]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - virt/kvm/kvm_main.c
KVM guest: exit idleness when handling KVM_PV_REASON_PAGE_NOT_PRESENT
[~shefty/rdma-dev.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 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
216 {
217         struct page *page;
218         int r;
219
220         mutex_init(&vcpu->mutex);
221         vcpu->cpu = -1;
222         vcpu->kvm = kvm;
223         vcpu->vcpu_id = id;
224         vcpu->pid = NULL;
225         init_waitqueue_head(&vcpu->wq);
226         kvm_async_pf_vcpu_init(vcpu);
227
228         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
229         if (!page) {
230                 r = -ENOMEM;
231                 goto fail;
232         }
233         vcpu->run = page_address(page);
234
235         kvm_vcpu_set_in_spin_loop(vcpu, false);
236         kvm_vcpu_set_dy_eligible(vcpu, false);
237
238         r = kvm_arch_vcpu_init(vcpu);
239         if (r < 0)
240                 goto fail_free_run;
241         return 0;
242
243 fail_free_run:
244         free_page((unsigned long)vcpu->run);
245 fail:
246         return r;
247 }
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 {
252         put_pid(vcpu->pid);
253         kvm_arch_vcpu_uninit(vcpu);
254         free_page((unsigned long)vcpu->run);
255 }
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 {
261         return container_of(mn, struct kvm, mmu_notifier);
262 }
263
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265                                              struct mm_struct *mm,
266                                              unsigned long address)
267 {
268         struct kvm *kvm = mmu_notifier_to_kvm(mn);
269         int need_tlb_flush, idx;
270
271         /*
272          * When ->invalidate_page runs, the linux pte has been zapped
273          * already but the page is still allocated until
274          * ->invalidate_page returns. So if we increase the sequence
275          * here the kvm page fault will notice if the spte can't be
276          * established because the page is going to be freed. If
277          * instead the kvm page fault establishes the spte before
278          * ->invalidate_page runs, kvm_unmap_hva will release it
279          * before returning.
280          *
281          * The sequence increase only need to be seen at spin_unlock
282          * time, and not at spin_lock time.
283          *
284          * Increasing the sequence after the spin_unlock would be
285          * unsafe because the kvm page fault could then establish the
286          * pte after kvm_unmap_hva returned, without noticing the page
287          * is going to be freed.
288          */
289         idx = srcu_read_lock(&kvm->srcu);
290         spin_lock(&kvm->mmu_lock);
291
292         kvm->mmu_notifier_seq++;
293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294         /* we've to flush the tlb before the pages can be freed */
295         if (need_tlb_flush)
296                 kvm_flush_remote_tlbs(kvm);
297
298         spin_unlock(&kvm->mmu_lock);
299         srcu_read_unlock(&kvm->srcu, idx);
300 }
301
302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303                                         struct mm_struct *mm,
304                                         unsigned long address,
305                                         pte_t pte)
306 {
307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
308         int idx;
309
310         idx = srcu_read_lock(&kvm->srcu);
311         spin_lock(&kvm->mmu_lock);
312         kvm->mmu_notifier_seq++;
313         kvm_set_spte_hva(kvm, address, pte);
314         spin_unlock(&kvm->mmu_lock);
315         srcu_read_unlock(&kvm->srcu, idx);
316 }
317
318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319                                                     struct mm_struct *mm,
320                                                     unsigned long start,
321                                                     unsigned long end)
322 {
323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
324         int need_tlb_flush = 0, idx;
325
326         idx = srcu_read_lock(&kvm->srcu);
327         spin_lock(&kvm->mmu_lock);
328         /*
329          * The count increase must become visible at unlock time as no
330          * spte can be established without taking the mmu_lock and
331          * count is also read inside the mmu_lock critical section.
332          */
333         kvm->mmu_notifier_count++;
334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335         need_tlb_flush |= kvm->tlbs_dirty;
336         /* we've to flush the tlb before the pages can be freed */
337         if (need_tlb_flush)
338                 kvm_flush_remote_tlbs(kvm);
339
340         spin_unlock(&kvm->mmu_lock);
341         srcu_read_unlock(&kvm->srcu, idx);
342 }
343
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345                                                   struct mm_struct *mm,
346                                                   unsigned long start,
347                                                   unsigned long end)
348 {
349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
350
351         spin_lock(&kvm->mmu_lock);
352         /*
353          * This sequence increase will notify the kvm page fault that
354          * the page that is going to be mapped in the spte could have
355          * been freed.
356          */
357         kvm->mmu_notifier_seq++;
358         smp_wmb();
359         /*
360          * The above sequence increase must be visible before the
361          * below count decrease, which is ensured by the smp_wmb above
362          * in conjunction with the smp_rmb in mmu_notifier_retry().
363          */
364         kvm->mmu_notifier_count--;
365         spin_unlock(&kvm->mmu_lock);
366
367         BUG_ON(kvm->mmu_notifier_count < 0);
368 }
369
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371                                               struct mm_struct *mm,
372                                               unsigned long address)
373 {
374         struct kvm *kvm = mmu_notifier_to_kvm(mn);
375         int young, idx;
376
377         idx = srcu_read_lock(&kvm->srcu);
378         spin_lock(&kvm->mmu_lock);
379
380         young = kvm_age_hva(kvm, address);
381         if (young)
382                 kvm_flush_remote_tlbs(kvm);
383
384         spin_unlock(&kvm->mmu_lock);
385         srcu_read_unlock(&kvm->srcu, idx);
386
387         return young;
388 }
389
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391                                        struct mm_struct *mm,
392                                        unsigned long address)
393 {
394         struct kvm *kvm = mmu_notifier_to_kvm(mn);
395         int young, idx;
396
397         idx = srcu_read_lock(&kvm->srcu);
398         spin_lock(&kvm->mmu_lock);
399         young = kvm_test_age_hva(kvm, address);
400         spin_unlock(&kvm->mmu_lock);
401         srcu_read_unlock(&kvm->srcu, idx);
402
403         return young;
404 }
405
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407                                      struct mm_struct *mm)
408 {
409         struct kvm *kvm = mmu_notifier_to_kvm(mn);
410         int idx;
411
412         idx = srcu_read_lock(&kvm->srcu);
413         kvm_arch_flush_shadow_all(kvm);
414         srcu_read_unlock(&kvm->srcu, idx);
415 }
416
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
419         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
421         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
422         .test_young             = kvm_mmu_notifier_test_young,
423         .change_pte             = kvm_mmu_notifier_change_pte,
424         .release                = kvm_mmu_notifier_release,
425 };
426
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
428 {
429         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
431 }
432
433 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
436 {
437         return 0;
438 }
439
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441
442 static void kvm_init_memslots_id(struct kvm *kvm)
443 {
444         int i;
445         struct kvm_memslots *slots = kvm->memslots;
446
447         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
448                 slots->id_to_index[i] = slots->memslots[i].id = i;
449 }
450
451 static struct kvm *kvm_create_vm(unsigned long type)
452 {
453         int r, i;
454         struct kvm *kvm = kvm_arch_alloc_vm();
455
456         if (!kvm)
457                 return ERR_PTR(-ENOMEM);
458
459         r = kvm_arch_init_vm(kvm, type);
460         if (r)
461                 goto out_err_nodisable;
462
463         r = hardware_enable_all();
464         if (r)
465                 goto out_err_nodisable;
466
467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
468         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
469         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
470 #endif
471
472         r = -ENOMEM;
473         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
474         if (!kvm->memslots)
475                 goto out_err_nosrcu;
476         kvm_init_memslots_id(kvm);
477         if (init_srcu_struct(&kvm->srcu))
478                 goto out_err_nosrcu;
479         for (i = 0; i < KVM_NR_BUSES; i++) {
480                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
481                                         GFP_KERNEL);
482                 if (!kvm->buses[i])
483                         goto out_err;
484         }
485
486         spin_lock_init(&kvm->mmu_lock);
487         kvm->mm = current->mm;
488         atomic_inc(&kvm->mm->mm_count);
489         kvm_eventfd_init(kvm);
490         mutex_init(&kvm->lock);
491         mutex_init(&kvm->irq_lock);
492         mutex_init(&kvm->slots_lock);
493         atomic_set(&kvm->users_count, 1);
494
495         r = kvm_init_mmu_notifier(kvm);
496         if (r)
497                 goto out_err;
498
499         raw_spin_lock(&kvm_lock);
500         list_add(&kvm->vm_list, &vm_list);
501         raw_spin_unlock(&kvm_lock);
502
503         return kvm;
504
505 out_err:
506         cleanup_srcu_struct(&kvm->srcu);
507 out_err_nosrcu:
508         hardware_disable_all();
509 out_err_nodisable:
510         for (i = 0; i < KVM_NR_BUSES; i++)
511                 kfree(kvm->buses[i]);
512         kfree(kvm->memslots);
513         kvm_arch_free_vm(kvm);
514         return ERR_PTR(r);
515 }
516
517 /*
518  * Avoid using vmalloc for a small buffer.
519  * Should not be used when the size is statically known.
520  */
521 void *kvm_kvzalloc(unsigned long size)
522 {
523         if (size > PAGE_SIZE)
524                 return vzalloc(size);
525         else
526                 return kzalloc(size, GFP_KERNEL);
527 }
528
529 void kvm_kvfree(const void *addr)
530 {
531         if (is_vmalloc_addr(addr))
532                 vfree(addr);
533         else
534                 kfree(addr);
535 }
536
537 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
538 {
539         if (!memslot->dirty_bitmap)
540                 return;
541
542         kvm_kvfree(memslot->dirty_bitmap);
543         memslot->dirty_bitmap = NULL;
544 }
545
546 /*
547  * Free any memory in @free but not in @dont.
