KVM: use kmalloc() for small dirty bitmaps
[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;
405         if (init_srcu_struct(&kvm->srcu))
406                 goto out_err;
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                         cleanup_srcu_struct(&kvm->srcu);
412                         goto out_err;
413                 }
414         }
415
416         r = kvm_init_mmu_notifier(kvm);
417         if (r) {
418                 cleanup_srcu_struct(&kvm->srcu);
419                 goto out_err;
420         }
421
422         kvm->mm = current->mm;
423         atomic_inc(&kvm->mm->mm_count);
424         spin_lock_init(&kvm->mmu_lock);
425         raw_spin_lock_init(&kvm->requests_lock);
426         kvm_eventfd_init(kvm);
427         mutex_init(&kvm->lock);
428         mutex_init(&kvm->irq_lock);
429         mutex_init(&kvm->slots_lock);
430         atomic_set(&kvm->users_count, 1);
431         spin_lock(&kvm_lock);
432         list_add(&kvm->vm_list, &vm_list);
433         spin_unlock(&kvm_lock);
434 out:
435         return kvm;
436
437 out_err:
438         hardware_disable_all();
439 out_err_nodisable:
440         for (i = 0; i < KVM_NR_BUSES; i++)
441                 kfree(kvm->buses[i]);
442         kfree(kvm->memslots);
443         kfree(kvm);
444         return ERR_PTR(r);
445 }
446
447 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
448 {
449         if (!memslot->dirty_bitmap)
450                 return;
451
452         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
453                 vfree(memslot->dirty_bitmap_head);
454         else
455                 kfree(memslot->dirty_bitmap_head);
456
457         memslot->dirty_bitmap = NULL;
458         memslot->dirty_bitmap_head = NULL;
459 }
460
461 /*
462  * Free any memory in @free but not in @dont.
463  */
464 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
465                                   struct kvm_memory_slot *dont)
466 {
467         int i;
468
469         if (!dont || free->rmap != dont->rmap)
470                 vfree(free->rmap);
471
472         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
473                 kvm_destroy_dirty_bitmap(free);
474
475
476         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
477                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
478                         vfree(free->lpage_info[i]);
479                         free->lpage_info[i] = NULL;
480                 }
481         }
482
483         free->npages = 0;
484         free->rmap = NULL;
485 }
486
487 void kvm_free_physmem(struct kvm *kvm)
488 {
489         int i;
490         struct kvm_memslots *slots = kvm->memslots;
491
492         for (i = 0; i < slots->nmemslots; ++i)
493                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
494
495         kfree(kvm->memslots);
496 }
497
498 static void kvm_destroy_vm(struct kvm *kvm)
499 {
500         int i;
501         struct mm_struct *mm = kvm->mm;
502
503         kvm_arch_sync_events(kvm);
504         spin_lock(&kvm_lock);
505         list_del(&kvm->vm_list);
506         spin_unlock(&kvm_lock);
507         kvm_free_irq_routing(kvm);
508         for (i = 0; i < KVM_NR_BUSES; i++)
509                 kvm_io_bus_destroy(kvm->buses[i]);
510         kvm_coalesced_mmio_free(kvm);
511 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
512         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
513 #else
514         kvm_arch_flush_shadow(kvm);
515 #endif
516         kvm_arch_destroy_vm(kvm);
517         hardware_disable_all();
518         mmdrop(mm);
519 }
520
521 void kvm_get_kvm(struct kvm *kvm)
522 {
523         atomic_inc(&kvm->users_count);
524 }
525 EXPORT_SYMBOL_GPL(kvm_get_kvm);
526
527 void kvm_put_kvm(struct kvm *kvm)
528 {
529         if (atomic_dec_and_test(&kvm->users_count))
530                 kvm_destroy_vm(kvm);
531 }
532 EXPORT_SYMBOL_GPL(kvm_put_kvm);
533
534
535 static int kvm_vm_release(struct inode *inode, struct file *filp)
536 {
537         struct kvm *kvm = filp->private_data;
538
539         kvm_irqfd_release(kvm);
540
541         kvm_put_kvm(kvm);
542         return 0;
543 }
544
545 /*
546  * Allocation size is twice as large as the actual dirty bitmap size.
547  * This makes it possible to do double buffering: see x86's
548  * kvm_vm_ioctl_get_dirty_log().
549  */
550 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
551 {
552         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
553
554         if (dirty_bytes > PAGE_SIZE)
555                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
556         else
557                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
558
559         if (!memslot->dirty_bitmap)
560                 return -ENOMEM;
561
562         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
563         return 0;
564 }
565
566 /*
567  * Allocate some memory and give it an address in the guest physical address
568  * space.
569  *
570  * Discontiguous memory is allowed, mostly for framebuffers.
571  *
572  * Must be called holding mmap_sem for write.
573  */
574 int __kvm_set_memory_region(struct kvm *kvm,
575                             struct kvm_userspace_memory_region *mem,
576                             int user_alloc)
577 {
578         int r, flush_shadow = 0;
579         gfn_t base_gfn;
580         unsigned long npages;
581         unsigned long i;
582         struct kvm_memory_slot *memslot;
583         struct kvm_memory_slot old, new;
584         struct kvm_memslots *slots, *old_memslots;
585
586         r = -EINVAL;
587         /* General sanity checks */
588         if (mem->memory_size & (PAGE_SIZE - 1))
589                 goto out;
590         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
591                 goto out;
592         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
593                 goto out;
594         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
595                 goto out;
596         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
597                 goto out;
598
599         memslot = &kvm->memslots->memslots[mem->slot];
600         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
601         npages = mem->memory_size >> PAGE_SHIFT;
602
603         r = -EINVAL;
604         if (npages > KVM_MEM_MAX_NR_PAGES)
605                 goto out;
606
607         if (!npages)
608                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
609
610         new = old = *memslot;
611
612         new.id = mem->slot;
613         new.base_gfn = base_gfn;
614         new.npages = npages;
615         new.flags = mem->flags;
616
617         /* Disallow changing a memory slot's size. */
618         r = -EINVAL;
619         if (npages && old.npages && npages != old.npages)
620                 goto out_free;
621
622         /* Check for overlaps */
623         r = -EEXIST;
624         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
625                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
626
627                 if (s == memslot || !s->npages)
628                         continue;
629                 if (!((base_gfn + npages <= s->base_gfn) ||
630                       (base_gfn >= s->base_gfn + s->npages)))
631                         goto out_free;
632         }
633
634         /* Free page dirty bitmap if unneeded */
635         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
636                 new.dirty_bitmap = NULL;
637
638         r = -ENOMEM;
639
640         /* Allocate if a slot is being created */
641 #ifndef CONFIG_S390
642         if (npages && !new.rmap) {
643                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
644
645                 if (!new.rmap)
646                         goto out_free;
647
648                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
649
650                 new.user_alloc = user_alloc;
651                 new.userspace_addr = mem->userspace_addr;
652         }
653         if (!npages)
654                 goto skip_lpage;
655
656         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
657                 unsigned long ugfn;
658                 unsigned long j;
659                 int lpages;
660                 int level = i + 2;
661
662                 /* Avoid unused variable warning if no large pages */
663                 (void)level;
664
665                 if (new.lpage_info[i])
666                         continue;
667
668                 lpages = 1 + ((base_gfn + npages - 1)
669                              >> KVM_HPAGE_GFN_SHIFT(level));
670                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
671
672                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
673
674                 if (!new.lpage_info[i])
675                         goto out_free;
676
677                 memset(new.lpage_info[i], 0,
678                        lpages * sizeof(*new.lpage_info[i]));
679
680                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
681                         new.lpage_info[i][0].write_count = 1;
682                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
683                         new.lpage_info[i][lpages - 1].write_count = 1;
684                 ugfn = new.userspace_addr >> PAGE_SHIFT;
685                 /*
686                  * If the gfn and userspace address are not aligned wrt each
687                  * other, or if explicitly asked to, disable large page
688                  * support for this slot
689                  */
690                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
691                     !largepages_enabled)
692                         for (j = 0; j < lpages; ++j)
693                                 new.lpage_info[i][j].write_count = 1;
694         }
695
696 skip_lpage:
697
698         /* Allocate page dirty bitmap if needed */
699         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
700                 if (kvm_create_dirty_bitmap(&new) < 0)
701                         goto out_free;
702                 /* destroy any largepage mappings for dirty tracking */
703                 if (old.npages)
704                         flush_shadow = 1;
705         }
706 #else  /* not defined CONFIG_S390 */
707         new.user_alloc = user_alloc;
708         if (user_alloc)
709                 new.userspace_addr = mem->userspace_addr;
710 #endif /* not defined CONFIG_S390 */
711
712         if (!npages) {
713                 r = -ENOMEM;
714                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
715                 if (!slots)
716                         goto out_free;
717                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
718                 if (mem->slot >= slots->nmemslots)
719                         slots->nmemslots = mem->slot + 1;
720                 slots->generation++;
721                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
722
723                 old_memslots = kvm->memslots;
724                 rcu_assign_pointer(kvm->memslots, slots);
725                 synchronize_srcu_expedited(&kvm->srcu);
726                 /* From this point no new shadow pages pointing to a deleted
727                  * memslot will be created.
