a matching vcpu_put()
115  */
116 void vcpu_load(struct kvm_vcpu *vcpu)
117 {
118         int cpu;
119
120         mutex_lock(&vcpu->mutex);
121         cpu = get_cpu();
122         preempt_notifier_register(&vcpu->preempt_notifier);
123         kvm_arch_vcpu_load(vcpu, cpu);
124         put_cpu();
125 }
126
127 void vcpu_put(struct kvm_vcpu *vcpu)
128 {
129         preempt_disable();
130         kvm_arch_vcpu_put(vcpu);
131         preempt_notifier_unregister(&vcpu->preempt_notifier);
132         preempt_enable();
133         mutex_unlock(&vcpu->mutex);
134 }
135
136 static void ack_flush(void *_completed)
137 {
138 }
139
140 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
141 {
142         int i, cpu, me;
143         cpumask_var_t cpus;
144         bool called = true;
145         struct kvm_vcpu *vcpu;
146
147         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
148
149         raw_spin_lock(&kvm->requests_lock);
150         me = smp_processor_id();
151         kvm_for_each_vcpu(i, vcpu, kvm) {
152                 if (kvm_make_check_request(req, vcpu))
153                         continue;
154                 cpu = vcpu->cpu;
155                 if (cpus != NULL && cpu != -1 && cpu != me)
156                         cpumask_set_cpu(cpu, cpus);
157         }
158         if (unlikely(cpus == NULL))
159                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
160         else if (!cpumask_empty(cpus))
161                 smp_call_function_many(cpus, ack_flush, NULL, 1);
162         else
163                 called = false;
164         raw_spin_unlock(&kvm->requests_lock);
165         free_cpumask_var(cpus);
166         return called;
167 }
168
169 void kvm_flush_remote_tlbs(struct kvm *kvm)
170 {
171         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
172                 ++kvm->stat.remote_tlb_flush;
173 }
174
175 void kvm_reload_remote_mmus(struct kvm *kvm)
176 {
177         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
178 }
179
180 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
181 {
182         struct page *page;
183         int r;
184
185         mutex_init(&vcpu->mutex);
186         vcpu->cpu = -1;
187         vcpu->kvm = kvm;
188         vcpu->vcpu_id = id;
189         init_waitqueue_head(&vcpu->wq);
190         kvm_async_pf_vcpu_init(vcpu);
191
192         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
193         if (!page) {
194                 r = -ENOMEM;
195                 goto fail;
196         }
197         vcpu->run = page_address(page);
198
199         r = kvm_arch_vcpu_init(vcpu);
200         if (r < 0)
201                 goto fail_free_run;
202         return 0;
203
204 fail_free_run:
205         free_page((unsigned long)vcpu->run);
206 fail:
207         return r;
208 }
209 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
210
211 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
212 {
213         kvm_arch_vcpu_uninit(vcpu);
214         free_page((unsigned long)vcpu->run);
215 }
216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
217
218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
219 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
220 {
221         return container_of(mn, struct kvm, mmu_notifier);
222 }
223
224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
225                                              struct mm_struct *mm,
226                                              unsigned long address)
227 {
228         struct kvm *kvm = mmu_notifier_to_kvm(mn);
229         int need_tlb_flush, idx;
230
231         /*
232          * When ->invalidate_page runs, the linux pte has been zapped
233          * already but the page is still allocated until
234          * ->invalidate_page returns. So if we increase the sequence
235          * here the kvm page fault will notice if the spte can't be
236          * established because the page is going to be freed. If
237          * instead the kvm page fault establishes the spte before
238          * ->invalidate_page runs, kvm_unmap_hva will release it
239          * before returning.
240          *
241          * The sequence increase only need to be seen at spin_unlock
242          * time, and not at spin_lock time.
243          *
244          * Increasing the sequence after the spin_unlock would be
245          * unsafe because the kvm page fault could then establish the
246          * pte after kvm_unmap_hva returned, without noticing the page
247          * is going to be freed.
248          */
249         idx = srcu_read_lock(&kvm->srcu);
250         spin_lock(&kvm->mmu_lock);
251         kvm->mmu_notifier_seq++;
252         need_tlb_flush = kvm_unmap_hva(kvm, address);
253         spin_unlock(&kvm->mmu_lock);
254         srcu_read_unlock(&kvm->srcu, idx);
255
256         /* we've to flush the tlb before the pages can be freed */
257         if (need_tlb_flush)
258                 kvm_flush_remote_tlbs(kvm);
259
260 }
261
262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
263                                         struct mm_struct *mm,
264                                         unsigned long address,
265                                         pte_t pte)
266 {
267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
268         int idx;
269
270         idx = srcu_read_lock(&kvm->srcu);
271         spin_lock(&kvm->mmu_lock);
272         kvm->mmu_notifier_seq++;
273         kvm_set_spte_hva(kvm, address, pte);
274         spin_unlock(&kvm->mmu_lock);
275         srcu_read_unlock(&kvm->srcu, idx);
276 }
277
278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
279                                                     struct mm_struct *mm,
280                                                     unsigned long start,
281                                                     unsigned long end)
282 {
283         struct kvm *kvm = mmu_notifier_to_kvm(mn);
284         int need_tlb_flush = 0, idx;
285
286         idx = srcu_read_lock(&kvm->srcu);
287         spin_lock(&kvm->mmu_lock);
288         /*
289          * The count increase must become visible at unlock time as no
290          * spte can be established without taking the mmu_lock and
291          * count is also read inside the mmu_lock critical section.
292          */
293         kvm->mmu_notifier_count++;
294         for (; start < end; start += PAGE_SIZE)
295                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
296         spin_unlock(&kvm->mmu_lock);
297         srcu_read_unlock(&kvm->srcu, idx);
298
299         /* we've to flush the tlb before the pages can be freed */
300         if (need_tlb_flush)
301                 kvm_flush_remote_tlbs(kvm);
302 }
303
304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
305                                                   struct mm_struct *mm,
306                                                   unsigned long start,
307                                                   unsigned long end)
308 {
309         struct kvm *kvm = mmu_notifier_to_kvm(mn);
310
311         spin_lock(&kvm->mmu_lock);
312         /*
313          * This sequence increase will notify the kvm page fault that
314          * the page that is going to be mapped in the spte could have
315          * been freed.
316          */
317         kvm->mmu_notifier_seq++;
318         /*
319          * The above sequence increase must be visible before the
320          * below count decrease but both values are read by the kvm
321          * page fault under mmu_lock spinlock so we don't need to add
322          * a smb_wmb() here in between the two.
323          */
324         kvm->mmu_notifier_count--;
325         spin_unlock(&kvm->mmu_lock);
326
327         BUG_ON(kvm->mmu_notifier_count < 0);
328 }
329
330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
331                                               struct mm_struct *mm,
332                                               unsigned long address)
333 {
334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
335         int young, idx;
336
337         idx = srcu_read_lock(&kvm->srcu);
338         spin_lock(&kvm->mmu_lock);
339         young = kvm_age_hva(kvm, address);
340         spin_unlock(&kvm->mmu_lock);
341         srcu_read_unlock(&kvm->srcu, idx);
342
343         if (young)
344                 kvm_flush_remote_tlbs(kvm);
345
346         return young;
347 }
348
349 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
350                                      struct mm_struct *mm)
351 {
352         struct kvm *kvm = mmu_notifier_to_kvm(mn);
353         int idx;
354
355         idx = srcu_read_lock(&kvm->srcu);
356         kvm_arch_flush_shadow(kvm);
357         srcu_read_unlock(&kvm->srcu, idx);
358 }
359
360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
361         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
362         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
363         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
364         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
365         .change_pte             = kvm_mmu_notifier_change_pte,
366         .release                = kvm_mmu_notifier_release,
367 };
368
369 static int kvm_init_mmu_notifier(struct kvm *kvm)
370 {
371         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
372         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
373 }
374
375 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
376
377 static int kvm_init_mmu_notifier(struct kvm *kvm)
378 {
379         return 0;
380 }
381
382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
383
384 static struct kvm *kvm_create_vm(void)
385 {
386         int r = 0, i;
387         struct kvm *kvm = kvm_arch_create_vm();
388
389         if (IS_ERR(kvm))
390                 goto out;
391
392         r = hardware_enable_all();
393         if (r)
394                 goto out_err_nodisable;
395
396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
397         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
398         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
399 #endif
400
401         r = -ENOMEM;
402         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
403         if (!kvm->memslots)
404                 goto out_err_nosrcu;
405         if (init_srcu_struct(&kvm->srcu))
406                 goto out_err_nosrcu;
407         for (i = 0; i < KVM_NR_BUSES; i++) {
408                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
409                                         GFP_KERNEL);
410                 if (!kvm->buses[i])
411                         goto out_err;
412         }
413
414         r = kvm_init_mmu_notifier(kvm);
415         if (r)
416                 goto out_err;
417
418         kvm->mm = current->mm;
419         atomic_inc(&kvm->mm->mm_count);
420         spin_lock_init(&kvm->mmu_lock);
421         raw_spin_lock_init(&kvm->requests_lock);
422         kvm_eventfd_init(kvm);
423         mutex_init(&kvm->lock);
424         mutex_init(&kvm->irq_lock);
425         mutex_init(&kvm->slots_lock);
426         atomic_set(&kvm->users_count, 1);
427         spin_lock(&kvm_lock);
428         list_add(&kvm->vm_list, &vm_list);
429         spin_unlock(&kvm_lock);
430 out:
431         return kvm;
432
433 out_err:
434         cleanup_srcu_struct(&kvm->srcu);
435 out_err_nosrcu:
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 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
446 {
447         if (!memslot->dirty_bitmap)
448                 return;
449
450         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
451                 vfree(memslot->dirty_bitmap_head);
452         else
453                 kfree(memslot->dirty_bitmap_head);
454
455         memslot->dirty_bitmap = NULL;
456         memslot->dirty_bitmap_head = NULL;
457 }
458
459 /*
460  * Free any memory in @free but not in @dont.
