bd31096e3698bf79ade9f3d270f503f7a3baec01
[linux-3.10.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/syscore_ops.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 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 bool kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 int vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         if (mutex_lock_killable(&vcpu->mutex))
139                 return -EINTR;
140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141                 /* The thread running this VCPU changed. */
142                 struct pid *oldpid = vcpu->pid;
143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144                 rcu_assign_pointer(vcpu->pid, newpid);
145                 synchronize_rcu();
146                 put_pid(oldpid);
147         }
148         cpu = get_cpu();
149         preempt_notifier_register(&vcpu->preempt_notifier);
150         kvm_arch_vcpu_load(vcpu, cpu);
151         put_cpu();
152         return 0;
153 }
154
155 void vcpu_put(struct kvm_vcpu *vcpu)
156 {
157         preempt_disable();
158         kvm_arch_vcpu_put(vcpu);
159         preempt_notifier_unregister(&vcpu->preempt_notifier);
160         preempt_enable();
161         mutex_unlock(&vcpu->mutex);
162 }
163
164 static void ack_flush(void *_completed)
165 {
166 }
167
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
169 {
170         int i, cpu, me;
171         cpumask_var_t cpus;
172         bool called = true;
173         struct kvm_vcpu *vcpu;
174
175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176
177         me = get_cpu();
178         kvm_for_each_vcpu(i, vcpu, kvm) {
179                 kvm_make_request(req, vcpu);
180                 cpu = vcpu->cpu;
181
182                 /* Set ->requests bit before we read ->mode */
183                 smp_mb();
184
185                 if (cpus != NULL && cpu != -1 && cpu != me &&
186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187                         cpumask_set_cpu(cpu, cpus);
188         }
189         if (unlikely(cpus == NULL))
190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191         else if (!cpumask_empty(cpus))
192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
193         else
194                 called = false;
195         put_cpu();
196         free_cpumask_var(cpus);
197         return called;
198 }
199
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
201 {
202         long dirty_count = kvm->tlbs_dirty;
203
204         smp_mb();
205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206                 ++kvm->stat.remote_tlb_flush;
207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 }
209
210 void kvm_reload_remote_mmus(struct kvm *kvm)
211 {
212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 }
214
215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216 {
217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218 }
219
220 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
221 {
222         struct page *page;
223         int r;
224
225         mutex_init(&vcpu->mutex);
226         vcpu->cpu = -1;
227         vcpu->kvm = kvm;
228         vcpu->vcpu_id = id;
229         vcpu->pid = NULL;
230         init_waitqueue_head(&vcpu->wq);
231         kvm_async_pf_vcpu_init(vcpu);
232
233         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
234         if (!page) {
235                 r = -ENOMEM;
236                 goto fail;
237         }
238         vcpu->run = page_address(page);
239
240         kvm_vcpu_set_in_spin_loop(vcpu, false);
241         kvm_vcpu_set_dy_eligible(vcpu, false);
242
243         r = kvm_arch_vcpu_init(vcpu);
244         if (r < 0)
245                 goto fail_free_run;
246         return 0;
247
248 fail_free_run:
249         free_page((unsigned long)vcpu->run);
250 fail:
251         return r;
252 }
253 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
254
255 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
256 {
257         put_pid(vcpu->pid);
258         kvm_arch_vcpu_uninit(vcpu);
259         free_page((unsigned long)vcpu->run);
260 }
261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
262
263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
265 {
266         return container_of(mn, struct kvm, mmu_notifier);
267 }
268
269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
270                                              struct mm_struct *mm,
271                                              unsigned long address)
272 {
273         struct kvm *kvm = mmu_notifier_to_kvm(mn);
274         int need_tlb_flush, idx;
275
276         /*
277          * When ->invalidate_page runs, the linux pte has been zapped
278          * already but the page is still allocated until
279          * ->invalidate_page returns. So if we increase the sequence
280          * here the kvm page fault will notice if the spte can't be
281          * established because the page is going to be freed. If
282          * instead the kvm page fault establishes the spte before
283          * ->invalidate_page runs, kvm_unmap_hva will release it
284          * before returning.
285          *
286          * The sequence increase only need to be seen at spin_unlock
287          * time, and not at spin_lock time.
288          *
289          * Increasing the sequence after the spin_unlock would be
290          * unsafe because the kvm page fault could then establish the
291          * pte after kvm_unmap_hva returned, without noticing the page
292          * is going to be freed.
293          */
294         idx = srcu_read_lock(&kvm->srcu);
295         spin_lock(&kvm->mmu_lock);
296
297         kvm->mmu_notifier_seq++;
298         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
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         spin_unlock(&kvm->mmu_lock);
304         srcu_read_unlock(&kvm->srcu, idx);
305 }
306
307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
308                                         struct mm_struct *mm,
309                                         unsigned long address,
310                                         pte_t pte)
311 {
312         struct kvm *kvm = mmu_notifier_to_kvm(mn);
313         int idx;
314
315         idx = srcu_read_lock(&kvm->srcu);
316         spin_lock(&kvm->mmu_lock);
317         kvm->mmu_notifier_seq++;
318         kvm_set_spte_hva(kvm, address, pte);
319         spin_unlock(&kvm->mmu_lock);
320         srcu_read_unlock(&kvm->srcu, idx);
321 }
322
323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
324                                                     struct mm_struct *mm,
325                                                     unsigned long start,
326                                                     unsigned long end)
327 {
328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
329         int need_tlb_flush = 0, idx;
330
331         idx = srcu_read_lock(&kvm->srcu);
332         spin_lock(&kvm->mmu_lock);
333         /*
334          * The count increase must become visible at unlock time as no
335          * spte can be established without taking the mmu_lock and
336          * count is also read inside the mmu_lock critical section.
337          */
338         kvm->mmu_notifier_count++;
339         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
340         need_tlb_flush |= kvm->tlbs_dirty;
341         /* we've to flush the tlb before the pages can be freed */
342         if (need_tlb_flush)
343                 kvm_flush_remote_tlbs(kvm);
344
345         spin_unlock(&kvm->mmu_lock);
346         srcu_read_unlock(&kvm->srcu, idx);
347 }
348
349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
350                                                   struct mm_struct *mm,
351                                                   unsigned long start,
352                                                   unsigned long end)
353 {
354         struct kvm *kvm = mmu_notifier_to_kvm(mn);
355
356         spin_lock(&kvm->mmu_lock);
357         /*
358          * This sequence increase will notify the kvm page fault that
359          * the page that is going to be mapped in the spte could have
360          * been freed.
361          */
362         kvm->mmu_notifier_seq++;
363         smp_wmb();
364         /*
365          * The above sequence increase must be visible before the
366          * below count decrease, which is ensured by the smp_wmb above
367          * in conjunction with the smp_rmb in mmu_notifier_retry().
368          */
369         kvm->mmu_notifier_count--;
370         spin_unlock(&kvm->mmu_lock);
371
372         BUG_ON(kvm->mmu_notifier_count < 0);
373 }
374
375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
376                                               struct mm_struct *mm,
377                                               unsigned long address)
378 {
379         struct kvm *kvm = mmu_notifier_to_kvm(mn);
380         int young, idx;
381
382         idx = srcu_read_lock(&kvm->srcu);
383         spin_lock(&kvm->mmu_lock);
384
385         young = kvm_age_hva(kvm, address);
386         if (young)
387                 kvm_flush_remote_tlbs(kvm);
388
389         spin_unlock(&kvm->mmu_lock);
390         srcu_read_unlock(&kvm->srcu, idx);
391
392         return young;
393 }
394
395 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
396                                        struct mm_struct *mm,
397                                        unsigned long address)
398 {
399         struct kvm *kvm = mmu_notifier_to_kvm(mn);
400         int young, idx;
401
402         idx = srcu_read_lock(&kvm->srcu);
403         spin_lock(&kvm->mmu_lock);
404         young = kvm_test_age_hva(kvm, address);
405         spin_unlock(&kvm->mmu_lock);
406         srcu_read_unlock(&kvm->srcu, idx);
407
408         return young;
409 }
410
411 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
412                                      struct mm_struct *mm)
413 {
414         struct kvm *kvm = mmu_notifier_to_kvm(mn);
415         int idx;
416
417         idx = srcu_read_lock(&kvm->srcu);
418         kvm_arch_flush_shadow_all(kvm);
419         srcu_read_unlock(&kvm->srcu, idx);
420 }
421
422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
423         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
424         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
425         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
426         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
427         .test_young             = kvm_mmu_notifier_test_young,
428         .change_pte             = kvm_mmu_notifier_change_pte,
429         .release                = kvm_mmu_notifier_release,
430 };
431
432 static int kvm_init_mmu_notifier(struct kvm *kvm)
433 {
434         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
435         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
436 }
437
438 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
439
440 static int kvm_init_mmu_notifier(struct kvm *kvm)
441 {
442         return 0;
443 }
444
445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
446
447 static void kvm_init_memslots_id(struct kvm *kvm)
448 {
449         int i;
450         struct kvm_memslots *slots = kvm->memslots;
451
452         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
453                 slots->id_to_index[i] = slots->memslots[i].id = i;
454 }
455
456 static struct kvm *kvm_create_vm(unsigned long type)
457 {
458         int r, i;
459         struct kvm *kvm = kvm_arch_alloc_vm();
460
461         if (!kvm)
462                 return ERR_PTR(-ENOMEM);
463
464         r = kvm_arch_init_vm(kvm, type);
465         if (r)
466                 goto out_err_nodisable;
467
468         r = hardware_enable_all();
469         if (r)
470                 goto out_err_nodisable;
471
472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
473         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
474         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
475 #endif
476
477         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
478
479         r = -ENOMEM;
480         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
481         if (!kvm->memslots)
482                 goto out_err_nosrcu;
483         kvm_init_memslots_id(kvm);
484         if (init_srcu_struct(&kvm->srcu))
485                 goto out_err_nosrcu;
486         for (i = 0; i < KVM_NR_BUSES; i++) {
487                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
488                                         GFP_KERNEL);
489                 if (!kvm->buses[i])
490                         goto out_err;
491         }
492
493         spin_lock_init(&kvm->mmu_lock);
494         kvm->mm = current->mm;
495         atomic_inc(&kvm->mm->mm_count);
496         kvm_eventfd_init(kvm);
497         mutex_init(&kvm->lock);
498         mutex_init(&kvm->irq_lock);
499         mutex_init(&kvm->slots_lock);
500         atomic_set(&kvm->users_count, 1);
501
502         r = kvm_init_mmu_notifier(kvm);
503         if (r)
504                 goto out_err;
505
506         raw_spin_lock(&kvm_lock);
507         list_add(&kvm->vm_list, &vm_list);
508         raw_spin_unlock(&kvm_lock);
509
510         return kvm;
511
512 out_err:
513         cleanup_srcu_struct(&kvm->srcu);
514 out_err_nosrcu:
515         hardware_disable_all();
516 out_err_nodisable:
517         for (i = 0; i < KVM_NR_BUSES; i++)
518                 kfree(kvm->buses[i]);
519         kfree(kvm->memslots);
520         kvm_arch_free_vm(kvm);
521         return ERR_PTR(r);
522 }
523
524 /*
525  * Avoid using vmalloc for a small buffer.