728                  *
729                  * validation of sp->gfn happens in:
730                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
731                  *      - kvm_is_visible_gfn (mmu_check_roots)
732                  */
733                 kvm_arch_flush_shadow(kvm);
734                 kfree(old_memslots);
735         }
736
737         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
738         if (r)
739                 goto out_free;
740
741         /* map the pages in iommu page table */
742         if (npages) {
743                 r = kvm_iommu_map_pages(kvm, &new);
744                 if (r)
745                         goto out_free;
746         }
747
748         r = -ENOMEM;
749         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
750         if (!slots)
751                 goto out_free;
752         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
753         if (mem->slot >= slots->nmemslots)
754                 slots->nmemslots = mem->slot + 1;
755         slots->generation++;
756
757         /* actual memory is freed via old in kvm_free_physmem_slot below */
758         if (!npages) {
759                 new.rmap = NULL;
760                 new.dirty_bitmap = NULL;
761                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
762                         new.lpage_info[i] = NULL;
763         }
764
765         slots->memslots[mem->slot] = new;
766         old_memslots = kvm->memslots;
767         rcu_assign_pointer(kvm->memslots, slots);
768         synchronize_srcu_expedited(&kvm->srcu);
769
770         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
771
772         kvm_free_physmem_slot(&old, &new);
773         kfree(old_memslots);
774
775         if (flush_shadow)
776                 kvm_arch_flush_shadow(kvm);
777
778         return 0;
779
780 out_free:
781         kvm_free_physmem_slot(&new, &old);
782 out:
783         return r;
784
785 }
786 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
787
788 int kvm_set_memory_region(struct kvm *kvm,
789                           struct kvm_userspace_memory_region *mem,
790                           int user_alloc)
791 {
792         int r;
793
794         mutex_lock(&kvm->slots_lock);
795         r = __kvm_set_memory_region(kvm, mem, user_alloc);
796         mutex_unlock(&kvm->slots_lock);
797         return r;
798 }
799 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
800
801 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
802                                    struct
803                                    kvm_userspace_memory_region *mem,
804                                    int user_alloc)
805 {
806         if (mem->slot >= KVM_MEMORY_SLOTS)
807                 return -EINVAL;
808         return kvm_set_memory_region(kvm, mem, user_alloc);
809 }
810
811 int kvm_get_dirty_log(struct kvm *kvm,
812                         struct kvm_dirty_log *log, int *is_dirty)
813 {
814         struct kvm_memory_slot *memslot;
815         int r, i;
816         unsigned long n;
817         unsigned long any = 0;
818
819         r = -EINVAL;
820         if (log->slot >= KVM_MEMORY_SLOTS)
821                 goto out;
822
823         memslot = &kvm->memslots->memslots[log->slot];
824         r = -ENOENT;
825         if (!memslot->dirty_bitmap)
826                 goto out;
827
828         n = kvm_dirty_bitmap_bytes(memslot);
829
830         for (i = 0; !any && i < n/sizeof(long); ++i)
831                 any = memslot->dirty_bitmap[i];
832
833         r = -EFAULT;
834         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
835                 goto out;
836
837         if (any)
838                 *is_dirty = 1;
839
840         r = 0;
841 out:
842         return r;
843 }
844
845 void kvm_disable_largepages(void)
846 {
847         largepages_enabled = false;
848 }
849 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
850
851 int is_error_page(struct page *page)
852 {
853         return page == bad_page || page == hwpoison_page || page == fault_page;
854 }
855 EXPORT_SYMBOL_GPL(is_error_page);
856
857 int is_error_pfn(pfn_t pfn)
858 {
859         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
860 }
861 EXPORT_SYMBOL_GPL(is_error_pfn);
862
863 int is_hwpoison_pfn(pfn_t pfn)
864 {
865         return pfn == hwpoison_pfn;
866 }
867 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
868
869 int is_fault_pfn(pfn_t pfn)
870 {
871         return pfn == fault_pfn;
872 }
873 EXPORT_SYMBOL_GPL(is_fault_pfn);
874
875 static inline unsigned long bad_hva(void)
876 {
877         return PAGE_OFFSET;
878 }
879
880 int kvm_is_error_hva(unsigned long addr)
881 {
882         return addr == bad_hva();
883 }
884 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
885
886 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
887                                                 gfn_t gfn)
888 {
889         int i;
890
891         for (i = 0; i < slots->nmemslots; ++i) {
892                 struct kvm_memory_slot *memslot = &slots->memslots[i];
893
894                 if (gfn >= memslot->base_gfn
895                     && gfn < memslot->base_gfn + memslot->npages)
896                         return memslot;
897         }
898         return NULL;
899 }
900
901 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
902 {
903         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
904 }
905 EXPORT_SYMBOL_GPL(gfn_to_memslot);
906
907 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
908 {
909         int i;
910         struct kvm_memslots *slots = kvm_memslots(kvm);
911
912         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
913                 struct kvm_memory_slot *memslot = &slots->memslots[i];
914
915                 if (memslot->flags & KVM_MEMSLOT_INVALID)
916                         continue;
917
918                 if (gfn >= memslot->base_gfn
919                     && gfn < memslot->base_gfn + memslot->npages)
920                         return 1;
921         }
922         return 0;
923 }
924 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
925
926 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
927 {
928         struct vm_area_struct *vma;
929         unsigned long addr, size;
930
931         size = PAGE_SIZE;
932
933         addr = gfn_to_hva(kvm, gfn);
934         if (kvm_is_error_hva(addr))
935                 return PAGE_SIZE;
936
937         down_read(&current->mm->mmap_sem);
938         vma = find_vma(current->mm, addr);
939         if (!vma)
940                 goto out;
941
942         size = vma_kernel_pagesize(vma);
943
944 out:
945         up_read(&current->mm->mmap_sem);
946
947         return size;
948 }
949
950 int memslot_id(struct kvm *kvm, gfn_t gfn)
951 {
952         int i;
953         struct kvm_memslots *slots = kvm_memslots(kvm);
954         struct kvm_memory_slot *memslot = NULL;
955
956         for (i = 0; i < slots->nmemslots; ++i) {
957                 memslot = &slots->memslots[i];
958
959                 if (gfn >= memslot->base_gfn
960                     && gfn < memslot->base_gfn + memslot->npages)
961                         break;
962         }
963
964         return memslot - slots->memslots;
965 }
966
967 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
968                                      gfn_t *nr_pages)
969 {
970         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
971                 return bad_hva();
972
973         if (nr_pages)
974                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
975
976         return gfn_to_hva_memslot(slot, gfn);
977 }
978
979 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
980 {
981         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
982 }
983 EXPORT_SYMBOL_GPL(gfn_to_hva);
984
985 static pfn_t get_fault_pfn(void)
986 {
987         get_page(fault_page);
988         return fault_pfn;
989 }
990
991 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
992                         bool *async, bool write_fault, bool *writable)
993 {
994         struct page *page[1];
995         int npages = 0;
996         pfn_t pfn;
997
