a83ca63d26fcdb89b6ebcba0e2fc40eaeba08cd4
[linux-3.10.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 bool kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 int vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         if (mutex_lock_killable(&vcpu->mutex))
139                 return -EINTR;
140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141                 /* The thread running this VCPU changed. */
142                 struct pid *oldpid = vcpu->pid;
143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144                 rcu_assign_pointer(vcpu->pid, newpid);
145                 synchronize_rcu();
146                 put_pid(oldpid);
147         }
148         cpu = get_cpu();
149         preempt_notifier_register(&vcpu->preempt_notifier);
150         kvm_arch_vcpu_load(vcpu, cpu);
151         put_cpu();
152         return 0;
153 }
154
155 void vcpu_put(struct kvm_vcpu *vcpu)
156 {
157         preempt_disable();
158         kvm_arch_vcpu_put(vcpu);
159         preempt_notifier_unregister(&vcpu->preempt_notifier);
160         preempt_enable();
161         mutex_unlock(&vcpu->mutex);
162 }
163
164 static void ack_flush(void *_completed)
165 {
166 }
167
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
169 {
170         int i, cpu, me;
171         cpumask_var_t cpus;
172         bool called = true;
173         struct kvm_vcpu *vcpu;
174
175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176
177         me = get_cpu();
178         kvm_for_each_vcpu(i, vcpu, kvm) {
179                 kvm_make_request(req, vcpu);
180                 cpu = vcpu->cpu;
181
182                 /* Set ->requests bit before we read ->mode */
183                 smp_mb();
184
185                 if (cpus != NULL && cpu != -1 && cpu != me &&
186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187                         cpumask_set_cpu(cpu, cpus);
188         }
189         if (unlikely(cpus == NULL))
190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191         else if (!cpumask_empty(cpus))
192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
193         else
194                 called = false;
195         put_cpu();
196         free_cpumask_var(cpus);
197         return called;
198 }
199
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
201 {
202         long dirty_count = kvm->tlbs_dirty;
203
204         smp_mb();
205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206                 ++kvm->stat.remote_tlb_flush;
207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 }
209
210 void kvm_reload_remote_mmus(struct kvm *kvm)
211 {
212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 }
214
215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216 {
217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218 }
219
220 void kvm_make_update_eoibitmap_request(struct kvm *kvm)
221 {
222         make_all_cpus_request(kvm, KVM_REQ_EOIBITMAP);
223 }
224
225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
226 {
227         struct page *page;
228         int r;
229
230         mutex_init(&vcpu->mutex);
231         vcpu->cpu = -1;
232         vcpu->kvm = kvm;
233         vcpu->vcpu_id = id;
234         vcpu->pid = NULL;
235         init_waitqueue_head(&vcpu->wq);
236         kvm_async_pf_vcpu_init(vcpu);
237
238         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
239         if (!page) {
240                 r = -ENOMEM;
241                 goto fail;
242         }
243         vcpu->run = page_address(page);
244
245         kvm_vcpu_set_in_spin_loop(vcpu, false);
246         kvm_vcpu_set_dy_eligible(vcpu, false);
247
248         r = kvm_arch_vcpu_init(vcpu);
249         if (r < 0)
250                 goto fail_free_run;
251         return 0;
252
253 fail_free_run:
254         free_page((unsigned long)vcpu->run);
255 fail:
256         return r;
257 }
258 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
259
260 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
261 {
262         put_pid(vcpu->pid);
263         kvm_arch_vcpu_uninit(vcpu);
264         free_page((unsigned long)vcpu->run);
265 }
266 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
267
268 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
269 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
270 {
271         return container_of(mn, struct kvm, mmu_notifier);
272 }
273
274 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
275                                              struct mm_struct *mm,
276                                              unsigned long address)
277 {
278         struct kvm *kvm = mmu_notifier_to_kvm(mn);
279         int need_tlb_flush, idx;
280
281         /*
282          * When ->invalidate_page runs, the linux pte has been zapped
283          * already but the page is still allocated until
284          * ->invalidate_page returns. So if we increase the sequence
285          * here the kvm page fault will notice if the spte can't be
286          * established because the page is going to be freed. If
287          * instead the kvm page fault establishes the spte before
288          * ->invalidate_page runs, kvm_unmap_hva will release it
289          * before returning.
290          *
291          * The sequence increase only need to be seen at spin_unlock
292          * time, and not at spin_lock time.
293          *
294          * Increasing the sequence after the spin_unlock would be
295          * unsafe because the kvm page fault could then establish the
296          * pte after kvm_unmap_hva returned, without noticing the page
297          * is going to be freed.
298          */
299         idx = srcu_read_lock(&kvm->srcu);
300         spin_lock(&kvm->mmu_lock);
301
302         kvm->mmu_notifier_seq++;
303         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
304         /* we've to flush the tlb before the pages can be freed */
305         if (need_tlb_flush)
306                 kvm_flush_remote_tlbs(kvm);
307
308         spin_unlock(&kvm->mmu_lock);
309         srcu_read_unlock(&kvm->srcu, idx);
310 }
311
312 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
313                                         struct mm_struct *mm,
314                                         unsigned long address,
315                                         pte_t pte)
316 {
317         struct kvm *kvm = mmu_notifier_to_kvm(mn);
318         int idx;
319
320         idx = srcu_read_lock(&kvm->srcu);
321         spin_lock(&kvm->mmu_lock);
322         kvm->mmu_notifier_seq++;
323         kvm_set_spte_hva(kvm, address, pte);
324         spin_unlock(&kvm->mmu_lock);
325         srcu_read_unlock(&kvm->srcu, idx);
326 }
327
328 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
329                                                     struct mm_struct *mm,
330                                                     unsigned long start,
331                                                     unsigned long end)
332 {
333         struct kvm *kvm = mmu_notifier_to_kvm(mn);
334         int need_tlb_flush = 0, idx;
335
336         idx = srcu_read_lock(&kvm->srcu);
337         spin_lock(&kvm->mmu_lock);
338         /*
339          * The count increase must become visible at unlock time as no
340          * spte can be established without taking the mmu_lock and
341          * count is also read inside the mmu_lock critical section.
342          */
343         kvm->mmu_notifier_count++;
344         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
345         need_tlb_flush |= kvm->tlbs_dirty;
346         /* we've to flush the tlb before the pages can be freed */
347         if (need_tlb_flush)
348                 kvm_flush_remote_tlbs(kvm);
349
350         spin_unlock(&kvm->mmu_lock);
351         srcu_read_unlock(&kvm->srcu, idx);
352 }
353
354 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
355                                                   struct mm_struct *mm,
356                                                   unsigned long start,
357                                                   unsigned long end)
358 {
359         struct kvm *kvm = mmu_notifier_to_kvm(mn);
360
361         spin_lock(&kvm->mmu_lock);
362         /*
363          * This sequence increase will notify the kvm page fault that
364          * the page that is going to be mapped in the spte could have
365          * been freed.