526  * Should not be used when the size is statically known.
527  */
528 void *kvm_kvzalloc(unsigned long size)
529 {
530         if (size > PAGE_SIZE)
531                 return vzalloc(size);
532         else
533                 return kzalloc(size, GFP_KERNEL);
534 }
535
536 void kvm_kvfree(const void *addr)
537 {
538         if (is_vmalloc_addr(addr))
539                 vfree(addr);
540         else
541                 kfree(addr);
542 }
543
544 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 {
546         if (!memslot->dirty_bitmap)
547                 return;
548
549         kvm_kvfree(memslot->dirty_bitmap);
550         memslot->dirty_bitmap = NULL;
551 }
552
553 /*
554  * Free any memory in @free but not in @dont.
555  */
556 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
557                                   struct kvm_memory_slot *dont)
558 {
559         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
560                 kvm_destroy_dirty_bitmap(free);
561
562         kvm_arch_free_memslot(free, dont);
563
564         free->npages = 0;
565 }
566
567 void kvm_free_physmem(struct kvm *kvm)
568 {
569         struct kvm_memslots *slots = kvm->memslots;
570         struct kvm_memory_slot *memslot;
571
572         kvm_for_each_memslot(memslot, slots)
573                 kvm_free_physmem_slot(memslot, NULL);
574
575         kfree(kvm->memslots);
576 }
577
578 static void kvm_destroy_vm(struct kvm *kvm)
579 {
580         int i;
581         struct mm_struct *mm = kvm->mm;
582
583         kvm_arch_sync_events(kvm);
584         raw_spin_lock(&kvm_lock);
585         list_del(&kvm->vm_list);
586         raw_spin_unlock(&kvm_lock);
587         kvm_free_irq_routing(kvm);
588         for (i = 0; i < KVM_NR_BUSES; i++)
589                 kvm_io_bus_destroy(kvm->buses[i]);
590         kvm_coalesced_mmio_free(kvm);
591 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
593 #else
594         kvm_arch_flush_shadow_all(kvm);
595 #endif
596         kvm_arch_destroy_vm(kvm);
597         kvm_free_physmem(kvm);
598         cleanup_srcu_struct(&kvm->srcu);
599         kvm_arch_free_vm(kvm);
600         hardware_disable_all();
601         mmdrop(mm);
602 }
603
604 void kvm_get_kvm(struct kvm *kvm)
605 {
606         atomic_inc(&kvm->users_count);
607 }
608 EXPORT_SYMBOL_GPL(kvm_get_kvm);
609
610 void kvm_put_kvm(struct kvm *kvm)
611 {
612         if (atomic_dec_and_test(&kvm->users_count))
613                 kvm_destroy_vm(kvm);
614 }
615 EXPORT_SYMBOL_GPL(kvm_put_kvm);
616
617
618 static int kvm_vm_release(struct inode *inode, struct file *filp)
619 {
620         struct kvm *kvm = filp->private_data;
621
622         kvm_irqfd_release(kvm);
623
624         kvm_put_kvm(kvm);
625         return 0;
626 }
627
628 /*
629  * Allocation size is twice as large as the actual dirty bitmap size.
630  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
631  */
632 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
633 {
634 #ifndef CONFIG_S390
635         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
636
637         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
638         if (!memslot->dirty_bitmap)
639                 return -ENOMEM;
640
641 #endif /* !CONFIG_S390 */
642         return 0;
643 }
644
645 static int cmp_memslot(const void *slot1, const void *slot2)
646 {
647         struct kvm_memory_slot *s1, *s2;
648
649         s1 = (struct kvm_memory_slot *)slot1;
650         s2 = (struct kvm_memory_slot *)slot2;
651
652         if (s1->npages < s2->npages)
653                 return 1;
654         if (s1->npages > s2->npages)
655                 return -1;
656
657         return 0;
658 }
659
660 /*
661  * Sort the memslots base on its size, so the larger slots
662  * will get better fit.
663  */
664 static void sort_memslots(struct kvm_memslots *slots)
665 {
666         int i;
667
668         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
669               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
670
671         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
672                 slots->id_to_index[slots->memslots[i].id] = i;
673 }
674
675 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
676 {
677         if (new) {
678                 int id = new->id;
679                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
680                 unsigned long npages = old->npages;
681
682                 *old = *new;
683                 if (new->npages != npages)
684                         sort_memslots(slots);
685         }
686
687         slots->generation++;
688 }
689
690 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
691 {
692         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
693
694 #ifdef KVM_CAP_READONLY_MEM
695         valid_flags |= KVM_MEM_READONLY;
696 #endif
697
698         if (mem->flags & ~valid_flags)
699                 return -EINVAL;
700
701         return 0;
702 }
703
704 /*
705  * Allocate some memory and give it an address in the guest physical address
706  * space.
707  *
708  * Discontiguous memory is allowed, mostly for framebuffers.
709  *
710  * Must be called holding mmap_sem for write.
711  */
712 int __kvm_set_memory_region(struct kvm *kvm,
713                             struct kvm_userspace_memory_region *mem,
714                             bool user_alloc)
715 {
716         int r;
717         gfn_t base_gfn;
718         unsigned long npages;
719         struct kvm_memory_slot *memslot, *slot;
720         struct kvm_memory_slot old, new;
721         struct kvm_memslots *slots = NULL, *old_memslots;
722
723         r = check_memory_region_flags(mem);
724         if (r)
725                 goto out;
726
727         r = -EINVAL;
728         /* General sanity checks */
729         if (mem->memory_size & (PAGE_SIZE - 1))
730                 goto out;
731         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
732                 goto out;
733         /* We can read the guest memory with __xxx_user() later on. */
734         if (user_alloc &&
735             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
736              !access_ok(VERIFY_WRITE,
737                         (void __user *)(unsigned long)mem->userspace_addr,
738                         mem->memory_size)))
739                 goto out;
740         if (mem->slot >= KVM_MEM_SLOTS_NUM)
741                 goto out;
742         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
743                 goto out;
744
745         memslot = id_to_memslot(kvm->memslots, mem->slot);
746         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
747         npages = mem->memory_size >> PAGE_SHIFT;
748
749         r = -EINVAL;
750         if (npages > KVM_MEM_MAX_NR_PAGES)
751                 goto out;
752
753         if (!npages)
754                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
755
756         new = old = *memslot;
757
758         new.id = mem->slot;
759         new.base_gfn = base_gfn;
760         new.npages = npages;
761         new.flags = mem->flags;
762
763         /*
764          * Disallow changing a memory slot's size or changing anything about
765          * zero sized slots that doesn't involve making them non-zero.
766          */
767         r = -EINVAL;
768         if (npages && old.npages && npages != old.npages)
769                 goto out_free;
770         if (!npages && !old.npages)
771                 goto out_free;
772
773         /* Check for overlaps */
774         r = -EEXIST;
775         kvm_for_each_memslot(slot, kvm->memslots) {
776                 if (slot->id >= KVM_USER_MEM_SLOTS || slot == memslot)
777                         continue;
778                 if (!((base_gfn + npages <= slot->base_gfn) ||
779                       (base_gfn >= slot->base_gfn + slot->npages)))
780                         goto out_free;
781         }
782
783         /* Free page dirty bitmap if unneeded */
784         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
785                 new.dirty_bitmap = NULL;
786
787         r = -ENOMEM;
788
789         /*
790          * Allocate if a slot is being created.  If modifying a slot,
791          * the userspace_addr cannot change.
792          */
793         if (!old.npages) {
794                 new.user_alloc = user_alloc;
795                 new.userspace_addr = mem->userspace_addr;
796
797                 if (kvm_arch_create_memslot(&new, npages))
798                         goto out_free;
799         } else if (npages && mem->userspace_addr != old.userspace_addr) {
800                 r = -EINVAL;
801                 goto out_free;
802         }
803
804         /* Allocate page dirty bitmap if needed */
805         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
806                 if (kvm_create_dirty_bitmap(&new) < 0)
807                         goto out_free;
808                 /* destroy any largepage mappings for dirty tracking */
809         }
810
811         if (!npages || base_gfn != old.base_gfn) {
812                 struct kvm_memory_slot *slot;
813
814                 r = -ENOMEM;
815                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
816                                 GFP_KERNEL);
817                 if (!slots)
818                         goto out_free;
819                 slot = id_to_memslot(slots, mem->slot);
820                 slot->flags |= KVM_MEMSLOT_INVALID;
821
822                 update_memslots(slots, NULL);
823
824                 old_memslots = kvm->memslots;
825                 rcu_assign_pointer(kvm->memslots, slots);
826                 synchronize_srcu_expedited(&kvm->srcu);
827                 /* slot was deleted or moved, clear iommu mapping */
828                 kvm_iommu_unmap_pages(kvm, &old);
829                 /* From this point no new shadow pages pointing to a deleted,
830                  * or moved, memslot will be created.
