KVM: Rename KVM_MEMORY_SLOTS -> KVM_USER_MEM_SLOTS
[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 = NULL, *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_USER_MEM_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                 /* slot was deleted or moved, clear iommu mapping */
826                 kvm_iommu_unmap_pages(kvm, &old);
827                 /* From this point no new shadow pages pointing to a deleted,
828                  * or moved, memslot will be created.
829                  *
830                  * validation of sp->gfn happens in:
831                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
832                  *      - kvm_is_visible_gfn (mmu_check_roots)
833                  */
834                 kvm_arch_flush_shadow_memslot(kvm, slot);
835                 slots = old_memslots;
836         }
837
838         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
839         if (r)
840                 goto out_slots;
841
842         r = -ENOMEM;
843         /*
844          * We can re-use the old_memslots from above, the only difference
845          * from the currently installed memslots is the invalid flag.  This
846          * will get overwritten by update_memslots anyway.
847          */
848         if (!slots) {
849                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
850                                 GFP_KERNEL);
851                 if (!slots)
852                         goto out_free;
853         }
854
855         /* map new memory slot into the iommu */
856         if (npages) {
857                 r = kvm_iommu_map_pages(kvm, &new);
858                 if (r)
859                         goto out_slots;
860         }
861
862         /* actual memory is freed via old in kvm_free_physmem_slot below */
863         if (!npages) {
864                 new.dirty_bitmap = NULL;
865                 memset(&new.arch, 0, sizeof(new.arch));
866         }
867
868         update_memslots(slots, &new);
869         old_memslots = kvm->memslots;
870         rcu_assign_pointer(kvm->memslots, slots);
871         synchronize_srcu_expedited(&kvm->srcu);
872
873         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
874
875         kvm_free_physmem_slot(&old, &new);
876         kfree(old_memslots);
877
878         return 0;
879
880 out_slots:
881         kfree(slots);
882 out_free:
883         kvm_free_physmem_slot(&new, &old);
884 out:
885         return r;
886
887 }
888 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
889
890 int kvm_set_memory_region(struct kvm *kvm,
891                           struct kvm_userspace_memory_region *mem,
892                           int user_alloc)
893 {
894         int r;
895
896         mutex_lock(&kvm->slots_lock);
897         r = __kvm_set_memory_region(kvm, mem, user_alloc);
898         mutex_unlock(&kvm->slots_lock);
899         return r;
900 }
901 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
902
903 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
904                                    struct
905                                    kvm_userspace_memory_region *mem,
906                                    int user_alloc)
907 {
908         if (mem->slot >= KVM_USER_MEM_SLOTS)
909                 return -EINVAL;
910         return kvm_set_memory_region(kvm, mem, user_alloc);
911 }
912
913 int kvm_get_dirty_log(struct kvm *kvm,
914                         struct kvm_dirty_log *log, int *is_dirty)
915 {
916         struct kvm_memory_slot *memslot;
917         int r, i;
918         unsigned long n;
919         unsigned long any = 0;
920
921         r = -EINVAL;
922         if (log->slot >= KVM_USER_MEM_SLOTS)
923                 goto out;
924
925         memslot = id_to_memslot(kvm->memslots, log->slot);
926         r = -ENOENT;
927         if (!memslot->dirty_bitmap)
928                 goto out;
929
930         n = kvm_dirty_bitmap_bytes(memslot);
931
932         for (i = 0; !any && i < n/sizeof(long); ++i)
933                 any = memslot->dirty_bitmap[i];
934
935         r = -EFAULT;
936         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
937                 goto out;
938
939         if (any)
940                 *is_dirty = 1;
941
942         r = 0;
943 out:
944         return r;
945 }
946
947 bool kvm_largepages_enabled(void)
948 {
949         return largepages_enabled;
950 }
951
952 void kvm_disable_largepages(void)
953 {
954         largepages_enabled = false;
955 }
956 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
957
958 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
959 {
960         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
961 }
962 EXPORT_SYMBOL_GPL(gfn_to_memslot);
963
964 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
965 {
966         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
967
968         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
969               memslot->flags & KVM_MEMSLOT_INVALID)
970                 return 0;
971
972         return 1;
973 }
974 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
975
976 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
977 {
978         struct vm_area_struct *vma;
979         unsigned long addr, size;
980
981         size = PAGE_SIZE;
982
983         addr = gfn_to_hva(kvm, gfn);
984         if (kvm_is_error_hva(addr))
985                 return PAGE_SIZE;
986
987         down_read(&current->mm->mmap_sem);
988         vma = find_vma(current->mm, addr);
989         if (!vma)
990                 goto out;
991
992         size = vma_kernel_pagesize(vma);
993
994 out:
995         up_read(&current->mm->mmap_sem);
996
997         return size;
998 }
999
1000 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1001 {
1002         return slot->flags & KVM_MEM_READONLY;
1003 }
1004
1005 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1006                                        gfn_t *nr_pages, bool write)
1007 {
1008         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1009                 return KVM_HVA_ERR_BAD;
1010
1011         if (memslot_is_readonly(slot) && write)
1012                 return KVM_HVA_ERR_RO_BAD;
1013
1014         if (nr_pages)
1015                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1016
1017         return __gfn_to_hva_memslot(slot, gfn);
1018 }
1019
1020 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1021                                      gfn_t *nr_pages)
1022 {
1023         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1024 }
1025
1026 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1027                                  gfn_t gfn)
1028 {
1029         return gfn_to_hva_many(slot, gfn, NULL);
1030 }
1031 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1032
1033 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1034 {
1035         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1036 }
1037 EXPORT_SYMBOL_GPL(gfn_to_hva);
1038
1039 /*
1040  * The hva returned by this function is only allowed to be read.
1041  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1042  */
1043 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1044 {
1045         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1046 }
1047
1048 static int kvm_read_hva(void *data, void __user *hva, int len)
1049 {
1050         return __copy_from_user(data, hva, len);
1051 }
1052
1053 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1054 {
1055         return __copy_from_user_inatomic(data, hva, len);
1056 }
1057
1058 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1059         unsigned long start, int write, struct page **page)
1060 {
1061         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1062
1063         if (write)
1064                 flags |= FOLL_WRITE;
1065
1066         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1067 }
1068
1069 static inline int check_user_page_hwpoison(unsigned long addr)
1070 {
1071         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1072
1073         rc = __get_user_pages(current, current->mm, addr, 1,
1074                               flags, NULL, NULL, NULL);
1075         return rc == -EHWPOISON;
1076 }
1077
1078 /*
1079  * The atomic path to get the writable pfn which will be stored in @pfn,
1080  * true indicates success, otherwise false is returned.
1081  */
1082 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1083                             bool write_fault, bool *writable, pfn_t *pfn)
1084 {
1085         struct page *page[1];
1086         int npages;
1087
1088         if (!(async || atomic))
1089                 return false;
1090
1091         /*
1092          * Fast pin a writable pfn only if it is a write fault request
1093          * or the caller allows to map a writable pfn for a read fault
1094          * request.
1095          */
1096         if (!(write_fault || writable))
1097                 return false;
1098
1099         npages = __get_user_pages_fast(addr, 1, 1, page);
1100         if (npages == 1) {
1101                 *pfn = page_to_pfn(page[0]);
1102
1103                 if (writable)
1104                         *writable = true;
1105                 return true;
1106         }
1107
1108         return false;
1109 }
1110
1111 /*
1112  * The slow path to get the pfn of the specified host virtual address,
1113  * 1 indicates success, -errno is returned if error is detected.