366          */
367         kvm->mmu_notifier_seq++;
368         smp_wmb();
369         /*
370          * The above sequence increase must be visible before the
371          * below count decrease, which is ensured by the smp_wmb above
372          * in conjunction with the smp_rmb in mmu_notifier_retry().
373          */
374         kvm->mmu_notifier_count--;
375         spin_unlock(&kvm->mmu_lock);
376
377         BUG_ON(kvm->mmu_notifier_count < 0);
378 }
379
380 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
381                                               struct mm_struct *mm,
382                                               unsigned long address)
383 {
384         struct kvm *kvm = mmu_notifier_to_kvm(mn);
385         int young, idx;
386
387         idx = srcu_read_lock(&kvm->srcu);
388         spin_lock(&kvm->mmu_lock);
389
390         young = kvm_age_hva(kvm, address);
391         if (young)
392                 kvm_flush_remote_tlbs(kvm);
393
394         spin_unlock(&kvm->mmu_lock);
395         srcu_read_unlock(&kvm->srcu, idx);
396
397         return young;
398 }
399
400 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
401                                        struct mm_struct *mm,
402                                        unsigned long address)
403 {
404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
405         int young, idx;
406
407         idx = srcu_read_lock(&kvm->srcu);
408         spin_lock(&kvm->mmu_lock);
409         young = kvm_test_age_hva(kvm, address);
410         spin_unlock(&kvm->mmu_lock);
411         srcu_read_unlock(&kvm->srcu, idx);
412
413         return young;
414 }
415
416 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
417                                      struct mm_struct *mm)
418 {
419         struct kvm *kvm = mmu_notifier_to_kvm(mn);
420         int idx;
421
422         idx = srcu_read_lock(&kvm->srcu);
423         kvm_arch_flush_shadow_all(kvm);
424         srcu_read_unlock(&kvm->srcu, idx);
425 }
426
427 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
428         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
429         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
430         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
431         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
432         .test_young             = kvm_mmu_notifier_test_young,
433         .change_pte             = kvm_mmu_notifier_change_pte,
434         .release                = kvm_mmu_notifier_release,
435 };
436
437 static int kvm_init_mmu_notifier(struct kvm *kvm)
438 {
439         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
440         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
441 }
442
443 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
444
445 static int kvm_init_mmu_notifier(struct kvm *kvm)
446 {
447         return 0;
448 }
449
450 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
451
452 static void kvm_init_memslots_id(struct kvm *kvm)
453 {
454         int i;
455         struct kvm_memslots *slots = kvm->memslots;
456
457         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
458                 slots->id_to_index[i] = slots->memslots[i].id = i;
459 }
460
461 static struct kvm *kvm_create_vm(unsigned long type)
462 {
463         int r, i;
464         struct kvm *kvm = kvm_arch_alloc_vm();
465
466         if (!kvm)
467                 return ERR_PTR(-ENOMEM);
468
469         r = kvm_arch_init_vm(kvm, type);
470         if (r)
471                 goto out_err_nodisable;
472
473         r = hardware_enable_all();
474         if (r)
475                 goto out_err_nodisable;
476
477 #ifdef CONFIG_HAVE_KVM_IRQCHIP
478         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
479         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
480 #endif
481
482         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
483
484         r = -ENOMEM;
485         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
486         if (!kvm->memslots)
487                 goto out_err_nosrcu;
488         kvm_init_memslots_id(kvm);
489         if (init_srcu_struct(&kvm->srcu))
490                 goto out_err_nosrcu;
491         for (i = 0; i < KVM_NR_BUSES; i++) {
492                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
493                                         GFP_KERNEL);
494                 if (!kvm->buses[i])
495                         goto out_err;
496         }
497
498         spin_lock_init(&kvm->mmu_lock);
499         kvm->mm = current->mm;
500         atomic_inc(&kvm->mm->mm_count);
501         kvm_eventfd_init(kvm);
502         mutex_init(&kvm->lock);
503         mutex_init(&kvm->irq_lock);
504         mutex_init(&kvm->slots_lock);
505         atomic_set(&kvm->users_count, 1);
506
507         r = kvm_init_mmu_notifier(kvm);
508         if (r)
509                 goto out_err;
510
511         raw_spin_lock(&kvm_lock);
512         list_add(&kvm->vm_list, &vm_list);
513         raw_spin_unlock(&kvm_lock);
514
515         return kvm;
516
517 out_err:
518         cleanup_srcu_struct(&kvm->srcu);
519 out_err_nosrcu:
520         hardware_disable_all();
521 out_err_nodisable:
522         for (i = 0; i < KVM_NR_BUSES; i++)
523                 kfree(kvm->buses[i]);
524         kfree(kvm->memslots);
525         kvm_arch_free_vm(kvm);
526         return ERR_PTR(r);
527 }
528
529 /*
530  * Avoid using vmalloc for a small buffer.
531  * Should not be used when the size is statically known.
532  */
533 void *kvm_kvzalloc(unsigned long size)
534 {
535         if (size > PAGE_SIZE)
536                 return vzalloc(size);
537         else
538                 return kzalloc(size, GFP_KERNEL);
539 }
540
541 void kvm_kvfree(const void *addr)
542 {
543         if (is_vmalloc_addr(addr))
544                 vfree(addr);
545         else
546                 kfree(addr);
547 }
548
549 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
550 {
551         if (!memslot->dirty_bitmap)
552                 return;
553
554         kvm_kvfree(memslot->dirty_bitmap);
555         memslot->dirty_bitmap = NULL;
556 }
557
558 /*
559  * Free any memory in @free but not in @dont.
560  */
561 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
562                                   struct kvm_memory_slot *dont)
563 {
564         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
565                 kvm_destroy_dirty_bitmap(free);
566
567         kvm_arch_free_memslot(free, dont);
568
569         free->npages = 0;
570 }
571
572 void kvm_free_physmem(struct kvm *kvm)
573 {
574         struct kvm_memslots *slots = kvm->memslots;
575         struct kvm_memory_slot *memslot;
576
577         kvm_for_each_memslot(memslot, slots)
578                 kvm_free_physmem_slot(memslot, NULL);
579
580         kfree(kvm->memslots);
581 }
582
583 static void kvm_destroy_vm(struct kvm *kvm)
584 {
585         int i;
586         struct mm_struct *mm = kvm->mm;
587
588         kvm_arch_sync_events(kvm);
589         raw_spin_lock(&kvm_lock);
590         list_del(&kvm->vm_list);
591         raw_spin_unlock(&kvm_lock);
592         kvm_free_irq_routing(kvm);
593         for (i = 0; i < KVM_NR_BUSES; i++)
594                 kvm_io_bus_destroy(kvm->buses[i]);
595         kvm_coalesced_mmio_free(kvm);
596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
598 #else
599         kvm_arch_flush_shadow_all(kvm);
600 #endif
601         kvm_arch_destroy_vm(kvm);
602         kvm_free_physmem(kvm);
603         cleanup_srcu_struct(&kvm->srcu);
604         kvm_arch_free_vm(kvm);
605         hardware_disable_all();
606         mmdrop(mm);
607 }
608
609 void kvm_get_kvm(struct kvm *kvm)
610 {
611         atomic_inc(&kvm->users_count);
612 }
613 EXPORT_SYMBOL_GPL(kvm_get_kvm);
614
615 void kvm_put_kvm(struct kvm *kvm)
616 {
617         if (atomic_dec_and_test(&kvm->users_count))
618                 kvm_destroy_vm(kvm);
619 }
620 EXPORT_SYMBOL_GPL(kvm_put_kvm);
621
622
623 static int kvm_vm_release(struct inode *inode, struct file *filp)
624 {
625         struct kvm *kvm = filp->private_data;
626
627         kvm_irqfd_release(kvm);
628
629         kvm_put_kvm(kvm);
630         return 0;
631 }
632
633 /*
634  * Allocation size is twice as large as the actual dirty bitmap size.