461  */
462 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
463                                   struct kvm_memory_slot *dont)
464 {
465         int i;
466
467         if (!dont || free->rmap != dont->rmap)
468                 vfree(free->rmap);
469
470         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
471                 kvm_destroy_dirty_bitmap(free);
472
473
474         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
475                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
476                         vfree(free->lpage_info[i]);
477                         free->lpage_info[i] = NULL;
478                 }
479         }
480
481         free->npages = 0;
482         free->rmap = NULL;
483 }
484
485 void kvm_free_physmem(struct kvm *kvm)
486 {
487         int i;
488         struct kvm_memslots *slots = kvm->memslots;
489
490         for (i = 0; i < slots->nmemslots; ++i)
491                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
492
493         kfree(kvm->memslots);
494 }
495
496 static void kvm_destroy_vm(struct kvm *kvm)
497 {
498         int i;
499         struct mm_struct *mm = kvm->mm;
500
501         kvm_arch_sync_events(kvm);
502         spin_lock(&kvm_lock);
503         list_del(&kvm->vm_list);
504         spin_unlock(&kvm_lock);
505         kvm_free_irq_routing(kvm);
506         for (i = 0; i < KVM_NR_BUSES; i++)
507                 kvm_io_bus_destroy(kvm->buses[i]);
508         kvm_coalesced_mmio_free(kvm);
509 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
510         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
511 #else
512         kvm_arch_flush_shadow(kvm);
513 #endif
514         kvm_arch_destroy_vm(kvm);
515         hardware_disable_all();
516         mmdrop(mm);
517 }
518
519 void kvm_get_kvm(struct kvm *kvm)
520 {
521         atomic_inc(&kvm->users_count);
522 }
523 EXPORT_SYMBOL_GPL(kvm_get_kvm);
524
525 void kvm_put_kvm(struct kvm *kvm)
526 {
527         if (atomic_dec_and_test(&kvm->users_count))
528                 kvm_destroy_vm(kvm);
529 }
530 EXPORT_SYMBOL_GPL(kvm_put_kvm);
531
532
533 static int kvm_vm_release(struct inode *inode, struct file *filp)
534 {
535         struct kvm *kvm = filp->private_data;
536
537         kvm_irqfd_release(kvm);
538
539         kvm_put_kvm(kvm);
540         return 0;
541 }
542
543 /*
544  * Allocation size is twice as large as the actual dirty bitmap size.
545  * This makes it possible to do double buffering: see x86's
546  * kvm_vm_ioctl_get_dirty_log().
547  */
548 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
549 {
550         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
551
552         if (dirty_bytes > PAGE_SIZE)
553                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
554         else
555                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
556
557         if (!memslot->dirty_bitmap)
558                 return -ENOMEM;
559
560         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
561         return 0;
562 }
563
564 /*
565  * Allocate some memory and give it an address in the guest physical address
566  * space.
567  *
568  * Discontiguous memory is allowed, mostly for framebuffers.
569  *
570  * Must be called holding mmap_sem for write.
571  */
572 int __kvm_set_memory_region(struct kvm *kvm,
573                             struct kvm_userspace_memory_region *mem,
574                             int user_alloc)
575 {
576         int r, flush_shadow = 0;
577         gfn_t base_gfn;
578         unsigned long npages;
579         unsigned long i;
580         struct kvm_memory_slot *memslot;
581         struct kvm_memory_slot old, new;
582         struct kvm_memslots *slots, *old_memslots;
583
584         r = -EINVAL;
585         /* General sanity checks */
586         if (mem->memory_size & (PAGE_SIZE - 1))
587                 goto out;
588         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
589                 goto out;
590         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
591                 goto out;
592         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
593                 goto out;
594         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
595                 goto out;
596
597         memslot = &kvm->memslots->memslots[mem->slot];
598         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
599         npages = mem->memory_size >> PAGE_SHIFT;
600
601         r = -EINVAL;
602         if (npages > KVM_MEM_MAX_NR_PAGES)
603                 goto out;
604
605         if (!npages)
606                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
607
608         new = old = *memslot;
609
610         new.id = mem->slot;
611         new.base_gfn = base_gfn;
612         new.npages = npages;
613         new.flags = mem->flags;
614
615         /* Disallow changing a memory slot's size. */
616         r = -EINVAL;
617         if (npages && old.npages && npages != old.npages)
618                 goto out_free;
619
620         /* Check for overlaps */
621         r = -EEXIST;
622         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
623                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
624
625                 if (s == memslot || !s->npages)
626                         continue;
627                 if (!((base_gfn + npages <= s->base_gfn) ||
628                       (base_gfn >= s->base_gfn + s->npages)))
629                         goto out_free;
630         }
631
632         /* Free page dirty bitmap if unneeded */
633         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
634                 new.dirty_bitmap = NULL;
635
636         r = -ENOMEM;
637
638         /* Allocate if a slot is being created */
639 #ifndef CONFIG_S390
640         if (npages && !new.rmap) {
641                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
642
643                 if (!new.rmap)
644                         goto out_free;
645
646                 new.user_alloc = user_alloc;
647                 new.userspace_addr = mem->userspace_addr;
648         }
649         if (!npages)
650                 goto skip_lpage;
651
652         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
653                 unsigned long ugfn;
654                 unsigned long j;
655                 int lpages;
656                 int level = i + 2;
657
658                 /* Avoid unused variable warning if no large pages */
659                 (void)level;
660
661                 if (new.lpage_info[i])
662                         continue;
663
664                 lpages = 1 + ((base_gfn + npages - 1)
665                              >> KVM_HPAGE_GFN_SHIFT(level));
666                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
667
668                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
669
670                 if (!new.lpage_info[i])
671                         goto out_free;
672
673                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
674                         new.lpage_info[i][0].write_count = 1;
675                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
676                         new.lpage_info[i][lpages - 1].write_count = 1;
677                 ugfn = new.userspace_addr >> PAGE_SHIFT;
678                 /*
679                  * If the gfn and userspace address are not aligned wrt each
680                  * other, or if explicitly asked to, disable large page
681                  * support for this slot
682                  */
683                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
684                     !largepages_enabled)
685                         for (j = 0; j < lpages; ++j)
686                                 new.lpage_info[i][j].write_count = 1;
687         }
688
689 skip_lpage:
690
691         /* Allocate page dirty bitmap if needed */
692         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
693                 if (kvm_create_dirty_bitmap(&new) < 0)
694                         goto out_free;
695                 /* destroy any largepage mappings for dirty tracking */
696                 if (old.npages)
697                         flush_shadow = 1;
698         }
699 #else  /* not defined CONFIG_S390 */
700         new.user_alloc = user_alloc;
701         if (user_alloc)
702                 new.userspace_addr = mem->userspace_addr;
703 #endif /* not defined CONFIG_S390 */
704
705         if (!npages) {
706                 r = -ENOMEM;
707                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
708                 if (!slots)
709                         goto out_free;
710                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
711                 if (mem->slot >= slots->nmemslots)
712                         slots->nmemslots = mem->slot + 1;
713                 slots->generation++;
714                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
715
716                 old_memslots = kvm->memslots;
717                 rcu_assign_pointer(kvm->memslots, slots);
718                 synchronize_srcu_expedited(&kvm->srcu);
719                 /* From this point no new shadow pages pointing to a deleted
720                  * memslot will be created.