548  */
549 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
550                                   struct kvm_memory_slot *dont)
551 {
552         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
553                 kvm_destroy_dirty_bitmap(free);
554
555         kvm_arch_free_memslot(free, dont);
556
557         free->npages = 0;
558 }
559
560 void kvm_free_physmem(struct kvm *kvm)
561 {
562         struct kvm_memslots *slots = kvm->memslots;
563         struct kvm_memory_slot *memslot;
564
565         kvm_for_each_memslot(memslot, slots)
566                 kvm_free_physmem_slot(memslot, NULL);
567
568         kfree(kvm->memslots);
569 }
570
571 static void kvm_destroy_vm(struct kvm *kvm)
572 {
573         int i;
574         struct mm_struct *mm = kvm->mm;
575
576         kvm_arch_sync_events(kvm);
577         raw_spin_lock(&kvm_lock);
578         list_del(&kvm->vm_list);
579         raw_spin_unlock(&kvm_lock);
580         kvm_free_irq_routing(kvm);
581         for (i = 0; i < KVM_NR_BUSES; i++)
582                 kvm_io_bus_destroy(kvm->buses[i]);
583         kvm_coalesced_mmio_free(kvm);
584 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
585         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
586 #else
587         kvm_arch_flush_shadow_all(kvm);
588 #endif
589         kvm_arch_destroy_vm(kvm);
590         kvm_free_physmem(kvm);
591         cleanup_srcu_struct(&kvm->srcu);
592         kvm_arch_free_vm(kvm);
593         hardware_disable_all();
594         mmdrop(mm);
595 }
596
597 void kvm_get_kvm(struct kvm *kvm)
598 {
599         atomic_inc(&kvm->users_count);
600 }
601 EXPORT_SYMBOL_GPL(kvm_get_kvm);
602
603 void kvm_put_kvm(struct kvm *kvm)
604 {
605         if (atomic_dec_and_test(&kvm->users_count))
606                 kvm_destroy_vm(kvm);
607 }
608 EXPORT_SYMBOL_GPL(kvm_put_kvm);
609
610
611 static int kvm_vm_release(struct inode *inode, struct file *filp)
612 {
613         struct kvm *kvm = filp->private_data;
614
615         kvm_irqfd_release(kvm);
616
617         kvm_put_kvm(kvm);
618         return 0;
619 }
620
621 /*
622  * Allocation size is twice as large as the actual dirty bitmap size.
623  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
624  */
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626 {
627 #ifndef CONFIG_S390
628         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
629
630         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
631         if (!memslot->dirty_bitmap)
632                 return -ENOMEM;
633
634 #endif /* !CONFIG_S390 */
635         return 0;
636 }
637
638 static int cmp_memslot(const void *slot1, const void *slot2)
639 {
640         struct kvm_memory_slot *s1, *s2;
641
642         s1 = (struct kvm_memory_slot *)slot1;
643         s2 = (struct kvm_memory_slot *)slot2;
644
645         if (s1->npages < s2->npages)
646                 return 1;
647         if (s1->npages > s2->npages)
648                 return -1;
649
650         return 0;
651 }
652
653 /*
654  * Sort the memslots base on its size, so the larger slots
655  * will get better fit.
656  */
657 static void sort_memslots(struct kvm_memslots *slots)
658 {
659         int i;
660
661         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
662               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
663
664         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
665                 slots->id_to_index[slots->memslots[i].id] = i;
666 }
667
668 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
669 {
670         if (new) {
671                 int id = new->id;
672                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
673                 unsigned long npages = old->npages;
674
675                 *old = *new;
676                 if (new->npages != npages)
677                         sort_memslots(slots);
678         }
679
680         slots->generation++;
681 }
682
683 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
684 {
685         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
686
687 #ifdef KVM_CAP_READONLY_MEM
688         valid_flags |= KVM_MEM_READONLY;
689 #endif
690
691         if (mem->flags & ~valid_flags)
692                 return -EINVAL;
693
694         return 0;
695 }
696
697 /*
698  * Allocate some memory and give it an address in the guest physical address
699  * space.
700  *
701  * Discontiguous memory is allowed, mostly for framebuffers.
702  *
703  * Must be called holding mmap_sem for write.
704  */
705 int __kvm_set_memory_region(struct kvm *kvm,
706                             struct kvm_userspace_memory_region *mem,
707                             int user_alloc)
708 {
709         int r;
710         gfn_t base_gfn;
711         unsigned long npages;
712         unsigned long i;
713         struct kvm_memory_slot *memslot;
714         struct kvm_memory_slot old, new;
715         struct kvm_memslots *slots, *old_memslots;
716
717         r = check_memory_region_flags(mem);
718         if (r)
719                 goto out;
720
721         r = -EINVAL;
722         /* General sanity checks */
723         if (mem->memory_size & (PAGE_SIZE - 1))
724                 goto out;
725         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
726                 goto out;
727         /* We can read the guest memory with __xxx_user() later on. */
728         if (user_alloc &&
729             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
730              !access_ok(VERIFY_WRITE,
731                         (void __user *)(unsigned long)mem->userspace_addr,
732                         mem->memory_size)))
733                 goto out;
734         if (mem->slot >= KVM_MEM_SLOTS_NUM)
735                 goto out;
736         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
737                 goto out;
738
739         memslot = id_to_memslot(kvm->memslots, mem->slot);
740         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
741         npages = mem->memory_size >> PAGE_SHIFT;
742
743         r = -EINVAL;
744         if (npages > KVM_MEM_MAX_NR_PAGES)
745                 goto out;
746
747         if (!npages)
748                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
749
750         new = old = *memslot;
751
752         new.id = mem->slot;
753         new.base_gfn = base_gfn;
754         new.npages = npages;
755         new.flags = mem->flags;
756
757         /* Disallow changing a memory slot's size. */
758         r = -EINVAL;
759         if (npages && old.npages && npages != old.npages)
760                 goto out_free;
761
762         /* Check for overlaps */
763         r = -EEXIST;
764         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
765                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
766
767                 if (s == memslot || !s->npages)
768                         continue;
769                 if (!((base_gfn + npages <= s->base_gfn) ||
770                       (base_gfn >= s->base_gfn + s->npages)))
771                         goto out_free;
772         }
773
774         /* Free page dirty bitmap if unneeded */
775         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
776                 new.dirty_bitmap = NULL;
777
778         r = -ENOMEM;
779
780         /* Allocate if a slot is being created */
781         if (npages && !old.npages) {
782                 new.user_alloc = user_alloc;
783                 new.userspace_addr = mem->userspace_addr;
784
785                 if (kvm_arch_create_memslot(&new, npages))
786                         goto out_free;
787         }
788
789         /* Allocate page dirty bitmap if needed */
790         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
791                 if (kvm_create_dirty_bitmap(&new) < 0)
792                         goto out_free;
793                 /* destroy any largepage mappings for dirty tracking */
794         }
795
796         if (!npages || base_gfn != old.base_gfn) {
797                 struct kvm_memory_slot *slot;
798
799                 r = -ENOMEM;
800                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
801                                 GFP_KERNEL);
802                 if (!slots)
803                         goto out_free;
804                 slot = id_to_memslot(slots, mem->slot);
805                 slot->flags |= KVM_MEMSLOT_INVALID;
806
807                 update_memslots(slots, NULL);
808
809                 old_memslots = kvm->memslots;
810                 rcu_assign_pointer(kvm->memslots, slots);
811                 synchronize_srcu_expedited(&kvm->srcu);
812                 /* From this point no new shadow pages pointing to a deleted,
813                  * or moved, memslot will be created.