1114  */
1115 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1116                            bool *writable, pfn_t *pfn)
1117 {
1118         struct page *page[1];
1119         int npages = 0;
1120
1121         might_sleep();
1122
1123         if (writable)
1124                 *writable = write_fault;
1125
1126         if (async) {
1127                 down_read(&current->mm->mmap_sem);
1128                 npages = get_user_page_nowait(current, current->mm,
1129                                               addr, write_fault, page);
1130                 up_read(&current->mm->mmap_sem);
1131         } else
1132                 npages = get_user_pages_fast(addr, 1, write_fault,
1133                                              page);
1134         if (npages != 1)
1135                 return npages;
1136
1137         /* map read fault as writable if possible */
1138         if (unlikely(!write_fault) && writable) {
1139                 struct page *wpage[1];
1140
1141                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1142                 if (npages == 1) {
1143                         *writable = true;
1144                         put_page(page[0]);
1145                         page[0] = wpage[0];
1146                 }
1147
1148                 npages = 1;
1149         }
1150         *pfn = page_to_pfn(page[0]);
1151         return npages;
1152 }
1153
1154 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1155 {
1156         if (unlikely(!(vma->vm_flags & VM_READ)))
1157                 return false;
1158
1159         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1160                 return false;
1161
1162         return true;
1163 }
1164
1165 /*
1166  * Pin guest page in memory and return its pfn.
1167  * @addr: host virtual address which maps memory to the guest
1168  * @atomic: whether this function can sleep
1169  * @async: whether this function need to wait IO complete if the
1170  *         host page is not in the memory
1171  * @write_fault: whether we should get a writable host page
1172  * @writable: whether it allows to map a writable host page for !@write_fault
1173  *
1174  * The function will map a writable host page for these two cases:
1175  * 1): @write_fault = true
1176  * 2): @write_fault = false && @writable, @writable will tell the caller
1177  *     whether the mapping is writable.
1178  */
1179 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1180                         bool write_fault, bool *writable)
1181 {
1182         struct vm_area_struct *vma;
1183         pfn_t pfn = 0;
1184         int npages;
1185
1186         /* we can do it either atomically or asynchronously, not both */
1187         BUG_ON(atomic && async);
1188
1189         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1190                 return pfn;
1191
1192         if (atomic)
1193                 return KVM_PFN_ERR_FAULT;
1194
1195         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1196         if (npages == 1)
1197                 return pfn;
1198
1199         down_read(&current->mm->mmap_sem);
1200         if (npages == -EHWPOISON ||
1201               (!async && check_user_page_hwpoison(addr))) {
1202                 pfn = KVM_PFN_ERR_HWPOISON;
1203                 goto exit;
1204         }
1205
1206         vma = find_vma_intersection(current->mm, addr, addr + 1);
1207
1208         if (vma == NULL)
1209                 pfn = KVM_PFN_ERR_FAULT;
1210         else if ((vma->vm_flags & VM_PFNMAP)) {
1211                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1212                         vma->vm_pgoff;
1213                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1214         } else {
1215                 if (async && vma_is_valid(vma, write_fault))
1216                         *async = true;
1217                 pfn = KVM_PFN_ERR_FAULT;
1218         }
1219 exit:
1220         up_read(&current->mm->mmap_sem);
1221         return pfn;
1222 }
1223
1224 static pfn_t
1225 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1226                      bool *async, bool write_fault, bool *writable)
1227 {
1228         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1229
1230         if (addr == KVM_HVA_ERR_RO_BAD)
1231                 return KVM_PFN_ERR_RO_FAULT;
1232
1233         if (kvm_is_error_hva(addr))
1234                 return KVM_PFN_NOSLOT;
1235
1236         /* Do not map writable pfn in the readonly memslot. */
1237         if (writable && memslot_is_readonly(slot)) {
1238                 *writable = false;
1239                 writable = NULL;
1240         }
1241
1242         return hva_to_pfn(addr, atomic, async, write_fault,
1243                           writable);
1244 }
1245
1246 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1247                           bool write_fault, bool *writable)
1248 {
1249         struct kvm_memory_slot *slot;
1250
1251         if (async)
1252                 *async = false;
1253
1254         slot = gfn_to_memslot(kvm, gfn);
1255
1256         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1257                                     writable);
1258 }
1259
1260 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1261 {
1262         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1263 }
1264 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1265
1266 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1267                        bool write_fault, bool *writable)
1268 {
1269         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1270 }
1271 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1272
1273 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1274 {
1275         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1276 }
1277 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1278
1279 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1280                       bool *writable)
1281 {
1282         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1283 }
1284 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1285
1286 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1287 {
1288         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1289 }
1290
1291 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1292 {
1293         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1294 }
1295 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1296
1297 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1298                                                                   int nr_pages)
1299 {
1300         unsigned long addr;
1301         gfn_t entry;
1302
1303         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1304         if (kvm_is_error_hva(addr))
1305                 return -1;
1306
1307         if (entry < nr_pages)
1308                 return 0;
1309
1310         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1311 }
1312 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1313
1314 static struct page *kvm_pfn_to_page(pfn_t pfn)
1315 {
1316         if (is_error_noslot_pfn(pfn))
1317                 return KVM_ERR_PTR_BAD_PAGE;
1318
1319         if (kvm_is_mmio_pfn(pfn)) {
1320                 WARN_ON(1);
1321                 return KVM_ERR_PTR_BAD_PAGE;
1322         }
1323
1324         return pfn_to_page(pfn);
1325 }
1326
1327 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1328 {
1329         pfn_t pfn;
1330
1331         pfn = gfn_to_pfn(kvm, gfn);
1332
1333         return kvm_pfn_to_page(pfn);
1334 }
1335
1336 EXPORT_SYMBOL_GPL(gfn_to_page);
1337
1338 void kvm_release_page_clean(struct page *page)
1339 {
1340         WARN_ON(is_error_page(page));
1341
1342         kvm_release_pfn_clean(page_to_pfn(page));
1343 }
1344 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1345
1346 void kvm_release_pfn_clean(pfn_t pfn)
1347 {
1348         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1349                 put_page(pfn_to_page(pfn));
1350 }
1351 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1352
1353 void kvm_release_page_dirty(struct page *page)
1354 {
1355         WARN_ON(is_error_page(page));
1356
1357         kvm_release_pfn_dirty(page_to_pfn(page));
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1360
1361 void kvm_release_pfn_dirty(pfn_t pfn)
1362 {
1363         kvm_set_pfn_dirty(pfn);
1364         kvm_release_pfn_clean(pfn);
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1367
1368 void kvm_set_page_dirty(struct page *page)
1369 {
1370         kvm_set_pfn_dirty(page_to_pfn(page));
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1373
1374 void kvm_set_pfn_dirty(pfn_t pfn)
1375 {
1376         if (!kvm_is_mmio_pfn(pfn)) {
1377                 struct page *page = pfn_to_page(pfn);
1378                 if (!PageReserved(page))
1379                         SetPageDirty(page);
1380         }
1381 }
1382 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1383
1384 void kvm_set_pfn_accessed(pfn_t pfn)
1385 {
1386         if (!kvm_is_mmio_pfn(pfn))
1387                 mark_page_accessed(pfn_to_page(pfn));
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1390
1391 void kvm_get_pfn(pfn_t pfn)
1392 {
1393         if (!kvm_is_mmio_pfn(pfn))
1394                 get_page(pfn_to_page(pfn));
1395 }
1396 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1397