721                  *
722                  * validation of sp->gfn happens in:
723                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
724                  *      - kvm_is_visible_gfn (mmu_check_roots)
725                  */
726                 kvm_arch_flush_shadow(kvm);
727                 kfree(old_memslots);
728         }
729
730         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
731         if (r)
732                 goto out_free;
733
734         /* map the pages in iommu page table */
735         if (npages) {
736                 r = kvm_iommu_map_pages(kvm, &new);
737                 if (r)
738                         goto out_free;
739         }
740
741         r = -ENOMEM;
742         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
743         if (!slots)
744                 goto out_free;
745         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
746         if (mem->slot >= slots->nmemslots)
747                 slots->nmemslots = mem->slot + 1;
748         slots->generation++;
749
750         /* actual memory is freed via old in kvm_free_physmem_slot below */
751         if (!npages) {
752                 new.rmap = NULL;
753                 new.dirty_bitmap = NULL;
754                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
755                         new.lpage_info[i] = NULL;
756         }
757
758         slots->memslots[mem->slot] = new;
759         old_memslots = kvm->memslots;
760         rcu_assign_pointer(kvm->memslots, slots);
761         synchronize_srcu_expedited(&kvm->srcu);
762
763         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
764
765         kvm_free_physmem_slot(&old, &new);
766         kfree(old_memslots);
767
768         if (flush_shadow)
769                 kvm_arch_flush_shadow(kvm);
770
771         return 0;
772
773 out_free:
774         kvm_free_physmem_slot(&new, &old);
775 out:
776         return r;
777
778 }
779 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
780
781 int kvm_set_memory_region(struct kvm *kvm,
782                           struct kvm_userspace_memory_region *mem,
783                           int user_alloc)
784 {
785         int r;
786
787         mutex_lock(&kvm->slots_lock);
788         r = __kvm_set_memory_region(kvm, mem, user_alloc);
789         mutex_unlock(&kvm->slots_lock);
790         return r;
791 }
792 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
793
794 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
795                                    struct
796                                    kvm_userspace_memory_region *mem,
797                                    int user_alloc)
798 {
799         if (mem->slot >= KVM_MEMORY_SLOTS)
800                 return -EINVAL;
801         return kvm_set_memory_region(kvm, mem, user_alloc);
802 }
803
804 int kvm_get_dirty_log(struct kvm *kvm,
805                         struct kvm_dirty_log *log, int *is_dirty)
806 {
807         struct kvm_memory_slot *memslot;
808         int r, i;
809         unsigned long n;
810         unsigned long any = 0;
811
812         r = -EINVAL;
813         if (log->slot >= KVM_MEMORY_SLOTS)
814                 goto out;
815
816         memslot = &kvm->memslots->memslots[log->slot];
817         r = -ENOENT;
818         if (!memslot->dirty_bitmap)
819                 goto out;
820
821         n = kvm_dirty_bitmap_bytes(memslot);
822
823         for (i = 0; !any && i < n/sizeof(long); ++i)
824                 any = memslot->dirty_bitmap[i];
825
826         r = -EFAULT;
827         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
828                 goto out;
829
830         if (any)
831                 *is_dirty = 1;
832
833         r = 0;
834 out:
835         return r;
836 }
837
838 void kvm_disable_largepages(void)
839 {
840         largepages_enabled = false;
841 }
842 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
843
844 int is_error_page(struct page *page)
845 {
846         return page == bad_page || page == hwpoison_page || page == fault_page;
847 }
848 EXPORT_SYMBOL_GPL(is_error_page);
849
850 int is_error_pfn(pfn_t pfn)
851 {
852         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
853 }
854 EXPORT_SYMBOL_GPL(is_error_pfn);
855
856 int is_hwpoison_pfn(pfn_t pfn)
857 {
858         return pfn == hwpoison_pfn;
859 }
860 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
861
862 int is_fault_pfn(pfn_t pfn)
863 {
864         return pfn == fault_pfn;
865 }
866 EXPORT_SYMBOL_GPL(is_fault_pfn);
867
868 static inline unsigned long bad_hva(void)
869 {
870         return PAGE_OFFSET;
871 }
872
873 int kvm_is_error_hva(unsigned long addr)
874 {
875         return addr == bad_hva();
876 }
877 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
878
879 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
880                                                 gfn_t gfn)
881 {
882         int i;
883
884         for (i = 0; i < slots->nmemslots; ++i) {
885                 struct kvm_memory_slot *memslot = &slots->memslots[i];
886
887                 if (gfn >= memslot->base_gfn
888                     && gfn < memslot->base_gfn + memslot->npages)
889                         return memslot;
890         }
891         return NULL;
892 }
893
894 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
895 {
896         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
897 }
898 EXPORT_SYMBOL_GPL(gfn_to_memslot);
899
900 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
901 {
902         int i;
903         struct kvm_memslots *slots = kvm_memslots(kvm);
904
905         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
906                 struct kvm_memory_slot *memslot = &slots->memslots[i];
907
908                 if (memslot->flags & KVM_MEMSLOT_INVALID)
909                         continue;
910
911                 if (gfn >= memslot->base_gfn
912                     && gfn < memslot->base_gfn + memslot->npages)
913                         return 1;
914         }
915         return 0;
916 }
917 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
918
919 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
920 {
921         struct vm_area_struct *vma;
922         unsigned long addr, size;
923
924         size = PAGE_SIZE;
925
926         addr = gfn_to_hva(kvm, gfn);
927         if (kvm_is_error_hva(addr))
928                 return PAGE_SIZE;
929
930         down_read(&current->mm->mmap_sem);
931         vma = find_vma(current->mm, addr);
932         if (!vma)
933                 goto out;
934
935         size = vma_kernel_pagesize(vma);
936
937 out:
938         up_read(&current->mm->mmap_sem);
939
940         return size;
941 }
942
943 int memslot_id(struct kvm *kvm, gfn_t gfn)
944 {
945         int i;
946         struct kvm_memslots *slots = kvm_memslots(kvm);
947         struct kvm_memory_slot *memslot = NULL;
948
949         for (i = 0; i < slots->nmemslots; ++i) {
950                 memslot = &slots->memslots[i];
951
952                 if (gfn >= memslot->base_gfn
953                     && gfn < memslot->base_gfn + memslot->npages)
954                         break;
955         }
956
957         return memslot - slots->memslots;
958 }
959
960 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
961                                      gfn_t *nr_pages)
962 {
963         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
964                 return bad_hva();
965
966         if (nr_pages)
967                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
968
969         return gfn_to_hva_memslot(slot, gfn);
970 }
971
972 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
973 {
974         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
975 }
976 EXPORT_SYMBOL_GPL(gfn_to_hva);
977
978 static pfn_t get_fault_pfn(void)
979 {
980         get_page(fault_page);
981         return fault_pfn;
982 }
983
984 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
985                         bool *async, bool write_fault, bool *writable)
986 {
987         struct page *page[1];
988         int npages = 0;
989         pfn_t pfn;
990
991         /* we can do it either atomically or asynchronously, not both */
992         BUG_ON(atomic && async);
993
994         BUG_ON(!write_fault && !writable);
995
996         if (writable)
997                 *writable = true;
998
999         if (atomic || async)
1000                 npages = __get_user_pages_fast(addr, 1, 1, page);
1001
1002         if (unlikely(npages != 1) && !atomic) {
1003                 might_sleep();
1004
1005                 if (writable)
1006                         *writable = write_fault;
1007
1008                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1009
1010                 /* map read fault as writable if possible */
1011                 if (unlikely(!write_fault) && npages == 1) {
1012                         struct page *wpage[1];
1013
1014                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1015                         if (npages == 1) {