814                  *
815                  * validation of sp->gfn happens in:
816                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
817                  *      - kvm_is_visible_gfn (mmu_check_roots)
818                  */
819                 kvm_arch_flush_shadow_memslot(kvm, slot);
820                 kfree(old_memslots);
821         }
822
823         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
824         if (r)
825                 goto out_free;
826
827         /* map/unmap the pages in iommu page table */
828         if (npages) {
829                 r = kvm_iommu_map_pages(kvm, &new);
830                 if (r)
831                         goto out_free;
832         } else
833                 kvm_iommu_unmap_pages(kvm, &old);
834
835         r = -ENOMEM;
836         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
837                         GFP_KERNEL);
838         if (!slots)
839                 goto out_free;
840
841         /* actual memory is freed via old in kvm_free_physmem_slot below */
842         if (!npages) {
843                 new.dirty_bitmap = NULL;
844                 memset(&new.arch, 0, sizeof(new.arch));
845         }
846
847         update_memslots(slots, &new);
848         old_memslots = kvm->memslots;
849         rcu_assign_pointer(kvm->memslots, slots);
850         synchronize_srcu_expedited(&kvm->srcu);
851
852         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
853
854         kvm_free_physmem_slot(&old, &new);
855         kfree(old_memslots);
856
857         return 0;
858
859 out_free:
860         kvm_free_physmem_slot(&new, &old);
861 out:
862         return r;
863
864 }
865 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
866
867 int kvm_set_memory_region(struct kvm *kvm,
868                           struct kvm_userspace_memory_region *mem,
869                           int user_alloc)
870 {
871         int r;
872
873         mutex_lock(&kvm->slots_lock);
874         r = __kvm_set_memory_region(kvm, mem, user_alloc);
875         mutex_unlock(&kvm->slots_lock);
876         return r;
877 }
878 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
879
880 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
881                                    struct
882                                    kvm_userspace_memory_region *mem,
883                                    int user_alloc)
884 {
885         if (mem->slot >= KVM_MEMORY_SLOTS)
886                 return -EINVAL;
887         return kvm_set_memory_region(kvm, mem, user_alloc);
888 }
889
890 int kvm_get_dirty_log(struct kvm *kvm,
891                         struct kvm_dirty_log *log, int *is_dirty)
892 {
893         struct kvm_memory_slot *memslot;
894         int r, i;
895         unsigned long n;
896         unsigned long any = 0;
897
898         r = -EINVAL;
899         if (log->slot >= KVM_MEMORY_SLOTS)
900                 goto out;
901
902         memslot = id_to_memslot(kvm->memslots, log->slot);
903         r = -ENOENT;
904         if (!memslot->dirty_bitmap)
905                 goto out;
906
907         n = kvm_dirty_bitmap_bytes(memslot);
908
909         for (i = 0; !any && i < n/sizeof(long); ++i)
910                 any = memslot->dirty_bitmap[i];
911
912         r = -EFAULT;
913         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
914                 goto out;
915
916         if (any)
917                 *is_dirty = 1;
918
919         r = 0;
920 out:
921         return r;
922 }
923
924 bool kvm_largepages_enabled(void)
925 {
926         return largepages_enabled;
927 }
928
929 void kvm_disable_largepages(void)
930 {
931         largepages_enabled = false;
932 }
933 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
934
935 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
936 {
937         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
938 }
939 EXPORT_SYMBOL_GPL(gfn_to_memslot);
940
941 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
942 {
943         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
944
945         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
946               memslot->flags & KVM_MEMSLOT_INVALID)
947                 return 0;
948
949         return 1;
950 }
951 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
952
953 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
954 {
955         struct vm_area_struct *vma;
956         unsigned long addr, size;
957
958         size = PAGE_SIZE;
959
960         addr = gfn_to_hva(kvm, gfn);
961         if (kvm_is_error_hva(addr))
962                 return PAGE_SIZE;
963
964         down_read(&current->mm->mmap_sem);
965         vma = find_vma(current->mm, addr);
966         if (!vma)
967                 goto out;
968
969         size = vma_kernel_pagesize(vma);
970
971 out:
972         up_read(&current->mm->mmap_sem);
973
974         return size;
975 }
976
977 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
978 {
979         return slot->flags & KVM_MEM_READONLY;
980 }
981
982 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
983                                        gfn_t *nr_pages, bool write)
984 {
985         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
986                 return KVM_HVA_ERR_BAD;
987
988         if (memslot_is_readonly(slot) && write)
989                 return KVM_HVA_ERR_RO_BAD;
990
991         if (nr_pages)
992                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
993
994         return __gfn_to_hva_memslot(slot, gfn);
995 }
996
997 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
998                                      gfn_t *nr_pages)
999 {
1000         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1001 }
1002
1003 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1004                                  gfn_t gfn)
1005 {
1006         return gfn_to_hva_many(slot, gfn, NULL);
1007 }
1008 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1009
1010 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1011 {
1012         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_hva);
1015
1016 /*
1017  * The hva returned by this function is only allowed to be read.
1018  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1019  */
1020 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1021 {
1022         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1023 }
1024
1025 static int kvm_read_hva(void *data, void __user *hva, int len)
1026 {
1027         return __copy_from_user(data, hva, len);
1028 }
1029
1030 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1031 {
1032         return __copy_from_user_inatomic(data, hva, len);
1033 }
1034
1035 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1036         unsigned long start, int write, struct page **page)
1037 {
1038         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1039
1040         if (write)
1041                 flags |= FOLL_WRITE;
1042
1043         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1044 }
1045
1046 static inline int check_user_page_hwpoison(unsigned long addr)
1047 {
1048         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1049
1050         rc = __get_user_pages(current, current->mm, addr, 1,
1051                               flags, NULL, NULL, NULL);
1052         return rc == -EHWPOISON;
1053 }
1054
1055 /*
1056  * The atomic path to get the writable pfn which will be stored in @pfn,
1057  * true indicates success, otherwise false is returned.
1058  */
1059 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1060                             bool write_fault, bool *writable, pfn_t *pfn)
1061 {
1062         struct page *page[1];
1063         int npages;
1064
1065         if (!(async || atomic))
1066                 return false;
1067
1068         /*
1069          * Fast pin a writable pfn only if it is a write fault request
1070          * or the caller allows to map a writable pfn for a read fault
1071          * request.
1072          */
1073         if (!(write_fault || writable))
1074                 return false;
1075
1076         npages = __get_user_pages_fast(addr, 1, 1, page);
1077         if (npages == 1) {
1078                 *pfn = page_to_pfn(page[0]);
1079
1080                 if (writable)
1081                         *writable = true;
1082                 return true;
1083         }
1084
1085         return false;
1086 }
1087
1088 /*
1089  * The slow path to get the pfn of the specified host virtual address,
1090  * 1 indicates success, -errno is returned if error is detected.
1091  */
1092 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1093                            bool *writable, pfn_t *pfn)
1094 {
1095         struct page *page[1];
1096         int npages = 0;
1097
1098         might_sleep();
1099
1100         if (writable)
1101                 *writable = write_fault;
1102
1103         if (async) {
1104                 down_read(&current->mm->mmap_sem);
1105                 npages = get_user_page_nowait(current, current->mm,
1106                                               addr, write_fault, page);
1107                 up_read(&current->mm->mmap_sem);
1108         } else
1109                 npages = get_user_pages_fast(addr, 1, write_fault,
1110                                              page);
1111         if (npages != 1)
1112                 return npages;
1113
1114         /* map read fault as writable if possible */
1115         if (unlikely(!write_fault) && writable) {
1116                 struct page *wpage[1];
1117
1118                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1119                 if (npages == 1) {
1120                         *writable = true;
1121                         put_page(page[0]);
1122                         page[0] = wpage[0];
1123                 }
1124
1125                 npages = 1;
1126         }
1127         *pfn = page_to_pfn(page[0]);
1128         return npages;
1129 }
1130
1131 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1132 {
1133         if (unlikely(!(vma->vm_flags & VM_READ)))
1134                 return false;
1135
1136         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1137                 return false;
1138
1139         return true;
1140 }
1141
1142 /*
1143  * Pin guest page in memory and return its pfn.
1144  * @addr: host virtual address which maps memory to the guest
1145  * @atomic: whether this function can sleep
1146  * @async: whether this function need to wait IO complete if the
1147  *         host page is not in the memory
1148  * @write_fault: whether we should get a writable host page
1149  * @writable: whether it allows to map a writable host page for !@write_fault
1150  *
1151  * The function will map a writable host page for these two cases:
1152  * 1): @write_fault = true
1153  * 2): @write_fault = false && @writable, @writable will tell the caller
1154  *     whether the mapping is writable.