998         /* we can do it either atomically or asynchronously, not both */
999         BUG_ON(atomic && async);
1000
1001         BUG_ON(!write_fault && !writable);
1002
1003         if (writable)
1004                 *writable = true;
1005
1006         if (atomic || async)
1007                 npages = __get_user_pages_fast(addr, 1, 1, page);
1008
1009         if (unlikely(npages != 1) && !atomic) {
1010                 might_sleep();
1011
1012                 if (writable)
1013                         *writable = write_fault;
1014
1015                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1016
1017                 /* map read fault as writable if possible */
1018                 if (unlikely(!write_fault) && npages == 1) {
1019                         struct page *wpage[1];
1020
1021                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1022                         if (npages == 1) {
1023                                 *writable = true;
1024                                 put_page(page[0]);
1025                                 page[0] = wpage[0];
1026                         }
1027                         npages = 1;
1028                 }
1029         }
1030
1031         if (unlikely(npages != 1)) {
1032                 struct vm_area_struct *vma;
1033
1034                 if (atomic)
1035                         return get_fault_pfn();
1036
1037                 down_read(&current->mm->mmap_sem);
1038                 if (is_hwpoison_address(addr)) {
1039                         up_read(&current->mm->mmap_sem);
1040                         get_page(hwpoison_page);
1041                         return page_to_pfn(hwpoison_page);
1042                 }
1043
1044                 vma = find_vma_intersection(current->mm, addr, addr+1);
1045
1046                 if (vma == NULL)
1047                         pfn = get_fault_pfn();
1048                 else if ((vma->vm_flags & VM_PFNMAP)) {
1049                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1050                                 vma->vm_pgoff;
1051                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1052                 } else {
1053                         if (async && (vma->vm_flags & VM_WRITE))
1054                                 *async = true;
1055                         pfn = get_fault_pfn();
1056                 }
1057                 up_read(&current->mm->mmap_sem);
1058         } else
1059                 pfn = page_to_pfn(page[0]);
1060
1061         return pfn;
1062 }
1063
1064 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1065 {
1066         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1067 }
1068 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1069
1070 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1071                           bool write_fault, bool *writable)
1072 {
1073         unsigned long addr;
1074
1075         if (async)
1076                 *async = false;
1077
1078         addr = gfn_to_hva(kvm, gfn);
1079         if (kvm_is_error_hva(addr)) {
1080                 get_page(bad_page);
1081                 return page_to_pfn(bad_page);
1082         }
1083
1084         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1085 }
1086
1087 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1088 {
1089         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1090 }
1091 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1092
1093 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1094                        bool write_fault, bool *writable)
1095 {
1096         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1097 }
1098 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1099
1100 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1101 {
1102         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1103 }
1104 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1105
1106 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1107                       bool *writable)
1108 {
1109         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1110 }
1111 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1112
1113 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1114                          struct kvm_memory_slot *slot, gfn_t gfn)
1115 {
1116         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1117         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1118 }
1119
1120 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1121                                                                   int nr_pages)
1122 {
1123         unsigned long addr;
1124         gfn_t entry;
1125
1126         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1127         if (kvm_is_error_hva(addr))
1128                 return -1;
1129
1130         if (entry < nr_pages)
1131                 return 0;
1132
1133         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1134 }
1135 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1136
1137 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1138 {
1139         pfn_t pfn;
1140
1141         pfn = gfn_to_pfn(kvm, gfn);
1142         if (!kvm_is_mmio_pfn(pfn))
1143                 return pfn_to_page(pfn);
1144
1145         WARN_ON(kvm_is_mmio_pfn(pfn));
1146
1147         get_page(bad_page);
1148         return bad_page;
1149 }
1150
1151 EXPORT_SYMBOL_GPL(gfn_to_page);
1152
1153 void kvm_release_page_clean(struct page *page)
1154 {
1155         kvm_release_pfn_clean(page_to_pfn(page));
1156 }
1157 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1158
1159 void kvm_release_pfn_clean(pfn_t pfn)
1160 {
1161         if (!kvm_is_mmio_pfn(pfn))
1162                 put_page(pfn_to_page(pfn));
1163 }
1164 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1165
1166 void kvm_release_page_dirty(struct page *page)
1167 {
1168         kvm_release_pfn_dirty(page_to_pfn(page));
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1171
1172 void kvm_release_pfn_dirty(pfn_t pfn)
1173 {
1174         kvm_set_pfn_dirty(pfn);
1175         kvm_release_pfn_clean(pfn);
1176 }
1177 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1178
1179 void kvm_set_page_dirty(struct page *page)
1180 {
1181         kvm_set_pfn_dirty(page_to_pfn(page));
1182 }
1183 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1184
1185 void kvm_set_pfn_dirty(pfn_t pfn)
1186 {
1187         if (!kvm_is_mmio_pfn(pfn)) {
1188                 struct page *page = pfn_to_page(pfn);
1189                 if (!PageReserved(page))
1190                         SetPageDirty(page);
1191         }
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1194
1195 void kvm_set_pfn_accessed(pfn_t pfn)
1196 {
1197         if (!kvm_is_mmio_pfn(pfn))
1198                 mark_page_accessed(pfn_to_page(pfn));
1199 }
1200 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1201
1202 void kvm_get_pfn(pfn_t pfn)
1203 {
1204         if (!kvm_is_mmio_pfn(pfn))
1205                 get_page(pfn_to_page(pfn));
1206 }
1207 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1208
1209 static int next_segment(unsigned long len, int offset)
1210 {
1211         if (len > PAGE_SIZE - offset)
1212                 return PAGE_SIZE - offset;
1213         else
1214                 return len;
1215 }
1216
1217 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1218                         int len)
1219 {
1220         int r;
1221         unsigned long addr;
1222
1223         addr = gfn_to_hva(kvm, gfn);
1224         if (kvm_is_error_hva(addr))
1225                 return -EFAULT;
1226         r = copy_from_user(data, (void __user *)addr + offset, len);
1227         if (r)
1228                 return -EFAULT;
1229         return 0;
1230 }
1231 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1232
1233 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1234 {
1235         gfn_t gfn = gpa >> PAGE_SHIFT;
1236         int seg;
1237         int offset = offset_in_page(gpa);
1238         int ret;
1239
1240         while ((seg = next_segment(len, offset)) != 0) {
1241                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1242                 if (ret < 0)