1016                                 *writable = true;
1017                                 put_page(page[0]);
1018                                 page[0] = wpage[0];
1019                         }
1020                         npages = 1;
1021                 }
1022         }
1023
1024         if (unlikely(npages != 1)) {
1025                 struct vm_area_struct *vma;
1026
1027                 if (atomic)
1028                         return get_fault_pfn();
1029
1030                 down_read(&current->mm->mmap_sem);
1031                 if (is_hwpoison_address(addr)) {
1032                         up_read(&current->mm->mmap_sem);
1033                         get_page(hwpoison_page);
1034                         return page_to_pfn(hwpoison_page);
1035                 }
1036
1037                 vma = find_vma_intersection(current->mm, addr, addr+1);
1038
1039                 if (vma == NULL)
1040                         pfn = get_fault_pfn();
1041                 else if ((vma->vm_flags & VM_PFNMAP)) {
1042                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1043                                 vma->vm_pgoff;
1044                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1045                 } else {
1046                         if (async && (vma->vm_flags & VM_WRITE))
1047                                 *async = true;
1048                         pfn = get_fault_pfn();
1049                 }
1050                 up_read(&current->mm->mmap_sem);
1051         } else
1052                 pfn = page_to_pfn(page[0]);
1053
1054         return pfn;
1055 }
1056
1057 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1058 {
1059         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1060 }
1061 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1062
1063 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1064                           bool write_fault, bool *writable)
1065 {
1066         unsigned long addr;
1067
1068         if (async)
1069                 *async = false;
1070
1071         addr = gfn_to_hva(kvm, gfn);
1072         if (kvm_is_error_hva(addr)) {
1073                 get_page(bad_page);
1074                 return page_to_pfn(bad_page);
1075         }
1076
1077         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1078 }
1079
1080 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1081 {
1082         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1083 }
1084 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1085
1086 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1087                        bool write_fault, bool *writable)
1088 {
1089         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1090 }
1091 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1092
1093 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1094 {
1095         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1096 }
1097 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1098
1099 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1100                       bool *writable)
1101 {
1102         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1103 }
1104 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1105
1106 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1107                          struct kvm_memory_slot *slot, gfn_t gfn)
1108 {
1109         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1110         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1111 }
1112
1113 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1114                                                                   int nr_pages)
1115 {
1116         unsigned long addr;
1117         gfn_t entry;
1118
1119         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1120         if (kvm_is_error_hva(addr))
1121                 return -1;
1122
1123         if (entry < nr_pages)
1124                 return 0;
1125
1126         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1127 }
1128 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1129
1130 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1131 {
1132         pfn_t pfn;
1133
1134         pfn = gfn_to_pfn(kvm, gfn);
1135         if (!kvm_is_mmio_pfn(pfn))
1136                 return pfn_to_page(pfn);
1137
1138         WARN_ON(kvm_is_mmio_pfn(pfn));
1139
1140         get_page(bad_page);
1141         return bad_page;
1142 }
1143
1144 EXPORT_SYMBOL_GPL(gfn_to_page);
1145
1146 void kvm_release_page_clean(struct page *page)
1147 {
1148         kvm_release_pfn_clean(page_to_pfn(page));
1149 }
1150 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1151
1152 void kvm_release_pfn_clean(pfn_t pfn)
1153 {
1154         if (!kvm_is_mmio_pfn(pfn))
1155                 put_page(pfn_to_page(pfn));
1156 }
1157 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1158
1159 void kvm_release_page_dirty(struct page *page)
1160 {
1161         kvm_release_pfn_dirty(page_to_pfn(page));
1162 }
1163 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1164
1165 void kvm_release_pfn_dirty(pfn_t pfn)
1166 {
1167         kvm_set_pfn_dirty(pfn);
1168         kvm_release_pfn_clean(pfn);
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1171
1172 void kvm_set_page_dirty(struct page *page)
1173 {
1174         kvm_set_pfn_dirty(page_to_pfn(page));
1175 }
1176 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1177
1178 void kvm_set_pfn_dirty(pfn_t pfn)
1179 {
1180         if (!kvm_is_mmio_pfn(pfn)) {
1181                 struct page *page = pfn_to_page(pfn);
1182                 if (!PageReserved(page))
1183                         SetPageDirty(page);
1184         }
1185 }
1186 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1187
1188 void kvm_set_pfn_accessed(pfn_t pfn)
1189 {
1190         if (!kvm_is_mmio_pfn(pfn))
1191                 mark_page_accessed(pfn_to_page(pfn));
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1194
1195 void kvm_get_pfn(pfn_t pfn)
1196 {
1197         if (!kvm_is_mmio_pfn(pfn))
1198                 get_page(pfn_to_page(pfn));
1199 }
1200 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1201
1202 static int next_segment(unsigned long len, int offset)
1203 {
1204         if (len > PAGE_SIZE - offset)
1205                 return PAGE_SIZE - offset;
1206         else
1207                 return len;
1208 }
1209
1210 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1211                         int len)
1212 {
1213         int r;
1214         unsigned long addr;
1215
1216         addr = gfn_to_hva(kvm, gfn);
1217         if (kvm_is_error_hva(addr))
1218                 return -EFAULT;
1219         r = copy_from_user(data, (void __user *)addr + offset, len);
1220         if (r)
1221                 return -EFAULT;
1222         return 0;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1225
1226 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1227 {
1228         gfn_t gfn = gpa >> PAGE_SHIFT;
1229         int seg;
1230         int offset = offset_in_page(gpa);
1231         int ret;
1232
1233         while ((seg = next_segment(len, offset)) != 0) {
1234                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1235                 if (ret < 0)
1236                         return ret;
1237                 offset = 0;
1238                 len -= seg;
1239                 data += seg;
1240                 ++gfn;
1241         }
1242         return 0;
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_read_guest);
1245
1246 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1247                           unsigned long len)
1248 {
1249         int r;
1250         unsigned long addr;
1251         gfn_t gfn = gpa >> PAGE_SHIFT;
1252         int offset = offset_in_page(gpa);
1253
1254         addr = gfn_to_hva(kvm, gfn);
1255         if (kvm_is_error_hva(addr))
1256                 return -EFAULT;
1257         pagefault_disable();
1258         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1259         pagefault_enable();
1260         if (r)
1261                 return -EFAULT;
1262         return 0;
1263 }
1264 EXPORT_SYMBOL(kvm_read_guest_atomic);
1265
1266 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1267                          int offset, int len)
1268 {
1269         int r;
1270         unsigned long addr;
1271
1272         addr = gfn_to_hva(kvm, gfn);
1273         if (kvm_is_error_hva(addr))
1274                 return -EFAULT;
1275         r = copy_to_user((void __user *)addr + offset, data, len);
1276         if (r)
1277                 return -EFAULT;
1278         mark_page_dirty(kvm, gfn);
1279         return 0;
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1282
1283 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1284                     unsigned long len)
1285 {
1286         gfn_t gfn = gpa >> PAGE_SHIFT;
1287         int seg;
1288         int offset = offset_in_page(gpa);
1289         int ret;
1290
1291         while ((seg = next_segment(len, offset)) != 0) {
1292                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1293                 if (ret < 0)