1398 static int next_segment(unsigned long len, int offset)
1399 {
1400         if (len > PAGE_SIZE - offset)
1401                 return PAGE_SIZE - offset;
1402         else
1403                 return len;
1404 }
1405
1406 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1407                         int len)
1408 {
1409         int r;
1410         unsigned long addr;
1411
1412         addr = gfn_to_hva_read(kvm, gfn);
1413         if (kvm_is_error_hva(addr))
1414                 return -EFAULT;
1415         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1416         if (r)
1417                 return -EFAULT;
1418         return 0;
1419 }
1420 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1421
1422 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1423 {
1424         gfn_t gfn = gpa >> PAGE_SHIFT;
1425         int seg;
1426         int offset = offset_in_page(gpa);
1427         int ret;
1428
1429         while ((seg = next_segment(len, offset)) != 0) {
1430                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1431                 if (ret < 0)
1432                         return ret;
1433                 offset = 0;
1434                 len -= seg;
1435                 data += seg;
1436                 ++gfn;
1437         }
1438         return 0;
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_read_guest);
1441
1442 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1443                           unsigned long len)
1444 {
1445         int r;
1446         unsigned long addr;
1447         gfn_t gfn = gpa >> PAGE_SHIFT;
1448         int offset = offset_in_page(gpa);
1449
1450         addr = gfn_to_hva_read(kvm, gfn);
1451         if (kvm_is_error_hva(addr))
1452                 return -EFAULT;
1453         pagefault_disable();
1454         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1455         pagefault_enable();
1456         if (r)
1457                 return -EFAULT;
1458         return 0;
1459 }
1460 EXPORT_SYMBOL(kvm_read_guest_atomic);
1461
1462 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1463                          int offset, int len)
1464 {
1465         int r;
1466         unsigned long addr;
1467
1468         addr = gfn_to_hva(kvm, gfn);
1469         if (kvm_is_error_hva(addr))
1470                 return -EFAULT;
1471         r = __copy_to_user((void __user *)addr + offset, data, len);
1472         if (r)
1473                 return -EFAULT;
1474         mark_page_dirty(kvm, gfn);
1475         return 0;
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1478
1479 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1480                     unsigned long len)
1481 {
1482         gfn_t gfn = gpa >> PAGE_SHIFT;
1483         int seg;
1484         int offset = offset_in_page(gpa);
1485         int ret;
1486
1487         while ((seg = next_segment(len, offset)) != 0) {
1488                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1489                 if (ret < 0)
1490                         return ret;
1491                 offset = 0;
1492                 len -= seg;
1493                 data += seg;
1494                 ++gfn;
1495         }
1496         return 0;
1497 }
1498
1499 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1500                               gpa_t gpa)
1501 {
1502         struct kvm_memslots *slots = kvm_memslots(kvm);
1503         int offset = offset_in_page(gpa);
1504         gfn_t gfn = gpa >> PAGE_SHIFT;
1505
1506         ghc->gpa = gpa;
1507         ghc->generation = slots->generation;
1508         ghc->memslot = gfn_to_memslot(kvm, gfn);
1509         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1510         if (!kvm_is_error_hva(ghc->hva))
1511                 ghc->hva += offset;
1512         else
1513                 return -EFAULT;
1514
1515         return 0;
1516 }
1517 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1518
1519 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1520                            void *data, unsigned long len)
1521 {
1522         struct kvm_memslots *slots = kvm_memslots(kvm);
1523         int r;
1524
1525         if (slots->generation != ghc->generation)
1526                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1527
1528         if (kvm_is_error_hva(ghc->hva))
1529                 return -EFAULT;
1530
1531         r = __copy_to_user((void __user *)ghc->hva, data, len);
1532         if (r)
1533                 return -EFAULT;
1534         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1535
1536         return 0;
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1539
1540 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1541                            void *data, unsigned long len)
1542 {
1543         struct kvm_memslots *slots = kvm_memslots(kvm);
1544         int r;
1545
1546         if (slots->generation != ghc->generation)
1547                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1548
1549         if (kvm_is_error_hva(ghc->hva))
1550                 return -EFAULT;
1551
1552         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1553         if (r)
1554                 return -EFAULT;
1555
1556         return 0;
1557 }
1558 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1559
1560 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1561 {
1562         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1563                                     offset, len);
1564 }
1565 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1566
1567 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1568 {
1569         gfn_t gfn = gpa >> PAGE_SHIFT;
1570         int seg;
1571         int offset = offset_in_page(gpa);
1572         int ret;
1573
1574         while ((seg = next_segment(len, offset)) != 0) {
1575                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1576                 if (ret < 0)
1577                         return ret;
1578                 offset = 0;
1579                 len -= seg;
1580                 ++gfn;
1581         }
1582         return 0;
1583 }
1584 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1585
1586 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1587                              gfn_t gfn)
1588 {
1589         if (memslot && memslot->dirty_bitmap) {
1590                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1591
1592                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1593         }
1594 }
1595
1596 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1597 {
1598         struct kvm_memory_slot *memslot;
1599
1600         memslot = gfn_to_memslot(kvm, gfn);
1601         mark_page_dirty_in_slot(kvm, memslot, gfn);
1602 }
1603
1604 /*
1605  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1606  */
1607 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1608 {
1609         DEFINE_WAIT(wait);
1610
1611         for (;;) {
1612                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1613
1614                 if (kvm_arch_vcpu_runnable(vcpu)) {
1615                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1616                         break;
1617                 }
1618                 if (kvm_cpu_has_pending_timer(vcpu))
1619                         break;
1620                 if (signal_pending(current))
1621                         break;
1622
1623                 schedule();
1624         }
1625
1626         finish_wait(&vcpu->wq, &wait);
1627 }
1628
1629 #ifndef CONFIG_S390
1630 /*
1631  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1632  */
1633 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1634 {
1635         int me;
1636         int cpu = vcpu->cpu;
1637         wait_queue_head_t *wqp;
1638
1639         wqp = kvm_arch_vcpu_wq(vcpu);
1640         if (waitqueue_active(wqp)) {
1641                 wake_up_interruptible(wqp);
1642                 ++vcpu->stat.halt_wakeup;
1643         }
1644
1645         me = get_cpu();
1646         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1647                 if (kvm_arch_vcpu_should_kick(vcpu))
1648                         smp_send_reschedule(cpu);
1649         put_cpu();
1650 }
1651 #endif /* !CONFIG_S390 */
1652
1653 void kvm_resched(struct kvm_vcpu *vcpu)
1654 {
1655         if (!need_resched())
1656                 return;
1657         cond_resched();
1658 }
1659 EXPORT_SYMBOL_GPL(kvm_resched);
1660
1661 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1662 {
1663         struct pid *pid;
1664         struct task_struct *task = NULL;
1665
1666         rcu_read_lock();
1667         pid = rcu_dereference(target->pid);
1668         if (pid)
1669                 task = get_pid_task(target->pid, PIDTYPE_PID);
1670         rcu_read_unlock();
1671         if (!task)
1672                 return false;
1673         if (task->flags & PF_VCPU) {
1674                 put_task_struct(task);
1675                 return false;
1676         }
1677         if (yield_to(task, 1)) {
1678                 put_task_struct(task);
1679                 return true;
1680         }
1681         put_task_struct(task);
1682         return false;
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1685
1686 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1687 /*
1688  * Helper that checks whether a VCPU is eligible for directed yield.