1243                         return ret;
1244                 offset = 0;
1245                 len -= seg;
1246                 data += seg;
1247                 ++gfn;
1248         }
1249         return 0;
1250 }
1251 EXPORT_SYMBOL_GPL(kvm_read_guest);
1252
1253 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1254                           unsigned long len)
1255 {
1256         int r;
1257         unsigned long addr;
1258         gfn_t gfn = gpa >> PAGE_SHIFT;
1259         int offset = offset_in_page(gpa);
1260
1261         addr = gfn_to_hva(kvm, gfn);
1262         if (kvm_is_error_hva(addr))
1263                 return -EFAULT;
1264         pagefault_disable();
1265         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1266         pagefault_enable();
1267         if (r)
1268                 return -EFAULT;
1269         return 0;
1270 }
1271 EXPORT_SYMBOL(kvm_read_guest_atomic);
1272
1273 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1274                          int offset, int len)
1275 {
1276         int r;
1277         unsigned long addr;
1278
1279         addr = gfn_to_hva(kvm, gfn);
1280         if (kvm_is_error_hva(addr))
1281                 return -EFAULT;
1282         r = copy_to_user((void __user *)addr + offset, data, len);
1283         if (r)
1284                 return -EFAULT;
1285         mark_page_dirty(kvm, gfn);
1286         return 0;
1287 }
1288 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1289
1290 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1291                     unsigned long len)
1292 {
1293         gfn_t gfn = gpa >> PAGE_SHIFT;
1294         int seg;
1295         int offset = offset_in_page(gpa);
1296         int ret;
1297
1298         while ((seg = next_segment(len, offset)) != 0) {
1299                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1300                 if (ret < 0)
1301                         return ret;
1302                 offset = 0;
1303                 len -= seg;
1304                 data += seg;
1305                 ++gfn;
1306         }
1307         return 0;
1308 }
1309
1310 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1311                               gpa_t gpa)
1312 {
1313         struct kvm_memslots *slots = kvm_memslots(kvm);
1314         int offset = offset_in_page(gpa);
1315         gfn_t gfn = gpa >> PAGE_SHIFT;
1316
1317         ghc->gpa = gpa;
1318         ghc->generation = slots->generation;
1319         ghc->memslot = __gfn_to_memslot(slots, gfn);
1320         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1321         if (!kvm_is_error_hva(ghc->hva))
1322                 ghc->hva += offset;
1323         else
1324                 return -EFAULT;
1325
1326         return 0;
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1329
1330 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1331                            void *data, unsigned long len)
1332 {
1333         struct kvm_memslots *slots = kvm_memslots(kvm);
1334         int r;
1335
1336         if (slots->generation != ghc->generation)
1337                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1338
1339         if (kvm_is_error_hva(ghc->hva))
1340                 return -EFAULT;
1341
1342         r = copy_to_user((void __user *)ghc->hva, data, len);
1343         if (r)
1344                 return -EFAULT;
1345         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1346
1347         return 0;
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1350
1351 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1352 {
1353         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1354 }
1355 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1356
1357 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1358 {
1359         gfn_t gfn = gpa >> PAGE_SHIFT;
1360         int seg;
1361         int offset = offset_in_page(gpa);
1362         int ret;
1363
1364         while ((seg = next_segment(len, offset)) != 0) {
1365                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1366                 if (ret < 0)
1367                         return ret;
1368                 offset = 0;
1369                 len -= seg;
1370                 ++gfn;
1371         }
1372         return 0;
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1375
1376 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1377                              gfn_t gfn)
1378 {
1379         if (memslot && memslot->dirty_bitmap) {
1380                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1381
1382                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1383         }
1384 }
1385
1386 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1387 {
1388         struct kvm_memory_slot *memslot;
1389
1390         memslot = gfn_to_memslot(kvm, gfn);
1391         mark_page_dirty_in_slot(kvm, memslot, gfn);
1392 }
1393
1394 /*
1395  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1396  */
1397 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1398 {
1399         DEFINE_WAIT(wait);
1400
1401         for (;;) {
1402                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1403
1404                 if (kvm_arch_vcpu_runnable(vcpu)) {
1405                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1406                         break;
1407                 }
1408                 if (kvm_cpu_has_pending_timer(vcpu))
1409                         break;
1410                 if (signal_pending(current))
1411                         break;
1412
1413                 schedule();
1414         }
1415
1416         finish_wait(&vcpu->wq, &wait);
1417 }
1418
1419 void kvm_resched(struct kvm_vcpu *vcpu)
1420 {
1421         if (!need_resched())
1422                 return;
1423         cond_resched();
1424 }
1425 EXPORT_SYMBOL_GPL(kvm_resched);
1426
1427 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1428 {
1429         ktime_t expires;
1430         DEFINE_WAIT(wait);
1431
1432         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1433
1434         /* Sleep for 100 us, and hope lock-holder got scheduled */
1435         expires = ktime_add_ns(ktime_get(), 100000UL);
1436         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1437
1438         finish_wait(&vcpu->wq, &wait);
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1441
1442 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1443 {
1444         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1445         struct page *page;
1446
1447         if (vmf->pgoff == 0)
1448                 page = virt_to_page(vcpu->run);
1449 #ifdef CONFIG_X86
1450         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1451                 page = virt_to_page(vcpu->arch.pio_data);
1452 #endif
1453 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1454         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1455                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1456 #endif
1457         else
1458                 return VM_FAULT_SIGBUS;
1459         get_page(page);
1460         vmf->page = page;
1461         return 0;
1462 }
1463
1464 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1465         .fault = kvm_vcpu_fault,
1466 };
1467
1468 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1469 {
1470         vma->vm_ops = &kvm_vcpu_vm_ops;
1471         return 0;
1472 }
1473
1474 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1475 {
1476         struct kvm_vcpu *vcpu = filp->private_data;
1477
1478         kvm_put_kvm(vcpu->kvm);
1479         return 0;
1480 }
1481
1482 static struct file_operations kvm_vcpu_fops = {
1483         .release        = kvm_vcpu_release,
1484         .unlocked_ioctl = kvm_vcpu_ioctl,
1485         .compat_ioctl   = kvm_vcpu_ioctl,
1486         .mmap           = kvm_vcpu_mmap,
1487         .llseek         = noop_llseek,
1488 };
1489
1490 /*
1491  * Allocates an inode for the vcpu.
1492  */
1493 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1494 {
1495         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1496 }
1497
1498 /*
1499  * Creates some virtual cpus.  Good luck creating more than one.
