thp: add compound_trans_head() helper
[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 bool kvm_rebooting;
94 EXPORT_SYMBOL_GPL(kvm_rebooting);
95
96 static bool largepages_enabled = true;
97
98 static struct page *hwpoison_page;
99 static pfn_t hwpoison_pfn;
100
101 static struct page *fault_page;
102 static pfn_t fault_pfn;
103
104 inline int kvm_is_mmio_pfn(pfn_t pfn)
105 {
106         if (pfn_valid(pfn)) {
107                 int reserved;
108                 struct page *tail = pfn_to_page(pfn);
109                 struct page *head = compound_trans_head(tail);
110                 reserved = PageReserved(head);
111                 if (head != tail) {
112                         /*
113                          * "head" is not a dangling pointer
114                          * (compound_trans_head takes care of that)
115                          * but the hugepage may have been splitted
116                          * from under us (and we may not hold a
117                          * reference count on the head page so it can
118                          * be reused before we run PageReferenced), so
119                          * we've to check PageTail before returning
120                          * what we just read.
121                          */
122                         smp_rmb();
123                         if (PageTail(tail))
124                                 return reserved;
125                 }
126                 return PageReserved(tail);
127         }
128
129         return true;
130 }
131
132 /*
133  * Switches to specified vcpu, until a matching vcpu_put()
134  */
135 void vcpu_load(struct kvm_vcpu *vcpu)
136 {
137         int cpu;
138
139         mutex_lock(&vcpu->mutex);
140         cpu = get_cpu();
141         preempt_notifier_register(&vcpu->preempt_notifier);
142         kvm_arch_vcpu_load(vcpu, cpu);
143         put_cpu();
144 }
145
146 void vcpu_put(struct kvm_vcpu *vcpu)
147 {
148         preempt_disable();
149         kvm_arch_vcpu_put(vcpu);
150         preempt_notifier_unregister(&vcpu->preempt_notifier);
151         preempt_enable();
152         mutex_unlock(&vcpu->mutex);
153 }
154
155 static void ack_flush(void *_completed)
156 {
157 }
158
159 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
160 {
161         int i, cpu, me;
162         cpumask_var_t cpus;
163         bool called = true;
164         struct kvm_vcpu *vcpu;
165
166         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
167
168         raw_spin_lock(&kvm->requests_lock);
169         me = smp_processor_id();
170         kvm_for_each_vcpu(i, vcpu, kvm) {
171                 if (kvm_make_check_request(req, vcpu))
172                         continue;
173                 cpu = vcpu->cpu;
174                 if (cpus != NULL && cpu != -1 && cpu != me)
175                         cpumask_set_cpu(cpu, cpus);
176         }
177         if (unlikely(cpus == NULL))
178                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179         else if (!cpumask_empty(cpus))
180                 smp_call_function_many(cpus, ack_flush, NULL, 1);
181         else
182                 called = false;
183         raw_spin_unlock(&kvm->requests_lock);
184         free_cpumask_var(cpus);
185         return called;
186 }
187
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
189 {
190         int dirty_count = kvm->tlbs_dirty;
191
192         smp_mb();
193         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194                 ++kvm->stat.remote_tlb_flush;
195         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
196 }
197
198 void kvm_reload_remote_mmus(struct kvm *kvm)
199 {
200         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
201 }
202
203 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
204 {
205         struct page *page;
206         int r;
207
208         mutex_init(&vcpu->mutex);
209         vcpu->cpu = -1;
210         vcpu->kvm = kvm;
211         vcpu->vcpu_id = id;
212         init_waitqueue_head(&vcpu->wq);
213         kvm_async_pf_vcpu_init(vcpu);
214
215         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
216         if (!page) {
217                 r = -ENOMEM;
218                 goto fail;
219         }
220         vcpu->run = page_address(page);
221
222         r = kvm_arch_vcpu_init(vcpu);
223         if (r < 0)
224                 goto fail_free_run;
225         return 0;
226
227 fail_free_run:
228         free_page((unsigned long)vcpu->run);
229 fail:
230         return r;
231 }
232 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
233
234 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
235 {
236         kvm_arch_vcpu_uninit(vcpu);
237         free_page((unsigned long)vcpu->run);
238 }
239 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
240
241 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
242 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
243 {
244         return container_of(mn, struct kvm, mmu_notifier);
245 }
246
247 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
248                                              struct mm_struct *mm,
249                                              unsigned long address)
250 {
251         struct kvm *kvm = mmu_notifier_to_kvm(mn);
252         int need_tlb_flush, idx;
253
254         /*
255          * When ->invalidate_page runs, the linux pte has been zapped
256          * already but the page is still allocated until
257          * ->invalidate_page returns. So if we increase the sequence
258          * here the kvm page fault will notice if the spte can't be
259          * established because the page is going to be freed. If
260          * instead the kvm page fault establishes the spte before
261          * ->invalidate_page runs, kvm_unmap_hva will release it
262          * before returning.
263          *
264          * The sequence increase only need to be seen at spin_unlock
265          * time, and not at spin_lock time.
266          *
267          * Increasing the sequence after the spin_unlock would be
268          * unsafe because the kvm page fault could then establish the
269          * pte after kvm_unmap_hva returned, without noticing the page
270          * is going to be freed.
271          */
272         idx = srcu_read_lock(&kvm->srcu);
273         spin_lock(&kvm->mmu_lock);
274         kvm->mmu_notifier_seq++;
275         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
276         spin_unlock(&kvm->mmu_lock);
277         srcu_read_unlock(&kvm->srcu, idx);
278
279         /* we've to flush the tlb before the pages can be freed */
280         if (need_tlb_flush)
281                 kvm_flush_remote_tlbs(kvm);
282
283 }
284
285 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
286                                         struct mm_struct *mm,
287                                         unsigned long address,
288                                         pte_t pte)
289 {
290         struct kvm *kvm = mmu_notifier_to_kvm(mn);
291         int idx;
292
293         idx = srcu_read_lock(&kvm->srcu);
294         spin_lock(&kvm->mmu_lock);
295         kvm->mmu_notifier_seq++;
296         kvm_set_spte_hva(kvm, address, pte);
297         spin_unlock(&kvm->mmu_lock);
298         srcu_read_unlock(&kvm->srcu, idx);
299 }
300
301 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
302                                                     struct mm_struct *mm,
303                                                     unsigned long start,
304                                                     unsigned long end)
305 {
306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
307         int need_tlb_flush = 0, idx;
308
309         idx = srcu_read_lock(&kvm->srcu);
310         spin_lock(&kvm->mmu_lock);
311         /*
312          * The count increase must become visible at unlock time as no
313          * spte can be established without taking the mmu_lock and
314          * count is also read inside the mmu_lock critical section.
315          */
316         kvm->mmu_notifier_count++;
317         for (; start < end; start += PAGE_SIZE)
318                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
319         need_tlb_flush |= kvm->tlbs_dirty;
320         spin_unlock(&kvm->mmu_lock);
321         srcu_read_unlock(&kvm->srcu, idx);
322
323         /* we've to flush the tlb before the pages can be freed */
324         if (need_tlb_flush)
325                 kvm_flush_remote_tlbs(kvm);
326 }
327
328 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
329                                                   struct mm_struct *mm,
330                                                   unsigned long start,
331                                                   unsigned long end)
332 {
333         struct kvm *kvm = mmu_notifier_to_kvm(mn);
334
335         spin_lock(&kvm->mmu_lock);
336         /*
337          * This sequence increase will notify the kvm page fault that
338          * the page that is going to be mapped in the spte could have
339          * been freed.
340          */
341         kvm->mmu_notifier_seq++;
342         /*
343          * The above sequence increase must be visible before the
344          * below count decrease but both values are read by the kvm
345          * page fault under mmu_lock spinlock so we don't need to add
346          * a smb_wmb() here in between the two.