1294                         return ret;
1295                 offset = 0;
1296                 len -= seg;
1297                 data += seg;
1298                 ++gfn;
1299         }
1300         return 0;
1301 }
1302
1303 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1304                               gpa_t gpa)
1305 {
1306         struct kvm_memslots *slots = kvm_memslots(kvm);
1307         int offset = offset_in_page(gpa);
1308         gfn_t gfn = gpa >> PAGE_SHIFT;
1309
1310         ghc->gpa = gpa;
1311         ghc->generation = slots->generation;
1312         ghc->memslot = __gfn_to_memslot(slots, gfn);
1313         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1314         if (!kvm_is_error_hva(ghc->hva))
1315                 ghc->hva += offset;
1316         else
1317                 return -EFAULT;
1318
1319         return 0;
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1322
1323 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1324                            void *data, unsigned long len)
1325 {
1326         struct kvm_memslots *slots = kvm_memslots(kvm);
1327         int r;
1328
1329         if (slots->generation != ghc->generation)
1330                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1331
1332         if (kvm_is_error_hva(ghc->hva))
1333                 return -EFAULT;
1334
1335         r = copy_to_user((void __user *)ghc->hva, data, len);
1336         if (r)
1337                 return -EFAULT;
1338         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1339
1340         return 0;
1341 }
1342 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1343
1344 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1345 {
1346         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1347                                     offset, len);
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1350
1351 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1352 {
1353         gfn_t gfn = gpa >> PAGE_SHIFT;
1354         int seg;
1355         int offset = offset_in_page(gpa);
1356         int ret;
1357
1358         while ((seg = next_segment(len, offset)) != 0) {
1359                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1360                 if (ret < 0)
1361                         return ret;
1362                 offset = 0;
1363                 len -= seg;
1364                 ++gfn;
1365         }
1366         return 0;
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1369
1370 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1371                              gfn_t gfn)
1372 {
1373         if (memslot && memslot->dirty_bitmap) {
1374                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1375
1376                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1377         }
1378 }
1379
1380 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1381 {
1382         struct kvm_memory_slot *memslot;
1383
1384         memslot = gfn_to_memslot(kvm, gfn);
1385         mark_page_dirty_in_slot(kvm, memslot, gfn);
1386 }
1387
1388 /*
1389  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1390  */
1391 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1392 {
1393         DEFINE_WAIT(wait);
1394
1395         for (;;) {
1396                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1397
1398                 if (kvm_arch_vcpu_runnable(vcpu)) {
1399                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1400                         break;
1401                 }
1402                 if (kvm_cpu_has_pending_timer(vcpu))
1403                         break;
1404                 if (signal_pending(current))
1405                         break;
1406
1407                 schedule();
1408         }
1409
1410         finish_wait(&vcpu->wq, &wait);
1411 }
1412
1413 void kvm_resched(struct kvm_vcpu *vcpu)
1414 {
1415         if (!need_resched())
1416                 return;
1417         cond_resched();
1418 }
1419 EXPORT_SYMBOL_GPL(kvm_resched);
1420
1421 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1422 {
1423         ktime_t expires;
1424         DEFINE_WAIT(wait);
1425
1426         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1427
1428         /* Sleep for 100 us, and hope lock-holder got scheduled */
1429         expires = ktime_add_ns(ktime_get(), 100000UL);
1430         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1431
1432         finish_wait(&vcpu->wq, &wait);
1433 }
1434 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1435
1436 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1437 {
1438         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1439         struct page *page;
1440
1441         if (vmf->pgoff == 0)
1442                 page = virt_to_page(vcpu->run);
1443 #ifdef CONFIG_X86
1444         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1445                 page = virt_to_page(vcpu->arch.pio_data);
1446 #endif
1447 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1448         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1449                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1450 #endif
1451         else
1452                 return VM_FAULT_SIGBUS;
1453         get_page(page);
1454         vmf->page = page;
1455         return 0;
1456 }
1457
1458 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1459         .fault = kvm_vcpu_fault,
1460 };
1461
1462 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1463 {
1464         vma->vm_ops = &kvm_vcpu_vm_ops;
1465         return 0;
1466 }
1467
1468 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1469 {
1470         struct kvm_vcpu *vcpu = filp->private_data;
1471
1472         kvm_put_kvm(vcpu->kvm);
1473         return 0;
1474 }
1475
1476 static struct file_operations kvm_vcpu_fops = {
1477         .release        = kvm_vcpu_release,
1478         .unlocked_ioctl = kvm_vcpu_ioctl,
1479         .compat_ioctl   = kvm_vcpu_ioctl,
1480         .mmap           = kvm_vcpu_mmap,
1481         .llseek         = noop_llseek,
1482 };
1483
1484 /*
1485  * Allocates an inode for the vcpu.
1486  */
1487 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1488 {
1489         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1490 }
1491
1492 /*
1493  * Creates some virtual cpus.  Good luck creating more than one.
1494  */
1495 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1496 {
1497         int r;
1498         struct kvm_vcpu *vcpu, *v;
1499
1500         vcpu = kvm_arch_vcpu_create(kvm, id);
1501         if (IS_ERR(vcpu))
1502                 return PTR_ERR(vcpu);
1503
1504         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1505
1506         r = kvm_arch_vcpu_setup(vcpu);
1507         if (r)
1508                 return r;
1509
1510         mutex_lock(&kvm->lock);
1511         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1512                 r = -EINVAL;
1513                 goto vcpu_destroy;
1514         }
1515
1516         kvm_for_each_vcpu(r, v, kvm)
1517                 if (v->vcpu_id == id) {
1518                         r = -EEXIST;
1519                         goto vcpu_destroy;
1520                 }
1521
1522         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1523
1524         /* Now it's all set up, let userspace reach it */
1525         kvm_get_kvm(kvm);
1526         r = create_vcpu_fd(vcpu);
1527         if (r < 0) {
1528                 kvm_put_kvm(kvm);
1529                 goto vcpu_destroy;
1530         }
1531
1532         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1533         smp_wmb();
1534         atomic_inc(&kvm->online_vcpus);
1535
1536 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1537         if (kvm->bsp_vcpu_id == id)
1538                 kvm->bsp_vcpu = vcpu;
1539 #endif
1540         mutex_unlock(&kvm->lock);
1541         return r;
1542
1543 vcpu_destroy:
1544         mutex_unlock(&kvm->lock);
1545         kvm_arch_vcpu_destroy(vcpu);
1546         return r;
1547 }
1548
1549 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1550 {
1551         if (sigset) {
1552                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1553                 vcpu->sigset_active = 1;
1554                 vcpu->sigset = *sigset;
1555         } else
1556                 vcpu->sigset_active = 0;
1557         return 0;
1558 }
1559
1560 static long kvm_vcpu_ioctl(struct file *filp,
1561                            unsigned int ioctl, unsigned long arg)
1562 {
1563         struct kvm_vcpu *vcpu = filp->private_data;
1564         void __user *argp = (void __user *)arg;
1565         int r;
1566         struct kvm_fpu *fpu = NULL;
1567         struct kvm_sregs *kvm_sregs = NULL;
1568
1569         if (vcpu->kvm->mm != current->mm)
1570                 return -EIO;
1571
1572 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1573         /*
1574          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1575          * so vcpu_load() would break it.