635  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
636  */
637 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
638 {
639 #ifndef CONFIG_S390
640         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
641
642         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
643         if (!memslot->dirty_bitmap)
644                 return -ENOMEM;
645
646 #endif /* !CONFIG_S390 */
647         return 0;
648 }
649
650 static int cmp_memslot(const void *slot1, const void *slot2)
651 {
652         struct kvm_memory_slot *s1, *s2;
653
654         s1 = (struct kvm_memory_slot *)slot1;
655         s2 = (struct kvm_memory_slot *)slot2;
656
657         if (s1->npages < s2->npages)
658                 return 1;
659         if (s1->npages > s2->npages)
660                 return -1;
661
662         return 0;
663 }
664
665 /*
666  * Sort the memslots base on its size, so the larger slots
667  * will get better fit.
668  */
669 static void sort_memslots(struct kvm_memslots *slots)
670 {
671         int i;
672
673         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
674               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
675
676         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
677                 slots->id_to_index[slots->memslots[i].id] = i;
678 }
679
680 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
681                      u64 last_generation)
682 {
683         if (new) {
684                 int id = new->id;
685                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
686                 unsigned long npages = old->npages;
687
688                 *old = *new;
689                 if (new->npages != npages)
690                         sort_memslots(slots);
691         }
692
693         slots->generation = last_generation + 1;
694 }
695
696 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
697 {
698         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
699
700 #ifdef KVM_CAP_READONLY_MEM
701         valid_flags |= KVM_MEM_READONLY;
702 #endif
703
704         if (mem->flags & ~valid_flags)
705                 return -EINVAL;
706
707         return 0;
708 }
709
710 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
711                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
712 {
713         struct kvm_memslots *old_memslots = kvm->memslots;
714
715         update_memslots(slots, new, kvm->memslots->generation);
716         rcu_assign_pointer(kvm->memslots, slots);
717         synchronize_srcu_expedited(&kvm->srcu);
718         return old_memslots; 
719 }
720
721 /*
722  * Allocate some memory and give it an address in the guest physical address
723  * space.
724  *
725  * Discontiguous memory is allowed, mostly for framebuffers.
726  *
727  * Must be called holding mmap_sem for write.
728  */
729 int __kvm_set_memory_region(struct kvm *kvm,
730                             struct kvm_userspace_memory_region *mem,
731                             bool user_alloc)
732 {
733         int r;
734         gfn_t base_gfn;
735         unsigned long npages;
736         struct kvm_memory_slot *slot;
737         struct kvm_memory_slot old, new;
738         struct kvm_memslots *slots = NULL, *old_memslots;
739         bool old_iommu_mapped;
740
741         r = check_memory_region_flags(mem);
742         if (r)
743                 goto out;
744
745         r = -EINVAL;
746         /* General sanity checks */
747         if (mem->memory_size & (PAGE_SIZE - 1))
748                 goto out;
749         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
750                 goto out;
751         /* We can read the guest memory with __xxx_user() later on. */
752         if (user_alloc &&
753             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
754              !access_ok(VERIFY_WRITE,
755                         (void __user *)(unsigned long)mem->userspace_addr,
756                         mem->memory_size)))
757                 goto out;
758         if (mem->slot >= KVM_MEM_SLOTS_NUM)
759                 goto out;
760         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
761                 goto out;
762
763         slot = id_to_memslot(kvm->memslots, mem->slot);
764         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
765         npages = mem->memory_size >> PAGE_SHIFT;
766
767         r = -EINVAL;
768         if (npages > KVM_MEM_MAX_NR_PAGES)
769                 goto out;
770
771         if (!npages)
772                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
773
774         new = old = *slot;
775
776         new.id = mem->slot;
777         new.base_gfn = base_gfn;
778         new.npages = npages;
779         new.flags = mem->flags;
780
781         old_iommu_mapped = old.npages;
782
783         /*
784          * Disallow changing a memory slot's size or changing anything about
785          * zero sized slots that doesn't involve making them non-zero.
786          */
787         r = -EINVAL;
788         if (npages && old.npages && npages != old.npages)
789                 goto out;
790         if (!npages && !old.npages)
791                 goto out;
792
793         if ((npages && !old.npages) || (base_gfn != old.base_gfn)) {
794                 /* Check for overlaps */
795                 r = -EEXIST;
796                 kvm_for_each_memslot(slot, kvm->memslots) {
797                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
798                             (slot->id == mem->slot))
799                                 continue;
800                         if (!((base_gfn + npages <= slot->base_gfn) ||
801                               (base_gfn >= slot->base_gfn + slot->npages)))
802                                 goto out;
803                 }
804         }
805
806         /* Free page dirty bitmap if unneeded */
807         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
808                 new.dirty_bitmap = NULL;
809
810         r = -ENOMEM;
811
812         /*
813          * Allocate if a slot is being created.  If modifying a slot,
814          * the userspace_addr cannot change.
815          */
816         if (!old.npages) {
817                 new.user_alloc = user_alloc;
818                 new.userspace_addr = mem->userspace_addr;
819
820                 if (kvm_arch_create_memslot(&new, npages))
821                         goto out_free;
822         } else if (npages && mem->userspace_addr != old.userspace_addr) {
823                 r = -EINVAL;
824                 goto out_free;
825         }
826
827         /* Allocate page dirty bitmap if needed */
828         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
829                 if (kvm_create_dirty_bitmap(&new) < 0)
830                         goto out_free;
831         }
832
833         if (!npages || base_gfn != old.base_gfn) {
834                 r = -ENOMEM;
835                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
836                                 GFP_KERNEL);
837                 if (!slots)
838                         goto out_free;
839                 slot = id_to_memslot(slots, mem->slot);
840                 slot->flags |= KVM_MEMSLOT_INVALID;
841
842                 old_memslots = install_new_memslots(kvm, slots, NULL);
843
844                 /* slot was deleted or moved, clear iommu mapping */
845                 kvm_iommu_unmap_pages(kvm, &old);
846                 old_iommu_mapped = false;
847                 /* From this point no new shadow pages pointing to a deleted,
848                  * or moved, memslot will be created.