347          */
348         kvm->mmu_notifier_count--;
349         spin_unlock(&kvm->mmu_lock);
350
351         BUG_ON(kvm->mmu_notifier_count < 0);
352 }
353
354 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
355                                               struct mm_struct *mm,
356                                               unsigned long address)
357 {
358         struct kvm *kvm = mmu_notifier_to_kvm(mn);
359         int young, idx;
360
361         idx = srcu_read_lock(&kvm->srcu);
362         spin_lock(&kvm->mmu_lock);
363         young = kvm_age_hva(kvm, address);
364         spin_unlock(&kvm->mmu_lock);
365         srcu_read_unlock(&kvm->srcu, idx);
366
367         if (young)
368                 kvm_flush_remote_tlbs(kvm);
369
370         return young;
371 }
372
373 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
374                                        struct mm_struct *mm,
375                                        unsigned long address)
376 {
377         struct kvm *kvm = mmu_notifier_to_kvm(mn);
378         int young, idx;
379
380         idx = srcu_read_lock(&kvm->srcu);
381         spin_lock(&kvm->mmu_lock);
382         young = kvm_test_age_hva(kvm, address);
383         spin_unlock(&kvm->mmu_lock);
384         srcu_read_unlock(&kvm->srcu, idx);
385
386         return young;
387 }
388
389 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
390                                      struct mm_struct *mm)
391 {
392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
393         int idx;
394
395         idx = srcu_read_lock(&kvm->srcu);
396         kvm_arch_flush_shadow(kvm);
397         srcu_read_unlock(&kvm->srcu, idx);
398 }
399
400 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
401         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
402         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
403         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
404         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
405         .test_young             = kvm_mmu_notifier_test_young,
406         .change_pte             = kvm_mmu_notifier_change_pte,
407         .release                = kvm_mmu_notifier_release,
408 };
409
410 static int kvm_init_mmu_notifier(struct kvm *kvm)
411 {
412         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
413         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
414 }
415
416 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
417
418 static int kvm_init_mmu_notifier(struct kvm *kvm)
419 {
420         return 0;
421 }
422
423 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
424
425 static struct kvm *kvm_create_vm(void)
426 {
427         int r, i;
428         struct kvm *kvm = kvm_arch_alloc_vm();
429
430         if (!kvm)
431                 return ERR_PTR(-ENOMEM);
432
433         r = kvm_arch_init_vm(kvm);
434         if (r)
435                 goto out_err_nodisable;
436
437         r = hardware_enable_all();
438         if (r)
439                 goto out_err_nodisable;
440
441 #ifdef CONFIG_HAVE_KVM_IRQCHIP
442         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
443         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
444 #endif
445
446         r = -ENOMEM;
447         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
448         if (!kvm->memslots)
449                 goto out_err_nosrcu;
450         if (init_srcu_struct(&kvm->srcu))
451                 goto out_err_nosrcu;
452         for (i = 0; i < KVM_NR_BUSES; i++) {
453                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
454                                         GFP_KERNEL);
455                 if (!kvm->buses[i])
456                         goto out_err;
457         }
458
459         r = kvm_init_mmu_notifier(kvm);
460         if (r)
461                 goto out_err;
462
463         kvm->mm = current->mm;
464         atomic_inc(&kvm->mm->mm_count);
465         spin_lock_init(&kvm->mmu_lock);
466         raw_spin_lock_init(&kvm->requests_lock);
467         kvm_eventfd_init(kvm);
468         mutex_init(&kvm->lock);
469         mutex_init(&kvm->irq_lock);
470         mutex_init(&kvm->slots_lock);
471         atomic_set(&kvm->users_count, 1);
472         spin_lock(&kvm_lock);
473         list_add(&kvm->vm_list, &vm_list);
474         spin_unlock(&kvm_lock);
475
476         return kvm;
477
478 out_err:
479         cleanup_srcu_struct(&kvm->srcu);
480 out_err_nosrcu:
481         hardware_disable_all();
482 out_err_nodisable:
483         for (i = 0; i < KVM_NR_BUSES; i++)
484                 kfree(kvm->buses[i]);
485         kfree(kvm->memslots);
486         kvm_arch_free_vm(kvm);
487         return ERR_PTR(r);
488 }
489
490 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
491 {
492         if (!memslot->dirty_bitmap)
493                 return;
494
495         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
496                 vfree(memslot->dirty_bitmap_head);
497         else
498                 kfree(memslot->dirty_bitmap_head);
499
500         memslot->dirty_bitmap = NULL;
501         memslot->dirty_bitmap_head = NULL;
502 }
503
504 /*
505  * Free any memory in @free but not in @dont.
506  */
507 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
508                                   struct kvm_memory_slot *dont)
509 {
510         int i;
511
512         if (!dont || free->rmap != dont->rmap)
513                 vfree(free->rmap);
514
515         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
516                 kvm_destroy_dirty_bitmap(free);
517
518
519         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
520                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
521                         vfree(free->lpage_info[i]);
522                         free->lpage_info[i] = NULL;
523                 }
524         }
525
526         free->npages = 0;
527         free->rmap = NULL;
528 }
529
530 void kvm_free_physmem(struct kvm *kvm)
531 {
532         int i;
533         struct kvm_memslots *slots = kvm->memslots;
534
535         for (i = 0; i < slots->nmemslots; ++i)
536                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
537
538         kfree(kvm->memslots);
539 }
540
541 static void kvm_destroy_vm(struct kvm *kvm)
542 {
543         int i;
544         struct mm_struct *mm = kvm->mm;
545
546         kvm_arch_sync_events(kvm);
547         spin_lock(&kvm_lock);
548         list_del(&kvm->vm_list);
549         spin_unlock(&kvm_lock);
550         kvm_free_irq_routing(kvm);
551         for (i = 0; i < KVM_NR_BUSES; i++)
552                 kvm_io_bus_destroy(kvm->buses[i]);
553         kvm_coalesced_mmio_free(kvm);
554 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
555         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
556 #else
557         kvm_arch_flush_shadow(kvm);
558 #endif
559         kvm_arch_destroy_vm(kvm);
560         kvm_free_physmem(kvm);
561         cleanup_srcu_struct(&kvm->srcu);
562         kvm_arch_free_vm(kvm);
563         hardware_disable_all();
564         mmdrop(mm);
565 }
566
567 void kvm_get_kvm(struct kvm *kvm)
568 {
569         atomic_inc(&kvm->users_count);
570 }
571 EXPORT_SYMBOL_GPL(kvm_get_kvm);
572
573 void kvm_put_kvm(struct kvm *kvm)
574 {
575         if (atomic_dec_and_test(&kvm->users_count))
576                 kvm_destroy_vm(kvm);
577 }
578 EXPORT_SYMBOL_GPL(kvm_put_kvm);
579
580
581 static int kvm_vm_release(struct inode *inode, struct file *filp)
582 {
583         struct kvm *kvm = filp->private_data;
584
585         kvm_irqfd_release(kvm);
586
587         kvm_put_kvm(kvm);
588         return 0;
589 }
590
591 /*
592  * Allocation size is twice as large as the actual dirty bitmap size.
593  * This makes it possible to do double buffering: see x86's
594  * kvm_vm_ioctl_get_dirty_log().
595  */
596 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
597 {
598         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
599
600         if (dirty_bytes > PAGE_SIZE)
601                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
602         else
603                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
604
605         if (!memslot->dirty_bitmap)
606                 return -ENOMEM;
607
608         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
609         return 0;
610 }
611
612 /*
613  * Allocate some memory and give it an address in the guest physical address
614  * space.
615  *
616  * Discontiguous memory is allowed, mostly for framebuffers.
617  *
618  * Must be called holding mmap_sem for write.