1155  */
1156 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1157                         bool write_fault, bool *writable)
1158 {
1159         struct vm_area_struct *vma;
1160         pfn_t pfn = 0;
1161         int npages;
1162
1163         /* we can do it either atomically or asynchronously, not both */
1164         BUG_ON(atomic && async);
1165
1166         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1167                 return pfn;
1168
1169         if (atomic)
1170                 return KVM_PFN_ERR_FAULT;
1171
1172         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1173         if (npages == 1)
1174                 return pfn;
1175
1176         down_read(&current->mm->mmap_sem);
1177         if (npages == -EHWPOISON ||
1178               (!async && check_user_page_hwpoison(addr))) {
1179                 pfn = KVM_PFN_ERR_HWPOISON;
1180                 goto exit;
1181         }
1182
1183         vma = find_vma_intersection(current->mm, addr, addr + 1);
1184
1185         if (vma == NULL)
1186                 pfn = KVM_PFN_ERR_FAULT;
1187         else if ((vma->vm_flags & VM_PFNMAP)) {
1188                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1189                         vma->vm_pgoff;
1190                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1191         } else {
1192                 if (async && vma_is_valid(vma, write_fault))
1193                         *async = true;
1194                 pfn = KVM_PFN_ERR_FAULT;
1195         }
1196 exit:
1197         up_read(&current->mm->mmap_sem);
1198         return pfn;
1199 }
1200
1201 static pfn_t
1202 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1203                      bool *async, bool write_fault, bool *writable)
1204 {
1205         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1206
1207         if (addr == KVM_HVA_ERR_RO_BAD)
1208                 return KVM_PFN_ERR_RO_FAULT;
1209
1210         if (kvm_is_error_hva(addr))
1211                 return KVM_PFN_ERR_BAD;
1212
1213         /* Do not map writable pfn in the readonly memslot. */
1214         if (writable && memslot_is_readonly(slot)) {
1215                 *writable = false;
1216                 writable = NULL;
1217         }
1218
1219         return hva_to_pfn(addr, atomic, async, write_fault,
1220                           writable);
1221 }
1222
1223 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1224                           bool write_fault, bool *writable)
1225 {
1226         struct kvm_memory_slot *slot;
1227
1228         if (async)
1229                 *async = false;
1230
1231         slot = gfn_to_memslot(kvm, gfn);
1232
1233         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1234                                     writable);
1235 }
1236
1237 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1238 {
1239         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1240 }
1241 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1242
1243 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1244                        bool write_fault, bool *writable)
1245 {
1246         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1247 }
1248 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1249
1250 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1251 {
1252         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1253 }
1254 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1255
1256 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1257                       bool *writable)
1258 {
1259         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1260 }
1261 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1262
1263 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1264 {
1265         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1266 }
1267
1268 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1269 {
1270         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1271 }
1272 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1273
1274 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1275                                                                   int nr_pages)
1276 {
1277         unsigned long addr;
1278         gfn_t entry;
1279
1280         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1281         if (kvm_is_error_hva(addr))
1282                 return -1;
1283
1284         if (entry < nr_pages)
1285                 return 0;
1286
1287         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1288 }
1289 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1290
1291 static struct page *kvm_pfn_to_page(pfn_t pfn)
1292 {
1293         if (is_error_pfn(pfn))
1294                 return KVM_ERR_PTR_BAD_PAGE;
1295
1296         if (kvm_is_mmio_pfn(pfn)) {
1297                 WARN_ON(1);
1298                 return KVM_ERR_PTR_BAD_PAGE;
1299         }
1300
1301         return pfn_to_page(pfn);
1302 }
1303
1304 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1305 {
1306         pfn_t pfn;
1307
1308         pfn = gfn_to_pfn(kvm, gfn);
1309
1310         return kvm_pfn_to_page(pfn);
1311 }
1312
1313 EXPORT_SYMBOL_GPL(gfn_to_page);
1314
1315 void kvm_release_page_clean(struct page *page)
1316 {
1317         WARN_ON(is_error_page(page));
1318
1319         kvm_release_pfn_clean(page_to_pfn(page));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1322
1323 void kvm_release_pfn_clean(pfn_t pfn)
1324 {
1325         if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1326                 put_page(pfn_to_page(pfn));
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1329
1330 void kvm_release_page_dirty(struct page *page)
1331 {
1332         WARN_ON(is_error_page(page));
1333
1334         kvm_release_pfn_dirty(page_to_pfn(page));
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1337
1338 void kvm_release_pfn_dirty(pfn_t pfn)
1339 {
1340         kvm_set_pfn_dirty(pfn);
1341         kvm_release_pfn_clean(pfn);
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1344
1345 void kvm_set_page_dirty(struct page *page)
1346 {
1347         kvm_set_pfn_dirty(page_to_pfn(page));
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1350
1351 void kvm_set_pfn_dirty(pfn_t pfn)
1352 {
1353         if (!kvm_is_mmio_pfn(pfn)) {
1354                 struct page *page = pfn_to_page(pfn);
1355                 if (!PageReserved(page))
1356                         SetPageDirty(page);
1357         }
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1360
1361 void kvm_set_pfn_accessed(pfn_t pfn)
1362 {
1363         if (!kvm_is_mmio_pfn(pfn))
1364                 mark_page_accessed(pfn_to_page(pfn));
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1367
1368 void kvm_get_pfn(pfn_t pfn)
1369 {
1370         if (!kvm_is_mmio_pfn(pfn))
1371                 get_page(pfn_to_page(pfn));
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1374
1375 static int next_segment(unsigned long len, int offset)
1376 {
1377         if (len > PAGE_SIZE - offset)
1378                 return PAGE_SIZE - offset;
1379         else
1380                 return len;
1381 }
1382
1383 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1384                         int len)
1385 {
1386         int r;
1387         unsigned long addr;
1388
1389         addr = gfn_to_hva_read(kvm, gfn);
1390         if (kvm_is_error_hva(addr))
1391                 return -EFAULT;
1392         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1393         if (r)
1394                 return -EFAULT;
1395         return 0;
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1398
1399 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1400 {
1401         gfn_t gfn = gpa >> PAGE_SHIFT;
1402         int seg;
1403         int offset = offset_in_page(gpa);
1404         int ret;
1405
1406         while ((seg = next_segment(len, offset)) != 0) {
1407                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1408                 if (ret < 0)
1409                         return ret;
1410                 offset = 0;
1411                 len -= seg;
1412                 data += seg;
1413                 ++gfn;
1414         }
1415         return 0;
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_read_guest);
1418
1419 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1420                           unsigned long len)
1421 {
1422         int r;
1423         unsigned long addr;
1424         gfn_t gfn = gpa >> PAGE_SHIFT;
1425         int offset = offset_in_page(gpa);
1426
1427         addr = gfn_to_hva_read(kvm, gfn);
1428         if (kvm_is_error_hva(addr))
1429                 return -EFAULT;
1430         pagefault_disable();
1431         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1432         pagefault_enable();
1433         if (r)
1434                 return -EFAULT;
1435         return 0;
1436 }
1437 EXPORT_SYMBOL(kvm_read_guest_atomic);
1438
1439 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1440                          int offset, int len)
1441 {
1442         int r;
1443         unsigned long addr;
1444
1445         addr = gfn_to_hva(kvm, gfn);
1446         if (kvm_is_error_hva(addr))
1447                 return -EFAULT;
1448         r = __copy_to_user((void __user *)addr + offset, data, len);
1449         if (r)
1450                 return -EFAULT;
1451         mark_page_dirty(kvm, gfn);
1452         return 0;
1453 }
1454 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1455
1456 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1457                     unsigned long len)
1458 {
1459         gfn_t gfn = gpa >> PAGE_SHIFT;
1460         int seg;
1461         int offset = offset_in_page(gpa);
1462         int ret;
1463
1464         while ((seg = next_segment(len, offset)) != 0) {
1465                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1466                 if (ret < 0)
1467                         return ret;
1468                 offset = 0;
1469                 len -= seg;
1470                 data += seg;
1471                 ++gfn;
1472         }
1473         return 0;
1474 }
1475
1476 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1477                               gpa_t gpa)
1478 {
1479         struct kvm_memslots *slots = kvm_memslots(kvm);
1480         int offset = offset_in_page(gpa);
1481         gfn_t gfn = gpa >> PAGE_SHIFT;
1482
1483         ghc->gpa = gpa;
1484         ghc->generation = slots->generation;
1485         ghc->memslot = gfn_to_memslot(kvm, gfn);
1486         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1487         if (!kvm_is_error_hva(ghc->hva))
1488                 ghc->hva += offset;
1489         else
1490                 return -EFAULT;
1491
1492         return 0;
1493 }
1494 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1495
1496 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1497                            void *data, unsigned long len)
1498 {
1499         struct kvm_memslots *slots = kvm_memslots(kvm);
1500         int r;
1501
1502         if (slots->generation != ghc->generation)
1503                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1504
1505         if (kvm_is_error_hva(ghc->hva))
1506                 return -EFAULT;
1507
1508         r = __copy_to_user((void __user *)ghc->hva, data, len);
1509         if (r)
1510                 return -EFAULT;
1511         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1512
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1516
1517 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1518                            void *data, unsigned long len)
1519 {
1520         struct kvm_memslots *slots = kvm_memslots(kvm);
1521         int r;
1522
1523         if (slots->generation != ghc->generation)
1524                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1525
1526         if (kvm_is_error_hva(ghc->hva))
1527                 return -EFAULT;
1528
1529         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1530         if (r)
1531                 return -EFAULT;
1532
1533         return 0;
1534 }
1535 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1536
1537 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1538 {
1539         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1540                                     offset, len);
1541 }
1542 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1543
1544 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1545 {
1546         gfn_t gfn = gpa >> PAGE_SHIFT;
1547         int seg;
1548         int offset = offset_in_page(gpa);
1549         int ret;
1550
1551         while ((seg = next_segment(len, offset)) != 0) {
1552                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1553                 if (ret < 0)
1554                         return ret;
1555                 offset = 0;
1556                 len -= seg;
1557                 ++gfn;
1558         }
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1562
1563 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1564                              gfn_t gfn)
1565 {
1566         if (memslot && memslot->dirty_bitmap) {
1567                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1568
1569                 /* TODO: introduce set_bit_le() and use it */
1570                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1571         }
1572 }
1573
1574 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1575 {
1576         struct kvm_memory_slot *memslot;
1577
1578         memslot = gfn_to_memslot(kvm, gfn);
1579         mark_page_dirty_in_slot(kvm, memslot, gfn);
1580 }
1581
1582 /*
1583  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1584  */
1585 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1586 {
1587         DEFINE_WAIT(wait);
1588
1589         for (;;) {
1590                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1591
1592                 if (kvm_arch_vcpu_runnable(vcpu)) {
1593                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1594                         break;
1595                 }
1596                 if (kvm_cpu_has_pending_timer(vcpu))
1597                         break;
1598                 if (signal_pending(current))
1599                         break;
1600
1601                 schedule();
1602         }
1603
1604         finish_wait(&vcpu->wq, &wait);
1605 }
1606
1607 #ifndef CONFIG_S390
1608 /*
1609  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1610  */
1611 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1612 {
1613         int me;
1614         int cpu = vcpu->cpu;
1615         wait_queue_head_t *wqp;
1616
1617         wqp = kvm_arch_vcpu_wq(vcpu);
1618         if (waitqueue_active(wqp)) {
1619                 wake_up_interruptible(wqp);
1620                 ++vcpu->stat.halt_wakeup;
1621         }
1622
1623         me = get_cpu();
1624         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1625                 if (kvm_arch_vcpu_should_kick(vcpu))
1626                         smp_send_reschedule(cpu);
1627         put_cpu();
1628 }
1629 #endif /* !CONFIG_S390 */
1630
1631 void kvm_resched(struct kvm_vcpu *vcpu)
1632 {
1633         if (!need_resched())
1634                 return;
1635         cond_resched();
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_resched);
1638
1639 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1640 {
1641         struct pid *pid;
1642         struct task_struct *task = NULL;
1643
1644         rcu_read_lock();
1645         pid = rcu_dereference(target->pid);
1646         if (pid)
1647                 task = get_pid_task(target->pid, PIDTYPE_PID);
1648         rcu_read_unlock();
1649         if (!task)
1650                 return false;
1651         if (task->flags & PF_VCPU) {
1652                 put_task_struct(task);
1653                 return false;
1654         }
1655         if (yield_to(task, 1)) {
1656                 put_task_struct(task);
1657                 return true;
1658         }
1659         put_task_struct(task);
1660         return false;
1661 }
1662 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1663
1664 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1665 /*
1666  * Helper that checks whether a VCPU is eligible for directed yield.