831                  *
832                  * validation of sp->gfn happens in:
833                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
834                  *      - kvm_is_visible_gfn (mmu_check_roots)
835                  */
836                 kvm_arch_flush_shadow_memslot(kvm, slot);
837                 slots = old_memslots;
838         }
839
840         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
841         if (r)
842                 goto out_slots;
843
844         r = -ENOMEM;
845         /*
846          * We can re-use the old_memslots from above, the only difference
847          * from the currently installed memslots is the invalid flag.  This
848          * will get overwritten by update_memslots anyway.
849          */
850         if (!slots) {
851                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
852                                 GFP_KERNEL);
853                 if (!slots)
854                         goto out_free;
855         }
856
857         /* map new memory slot into the iommu */
858         if (npages) {
859                 r = kvm_iommu_map_pages(kvm, &new);
860                 if (r)
861                         goto out_slots;
862         }
863
864         /* actual memory is freed via old in kvm_free_physmem_slot below */
865         if (!npages) {
866                 new.dirty_bitmap = NULL;
867                 memset(&new.arch, 0, sizeof(new.arch));
868         }
869
870         update_memslots(slots, &new);
871         old_memslots = kvm->memslots;
872         rcu_assign_pointer(kvm->memslots, slots);
873         synchronize_srcu_expedited(&kvm->srcu);
874
875         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
876
877         kvm_free_physmem_slot(&old, &new);
878         kfree(old_memslots);
879
880         return 0;
881
882 out_slots:
883         kfree(slots);
884 out_free:
885         kvm_free_physmem_slot(&new, &old);
886 out:
887         return r;
888
889 }
890 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
891
892 int kvm_set_memory_region(struct kvm *kvm,
893                           struct kvm_userspace_memory_region *mem,
894                           bool user_alloc)
895 {
896         int r;
897
898         mutex_lock(&kvm->slots_lock);
899         r = __kvm_set_memory_region(kvm, mem, user_alloc);
900         mutex_unlock(&kvm->slots_lock);
901         return r;
902 }
903 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
904
905 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
906                                    struct
907                                    kvm_userspace_memory_region *mem,
908                                    bool user_alloc)
909 {
910         if (mem->slot >= KVM_USER_MEM_SLOTS)
911                 return -EINVAL;
912         return kvm_set_memory_region(kvm, mem, user_alloc);
913 }
914
915 int kvm_get_dirty_log(struct kvm *kvm,
916                         struct kvm_dirty_log *log, int *is_dirty)
917 {
918         struct kvm_memory_slot *memslot;
919         int r, i;
920         unsigned long n;
921         unsigned long any = 0;
922
923         r = -EINVAL;
924         if (log->slot >= KVM_USER_MEM_SLOTS)
925                 goto out;
926
927         memslot = id_to_memslot(kvm->memslots, log->slot);
928         r = -ENOENT;
929         if (!memslot->dirty_bitmap)
930                 goto out;
931
932         n = kvm_dirty_bitmap_bytes(memslot);
933
934         for (i = 0; !any && i < n/sizeof(long); ++i)
935                 any = memslot->dirty_bitmap[i];
936
937         r = -EFAULT;
938         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
939                 goto out;
940
941         if (any)
942                 *is_dirty = 1;
943
944         r = 0;
945 out:
946         return r;
947 }
948
949 bool kvm_largepages_enabled(void)
950 {
951         return largepages_enabled;
952 }
953
954 void kvm_disable_largepages(void)
955 {
956         largepages_enabled = false;
957 }
958 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
959
960 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
961 {
962         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
963 }
964 EXPORT_SYMBOL_GPL(gfn_to_memslot);
965
966 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
967 {
968         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
969
970         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
971               memslot->flags & KVM_MEMSLOT_INVALID)
972                 return 0;
973
974         return 1;
975 }
976 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
977
978 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
979 {
980         struct vm_area_struct *vma;
981         unsigned long addr, size;
982
983         size = PAGE_SIZE;
984
985         addr = gfn_to_hva(kvm, gfn);
986         if (kvm_is_error_hva(addr))
987                 return PAGE_SIZE;
988
989         down_read(&current->mm->mmap_sem);
990         vma = find_vma(current->mm, addr);
991         if (!vma)
992                 goto out;
993
994         size = vma_kernel_pagesize(vma);
995
996 out:
997         up_read(&current->mm->mmap_sem);
998
999         return size;
1000 }
1001
1002 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1003 {
1004         return slot->flags & KVM_MEM_READONLY;
1005 }
1006
1007 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1008                                        gfn_t *nr_pages, bool write)
1009 {
1010         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1011                 return KVM_HVA_ERR_BAD;
1012
1013         if (memslot_is_readonly(slot) && write)
1014                 return KVM_HVA_ERR_RO_BAD;
1015
1016         if (nr_pages)
1017                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1018
1019         return __gfn_to_hva_memslot(slot, gfn);
1020 }
1021
1022 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1023                                      gfn_t *nr_pages)
1024 {
1025         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1026 }
1027
1028 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1029                                  gfn_t gfn)
1030 {
1031         return gfn_to_hva_many(slot, gfn, NULL);
1032 }
1033 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1034
1035 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1036 {
1037         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1038 }
1039 EXPORT_SYMBOL_GPL(gfn_to_hva);
1040
1041 /*
1042  * The hva returned by this function is only allowed to be read.
1043  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1044  */
1045 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1046 {
1047         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1048 }
1049
1050 static int kvm_read_hva(void *data, void __user *hva, int len)
1051 {
1052         return __copy_from_user(data, hva, len);
1053 }
1054
1055 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1056 {
1057         return __copy_from_user_inatomic(data, hva, len);
1058 }
1059
1060 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1061         unsigned long start, int write, struct page **page)
1062 {
1063         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1064
1065         if (write)
1066                 flags |= FOLL_WRITE;
1067
1068         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1069 }
1070
1071 static inline int check_user_page_hwpoison(unsigned long addr)
1072 {
1073         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1074
1075         rc = __get_user_pages(current, current->mm, addr, 1,
1076                               flags, NULL, NULL, NULL);
1077         return rc == -EHWPOISON;
1078 }
1079
1080 /*
1081  * The atomic path to get the writable pfn which will be stored in @pfn,
1082  * true indicates success, otherwise false is returned.
1083  */
1084 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1085                             bool write_fault, bool *writable, pfn_t *pfn)
1086 {
1087         struct page *page[1];
1088         int npages;
1089
1090         if (!(async || atomic))
1091                 return false;
1092
1093         /*
1094          * Fast pin a writable pfn only if it is a write fault request
1095          * or the caller allows to map a writable pfn for a read fault
1096          * request.
1097          */
1098         if (!(write_fault || writable))
1099                 return false;
1100
1101         npages = __get_user_pages_fast(addr, 1, 1, page);
1102         if (npages == 1) {
1103                 *pfn = page_to_pfn(page[0]);
1104
1105                 if (writable)
1106                         *writable = true;
1107                 return true;
1108         }
1109
1110         return false;
1111 }
1112
1113 /*
1114  * The slow path to get the pfn of the specified host virtual address,
1115  * 1 indicates success, -errno is returned if error is detected.
1116  */
1117 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1118                            bool *writable, pfn_t *pfn)
1119 {
1120         struct page *page[1];
1121         int npages = 0;
1122
1123         might_sleep();
1124
1125         if (writable)
1126                 *writable = write_fault;
1127
1128         if (async) {
1129                 down_read(&current->mm->mmap_sem);
1130                 npages = get_user_page_nowait(current, current->mm,
1131                                               addr, write_fault, page);
1132                 up_read(&current->mm->mmap_sem);
1133         } else
1134                 npages = get_user_pages_fast(addr, 1, write_fault,
1135                                              page);
1136         if (npages != 1)
1137                 return npages;
1138
1139         /* map read fault as writable if possible */
1140         if (unlikely(!write_fault) && writable) {
1141                 struct page *wpage[1];
1142
1143                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1144                 if (npages == 1) {
1145                         *writable = true;
1146                         put_page(page[0]);
1147                         page[0] = wpage[0];
1148                 }
1149
1150                 npages = 1;
1151         }
1152         *pfn = page_to_pfn(page[0]);
1153         return npages;
1154 }
1155
1156 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1157 {
1158         if (unlikely(!(vma->vm_flags & VM_READ)))
1159                 return false;
1160
1161         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1162                 return false;
1163
1164         return true;
1165 }
1166
1167 /*
1168  * Pin guest page in memory and return its pfn.
1169  * @addr: host virtual address which maps memory to the guest
1170  * @atomic: whether this function can sleep
1171  * @async: whether this function need to wait IO complete if the
1172  *         host page is not in the memory
1173  * @write_fault: whether we should get a writable host page
1174  * @writable: whether it allows to map a writable host page for !@write_fault
1175  *
1176  * The function will map a writable host page for these two cases:
1177  * 1): @write_fault = true
1178  * 2): @write_fault = false && @writable, @writable will tell the caller
1179  *     whether the mapping is writable.