619  */
620 int __kvm_set_memory_region(struct kvm *kvm,
621                             struct kvm_userspace_memory_region *mem,
622                             int user_alloc)
623 {
624         int r, flush_shadow = 0;
625         gfn_t base_gfn;
626         unsigned long npages;
627         unsigned long i;
628         struct kvm_memory_slot *memslot;
629         struct kvm_memory_slot old, new;
630         struct kvm_memslots *slots, *old_memslots;
631
632         r = -EINVAL;
633         /* General sanity checks */
634         if (mem->memory_size & (PAGE_SIZE - 1))
635                 goto out;
636         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
637                 goto out;
638         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
639                 goto out;
640         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
641                 goto out;
642         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
643                 goto out;
644
645         memslot = &kvm->memslots->memslots[mem->slot];
646         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
647         npages = mem->memory_size >> PAGE_SHIFT;
648
649         r = -EINVAL;
650         if (npages > KVM_MEM_MAX_NR_PAGES)
651                 goto out;
652
653         if (!npages)
654                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
655
656         new = old = *memslot;
657
658         new.id = mem->slot;
659         new.base_gfn = base_gfn;
660         new.npages = npages;
661         new.flags = mem->flags;
662
663         /* Disallow changing a memory slot's size. */
664         r = -EINVAL;
665         if (npages && old.npages && npages != old.npages)
666                 goto out_free;
667
668         /* Check for overlaps */
669         r = -EEXIST;
670         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
671                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
672
673                 if (s == memslot || !s->npages)
674                         continue;
675                 if (!((base_gfn + npages <= s->base_gfn) ||
676                       (base_gfn >= s->base_gfn + s->npages)))
677                         goto out_free;
678         }
679
680         /* Free page dirty bitmap if unneeded */
681         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
682                 new.dirty_bitmap = NULL;
683
684         r = -ENOMEM;
685
686         /* Allocate if a slot is being created */
687 #ifndef CONFIG_S390
688         if (npages && !new.rmap) {
689                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
690
691                 if (!new.rmap)
692                         goto out_free;
693
694                 new.user_alloc = user_alloc;
695                 new.userspace_addr = mem->userspace_addr;
696         }
697         if (!npages)
698                 goto skip_lpage;
699
700         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
701                 unsigned long ugfn;
702                 unsigned long j;
703                 int lpages;
704                 int level = i + 2;
705
706                 /* Avoid unused variable warning if no large pages */
707                 (void)level;
708
709                 if (new.lpage_info[i])
710                         continue;
711
712                 lpages = 1 + ((base_gfn + npages - 1)
713                              >> KVM_HPAGE_GFN_SHIFT(level));
714                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
715
716                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
717
718                 if (!new.lpage_info[i])
719                         goto out_free;
720
721                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
722                         new.lpage_info[i][0].write_count = 1;
723                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
724                         new.lpage_info[i][lpages - 1].write_count = 1;
725                 ugfn = new.userspace_addr >> PAGE_SHIFT;
726                 /*
727                  * If the gfn and userspace address are not aligned wrt each
728                  * other, or if explicitly asked to, disable large page
729                  * support for this slot
730                  */
731                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
732                     !largepages_enabled)
733                         for (j = 0; j < lpages; ++j)
734                                 new.lpage_info[i][j].write_count = 1;
735         }
736
737 skip_lpage:
738
739         /* Allocate page dirty bitmap if needed */
740         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
741                 if (kvm_create_dirty_bitmap(&new) < 0)
742                         goto out_free;
743                 /* destroy any largepage mappings for dirty tracking */
744                 if (old.npages)
745                         flush_shadow = 1;
746         }
747 #else  /* not defined CONFIG_S390 */
748         new.user_alloc = user_alloc;
749         if (user_alloc)
750                 new.userspace_addr = mem->userspace_addr;
751 #endif /* not defined CONFIG_S390 */
752
753         if (!npages) {
754                 r = -ENOMEM;
755                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
756                 if (!slots)
757                         goto out_free;
758                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
759                 if (mem->slot >= slots->nmemslots)
760                         slots->nmemslots = mem->slot + 1;
761                 slots->generation++;
762                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
763
764                 old_memslots = kvm->memslots;
765                 rcu_assign_pointer(kvm->memslots, slots);
766                 synchronize_srcu_expedited(&kvm->srcu);
767                 /* From this point no new shadow pages pointing to a deleted
768                  * memslot will be created.
769                  *
770                  * validation of sp->gfn happens in:
771                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
772                  *      - kvm_is_visible_gfn (mmu_check_roots)
773                  */
774                 kvm_arch_flush_shadow(kvm);
775                 kfree(old_memslots);
776         }
777
778         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
779         if (r)
780                 goto out_free;
781
782         /* map the pages in iommu page table */
783         if (npages) {
784                 r = kvm_iommu_map_pages(kvm, &new);
785                 if (r)
786                         goto out_free;
787         }
788
789         r = -ENOMEM;
790         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
791         if (!slots)
792                 goto out_free;
793         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
794         if (mem->slot >= slots->nmemslots)
795                 slots->nmemslots = mem->slot + 1;
796         slots->generation++;
797
798         /* actual memory is freed via old in kvm_free_physmem_slot below */
799         if (!npages) {
800                 new.rmap = NULL;
801                 new.dirty_bitmap = NULL;
802                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
803                         new.lpage_info[i] = NULL;
804         }
805
806         slots->memslots[mem->slot] = new;
807         old_memslots = kvm->memslots;
808         rcu_assign_pointer(kvm->memslots, slots);
809         synchronize_srcu_expedited(&kvm->srcu);
810
811         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
812
813         kvm_free_physmem_slot(&old, &new);
814         kfree(old_memslots);
815
816         if (flush_shadow)
817                 kvm_arch_flush_shadow(kvm);
818
819         return 0;
820
821 out_free:
822         kvm_free_physmem_slot(&new, &old);
823 out:
824         return r;
825
826 }
827 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
828
829 int kvm_set_memory_region(struct kvm *kvm,
830                           struct kvm_userspace_memory_region *mem,
831                           int user_alloc)
832 {
833         int r;
834
835         mutex_lock(&kvm->slots_lock);
836         r = __kvm_set_memory_region(kvm, mem, user_alloc);
837         mutex_unlock(&kvm->slots_lock);
838         return r;
839 }
840 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
841
842 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
843                                    struct
844                                    kvm_userspace_memory_region *mem,
845                                    int user_alloc)
846 {
847         if (mem->slot >= KVM_MEMORY_SLOTS)
848                 return -EINVAL;
849         return kvm_set_memory_region(kvm, mem, user_alloc);
850 }
851
852 int kvm_get_dirty_log(struct kvm *kvm,
853                         struct kvm_dirty_log *log, int *is_dirty)
854 {
855         struct kvm_memory_slot *memslot;
856         int r, i;
857         unsigned long n;
858         unsigned long any = 0;
859
860         r = -EINVAL;
861         if (log->slot >= KVM_MEMORY_SLOTS)
862                 goto out;
863
864         memslot = &kvm->memslots->memslots[log->slot];
865         r = -ENOENT;
866         if (!memslot->dirty_bitmap)
867                 goto out;
868
869         n = kvm_dirty_bitmap_bytes(memslot);
870
871         for (i = 0; !any && i < n/sizeof(long); ++i)
872                 any = memslot->dirty_bitmap[i];
873
874         r = -EFAULT;
875         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
876                 goto out;
877
878         if (any)
879                 *is_dirty = 1;
880
881         r = 0;
882 out:
883         return r;
884 }
885
886 void kvm_disable_largepages(void)
887 {
888         largepages_enabled = false;
889 }
890 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
891
892 int is_error_page(struct page *page)
893 {
894         return page == bad_page || page == hwpoison_page || page == fault_page;
895 }
896 EXPORT_SYMBOL_GPL(is_error_page);
897
898 int is_error_pfn(pfn_t pfn)
899 {
900         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
901 }
902 EXPORT_SYMBOL_GPL(is_error_pfn);
903
904 int is_hwpoison_pfn(pfn_t pfn)
905 {
906         return pfn == hwpoison_pfn;
907 }
908 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
909
910 int is_fault_pfn(pfn_t pfn)
911 {
912         return pfn == fault_pfn;
913 }
914 EXPORT_SYMBOL_GPL(is_fault_pfn);
915
916 static inline unsigned long bad_hva(void)
917 {
918         return PAGE_OFFSET;
919 }
920
921 int kvm_is_error_hva(unsigned long addr)
922 {
923         return addr == bad_hva();
924 }
925 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
926
927 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
928                                                 gfn_t gfn)
929 {
930         int i;
931
932         for (i = 0; i < slots->nmemslots; ++i) {
933                 struct kvm_memory_slot *memslot = &slots->memslots[i];
934
935                 if (gfn >= memslot->base_gfn