1689  * Most eligible candidate to yield is decided by following heuristics:
1690  *
1691  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1692  *  (preempted lock holder), indicated by @in_spin_loop.
1693  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1694  *
1695  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1696  *  chance last time (mostly it has become eligible now since we have probably
1697  *  yielded to lockholder in last iteration. This is done by toggling
1698  *  @dy_eligible each time a VCPU checked for eligibility.)
1699  *
1700  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1701  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1702  *  burning. Giving priority for a potential lock-holder increases lock
1703  *  progress.
1704  *
1705  *  Since algorithm is based on heuristics, accessing another VCPU data without
1706  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1707  *  and continue with next VCPU and so on.
1708  */
1709 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1710 {
1711         bool eligible;
1712
1713         eligible = !vcpu->spin_loop.in_spin_loop ||
1714                         (vcpu->spin_loop.in_spin_loop &&
1715                          vcpu->spin_loop.dy_eligible);
1716
1717         if (vcpu->spin_loop.in_spin_loop)
1718                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1719
1720         return eligible;
1721 }
1722 #endif
1723 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1724 {
1725         struct kvm *kvm = me->kvm;
1726         struct kvm_vcpu *vcpu;
1727         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1728         int yielded = 0;
1729         int pass;
1730         int i;
1731
1732         kvm_vcpu_set_in_spin_loop(me, true);
1733         /*
1734          * We boost the priority of a VCPU that is runnable but not
1735          * currently running, because it got preempted by something
1736          * else and called schedule in __vcpu_run.  Hopefully that
1737          * VCPU is holding the lock that we need and will release it.
1738          * We approximate round-robin by starting at the last boosted VCPU.
1739          */
1740         for (pass = 0; pass < 2 && !yielded; pass++) {
1741                 kvm_for_each_vcpu(i, vcpu, kvm) {
1742                         if (!pass && i <= last_boosted_vcpu) {
1743                                 i = last_boosted_vcpu;
1744                                 continue;
1745                         } else if (pass && i > last_boosted_vcpu)
1746                                 break;
1747                         if (vcpu == me)
1748                                 continue;
1749                         if (waitqueue_active(&vcpu->wq))
1750                                 continue;
1751                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1752                                 continue;
1753                         if (kvm_vcpu_yield_to(vcpu)) {
1754                                 kvm->last_boosted_vcpu = i;
1755                                 yielded = 1;
1756                                 break;
1757                         }
1758                 }
1759         }
1760         kvm_vcpu_set_in_spin_loop(me, false);
1761
1762         /* Ensure vcpu is not eligible during next spinloop */
1763         kvm_vcpu_set_dy_eligible(me, false);
1764 }
1765 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1766
1767 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1768 {
1769         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1770         struct page *page;
1771
1772         if (vmf->pgoff == 0)
1773                 page = virt_to_page(vcpu->run);
1774 #ifdef CONFIG_X86
1775         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1776                 page = virt_to_page(vcpu->arch.pio_data);
1777 #endif
1778 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1779         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1780                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1781 #endif
1782         else
1783                 return kvm_arch_vcpu_fault(vcpu, vmf);
1784         get_page(page);
1785         vmf->page = page;
1786         return 0;
1787 }
1788
1789 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1790         .fault = kvm_vcpu_fault,
1791 };
1792
1793 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1794 {
1795         vma->vm_ops = &kvm_vcpu_vm_ops;
1796         return 0;
1797 }
1798
1799 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1800 {
1801         struct kvm_vcpu *vcpu = filp->private_data;
1802
1803         kvm_put_kvm(vcpu->kvm);
1804         return 0;
1805 }
1806
1807 static struct file_operations kvm_vcpu_fops = {
1808         .release        = kvm_vcpu_release,
1809         .unlocked_ioctl = kvm_vcpu_ioctl,
1810 #ifdef CONFIG_COMPAT
1811         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1812 #endif
1813         .mmap           = kvm_vcpu_mmap,
1814         .llseek         = noop_llseek,
1815 };
1816
1817 /*
1818  * Allocates an inode for the vcpu.
1819  */
1820 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1821 {
1822         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1823 }
1824
1825 /*
1826  * Creates some virtual cpus.  Good luck creating more than one.
1827  */
1828 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1829 {
1830         int r;
1831         struct kvm_vcpu *vcpu, *v;
1832
1833         vcpu = kvm_arch_vcpu_create(kvm, id);
1834         if (IS_ERR(vcpu))
1835                 return PTR_ERR(vcpu);
1836
1837         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1838
1839         r = kvm_arch_vcpu_setup(vcpu);
1840         if (r)
1841                 goto vcpu_destroy;
1842
1843         mutex_lock(&kvm->lock);
1844         if (!kvm_vcpu_compatible(vcpu)) {
1845                 r = -EINVAL;
1846                 goto unlock_vcpu_destroy;
1847         }
1848         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1849                 r = -EINVAL;
1850                 goto unlock_vcpu_destroy;
1851         }
1852
1853         kvm_for_each_vcpu(r, v, kvm)
1854                 if (v->vcpu_id == id) {
1855                         r = -EEXIST;
1856                         goto unlock_vcpu_destroy;
1857                 }
1858
1859         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1860
1861         /* Now it's all set up, let userspace reach it */
1862         kvm_get_kvm(kvm);
1863         r = create_vcpu_fd(vcpu);
1864         if (r < 0) {
1865                 kvm_put_kvm(kvm);
1866                 goto unlock_vcpu_destroy;
1867         }
1868
1869         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1870         smp_wmb();
1871         atomic_inc(&kvm->online_vcpus);
1872
1873         mutex_unlock(&kvm->lock);
1874         kvm_arch_vcpu_postcreate(vcpu);
1875         return r;
1876
1877 unlock_vcpu_destroy:
1878         mutex_unlock(&kvm->lock);
1879 vcpu_destroy:
1880         kvm_arch_vcpu_destroy(vcpu);
1881         return r;
1882 }
1883
1884 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1885 {
1886         if (sigset) {
1887                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1888                 vcpu->sigset_active = 1;
1889                 vcpu->sigset = *sigset;
1890         } else
1891                 vcpu->sigset_active = 0;
1892         return 0;
1893 }
1894
1895 static long kvm_vcpu_ioctl(struct file *filp,
1896                            unsigned int ioctl, unsigned long arg)
1897 {
1898         struct kvm_vcpu *vcpu = filp->private_data;
1899         void __user *argp = (void __user *)arg;
1900         int r;
1901         struct kvm_fpu *fpu = NULL;
1902         struct kvm_sregs *kvm_sregs = NULL;
1903
1904         if (vcpu->kvm->mm != current->mm)
1905                 return -EIO;
1906
1907 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1908         /*
1909          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1910          * so vcpu_load() would break it.