849                  *
850                  * validation of sp->gfn happens in:
851                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
852                  *      - kvm_is_visible_gfn (mmu_check_roots)
853                  */
854                 kvm_arch_flush_shadow_memslot(kvm, slot);
855                 slots = old_memslots;
856         }
857
858         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
859         if (r)
860                 goto out_slots;
861
862         r = -ENOMEM;
863         /*
864          * We can re-use the old_memslots from above, the only difference
865          * from the currently installed memslots is the invalid flag.  This
866          * will get overwritten by update_memslots anyway.
867          */
868         if (!slots) {
869                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
870                                 GFP_KERNEL);
871                 if (!slots)
872                         goto out_free;
873         }
874
875         /*
876          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
877          * un-mapped and re-mapped if their base changes or if flags that the
878          * iommu cares about change (read-only).  Base change unmapping is
879          * handled above with slot deletion, so we only unmap incompatible
880          * flags here.  Anything else the iommu might care about for existing
881          * slots (size changes, userspace addr changes) is disallowed above,
882          * so any other attribute changes getting here can be skipped.
883          */
884         if (npages) {
885                 if (old_iommu_mapped &&
886                     ((new.flags ^ old.flags) & KVM_MEM_READONLY)) {
887                         kvm_iommu_unmap_pages(kvm, &old);
888                         old_iommu_mapped = false;
889                 }
890
891                 if (!old_iommu_mapped) {
892                         r = kvm_iommu_map_pages(kvm, &new);
893                         if (r)
894                                 goto out_slots;
895                 }
896         }
897
898         /* actual memory is freed via old in kvm_free_physmem_slot below */
899         if (!npages) {
900                 new.dirty_bitmap = NULL;
901                 memset(&new.arch, 0, sizeof(new.arch));
902         }
903
904         old_memslots = install_new_memslots(kvm, slots, &new);
905
906         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
907
908         kvm_free_physmem_slot(&old, &new);
909         kfree(old_memslots);
910
911         return 0;
912
913 out_slots:
914         kfree(slots);
915 out_free:
916         kvm_free_physmem_slot(&new, &old);
917 out:
918         return r;
919 }
920 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
921
922 int kvm_set_memory_region(struct kvm *kvm,
923                           struct kvm_userspace_memory_region *mem,
924                           bool user_alloc)
925 {
926         int r;
927
928         mutex_lock(&kvm->slots_lock);
929         r = __kvm_set_memory_region(kvm, mem, user_alloc);
930         mutex_unlock(&kvm->slots_lock);
931         return r;
932 }
933 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
934
935 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
936                                    struct
937                                    kvm_userspace_memory_region *mem,
938                                    bool user_alloc)
939 {
940         if (mem->slot >= KVM_USER_MEM_SLOTS)
941                 return -EINVAL;
942         return kvm_set_memory_region(kvm, mem, user_alloc);
943 }
944
945 int kvm_get_dirty_log(struct kvm *kvm,
946                         struct kvm_dirty_log *log, int *is_dirty)
947 {
948         struct kvm_memory_slot *memslot;
949         int r, i;
950         unsigned long n;
951         unsigned long any = 0;
952
953         r = -EINVAL;
954         if (log->slot >= KVM_USER_MEM_SLOTS)
955                 goto out;
956
957         memslot = id_to_memslot(kvm->memslots, log->slot);
958         r = -ENOENT;
959         if (!memslot->dirty_bitmap)
960                 goto out;
961
962         n = kvm_dirty_bitmap_bytes(memslot);
963
964         for (i = 0; !any && i < n/sizeof(long); ++i)
965                 any = memslot->dirty_bitmap[i];
966
967         r = -EFAULT;
968         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
969                 goto out;
970
971         if (any)
972                 *is_dirty = 1;
973
974         r = 0;
975 out:
976         return r;
977 }
978
979 bool kvm_largepages_enabled(void)
980 {
981         return largepages_enabled;
982 }
983
984 void kvm_disable_largepages(void)
985 {
986         largepages_enabled = false;
987 }
988 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
989
990 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
991 {
992         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
993 }
994 EXPORT_SYMBOL_GPL(gfn_to_memslot);
995
996 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
997 {
998         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
999
1000         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1001               memslot->flags & KVM_MEMSLOT_INVALID)
1002                 return 0;
1003
1004         return 1;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1007
1008 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1009 {
1010         struct vm_area_struct *vma;
1011         unsigned long addr, size;
1012
1013         size = PAGE_SIZE;
1014
1015         addr = gfn_to_hva(kvm, gfn);
1016         if (kvm_is_error_hva(addr))
1017                 return PAGE_SIZE;
1018
1019         down_read(&current->mm->mmap_sem);
1020         vma = find_vma(current->mm, addr);
1021         if (!vma)
1022                 goto out;
1023
1024         size = vma_kernel_pagesize(vma);
1025
1026 out:
1027         up_read(&current->mm->mmap_sem);
1028
1029         return size;
1030 }
1031
1032 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1033 {
1034         return slot->flags & KVM_MEM_READONLY;
1035 }
1036
1037 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1038                                        gfn_t *nr_pages, bool write)
1039 {
1040         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1041                 return KVM_HVA_ERR_BAD;
1042
1043         if (memslot_is_readonly(slot) && write)
1044                 return KVM_HVA_ERR_RO_BAD;
1045
1046         if (nr_pages)
1047                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1048
1049         return __gfn_to_hva_memslot(slot, gfn);
1050 }
1051
1052 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1053                                      gfn_t *nr_pages)
1054 {
1055         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1056 }
1057
1058 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1059                                  gfn_t gfn)
1060 {
1061         return gfn_to_hva_many(slot, gfn, NULL);
1062 }
1063 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1064
1065 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1066 {
1067         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1068 }
1069 EXPORT_SYMBOL_GPL(gfn_to_hva);
1070
1071 /*
1072  * The hva returned by this function is only allowed to be read.
1073  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1074  */
1075 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1076 {
1077         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1078 }
1079
1080 static int kvm_read_hva(void *data, void __user *hva, int len)
1081 {
1082         return __copy_from_user(data, hva, len);
1083 }
1084
1085 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1086 {
1087         return __copy_from_user_inatomic(data, hva, len);
1088 }
1089
1090 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1091         unsigned long start, int write, struct page **page)
1092 {
1093         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1094
1095         if (write)
1096                 flags |= FOLL_WRITE;
1097
1098         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1099 }
1100
1101 static inline int check_user_page_hwpoison(unsigned long addr)
1102 {
1103         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1104
1105         rc = __get_user_pages(current, current->mm, addr, 1,
1106                               flags, NULL, NULL, NULL);
1107         return rc == -EHWPOISON;
1108 }
1109
1110 /*
1111  * The atomic path to get the writable pfn which will be stored in @pfn,
1112  * true indicates success, otherwise false is returned.