936                     && gfn < memslot->base_gfn + memslot->npages)
937                         return memslot;
938         }
939         return NULL;
940 }
941
942 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
943 {
944         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
945 }
946 EXPORT_SYMBOL_GPL(gfn_to_memslot);
947
948 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
949 {
950         int i;
951         struct kvm_memslots *slots = kvm_memslots(kvm);
952
953         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
954                 struct kvm_memory_slot *memslot = &slots->memslots[i];
955
956                 if (memslot->flags & KVM_MEMSLOT_INVALID)
957                         continue;
958
959                 if (gfn >= memslot->base_gfn
960                     && gfn < memslot->base_gfn + memslot->npages)
961                         return 1;
962         }
963         return 0;
964 }
965 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
966
967 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
968 {
969         struct vm_area_struct *vma;
970         unsigned long addr, size;
971
972         size = PAGE_SIZE;
973
974         addr = gfn_to_hva(kvm, gfn);
975         if (kvm_is_error_hva(addr))
976                 return PAGE_SIZE;
977
978         down_read(&current->mm->mmap_sem);
979         vma = find_vma(current->mm, addr);
980         if (!vma)
981                 goto out;
982
983         size = vma_kernel_pagesize(vma);
984
985 out:
986         up_read(&current->mm->mmap_sem);
987
988         return size;
989 }
990
991 int memslot_id(struct kvm *kvm, gfn_t gfn)
992 {
993         int i;
994         struct kvm_memslots *slots = kvm_memslots(kvm);
995         struct kvm_memory_slot *memslot = NULL;
996
997         for (i = 0; i < slots->nmemslots; ++i) {
998                 memslot = &slots->memslots[i];
999
1000                 if (gfn >= memslot->base_gfn
1001                     && gfn < memslot->base_gfn + memslot->npages)
1002                         break;
1003         }
1004
1005         return memslot - slots->memslots;
1006 }
1007
1008 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1009                                      gfn_t *nr_pages)
1010 {
1011         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1012                 return bad_hva();
1013
1014         if (nr_pages)
1015                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1016
1017         return gfn_to_hva_memslot(slot, gfn);
1018 }
1019
1020 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1021 {
1022         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1023 }
1024 EXPORT_SYMBOL_GPL(gfn_to_hva);
1025
1026 static pfn_t get_fault_pfn(void)
1027 {
1028         get_page(fault_page);
1029         return fault_pfn;
1030 }
1031
1032 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1033                         bool *async, bool write_fault, bool *writable)
1034 {
1035         struct page *page[1];
1036         int npages = 0;
1037         pfn_t pfn;
1038
1039         /* we can do it either atomically or asynchronously, not both */
1040         BUG_ON(atomic && async);
1041
1042         BUG_ON(!write_fault && !writable);
1043
1044         if (writable)
1045                 *writable = true;
1046
1047         if (atomic || async)
1048                 npages = __get_user_pages_fast(addr, 1, 1, page);
1049
1050         if (unlikely(npages != 1) && !atomic) {
1051                 might_sleep();
1052
1053                 if (writable)
1054                         *writable = write_fault;
1055
1056                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1057
1058                 /* map read fault as writable if possible */
1059                 if (unlikely(!write_fault) && npages == 1) {
1060                         struct page *wpage[1];
1061
1062                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1063                         if (npages == 1) {
1064                                 *writable = true;
1065                                 put_page(page[0]);
1066                                 page[0] = wpage[0];
1067                         }
1068                         npages = 1;
1069                 }
1070         }
1071
1072         if (unlikely(npages != 1)) {
1073                 struct vm_area_struct *vma;
1074
1075                 if (atomic)
1076                         return get_fault_pfn();
1077
1078                 down_read(&current->mm->mmap_sem);
1079                 if (is_hwpoison_address(addr)) {
1080                         up_read(&current->mm->mmap_sem);
1081                         get_page(hwpoison_page);
1082                         return page_to_pfn(hwpoison_page);
1083                 }
1084
1085                 vma = find_vma_intersection(current->mm, addr, addr+1);
1086
1087                 if (vma == NULL)
1088                         pfn = get_fault_pfn();
1089                 else if ((vma->vm_flags & VM_PFNMAP)) {
1090                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1091                                 vma->vm_pgoff;
1092                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1093                 } else {
1094                         if (async && (vma->vm_flags & VM_WRITE))
1095                                 *async = true;
1096                         pfn = get_fault_pfn();
1097                 }
1098                 up_read(&current->mm->mmap_sem);
1099         } else
1100                 pfn = page_to_pfn(page[0]);
1101
1102         return pfn;
1103 }
1104
1105 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1106 {
1107         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1108 }
1109 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1110
1111 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1112                           bool write_fault, bool *writable)
1113 {
1114         unsigned long addr;
1115
1116         if (async)
1117                 *async = false;
1118
1119         addr = gfn_to_hva(kvm, gfn);
1120         if (kvm_is_error_hva(addr)) {
1121                 get_page(bad_page);
1122                 return page_to_pfn(bad_page);
1123         }
1124
1125         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1126 }
1127
1128 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1129 {
1130         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1131 }
1132 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1133
1134 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1135                        bool write_fault, bool *writable)
1136 {
1137         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1138 }
1139 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1140
1141 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1142 {
1143         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1144 }
1145 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1146
1147 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1148                       bool *writable)
1149 {
1150         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1151 }
1152 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1153
1154 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1155                          struct kvm_memory_slot *slot, gfn_t gfn)
1156 {
1157         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1158         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1159 }
1160
1161 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1162                                                                   int nr_pages)
1163 {
1164         unsigned long addr;
1165         gfn_t entry;
1166
1167         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1168         if (kvm_is_error_hva(addr))
1169                 return -1;
1170
1171         if (entry < nr_pages)
1172                 return 0;
1173
1174         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1175 }
1176 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1177
1178 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1179 {
1180         pfn_t pfn;
1181
1182         pfn = gfn_to_pfn(kvm, gfn);
1183         if (!kvm_is_mmio_pfn(pfn))
1184                 return pfn_to_page(pfn);
1185
1186         WARN_ON(kvm_is_mmio_pfn(pfn));
1187
1188         get_page(bad_page);
1189         return bad_page;
1190 }
1191
1192 EXPORT_SYMBOL_GPL(gfn_to_page);
1193
1194 void kvm_release_page_clean(struct page *page)
1195 {
1196         kvm_release_pfn_clean(page_to_pfn(page));
1197 }
1198 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1199
1200 void kvm_release_pfn_clean(pfn_t pfn)
1201 {
1202         if (!kvm_is_mmio_pfn(pfn))
1203                 put_page(pfn_to_page(pfn));
1204 }
1205 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1206
1207 void kvm_release_page_dirty(struct page *page)
1208 {
1209         kvm_release_pfn_dirty(page_to_pfn(page));
1210 }
1211 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1212
1213 void kvm_release_pfn_dirty(pfn_t pfn)
1214 {
1215         kvm_set_pfn_dirty(pfn);
1216         kvm_release_pfn_clean(pfn);
1217 }
1218 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1219
1220 void kvm_set_page_dirty(struct page *page)
1221 {
1222         kvm_set_pfn_dirty(page_to_pfn(page));
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1225
1226 void kvm_set_pfn_dirty(pfn_t pfn)
1227 {
1228         if (!kvm_is_mmio_pfn(pfn)) {
1229                 struct page *page = pfn_to_page(pfn);
1230                 if (!PageReserved(page))
1231                         SetPageDirty(page);
1232         }
1233 }
1234 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1235
1236 void kvm_set_pfn_accessed(pfn_t pfn)
1237 {
1238         if (!kvm_is_mmio_pfn(pfn))
1239                 mark_page_accessed(pfn_to_page(pfn));
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1242
1243 void kvm_get_pfn(pfn_t pfn)
1244 {
1245         if (!kvm_is_mmio_pfn(pfn))
1246                 get_page(pfn_to_page(pfn));
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1249
1250 static int next_segment(unsigned long len, int offset)
1251 {
1252         if (len > PAGE_SIZE - offset)
1253                 return PAGE_SIZE - offset;
1254         else
1255                 return len;
1256 }
1257
1258 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1259                         int len)
1260 {
1261         int r;
1262         unsigned long addr;
1263
1264         addr = gfn_to_hva(kvm, gfn);
1265         if (kvm_is_error_hva(addr))
1266                 return -EFAULT;
1267         r = copy_from_user(data, (void __user *)addr + offset, len);
1268         if (r)
1269                 return -EFAULT;
1270         return 0;
1271 }
1272 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1273
1274 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1275 {
1276         gfn_t gfn = gpa >> PAGE_SHIFT;
1277         int seg;
1278         int offset = offset_in_page(gpa);
1279         int ret;
1280