1500  */
1501 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1502 {
1503         int r;
1504         struct kvm_vcpu *vcpu, *v;
1505
1506         vcpu = kvm_arch_vcpu_create(kvm, id);
1507         if (IS_ERR(vcpu))
1508                 return PTR_ERR(vcpu);
1509
1510         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1511
1512         r = kvm_arch_vcpu_setup(vcpu);
1513         if (r)
1514                 return r;
1515
1516         mutex_lock(&kvm->lock);
1517         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1518                 r = -EINVAL;
1519                 goto vcpu_destroy;
1520         }
1521
1522         kvm_for_each_vcpu(r, v, kvm)
1523                 if (v->vcpu_id == id) {
1524                         r = -EEXIST;
1525                         goto vcpu_destroy;
1526                 }
1527
1528         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1529
1530         /* Now it's all set up, let userspace reach it */
1531         kvm_get_kvm(kvm);
1532         r = create_vcpu_fd(vcpu);
1533         if (r < 0) {
1534                 kvm_put_kvm(kvm);
1535                 goto vcpu_destroy;
1536         }
1537
1538         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1539         smp_wmb();
1540         atomic_inc(&kvm->online_vcpus);
1541
1542 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1543         if (kvm->bsp_vcpu_id == id)
1544                 kvm->bsp_vcpu = vcpu;
1545 #endif
1546         mutex_unlock(&kvm->lock);
1547         return r;
1548
1549 vcpu_destroy:
1550         mutex_unlock(&kvm->lock);
1551         kvm_arch_vcpu_destroy(vcpu);
1552         return r;
1553 }
1554
1555 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1556 {
1557         if (sigset) {
1558                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1559                 vcpu->sigset_active = 1;
1560                 vcpu->sigset = *sigset;
1561         } else
1562                 vcpu->sigset_active = 0;
1563         return 0;
1564 }
1565
1566 static long kvm_vcpu_ioctl(struct file *filp,
1567                            unsigned int ioctl, unsigned long arg)
1568 {
1569         struct kvm_vcpu *vcpu = filp->private_data;
1570         void __user *argp = (void __user *)arg;
1571         int r;
1572         struct kvm_fpu *fpu = NULL;
1573         struct kvm_sregs *kvm_sregs = NULL;
1574
1575         if (vcpu->kvm->mm != current->mm)
1576                 return -EIO;
1577
1578 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1579         /*
1580          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1581          * so vcpu_load() would break it.
1582          */
1583         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1584                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1585 #endif
1586
1587
1588         vcpu_load(vcpu);
1589         switch (ioctl) {
1590         case KVM_RUN:
1591                 r = -EINVAL;
1592                 if (arg)
1593                         goto out;
1594                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1595                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1596                 break;
1597         case KVM_GET_REGS: {
1598                 struct kvm_regs *kvm_regs;
1599
1600                 r = -ENOMEM;
1601                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1602                 if (!kvm_regs)
1603                         goto out;
1604                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1605                 if (r)
1606                         goto out_free1;
1607                 r = -EFAULT;
1608                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1609                         goto out_free1;
1610                 r = 0;
1611 out_free1:
1612                 kfree(kvm_regs);
1613                 break;
1614         }
1615         case KVM_SET_REGS: {
1616                 struct kvm_regs *kvm_regs;
1617
1618                 r = -ENOMEM;
1619                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1620                 if (!kvm_regs)
1621                         goto out;
1622                 r = -EFAULT;
1623                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1624                         goto out_free2;
1625                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1626                 if (r)
1627                         goto out_free2;
1628                 r = 0;
1629 out_free2:
1630                 kfree(kvm_regs);
1631                 break;
1632         }
1633         case KVM_GET_SREGS: {
1634                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1635                 r = -ENOMEM;
1636                 if (!kvm_sregs)
1637                         goto out;
1638                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1639                 if (r)
1640                         goto out;
1641                 r = -EFAULT;
1642                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1643                         goto out;
1644                 r = 0;
1645                 break;
1646         }
1647         case KVM_SET_SREGS: {
1648                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1649                 r = -ENOMEM;
1650                 if (!kvm_sregs)
1651                         goto out;
1652                 r = -EFAULT;
1653                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1654                         goto out;
1655                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1656                 if (r)
1657                         goto out;
1658                 r = 0;
1659                 break;
1660         }
1661         case KVM_GET_MP_STATE: {
1662                 struct kvm_mp_state mp_state;
1663
1664                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1665                 if (r)
1666                         goto out;
1667                 r = -EFAULT;
1668                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1669                         goto out;
1670                 r = 0;
1671                 break;
1672         }
1673         case KVM_SET_MP_STATE: {
1674                 struct kvm_mp_state mp_state;
1675
1676                 r = -EFAULT;
1677                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1678                         goto out;
1679                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1680                 if (r)
1681                         goto out;
1682                 r = 0;
1683                 break;
1684         }
1685         case KVM_TRANSLATE: {
1686                 struct kvm_translation tr;
1687
1688                 r = -EFAULT;
1689                 if (copy_from_user(&tr, argp, sizeof tr))
1690                         goto out;
1691                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1692                 if (r)
1693                         goto out;
1694                 r = -EFAULT;
1695                 if (copy_to_user(argp, &tr, sizeof tr))
1696                         goto out;
1697                 r = 0;
1698                 break;
1699         }
1700         case KVM_SET_GUEST_DEBUG: {
1701                 struct kvm_guest_debug dbg;
1702
1703                 r = -EFAULT;
1704                 if (copy_from_user(&dbg, argp, sizeof dbg))
1705                         goto out;
1706                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1707                 if (r)
1708                         goto out;
1709                 r = 0;
1710                 break;
1711         }
1712         case KVM_SET_SIGNAL_MASK: {
1713                 struct kvm_signal_mask __user *sigmask_arg = argp;
1714                 struct kvm_signal_mask kvm_sigmask;
1715                 sigset_t sigset, *p;
1716
1717                 p = NULL;
1718                 if (argp) {
1719                         r = -EFAULT;
1720                         if (copy_from_user(&kvm_sigmask, argp,
1721                                            sizeof kvm_sigmask))
1722                                 goto out;
1723                         r = -EINVAL;
1724                         if (kvm_sigmask.len != sizeof sigset)
1725                                 goto out;
1726                         r = -EFAULT;
1727                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1728                                            sizeof sigset))
1729                                 goto out;
1730                         p = &sigset;
1731                 }
1732                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1733                 break;
1734         }
1735         case KVM_GET_FPU: {