1911          */
1912         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1913                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1914 #endif
1915
1916
1917         r = vcpu_load(vcpu);
1918         if (r)
1919                 return r;
1920         switch (ioctl) {
1921         case KVM_RUN:
1922                 r = -EINVAL;
1923                 if (arg)
1924                         goto out;
1925                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1926                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1927                 break;
1928         case KVM_GET_REGS: {
1929                 struct kvm_regs *kvm_regs;
1930
1931                 r = -ENOMEM;
1932                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1933                 if (!kvm_regs)
1934                         goto out;
1935                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1936                 if (r)
1937                         goto out_free1;
1938                 r = -EFAULT;
1939                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1940                         goto out_free1;
1941                 r = 0;
1942 out_free1:
1943                 kfree(kvm_regs);
1944                 break;
1945         }
1946         case KVM_SET_REGS: {
1947                 struct kvm_regs *kvm_regs;
1948
1949                 r = -ENOMEM;
1950                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1951                 if (IS_ERR(kvm_regs)) {
1952                         r = PTR_ERR(kvm_regs);
1953                         goto out;
1954                 }
1955                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1956                 kfree(kvm_regs);
1957                 break;
1958         }
1959         case KVM_GET_SREGS: {
1960                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1961                 r = -ENOMEM;
1962                 if (!kvm_sregs)
1963                         goto out;
1964                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1965                 if (r)
1966                         goto out;
1967                 r = -EFAULT;
1968                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1969                         goto out;
1970                 r = 0;
1971                 break;
1972         }
1973         case KVM_SET_SREGS: {
1974                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1975                 if (IS_ERR(kvm_sregs)) {
1976                         r = PTR_ERR(kvm_sregs);
1977                         kvm_sregs = NULL;
1978                         goto out;
1979                 }
1980                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1981                 break;
1982         }
1983         case KVM_GET_MP_STATE: {
1984                 struct kvm_mp_state mp_state;
1985
1986                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1987                 if (r)
1988                         goto out;
1989                 r = -EFAULT;
1990                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1991                         goto out;
1992                 r = 0;
1993                 break;
1994         }
1995         case KVM_SET_MP_STATE: {
1996                 struct kvm_mp_state mp_state;
1997
1998                 r = -EFAULT;
1999                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2000                         goto out;
2001                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2002                 break;
2003         }
2004         case KVM_TRANSLATE: {
2005                 struct kvm_translation tr;
2006
2007                 r = -EFAULT;
2008                 if (copy_from_user(&tr, argp, sizeof tr))
2009                         goto out;
2010                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2011                 if (r)
2012                         goto out;
2013                 r = -EFAULT;
2014                 if (copy_to_user(argp, &tr, sizeof tr))
2015                         goto out;
2016                 r = 0;
2017                 break;
2018         }
2019         case KVM_SET_GUEST_DEBUG: {
2020                 struct kvm_guest_debug dbg;
2021
2022                 r = -EFAULT;
2023                 if (copy_from_user(&dbg, argp, sizeof dbg))
2024                         goto out;
2025                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2026                 break;
2027         }
2028         case KVM_SET_SIGNAL_MASK: {
2029                 struct kvm_signal_mask __user *sigmask_arg = argp;
2030                 struct kvm_signal_mask kvm_sigmask;
2031                 sigset_t sigset, *p;
2032
2033                 p = NULL;
2034                 if (argp) {
2035                         r = -EFAULT;
2036                         if (copy_from_user(&kvm_sigmask, argp,
2037                                            sizeof kvm_sigmask))
2038                                 goto out;
2039                         r = -EINVAL;
2040                         if (kvm_sigmask.len != sizeof sigset)
2041                                 goto out;
2042                         r = -EFAULT;
2043                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2044                                            sizeof sigset))
2045                                 goto out;
2046                         p = &sigset;
2047                 }
2048                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2049                 break;
2050         }
2051         case KVM_GET_FPU: {
2052                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2053                 r = -ENOMEM;
2054                 if (!fpu)
2055                         goto out;
2056                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2057                 if (r)
2058                         goto out;
2059                 r = -EFAULT;
2060                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2061                         goto out;
2062                 r = 0;
2063                 break;
2064         }
2065         case KVM_SET_FPU: {
2066                 fpu = memdup_user(argp, sizeof(*fpu));
2067                 if (IS_ERR(fpu)) {
2068                         r = PTR_ERR(fpu);
2069                         fpu = NULL;
2070                         goto out;
2071                 }
2072                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2073                 break;
2074         }
2075         default:
2076                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2077         }
2078 out:
2079         vcpu_put(vcpu);
2080         kfree(fpu);
2081         kfree(kvm_sregs);
2082         return r;
2083 }
2084
2085 #ifdef CONFIG_COMPAT
2086 static long kvm_vcpu_compat_ioctl(struct file *filp,
2087                                   unsigned int ioctl, unsigned long arg)
2088 {
2089         struct kvm_vcpu *vcpu = filp->private_data;
2090         void __user *argp = compat_ptr(arg);
2091         int r;
2092
2093         if (vcpu->kvm->mm != current->mm)
2094                 return -EIO;
2095
2096         switch (ioctl) {
2097         case KVM_SET_SIGNAL_MASK: {
2098                 struct kvm_signal_mask __user *sigmask_arg = argp;
2099                 struct kvm_signal_mask kvm_sigmask;
2100                 compat_sigset_t csigset;
2101                 sigset_t sigset;
2102
2103                 if (argp) {
2104                         r = -EFAULT;
2105                         if (copy_from_user(&kvm_sigmask, argp,
2106                                            sizeof kvm_sigmask))
2107                                 goto out;
2108                         r = -EINVAL;
2109                         if (kvm_sigmask.len != sizeof csigset)
2110                                 goto out;
2111                         r = -EFAULT;
2112                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2113                                            sizeof csigset))
2114                                 goto out;
2115                         sigset_from_compat(&sigset, &csigset);
2116                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2117                 } else
2118                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2119                 break;
2120         }
2121         default:
2122                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2123         }
2124
2125 out:
2126         return r;
2127 }
2128 #endif
2129
2130 static long kvm_vm_ioctl(struct file *filp,
2131                            unsigned int ioctl, unsigned long arg)
2132 {
2133         struct kvm *kvm = filp->private_data;
2134         void __user *argp = (void __user *)arg;
2135         int r;
2136
2137         if (kvm->mm != current->mm)
2138                 return -EIO;
2139         switch (ioctl) {
2140         case KVM_CREATE_VCPU:
2141                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2142                 break;
2143         case KVM_SET_USER_MEMORY_REGION: {
2144                 struct kvm_userspace_memory_region kvm_userspace_mem;
2145
2146                 r = -EFAULT;
2147                 if (copy_from_user(&kvm_userspace_mem, argp,
2148                                                 sizeof kvm_userspace_mem))
2149                         goto out;
2150
2151                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2152                 break;
2153         }
2154         case KVM_GET_DIRTY_LOG: {
2155                 struct kvm_dirty_log log;
2156
2157                 r = -EFAULT;
2158                 if (copy_from_user(&log, argp, sizeof log))
2159                         goto out;
2160                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2161                 break;
2162         }
2163 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2164         case KVM_REGISTER_COALESCED_MMIO: {
2165                 struct kvm_coalesced_mmio_zone zone;
2166                 r = -EFAULT;
2167                 if (copy_from_user(&zone, argp, sizeof zone))
2168                         goto out;
2169                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2170                 break;
2171         }
2172         case KVM_UNREGISTER_COALESCED_MMIO: {
2173                 struct kvm_coalesced_mmio_zone zone;
2174                 r = -EFAULT;
2175                 if (copy_from_user(&zone, argp, sizeof zone))
2176                         goto out;
2177                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2178                 break;
2179         }
2180 #endif
2181         case KVM_IRQFD: {
2182                 struct kvm_irqfd data;
2183
2184                 r = -EFAULT;
2185                 if (copy_from_user(&data, argp, sizeof data))
2186                         goto out;
2187                 r = kvm_irqfd(kvm, &data);
2188                 break;
2189         }
2190         case KVM_IOEVENTFD: {
2191                 struct kvm_ioeventfd data;
2192
2193                 r = -EFAULT;
2194                 if (copy_from_user(&data, argp, sizeof data))
2195                         goto out;
2196                 r = kvm_ioeventfd(kvm, &data);
2197                 break;
2198         }
2199 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2200         case KVM_SET_BOOT_CPU_ID:
2201                 r = 0;
2202                 mutex_lock(&kvm->lock);
2203                 if (atomic_read(&kvm->online_vcpus) != 0)
2204                         r = -EBUSY;
2205                 else
2206                         kvm->bsp_vcpu_id = arg;
2207                 mutex_unlock(&kvm->lock);
2208                 break;
2209 #endif
2210 #ifdef CONFIG_HAVE_KVM_MSI
2211         case KVM_SIGNAL_MSI: {
2212                 struct kvm_msi msi;
2213
2214                 r = -EFAULT;
2215                 if (copy_from_user(&msi, argp, sizeof msi))
2216                         goto out;
2217                 r = kvm_send_userspace_msi(kvm, &msi);
2218                 break;
2219         }
2220 #endif
2221 #ifdef __KVM_HAVE_IRQ_LINE
2222         case KVM_IRQ_LINE_STATUS:
2223         case KVM_IRQ_LINE: {
2224                 struct kvm_irq_level irq_event;
2225
2226                 r = -EFAULT;
2227                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2228                         goto out;
2229
2230                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2231                 if (r)
2232                         goto out;
2233
2234                 r = -EFAULT;
2235                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2236                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2237                                 goto out;
2238                 }
2239
2240                 r = 0;
2241                 break;
2242         }
2243 #endif
2244         default:
2245                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2246                 if (r == -ENOTTY)
2247                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2248         }
2249 out:
2250         return r;
2251 }
2252
2253 #ifdef CONFIG_COMPAT
2254 struct compat_kvm_dirty_log {
2255         __u32 slot;
2256         __u32 padding1;
2257         union {
2258                 compat_uptr_t dirty_bitmap; /* one bit per page */
2259                 __u64 padding2;
2260         };
2261 };
2262
2263 static long kvm_vm_compat_ioctl(struct file *filp,
2264                            unsigned int ioctl, unsigned long arg)
2265 {
2266         struct kvm *kvm = filp->private_data;
2267         int r;
2268
2269         if (kvm->mm != current->mm)
2270                 return -EIO;
2271         switch (ioctl) {
2272         case KVM_GET_DIRTY_LOG: {
2273                 struct compat_kvm_dirty_log compat_log;
2274                 struct kvm_dirty_log log;
2275
2276                 r = -EFAULT;
2277                 if (copy_from_user(&compat_log, (void __user *)arg,
2278                                    sizeof(compat_log)))
2279                         goto out;
2280                 log.slot         = compat_log.slot;
2281                 log.padding1     = compat_log.padding1;
2282                 log.padding2     = compat_log.padding2;
2283                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2284
2285                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2286                 break;
2287         }
2288         default:
2289                 r = kvm_vm_ioctl(filp, ioctl, arg);
2290         }
2291
2292 out:
2293         return r;
2294 }
2295 #endif
2296
2297 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2298 {
2299         struct page *page[1];
2300         unsigned long addr;
2301         int npages;
2302         gfn_t gfn = vmf->pgoff;
2303         struct kvm *kvm = vma->vm_file->private_data;
2304
2305         addr = gfn_to_hva(kvm, gfn);
2306         if (kvm_is_error_hva(addr))
2307                 return VM_FAULT_SIGBUS;
2308
2309         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2310                                 NULL);
2311         if (unlikely(npages != 1))
2312                 return VM_FAULT_SIGBUS;
2313
2314         vmf->page = page[0];
2315         return 0;
2316 }
2317
2318 static const struct vm_operations_struct kvm_vm_vm_ops = {
2319         .fault = kvm_vm_fault,
2320 };
2321
2322 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2323 {
2324         vma->vm_ops = &kvm_vm_vm_ops;
2325         return 0;
2326 }
2327
2328 static struct file_operations kvm_vm_fops = {
2329         .release        = kvm_vm_release,
2330         .unlocked_ioctl = kvm_vm_ioctl,
2331 #ifdef CONFIG_COMPAT
2332         .compat_ioctl   = kvm_vm_compat_ioctl,
2333 #endif
2334         .mmap           = kvm_vm_mmap,
2335         .llseek         = noop_llseek,
2336 };
2337
2338 static int kvm_dev_ioctl_create_vm(unsigned long type)
2339 {
2340         int r;
2341         struct kvm *kvm;
2342
2343         kvm = kvm_create_vm(type);
2344         if (IS_ERR(kvm))
2345                 return PTR_ERR(kvm);
2346 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2347         r = kvm_coalesced_mmio_init(kvm);
2348         if (r < 0) {
2349                 kvm_put_kvm(kvm);
2350                 return r;
2351         }
2352 #endif
2353         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2354         if (r < 0)
2355                 kvm_put_kvm(kvm);
2356
2357         return r;
2358 }
2359
2360 static long kvm_dev_ioctl_check_extension_generic(long arg)
2361 {
2362         switch (arg) {
2363         case KVM_CAP_USER_MEMORY:
2364         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2365         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2366 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2367         case KVM_CAP_SET_BOOT_CPU_ID:
2368 #endif
2369         case KVM_CAP_INTERNAL_ERROR_DATA:
2370 #ifdef CONFIG_HAVE_KVM_MSI
2371         case KVM_CAP_SIGNAL_MSI:
2372 #endif
2373                 return 1;
2374 #ifdef KVM_CAP_IRQ_ROUTING
2375         case KVM_CAP_IRQ_ROUTING:
2376                 return KVM_MAX_IRQ_ROUTES;
2377 #endif
2378         default:
2379                 break;
2380         }
2381         return kvm_dev_ioctl_check_extension(arg);
2382 }
2383
2384 static long kvm_dev_ioctl(struct file *filp,
2385                           unsigned int ioctl, unsigned long arg)
2386 {
2387         long r = -EINVAL;
2388
2389         switch (ioctl) {
2390         case KVM_GET_API_VERSION:
2391                 r = -EINVAL;
2392                 if (arg)
2393                         goto out;
2394                 r = KVM_API_VERSION;
2395                 break;
2396         case KVM_CREATE_VM:
2397                 r = kvm_dev_ioctl_create_vm(arg);
2398                 break;
2399         case KVM_CHECK_EXTENSION:
2400                 r = kvm_dev_ioctl_check_extension_generic(arg);
2401                 break;
2402         case KVM_GET_VCPU_MMAP_SIZE:
2403                 r = -EINVAL;
2404                 if (arg)
2405                         goto out;
2406                 r = PAGE_SIZE;     /* struct kvm_run */
2407 #ifdef CONFIG_X86
2408                 r += PAGE_SIZE;    /* pio data page */
2409 #endif
2410 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2411                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2412 #endif
2413                 break;
2414         case KVM_TRACE_ENABLE:
2415         case KVM_TRACE_PAUSE:
2416         case KVM_TRACE_DISABLE:
2417                 r = -EOPNOTSUPP;
2418                 break;
2419         default:
2420                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2421         }
2422 out:
2423         return r;
2424 }
2425
2426 static struct file_operations kvm_chardev_ops = {
2427         .unlocked_ioctl = kvm_dev_ioctl,
2428         .compat_ioctl   = kvm_dev_ioctl,
2429         .llseek         = noop_llseek,
2430 };
2431
2432 static struct miscdevice kvm_dev = {
2433         KVM_MINOR,
2434         "kvm",
2435         &kvm_chardev_ops,
2436 };
2437
2438 static void hardware_enable_nolock(void *junk)
2439 {
2440         int cpu = raw_smp_processor_id();
2441         int r;
2442
2443         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2444                 return;
2445
2446         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2447
2448         r = kvm_arch_hardware_enable(NULL);
2449
2450         if (r) {
2451                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2452                 atomic_inc(&hardware_enable_failed);
2453                 printk(KERN_INFO "kvm: enabling virtualization on "
2454                                  "CPU%d failed\n", cpu);
2455         }
2456 }
2457
2458 static void hardware_enable(void *junk)
2459 {
2460         raw_spin_lock(&kvm_lock);
2461         hardware_enable_nolock(junk);
2462         raw_spin_unlock(&kvm_lock);
2463 }
2464
2465 static void hardware_disable_nolock(void *junk)
2466 {
2467         int cpu = raw_smp_processor_id();
2468
2469         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2470                 return;
2471         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2472         kvm_arch_hardware_disable(NULL);
2473 }
2474
2475 static void hardware_disable(void *junk)
2476 {
2477         raw_spin_lock(&kvm_lock);
2478         hardware_disable_nolock(junk);
2479         raw_spin_unlock(&kvm_lock);
2480 }
2481
2482 static void hardware_disable_all_nolock(void)
2483 {
2484         BUG_ON(!kvm_usage_count);
2485
2486         kvm_usage_count--;
2487         if (!kvm_usage_count)
2488                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2489 }
2490
2491 static void hardware_disable_all(void)
2492 {
2493         raw_spin_lock(&kvm_lock);
2494         hardware_disable_all_nolock();
2495         raw_spin_unlock(&kvm_lock);
2496 }
2497
2498 static int hardware_enable_all(void)
2499 {
2500         int r = 0;
2501
2502         raw_spin_lock(&kvm_lock);
2503
2504         kvm_usage_count++;
2505         if (kvm_usage_count == 1) {
2506                 atomic_set(&hardware_enable_failed, 0);
2507                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2508
2509                 if (atomic_read(&hardware_enable_failed)) {
2510                         hardware_disable_all_nolock();
2511                         r = -EBUSY;
2512                 }
2513         }
2514
2515         raw_spin_unlock(&kvm_lock);
2516
2517         return r;
2518 }
2519
2520 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2521                            void *v)
2522 {
2523         int cpu = (long)v;
2524
2525         if (!kvm_usage_count)
2526                 return NOTIFY_OK;
2527
2528         val &= ~CPU_TASKS_FROZEN;
2529         switch (val) {
2530         case CPU_DYING:
2531                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2532                        cpu);
2533                 hardware_disable(NULL);
2534                 break;
2535         case CPU_STARTING:
2536                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2537                        cpu);
2538                 hardware_enable(NULL);
2539                 break;
2540         }
2541         return NOTIFY_OK;
2542 }
2543
2544
2545 asmlinkage void kvm_spurious_fault(void)
2546 {
2547         /* Fault while not rebooting.  We want the trace. */
2548         BUG();
2549 }
2550 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2551
2552 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2553                       void *v)
2554 {
2555         /*
2556          * Some (well, at least mine) BIOSes hang on reboot if
2557          * in vmx root mode.
2558          *
2559          * And Intel TXT required VMX off for all cpu when system shutdown.
2560          */
2561         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2562         kvm_rebooting = true;
2563         on_each_cpu(hardware_disable_nolock, NULL, 1);
2564         return NOTIFY_OK;
2565 }
2566
2567 static struct notifier_block kvm_reboot_notifier = {
2568         .notifier_call = kvm_reboot,
2569         .priority = 0,
2570 };
2571
2572 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2573 {
2574         int i;
2575
2576         for (i = 0; i < bus->dev_count; i++) {
2577                 struct kvm_io_device *pos = bus->range[i].dev;
2578
2579                 kvm_iodevice_destructor(pos);
2580         }
2581         kfree(bus);
2582 }
2583
2584 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2585 {
2586         const struct kvm_io_range *r1 = p1;
2587         const struct kvm_io_range *r2 = p2;
2588
2589         if (r1->addr < r2->addr)
2590                 return -1;
2591         if (r1->addr + r1->len > r2->addr + r2->len)
2592                 return 1;
2593         return 0;
2594 }
2595
2596 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2597                           gpa_t addr, int len)
2598 {
2599         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2600                 .addr = addr,
2601                 .len = len,
2602                 .dev = dev,
2603         };
2604
2605         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2606                 kvm_io_bus_sort_cmp, NULL);
2607
2608         return 0;
2609 }
2610
2611 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2612                              gpa_t addr, int len)
2613 {
2614         struct kvm_io_range *range, key;
2615         int off;
2616
2617         key = (struct kvm_io_range) {
2618                 .addr = addr,
2619                 .len = len,
2620         };
2621
2622         range = bsearch(&key, bus->range, bus->dev_count,
2623                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2624         if (range == NULL)
2625                 return -ENOENT;
2626
2627         off = range - bus->range;
2628
2629         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2630                 off--;
2631
2632         return off;
2633 }
2634
2635 /* kvm_io_bus_write - called under kvm->slots_lock */
2636 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2637                      int len, const void *val)
2638 {
2639         int idx;
2640         struct kvm_io_bus *bus;
2641         struct kvm_io_range range;
2642
2643         range = (struct kvm_io_range) {
2644                 .addr = addr,
2645                 .len = len,
2646         };
2647
2648         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2649         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2650         if (idx < 0)
2651                 return -EOPNOTSUPP;
2652
2653         while (idx < bus->dev_count &&
2654                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2655                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2656                         return 0;
2657                 idx++;
2658         }
2659
2660         return -EOPNOTSUPP;
2661 }
2662
2663 /* kvm_io_bus_read - called under kvm->slots_lock */
2664 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2665                     int len, void *val)
2666 {
2667         int idx;
2668         struct kvm_io_bus *bus;
2669         struct kvm_io_range range;
2670
2671         range = (struct kvm_io_range) {
2672                 .addr = addr,
2673                 .len = len,
2674         };
2675
2676         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2677         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2678         if (idx < 0)
2679                 return -EOPNOTSUPP;
2680
2681         while (idx < bus->dev_count &&
2682                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2683                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2684                         return 0;
2685                 idx++;
2686         }
2687
2688         return -EOPNOTSUPP;
2689 }
2690
2691 /* Caller must hold slots_lock. */