1180  */
1181 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1182                         bool write_fault, bool *writable)
1183 {
1184         struct vm_area_struct *vma;
1185         pfn_t pfn = 0;
1186         int npages;
1187
1188         /* we can do it either atomically or asynchronously, not both */
1189         BUG_ON(atomic && async);
1190
1191         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1192                 return pfn;
1193
1194         if (atomic)
1195                 return KVM_PFN_ERR_FAULT;
1196
1197         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1198         if (npages == 1)
1199                 return pfn;
1200
1201         down_read(&current->mm->mmap_sem);
1202         if (npages == -EHWPOISON ||
1203               (!async && check_user_page_hwpoison(addr))) {
1204                 pfn = KVM_PFN_ERR_HWPOISON;
1205                 goto exit;
1206         }
1207
1208         vma = find_vma_intersection(current->mm, addr, addr + 1);
1209
1210         if (vma == NULL)
1211                 pfn = KVM_PFN_ERR_FAULT;
1212         else if ((vma->vm_flags & VM_PFNMAP)) {
1213                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1214                         vma->vm_pgoff;
1215                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1216         } else {
1217                 if (async && vma_is_valid(vma, write_fault))
1218                         *async = true;
1219                 pfn = KVM_PFN_ERR_FAULT;
1220         }
1221 exit:
1222         up_read(&current->mm->mmap_sem);
1223         return pfn;
1224 }
1225
1226 static pfn_t
1227 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1228                      bool *async, bool write_fault, bool *writable)
1229 {
1230         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1231
1232         if (addr == KVM_HVA_ERR_RO_BAD)
1233                 return KVM_PFN_ERR_RO_FAULT;
1234
1235         if (kvm_is_error_hva(addr))
1236                 return KVM_PFN_NOSLOT;
1237
1238         /* Do not map writable pfn in the readonly memslot. */
1239         if (writable && memslot_is_readonly(slot)) {
1240                 *writable = false;
1241                 writable = NULL;
1242         }
1243
1244         return hva_to_pfn(addr, atomic, async, write_fault,
1245                           writable);
1246 }
1247
1248 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1249                           bool write_fault, bool *writable)
1250 {
1251         struct kvm_memory_slot *slot;
1252
1253         if (async)
1254                 *async = false;
1255
1256         slot = gfn_to_memslot(kvm, gfn);
1257
1258         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1259                                     writable);
1260 }
1261
1262 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1263 {
1264         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1265 }
1266 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1267
1268 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1269                        bool write_fault, bool *writable)
1270 {
1271         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1272 }
1273 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1274
1275 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1276 {
1277         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1278 }
1279 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1280
1281 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1282                       bool *writable)
1283 {
1284         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1285 }
1286 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1287
1288 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1289 {
1290         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1291 }
1292
1293 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1294 {
1295         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1296 }
1297 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1298
1299 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1300                                                                   int nr_pages)
1301 {
1302         unsigned long addr;
1303         gfn_t entry;
1304
1305         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1306         if (kvm_is_error_hva(addr))
1307                 return -1;
1308
1309         if (entry < nr_pages)
1310                 return 0;
1311
1312         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1313 }
1314 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1315
1316 static struct page *kvm_pfn_to_page(pfn_t pfn)
1317 {
1318         if (is_error_noslot_pfn(pfn))
1319                 return KVM_ERR_PTR_BAD_PAGE;
1320
1321         if (kvm_is_mmio_pfn(pfn)) {
1322                 WARN_ON(1);
1323                 return KVM_ERR_PTR_BAD_PAGE;
1324         }
1325
1326         return pfn_to_page(pfn);
1327 }
1328
1329 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1330 {
1331         pfn_t pfn;
1332
1333         pfn = gfn_to_pfn(kvm, gfn);
1334
1335         return kvm_pfn_to_page(pfn);
1336 }
1337
1338 EXPORT_SYMBOL_GPL(gfn_to_page);
1339
1340 void kvm_release_page_clean(struct page *page)
1341 {
1342         WARN_ON(is_error_page(page));
1343
1344         kvm_release_pfn_clean(page_to_pfn(page));
1345 }
1346 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1347
1348 void kvm_release_pfn_clean(pfn_t pfn)
1349 {
1350         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1351                 put_page(pfn_to_page(pfn));
1352 }
1353 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1354
1355 void kvm_release_page_dirty(struct page *page)
1356 {
1357         WARN_ON(is_error_page(page));
1358
1359         kvm_release_pfn_dirty(page_to_pfn(page));
1360 }
1361 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1362
1363 void kvm_release_pfn_dirty(pfn_t pfn)
1364 {
1365         kvm_set_pfn_dirty(pfn);
1366         kvm_release_pfn_clean(pfn);
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1369
1370 void kvm_set_page_dirty(struct page *page)
1371 {
1372         kvm_set_pfn_dirty(page_to_pfn(page));
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1375
1376 void kvm_set_pfn_dirty(pfn_t pfn)
1377 {
1378         if (!kvm_is_mmio_pfn(pfn)) {
1379                 struct page *page = pfn_to_page(pfn);
1380                 if (!PageReserved(page))
1381                         SetPageDirty(page);
1382         }
1383 }
1384 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1385
1386 void kvm_set_pfn_accessed(pfn_t pfn)
1387 {
1388         if (!kvm_is_mmio_pfn(pfn))
1389                 mark_page_accessed(pfn_to_page(pfn));
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1392
1393 void kvm_get_pfn(pfn_t pfn)
1394 {
1395         if (!kvm_is_mmio_pfn(pfn))
1396                 get_page(pfn_to_page(pfn));
1397 }
1398 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1399
1400 static int next_segment(unsigned long len, int offset)
1401 {
1402         if (len > PAGE_SIZE - offset)
1403                 return PAGE_SIZE - offset;
1404         else
1405                 return len;
1406 }
1407
1408 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1409                         int len)
1410 {
1411         int r;
1412         unsigned long addr;
1413
1414         addr = gfn_to_hva_read(kvm, gfn);
1415         if (kvm_is_error_hva(addr))
1416                 return -EFAULT;
1417         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1418         if (r)
1419                 return -EFAULT;
1420         return 0;
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1423
1424 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1425 {
1426         gfn_t gfn = gpa >> PAGE_SHIFT;
1427         int seg;
1428         int offset = offset_in_page(gpa);
1429         int ret;
1430
1431         while ((seg = next_segment(len, offset)) != 0) {
1432                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1433                 if (ret < 0)
1434                         return ret;
1435                 offset = 0;
1436                 len -= seg;
1437                 data += seg;
1438                 ++gfn;
1439         }
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_read_guest);
1443
1444 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1445                           unsigned long len)
1446 {
1447         int r;
1448         unsigned long addr;
1449         gfn_t gfn = gpa >> PAGE_SHIFT;
1450         int offset = offset_in_page(gpa);
1451
1452         addr = gfn_to_hva_read(kvm, gfn);
1453         if (kvm_is_error_hva(addr))
1454                 return -EFAULT;
1455         pagefault_disable();
1456         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1457         pagefault_enable();
1458         if (r)
1459                 return -EFAULT;
1460         return 0;
1461 }
1462 EXPORT_SYMBOL(kvm_read_guest_atomic);
1463
1464 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1465                          int offset, int len)
1466 {
1467         int r;
1468         unsigned long addr;
1469
1470         addr = gfn_to_hva(kvm, gfn);
1471         if (kvm_is_error_hva(addr))
1472                 return -EFAULT;
1473         r = __copy_to_user((void __user *)addr + offset, data, len);
1474         if (r)
1475                 return -EFAULT;
1476         mark_page_dirty(kvm, gfn);
1477         return 0;
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1480
1481 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1482                     unsigned long len)
1483 {
1484         gfn_t gfn = gpa >> PAGE_SHIFT;
1485         int seg;
1486         int offset = offset_in_page(gpa);
1487         int ret;
1488
1489         while ((seg = next_segment(len, offset)) != 0) {
1490                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1491                 if (ret < 0)
1492                         return ret;
1493                 offset = 0;
1494                 len -= seg;
1495                 data += seg;
1496                 ++gfn;
1497         }
1498         return 0;
1499 }
1500
1501 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1502                               gpa_t gpa)
1503 {
1504         struct kvm_memslots *slots = kvm_memslots(kvm);
1505         int offset = offset_in_page(gpa);
1506         gfn_t gfn = gpa >> PAGE_SHIFT;
1507
1508         ghc->gpa = gpa;
1509         ghc->generation = slots->generation;
1510         ghc->memslot = gfn_to_memslot(kvm, gfn);
1511         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1512         if (!kvm_is_error_hva(ghc->hva))
1513                 ghc->hva += offset;
1514         else
1515                 return -EFAULT;
1516
1517         return 0;
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1520
1521 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1522                            void *data, unsigned long len)
1523 {
1524         struct kvm_memslots *slots = kvm_memslots(kvm);
1525         int r;
1526
1527         if (slots->generation != ghc->generation)
1528                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1529
1530         if (kvm_is_error_hva(ghc->hva))
1531                 return -EFAULT;
1532
1533         r = __copy_to_user((void __user *)ghc->hva, data, len);
1534         if (r)
1535                 return -EFAULT;
1536         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1537
1538         return 0;
1539 }
1540 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1541
1542 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1543                            void *data, unsigned long len)
1544 {
1545         struct kvm_memslots *slots = kvm_memslots(kvm);
1546         int r;
1547
1548         if (slots->generation != ghc->generation)
1549                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1550
1551         if (kvm_is_error_hva(ghc->hva))
1552                 return -EFAULT;
1553
1554         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1555         if (r)
1556                 return -EFAULT;
1557
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1561
1562 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1563 {
1564         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1565                                     offset, len);
1566 }
1567 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1568
1569 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1570 {
1571         gfn_t gfn = gpa >> PAGE_SHIFT;
1572         int seg;
1573         int offset = offset_in_page(gpa);
1574         int ret;
1575
1576         while ((seg = next_segment(len, offset)) != 0) {
1577                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1578                 if (ret < 0)
1579                         return ret;
1580                 offset = 0;
1581                 len -= seg;
1582                 ++gfn;
1583         }
1584         return 0;
1585 }
1586 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1587
1588 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1589                              gfn_t gfn)
1590 {
1591         if (memslot && memslot->dirty_bitmap) {
1592                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1593
1594                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1595         }
1596 }
1597
1598 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1599 {
1600         struct kvm_memory_slot *memslot;
1601
1602         memslot = gfn_to_memslot(kvm, gfn);
1603         mark_page_dirty_in_slot(kvm, memslot, gfn);
1604 }
1605
1606 /*
1607  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1608  */
1609 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1610 {
1611         DEFINE_WAIT(wait);
1612
1613         for (;;) {
1614                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1615
1616                 if (kvm_arch_vcpu_runnable(vcpu)) {
1617                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1618                         break;
1619                 }
1620                 if (kvm_cpu_has_pending_timer(vcpu))
1621                         break;
1622                 if (signal_pending(current))
1623                         break;
1624
1625                 schedule();
1626         }
1627
1628         finish_wait(&vcpu->wq, &wait);
1629 }
1630
1631 #ifndef CONFIG_S390
1632 /*
1633  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1634  */
1635 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1636 {
1637         int me;
1638         int cpu = vcpu->cpu;
1639         wait_queue_head_t *wqp;
1640
1641         wqp = kvm_arch_vcpu_wq(vcpu);
1642         if (waitqueue_active(wqp)) {
1643                 wake_up_interruptible(wqp);
1644                 ++vcpu->stat.halt_wakeup;
1645         }
1646
1647         me = get_cpu();
1648         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1649                 if (kvm_arch_vcpu_should_kick(vcpu))
1650                         smp_send_reschedule(cpu);
1651         put_cpu();
1652 }
1653 #endif /* !CONFIG_S390 */
1654
1655 void kvm_resched(struct kvm_vcpu *vcpu)
1656 {
1657         if (!need_resched())
1658                 return;
1659         cond_resched();
1660 }
1661 EXPORT_SYMBOL_GPL(kvm_resched);
1662
1663 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1664 {
1665         struct pid *pid;
1666         struct task_struct *task = NULL;
1667
1668         rcu_read_lock();
1669         pid = rcu_dereference(target->pid);
1670         if (pid)
1671                 task = get_pid_task(target->pid, PIDTYPE_PID);
1672         rcu_read_unlock();
1673         if (!task)
1674                 return false;
1675         if (task->flags & PF_VCPU) {
1676                 put_task_struct(task);
1677                 return false;
1678         }
1679         if (yield_to(task, 1)) {
1680                 put_task_struct(task);
1681                 return true;
1682         }
1683         put_task_struct(task);
1684         return false;
1685 }
1686 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1687
1688 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1689 /*
1690  * Helper that checks whether a VCPU is eligible for directed yield.