1576          */
1577         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1578                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1579 #endif
1580
1581
1582         vcpu_load(vcpu);
1583         switch (ioctl) {
1584         case KVM_RUN:
1585                 r = -EINVAL;
1586                 if (arg)
1587                         goto out;
1588                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1589                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1590                 break;
1591         case KVM_GET_REGS: {
1592                 struct kvm_regs *kvm_regs;
1593
1594                 r = -ENOMEM;
1595                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1596                 if (!kvm_regs)
1597                         goto out;
1598                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1599                 if (r)
1600                         goto out_free1;
1601                 r = -EFAULT;
1602                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1603                         goto out_free1;
1604                 r = 0;
1605 out_free1:
1606                 kfree(kvm_regs);
1607                 break;
1608         }
1609         case KVM_SET_REGS: {
1610                 struct kvm_regs *kvm_regs;
1611
1612                 r = -ENOMEM;
1613                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1614                 if (!kvm_regs)
1615                         goto out;
1616                 r = -EFAULT;
1617                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1618                         goto out_free2;
1619                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1620                 if (r)
1621                         goto out_free2;
1622                 r = 0;
1623 out_free2:
1624                 kfree(kvm_regs);
1625                 break;
1626         }
1627         case KVM_GET_SREGS: {
1628                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1629                 r = -ENOMEM;
1630                 if (!kvm_sregs)
1631                         goto out;
1632                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1633                 if (r)
1634                         goto out;
1635                 r = -EFAULT;
1636                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1637                         goto out;
1638                 r = 0;
1639                 break;
1640         }
1641         case KVM_SET_SREGS: {
1642                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1643                 r = -ENOMEM;
1644                 if (!kvm_sregs)
1645                         goto out;
1646                 r = -EFAULT;
1647                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1648                         goto out;
1649                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1650                 if (r)
1651                         goto out;
1652                 r = 0;
1653                 break;
1654         }
1655         case KVM_GET_MP_STATE: {
1656                 struct kvm_mp_state mp_state;
1657
1658                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1659                 if (r)
1660                         goto out;
1661                 r = -EFAULT;
1662                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1663                         goto out;
1664                 r = 0;
1665                 break;
1666         }
1667         case KVM_SET_MP_STATE: {
1668                 struct kvm_mp_state mp_state;
1669
1670                 r = -EFAULT;
1671                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1672                         goto out;
1673                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1674                 if (r)
1675                         goto out;
1676                 r = 0;
1677                 break;
1678         }
1679         case KVM_TRANSLATE: {
1680                 struct kvm_translation tr;
1681
1682                 r = -EFAULT;
1683                 if (copy_from_user(&tr, argp, sizeof tr))
1684                         goto out;
1685                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1686                 if (r)
1687                         goto out;
1688                 r = -EFAULT;
1689                 if (copy_to_user(argp, &tr, sizeof tr))
1690                         goto out;
1691                 r = 0;
1692                 break;
1693         }
1694         case KVM_SET_GUEST_DEBUG: {
1695                 struct kvm_guest_debug dbg;
1696
1697                 r = -EFAULT;
1698                 if (copy_from_user(&dbg, argp, sizeof dbg))
1699                         goto out;
1700                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1701                 if (r)
1702                         goto out;
1703                 r = 0;
1704                 break;
1705         }
1706         case KVM_SET_SIGNAL_MASK: {
1707                 struct kvm_signal_mask __user *sigmask_arg = argp;
1708                 struct kvm_signal_mask kvm_sigmask;
1709                 sigset_t sigset, *p;
1710
1711                 p = NULL;
1712                 if (argp) {
1713                         r = -EFAULT;
1714                         if (copy_from_user(&kvm_sigmask, argp,
1715                                            sizeof kvm_sigmask))
1716                                 goto out;
1717                         r = -EINVAL;
1718                         if (kvm_sigmask.len != sizeof sigset)
1719                                 goto out;
1720                         r = -EFAULT;
1721                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1722                                            sizeof sigset))
1723                                 goto out;
1724                         p = &sigset;
1725                 }
1726                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1727                 break;
1728         }
1729         case KVM_GET_FPU: {
1730                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1731                 r = -ENOMEM;
1732                 if (!fpu)
1733                         goto out;
1734                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1735                 if (r)
1736                         goto out;
1737                 r = -EFAULT;
1738                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1739                         goto out;
1740                 r = 0;
1741                 break;
1742         }
1743         case KVM_SET_FPU: {
1744                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1745                 r = -ENOMEM;
1746                 if (!fpu)
1747                         goto out;
1748                 r = -EFAULT;
1749                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1750                         goto out;
1751                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1752                 if (r)
1753                         goto out;
1754                 r = 0;
1755                 break;
1756         }
1757         default:
1758                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1759         }
1760 out:
1761         vcpu_put(vcpu);
1762         kfree(fpu);
1763         kfree(kvm_sregs);
1764         return r;
1765 }
1766
1767 static long kvm_vm_ioctl(struct file *filp,
1768                            unsigned int ioctl, unsigned long arg)
1769 {
1770         struct kvm *kvm = filp->private_data;
1771         void __user *argp = (void __user *)arg;
1772         int r;
1773
1774         if (kvm->mm != current->mm)
1775                 return -EIO;
1776         switch (ioctl) {
1777         case KVM_CREATE_VCPU:
1778                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1779                 if (r < 0)
1780                         goto out;
1781                 break;
1782         case KVM_SET_USER_MEMORY_REGION: {
1783                 struct kvm_userspace_memory_region kvm_userspace_mem;
1784
1785                 r = -EFAULT;
1786                 if (copy_from_user(&kvm_userspace_mem, argp,
1787                                                 sizeof kvm_userspace_mem))
1788                         goto out;
1789
1790                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1791                 if (r)
1792                         goto out;
1793                 break;
1794         }
1795         case KVM_GET_DIRTY_LOG: {
1796                 struct kvm_dirty_log log;
1797
1798                 r = -EFAULT;
1799                 if (copy_from_user(&log, argp, sizeof log))
1800                         goto out;
1801                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1802                 if (r)
1803                         goto out;
1804                 break;
1805         }
1806 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1807         case KVM_REGISTER_COALESCED_MMIO: {
1808                 struct kvm_coalesced_mmio_zone zone;
1809                 r = -EFAULT;
1810                 if (copy_from_user(&zone, argp, sizeof zone))
1811                         goto out;
1812                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1813                 if (r)
1814                         goto out;
1815                 r = 0;
1816                 break;
1817         }
1818         case KVM_UNREGISTER_COALESCED_MMIO: {
1819                 struct kvm_coalesced_mmio_zone zone;
1820                 r = -EFAULT;
1821                 if (copy_from_user(&zone, argp, sizeof zone))
1822                         goto out;
1823                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1824                 if (r)
1825                         goto out;
1826                 r = 0;
1827                 break;
1828         }
1829 #endif
1830         case KVM_IRQFD: {
1831                 struct kvm_irqfd data;
1832
1833                 r = -EFAULT;
1834                 if (copy_from_user(&data, argp, sizeof data))
1835                         goto out;
1836                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1837                 break;
1838         }
1839         case KVM_IOEVENTFD: {
1840                 struct kvm_ioeventfd data;
1841
1842                 r = -EFAULT;
1843                 if (copy_from_user(&data, argp, sizeof data))
1844                         goto out;