1667  * Most eligible candidate to yield is decided by following heuristics:
1668  *
1669  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1670  *  (preempted lock holder), indicated by @in_spin_loop.
1671  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1672  *
1673  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1674  *  chance last time (mostly it has become eligible now since we have probably
1675  *  yielded to lockholder in last iteration. This is done by toggling
1676  *  @dy_eligible each time a VCPU checked for eligibility.)
1677  *
1678  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1679  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1680  *  burning. Giving priority for a potential lock-holder increases lock
1681  *  progress.
1682  *
1683  *  Since algorithm is based on heuristics, accessing another VCPU data without
1684  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1685  *  and continue with next VCPU and so on.
1686  */
1687 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1688 {
1689         bool eligible;
1690
1691         eligible = !vcpu->spin_loop.in_spin_loop ||
1692                         (vcpu->spin_loop.in_spin_loop &&
1693                          vcpu->spin_loop.dy_eligible);
1694
1695         if (vcpu->spin_loop.in_spin_loop)
1696                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1697
1698         return eligible;
1699 }
1700 #endif
1701 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1702 {
1703         struct kvm *kvm = me->kvm;
1704         struct kvm_vcpu *vcpu;
1705         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1706         int yielded = 0;
1707         int pass;
1708         int i;
1709
1710         kvm_vcpu_set_in_spin_loop(me, true);
1711         /*
1712          * We boost the priority of a VCPU that is runnable but not
1713          * currently running, because it got preempted by something
1714          * else and called schedule in __vcpu_run.  Hopefully that
1715          * VCPU is holding the lock that we need and will release it.
1716          * We approximate round-robin by starting at the last boosted VCPU.
1717          */
1718         for (pass = 0; pass < 2 && !yielded; pass++) {
1719                 kvm_for_each_vcpu(i, vcpu, kvm) {
1720                         if (!pass && i <= last_boosted_vcpu) {
1721                                 i = last_boosted_vcpu;
1722                                 continue;
1723                         } else if (pass && i > last_boosted_vcpu)
1724                                 break;
1725                         if (vcpu == me)
1726                                 continue;
1727                         if (waitqueue_active(&vcpu->wq))
1728                                 continue;
1729                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1730                                 continue;
1731                         if (kvm_vcpu_yield_to(vcpu)) {
1732                                 kvm->last_boosted_vcpu = i;
1733                                 yielded = 1;
1734                                 break;
1735                         }
1736                 }
1737         }
1738         kvm_vcpu_set_in_spin_loop(me, false);
1739
1740         /* Ensure vcpu is not eligible during next spinloop */
1741         kvm_vcpu_set_dy_eligible(me, false);
1742 }
1743 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1744
1745 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1746 {
1747         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1748         struct page *page;
1749
1750         if (vmf->pgoff == 0)
1751                 page = virt_to_page(vcpu->run);
1752 #ifdef CONFIG_X86
1753         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1754                 page = virt_to_page(vcpu->arch.pio_data);
1755 #endif
1756 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1757         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1758                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1759 #endif
1760         else
1761                 return kvm_arch_vcpu_fault(vcpu, vmf);
1762         get_page(page);
1763         vmf->page = page;
1764         return 0;
1765 }
1766
1767 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1768         .fault = kvm_vcpu_fault,
1769 };
1770
1771 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1772 {
1773         vma->vm_ops = &kvm_vcpu_vm_ops;
1774         return 0;
1775 }
1776
1777 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1778 {
1779         struct kvm_vcpu *vcpu = filp->private_data;
1780
1781         kvm_put_kvm(vcpu->kvm);
1782         return 0;
1783 }
1784
1785 static struct file_operations kvm_vcpu_fops = {
1786         .release        = kvm_vcpu_release,
1787         .unlocked_ioctl = kvm_vcpu_ioctl,
1788 #ifdef CONFIG_COMPAT
1789         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1790 #endif
1791         .mmap           = kvm_vcpu_mmap,
1792         .llseek         = noop_llseek,
1793 };
1794
1795 /*
1796  * Allocates an inode for the vcpu.
1797  */
1798 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1799 {
1800         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1801 }
1802
1803 /*
1804  * Creates some virtual cpus.  Good luck creating more than one.
1805  */
1806 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1807 {
1808         int r;
1809         struct kvm_vcpu *vcpu, *v;
1810
1811         vcpu = kvm_arch_vcpu_create(kvm, id);
1812         if (IS_ERR(vcpu))
1813                 return PTR_ERR(vcpu);
1814
1815         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1816
1817         r = kvm_arch_vcpu_setup(vcpu);
1818         if (r)
1819                 goto vcpu_destroy;
1820
1821         mutex_lock(&kvm->lock);
1822         if (!kvm_vcpu_compatible(vcpu)) {
1823                 r = -EINVAL;
1824                 goto unlock_vcpu_destroy;
1825         }
1826         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1827                 r = -EINVAL;
1828                 goto unlock_vcpu_destroy;
1829         }
1830
1831         kvm_for_each_vcpu(r, v, kvm)
1832                 if (v->vcpu_id == id) {
1833                         r = -EEXIST;
1834                         goto unlock_vcpu_destroy;
1835                 }
1836
1837         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1838
1839         /* Now it's all set up, let userspace reach it */
1840         kvm_get_kvm(kvm);
1841         r = create_vcpu_fd(vcpu);
1842         if (r < 0) {
1843                 kvm_put_kvm(kvm);
1844                 goto unlock_vcpu_destroy;
1845         }
1846
1847         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1848         smp_wmb();
1849         atomic_inc(&kvm->online_vcpus);
1850
1851         mutex_unlock(&kvm->lock);
1852         return r;
1853
1854 unlock_vcpu_destroy:
1855         mutex_unlock(&kvm->lock);
1856 vcpu_destroy:
1857         kvm_arch_vcpu_destroy(vcpu);
1858         return r;
1859 }
1860
1861 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1862 {
1863         if (sigset) {
1864                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1865                 vcpu->sigset_active = 1;
1866                 vcpu->sigset = *sigset;
1867         } else
1868                 vcpu->sigset_active = 0;
1869         return 0;
1870 }
1871
1872 static long kvm_vcpu_ioctl(struct file *filp,
1873                            unsigned int ioctl, unsigned long arg)
1874 {
1875         struct kvm_vcpu *vcpu = filp->private_data;
1876         void __user *argp = (void __user *)arg;
1877         int r;
1878         struct kvm_fpu *fpu = NULL;
1879         struct kvm_sregs *kvm_sregs = NULL;
1880
1881         if (vcpu->kvm->mm != current->mm)
1882                 return -EIO;
1883
1884 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1885         /*
1886          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1887          * so vcpu_load() would break it.