1113  */
1114 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1115                             bool write_fault, bool *writable, pfn_t *pfn)
1116 {
1117         struct page *page[1];
1118         int npages;
1119
1120         if (!(async || atomic))
1121                 return false;
1122
1123         /*
1124          * Fast pin a writable pfn only if it is a write fault request
1125          * or the caller allows to map a writable pfn for a read fault
1126          * request.
1127          */
1128         if (!(write_fault || writable))
1129                 return false;
1130
1131         npages = __get_user_pages_fast(addr, 1, 1, page);
1132         if (npages == 1) {
1133                 *pfn = page_to_pfn(page[0]);
1134
1135                 if (writable)
1136                         *writable = true;
1137                 return true;
1138         }
1139
1140         return false;
1141 }
1142
1143 /*
1144  * The slow path to get the pfn of the specified host virtual address,
1145  * 1 indicates success, -errno is returned if error is detected.
1146  */
1147 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1148                            bool *writable, pfn_t *pfn)
1149 {
1150         struct page *page[1];
1151         int npages = 0;
1152
1153         might_sleep();
1154
1155         if (writable)
1156                 *writable = write_fault;
1157
1158         if (async) {
1159                 down_read(&current->mm->mmap_sem);
1160                 npages = get_user_page_nowait(current, current->mm,
1161                                               addr, write_fault, page);
1162                 up_read(&current->mm->mmap_sem);
1163         } else
1164                 npages = get_user_pages_fast(addr, 1, write_fault,
1165                                              page);
1166         if (npages != 1)
1167                 return npages;
1168
1169         /* map read fault as writable if possible */
1170         if (unlikely(!write_fault) && writable) {
1171                 struct page *wpage[1];
1172
1173                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1174                 if (npages == 1) {
1175                         *writable = true;
1176                         put_page(page[0]);
1177                         page[0] = wpage[0];
1178                 }
1179
1180                 npages = 1;
1181         }
1182         *pfn = page_to_pfn(page[0]);
1183         return npages;
1184 }
1185
1186 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1187 {
1188         if (unlikely(!(vma->vm_flags & VM_READ)))
1189                 return false;
1190
1191         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1192                 return false;
1193
1194         return true;
1195 }
1196
1197 /*
1198  * Pin guest page in memory and return its pfn.
1199  * @addr: host virtual address which maps memory to the guest
1200  * @atomic: whether this function can sleep
1201  * @async: whether this function need to wait IO complete if the
1202  *         host page is not in the memory
1203  * @write_fault: whether we should get a writable host page
1204  * @writable: whether it allows to map a writable host page for !@write_fault
1205  *
1206  * The function will map a writable host page for these two cases:
1207  * 1): @write_fault = true
1208  * 2): @write_fault = false && @writable, @writable will tell the caller
1209  *     whether the mapping is writable.
1210  */
1211 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1212                         bool write_fault, bool *writable)
1213 {
1214         struct vm_area_struct *vma;
1215         pfn_t pfn = 0;
1216         int npages;
1217
1218         /* we can do it either atomically or asynchronously, not both */
1219         BUG_ON(atomic && async);
1220
1221         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1222                 return pfn;
1223
1224         if (atomic)
1225                 return KVM_PFN_ERR_FAULT;
1226
1227         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1228         if (npages == 1)
1229                 return pfn;
1230
1231         down_read(&current->mm->mmap_sem);
1232         if (npages == -EHWPOISON ||
1233               (!async && check_user_page_hwpoison(addr))) {
1234                 pfn = KVM_PFN_ERR_HWPOISON;
1235                 goto exit;
1236         }
1237
1238         vma = find_vma_intersection(current->mm, addr, addr + 1);
1239
1240         if (vma == NULL)
1241                 pfn = KVM_PFN_ERR_FAULT;
1242         else if ((vma->vm_flags & VM_PFNMAP)) {
1243                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1244                         vma->vm_pgoff;
1245                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1246         } else {
1247                 if (async && vma_is_valid(vma, write_fault))
1248                         *async = true;
1249                 pfn = KVM_PFN_ERR_FAULT;
1250         }
1251 exit:
1252         up_read(&current->mm->mmap_sem);
1253         return pfn;
1254 }
1255
1256 static pfn_t
1257 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1258                      bool *async, bool write_fault, bool *writable)
1259 {
1260         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1261
1262         if (addr == KVM_HVA_ERR_RO_BAD)
1263                 return KVM_PFN_ERR_RO_FAULT;
1264
1265         if (kvm_is_error_hva(addr))
1266                 return KVM_PFN_NOSLOT;
1267
1268         /* Do not map writable pfn in the readonly memslot. */
1269         if (writable && memslot_is_readonly(slot)) {
1270                 *writable = false;
1271                 writable = NULL;
1272         }
1273
1274         return hva_to_pfn(addr, atomic, async, write_fault,
1275                           writable);
1276 }
1277
1278 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1279                           bool write_fault, bool *writable)
1280 {
1281         struct kvm_memory_slot *slot;
1282
1283         if (async)
1284                 *async = false;
1285
1286         slot = gfn_to_memslot(kvm, gfn);
1287
1288         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1289                                     writable);
1290 }
1291
1292 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1293 {
1294         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1295 }
1296 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1297
1298 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1299                        bool write_fault, bool *writable)
1300 {
1301         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1302 }
1303 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1304
1305 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1306 {
1307         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1308 }
1309 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1310
1311 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1312                       bool *writable)
1313 {
1314         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1315 }
1316 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1317
1318 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1319 {
1320         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1321 }
1322
1323 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1324 {
1325         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1326 }
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1328
1329 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1330                                                                   int nr_pages)
1331 {
1332         unsigned long addr;
1333         gfn_t entry;
1334
1335         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1336         if (kvm_is_error_hva(addr))
1337                 return -1;
1338
1339         if (entry < nr_pages)
1340                 return 0;
1341
1342         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1343 }
1344 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1345
1346 static struct page *kvm_pfn_to_page(pfn_t pfn)
1347 {
1348         if (is_error_noslot_pfn(pfn))
1349                 return KVM_ERR_PTR_BAD_PAGE;
1350
1351         if (kvm_is_mmio_pfn(pfn)) {
1352                 WARN_ON(1);
1353                 return KVM_ERR_PTR_BAD_PAGE;
1354         }
1355
1356         return pfn_to_page(pfn);
1357 }