1281         while ((seg = next_segment(len, offset)) != 0) {
1282                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1283                 if (ret < 0)
1284                         return ret;
1285                 offset = 0;
1286                 len -= seg;
1287                 data += seg;
1288                 ++gfn;
1289         }
1290         return 0;
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_read_guest);
1293
1294 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1295                           unsigned long len)
1296 {
1297         int r;
1298         unsigned long addr;
1299         gfn_t gfn = gpa >> PAGE_SHIFT;
1300         int offset = offset_in_page(gpa);
1301
1302         addr = gfn_to_hva(kvm, gfn);
1303         if (kvm_is_error_hva(addr))
1304                 return -EFAULT;
1305         pagefault_disable();
1306         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1307         pagefault_enable();
1308         if (r)
1309                 return -EFAULT;
1310         return 0;
1311 }
1312 EXPORT_SYMBOL(kvm_read_guest_atomic);
1313
1314 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1315                          int offset, int len)
1316 {
1317         int r;
1318         unsigned long addr;
1319
1320         addr = gfn_to_hva(kvm, gfn);
1321         if (kvm_is_error_hva(addr))
1322                 return -EFAULT;
1323         r = copy_to_user((void __user *)addr + offset, data, len);
1324         if (r)
1325                 return -EFAULT;
1326         mark_page_dirty(kvm, gfn);
1327         return 0;
1328 }
1329 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1330
1331 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1332                     unsigned long len)
1333 {
1334         gfn_t gfn = gpa >> PAGE_SHIFT;
1335         int seg;
1336         int offset = offset_in_page(gpa);
1337         int ret;
1338
1339         while ((seg = next_segment(len, offset)) != 0) {
1340                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1341                 if (ret < 0)
1342                         return ret;
1343                 offset = 0;
1344                 len -= seg;
1345                 data += seg;
1346                 ++gfn;
1347         }
1348         return 0;
1349 }
1350
1351 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1352                               gpa_t gpa)
1353 {
1354         struct kvm_memslots *slots = kvm_memslots(kvm);
1355         int offset = offset_in_page(gpa);
1356         gfn_t gfn = gpa >> PAGE_SHIFT;
1357
1358         ghc->gpa = gpa;
1359         ghc->generation = slots->generation;
1360         ghc->memslot = __gfn_to_memslot(slots, gfn);
1361         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1362         if (!kvm_is_error_hva(ghc->hva))
1363                 ghc->hva += offset;
1364         else
1365                 return -EFAULT;
1366
1367         return 0;
1368 }
1369 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1370
1371 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1372                            void *data, unsigned long len)
1373 {
1374         struct kvm_memslots *slots = kvm_memslots(kvm);
1375         int r;
1376
1377         if (slots->generation != ghc->generation)
1378                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1379
1380         if (kvm_is_error_hva(ghc->hva))
1381                 return -EFAULT;
1382
1383         r = copy_to_user((void __user *)ghc->hva, data, len);
1384         if (r)
1385                 return -EFAULT;
1386         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1387
1388         return 0;
1389 }
1390 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1391
1392 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1393 {
1394         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1395                                     offset, len);
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1398
1399 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1400 {
1401         gfn_t gfn = gpa >> PAGE_SHIFT;
1402         int seg;
1403         int offset = offset_in_page(gpa);
1404         int ret;
1405
1406         while ((seg = next_segment(len, offset)) != 0) {
1407                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1408                 if (ret < 0)
1409                         return ret;
1410                 offset = 0;
1411                 len -= seg;
1412                 ++gfn;
1413         }
1414         return 0;
1415 }
1416 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1417
1418 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1419                              gfn_t gfn)
1420 {
1421         if (memslot && memslot->dirty_bitmap) {
1422                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1423
1424                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1425         }
1426 }
1427
1428 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1429 {
1430         struct kvm_memory_slot *memslot;
1431
1432         memslot = gfn_to_memslot(kvm, gfn);
1433         mark_page_dirty_in_slot(kvm, memslot, gfn);
1434 }
1435
1436 /*
1437  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1438  */
1439 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1440 {
1441         DEFINE_WAIT(wait);
1442
1443         for (;;) {
1444                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1445
1446                 if (kvm_arch_vcpu_runnable(vcpu)) {
1447                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1448                         break;
1449                 }
1450                 if (kvm_cpu_has_pending_timer(vcpu))
1451                         break;
1452                 if (signal_pending(current))
1453                         break;
1454
1455                 schedule();
1456         }
1457
1458         finish_wait(&vcpu->wq, &wait);
1459 }
1460
1461 void kvm_resched(struct kvm_vcpu *vcpu)
1462 {
1463         if (!need_resched())
1464                 return;
1465         cond_resched();
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_resched);
1468
1469 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1470 {
1471         ktime_t expires;
1472         DEFINE_WAIT(wait);
1473
1474         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1475
1476         /* Sleep for 100 us, and hope lock-holder got scheduled */
1477         expires = ktime_add_ns(ktime_get(), 100000UL);
1478         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1479
1480         finish_wait(&vcpu->wq, &wait);
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1483
1484 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1485 {
1486         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1487         struct page *page;
1488
1489         if (vmf->pgoff == 0)
1490                 page = virt_to_page(vcpu->run);
1491 #ifdef CONFIG_X86
1492         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1493                 page = virt_to_page(vcpu->arch.pio_data);
1494 #endif
1495 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1496         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1497                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1498 #endif
1499         else
1500                 return VM_FAULT_SIGBUS;
1501         get_page(page);
1502         vmf->page = page;
1503         return 0;
1504 }
1505
1506 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1507         .fault = kvm_vcpu_fault,
1508 };
1509
1510 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1511 {
1512         vma->vm_ops = &kvm_vcpu_vm_ops;
1513         return 0;
1514 }
1515
1516 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1517 {
1518         struct kvm_vcpu *vcpu = filp->private_data;
1519
1520         kvm_put_kvm(vcpu->kvm);
1521         return 0;
1522 }
1523
1524 static struct file_operations kvm_vcpu_fops = {
1525         .release        = kvm_vcpu_release,
1526         .unlocked_ioctl = kvm_vcpu_ioctl,
1527         .compat_ioctl   = kvm_vcpu_ioctl,
1528         .mmap           = kvm_vcpu_mmap,
1529         .llseek         = noop_llseek,
1530 };
1531
1532 /*
1533  * Allocates an inode for the vcpu.
1534  */
1535 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1536 {
1537         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1538 }
1539
1540 /*
1541  * Creates some virtual cpus.  Good luck creating more than one.
1542  */
1543 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1544 {
1545         int r;
1546         struct kvm_vcpu *vcpu, *v;
1547
1548         vcpu = kvm_arch_vcpu_create(kvm, id);
1549         if (IS_ERR(vcpu))
1550                 return PTR_ERR(vcpu);
1551
1552         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1553
1554         r = kvm_arch_vcpu_setup(vcpu);
1555         if (r)
1556                 return r;
1557
1558         mutex_lock(&kvm->lock);
1559         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1560                 r = -EINVAL;
1561                 goto vcpu_destroy;
1562         }
1563
1564         kvm_for_each_vcpu(r, v, kvm)
1565                 if (v->vcpu_id == id) {
1566                         r = -EEXIST;
1567                         goto vcpu_destroy;
1568                 }
1569
1570         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1571
1572         /* Now it's all set up, let userspace reach it */
1573         kvm_get_kvm(kvm);
1574         r = create_vcpu_fd(vcpu);
1575         if (r < 0) {
1576                 kvm_put_kvm(kvm);
1577                 goto vcpu_destroy;
1578         }
1579
1580         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1581         smp_wmb();
1582         atomic_inc(&kvm->online_vcpus);
1583
1584 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1585         if (kvm->bsp_vcpu_id == id)
1586                 kvm->bsp_vcpu = vcpu;
1587 #endif
1588         mutex_unlock(&kvm->lock);
1589         return r;
1590
1591 vcpu_destroy:
1592         mutex_unlock(&kvm->lock);
1593         kvm_arch_vcpu_destroy(vcpu);
1594         return r;
1595 }
1596
1597 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1598 {
1599         if (sigset) {
1600                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1601                 vcpu->sigset_active = 1;
1602                 vcpu->sigset = *sigset;
1603         } else
1604                 vcpu->sigset_active = 0;
1605         return 0;
1606 }
1607
1608 static long kvm_vcpu_ioctl(struct file *filp,
1609                            unsigned int ioctl, unsigned long arg)
1610 {
1611         struct kvm_vcpu *vcpu = filp->private_data;
1612         void __user *argp = (void __user *)arg;
1613         int r;
1614         struct kvm_fpu *fpu = NULL;
1615         struct kvm_sregs *kvm_sregs = NULL;
1616
1617         if (vcpu->kvm->mm != current->mm)
1618                 return -EIO;
1619
1620 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1621         /*
1622          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1623          * so vcpu_load() would break it.