1845                 r = kvm_ioeventfd(kvm, &data);
1846                 break;
1847         }
1848 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1849         case KVM_SET_BOOT_CPU_ID:
1850                 r = 0;
1851                 mutex_lock(&kvm->lock);
1852                 if (atomic_read(&kvm->online_vcpus) != 0)
1853                         r = -EBUSY;
1854                 else
1855                         kvm->bsp_vcpu_id = arg;
1856                 mutex_unlock(&kvm->lock);
1857                 break;
1858 #endif
1859         default:
1860                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1861                 if (r == -ENOTTY)
1862                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1863         }
1864 out:
1865         return r;
1866 }
1867
1868 #ifdef CONFIG_COMPAT
1869 struct compat_kvm_dirty_log {
1870         __u32 slot;
1871         __u32 padding1;
1872         union {
1873                 compat_uptr_t dirty_bitmap; /* one bit per page */
1874                 __u64 padding2;
1875         };
1876 };
1877
1878 static long kvm_vm_compat_ioctl(struct file *filp,
1879                            unsigned int ioctl, unsigned long arg)
1880 {
1881         struct kvm *kvm = filp->private_data;
1882         int r;
1883
1884         if (kvm->mm != current->mm)
1885                 return -EIO;
1886         switch (ioctl) {
1887         case KVM_GET_DIRTY_LOG: {
1888                 struct compat_kvm_dirty_log compat_log;
1889                 struct kvm_dirty_log log;
1890
1891                 r = -EFAULT;
1892                 if (copy_from_user(&compat_log, (void __user *)arg,
1893                                    sizeof(compat_log)))
1894                         goto out;
1895                 log.slot         = compat_log.slot;
1896                 log.padding1     = compat_log.padding1;
1897                 log.padding2     = compat_log.padding2;
1898                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1899
1900                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1901                 if (r)
1902                         goto out;
1903                 break;
1904         }
1905         default:
1906                 r = kvm_vm_ioctl(filp, ioctl, arg);
1907         }
1908
1909 out:
1910         return r;
1911 }
1912 #endif
1913
1914 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1915 {
1916         struct page *page[1];
1917         unsigned long addr;
1918         int npages;
1919         gfn_t gfn = vmf->pgoff;
1920         struct kvm *kvm = vma->vm_file->private_data;
1921
1922         addr = gfn_to_hva(kvm, gfn);
1923         if (kvm_is_error_hva(addr))
1924                 return VM_FAULT_SIGBUS;
1925
1926         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1927                                 NULL);
1928         if (unlikely(npages != 1))
1929                 return VM_FAULT_SIGBUS;
1930
1931         vmf->page = page[0];
1932         return 0;
1933 }
1934
1935 static const struct vm_operations_struct kvm_vm_vm_ops = {
1936         .fault = kvm_vm_fault,
1937 };
1938
1939 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1940 {
1941         vma->vm_ops = &kvm_vm_vm_ops;
1942         return 0;
1943 }
1944
1945 static struct file_operations kvm_vm_fops = {
1946         .release        = kvm_vm_release,
1947         .unlocked_ioctl = kvm_vm_ioctl,
1948 #ifdef CONFIG_COMPAT
1949         .compat_ioctl   = kvm_vm_compat_ioctl,
1950 #endif
1951         .mmap           = kvm_vm_mmap,
1952         .llseek         = noop_llseek,
1953 };
1954
1955 static int kvm_dev_ioctl_create_vm(void)
1956 {
1957         int r;
1958         struct kvm *kvm;
1959
1960         kvm = kvm_create_vm();
1961         if (IS_ERR(kvm))
1962                 return PTR_ERR(kvm);
1963 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1964         r = kvm_coalesced_mmio_init(kvm);
1965         if (r < 0) {
1966                 kvm_put_kvm(kvm);
1967                 return r;
1968         }
1969 #endif
1970         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1971         if (r < 0)
1972                 kvm_put_kvm(kvm);
1973
1974         return r;
1975 }
1976
1977 static long kvm_dev_ioctl_check_extension_generic(long arg)
1978 {
1979         switch (arg) {
1980         case KVM_CAP_USER_MEMORY:
1981         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1982         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1983 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1984         case KVM_CAP_SET_BOOT_CPU_ID:
1985 #endif
1986         case KVM_CAP_INTERNAL_ERROR_DATA:
1987                 return 1;
1988 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1989         case KVM_CAP_IRQ_ROUTING:
1990                 return KVM_MAX_IRQ_ROUTES;
1991 #endif
1992         default:
1993                 break;
1994         }
1995         return kvm_dev_ioctl_check_extension(arg);
1996 }
1997
1998 static long kvm_dev_ioctl(struct file *filp,
1999                           unsigned int ioctl, unsigned long arg)
2000 {
2001         long r = -EINVAL;
2002
2003         switch (ioctl) {
2004         case KVM_GET_API_VERSION:
2005                 r = -EINVAL;
2006                 if (arg)
2007                         goto out;
2008                 r = KVM_API_VERSION;
2009                 break;
2010         case KVM_CREATE_VM:
2011                 r = -EINVAL;
2012                 if (arg)
2013                         goto out;
2014                 r = kvm_dev_ioctl_create_vm();
2015                 break;
2016         case KVM_CHECK_EXTENSION:
2017                 r = kvm_dev_ioctl_check_extension_generic(arg);
2018                 break;
2019         case KVM_GET_VCPU_MMAP_SIZE:
2020                 r = -EINVAL;
2021                 if (arg)
2022                         goto out;
2023                 r = PAGE_SIZE;     /* struct kvm_run */
2024 #ifdef CONFIG_X86
2025                 r += PAGE_SIZE;    /* pio data page */
2026 #endif
2027 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2028                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2029 #endif
2030                 break;
2031         case KVM_TRACE_ENABLE:
2032         case KVM_TRACE_PAUSE:
2033         case KVM_TRACE_DISABLE:
2034                 r = -EOPNOTSUPP;
2035                 break;
2036         default:
2037                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2038         }
2039 out:
2040         return r;
2041 }
2042
2043 static struct file_operations kvm_chardev_ops = {
2044         .unlocked_ioctl = kvm_dev_ioctl,
2045         .compat_ioctl   = kvm_dev_ioctl,
2046         .llseek         = noop_llseek,
2047 };
2048
2049 static struct miscdevice kvm_dev = {
2050         KVM_MINOR,
2051         "kvm",
2052         &kvm_chardev_ops,
2053 };
2054
2055 static void hardware_enable(void *junk)
2056 {
2057         int cpu = raw_smp_processor_id();
2058         int r;
2059
2060         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2061                 return;
2062
2063         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2064
2065         r = kvm_arch_hardware_enable(NULL);
2066
2067         if (r) {
2068                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2069                 atomic_inc(&hardware_enable_failed);
2070                 printk(KERN_INFO "kvm: enabling virtualization on "
2071                                  "CPU%d failed\n", cpu);
2072         }
2073 }
2074
2075 static void hardware_disable(void *junk)
2076 {
2077         int cpu = raw_smp_processor_id();
2078
2079         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2080                 return;
2081         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2082         kvm_arch_hardware_disable(NULL);
2083 }
2084
2085 static void hardware_disable_all_nolock(void)
2086 {
2087         BUG_ON(!kvm_usage_count);
2088
2089         kvm_usage_count--;
2090         if (!kvm_usage_count)
2091                 on_each_cpu(hardware_disable, NULL, 1);
2092 }
2093
2094 static void hardware_disable_all(void)
2095 {
2096         spin_lock(&kvm_lock);
2097         hardware_disable_all_nolock();
2098         spin_unlock(&kvm_lock);
2099 }
2100
2101 static int hardware_enable_all(void)
2102 {
2103         int r = 0;
2104
2105         spin_lock(&kvm_lock);
2106
2107         kvm_usage_count++;
2108         if (kvm_usage_count == 1) {
2109                 atomic_set(&hardware_enable_failed, 0);
2110                 on_each_cpu(hardware_enable, NULL, 1);
2111
2112                 if (atomic_read(&hardware_enable_failed)) {
2113                         hardware_disable_all_nolock();
2114                         r = -EBUSY;
2115                 }
2116         }
2117
2118         spin_unlock(&kvm_lock);
2119
2120         return r;
2121 }
2122
2123 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2124                            void *v)
2125 {
2126         int cpu = (long)v;
2127
2128         if (!kvm_usage_count)
2129                 return NOTIFY_OK;
2130
2131         val &= ~CPU_TASKS_FROZEN;
2132         switch (val) {
2133         case CPU_DYING:
2134                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2135                        cpu);
2136                 hardware_disable(NULL);
2137                 break;
2138         case CPU_STARTING:
2139                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2140                        cpu);
2141                 spin_lock(&kvm_lock);
2142                 hardware_enable(NULL);
2143                 spin_unlock(&kvm_lock);
2144                 break;
2145         }
2146         return NOTIFY_OK;
2147 }
2148
2149
2150 asmlinkage void kvm_handle_fault_on_reboot(void)
2151 {
2152         if (kvm_rebooting) {
2153                 /* spin while reset goes on */
2154                 local_irq_enable();
2155                 while (true)
2156                         cpu_relax();
2157         }
2158         /* Fault while not rebooting.  We want the trace. */
2159         BUG();
2160 }
2161 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2162
2163 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2164                       void *v)
2165 {
2166         /*
2167          * Some (well, at least mine) BIOSes hang on reboot if
2168          * in vmx root mode.
2169          *
2170          * And Intel TXT required VMX off for all cpu when system shutdown.