1691  * Most eligible candidate to yield is decided by following heuristics:
1692  *
1693  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1694  *  (preempted lock holder), indicated by @in_spin_loop.
1695  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1696  *
1697  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1698  *  chance last time (mostly it has become eligible now since we have probably
1699  *  yielded to lockholder in last iteration. This is done by toggling
1700  *  @dy_eligible each time a VCPU checked for eligibility.)
1701  *
1702  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1703  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1704  *  burning. Giving priority for a potential lock-holder increases lock
1705  *  progress.
1706  *
1707  *  Since algorithm is based on heuristics, accessing another VCPU data without
1708  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1709  *  and continue with next VCPU and so on.
1710  */
1711 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1712 {
1713         bool eligible;
1714
1715         eligible = !vcpu->spin_loop.in_spin_loop ||
1716                         (vcpu->spin_loop.in_spin_loop &&
1717                          vcpu->spin_loop.dy_eligible);
1718
1719         if (vcpu->spin_loop.in_spin_loop)
1720                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1721
1722         return eligible;
1723 }
1724 #endif
1725 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1726 {
1727         struct kvm *kvm = me->kvm;
1728         struct kvm_vcpu *vcpu;
1729         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1730         int yielded = 0;
1731         int pass;
1732         int i;
1733
1734         kvm_vcpu_set_in_spin_loop(me, true);
1735         /*
1736          * We boost the priority of a VCPU that is runnable but not
1737          * currently running, because it got preempted by something
1738          * else and called schedule in __vcpu_run.  Hopefully that
1739          * VCPU is holding the lock that we need and will release it.
1740          * We approximate round-robin by starting at the last boosted VCPU.
1741          */
1742         for (pass = 0; pass < 2 && !yielded; pass++) {
1743                 kvm_for_each_vcpu(i, vcpu, kvm) {
1744                         if (!pass && i <= last_boosted_vcpu) {
1745                                 i = last_boosted_vcpu;
1746                                 continue;
1747                         } else if (pass && i > last_boosted_vcpu)
1748                                 break;
1749                         if (vcpu == me)
1750                                 continue;
1751                         if (waitqueue_active(&vcpu->wq))
1752                                 continue;
1753                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1754                                 continue;
1755                         if (kvm_vcpu_yield_to(vcpu)) {
1756                                 kvm->last_boosted_vcpu = i;
1757                                 yielded = 1;
1758                                 break;
1759                         }
1760                 }
1761         }
1762         kvm_vcpu_set_in_spin_loop(me, false);
1763
1764         /* Ensure vcpu is not eligible during next spinloop */
1765         kvm_vcpu_set_dy_eligible(me, false);
1766 }
1767 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1768
1769 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1770 {
1771         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1772         struct page *page;
1773
1774         if (vmf->pgoff == 0)
1775                 page = virt_to_page(vcpu->run);
1776 #ifdef CONFIG_X86
1777         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1778                 page = virt_to_page(vcpu->arch.pio_data);
1779 #endif
1780 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1781         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1782                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1783 #endif
1784         else
1785                 return kvm_arch_vcpu_fault(vcpu, vmf);
1786         get_page(page);
1787         vmf->page = page;
1788         return 0;
1789 }
1790
1791 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1792         .fault = kvm_vcpu_fault,
1793 };
1794
1795 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1796 {
1797         vma->vm_ops = &kvm_vcpu_vm_ops;
1798         return 0;
1799 }
1800
1801 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1802 {
1803         struct kvm_vcpu *vcpu = filp->private_data;
1804
1805         kvm_put_kvm(vcpu->kvm);
1806         return 0;
1807 }
1808
1809 static struct file_operations kvm_vcpu_fops = {
1810         .release        = kvm_vcpu_release,
1811         .unlocked_ioctl = kvm_vcpu_ioctl,
1812 #ifdef CONFIG_COMPAT
1813         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1814 #endif
1815         .mmap           = kvm_vcpu_mmap,
1816         .llseek         = noop_llseek,
1817 };
1818
1819 /*
1820  * Allocates an inode for the vcpu.
1821  */
1822 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1823 {
1824         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1825 }
1826
1827 /*
1828  * Creates some virtual cpus.  Good luck creating more than one.
1829  */
1830 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1831 {
1832         int r;
1833         struct kvm_vcpu *vcpu, *v;
1834
1835         vcpu = kvm_arch_vcpu_create(kvm, id);
1836         if (IS_ERR(vcpu))
1837                 return PTR_ERR(vcpu);
1838
1839         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1840
1841         r = kvm_arch_vcpu_setup(vcpu);
1842         if (r)
1843                 goto vcpu_destroy;
1844
1845         mutex_lock(&kvm->lock);
1846         if (!kvm_vcpu_compatible(vcpu)) {
1847                 r = -EINVAL;
1848                 goto unlock_vcpu_destroy;
1849         }
1850         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1851                 r = -EINVAL;
1852                 goto unlock_vcpu_destroy;
1853         }
1854
1855         kvm_for_each_vcpu(r, v, kvm)
1856                 if (v->vcpu_id == id) {
1857                         r = -EEXIST;
1858                         goto unlock_vcpu_destroy;
1859                 }
1860
1861         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1862
1863         /* Now it's all set up, let userspace reach it */
1864         kvm_get_kvm(kvm);
1865         r = create_vcpu_fd(vcpu);
1866         if (r < 0) {
1867                 kvm_put_kvm(kvm);
1868                 goto unlock_vcpu_destroy;
1869         }
1870
1871         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1872         smp_wmb();
1873         atomic_inc(&kvm->online_vcpus);
1874
1875         mutex_unlock(&kvm->lock);
1876         kvm_arch_vcpu_postcreate(vcpu);
1877         return r;
1878
1879 unlock_vcpu_destroy:
1880         mutex_unlock(&kvm->lock);
1881 vcpu_destroy:
1882         kvm_arch_vcpu_destroy(vcpu);
1883         return r;
1884 }
1885
1886 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1887 {
1888         if (sigset) {
1889                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1890                 vcpu->sigset_active = 1;
1891                 vcpu->sigset = *sigset;
1892         } else
1893                 vcpu->sigset_active = 0;
1894         return 0;
1895 }
1896
1897 static long kvm_vcpu_ioctl(struct file *filp,
1898                            unsigned int ioctl, unsigned long arg)
1899 {
1900         struct kvm_vcpu *vcpu = filp->private_data;
1901         void __user *argp = (void __user *)arg;
1902         int r;
1903         struct kvm_fpu *fpu = NULL;
1904         struct kvm_sregs *kvm_sregs = NULL;
1905
1906         if (vcpu->kvm->mm != current->mm)
1907                 return -EIO;
1908
1909 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1910         /*
1911          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1912          * so vcpu_load() would break it.
