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