1888          */
1889         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1890                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1891 #endif
1892
1893
1894         r = vcpu_load(vcpu);
1895         if (r)
1896                 return r;
1897         switch (ioctl) {
1898         case KVM_RUN:
1899                 r = -EINVAL;
1900                 if (arg)
1901                         goto out;
1902                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1903                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1904                 break;
1905         case KVM_GET_REGS: {
1906                 struct kvm_regs *kvm_regs;
1907
1908                 r = -ENOMEM;
1909                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1910                 if (!kvm_regs)
1911                         goto out;
1912                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1913                 if (r)
1914                         goto out_free1;
1915                 r = -EFAULT;
1916                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1917                         goto out_free1;
1918                 r = 0;
1919 out_free1:
1920                 kfree(kvm_regs);
1921                 break;
1922         }
1923         case KVM_SET_REGS: {
1924                 struct kvm_regs *kvm_regs;
1925
1926                 r = -ENOMEM;
1927                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1928                 if (IS_ERR(kvm_regs)) {
1929                         r = PTR_ERR(kvm_regs);
1930                         goto out;
1931                 }
1932                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1933                 if (r)
1934                         goto out_free2;
1935                 r = 0;
1936 out_free2:
1937                 kfree(kvm_regs);
1938                 break;
1939         }
1940         case KVM_GET_SREGS: {
1941                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1942                 r = -ENOMEM;
1943                 if (!kvm_sregs)
1944                         goto out;
1945                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1946                 if (r)
1947                         goto out;
1948                 r = -EFAULT;
1949                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1950                         goto out;
1951                 r = 0;
1952                 break;
1953         }
1954         case KVM_SET_SREGS: {
1955                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1956                 if (IS_ERR(kvm_sregs)) {
1957                         r = PTR_ERR(kvm_sregs);
1958                         goto out;
1959                 }
1960                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1961                 if (r)
1962                         goto out;
1963                 r = 0;
1964                 break;
1965         }
1966         case KVM_GET_MP_STATE: {
1967                 struct kvm_mp_state mp_state;
1968
1969                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1970                 if (r)
1971                         goto out;
1972                 r = -EFAULT;
1973                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1974                         goto out;
1975                 r = 0;
1976                 break;
1977         }
1978         case KVM_SET_MP_STATE: {
1979                 struct kvm_mp_state mp_state;
1980
1981                 r = -EFAULT;
1982                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1983                         goto out;
1984                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1985                 if (r)
1986                         goto out;
1987                 r = 0;
1988                 break;
1989         }
1990         case KVM_TRANSLATE: {
1991                 struct kvm_translation tr;
1992
1993                 r = -EFAULT;
1994                 if (copy_from_user(&tr, argp, sizeof tr))
1995                         goto out;
1996                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1997                 if (r)
1998                         goto out;
1999                 r = -EFAULT;
2000                 if (copy_to_user(argp, &tr, sizeof tr))
2001                         goto out;
2002                 r = 0;
2003                 break;
2004         }
2005         case KVM_SET_GUEST_DEBUG: {
2006                 struct kvm_guest_debug dbg;
2007
2008                 r = -EFAULT;
2009                 if (copy_from_user(&dbg, argp, sizeof dbg))
2010                         goto out;
2011                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2012                 if (r)
2013                         goto out;
2014                 r = 0;
2015                 break;
2016         }
2017         case KVM_SET_SIGNAL_MASK: {
2018                 struct kvm_signal_mask __user *sigmask_arg = argp;
2019                 struct kvm_signal_mask kvm_sigmask;
2020                 sigset_t sigset, *p;
2021
2022                 p = NULL;
2023                 if (argp) {
2024                         r = -EFAULT;
2025                         if (copy_from_user(&kvm_sigmask, argp,
2026                                            sizeof kvm_sigmask))
2027                                 goto out;
2028                         r = -EINVAL;
2029                         if (kvm_sigmask.len != sizeof sigset)
2030                                 goto out;
2031                         r = -EFAULT;
2032                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2033                                            sizeof sigset))
2034                                 goto out;
2035                         p = &sigset;
2036                 }
2037                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2038                 break;
2039         }
2040         case KVM_GET_FPU: {
2041                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2042                 r = -ENOMEM;
2043                 if (!fpu)
2044                         goto out;
2045                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2046                 if (r)
2047                         goto out;
2048                 r = -EFAULT;
2049                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2050                         goto out;
2051                 r = 0;
2052                 break;
2053         }
2054         case KVM_SET_FPU: {
2055                 fpu = memdup_user(argp, sizeof(*fpu));
2056                 if (IS_ERR(fpu)) {
2057                         r = PTR_ERR(fpu);
2058                         goto out;
2059                 }
2060                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2061                 if (r)
2062                         goto out;
2063                 r = 0;
2064                 break;
2065         }
2066         default:
2067                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2068         }
2069 out:
2070         vcpu_put(vcpu);
2071         kfree(fpu);
2072         kfree(kvm_sregs);
2073         return r;
2074 }
2075
2076 #ifdef CONFIG_COMPAT
2077 static long kvm_vcpu_compat_ioctl(struct file *filp,
2078                                   unsigned int ioctl, unsigned long arg)
2079 {
2080         struct kvm_vcpu *vcpu = filp->private_data;
2081         void __user *argp = compat_ptr(arg);
2082         int r;
2083
2084         if (vcpu->kvm->mm != current->mm)
2085                 return -EIO;
2086
2087         switch (ioctl) {
2088         case KVM_SET_SIGNAL_MASK: {
2089                 struct kvm_signal_mask __user *sigmask_arg = argp;
2090                 struct kvm_signal_mask kvm_sigmask;
2091                 compat_sigset_t csigset;
2092                 sigset_t sigset;
2093
2094                 if (argp) {
2095                         r = -EFAULT;
2096                         if (copy_from_user(&kvm_sigmask, argp,
2097                                            sizeof kvm_sigmask))
2098                                 goto out;
2099                         r = -EINVAL;
2100                         if (kvm_sigmask.len != sizeof csigset)
2101                                 goto out;
2102                         r = -EFAULT;
2103                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2104                                            sizeof csigset))
2105                                 goto out;
2106                         sigset_from_compat(&sigset, &csigset);
2107                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2108                 } else
2109                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2110                 break;
2111         }
2112         default:
2113                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2114         }
2115
2116 out:
2117         return r;
2118 }
2119 #endif
2120
2121 static long kvm_vm_ioctl(struct file *filp,
2122                            unsigned int ioctl, unsigned long arg)
2123 {
2124         struct kvm *kvm = filp->private_data;
2125         void __user *argp = (void __user *)arg;
2126         int r;
2127
2128         if (kvm->mm != current->mm)
2129                 return -EIO;
2130         switch (ioctl) {
2131         case KVM_CREATE_VCPU:
2132                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2133                 if (r < 0)
2134                         goto out;
2135                 break;
2136         case KVM_SET_USER_MEMORY_REGION: {
2137                 struct kvm_userspace_memory_region kvm_userspace_mem;
2138
2139                 r = -EFAULT;
2140                 if (copy_from_user(&kvm_userspace_mem, argp,
2141                                                 sizeof kvm_userspace_mem))
2142                         goto out;
2143
2144                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2145                 if (r)
2146                         goto out;
2147                 break;
2148         }
2149         case KVM_GET_DIRTY_LOG: {
2150                 struct kvm_dirty_log log;
2151
2152                 r = -EFAULT;
2153                 if (copy_from_user(&log, argp, sizeof log))
2154                         goto out;
2155                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2156                 if (r)
2157                         goto out;
2158                 break;
2159         }
2160 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2161         case KVM_REGISTER_COALESCED_MMIO: {
2162                 struct kvm_coalesced_mmio_zone zone;
2163                 r = -EFAULT;
2164                 if (copy_from_user(&zone, argp, sizeof zone))
2165                         goto out;
2166                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2167                 if (r)
2168                         goto out;
2169                 r = 0;
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                 if (r)
2179                         goto out;
2180                 r = 0;
2181                 break;
2182         }
2183 #endif
2184         case KVM_IRQFD: {
2185                 struct kvm_irqfd data;
2186
2187                 r = -EFAULT;
2188                 if (copy_from_user(&data, argp, sizeof data))
2189                         goto out;
2190                 r = kvm_irqfd(kvm, &data);
2191                 break;
2192         }
2193         case KVM_IOEVENTFD: {
2194                 struct kvm_ioeventfd data;
2195
2196                 r = -EFAULT;
2197                 if (copy_from_user(&data, argp, sizeof data))
2198                         goto out;