1624          */
1625         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1626                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1627 #endif
1628
1629
1630         vcpu_load(vcpu);
1631         switch (ioctl) {
1632         case KVM_RUN:
1633                 r = -EINVAL;
1634                 if (arg)
1635                         goto out;
1636                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1637                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1638                 break;
1639         case KVM_GET_REGS: {
1640                 struct kvm_regs *kvm_regs;
1641
1642                 r = -ENOMEM;
1643                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1644                 if (!kvm_regs)
1645                         goto out;
1646                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1647                 if (r)
1648                         goto out_free1;
1649                 r = -EFAULT;
1650                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1651                         goto out_free1;
1652                 r = 0;
1653 out_free1:
1654                 kfree(kvm_regs);
1655                 break;
1656         }
1657         case KVM_SET_REGS: {
1658                 struct kvm_regs *kvm_regs;
1659
1660                 r = -ENOMEM;
1661                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1662                 if (!kvm_regs)
1663                         goto out;
1664                 r = -EFAULT;
1665                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1666                         goto out_free2;
1667                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1668                 if (r)
1669                         goto out_free2;
1670                 r = 0;
1671 out_free2:
1672                 kfree(kvm_regs);
1673                 break;
1674         }
1675         case KVM_GET_SREGS: {
1676                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1677                 r = -ENOMEM;
1678                 if (!kvm_sregs)
1679                         goto out;
1680                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1681                 if (r)
1682                         goto out;
1683                 r = -EFAULT;
1684                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1685                         goto out;
1686                 r = 0;
1687                 break;
1688         }
1689         case KVM_SET_SREGS: {
1690                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1691                 r = -ENOMEM;
1692                 if (!kvm_sregs)
1693                         goto out;
1694                 r = -EFAULT;
1695                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1696                         goto out;
1697                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1698                 if (r)
1699                         goto out;
1700                 r = 0;
1701                 break;
1702         }
1703         case KVM_GET_MP_STATE: {
1704                 struct kvm_mp_state mp_state;
1705
1706                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1707                 if (r)
1708                         goto out;
1709                 r = -EFAULT;
1710                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1711                         goto out;
1712                 r = 0;
1713                 break;
1714         }
1715         case KVM_SET_MP_STATE: {
1716                 struct kvm_mp_state mp_state;
1717
1718                 r = -EFAULT;
1719                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1720                         goto out;
1721                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1722                 if (r)
1723                         goto out;
1724                 r = 0;
1725                 break;
1726         }
1727         case KVM_TRANSLATE: {
1728                 struct kvm_translation tr;
1729
1730                 r = -EFAULT;
1731                 if (copy_from_user(&tr, argp, sizeof tr))
1732                         goto out;
1733                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1734                 if (r)
1735                         goto out;
1736                 r = -EFAULT;
1737                 if (copy_to_user(argp, &tr, sizeof tr))
1738                         goto out;
1739                 r = 0;
1740                 break;
1741         }
1742         case KVM_SET_GUEST_DEBUG: {
1743                 struct kvm_guest_debug dbg;
1744
1745                 r = -EFAULT;
1746                 if (copy_from_user(&dbg, argp, sizeof dbg))
1747                         goto out;
1748                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1749                 if (r)
1750                         goto out;
1751                 r = 0;
1752                 break;
1753         }
1754         case KVM_SET_SIGNAL_MASK: {
1755                 struct kvm_signal_mask __user *sigmask_arg = argp;
1756                 struct kvm_signal_mask kvm_sigmask;
1757                 sigset_t sigset, *p;
1758
1759                 p = NULL;
1760                 if (argp) {
1761                         r = -EFAULT;
1762                         if (copy_from_user(&kvm_sigmask, argp,
1763                                            sizeof kvm_sigmask))
1764                                 goto out;
1765                         r = -EINVAL;
1766                         if (kvm_sigmask.len != sizeof sigset)
1767                                 goto out;
1768                         r = -EFAULT;
1769                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1770                                            sizeof sigset))
1771                                 goto out;
1772                         p = &sigset;
1773                 }
1774                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1775                 break;
1776         }
1777         case KVM_GET_FPU: {
1778                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1779                 r = -ENOMEM;
1780                 if (!fpu)
1781                         goto out;
1782                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1783                 if (r)
1784                         goto out;
1785                 r = -EFAULT;
1786                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1787                         goto out;
1788                 r = 0;
1789                 break;
1790         }
1791         case KVM_SET_FPU: {
1792                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1793                 r = -ENOMEM;
1794                 if (!fpu)
1795                         goto out;
1796                 r = -EFAULT;
1797                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1798                         goto out;
1799                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1800                 if (r)
1801                         goto out;
1802                 r = 0;
1803                 break;
1804         }
1805         default:
1806                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1807         }
1808 out:
1809         vcpu_put(vcpu);
1810         kfree(fpu);
1811         kfree(kvm_sregs);
1812         return r;
1813 }
1814
1815 static long kvm_vm_ioctl(struct file *filp,
1816                            unsigned int ioctl, unsigned long arg)
1817 {
1818         struct kvm *kvm = filp->private_data;
1819         void __user *argp = (void __user *)arg;
1820         int r;
1821
1822         if (kvm->mm != current->mm)
1823                 return -EIO;
1824         switch (ioctl) {
1825         case KVM_CREATE_VCPU:
1826                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1827                 if (r < 0)
1828                         goto out;
1829                 break;
1830         case KVM_SET_USER_MEMORY_REGION: {
1831                 struct kvm_userspace_memory_region kvm_userspace_mem;
1832
1833                 r = -EFAULT;
1834                 if (copy_from_user(&kvm_userspace_mem, argp,
1835                                                 sizeof kvm_userspace_mem))
1836                         goto out;
1837
1838                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1839                 if (r)
1840                         goto out;
1841                 break;
1842         }
1843         case KVM_GET_DIRTY_LOG: {
1844                 struct kvm_dirty_log log;
1845
1846                 r = -EFAULT;
1847                 if (copy_from_user(&log, argp, sizeof log))
1848                         goto out;
1849                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1850                 if (r)
1851                         goto out;
1852                 break;
1853         }
1854 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1855         case KVM_REGISTER_COALESCED_MMIO: {
1856                 struct kvm_coalesced_mmio_zone zone;
1857                 r = -EFAULT;
1858                 if (copy_from_user(&zone, argp, sizeof zone))
1859                         goto out;
1860                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1861                 if (r)
1862                         goto out;
1863                 r = 0;
1864                 break;
1865         }
1866         case KVM_UNREGISTER_COALESCED_MMIO: {
1867                 struct kvm_coalesced_mmio_zone zone;
1868                 r = -EFAULT;
1869                 if (copy_from_user(&zone, argp, sizeof zone))
1870                         goto out;
1871                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1872                 if (r)
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877 #endif
1878         case KVM_IRQFD: {
1879                 struct kvm_irqfd data;
1880
1881                 r = -EFAULT;
1882                 if (copy_from_user(&data, argp, sizeof data))
1883                         goto out;
1884                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1885                 break;
1886         }
1887         case KVM_IOEVENTFD: {
1888                 struct kvm_ioeventfd data;
1889
1890                 r = -EFAULT;
1891                 if (copy_from_user(&data, argp, sizeof data))
1892                         goto out;
1893                 r = kvm_ioeventfd(kvm, &data);