1913          */
1914         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1915                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1916 #endif
1917
1918
1919         r = vcpu_load(vcpu);
1920         if (r)
1921                 return r;
1922         switch (ioctl) {
1923         case KVM_RUN:
1924                 r = -EINVAL;
1925                 if (arg)
1926                         goto out;
1927                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1928                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1929                 break;
1930         case KVM_GET_REGS: {
1931                 struct kvm_regs *kvm_regs;
1932
1933                 r = -ENOMEM;
1934                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1935                 if (!kvm_regs)
1936                         goto out;
1937                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1938                 if (r)
1939                         goto out_free1;
1940                 r = -EFAULT;
1941                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1942                         goto out_free1;
1943                 r = 0;
1944 out_free1:
1945                 kfree(kvm_regs);
1946                 break;
1947         }
1948         case KVM_SET_REGS: {
1949                 struct kvm_regs *kvm_regs;
1950
1951                 r = -ENOMEM;
1952                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1953                 if (IS_ERR(kvm_regs)) {
1954                         r = PTR_ERR(kvm_regs);
1955                         goto out;
1956                 }
1957                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1958                 kfree(kvm_regs);
1959                 break;
1960         }
1961         case KVM_GET_SREGS: {
1962                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1963                 r = -ENOMEM;
1964                 if (!kvm_sregs)
1965                         goto out;
1966                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1967                 if (r)
1968                         goto out;
1969                 r = -EFAULT;
1970                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1971                         goto out;
1972                 r = 0;
1973                 break;
1974         }
1975         case KVM_SET_SREGS: {
1976                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1977                 if (IS_ERR(kvm_sregs)) {
1978                         r = PTR_ERR(kvm_sregs);
1979                         kvm_sregs = NULL;
1980                         goto out;
1981                 }
1982                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1983                 break;
1984         }
1985         case KVM_GET_MP_STATE: {
1986                 struct kvm_mp_state mp_state;
1987
1988                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1989                 if (r)
1990                         goto out;
1991                 r = -EFAULT;
1992                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1993                         goto out;
1994                 r = 0;
1995                 break;
1996         }
1997         case KVM_SET_MP_STATE: {
1998                 struct kvm_mp_state mp_state;
1999
2000                 r = -EFAULT;
2001                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2002                         goto out;
2003                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2004                 break;
2005         }
2006         case KVM_TRANSLATE: {
2007                 struct kvm_translation tr;
2008
2009                 r = -EFAULT;
2010                 if (copy_from_user(&tr, argp, sizeof tr))
2011                         goto out;
2012                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2013                 if (r)
2014                         goto out;
2015                 r = -EFAULT;
2016                 if (copy_to_user(argp, &tr, sizeof tr))
2017                         goto out;
2018                 r = 0;
2019                 break;
2020         }
2021         case KVM_SET_GUEST_DEBUG: {
2022                 struct kvm_guest_debug dbg;
2023
2024                 r = -EFAULT;
2025                 if (copy_from_user(&dbg, argp, sizeof dbg))
2026                         goto out;
2027                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2028                 break;
2029         }
2030         case KVM_SET_SIGNAL_MASK: {
2031                 struct kvm_signal_mask __user *sigmask_arg = argp;
2032                 struct kvm_signal_mask kvm_sigmask;
2033                 sigset_t sigset, *p;
2034
2035                 p = NULL;
2036                 if (argp) {
2037                         r = -EFAULT;
2038                         if (copy_from_user(&kvm_sigmask, argp,
2039                                            sizeof kvm_sigmask))
2040                                 goto out;
2041                         r = -EINVAL;
2042                         if (kvm_sigmask.len != sizeof sigset)
2043                                 goto out;
2044                         r = -EFAULT;
2045                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2046                                            sizeof sigset))
2047                                 goto out;
2048                         p = &sigset;
2049                 }
2050                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2051                 break;
2052         }
2053         case KVM_GET_FPU: {
2054                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2055                 r = -ENOMEM;
2056                 if (!fpu)
2057                         goto out;
2058                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2059                 if (r)
2060                         goto out;
2061                 r = -EFAULT;
2062                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2063                         goto out;
2064                 r = 0;
2065                 break;
2066         }
2067         case KVM_SET_FPU: {
2068                 fpu = memdup_user(argp, sizeof(*fpu));
2069                 if (IS_ERR(fpu)) {
2070                         r = PTR_ERR(fpu);
2071                         fpu = NULL;
2072                         goto out;
2073                 }
2074                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2075                 break;
2076         }
2077         default:
2078                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2079         }
2080 out:
2081         vcpu_put(vcpu);
2082         kfree(fpu);
2083         kfree(kvm_sregs);
2084         return r;
2085 }
2086
2087 #ifdef CONFIG_COMPAT
2088 static long kvm_vcpu_compat_ioctl(struct file *filp,
2089                                   unsigned int ioctl, unsigned long arg)
2090 {
2091         struct kvm_vcpu *vcpu = filp->private_data;
2092         void __user *argp = compat_ptr(arg);
2093         int r;
2094
2095         if (vcpu->kvm->mm != current->mm)
2096                 return -EIO;
2097
2098         switch (ioctl) {
2099         case KVM_SET_SIGNAL_MASK: {
2100                 struct kvm_signal_mask __user *sigmask_arg = argp;
2101                 struct kvm_signal_mask kvm_sigmask;
2102                 compat_sigset_t csigset;
2103                 sigset_t sigset;
2104
2105                 if (argp) {
2106                         r = -EFAULT;
2107                         if (copy_from_user(&kvm_sigmask, argp,
2108                                            sizeof kvm_sigmask))
2109                                 goto out;
2110                         r = -EINVAL;
2111                         if (kvm_sigmask.len != sizeof csigset)
2112                                 goto out;
2113                         r = -EFAULT;
2114                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2115                                            sizeof csigset))
2116                                 goto out;
2117                         sigset_from_compat(&sigset, &csigset);
2118                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2119                 } else
2120                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2121                 break;
2122         }
2123         default:
2124                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2125         }
2126
2127 out:
2128         return r;
2129 }
2130 #endif
2131
2132 static long kvm_vm_ioctl(struct file *filp,
2133                            unsigned int ioctl, unsigned long arg)
2134 {
2135         struct kvm *kvm = filp->private_data;
2136         void __user *argp = (void __user *)arg;
2137         int r;
2138
2139         if (kvm->mm != current->mm)
2140                 return -EIO;
2141         switch (ioctl) {
2142         case KVM_CREATE_VCPU:
2143                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2144                 break;
2145         case KVM_SET_USER_MEMORY_REGION: {
2146                 struct kvm_userspace_memory_region kvm_userspace_mem;
2147
2148                 r = -EFAULT;
2149                 if (copy_from_user(&kvm_userspace_mem, argp,
2150                                                 sizeof kvm_userspace_mem))
2151                         goto out;
2152
2153                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2154                 break;
2155         }
2156         case KVM_GET_DIRTY_LOG: {
2157                 struct kvm_dirty_log log;
2158
2159                 r = -EFAULT;
2160                 if (copy_from_user(&log, argp, sizeof log))
2161                         goto out;
2162                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2163                 break;
2164         }
2165 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2166         case KVM_REGISTER_COALESCED_MMIO: {
2167                 struct kvm_coalesced_mmio_zone zone;
2168                 r = -EFAULT;
2169                 if (copy_from_user(&zone, argp, sizeof zone))
2170                         goto out;
2171                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2172                 break;
2173         }
2174         case KVM_UNREGISTER_COALESCED_MMIO: {
2175                 struct kvm_coalesced_mmio_zone zone;
2176                 r = -EFAULT;
2177                 if (copy_from_user(&zone, argp, sizeof zone))
2178                         goto out;
2179                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2180                 break;
2181         }
2182 #endif
2183         case KVM_IRQFD: {
2184                 struct kvm_irqfd data;
2185
2186                 r = -EFAULT;
2187                 if (copy_from_user(&data, argp, sizeof data))
2188                         goto out;
2189                 r = kvm_irqfd(kvm, &data);
2190                 break;
2191         }
2192         case KVM_IOEVENTFD: {
2193                 struct kvm_ioeventfd data;
2194
2195                 r = -EFAULT;
2196                 if (copy_from_user(&data, argp, sizeof data))
2197                         goto out;
2198                 r = kvm_ioeventfd(kvm, &data);
2199                 break;
2200         }
2201 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2202         case KVM_SET_BOOT_CPU_ID:
2203                 r = 0;