2199                 r = kvm_ioeventfd(kvm, &data);
2200                 break;
2201         }
2202 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2203         case KVM_SET_BOOT_CPU_ID:
2204                 r = 0;
2205                 mutex_lock(&kvm->lock);
2206                 if (atomic_read(&kvm->online_vcpus) != 0)
2207                         r = -EBUSY;
2208                 else
2209                         kvm->bsp_vcpu_id = arg;
2210                 mutex_unlock(&kvm->lock);
2211                 break;
2212 #endif
2213 #ifdef CONFIG_HAVE_KVM_MSI
2214         case KVM_SIGNAL_MSI: {
2215                 struct kvm_msi msi;
2216
2217                 r = -EFAULT;
2218                 if (copy_from_user(&msi, argp, sizeof msi))
2219                         goto out;
2220                 r = kvm_send_userspace_msi(kvm, &msi);
2221                 break;
2222         }
2223 #endif
2224 #ifdef __KVM_HAVE_IRQ_LINE
2225         case KVM_IRQ_LINE_STATUS:
2226         case KVM_IRQ_LINE: {
2227                 struct kvm_irq_level irq_event;
2228
2229                 r = -EFAULT;
2230                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2231                         goto out;
2232
2233                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2234                 if (r)
2235                         goto out;
2236
2237                 r = -EFAULT;
2238                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2239                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2240                                 goto out;
2241                 }
2242
2243                 r = 0;
2244                 break;
2245         }
2246 #endif
2247         default:
2248                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2249                 if (r == -ENOTTY)
2250                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2251         }
2252 out:
2253         return r;
2254 }
2255
2256 #ifdef CONFIG_COMPAT
2257 struct compat_kvm_dirty_log {
2258         __u32 slot;
2259         __u32 padding1;
2260         union {
2261                 compat_uptr_t dirty_bitmap; /* one bit per page */
2262                 __u64 padding2;
2263         };
2264 };
2265
2266 static long kvm_vm_compat_ioctl(struct file *filp,
2267                            unsigned int ioctl, unsigned long arg)
2268 {
2269         struct kvm *kvm = filp->private_data;
2270         int r;
2271
2272         if (kvm->mm != current->mm)
2273                 return -EIO;
2274         switch (ioctl) {
2275         case KVM_GET_DIRTY_LOG: {
2276                 struct compat_kvm_dirty_log compat_log;
2277                 struct kvm_dirty_log log;
2278
2279                 r = -EFAULT;
2280                 if (copy_from_user(&compat_log, (void __user *)arg,
2281                                    sizeof(compat_log)))
2282                         goto out;
2283                 log.slot         = compat_log.slot;
2284                 log.padding1     = compat_log.padding1;
2285                 log.padding2     = compat_log.padding2;
2286                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2287
2288                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2289                 if (r)
2290                         goto out;
2291                 break;
2292         }
2293         default:
2294                 r = kvm_vm_ioctl(filp, ioctl, arg);
2295         }
2296
2297 out:
2298         return r;
2299 }
2300 #endif
2301
2302 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2303 {
2304         struct page *page[1];
2305         unsigned long addr;
2306         int npages;
2307         gfn_t gfn = vmf->pgoff;
2308         struct kvm *kvm = vma->vm_file->private_data;
2309
2310         addr = gfn_to_hva(kvm, gfn);
2311         if (kvm_is_error_hva(addr))
2312                 return VM_FAULT_SIGBUS;
2313
2314         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2315                                 NULL);
2316         if (unlikely(npages != 1))
2317                 return VM_FAULT_SIGBUS;
2318
2319         vmf->page = page[0];
2320         return 0;
2321 }
2322
2323 static const struct vm_operations_struct kvm_vm_vm_ops = {
2324         .fault = kvm_vm_fault,
2325 };
2326
2327 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2328 {
2329         vma->vm_ops = &kvm_vm_vm_ops;
2330         return 0;
2331 }
2332
2333 static struct file_operations kvm_vm_fops = {
2334         .release        = kvm_vm_release,
2335         .unlocked_ioctl = kvm_vm_ioctl,
2336 #ifdef CONFIG_COMPAT
2337         .compat_ioctl   = kvm_vm_compat_ioctl,
2338 #endif
2339         .mmap           = kvm_vm_mmap,
2340         .llseek         = noop_llseek,
2341 };
2342
2343 static int kvm_dev_ioctl_create_vm(unsigned long type)
2344 {
2345         int r;
2346         struct kvm *kvm;
2347
2348         kvm = kvm_create_vm(type);
2349         if (IS_ERR(kvm))
2350                 return PTR_ERR(kvm);
2351 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2352         r = kvm_coalesced_mmio_init(kvm);
2353         if (r < 0) {
2354                 kvm_put_kvm(kvm);
2355                 return r;
2356         }
2357 #endif
2358         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2359         if (r < 0)
2360                 kvm_put_kvm(kvm);
2361
2362         return r;
2363 }
2364
2365 static long kvm_dev_ioctl_check_extension_generic(long arg)
2366 {
2367         switch (arg) {
2368         case KVM_CAP_USER_MEMORY:
2369         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2370         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2371 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2372         case KVM_CAP_SET_BOOT_CPU_ID:
2373 #endif
2374         case KVM_CAP_INTERNAL_ERROR_DATA:
2375 #ifdef CONFIG_HAVE_KVM_MSI
2376         case KVM_CAP_SIGNAL_MSI:
2377 #endif
2378                 return 1;
2379 #ifdef KVM_CAP_IRQ_ROUTING
2380         case KVM_CAP_IRQ_ROUTING:
2381                 return KVM_MAX_IRQ_ROUTES;
2382 #endif
2383         default:
2384                 break;
2385         }
2386         return kvm_dev_ioctl_check_extension(arg);
2387 }
2388
2389 static long kvm_dev_ioctl(struct file *filp,
2390                           unsigned int ioctl, unsigned long arg)
2391 {
2392         long r = -EINVAL;
2393
2394         switch (ioctl) {
2395         case KVM_GET_API_VERSION:
2396                 r = -EINVAL;
2397                 if (arg)
2398                         goto out;
2399                 r = KVM_API_VERSION;
2400                 break;
2401         case KVM_CREATE_VM:
2402                 r = kvm_dev_ioctl_create_vm(arg);
2403                 break;
2404         case KVM_CHECK_EXTENSION:
2405                 r = kvm_dev_ioctl_check_extension_generic(arg);
2406                 break;
2407         case KVM_GET_VCPU_MMAP_SIZE:
2408                 r = -EINVAL;
2409                 if (arg)
2410                         goto out;
2411                 r = PAGE_SIZE;     /* struct kvm_run */
2412 #ifdef CONFIG_X86
2413                 r += PAGE_SIZE;    /* pio data page */
2414 #endif
2415 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2416                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2417 #endif
2418                 break;
2419         case KVM_TRACE_ENABLE:
2420         case KVM_TRACE_PAUSE:
2421         case KVM_TRACE_DISABLE:
2422                 r = -EOPNOTSUPP;
2423                 break;
2424         default:
2425                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2426         }
2427 out:
2428         return r;
2429 }
2430
2431 static struct file_operations kvm_chardev_ops = {
2432         .unlocked_ioctl = kvm_dev_ioctl,
2433         .compat_ioctl   = kvm_dev_ioctl,
2434         .llseek         = noop_llseek,
2435 };
2436
2437 static struct miscdevice kvm_dev = {
2438         KVM_MINOR,
2439         "kvm",
2440         &kvm_chardev_ops,
2441 };
2442
2443 static void hardware_enable_nolock(void *junk)
2444 {
2445         int cpu = raw_smp_processor_id();
2446         int r;
2447
2448         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2449                 return;
2450
2451         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2452
2453         r = kvm_arch_hardware_enable(NULL);
2454
2455         if (r) {
2456                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2457                 atomic_inc(&hardware_enable_failed);
2458                 printk(KERN_INFO "kvm: enabling virtualization on "
2459                                  "CPU%d failed\n", cpu);
2460         }
2461 }
2462
2463 static void hardware_enable(void *junk)
2464 {
2465         raw_spin_lock(&kvm_lock);
2466         hardware_enable_nolock(junk);
2467         raw_spin_unlock(&kvm_lock);
2468 }
2469
2470 static void hardware_disable_nolock(void *junk)
2471 {
2472         int cpu = raw_smp_processor_id();
2473
2474         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2475                 return;
2476         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2477         kvm_arch_hardware_disable(NULL);
2478 }
2479
2480 static void hardware_disable(void *junk)
2481 {
2482         raw_spin_lock(&kvm_lock);
2483         hardware_disable_nolock(junk);
2484         raw_spin_unlock(&kvm_lock);
2485 }
2486
2487 static void hardware_disable_all_nolock(void)
2488 {
2489         BUG_ON(!kvm_usage_count);
2490
2491         kvm_usage_count--;
2492         if (!kvm_usage_count)
2493                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2494 }
2495
2496 static void hardware_disable_all(void)
2497 {
2498         raw_spin_lock(&kvm_lock);
2499         hardware_disable_all_nolock();
2500         raw_spin_unlock(&kvm_lock);
2501 }
2502
2503 static int hardware_enable_all(void)
2504 {
2505         int r = 0;
2506
2507         raw_spin_lock(&kvm_lock);
2508
2509         kvm_usage_count++;
2510         if (kvm_usage_count == 1) {
2511                 atomic_set(&hardware_enable_failed, 0);
2512                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2513
2514                 if (atomic_read(&hardware_enable_failed)) {
2515                         hardware_disable_all_nolock();
2516                         r = -EBUSY;
2517                 }
2518         }
2519
2520         raw_spin_unlock(&kvm_lock);
2521
2522         return r;
2523 }
2524
2525 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2526                            void *v)
2527 {
2528         int cpu = (long)v;
2529
2530         if (!kvm_usage_count)
2531                 return NOTIFY_OK;
2532
2533         val &= ~CPU_TASKS_FROZEN;
2534         switch (val) {
2535         case CPU_DYING:
2536                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2537                        cpu);
2538                 hardware_disable(NULL);
2539                 break;
2540         case CPU_STARTING:
2541                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2542                        cpu);
2543                 hardware_enable(NULL);
2544                 break;
2545         }
2546         return NOTIFY_OK;
2547 }
2548
2549
2550 asmlinkage void kvm_spurious_fault(void)
2551 {
2552         /* Fault while not rebooting.  We want the trace. */
2553         BUG();
2554 }
2555 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2556
2557 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2558                       void *v)
2559 {
2560         /*
2561          * Some (well, at least mine) BIOSes hang on reboot if
2562          * in vmx root mode.
2563          *
2564          * And Intel TXT required VMX off for all cpu when system shutdown.