1894                 break;
1895         }
1896 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1897         case KVM_SET_BOOT_CPU_ID:
1898                 r = 0;
1899                 mutex_lock(&kvm->lock);
1900                 if (atomic_read(&kvm->online_vcpus) != 0)
1901                         r = -EBUSY;
1902                 else
1903                         kvm->bsp_vcpu_id = arg;
1904                 mutex_unlock(&kvm->lock);
1905                 break;
1906 #endif
1907         default:
1908                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1909                 if (r == -ENOTTY)
1910                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1911         }
1912 out:
1913         return r;
1914 }
1915
1916 #ifdef CONFIG_COMPAT
1917 struct compat_kvm_dirty_log {
1918         __u32 slot;
1919         __u32 padding1;
1920         union {
1921                 compat_uptr_t dirty_bitmap; /* one bit per page */
1922                 __u64 padding2;
1923         };
1924 };
1925
1926 static long kvm_vm_compat_ioctl(struct file *filp,
1927                            unsigned int ioctl, unsigned long arg)
1928 {
1929         struct kvm *kvm = filp->private_data;
1930         int r;
1931
1932         if (kvm->mm != current->mm)
1933                 return -EIO;
1934         switch (ioctl) {
1935         case KVM_GET_DIRTY_LOG: {
1936                 struct compat_kvm_dirty_log compat_log;
1937                 struct kvm_dirty_log log;
1938
1939                 r = -EFAULT;
1940                 if (copy_from_user(&compat_log, (void __user *)arg,
1941                                    sizeof(compat_log)))
1942                         goto out;
1943                 log.slot         = compat_log.slot;
1944                 log.padding1     = compat_log.padding1;
1945                 log.padding2     = compat_log.padding2;
1946                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1947
1948                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1949                 if (r)
1950                         goto out;
1951                 break;
1952         }
1953         default:
1954                 r = kvm_vm_ioctl(filp, ioctl, arg);
1955         }
1956
1957 out:
1958         return r;
1959 }
1960 #endif
1961
1962 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1963 {
1964         struct page *page[1];
1965         unsigned long addr;
1966         int npages;
1967         gfn_t gfn = vmf->pgoff;
1968         struct kvm *kvm = vma->vm_file->private_data;
1969
1970         addr = gfn_to_hva(kvm, gfn);
1971         if (kvm_is_error_hva(addr))
1972                 return VM_FAULT_SIGBUS;
1973
1974         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1975                                 NULL);
1976         if (unlikely(npages != 1))
1977                 return VM_FAULT_SIGBUS;
1978
1979         vmf->page = page[0];
1980         return 0;
1981 }
1982
1983 static const struct vm_operations_struct kvm_vm_vm_ops = {
1984         .fault = kvm_vm_fault,
1985 };
1986
1987 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1988 {
1989         vma->vm_ops = &kvm_vm_vm_ops;
1990         return 0;
1991 }
1992
1993 static struct file_operations kvm_vm_fops = {
1994         .release        = kvm_vm_release,
1995         .unlocked_ioctl = kvm_vm_ioctl,
1996 #ifdef CONFIG_COMPAT
1997         .compat_ioctl   = kvm_vm_compat_ioctl,
1998 #endif
1999         .mmap           = kvm_vm_mmap,
2000         .llseek         = noop_llseek,
2001 };
2002
2003 static int kvm_dev_ioctl_create_vm(void)
2004 {
2005         int r;
2006         struct kvm *kvm;
2007
2008         kvm = kvm_create_vm();
2009         if (IS_ERR(kvm))
2010                 return PTR_ERR(kvm);
2011 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2012         r = kvm_coalesced_mmio_init(kvm);
2013         if (r < 0) {
2014                 kvm_put_kvm(kvm);
2015                 return r;
2016         }
2017 #endif
2018         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2019         if (r < 0)
2020                 kvm_put_kvm(kvm);
2021
2022         return r;
2023 }
2024
2025 static long kvm_dev_ioctl_check_extension_generic(long arg)
2026 {
2027         switch (arg) {
2028         case KVM_CAP_USER_MEMORY:
2029         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2030         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2031 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2032         case KVM_CAP_SET_BOOT_CPU_ID:
2033 #endif
2034         case KVM_CAP_INTERNAL_ERROR_DATA:
2035                 return 1;
2036 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2037         case KVM_CAP_IRQ_ROUTING:
2038                 return KVM_MAX_IRQ_ROUTES;
2039 #endif
2040         default:
2041                 break;
2042         }
2043         return kvm_dev_ioctl_check_extension(arg);
2044 }
2045
2046 static long kvm_dev_ioctl(struct file *filp,
2047                           unsigned int ioctl, unsigned long arg)
2048 {
2049         long r = -EINVAL;
2050
2051         switch (ioctl) {
2052         case KVM_GET_API_VERSION:
2053                 r = -EINVAL;
2054                 if (arg)
2055                         goto out;
2056                 r = KVM_API_VERSION;
2057                 break;
2058         case KVM_CREATE_VM:
2059                 r = -EINVAL;
2060                 if (arg)
2061                         goto out;
2062                 r = kvm_dev_ioctl_create_vm();
2063                 break;
2064         case KVM_CHECK_EXTENSION:
2065                 r = kvm_dev_ioctl_check_extension_generic(arg);
2066                 break;
2067         case KVM_GET_VCPU_MMAP_SIZE:
2068                 r = -EINVAL;
2069                 if (arg)
2070                         goto out;
2071                 r = PAGE_SIZE;     /* struct kvm_run */
2072 #ifdef CONFIG_X86
2073                 r += PAGE_SIZE;    /* pio data page */
2074 #endif
2075 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2076                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2077 #endif
2078                 break;
2079         case KVM_TRACE_ENABLE:
2080         case KVM_TRACE_PAUSE:
2081         case KVM_TRACE_DISABLE:
2082                 r = -EOPNOTSUPP;
2083                 break;
2084         default:
2085                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2086         }
2087 out:
2088         return r;
2089 }
2090
2091 static struct file_operations kvm_chardev_ops = {
2092         .unlocked_ioctl = kvm_dev_ioctl,
2093         .compat_ioctl   = kvm_dev_ioctl,
2094         .llseek         = noop_llseek,
2095 };
2096
2097 static struct miscdevice kvm_dev = {
2098         KVM_MINOR,
2099         "kvm",
2100         &kvm_chardev_ops,
2101 };
2102
2103 static void hardware_enable_nolock(void *junk)
2104 {
2105         int cpu = raw_smp_processor_id();
2106         int r;
2107
2108         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2109                 return;
2110
2111         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2112
2113         r = kvm_arch_hardware_enable(NULL);
2114
2115         if (r) {
2116                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2117                 atomic_inc(&hardware_enable_failed);
2118                 printk(KERN_INFO "kvm: enabling virtualization on "
2119                                  "CPU%d failed\n", cpu);
2120         }
2121 }
2122
2123 static void hardware_enable(void *junk)
2124 {
2125         spin_lock(&kvm_lock);
2126         hardware_enable_nolock(junk);
2127         spin_unlock(&kvm_lock);
2128 }
2129
2130 static void hardware_disable_nolock(void *junk)
2131 {
2132         int cpu = raw_smp_processor_id();
2133
2134         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2135                 return;
2136         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2137         kvm_arch_hardware_disable(NULL);
2138 }
2139
2140 static void hardware_disable(void *junk)
2141 {
2142         spin_lock(&kvm_lock);
2143         hardware_disable_nolock(junk);
2144         spin_unlock(&kvm_lock);
2145 }
2146
2147 static void hardware_disable_all_nolock(void)
2148 {
2149         BUG_ON(!kvm_usage_count);
2150
2151         kvm_usage_count--;
2152         if (!kvm_usage_count)
2153                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2154 }
2155
2156 static void hardware_disable_all(void)
2157 {
2158         spin_lock(&kvm_lock);
2159         hardware_disable_all_nolock();
2160         spin_unlock(&kvm_lock);
2161 }
2162
2163 static int hardware_enable_all(void)
2164 {
2165         int r = 0;
2166
2167         spin_lock(&kvm_lock);
2168
2169         kvm_usage_count++;
2170         if (kvm_usage_count == 1) {
2171                 atomic_set(&hardware_enable_failed, 0);
2172                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2173
2174                 if (atomic_read(&hardware_enable_failed)) {
2175                         hardware_disable_all_nolock();
2176                         r = -EBUSY;
2177                 }
2178         }
2179
2180         spin_unlock(&kvm_lock);
2181
2182         return r;
2183 }
2184
2185 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2186                            void *v)
2187 {
2188         int cpu = (long)v;
2189
2190         if (!kvm_usage_count)
2191                 return NOTIFY_OK;
2192
2193         val &= ~CPU_TASKS_FROZEN;
2194         switch (val) {
2195         case CPU_DYING:
2196                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2197                        cpu);
2198                 hardware_disable(NULL);
2199                 break;
2200         case CPU_STARTING:
2201                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2202                        cpu);
2203                 hardware_enable(NULL);
2204                 break;
2205         }
2206         return NOTIFY_OK;
2207 }
2208
2209
2210 asmlinkage void kvm_spurious_fault(void)
2211 {
2212         /* Fault while not rebooting.  We want the trace. */
2213         BUG();
2214 }
2215 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2216
2217 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2218                       void *v)
2219 {
2220         /*
2221          * Some (well, at least mine) BIOSes hang on reboot if
2222          * in vmx root mode.
2223          *
2224          * And Intel TXT required VMX off for all cpu when system shutdown.