1358
1359 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1360 {
1361         pfn_t pfn;
1362
1363         pfn = gfn_to_pfn(kvm, gfn);
1364
1365         return kvm_pfn_to_page(pfn);
1366 }
1367
1368 EXPORT_SYMBOL_GPL(gfn_to_page);
1369
1370 void kvm_release_page_clean(struct page *page)
1371 {
1372         WARN_ON(is_error_page(page));
1373
1374         kvm_release_pfn_clean(page_to_pfn(page));
1375 }
1376 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1377
1378 void kvm_release_pfn_clean(pfn_t pfn)
1379 {
1380         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1381                 put_page(pfn_to_page(pfn));
1382 }
1383 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1384
1385 void kvm_release_page_dirty(struct page *page)
1386 {
1387         WARN_ON(is_error_page(page));
1388
1389         kvm_release_pfn_dirty(page_to_pfn(page));
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1392
1393 void kvm_release_pfn_dirty(pfn_t pfn)
1394 {
1395         kvm_set_pfn_dirty(pfn);
1396         kvm_release_pfn_clean(pfn);
1397 }
1398 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1399
1400 void kvm_set_page_dirty(struct page *page)
1401 {
1402         kvm_set_pfn_dirty(page_to_pfn(page));
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1405
1406 void kvm_set_pfn_dirty(pfn_t pfn)
1407 {
1408         if (!kvm_is_mmio_pfn(pfn)) {
1409                 struct page *page = pfn_to_page(pfn);
1410                 if (!PageReserved(page))
1411                         SetPageDirty(page);
1412         }
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1415
1416 void kvm_set_pfn_accessed(pfn_t pfn)
1417 {
1418         if (!kvm_is_mmio_pfn(pfn))
1419                 mark_page_accessed(pfn_to_page(pfn));
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1422
1423 void kvm_get_pfn(pfn_t pfn)
1424 {
1425         if (!kvm_is_mmio_pfn(pfn))
1426                 get_page(pfn_to_page(pfn));
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1429
1430 static int next_segment(unsigned long len, int offset)
1431 {
1432         if (len > PAGE_SIZE - offset)
1433                 return PAGE_SIZE - offset;
1434         else
1435                 return len;
1436 }
1437
1438 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1439                         int len)
1440 {
1441         int r;
1442         unsigned long addr;
1443
1444         addr = gfn_to_hva_read(kvm, gfn);
1445         if (kvm_is_error_hva(addr))
1446                 return -EFAULT;
1447         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1448         if (r)
1449                 return -EFAULT;
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1453
1454 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1455 {
1456         gfn_t gfn = gpa >> PAGE_SHIFT;
1457         int seg;
1458         int offset = offset_in_page(gpa);
1459         int ret;
1460
1461         while ((seg = next_segment(len, offset)) != 0) {
1462                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1463                 if (ret < 0)
1464                         return ret;
1465                 offset = 0;
1466                 len -= seg;
1467                 data += seg;
1468                 ++gfn;
1469         }
1470         return 0;
1471 }
1472 EXPORT_SYMBOL_GPL(kvm_read_guest);
1473
1474 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1475                           unsigned long len)
1476 {
1477         int r;
1478         unsigned long addr;
1479         gfn_t gfn = gpa >> PAGE_SHIFT;
1480         int offset = offset_in_page(gpa);
1481
1482         addr = gfn_to_hva_read(kvm, gfn);
1483         if (kvm_is_error_hva(addr))
1484                 return -EFAULT;
1485         pagefault_disable();
1486         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1487         pagefault_enable();
1488         if (r)
1489                 return -EFAULT;
1490         return 0;
1491 }
1492 EXPORT_SYMBOL(kvm_read_guest_atomic);
1493
1494 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1495                          int offset, int len)
1496 {
1497         int r;
1498         unsigned long addr;
1499
1500         addr = gfn_to_hva(kvm, gfn);
1501         if (kvm_is_error_hva(addr))
1502                 return -EFAULT;
1503         r = __copy_to_user((void __user *)addr + offset, data, len);
1504         if (r)
1505                 return -EFAULT;
1506         mark_page_dirty(kvm, gfn);
1507         return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1510
1511 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1512                     unsigned long len)
1513 {
1514         gfn_t gfn = gpa >> PAGE_SHIFT;
1515         int seg;
1516         int offset = offset_in_page(gpa);
1517         int ret;
1518
1519         while ((seg = next_segment(len, offset)) != 0) {
1520                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1521                 if (ret < 0)
1522                         return ret;
1523                 offset = 0;
1524                 len -= seg;
1525                 data += seg;
1526                 ++gfn;
1527         }
1528         return 0;
1529 }
1530
1531 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1532                               gpa_t gpa)
1533 {
1534         struct kvm_memslots *slots = kvm_memslots(kvm);
1535         int offset = offset_in_page(gpa);
1536         gfn_t gfn = gpa >> PAGE_SHIFT;
1537
1538         ghc->gpa = gpa;
1539         ghc->generation = slots->generation;
1540         ghc->memslot = gfn_to_memslot(kvm, gfn);
1541         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1542         if (!kvm_is_error_hva(ghc->hva))
1543                 ghc->hva += offset;
1544         else
1545                 return -EFAULT;
1546
1547         return 0;
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1550
1551 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1552                            void *data, unsigned long len)
1553 {
1554         struct kvm_memslots *slots = kvm_memslots(kvm);
1555         int r;
1556
1557         if (slots->generation != ghc->generation)
1558                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1559
1560         if (kvm_is_error_hva(ghc->hva))
1561                 return -EFAULT;
1562
1563         r = __copy_to_user((void __user *)ghc->hva, data, len);
1564         if (r)
1565                 return -EFAULT;
1566         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1567
1568         return 0;
1569 }
1570 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1571
1572 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1573                            void *data, unsigned long len)
1574 {
1575         struct kvm_memslots *slots = kvm_memslots(kvm);
1576         int r;
1577
1578         if (slots->generation != ghc->generation)
1579                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1580
1581         if (kvm_is_error_hva(ghc->hva))
1582                 return -EFAULT;
1583
1584         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1585         if (r)
1586                 return -EFAULT;
1587
1588         return 0;
1589 }
1590 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1591
1592 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1593 {
1594         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1595                                     offset, len);
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1598
1599 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1600 {
1601         gfn_t gfn = gpa >> PAGE_SHIFT;
1602         int seg;
1603         int offset = offset_in_page(gpa);
1604         int ret;
1605
1606         while ((seg = next_segment(len, offset)) != 0) {
1607                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1608                 if (ret < 0)
1609                         return ret;
1610                 offset = 0;
1611                 len -= seg;
1612                 ++gfn;
1613         }
1614         return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1617
1618 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1619                              gfn_t gfn)
1620 {
1621         if (memslot && memslot->dirty_bitmap) {
1622                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1623
1624                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1625         }