2565          */
2566         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2567         kvm_rebooting = true;
2568         on_each_cpu(hardware_disable_nolock, NULL, 1);
2569         return NOTIFY_OK;
2570 }
2571
2572 static struct notifier_block kvm_reboot_notifier = {
2573         .notifier_call = kvm_reboot,
2574         .priority = 0,
2575 };
2576
2577 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2578 {
2579         int i;
2580
2581         for (i = 0; i < bus->dev_count; i++) {
2582                 struct kvm_io_device *pos = bus->range[i].dev;
2583
2584                 kvm_iodevice_destructor(pos);
2585         }
2586         kfree(bus);
2587 }
2588
2589 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2590 {
2591         const struct kvm_io_range *r1 = p1;
2592         const struct kvm_io_range *r2 = p2;
2593
2594         if (r1->addr < r2->addr)
2595                 return -1;
2596         if (r1->addr + r1->len > r2->addr + r2->len)
2597                 return 1;
2598         return 0;
2599 }
2600
2601 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2602                           gpa_t addr, int len)
2603 {
2604         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2605                 .addr = addr,
2606                 .len = len,
2607                 .dev = dev,
2608         };
2609
2610         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2611                 kvm_io_bus_sort_cmp, NULL);
2612
2613         return 0;
2614 }
2615
2616 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2617                              gpa_t addr, int len)
2618 {
2619         struct kvm_io_range *range, key;
2620         int off;
2621
2622         key = (struct kvm_io_range) {
2623                 .addr = addr,
2624                 .len = len,
2625         };
2626
2627         range = bsearch(&key, bus->range, bus->dev_count,
2628                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2629         if (range == NULL)
2630                 return -ENOENT;
2631
2632         off = range - bus->range;
2633
2634         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2635                 off--;
2636
2637         return off;
2638 }
2639
2640 /* kvm_io_bus_write - called under kvm->slots_lock */
2641 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2642                      int len, const void *val)
2643 {
2644         int idx;
2645         struct kvm_io_bus *bus;
2646         struct kvm_io_range range;
2647
2648         range = (struct kvm_io_range) {
2649                 .addr = addr,
2650                 .len = len,
2651         };
2652
2653         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2654         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2655         if (idx < 0)
2656                 return -EOPNOTSUPP;
2657
2658         while (idx < bus->dev_count &&
2659                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2660                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2661                         return 0;
2662                 idx++;
2663         }
2664
2665         return -EOPNOTSUPP;
2666 }
2667
2668 /* kvm_io_bus_read - called under kvm->slots_lock */
2669 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2670                     int len, void *val)
2671 {
2672         int idx;
2673         struct kvm_io_bus *bus;
2674         struct kvm_io_range range;
2675
2676         range = (struct kvm_io_range) {
2677                 .addr = addr,
2678                 .len = len,
2679         };
2680
2681         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2682         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2683         if (idx < 0)
2684                 return -EOPNOTSUPP;
2685
2686         while (idx < bus->dev_count &&
2687                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2688                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2689                         return 0;
2690                 idx++;
2691         }
2692
2693         return -EOPNOTSUPP;
2694 }
2695
2696 /* Caller must hold slots_lock. */
2697 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2698                             int len, struct kvm_io_device *dev)
2699 {
2700         struct kvm_io_bus *new_bus, *bus;
2701
2702         bus = kvm->buses[bus_idx];
2703         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2704                 return -ENOSPC;
2705
2706         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2707                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2708         if (!new_bus)
2709                 return -ENOMEM;
2710         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2711                sizeof(struct kvm_io_range)));
2712         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2713         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2714         synchronize_srcu_expedited(&kvm->srcu);
2715         kfree(bus);
2716
2717         return 0;
2718 }
2719
2720 /* Caller must hold slots_lock. */
2721 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2722                               struct kvm_io_device *dev)
2723 {
2724         int i, r;
2725         struct kvm_io_bus *new_bus, *bus;
2726
2727         bus = kvm->buses[bus_idx];
2728         r = -ENOENT;
2729         for (i = 0; i < bus->dev_count; i++)
2730                 if (bus->range[i].dev == dev) {
2731                         r = 0;
2732                         break;
2733                 }
2734
2735         if (r)
2736                 return r;
2737
2738         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2739                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2740         if (!new_bus)
2741                 return -ENOMEM;
2742
2743         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2744         new_bus->dev_count--;
2745         memcpy(new_bus->range + i, bus->range + i + 1,
2746                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2747
2748         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2749         synchronize_srcu_expedited(&kvm->srcu);
2750         kfree(bus);
2751         return r;
2752 }
2753
2754 static struct notifier_block kvm_cpu_notifier = {
2755         .notifier_call = kvm_cpu_hotplug,
2756 };
2757
2758 static int vm_stat_get(void *_offset, u64 *val)
2759 {
2760         unsigned offset = (long)_offset;
2761         struct kvm *kvm;
2762
2763         *val = 0;
2764         raw_spin_lock(&kvm_lock);
2765         list_for_each_entry(kvm, &vm_list, vm_list)
2766                 *val += *(u32 *)((void *)kvm + offset);
2767         raw_spin_unlock(&kvm_lock);
2768         return 0;
2769 }
2770
2771 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2772
2773 static int vcpu_stat_get(void *_offset, u64 *val)
2774 {
2775         unsigned offset = (long)_offset;
2776         struct kvm *kvm;
2777         struct kvm_vcpu *vcpu;
2778         int i;
2779
2780         *val = 0;
2781         raw_spin_lock(&kvm_lock);
2782         list_for_each_entry(kvm, &vm_list, vm_list)
2783                 kvm_for_each_vcpu(i, vcpu, kvm)
2784                         *val += *(u32 *)((void *)vcpu + offset);
2785
2786         raw_spin_unlock(&kvm_lock);
2787         return 0;
2788 }
2789
2790 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2791
2792 static const struct file_operations *stat_fops[] = {
2793         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2794         [KVM_STAT_VM]   = &vm_stat_fops,
2795 };
2796
2797 static int kvm_init_debug(void)
2798 {
2799         int r = -EFAULT;
2800         struct kvm_stats_debugfs_item *p;
2801
2802         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2803         if (kvm_debugfs_dir == NULL)
2804                 goto out;
2805
2806         for (p = debugfs_entries; p->name; ++p) {
2807                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2808                                                 (void *)(long)p->offset,
2809                                                 stat_fops[p->kind]);
2810                 if (p->dentry == NULL)
2811                         goto out_dir;
2812         }
2813
2814         return 0;
2815
2816 out_dir:
2817         debugfs_remove_recursive(kvm_debugfs_dir);
2818 out:
2819         return r;
2820 }
2821
2822 static void kvm_exit_debug(void)
2823 {
2824         struct kvm_stats_debugfs_item *p;
2825
2826         for (p = debugfs_entries; p->name; ++p)
2827                 debugfs_remove(p->dentry);
2828         debugfs_remove(kvm_debugfs_dir);
2829 }
2830
2831 static int kvm_suspend(void)
2832 {
2833         if (kvm_usage_count)
2834                 hardware_disable_nolock(NULL);
2835         return 0;
2836 }
2837
2838 static void kvm_resume(void)
2839 {
2840         if (kvm_usage_count) {
2841                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2842                 hardware_enable_nolock(NULL);
2843         }
2844 }
2845
2846 static struct syscore_ops kvm_syscore_ops = {
2847         .suspend = kvm_suspend,
2848         .resume = kvm_resume,
2849 };
2850
2851 static inline
2852 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2853 {
2854         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2855 }
2856
2857 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2858 {
2859         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2860
2861         kvm_arch_vcpu_load(vcpu, cpu);
2862 }
2863
2864 static void kvm_sched_out(struct preempt_notifier *pn,
2865                           struct task_struct *next)
2866 {
2867         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2868
2869         kvm_arch_vcpu_put(vcpu);
2870 }
2871
2872 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2873                   struct module *module)
2874 {
2875         int r;
2876         int cpu;
2877
2878         r = kvm_arch_init(opaque);
2879         if (r)
2880                 goto out_fail;
2881
2882         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2883                 r = -ENOMEM;
2884                 goto out_free_0;
2885         }
2886
2887         r = kvm_arch_hardware_setup();
2888         if (r < 0)
2889                 goto out_free_0a;
2890
2891         for_each_online_cpu(cpu) {
2892                 smp_call_function_single(cpu,
2893                                 kvm_arch_check_processor_compat,
2894                                 &r, 1);
2895                 if (r < 0)
2896                         goto out_free_1;
2897         }
2898
2899         r = register_cpu_notifier(&kvm_cpu_notifier);
2900         if (r)
2901                 goto out_free_2;
2902         register_reboot_notifier(&kvm_reboot_notifier);
2903
2904         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2905         if (!vcpu_align)
2906                 vcpu_align = __alignof__(struct kvm_vcpu);
2907         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2908                                            0, NULL);
2909         if (!kvm_vcpu_cache) {
2910                 r = -ENOMEM;
2911                 goto out_free_3;
2912         }
2913
2914         r = kvm_async_pf_init();
2915         if (r)
2916                 goto out_free;
2917
2918         kvm_chardev_ops.owner = module;
2919         kvm_vm_fops.owner = module;
2920         kvm_vcpu_fops.owner = module;
2921
2922         r = misc_register(&kvm_dev);
2923         if (r) {
2924                 printk(KERN_ERR "kvm: misc device register failed\n");
2925                 goto out_unreg;
2926         }
2927
2928         register_syscore_ops(&kvm_syscore_ops);
2929
2930         kvm_preempt_ops.sched_in = kvm_sched_in;
2931         kvm_preempt_ops.sched_out = kvm_sched_out;
2932
2933         r = kvm_init_debug();
2934         if (r) {
2935                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2936                 goto out_undebugfs;
2937         }
2938
2939         return 0;
2940
2941 out_undebugfs:
2942         unregister_syscore_ops(&kvm_syscore_ops);
2943 out_unreg:
2944         kvm_async_pf_deinit();
2945 out_free:
2946         kmem_cache_destroy(kvm_vcpu_cache);
2947 out_free_3:
2948         unregister_reboot_notifier(&kvm_reboot_notifier);
2949         unregister_cpu_notifier(&kvm_cpu_notifier);
2950 out_free_2:
2951 out_free_1:
2952         kvm_arch_hardware_unsetup();
2953 out_free_0a:
2954         free_cpumask_var(cpus_hardware_enabled);
2955 out_free_0:
2956         kvm_arch_exit();
2957 out_fail:
2958         return r;
2959 }
2960 EXPORT_SYMBOL_GPL(kvm_init);
2961
2962 void kvm_exit(void)
2963 {
2964         kvm_exit_debug();
2965         misc_deregister(&kvm_dev);
2966         kmem_cache_destroy(kvm_vcpu_cache);
2967         kvm_async_pf_deinit();
2968         unregister_syscore_ops(&kvm_syscore_ops);
2969         unregister_reboot_notifier(&kvm_reboot_notifier);
2970         unregister_cpu_notifier(&kvm_cpu_notifier);
2971         on_each_cpu(hardware_disable_nolock, NULL, 1);
2972         kvm_arch_hardware_unsetup();
2973         kvm_arch_exit();
2974         free_cpumask_var(cpus_hardware_enabled);
2975 }
2976 EXPORT_SYMBOL_GPL(kvm_exit);