2204                 mutex_lock(&kvm->lock);
2205                 if (atomic_read(&kvm->online_vcpus) != 0)
2206                         r = -EBUSY;
2207                 else
2208                         kvm->bsp_vcpu_id = arg;
2209                 mutex_unlock(&kvm->lock);
2210                 break;
2211 #endif
2212 #ifdef CONFIG_HAVE_KVM_MSI
2213         case KVM_SIGNAL_MSI: {
2214                 struct kvm_msi msi;
2215
2216                 r = -EFAULT;
2217                 if (copy_from_user(&msi, argp, sizeof msi))
2218                         goto out;
2219                 r = kvm_send_userspace_msi(kvm, &msi);
2220                 break;
2221         }
2222 #endif
2223 #ifdef __KVM_HAVE_IRQ_LINE
2224         case KVM_IRQ_LINE_STATUS:
2225         case KVM_IRQ_LINE: {
2226                 struct kvm_irq_level irq_event;
2227
2228                 r = -EFAULT;
2229                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2230                         goto out;
2231
2232                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2233                 if (r)
2234                         goto out;
2235
2236                 r = -EFAULT;
2237                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2238                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2239                                 goto out;
2240                 }
2241
2242                 r = 0;
2243                 break;
2244         }
2245 #endif
2246         default:
2247                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2248                 if (r == -ENOTTY)
2249                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2250         }
2251 out:
2252         return r;
2253 }
2254
2255 #ifdef CONFIG_COMPAT
2256 struct compat_kvm_dirty_log {
2257         __u32 slot;
2258         __u32 padding1;
2259         union {
2260                 compat_uptr_t dirty_bitmap; /* one bit per page */
2261                 __u64 padding2;
2262         };
2263 };
2264
2265 static long kvm_vm_compat_ioctl(struct file *filp,
2266                            unsigned int ioctl, unsigned long arg)
2267 {
2268         struct kvm *kvm = filp->private_data;
2269         int r;
2270
2271         if (kvm->mm != current->mm)
2272                 return -EIO;
2273         switch (ioctl) {
2274         case KVM_GET_DIRTY_LOG: {
2275                 struct compat_kvm_dirty_log compat_log;
2276                 struct kvm_dirty_log log;
2277
2278                 r = -EFAULT;
2279                 if (copy_from_user(&compat_log, (void __user *)arg,
2280                                    sizeof(compat_log)))
2281                         goto out;
2282                 log.slot         = compat_log.slot;
2283                 log.padding1     = compat_log.padding1;
2284                 log.padding2     = compat_log.padding2;
2285                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2286
2287                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2288                 break;
2289         }
2290         default:
2291                 r = kvm_vm_ioctl(filp, ioctl, arg);
2292         }
2293
2294 out:
2295         return r;
2296 }
2297 #endif
2298
2299 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2300 {
2301         struct page *page[1];
2302         unsigned long addr;
2303         int npages;
2304         gfn_t gfn = vmf->pgoff;
2305         struct kvm *kvm = vma->vm_file->private_data;
2306
2307         addr = gfn_to_hva(kvm, gfn);
2308         if (kvm_is_error_hva(addr))
2309                 return VM_FAULT_SIGBUS;
2310
2311         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2312                                 NULL);
2313         if (unlikely(npages != 1))
2314                 return VM_FAULT_SIGBUS;
2315
2316         vmf->page = page[0];
2317         return 0;
2318 }
2319
2320 static const struct vm_operations_struct kvm_vm_vm_ops = {
2321         .fault = kvm_vm_fault,
2322 };
2323
2324 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2325 {
2326         vma->vm_ops = &kvm_vm_vm_ops;
2327         return 0;
2328 }
2329
2330 static struct file_operations kvm_vm_fops = {
2331         .release        = kvm_vm_release,
2332         .unlocked_ioctl = kvm_vm_ioctl,
2333 #ifdef CONFIG_COMPAT
2334         .compat_ioctl   = kvm_vm_compat_ioctl,
2335 #endif
2336         .mmap           = kvm_vm_mmap,
2337         .llseek         = noop_llseek,
2338 };
2339
2340 static int kvm_dev_ioctl_create_vm(unsigned long type)
2341 {
2342         int r;
2343         struct kvm *kvm;
2344
2345         kvm = kvm_create_vm(type);
2346         if (IS_ERR(kvm))
2347                 return PTR_ERR(kvm);
2348 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2349         r = kvm_coalesced_mmio_init(kvm);
2350         if (r < 0) {
2351                 kvm_put_kvm(kvm);
2352                 return r;
2353         }
2354 #endif
2355         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2356         if (r < 0)
2357                 kvm_put_kvm(kvm);
2358
2359         return r;
2360 }
2361
2362 static long kvm_dev_ioctl_check_extension_generic(long arg)
2363 {
2364         switch (arg) {
2365         case KVM_CAP_USER_MEMORY:
2366         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2367         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2368 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2369         case KVM_CAP_SET_BOOT_CPU_ID:
2370 #endif
2371         case KVM_CAP_INTERNAL_ERROR_DATA:
2372 #ifdef CONFIG_HAVE_KVM_MSI
2373         case KVM_CAP_SIGNAL_MSI:
2374 #endif
2375                 return 1;
2376 #ifdef KVM_CAP_IRQ_ROUTING
2377         case KVM_CAP_IRQ_ROUTING:
2378                 return KVM_MAX_IRQ_ROUTES;
2379 #endif
2380         default:
2381                 break;
2382         }
2383         return kvm_dev_ioctl_check_extension(arg);
2384 }
2385
2386 static long kvm_dev_ioctl(struct file *filp,
2387                           unsigned int ioctl, unsigned long arg)
2388 {
2389         long r = -EINVAL;
2390
2391         switch (ioctl) {
2392         case KVM_GET_API_VERSION:
2393                 r = -EINVAL;
2394                 if (arg)
2395                         goto out;
2396                 r = KVM_API_VERSION;
2397                 break;
2398         case KVM_CREATE_VM:
2399                 r = kvm_dev_ioctl_create_vm(arg);
2400                 break;
2401         case KVM_CHECK_EXTENSION:
2402                 r = kvm_dev_ioctl_check_extension_generic(arg);
2403                 break;
2404         case KVM_GET_VCPU_MMAP_SIZE:
2405                 r = -EINVAL;
2406                 if (arg)
2407                         goto out;
2408                 r = PAGE_SIZE;     /* struct kvm_run */
2409 #ifdef CONFIG_X86
2410                 r += PAGE_SIZE;    /* pio data page */
2411 #endif
2412 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2413                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2414 #endif
2415                 break;
2416         case KVM_TRACE_ENABLE:
2417         case KVM_TRACE_PAUSE:
2418         case KVM_TRACE_DISABLE:
2419                 r = -EOPNOTSUPP;
2420                 break;
2421         default:
2422                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2423         }
2424 out:
2425         return r;
2426 }
2427
2428 static struct file_operations kvm_chardev_ops = {
2429         .unlocked_ioctl = kvm_dev_ioctl,
2430         .compat_ioctl   = kvm_dev_ioctl,
2431         .llseek         = noop_llseek,
2432 };
2433
2434 static struct miscdevice kvm_dev = {
2435         KVM_MINOR,
2436         "kvm",
2437         &kvm_chardev_ops,
2438 };
2439
2440 static void hardware_enable_nolock(void *junk)
2441 {
2442         int cpu = raw_smp_processor_id();
2443         int r;
2444
2445         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2446                 return;
2447
2448         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2449
2450         r = kvm_arch_hardware_enable(NULL);
2451
2452         if (r) {
2453                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2454                 atomic_inc(&hardware_enable_failed);
2455                 printk(KERN_INFO "kvm: enabling virtualization on "
2456                                  "CPU%d failed\n", cpu);
2457         }
2458 }
2459
2460 static void hardware_enable(void *junk)
2461 {
2462         raw_spin_lock(&kvm_lock);
2463         hardware_enable_nolock(junk);
2464         raw_spin_unlock(&kvm_lock);
2465 }
2466
2467 static void hardware_disable_nolock(void *junk)
2468 {
2469         int cpu = raw_smp_processor_id();
2470
2471         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2472                 return;
2473         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2474         kvm_arch_hardware_disable(NULL);
2475 }
2476
2477 static void hardware_disable(void *junk)
2478 {
2479         raw_spin_lock(&kvm_lock);
2480         hardware_disable_nolock(junk);
2481         raw_spin_unlock(&kvm_lock);
2482 }
2483
2484 static void hardware_disable_all_nolock(void)
2485 {
2486         BUG_ON(!kvm_usage_count);
2487
2488         kvm_usage_count--;
2489         if (!kvm_usage_count)
2490                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2491 }
2492
2493 static void hardware_disable_all(void)
2494 {
2495         raw_spin_lock(&kvm_lock);
2496         hardware_disable_all_nolock();
2497         raw_spin_unlock(&kvm_lock);
2498 }
2499
2500 static int hardware_enable_all(void)
2501 {
2502         int r = 0;
2503
2504         raw_spin_lock(&kvm_lock);
2505
2506         kvm_usage_count++;
2507         if (kvm_usage_count == 1) {
2508                 atomic_set(&hardware_enable_failed, 0);
2509                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2510
2511                 if (atomic_read(&hardware_enable_failed)) {
2512                         hardware_disable_all_nolock();
2513                         r = -EBUSY;
2514                 }
2515         }
2516
2517         raw_spin_unlock(&kvm_lock);
2518
2519         return r;
2520 }
2521
2522 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2523                            void *v)
2524 {
2525         int cpu = (long)v;
2526
2527         if (!kvm_usage_count)
2528                 return NOTIFY_OK;
2529
2530         val &= ~CPU_TASKS_FROZEN;
2531         switch (val) {
2532         case CPU_DYING:
2533                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2534                        cpu);
2535                 hardware_disable(NULL);
2536                 break;
2537         case CPU_STARTING:
2538                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2539                        cpu);
2540                 hardware_enable(NULL);
2541                 break;
2542         }
2543         return NOTIFY_OK;
2544 }
2545
2546
2547 asmlinkage void kvm_spurious_fault(void)
2548 {
2549         /* Fault while not rebooting.  We want the trace. */
2550         BUG();
2551 }
2552 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2553
2554 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2555                       void *v)
2556 {
2557         /*
2558          * Some (well, at least mine) BIOSes hang on reboot if
2559          * in vmx root mode.
2560          *
2561          * And Intel TXT required VMX off for all cpu when system shutdown.