1626 }
1627
1628 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1629 {
1630         struct kvm_memory_slot *memslot;
1631
1632         memslot = gfn_to_memslot(kvm, gfn);
1633         mark_page_dirty_in_slot(kvm, memslot, gfn);
1634 }
1635
1636 /*
1637  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1638  */
1639 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1640 {
1641         DEFINE_WAIT(wait);
1642
1643         for (;;) {
1644                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1645
1646                 if (kvm_arch_vcpu_runnable(vcpu)) {
1647                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1648                         break;
1649                 }
1650                 if (kvm_cpu_has_pending_timer(vcpu))
1651                         break;
1652                 if (signal_pending(current))
1653                         break;
1654
1655                 schedule();
1656         }
1657
1658         finish_wait(&vcpu->wq, &wait);
1659 }
1660
1661 #ifndef CONFIG_S390
1662 /*
1663  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1664  */
1665 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1666 {
1667         int me;
1668         int cpu = vcpu->cpu;
1669         wait_queue_head_t *wqp;
1670
1671         wqp = kvm_arch_vcpu_wq(vcpu);
1672         if (waitqueue_active(wqp)) {
1673                 wake_up_interruptible(wqp);
1674                 ++vcpu->stat.halt_wakeup;
1675         }
1676
1677         me = get_cpu();
1678         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1679                 if (kvm_arch_vcpu_should_kick(vcpu))
1680                         smp_send_reschedule(cpu);
1681         put_cpu();
1682 }
1683 #endif /* !CONFIG_S390 */
1684
1685 void kvm_resched(struct kvm_vcpu *vcpu)
1686 {
1687         if (!need_resched())
1688                 return;
1689         cond_resched();
1690 }
1691 EXPORT_SYMBOL_GPL(kvm_resched);
1692
1693 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1694 {
1695         struct pid *pid;
1696         struct task_struct *task = NULL;
1697         bool ret = false;
1698
1699         rcu_read_lock();
1700         pid = rcu_dereference(target->pid);
1701         if (pid)
1702                 task = get_pid_task(target->pid, PIDTYPE_PID);
1703         rcu_read_unlock();
1704         if (!task)
1705                 return ret;
1706         if (task->flags & PF_VCPU) {
1707                 put_task_struct(task);
1708                 return ret;
1709         }
1710         ret = yield_to(task, 1);
1711         put_task_struct(task);
1712
1713         return ret;
1714 }
1715 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1716
1717 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1718 /*
1719  * Helper that checks whether a VCPU is eligible for directed yield.
1720  * Most eligible candidate to yield is decided by following heuristics:
1721  *
1722  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1723  *  (preempted lock holder), indicated by @in_spin_loop.
1724  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1725  *
1726  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1727  *  chance last time (mostly it has become eligible now since we have probably
1728  *  yielded to lockholder in last iteration. This is done by toggling
1729  *  @dy_eligible each time a VCPU checked for eligibility.)
1730  *
1731  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1732  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1733  *  burning. Giving priority for a potential lock-holder increases lock
1734  *  progress.
1735  *
1736  *  Since algorithm is based on heuristics, accessing another VCPU data without
1737  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1738  *  and continue with next VCPU and so on.
1739  */
1740 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1741 {
1742         bool eligible;
1743
1744         eligible = !vcpu->spin_loop.in_spin_loop ||
1745                         (vcpu->spin_loop.in_spin_loop &&
1746                          vcpu->spin_loop.dy_eligible);
1747
1748         if (vcpu->spin_loop.in_spin_loop)
1749                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1750
1751         return eligible;
1752 }
1753 #endif
1754
1755 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1756 {
1757         struct kvm *kvm = me->kvm;
1758         struct kvm_vcpu *vcpu;
1759         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1760         int yielded = 0;
1761         int try = 3;
1762         int pass;
1763         int i;
1764
1765         kvm_vcpu_set_in_spin_loop(me, true);
1766         /*
1767          * We boost the priority of a VCPU that is runnable but not
1768          * currently running, because it got preempted by something
1769          * else and called schedule in __vcpu_run.  Hopefully that
1770          * VCPU is holding the lock that we need and will release it.
1771          * We approximate round-robin by starting at the last boosted VCPU.
1772          */
1773         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1774                 kvm_for_each_vcpu(i, vcpu, kvm) {
1775                         if (!pass && i <= last_boosted_vcpu) {
1776                                 i = last_boosted_vcpu;
1777                                 continue;
1778                         } else if (pass && i > last_boosted_vcpu)
1779                                 break;
1780                         if (vcpu == me)
1781                                 continue;
1782                         if (waitqueue_active(&vcpu->wq))
1783                                 continue;
1784                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1785                                 continue;
1786
1787                         yielded = kvm_vcpu_yield_to(vcpu);
1788                         if (yielded > 0) {
1789                                 kvm->last_boosted_vcpu = i;
1790                                 break;
1791                         } else if (yielded < 0) {
1792                                 try--;
1793                                 if (!try)
1794                                         break;
1795                         }
1796                 }
1797         }
1798         kvm_vcpu_set_in_spin_loop(me, false);
1799
1800         /* Ensure vcpu is not eligible during next spinloop */
1801         kvm_vcpu_set_dy_eligible(me, false);
1802 }
1803 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1804
1805 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1806 {
1807         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1808         struct page *page;
1809
1810         if (vmf->pgoff == 0)
1811                 page = virt_to_page(vcpu->run);
1812 #ifdef CONFIG_X86
1813         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1814                 page = virt_to_page(vcpu->arch.pio_data);
1815 #endif
1816 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1817         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1818                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1819 #endif
1820         else
1821                 return kvm_arch_vcpu_fault(vcpu, vmf);
1822         get_page(page);
1823         vmf->page = page;
1824         return 0;
1825 }
1826
1827 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1828         .fault = kvm_vcpu_fault,
1829 };
1830
1831 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1832 {
1833         vma->vm_ops = &kvm_vcpu_vm_ops;
1834         return 0;
1835 }
1836
1837 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1838 {
1839         struct kvm_vcpu *vcpu = filp->private_data;
1840
1841         kvm_put_kvm(vcpu->kvm);
1842         return 0;
1843 }
1844
1845 static struct file_operations kvm_vcpu_fops = {
1846         .release        = kvm_vcpu_release,
1847         .unlocked_ioctl = kvm_vcpu_ioctl,
1848 #ifdef CONFIG_COMPAT
1849         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1850 #endif
1851         .mmap           = kvm_vcpu_mmap,
1852         .llseek         = noop_llseek,
1853 };
1854
1855 /*
1856  * Allocates an inode for the vcpu.
1857  */
1858 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1859 {
1860         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1861 }
1862
1863 /*
1864  * Creates some virtual cpus.  Good luck creating more than one.