2225          */
2226         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2227         kvm_rebooting = true;
2228         on_each_cpu(hardware_disable_nolock, NULL, 1);
2229         return NOTIFY_OK;
2230 }
2231
2232 static struct notifier_block kvm_reboot_notifier = {
2233         .notifier_call = kvm_reboot,
2234         .priority = 0,
2235 };
2236
2237 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2238 {
2239         int i;
2240
2241         for (i = 0; i < bus->dev_count; i++) {
2242                 struct kvm_io_device *pos = bus->devs[i];
2243
2244                 kvm_iodevice_destructor(pos);
2245         }
2246         kfree(bus);
2247 }
2248
2249 /* kvm_io_bus_write - called under kvm->slots_lock */
2250 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2251                      int len, const void *val)
2252 {
2253         int i;
2254         struct kvm_io_bus *bus;
2255
2256         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2257         for (i = 0; i < bus->dev_count; i++)
2258                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2259                         return 0;
2260         return -EOPNOTSUPP;
2261 }
2262
2263 /* kvm_io_bus_read - called under kvm->slots_lock */
2264 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2265                     int len, void *val)
2266 {
2267         int i;
2268         struct kvm_io_bus *bus;
2269
2270         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2271         for (i = 0; i < bus->dev_count; i++)
2272                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2273                         return 0;
2274         return -EOPNOTSUPP;
2275 }
2276
2277 /* Caller must hold slots_lock. */
2278 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2279                             struct kvm_io_device *dev)
2280 {
2281         struct kvm_io_bus *new_bus, *bus;
2282
2283         bus = kvm->buses[bus_idx];
2284         if (bus->dev_count > NR_IOBUS_DEVS-1)
2285                 return -ENOSPC;
2286
2287         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2288         if (!new_bus)
2289                 return -ENOMEM;
2290         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2291         new_bus->devs[new_bus->dev_count++] = dev;
2292         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2293         synchronize_srcu_expedited(&kvm->srcu);
2294         kfree(bus);
2295
2296         return 0;
2297 }
2298
2299 /* Caller must hold slots_lock. */
2300 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2301                               struct kvm_io_device *dev)
2302 {
2303         int i, r;
2304         struct kvm_io_bus *new_bus, *bus;
2305
2306         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2307         if (!new_bus)
2308                 return -ENOMEM;
2309
2310         bus = kvm->buses[bus_idx];
2311         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2312
2313         r = -ENOENT;
2314         for (i = 0; i < new_bus->dev_count; i++)
2315                 if (new_bus->devs[i] == dev) {
2316                         r = 0;
2317                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2318                         break;
2319                 }
2320
2321         if (r) {
2322                 kfree(new_bus);
2323                 return r;
2324         }
2325
2326         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2327         synchronize_srcu_expedited(&kvm->srcu);
2328         kfree(bus);
2329         return r;
2330 }
2331
2332 static struct notifier_block kvm_cpu_notifier = {
2333         .notifier_call = kvm_cpu_hotplug,
2334 };
2335
2336 static int vm_stat_get(void *_offset, u64 *val)
2337 {
2338         unsigned offset = (long)_offset;
2339         struct kvm *kvm;
2340
2341         *val = 0;
2342         spin_lock(&kvm_lock);
2343         list_for_each_entry(kvm, &vm_list, vm_list)
2344                 *val += *(u32 *)((void *)kvm + offset);
2345         spin_unlock(&kvm_lock);
2346         return 0;
2347 }
2348
2349 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2350
2351 static int vcpu_stat_get(void *_offset, u64 *val)
2352 {
2353         unsigned offset = (long)_offset;
2354         struct kvm *kvm;
2355         struct kvm_vcpu *vcpu;
2356         int i;
2357
2358         *val = 0;
2359         spin_lock(&kvm_lock);
2360         list_for_each_entry(kvm, &vm_list, vm_list)
2361                 kvm_for_each_vcpu(i, vcpu, kvm)
2362                         *val += *(u32 *)((void *)vcpu + offset);
2363
2364         spin_unlock(&kvm_lock);
2365         return 0;
2366 }
2367
2368 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2369
2370 static const struct file_operations *stat_fops[] = {
2371         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2372         [KVM_STAT_VM]   = &vm_stat_fops,
2373 };
2374
2375 static void kvm_init_debug(void)
2376 {
2377         struct kvm_stats_debugfs_item *p;
2378
2379         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2380         for (p = debugfs_entries; p->name; ++p)
2381                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2382                                                 (void *)(long)p->offset,
2383                                                 stat_fops[p->kind]);
2384 }
2385
2386 static void kvm_exit_debug(void)
2387 {
2388         struct kvm_stats_debugfs_item *p;
2389
2390         for (p = debugfs_entries; p->name; ++p)
2391                 debugfs_remove(p->dentry);
2392         debugfs_remove(kvm_debugfs_dir);
2393 }
2394
2395 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2396 {
2397         if (kvm_usage_count)
2398                 hardware_disable_nolock(NULL);
2399         return 0;
2400 }
2401
2402 static int kvm_resume(struct sys_device *dev)
2403 {
2404         if (kvm_usage_count) {
2405                 WARN_ON(spin_is_locked(&kvm_lock));
2406                 hardware_enable_nolock(NULL);
2407         }
2408         return 0;
2409 }
2410
2411 static struct sysdev_class kvm_sysdev_class = {
2412         .name = "kvm",
2413         .suspend = kvm_suspend,
2414         .resume = kvm_resume,
2415 };
2416
2417 static struct sys_device kvm_sysdev = {
2418         .id = 0,
2419         .cls = &kvm_sysdev_class,
2420 };
2421
2422 struct page *bad_page;
2423 pfn_t bad_pfn;
2424
2425 static inline
2426 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2427 {
2428         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2429 }
2430
2431 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2432 {
2433         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2434
2435         kvm_arch_vcpu_load(vcpu, cpu);
2436 }
2437
2438 static void kvm_sched_out(struct preempt_notifier *pn,
2439                           struct task_struct *next)
2440 {
2441         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2442
2443         kvm_arch_vcpu_put(vcpu);
2444 }
2445
2446 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2447                   struct module *module)
2448 {
2449         int r;
2450         int cpu;
2451
2452         r = kvm_arch_init(opaque);
2453         if (r)
2454                 goto out_fail;
2455
2456         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2457
2458         if (bad_page == NULL) {
2459                 r = -ENOMEM;
2460                 goto out;
2461         }
2462
2463         bad_pfn = page_to_pfn(bad_page);
2464
2465         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2466
2467         if (hwpoison_page == NULL) {
2468                 r = -ENOMEM;
2469                 goto out_free_0;
2470         }
2471
2472         hwpoison_pfn = page_to_pfn(hwpoison_page);
2473
2474         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2475
2476         if (fault_page == NULL) {
2477                 r = -ENOMEM;
2478                 goto out_free_0;
2479         }
2480
2481         fault_pfn = page_to_pfn(fault_page);
2482
2483         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2484                 r = -ENOMEM;
2485                 goto out_free_0;
2486         }
2487
2488         r = kvm_arch_hardware_setup();
2489         if (r < 0)
2490                 goto out_free_0a;
2491
2492         for_each_online_cpu(cpu) {
2493                 smp_call_function_single(cpu,
2494                                 kvm_arch_check_processor_compat,
2495                                 &r, 1);
2496                 if (r < 0)
2497                         goto out_free_1;
2498         }
2499
2500         r = register_cpu_notifier(&kvm_cpu_notifier);
2501         if (r)
2502                 goto out_free_2;
2503         register_reboot_notifier(&kvm_reboot_notifier);
2504
2505         r = sysdev_class_register(&kvm_sysdev_class);
2506         if (r)
2507                 goto out_free_3;
2508
2509         r = sysdev_register(&kvm_sysdev);
2510         if (r)
2511                 goto out_free_4;
2512
2513         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2514         if (!vcpu_align)
2515                 vcpu_align = __alignof__(struct kvm_vcpu);
2516         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2517                                            0, NULL);
2518         if (!kvm_vcpu_cache) {
2519                 r = -ENOMEM;
2520                 goto out_free_5;
2521         }
2522
2523         r = kvm_async_pf_init();
2524         if (r)
2525                 goto out_free;
2526
2527         kvm_chardev_ops.owner = module;
2528         kvm_vm_fops.owner = module;
2529         kvm_vcpu_fops.owner = module;
2530
2531         r = misc_register(&kvm_dev);
2532         if (r) {
2533                 printk(KERN_ERR "kvm: misc device register failed\n");
2534                 goto out_unreg;
2535         }
2536
2537         kvm_preempt_ops.sched_in = kvm_sched_in;
2538         kvm_preempt_ops.sched_out = kvm_sched_out;
2539
2540         kvm_init_debug();
2541
2542         return 0;
2543
2544 out_unreg:
2545         kvm_async_pf_deinit();
2546 out_free:
2547         kmem_cache_destroy(kvm_vcpu_cache);
2548 out_free_5:
2549         sysdev_unregister(&kvm_sysdev);
2550 out_free_4:
2551         sysdev_class_unregister(&kvm_sysdev_class);
2552 out_free_3:
2553         unregister_reboot_notifier(&kvm_reboot_notifier);
2554         unregister_cpu_notifier(&kvm_cpu_notifier);
2555 out_free_2:
2556 out_free_1:
2557         kvm_arch_hardware_unsetup();
2558 out_free_0a:
2559         free_cpumask_var(cpus_hardware_enabled);
2560 out_free_0:
2561         if (fault_page)
2562                 __free_page(fault_page);
2563         if (hwpoison_page)
2564                 __free_page(hwpoison_page);
2565         __free_page(bad_page);
2566 out:
2567         kvm_arch_exit();
2568 out_fail:
2569         return r;
2570 }
2571 EXPORT_SYMBOL_GPL(kvm_init);
2572
2573 void kvm_exit(void)
2574 {
2575         kvm_exit_debug();
2576         misc_deregister(&kvm_dev);
2577         kmem_cache_destroy(kvm_vcpu_cache);
2578         kvm_async_pf_deinit();
2579         sysdev_unregister(&kvm_sysdev);
2580         sysdev_class_unregister(&kvm_sysdev_class);
2581         unregister_reboot_notifier(&kvm_reboot_notifier);
2582         unregister_cpu_notifier(&kvm_cpu_notifier);
2583         on_each_cpu(hardware_disable_nolock, NULL, 1);
2584         kvm_arch_hardware_unsetup();
2585         kvm_arch_exit();
2586         free_cpumask_var(cpus_hardware_enabled);
2587         __free_page(hwpoison_page);
2588         __free_page(bad_page);
2589 }
2590 EXPORT_SYMBOL_GPL(kvm_exit);