2562          */
2563         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2564         kvm_rebooting = true;
2565         on_each_cpu(hardware_disable_nolock, NULL, 1);
2566         return NOTIFY_OK;
2567 }
2568
2569 static struct notifier_block kvm_reboot_notifier = {
2570         .notifier_call = kvm_reboot,
2571         .priority = 0,
2572 };
2573
2574 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2575 {
2576         int i;
2577
2578         for (i = 0; i < bus->dev_count; i++) {
2579                 struct kvm_io_device *pos = bus->range[i].dev;
2580
2581                 kvm_iodevice_destructor(pos);
2582         }
2583         kfree(bus);
2584 }
2585
2586 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2587 {
2588         const struct kvm_io_range *r1 = p1;
2589         const struct kvm_io_range *r2 = p2;
2590
2591         if (r1->addr < r2->addr)
2592                 return -1;
2593         if (r1->addr + r1->len > r2->addr + r2->len)
2594                 return 1;
2595         return 0;
2596 }
2597
2598 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2599                           gpa_t addr, int len)
2600 {
2601         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2602                 .addr = addr,
2603                 .len = len,
2604                 .dev = dev,
2605         };
2606
2607         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2608                 kvm_io_bus_sort_cmp, NULL);
2609
2610         return 0;
2611 }
2612
2613 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2614                              gpa_t addr, int len)
2615 {
2616         struct kvm_io_range *range, key;
2617         int off;
2618
2619         key = (struct kvm_io_range) {
2620                 .addr = addr,
2621                 .len = len,
2622         };
2623
2624         range = bsearch(&key, bus->range, bus->dev_count,
2625                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2626         if (range == NULL)
2627                 return -ENOENT;
2628
2629         off = range - bus->range;
2630
2631         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2632                 off--;
2633
2634         return off;
2635 }
2636
2637 /* kvm_io_bus_write - called under kvm->slots_lock */
2638 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2639                      int len, const void *val)
2640 {
2641         int idx;
2642         struct kvm_io_bus *bus;
2643         struct kvm_io_range range;
2644
2645         range = (struct kvm_io_range) {
2646                 .addr = addr,
2647                 .len = len,
2648         };
2649
2650         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2651         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2652         if (idx < 0)
2653                 return -EOPNOTSUPP;
2654
2655         while (idx < bus->dev_count &&
2656                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2657                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2658                         return 0;
2659                 idx++;
2660         }
2661
2662         return -EOPNOTSUPP;
2663 }
2664
2665 /* kvm_io_bus_read - called under kvm->slots_lock */
2666 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2667                     int len, void *val)
2668 {
2669         int idx;
2670         struct kvm_io_bus *bus;
2671         struct kvm_io_range range;
2672
2673         range = (struct kvm_io_range) {
2674                 .addr = addr,
2675                 .len = len,
2676         };
2677
2678         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2679         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2680         if (idx < 0)
2681                 return -EOPNOTSUPP;
2682
2683         while (idx < bus->dev_count &&
2684                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2685                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2686                         return 0;
2687                 idx++;
2688         }
2689
2690         return -EOPNOTSUPP;
2691 }
2692
2693 /* Caller must hold slots_lock. */
2694 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2695                             int len, struct kvm_io_device *dev)
2696 {
2697         struct kvm_io_bus *new_bus, *bus;
2698
2699         bus = kvm->buses[bus_idx];
2700         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2701                 return -ENOSPC;
2702
2703         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2704                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2705         if (!new_bus)
2706                 return -ENOMEM;
2707         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2708                sizeof(struct kvm_io_range)));
2709         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2710         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2711         synchronize_srcu_expedited(&kvm->srcu);
2712         kfree(bus);
2713
2714         return 0;
2715 }
2716
2717 /* Caller must hold slots_lock. */
2718 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2719                               struct kvm_io_device *dev)
2720 {
2721         int i, r;
2722         struct kvm_io_bus *new_bus, *bus;
2723
2724         bus = kvm->buses[bus_idx];
2725         r = -ENOENT;
2726         for (i = 0; i < bus->dev_count; i++)
2727                 if (bus->range[i].dev == dev) {
2728                         r = 0;
2729                         break;
2730                 }
2731
2732         if (r)
2733                 return r;
2734
2735         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2736                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2737         if (!new_bus)
2738                 return -ENOMEM;
2739
2740         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2741         new_bus->dev_count--;
2742         memcpy(new_bus->range + i, bus->range + i + 1,
2743                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2744
2745         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2746         synchronize_srcu_expedited(&kvm->srcu);
2747         kfree(bus);
2748         return r;
2749 }
2750
2751 static struct notifier_block kvm_cpu_notifier = {
2752         .notifier_call = kvm_cpu_hotplug,
2753 };
2754
2755 static int vm_stat_get(void *_offset, u64 *val)
2756 {
2757         unsigned offset = (long)_offset;
2758         struct kvm *kvm;
2759
2760         *val = 0;
2761         raw_spin_lock(&kvm_lock);
2762         list_for_each_entry(kvm, &vm_list, vm_list)
2763                 *val += *(u32 *)((void *)kvm + offset);
2764         raw_spin_unlock(&kvm_lock);
2765         return 0;
2766 }
2767
2768 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2769
2770 static int vcpu_stat_get(void *_offset, u64 *val)
2771 {
2772         unsigned offset = (long)_offset;
2773         struct kvm *kvm;
2774         struct kvm_vcpu *vcpu;
2775         int i;
2776
2777         *val = 0;
2778         raw_spin_lock(&kvm_lock);
2779         list_for_each_entry(kvm, &vm_list, vm_list)
2780                 kvm_for_each_vcpu(i, vcpu, kvm)
2781                         *val += *(u32 *)((void *)vcpu + offset);
2782
2783         raw_spin_unlock(&kvm_lock);
2784         return 0;
2785 }
2786
2787 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2788
2789 static const struct file_operations *stat_fops[] = {
2790         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2791         [KVM_STAT_VM]   = &vm_stat_fops,
2792 };
2793
2794 static int kvm_init_debug(void)
2795 {
2796         int r = -EFAULT;
2797         struct kvm_stats_debugfs_item *p;
2798
2799         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2800         if (kvm_debugfs_dir == NULL)
2801                 goto out;
2802
2803         for (p = debugfs_entries; p->name; ++p) {
2804                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2805                                                 (void *)(long)p->offset,
2806                                                 stat_fops[p->kind]);
2807                 if (p->dentry == NULL)
2808                         goto out_dir;
2809         }
2810
2811         return 0;
2812
2813 out_dir:
2814         debugfs_remove_recursive(kvm_debugfs_dir);
2815 out:
2816         return r;
2817 }
2818
2819 static void kvm_exit_debug(void)
2820 {
2821         struct kvm_stats_debugfs_item *p;
2822
2823         for (p = debugfs_entries; p->name; ++p)
2824                 debugfs_remove(p->dentry);
2825         debugfs_remove(kvm_debugfs_dir);
2826 }
2827
2828 static int kvm_suspend(void)
2829 {
2830         if (kvm_usage_count)
2831                 hardware_disable_nolock(NULL);
2832         return 0;
2833 }
2834
2835 static void kvm_resume(void)
2836 {
2837         if (kvm_usage_count) {
2838                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2839                 hardware_enable_nolock(NULL);
2840         }
2841 }
2842
2843 static struct syscore_ops kvm_syscore_ops = {
2844         .suspend = kvm_suspend,
2845         .resume = kvm_resume,
2846 };
2847
2848 static inline
2849 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2850 {
2851         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2852 }
2853
2854 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2855 {
2856         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2857
2858         kvm_arch_vcpu_load(vcpu, cpu);
2859 }
2860
2861 static void kvm_sched_out(struct preempt_notifier *pn,
2862                           struct task_struct *next)
2863 {
2864         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2865
2866         kvm_arch_vcpu_put(vcpu);
2867 }
2868
2869 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2870                   struct module *module)
2871 {
2872         int r;
2873         int cpu;
2874
2875         r = kvm_arch_init(opaque);
2876         if (r)
2877                 goto out_fail;
2878
2879         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2880                 r = -ENOMEM;
2881                 goto out_free_0;
2882         }
2883
2884         r = kvm_arch_hardware_setup();
2885         if (r < 0)
2886                 goto out_free_0a;
2887
2888         for_each_online_cpu(cpu) {
2889                 smp_call_function_single(cpu,
2890                                 kvm_arch_check_processor_compat,
2891                                 &r, 1);
2892                 if (r < 0)
2893                         goto out_free_1;
2894         }
2895
2896         r = register_cpu_notifier(&kvm_cpu_notifier);
2897         if (r)
2898                 goto out_free_2;
2899         register_reboot_notifier(&kvm_reboot_notifier);
2900
2901         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2902         if (!vcpu_align)
2903                 vcpu_align = __alignof__(struct kvm_vcpu);
2904         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2905                                            0, NULL);
2906         if (!kvm_vcpu_cache) {
2907                 r = -ENOMEM;
2908                 goto out_free_3;
2909         }
2910
2911         r = kvm_async_pf_init();
2912         if (r)
2913                 goto out_free;
2914
2915         kvm_chardev_ops.owner = module;
2916         kvm_vm_fops.owner = module;
2917         kvm_vcpu_fops.owner = module;
2918
2919         r = misc_register(&kvm_dev);
2920         if (r) {
2921                 printk(KERN_ERR "kvm: misc device register failed\n");
2922                 goto out_unreg;
2923         }
2924
2925         register_syscore_ops(&kvm_syscore_ops);
2926
2927         kvm_preempt_ops.sched_in = kvm_sched_in;
2928         kvm_preempt_ops.sched_out = kvm_sched_out;
2929
2930         r = kvm_init_debug();
2931         if (r) {
2932                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2933                 goto out_undebugfs;
2934         }
2935
2936         return 0;
2937
2938 out_undebugfs:
2939         unregister_syscore_ops(&kvm_syscore_ops);
2940 out_unreg:
2941         kvm_async_pf_deinit();
2942 out_free:
2943         kmem_cache_destroy(kvm_vcpu_cache);
2944 out_free_3:
2945         unregister_reboot_notifier(&kvm_reboot_notifier);
2946         unregister_cpu_notifier(&kvm_cpu_notifier);
2947 out_free_2:
2948 out_free_1:
2949         kvm_arch_hardware_unsetup();
2950 out_free_0a:
2951         free_cpumask_var(cpus_hardware_enabled);
2952 out_free_0:
2953         kvm_arch_exit();
2954 out_fail:
2955         return r;
2956 }
2957 EXPORT_SYMBOL_GPL(kvm_init);
2958
2959 void kvm_exit(void)
2960 {
2961         kvm_exit_debug();
2962         misc_deregister(&kvm_dev);
2963         kmem_cache_destroy(kvm_vcpu_cache);
2964         kvm_async_pf_deinit();
2965         unregister_syscore_ops(&kvm_syscore_ops);
2966         unregister_reboot_notifier(&kvm_reboot_notifier);
2967         unregister_cpu_notifier(&kvm_cpu_notifier);
2968         on_each_cpu(hardware_disable_nolock, NULL, 1);
2969         kvm_arch_hardware_unsetup();
2970         kvm_arch_exit();
2971         free_cpumask_var(cpus_hardware_enabled);
2972 }
2973 EXPORT_SYMBOL_GPL(kvm_exit);