1736                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1737                 r = -ENOMEM;
1738                 if (!fpu)
1739                         goto out;
1740                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1741                 if (r)
1742                         goto out;
1743                 r = -EFAULT;
1744                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1745                         goto out;
1746                 r = 0;
1747                 break;
1748         }
1749         case KVM_SET_FPU: {
1750                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1751                 r = -ENOMEM;
1752                 if (!fpu)
1753                         goto out;
1754                 r = -EFAULT;
1755                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1756                         goto out;
1757                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1758                 if (r)
1759                         goto out;
1760                 r = 0;
1761                 break;
1762         }
1763         default:
1764                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1765         }
1766 out:
1767         vcpu_put(vcpu);
1768         kfree(fpu);
1769         kfree(kvm_sregs);
1770         return r;
1771 }
1772
1773 static long kvm_vm_ioctl(struct file *filp,
1774                            unsigned int ioctl, unsigned long arg)
1775 {
1776         struct kvm *kvm = filp->private_data;
1777         void __user *argp = (void __user *)arg;
1778         int r;
1779
1780         if (kvm->mm != current->mm)
1781                 return -EIO;
1782         switch (ioctl) {
1783         case KVM_CREATE_VCPU:
1784                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1785                 if (r < 0)
1786                         goto out;
1787                 break;
1788         case KVM_SET_USER_MEMORY_REGION: {
1789                 struct kvm_userspace_memory_region kvm_userspace_mem;
1790
1791                 r = -EFAULT;
1792                 if (copy_from_user(&kvm_userspace_mem, argp,
1793                                                 sizeof kvm_userspace_mem))
1794                         goto out;
1795
1796                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1797                 if (r)
1798                         goto out;
1799                 break;
1800         }
1801         case KVM_GET_DIRTY_LOG: {
1802                 struct kvm_dirty_log log;
1803
1804                 r = -EFAULT;
1805                 if (copy_from_user(&log, argp, sizeof log))
1806                         goto out;
1807                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1808                 if (r)
1809                         goto out;
1810                 break;
1811         }
1812 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1813         case KVM_REGISTER_COALESCED_MMIO: {
1814                 struct kvm_coalesced_mmio_zone zone;
1815                 r = -EFAULT;
1816                 if (copy_from_user(&zone, argp, sizeof zone))
1817                         goto out;
1818                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1819                 if (r)
1820                         goto out;
1821                 r = 0;
1822                 break;
1823         }
1824         case KVM_UNREGISTER_COALESCED_MMIO: {
1825                 struct kvm_coalesced_mmio_zone zone;
1826                 r = -EFAULT;
1827                 if (copy_from_user(&zone, argp, sizeof zone))
1828                         goto out;
1829                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1830                 if (r)
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835 #endif
1836         case KVM_IRQFD: {
1837                 struct kvm_irqfd data;
1838
1839                 r = -EFAULT;
1840                 if (copy_from_user(&data, argp, sizeof data))
1841                         goto out;
1842                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1843                 break;
1844         }
1845         case KVM_IOEVENTFD: {
1846                 struct kvm_ioeventfd data;
1847
1848                 r = -EFAULT;
1849                 if (copy_from_user(&data, argp, sizeof data))
1850                         goto out;
1851                 r = kvm_ioeventfd(kvm, &data);
1852                 break;
1853         }
1854 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1855         case KVM_SET_BOOT_CPU_ID:
1856                 r = 0;
1857                 mutex_lock(&kvm->lock);
1858                 if (atomic_read(&kvm->online_vcpus) != 0)
1859                         r = -EBUSY;
1860                 else
1861                         kvm->bsp_vcpu_id = arg;
1862                 mutex_unlock(&kvm->lock);
1863                 break;
1864 #endif
1865         default:
1866                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1867                 if (r == -ENOTTY)
1868                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1869         }
1870 out:
1871         return r;
1872 }
1873
1874 #ifdef CONFIG_COMPAT
1875 struct compat_kvm_dirty_log {
1876         __u32 slot;
1877         __u32 padding1;
1878         union {
1879                 compat_uptr_t dirty_bitmap; /* one bit per page */
1880                 __u64 padding2;
1881         };
1882 };
1883
1884 static long kvm_vm_compat_ioctl(struct file *filp,
1885                            unsigned int ioctl, unsigned long arg)
1886 {
1887         struct kvm *kvm = filp->private_data;
1888         int r;
1889
1890         if (kvm->mm != current->mm)
1891                 return -EIO;
1892         switch (ioctl) {
1893         case KVM_GET_DIRTY_LOG: {
1894                 struct compat_kvm_dirty_log compat_log;
1895                 struct kvm_dirty_log log;
1896
1897                 r = -EFAULT;
1898                 if (copy_from_user(&compat_log, (void __user *)arg,
1899                                    sizeof(compat_log)))
1900                         goto out;
1901                 log.slot         = compat_log.slot;
1902                 log.padding1     = compat_log.padding1;
1903                 log.padding2     = compat_log.padding2;
1904                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1905
1906                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1907                 if (r)
1908                         goto out;
1909                 break;
1910         }
1911         default:
1912                 r = kvm_vm_ioctl(filp, ioctl, arg);
1913         }
1914
1915 out:
1916         return r;
1917 }
1918 #endif
1919
1920 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1921 {
1922         struct page *page[1];
1923         unsigned long addr;
1924         int npages;
1925         gfn_t gfn = vmf->pgoff;
1926         struct kvm *kvm = vma->vm_file->private_data;
1927
1928         addr = gfn_to_hva(kvm, gfn);
1929         if (kvm_is_error_hva(addr))
1930                 return VM_FAULT_SIGBUS;
1931
1932         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1933                                 NULL);
1934         if (unlikely(npages != 1))
1935                 return VM_FAULT_SIGBUS;
1936
1937         vmf->page = page[0];
1938         return 0;
1939 }
1940
1941 static const struct vm_operations_struct kvm_vm_vm_ops = {
1942         .fault = kvm_vm_fault,
1943 };
1944
1945 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1946 {
1947         vma->vm_ops = &kvm_vm_vm_ops;
1948         return 0;
1949 }
1950
1951 static struct file_operations kvm_vm_fops = {
1952         .release        = kvm_vm_release,
1953         .unlocked_ioctl = kvm_vm_ioctl,
1954 #ifdef CONFIG_COMPAT
1955         .compat_ioctl   = kvm_vm_compat_ioctl,
1956 #endif
1957         .mmap           = kvm_vm_mmap,
1958         .llseek         = noop_llseek,
1959 };
1960
1961 static int kvm_dev_ioctl_create_vm(void)
1962 {
1963         int fd, r;
1964         struct kvm *kvm;
1965
1966         kvm = kvm_create_vm();
1967         if (IS_ERR(kvm))
1968                 return PTR_ERR(kvm);
1969 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1970         r = kvm_coalesced_mmio_init(kvm);
1971         if (r < 0) {
1972                 kvm_put_kvm(kvm);
1973                 return r;
1974         }
1975 #endif
1976         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1977         if (fd < 0)
1978                 kvm_put_kvm(kvm);
1979
1980         return fd;
1981 }
1982
1983 static long kvm_dev_ioctl_check_extension_generic(long arg)
1984 {
1985         switch (arg) {
1986         case KVM_CAP_USER_MEMORY:
1987         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1988         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1989 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1990         case KVM_CAP_SET_BOOT_CPU_ID:
1991 #endif
1992         case KVM_CAP_INTERNAL_ERROR_DATA:
1993                 return 1;
1994 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1995         case KVM_CAP_IRQ_ROUTING:
1996                 return KVM_MAX_IRQ_ROUTES;
1997 #endif
1998         default:
1999                 break;
2000         }
2001         return kvm_dev_ioctl_check_extension(arg);
2002 }
2003
2004 static long kvm_dev_ioctl(struct file *filp,
2005                           unsigned int ioctl, unsigned long arg)
2006 {
2007         long r = -EINVAL;
2008
2009         switch (ioctl) {
2010         case KVM_GET_API_VERSION:
2011                 r = -EINVAL;
2012                 if (arg)
2013                         goto out;
2014                 r = KVM_API_VERSION;
2015                 break;
2016         case KVM_CREATE_VM:
2017                 r = -EINVAL;
2018                 if (arg)
2019                         goto out;
2020                 r = kvm_dev_ioctl_create_vm();
2021                 break;
2022         case KVM_CHECK_EXTENSION:
2023                 r = kvm_dev_ioctl_check_extension_generic(arg);
2024                 break;
2025         case KVM_GET_VCPU_MMAP_SIZE:
2026                 r = -EINVAL;
2027                 if (arg)
2028                         goto out;
2029                 r = PAGE_SIZE;     /* struct kvm_run */
2030 #ifdef CONFIG_X86
2031                 r += PAGE_SIZE;    /* pio data page */
2032 #endif
2033 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2034                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2035 #endif
2036                 break;
2037         case KVM_TRACE_ENABLE:
2038         case KVM_TRACE_PAUSE:
2039         case KVM_TRACE_DISABLE:
2040                 r = -EOPNOTSUPP;
2041                 break;
2042         default:
2043                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2044         }
2045 out:
2046         return r;
2047 }
2048
2049 static struct file_operations kvm_chardev_ops = {
2050         .unlocked_ioctl = kvm_dev_ioctl,
2051         .compat_ioctl   = kvm_dev_ioctl,
2052         .llseek         = noop_llseek,
2053 };
2054
2055 static struct miscdevice kvm_dev = {
2056         KVM_MINOR,
2057         "kvm",
2058         &kvm_chardev_ops,
2059 };
2060
2061 static void hardware_enable(void *junk)
2062 {
2063         int cpu = raw_smp_processor_id();
2064         int r;
2065
2066         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2067                 return;
2068
2069         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2070
2071         r = kvm_arch_hardware_enable(NULL);
2072
2073         if (r) {
2074                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2075                 atomic_inc(&hardware_enable_failed);
2076                 printk(KERN_INFO "kvm: enabling virtualization on "
2077                                  "CPU%d failed\n", cpu);
2078         }
2079 }
2080
2081 static void hardware_disable(void *junk)
2082 {
2083         int cpu = raw_smp_processor_id();
2084
2085         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2086                 return;
2087         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2088         kvm_arch_hardware_disable(NULL);
2089 }
2090
2091 static void hardware_disable_all_nolock(void)
2092 {
2093         BUG_ON(!kvm_usage_count);
2094
2095         kvm_usage_count--;
2096         if (!kvm_usage_count)
2097                 on_each_cpu(hardware_disable, NULL, 1);
2098 }
2099
2100 static void hardware_disable_all(void)
2101 {
2102         spin_lock(&kvm_lock);
2103         hardware_disable_all_nolock();
2104         spin_unlock(&kvm_lock);
2105 }
2106
2107 static int hardware_enable_all(void)
2108 {
2109         int r = 0;
2110
2111         spin_lock(&kvm_lock);
2112
2113         kvm_usage_count++;
2114         if (kvm_usage_count == 1) {
2115                 atomic_set(&hardware_enable_failed, 0);
2116                 on_each_cpu(hardware_enable, NULL, 1);
2117
2118                 if (atomic_read(&hardware_enable_failed)) {
2119                         hardware_disable_all_nolock();
2120                         r = -EBUSY;
2121                 }
2122         }
2123
2124         spin_unlock(&kvm_lock);
2125
2126         return r;
2127 }
2128
2129 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2130                            void *v)
2131 {
2132         int cpu = (long)v;
2133
2134         if (!kvm_usage_count)
2135                 return NOTIFY_OK;
2136
2137         val &= ~CPU_TASKS_FROZEN;
2138         switch (val) {
2139         case CPU_DYING:
2140                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2141                        cpu);
2142                 hardware_disable(NULL);
2143                 break;
2144         case CPU_STARTING:
2145                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2146                        cpu);
2147                 spin_lock(&kvm_lock);
2148                 hardware_enable(NULL);
2149                 spin_unlock(&kvm_lock);
2150                 break;
2151         }
2152         return NOTIFY_OK;
2153 }
2154
2155
2156 asmlinkage void kvm_handle_fault_on_reboot(void)
2157 {
2158         if (kvm_rebooting) {
2159                 /* spin while reset goes on */
2160                 local_irq_enable();
2161                 while (true)
2162                         cpu_relax();
2163         }
2164         /* Fault while not rebooting.  We want the trace. */
2165         BUG();
2166 }
2167 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2168
2169 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2170                       void *v)
2171 {
2172         /*
2173          * Some (well, at least mine) BIOSes hang on reboot if
2174          * in vmx root mode.
2175          *
2176          * And Intel TXT required VMX off for all cpu when system shutdown.
2177          */
2178         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2179         kvm_rebooting = true;
2180         on_each_cpu(hardware_disable, NULL, 1);
2181         return NOTIFY_OK;
2182 }
2183
2184 static struct notifier_block kvm_reboot_notifier = {
2185         .notifier_call = kvm_reboot,
2186         .priority = 0,
2187 };
2188
2189 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2190 {
2191         int i;
2192
2193         for (i = 0; i < bus->dev_count; i++) {
2194                 struct kvm_io_device *pos = bus->devs[i];
2195
2196                 kvm_iodevice_destructor(pos);
2197         }
2198         kfree(bus);
2199 }
2200
2201 /* kvm_io_bus_write - called under kvm->slots_lock */
2202 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2203                      int len, const void *val)
2204 {
2205         int i;
2206         struct kvm_io_bus *bus;
2207
2208         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2209         for (i = 0; i < bus->dev_count; i++)
2210                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2211                         return 0;
2212         return -EOPNOTSUPP;
2213 }
2214
2215 /* kvm_io_bus_read - called under kvm->slots_lock */
2216 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2217                     int len, void *val)
2218 {
2219         int i;
2220         struct kvm_io_bus *bus;
2221
2222         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2223         for (i = 0; i < bus->dev_count; i++)
2224                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2225                         return 0;
2226         return -EOPNOTSUPP;
2227 }
2228
2229 /* Caller must hold slots_lock. */
2230 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2231                             struct kvm_io_device *dev)
2232 {
2233         struct kvm_io_bus *new_bus, *bus;
2234
2235         bus = kvm->buses[bus_idx];
2236         if (bus->dev_count > NR_IOBUS_DEVS-1)
2237                 return -ENOSPC;
2238
2239         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2240         if (!new_bus)
2241                 return -ENOMEM;
2242         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2243         new_bus->devs[new_bus->dev_count++] = dev;
2244         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2245         synchronize_srcu_expedited(&kvm->srcu);
2246         kfree(bus);
2247
2248         return 0;
2249 }
2250
2251 /* Caller must hold slots_lock. */
2252 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2253                               struct kvm_io_device *dev)
2254 {
2255         int i, r;
2256         struct kvm_io_bus *new_bus, *bus;
2257
2258         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2259         if (!new_bus)