2171          */
2172         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2173         kvm_rebooting = true;
2174         on_each_cpu(hardware_disable, NULL, 1);
2175         return NOTIFY_OK;
2176 }
2177
2178 static struct notifier_block kvm_reboot_notifier = {
2179         .notifier_call = kvm_reboot,
2180         .priority = 0,
2181 };
2182
2183 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2184 {
2185         int i;
2186
2187         for (i = 0; i < bus->dev_count; i++) {
2188                 struct kvm_io_device *pos = bus->devs[i];
2189
2190                 kvm_iodevice_destructor(pos);
2191         }
2192         kfree(bus);
2193 }
2194
2195 /* kvm_io_bus_write - called under kvm->slots_lock */
2196 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2197                      int len, const void *val)
2198 {
2199         int i;
2200         struct kvm_io_bus *bus;
2201
2202         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2203         for (i = 0; i < bus->dev_count; i++)
2204                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2205                         return 0;
2206         return -EOPNOTSUPP;
2207 }
2208
2209 /* kvm_io_bus_read - called under kvm->slots_lock */
2210 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2211                     int len, void *val)
2212 {
2213         int i;
2214         struct kvm_io_bus *bus;
2215
2216         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2217         for (i = 0; i < bus->dev_count; i++)
2218                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2219                         return 0;
2220         return -EOPNOTSUPP;
2221 }
2222
2223 /* Caller must hold slots_lock. */
2224 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2225                             struct kvm_io_device *dev)
2226 {
2227         struct kvm_io_bus *new_bus, *bus;
2228
2229         bus = kvm->buses[bus_idx];
2230         if (bus->dev_count > NR_IOBUS_DEVS-1)
2231                 return -ENOSPC;
2232
2233         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2234         if (!new_bus)
2235                 return -ENOMEM;
2236         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2237         new_bus->devs[new_bus->dev_count++] = dev;
2238         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2239         synchronize_srcu_expedited(&kvm->srcu);
2240         kfree(bus);
2241
2242         return 0;
2243 }
2244
2245 /* Caller must hold slots_lock. */
2246 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2247                               struct kvm_io_device *dev)
2248 {
2249         int i, r;
2250         struct kvm_io_bus *new_bus, *bus;
2251
2252         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2253         if (!new_bus)
2254                 return -ENOMEM;
2255
2256         bus = kvm->buses[bus_idx];
2257         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2258
2259         r = -ENOENT;
2260         for (i = 0; i < new_bus->dev_count; i++)
2261                 if (new_bus->devs[i] == dev) {
2262                         r = 0;
2263                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2264                         break;
2265                 }
2266
2267         if (r) {
2268                 kfree(new_bus);
2269                 return r;
2270         }
2271
2272         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2273         synchronize_srcu_expedited(&kvm->srcu);
2274         kfree(bus);
2275         return r;
2276 }
2277
2278 static struct notifier_block kvm_cpu_notifier = {
2279         .notifier_call = kvm_cpu_hotplug,
2280 };
2281
2282 static int vm_stat_get(void *_offset, u64 *val)
2283 {
2284         unsigned offset = (long)_offset;
2285         struct kvm *kvm;
2286
2287         *val = 0;
2288         spin_lock(&kvm_lock);
2289         list_for_each_entry(kvm, &vm_list, vm_list)
2290                 *val += *(u32 *)((void *)kvm + offset);
2291         spin_unlock(&kvm_lock);
2292         return 0;
2293 }
2294
2295 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2296
2297 static int vcpu_stat_get(void *_offset, u64 *val)
2298 {
2299         unsigned offset = (long)_offset;
2300         struct kvm *kvm;
2301         struct kvm_vcpu *vcpu;
2302         int i;
2303
2304         *val = 0;
2305         spin_lock(&kvm_lock);
2306         list_for_each_entry(kvm, &vm_list, vm_list)
2307                 kvm_for_each_vcpu(i, vcpu, kvm)
2308                         *val += *(u32 *)((void *)vcpu + offset);
2309
2310         spin_unlock(&kvm_lock);
2311         return 0;
2312 }
2313
2314 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2315
2316 static const struct file_operations *stat_fops[] = {
2317         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2318         [KVM_STAT_VM]   = &vm_stat_fops,
2319 };
2320
2321 static void kvm_init_debug(void)
2322 {
2323         struct kvm_stats_debugfs_item *p;
2324
2325         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2326         for (p = debugfs_entries; p->name; ++p)
2327                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2328                                                 (void *)(long)p->offset,
2329                                                 stat_fops[p->kind]);
2330 }
2331
2332 static void kvm_exit_debug(void)
2333 {
2334         struct kvm_stats_debugfs_item *p;
2335
2336         for (p = debugfs_entries; p->name; ++p)
2337                 debugfs_remove(p->dentry);
2338         debugfs_remove(kvm_debugfs_dir);
2339 }
2340
2341 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2342 {
2343         if (kvm_usage_count)
2344                 hardware_disable(NULL);
2345         return 0;
2346 }
2347
2348 static int kvm_resume(struct sys_device *dev)
2349 {
2350         if (kvm_usage_count) {
2351                 WARN_ON(spin_is_locked(&kvm_lock));
2352                 hardware_enable(NULL);
2353         }
2354         return 0;
2355 }
2356
2357 static struct sysdev_class kvm_sysdev_class = {
2358         .name = "kvm",
2359         .suspend = kvm_suspend,
2360         .resume = kvm_resume,
2361 };
2362
2363 static struct sys_device kvm_sysdev = {
2364         .id = 0,
2365         .cls = &kvm_sysdev_class,
2366 };
2367
2368 struct page *bad_page;
2369 pfn_t bad_pfn;
2370
2371 static inline
2372 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2373 {
2374         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2375 }
2376
2377 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2378 {
2379         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2380
2381         kvm_arch_vcpu_load(vcpu, cpu);
2382 }
2383
2384 static void kvm_sched_out(struct preempt_notifier *pn,
2385                           struct task_struct *next)
2386 {
2387         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2388
2389         kvm_arch_vcpu_put(vcpu);
2390 }
2391
2392 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2393                   struct module *module)
2394 {
2395         int r;
2396         int cpu;
2397
2398         r = kvm_arch_init(opaque);
2399         if (r)
2400                 goto out_fail;
2401
2402         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2403
2404         if (bad_page == NULL) {
2405                 r = -ENOMEM;
2406                 goto out;
2407         }
2408
2409         bad_pfn = page_to_pfn(bad_page);
2410
2411         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2412
2413         if (hwpoison_page == NULL) {
2414                 r = -ENOMEM;
2415                 goto out_free_0;
2416         }
2417
2418         hwpoison_pfn = page_to_pfn(hwpoison_page);
2419
2420         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2421
2422         if (fault_page == NULL) {
2423                 r = -ENOMEM;
2424                 goto out_free_0;
2425         }
2426
2427         fault_pfn = page_to_pfn(fault_page);
2428
2429         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2430                 r = -ENOMEM;
2431                 goto out_free_0;
2432         }
2433
2434         r = kvm_arch_hardware_setup();
2435         if (r < 0)
2436                 goto out_free_0a;
2437
2438         for_each_online_cpu(cpu) {
2439                 smp_call_function_single(cpu,
2440                                 kvm_arch_check_processor_compat,
2441                                 &r, 1);
2442                 if (r < 0)
2443                         goto out_free_1;
2444         }
2445
2446         r = register_cpu_notifier(&kvm_cpu_notifier);
2447         if (r)
2448                 goto out_free_2;
2449         register_reboot_notifier(&kvm_reboot_notifier);
2450
2451         r = sysdev_class_register(&kvm_sysdev_class);
2452         if (r)
2453                 goto out_free_3;
2454
2455         r = sysdev_register(&kvm_sysdev);
2456         if (r)
2457                 goto out_free_4;
2458
2459         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2460         if (!vcpu_align)
2461                 vcpu_align = __alignof__(struct kvm_vcpu);
2462         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2463                                            0, NULL);
2464         if (!kvm_vcpu_cache) {
2465                 r = -ENOMEM;
2466                 goto out_free_5;
2467         }
2468
2469         r = kvm_async_pf_init();
2470         if (r)
2471                 goto out_free;
2472
2473         kvm_chardev_ops.owner = module;
2474         kvm_vm_fops.owner = module;
2475         kvm_vcpu_fops.owner = module;
2476
2477         r = misc_register(&kvm_dev);
2478         if (r) {
2479                 printk(KERN_ERR "kvm: misc device register failed\n");
2480                 goto out_unreg;
2481         }
2482
2483         kvm_preempt_ops.sched_in = kvm_sched_in;
2484         kvm_preempt_ops.sched_out = kvm_sched_out;
2485
2486         kvm_init_debug();
2487
2488         return 0;
2489
2490 out_unreg:
2491         kvm_async_pf_deinit();
2492 out_free:
2493         kmem_cache_destroy(kvm_vcpu_cache);
2494 out_free_5:
2495         sysdev_unregister(&kvm_sysdev);
2496 out_free_4:
2497         sysdev_class_unregister(&kvm_sysdev_class);
2498 out_free_3:
2499         unregister_reboot_notifier(&kvm_reboot_notifier);
2500         unregister_cpu_notifier(&kvm_cpu_notifier);
2501 out_free_2:
2502 out_free_1:
2503         kvm_arch_hardware_unsetup();
2504 out_free_0a:
2505         free_cpumask_var(cpus_hardware_enabled);
2506 out_free_0:
2507         if (fault_page)
2508                 __free_page(fault_page);
2509         if (hwpoison_page)
2510                 __free_page(hwpoison_page);
2511         __free_page(bad_page);
2512 out:
2513         kvm_arch_exit();
2514 out_fail:
2515         return r;
2516 }
2517 EXPORT_SYMBOL_GPL(kvm_init);
2518
2519 void kvm_exit(void)
2520 {
2521         kvm_exit_debug();
2522         misc_deregister(&kvm_dev);
2523         kmem_cache_destroy(kvm_vcpu_cache);
2524         kvm_async_pf_deinit();
2525         sysdev_unregister(&kvm_sysdev);
2526         sysdev_class_unregister(&kvm_sysdev_class);
2527         unregister_reboot_notifier(&kvm_reboot_notifier);
2528         unregister_cpu_notifier(&kvm_cpu_notifier);
2529         on_each_cpu(hardware_disable, NULL, 1);
2530         kvm_arch_hardware_unsetup();
2531         kvm_arch_exit();
2532         free_cpumask_var(cpus_hardware_enabled);
2533         __free_page(hwpoison_page);
2534         __free_page(bad_page);
2535 }
2536 EXPORT_SYMBOL_GPL(kvm_exit);