2692 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2693                             int len, struct kvm_io_device *dev)
2694 {
2695         struct kvm_io_bus *new_bus, *bus;
2696
2697         bus = kvm->buses[bus_idx];
2698         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2699                 return -ENOSPC;
2700
2701         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2702                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2703         if (!new_bus)
2704                 return -ENOMEM;
2705         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2706                sizeof(struct kvm_io_range)));
2707         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2708         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2709         synchronize_srcu_expedited(&kvm->srcu);
2710         kfree(bus);
2711
2712         return 0;
2713 }
2714
2715 /* Caller must hold slots_lock. */
2716 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2717                               struct kvm_io_device *dev)
2718 {
2719         int i, r;
2720         struct kvm_io_bus *new_bus, *bus;
2721
2722         bus = kvm->buses[bus_idx];
2723         r = -ENOENT;
2724         for (i = 0; i < bus->dev_count; i++)
2725                 if (bus->range[i].dev == dev) {
2726                         r = 0;
2727                         break;
2728                 }
2729
2730         if (r)
2731                 return r;
2732
2733         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2734                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2735         if (!new_bus)
2736                 return -ENOMEM;
2737
2738         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2739         new_bus->dev_count--;
2740         memcpy(new_bus->range + i, bus->range + i + 1,
2741                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2742
2743         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2744         synchronize_srcu_expedited(&kvm->srcu);
2745         kfree(bus);
2746         return r;
2747 }
2748
2749 static struct notifier_block kvm_cpu_notifier = {
2750         .notifier_call = kvm_cpu_hotplug,
2751 };
2752
2753 static int vm_stat_get(void *_offset, u64 *val)
2754 {
2755         unsigned offset = (long)_offset;
2756         struct kvm *kvm;
2757
2758         *val = 0;
2759         raw_spin_lock(&kvm_lock);
2760         list_for_each_entry(kvm, &vm_list, vm_list)
2761                 *val += *(u32 *)((void *)kvm + offset);
2762         raw_spin_unlock(&kvm_lock);
2763         return 0;
2764 }
2765
2766 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2767
2768 static int vcpu_stat_get(void *_offset, u64 *val)
2769 {
2770         unsigned offset = (long)_offset;
2771         struct kvm *kvm;
2772         struct kvm_vcpu *vcpu;
2773         int i;
2774
2775         *val = 0;
2776         raw_spin_lock(&kvm_lock);
2777         list_for_each_entry(kvm, &vm_list, vm_list)
2778                 kvm_for_each_vcpu(i, vcpu, kvm)
2779                         *val += *(u32 *)((void *)vcpu + offset);
2780
2781         raw_spin_unlock(&kvm_lock);
2782         return 0;
2783 }
2784
2785 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2786
2787 static const struct file_operations *stat_fops[] = {
2788         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2789         [KVM_STAT_VM]   = &vm_stat_fops,
2790 };
2791
2792 static int kvm_init_debug(void)
2793 {
2794         int r = -EFAULT;
2795         struct kvm_stats_debugfs_item *p;
2796
2797         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2798         if (kvm_debugfs_dir == NULL)
2799                 goto out;
2800
2801         for (p = debugfs_entries; p->name; ++p) {
2802                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2803                                                 (void *)(long)p->offset,
2804                                                 stat_fops[p->kind]);
2805                 if (p->dentry == NULL)
2806                         goto out_dir;
2807         }
2808
2809         return 0;
2810
2811 out_dir:
2812         debugfs_remove_recursive(kvm_debugfs_dir);
2813 out:
2814         return r;
2815 }
2816
2817 static void kvm_exit_debug(void)
2818 {
2819         struct kvm_stats_debugfs_item *p;
2820
2821         for (p = debugfs_entries; p->name; ++p)
2822                 debugfs_remove(p->dentry);
2823         debugfs_remove(kvm_debugfs_dir);
2824 }
2825
2826 static int kvm_suspend(void)
2827 {
2828         if (kvm_usage_count)
2829                 hardware_disable_nolock(NULL);
2830         return 0;
2831 }
2832
2833 static void kvm_resume(void)
2834 {
2835         if (kvm_usage_count) {
2836                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2837                 hardware_enable_nolock(NULL);
2838         }
2839 }
2840
2841 static struct syscore_ops kvm_syscore_ops = {
2842         .suspend = kvm_suspend,
2843         .resume = kvm_resume,
2844 };
2845
2846 static inline
2847 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2848 {
2849         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2850 }
2851
2852 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2853 {
2854         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2855
2856         kvm_arch_vcpu_load(vcpu, cpu);
2857 }
2858
2859 static void kvm_sched_out(struct preempt_notifier *pn,
2860                           struct task_struct *next)
2861 {
2862         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2863
2864         kvm_arch_vcpu_put(vcpu);
2865 }
2866
2867 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2868                   struct module *module)
2869 {
2870         int r;
2871         int cpu;
2872
2873         r = kvm_arch_init(opaque);
2874         if (r)
2875                 goto out_fail;
2876
2877         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2878                 r = -ENOMEM;
2879                 goto out_free_0;
2880         }
2881
2882         r = kvm_arch_hardware_setup();
2883         if (r < 0)
2884                 goto out_free_0a;
2885
2886         for_each_online_cpu(cpu) {
2887                 smp_call_function_single(cpu,
2888                                 kvm_arch_check_processor_compat,
2889                                 &r, 1);
2890                 if (r < 0)
2891                         goto out_free_1;
2892         }
2893
2894         r = register_cpu_notifier(&kvm_cpu_notifier);
2895         if (r)
2896                 goto out_free_2;
2897         register_reboot_notifier(&kvm_reboot_notifier);
2898
2899         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2900         if (!vcpu_align)
2901                 vcpu_align = __alignof__(struct kvm_vcpu);
2902         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2903                                            0, NULL);
2904         if (!kvm_vcpu_cache) {
2905                 r = -ENOMEM;
2906                 goto out_free_3;
2907         }
2908
2909         r = kvm_async_pf_init();
2910         if (r)
2911                 goto out_free;
2912
2913         kvm_chardev_ops.owner = module;
2914         kvm_vm_fops.owner = module;
2915         kvm_vcpu_fops.owner = module;
2916
2917         r = misc_register(&kvm_dev);
2918         if (r) {
2919                 printk(KERN_ERR "kvm: misc device register failed\n");
2920                 goto out_unreg;
2921         }
2922
2923         register_syscore_ops(&kvm_syscore_ops);
2924
2925         kvm_preempt_ops.sched_in = kvm_sched_in;
2926         kvm_preempt_ops.sched_out = kvm_sched_out;
2927
2928         r = kvm_init_debug();
2929         if (r) {
2930                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2931                 goto out_undebugfs;
2932         }
2933
2934         return 0;
2935
2936 out_undebugfs:
2937         unregister_syscore_ops(&kvm_syscore_ops);
2938 out_unreg:
2939         kvm_async_pf_deinit();
2940 out_free:
2941         kmem_cache_destroy(kvm_vcpu_cache);
2942 out_free_3:
2943         unregister_reboot_notifier(&kvm_reboot_notifier);
2944         unregister_cpu_notifier(&kvm_cpu_notifier);
2945 out_free_2:
2946 out_free_1:
2947         kvm_arch_hardware_unsetup();
2948 out_free_0a:
2949         free_cpumask_var(cpus_hardware_enabled);
2950 out_free_0:
2951         kvm_arch_exit();
2952 out_fail:
2953         return r;
2954 }
2955 EXPORT_SYMBOL_GPL(kvm_init);
2956
2957 void kvm_exit(void)
2958 {
2959         kvm_exit_debug();
2960         misc_deregister(&kvm_dev);
2961         kmem_cache_destroy(kvm_vcpu_cache);
2962         kvm_async_pf_deinit();
2963         unregister_syscore_ops(&kvm_syscore_ops);
2964         unregister_reboot_notifier(&kvm_reboot_notifier);
2965         unregister_cpu_notifier(&kvm_cpu_notifier);
2966         on_each_cpu(hardware_disable_nolock, NULL, 1);
2967         kvm_arch_hardware_unsetup();
2968         kvm_arch_exit();
2969         free_cpumask_var(cpus_hardware_enabled);
2970 }
2971 EXPORT_SYMBOL_GPL(kvm_exit);