2260                 return -ENOMEM;
2261
2262         bus = kvm->buses[bus_idx];
2263         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2264
2265         r = -ENOENT;
2266         for (i = 0; i < new_bus->dev_count; i++)
2267                 if (new_bus->devs[i] == dev) {
2268                         r = 0;
2269                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2270                         break;
2271                 }
2272
2273         if (r) {
2274                 kfree(new_bus);
2275                 return r;
2276         }
2277
2278         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2279         synchronize_srcu_expedited(&kvm->srcu);
2280         kfree(bus);
2281         return r;
2282 }
2283
2284 static struct notifier_block kvm_cpu_notifier = {
2285         .notifier_call = kvm_cpu_hotplug,
2286 };
2287
2288 static int vm_stat_get(void *_offset, u64 *val)
2289 {
2290         unsigned offset = (long)_offset;
2291         struct kvm *kvm;
2292
2293         *val = 0;
2294         spin_lock(&kvm_lock);
2295         list_for_each_entry(kvm, &vm_list, vm_list)
2296                 *val += *(u32 *)((void *)kvm + offset);
2297         spin_unlock(&kvm_lock);
2298         return 0;
2299 }
2300
2301 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2302
2303 static int vcpu_stat_get(void *_offset, u64 *val)
2304 {
2305         unsigned offset = (long)_offset;
2306         struct kvm *kvm;
2307         struct kvm_vcpu *vcpu;
2308         int i;
2309
2310         *val = 0;
2311         spin_lock(&kvm_lock);
2312         list_for_each_entry(kvm, &vm_list, vm_list)
2313                 kvm_for_each_vcpu(i, vcpu, kvm)
2314                         *val += *(u32 *)((void *)vcpu + offset);
2315
2316         spin_unlock(&kvm_lock);
2317         return 0;
2318 }
2319
2320 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2321
2322 static const struct file_operations *stat_fops[] = {
2323         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2324         [KVM_STAT_VM]   = &vm_stat_fops,
2325 };
2326
2327 static void kvm_init_debug(void)
2328 {
2329         struct kvm_stats_debugfs_item *p;
2330
2331         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2332         for (p = debugfs_entries; p->name; ++p)
2333                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2334                                                 (void *)(long)p->offset,
2335                                                 stat_fops[p->kind]);
2336 }
2337
2338 static void kvm_exit_debug(void)
2339 {
2340         struct kvm_stats_debugfs_item *p;
2341
2342         for (p = debugfs_entries; p->name; ++p)
2343                 debugfs_remove(p->dentry);
2344         debugfs_remove(kvm_debugfs_dir);
2345 }
2346
2347 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2348 {
2349         if (kvm_usage_count)
2350                 hardware_disable(NULL);
2351         return 0;
2352 }
2353
2354 static int kvm_resume(struct sys_device *dev)
2355 {
2356         if (kvm_usage_count) {
2357                 WARN_ON(spin_is_locked(&kvm_lock));
2358                 hardware_enable(NULL);
2359         }
2360         return 0;
2361 }
2362
2363 static struct sysdev_class kvm_sysdev_class = {
2364         .name = "kvm",
2365         .suspend = kvm_suspend,
2366         .resume = kvm_resume,
2367 };
2368
2369 static struct sys_device kvm_sysdev = {
2370         .id = 0,
2371         .cls = &kvm_sysdev_class,
2372 };
2373
2374 struct page *bad_page;
2375 pfn_t bad_pfn;
2376
2377 static inline
2378 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2379 {
2380         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2381 }
2382
2383 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2384 {
2385         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2386
2387         kvm_arch_vcpu_load(vcpu, cpu);
2388 }
2389
2390 static void kvm_sched_out(struct preempt_notifier *pn,
2391                           struct task_struct *next)
2392 {
2393         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2394
2395         kvm_arch_vcpu_put(vcpu);
2396 }
2397
2398 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2399                   struct module *module)
2400 {
2401         int r;
2402         int cpu;
2403
2404         r = kvm_arch_init(opaque);
2405         if (r)
2406                 goto out_fail;
2407
2408         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2409
2410         if (bad_page == NULL) {
2411                 r = -ENOMEM;
2412                 goto out;
2413         }
2414
2415         bad_pfn = page_to_pfn(bad_page);
2416
2417         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2418
2419         if (hwpoison_page == NULL) {
2420                 r = -ENOMEM;
2421                 goto out_free_0;
2422         }
2423
2424         hwpoison_pfn = page_to_pfn(hwpoison_page);
2425
2426         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2427
2428         if (fault_page == NULL) {
2429                 r = -ENOMEM;
2430                 goto out_free_0;
2431         }
2432
2433         fault_pfn = page_to_pfn(fault_page);
2434
2435         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2436                 r = -ENOMEM;
2437                 goto out_free_0;
2438         }
2439
2440         r = kvm_arch_hardware_setup();
2441         if (r < 0)
2442                 goto out_free_0a;
2443
2444         for_each_online_cpu(cpu) {
2445                 smp_call_function_single(cpu,
2446                                 kvm_arch_check_processor_compat,
2447                                 &r, 1);
2448                 if (r < 0)
2449                         goto out_free_1;
2450         }
2451
2452         r = register_cpu_notifier(&kvm_cpu_notifier);
2453         if (r)
2454                 goto out_free_2;
2455         register_reboot_notifier(&kvm_reboot_notifier);
2456
2457         r = sysdev_class_register(&kvm_sysdev_class);
2458         if (r)
2459                 goto out_free_3;
2460
2461         r = sysdev_register(&kvm_sysdev);
2462         if (r)
2463                 goto out_free_4;
2464
2465         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2466         if (!vcpu_align)
2467                 vcpu_align = __alignof__(struct kvm_vcpu);
2468         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2469                                            0, NULL);
2470         if (!kvm_vcpu_cache) {
2471                 r = -ENOMEM;
2472                 goto out_free_5;
2473         }
2474
2475         r = kvm_async_pf_init();
2476         if (r)
2477                 goto out_free;
2478
2479         kvm_chardev_ops.owner = module;
2480         kvm_vm_fops.owner = module;
2481         kvm_vcpu_fops.owner = module;
2482
2483         r = misc_register(&kvm_dev);
2484         if (r) {
2485                 printk(KERN_ERR "kvm: misc device register failed\n");
2486                 goto out_unreg;
2487         }
2488
2489         kvm_preempt_ops.sched_in = kvm_sched_in;
2490         kvm_preempt_ops.sched_out = kvm_sched_out;
2491
2492         kvm_init_debug();
2493
2494         return 0;
2495
2496 out_unreg:
2497         kvm_async_pf_deinit();
2498 out_free:
2499         kmem_cache_destroy(kvm_vcpu_cache);
2500 out_free_5:
2501         sysdev_unregister(&kvm_sysdev);
2502 out_free_4:
2503         sysdev_class_unregister(&kvm_sysdev_class);
2504 out_free_3:
2505         unregister_reboot_notifier(&kvm_reboot_notifier);
2506         unregister_cpu_notifier(&kvm_cpu_notifier);
2507 out_free_2:
2508 out_free_1:
2509         kvm_arch_hardware_unsetup();
2510 out_free_0a:
2511         free_cpumask_var(cpus_hardware_enabled);
2512 out_free_0:
2513         if (fault_page)
2514                 __free_page(fault_page);
2515         if (hwpoison_page)
2516                 __free_page(hwpoison_page);
2517         __free_page(bad_page);
2518 out:
2519         kvm_arch_exit();
2520 out_fail:
2521         return r;
2522 }
2523 EXPORT_SYMBOL_GPL(kvm_init);
2524
2525 void kvm_exit(void)
2526 {
2527         kvm_exit_debug();
2528         misc_deregister(&kvm_dev);
2529         kmem_cache_destroy(kvm_vcpu_cache);
2530         kvm_async_pf_deinit();
2531         sysdev_unregister(&kvm_sysdev);
2532         sysdev_class_unregister(&kvm_sysdev_class);
2533         unregister_reboot_notifier(&kvm_reboot_notifier);
2534         unregister_cpu_notifier(&kvm_cpu_notifier);
2535         on_each_cpu(hardware_disable, NULL, 1);
2536         kvm_arch_hardware_unsetup();
2537         kvm_arch_exit();
2538         free_cpumask_var(cpus_hardware_enabled);
2539         __free_page(hwpoison_page);
2540         __free_page(bad_page);
2541 }
2542 EXPORT_SYMBOL_GPL(kvm_exit);