1865  */
1866 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1867 {
1868         int r;
1869         struct kvm_vcpu *vcpu, *v;
1870
1871         vcpu = kvm_arch_vcpu_create(kvm, id);
1872         if (IS_ERR(vcpu))
1873                 return PTR_ERR(vcpu);
1874
1875         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1876
1877         r = kvm_arch_vcpu_setup(vcpu);
1878         if (r)
1879                 goto vcpu_destroy;
1880
1881         mutex_lock(&kvm->lock);
1882         if (!kvm_vcpu_compatible(vcpu)) {
1883                 r = -EINVAL;
1884                 goto unlock_vcpu_destroy;
1885         }
1886         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1887                 r = -EINVAL;
1888                 goto unlock_vcpu_destroy;
1889         }
1890
1891         kvm_for_each_vcpu(r, v, kvm)
1892                 if (v->vcpu_id == id) {
1893                         r = -EEXIST;
1894                         goto unlock_vcpu_destroy;
1895                 }
1896
1897         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1898
1899         /* Now it's all set up, let userspace reach it */
1900         kvm_get_kvm(kvm);
1901         r = create_vcpu_fd(vcpu);
1902         if (r < 0) {
1903                 kvm_put_kvm(kvm);
1904                 goto unlock_vcpu_destroy;
1905         }
1906
1907         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1908         smp_wmb();
1909         atomic_inc(&kvm->online_vcpus);
1910
1911         mutex_unlock(&kvm->lock);
1912         kvm_arch_vcpu_postcreate(vcpu);
1913         return r;
1914
1915 unlock_vcpu_destroy:
1916         mutex_unlock(&kvm->lock);
1917 vcpu_destroy:
1918         kvm_arch_vcpu_destroy(vcpu);
1919         return r;
1920 }
1921
1922 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1923 {
1924         if (sigset) {
1925                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1926                 vcpu->sigset_active = 1;
1927                 vcpu->sigset = *sigset;
1928         } else
1929                 vcpu->sigset_active = 0;
1930         return 0;
1931 }
1932
1933 static long kvm_vcpu_ioctl(struct file *filp,
1934                            unsigned int ioctl, unsigned long arg)
1935 {
1936         struct kvm_vcpu *vcpu = filp->private_data;
1937         void __user *argp = (void __user *)arg;
1938         int r;
1939         struct kvm_fpu *fpu = NULL;
1940         struct kvm_sregs *kvm_sregs = NULL;
1941
1942         if (vcpu->kvm->mm != current->mm)
1943                 return -EIO;
1944
1945 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1946         /*
1947          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1948          * so vcpu_load() would break it.
1949          */
1950         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1951                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1952 #endif
1953
1954
1955         r = vcpu_load(vcpu);
1956         if (r)
1957                 return r;
1958         switch (ioctl) {
1959         case KVM_RUN:
1960                 r = -EINVAL;
1961                 if (arg)
1962                         goto out;
1963                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1964                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1965                 break;
1966         case KVM_GET_REGS: {
1967                 struct kvm_regs *kvm_regs;
1968
1969                 r = -ENOMEM;
1970                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1971                 if (!kvm_regs)
1972                         goto out;
1973                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1974                 if (r)
1975                         goto out_free1;
1976                 r = -EFAULT;
1977                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1978                         goto out_free1;
1979                 r = 0;
1980 out_free1:
1981                 kfree(kvm_regs);
1982                 break;
1983         }
1984         case KVM_SET_REGS: {
1985                 struct kvm_regs *kvm_regs;
1986
1987                 r = -ENOMEM;
1988                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1989                 if (IS_ERR(kvm_regs)) {
1990                         r = PTR_ERR(kvm_regs);
1991                         goto out;
1992                 }
1993                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1994                 kfree(kvm_regs);
1995                 break;
1996         }
1997         case KVM_GET_SREGS: {
1998                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1999                 r = -ENOMEM;
2000                 if (!kvm_sregs)
2001                         goto out;
2002                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2003                 if (r)
2004                         goto out;
2005                 r = -EFAULT;
2006                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2007                         goto out;
2008                 r = 0;
2009                 break;
2010         }
2011         case KVM_SET_SREGS: {
2012                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2013                 if (IS_ERR(kvm_sregs)) {
2014                         r = PTR_ERR(kvm_sregs);
2015                         kvm_sregs = NULL;
2016                         goto out;
2017                 }
2018                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2019                 break;
2020         }
2021         case KVM_GET_MP_STATE: {
2022                 struct kvm_mp_state mp_state;
2023
2024                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2025                 if (r)
2026                         goto out;
2027                 r = -EFAULT;
2028                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2029                         goto out;
2030                 r = 0;
2031                 break;
2032         }
2033         case KVM_SET_MP_STATE: {
2034                 struct kvm_mp_state mp_state;
2035
2036                 r = -EFAULT;
2037                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2038                         goto out;
2039                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2040                 break;
2041         }
2042         case KVM_TRANSLATE: {
2043                 struct kvm_translation tr;
2044
2045                 r = -EFAULT;
2046                 if (copy_from_user(&tr, argp, sizeof tr))
2047                         goto out;
2048                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2049                 if (r)
2050                         goto out;
2051                 r = -EFAULT;
2052                 if (copy_to_user(argp, &tr, sizeof tr))
2053                         goto out;
2054                 r = 0;
2055                 break;
2056         }
2057         case KVM_SET_GUEST_DEBUG: {
2058                 struct kvm_guest_debug dbg;
2059
2060                 r = -EFAULT;
2061                 if (copy_from_user(&dbg, argp, sizeof dbg))
2062                         goto out;
2063                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2064                 break;
2065         }
2066         case KVM_SET_SIGNAL_MASK: {
2067                 struct kvm_signal_mask __user *sigmask_arg = argp;
2068                 struct kvm_signal_mask kvm_sigmask;
2069                 sigset_t sigset, *p;
2070
2071                 p = NULL;
2072                 if (argp) {
2073                         r = -EFAULT;
2074                         if (copy_from_user(&kvm_sigmask, argp,
2075                                            sizeof kvm_sigmask))
2076                                 goto out;
2077                         r = -EINVAL;
2078                         if (kvm_sigmask.len != sizeof sigset)
2079                                 goto out;
2080                         r = -EFAULT;
2081                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2082                                            sizeof sigset))
2083                                 goto out;
2084                         p = &sigset;
2085                 }
2086                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2087                 break;
2088         }
2089         case KVM_GET_FPU: {
2090                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2091                 r = -ENOMEM;
2092                 if (!fpu)
2093                         goto out;
2094                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2095                 if (r)
2096                         goto out;
2097                 r = -EFAULT;
2098                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2099                         goto out;
2100                 r = 0;
2101                 break;
2102         }
2103         case KVM_SET_FPU: {
2104                 fpu = memdup_user(argp, sizeof(*fpu));
2105                 if (IS_ERR(fpu)) {
2106                         r = PTR_ERR(fpu);
2107                         fpu = NULL;
2108                         goto out;
2109                 }
2110                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2111                 break;
2112         }
2113         default:
2114                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2115         }
2116 out:
2117         vcpu_put(vcpu);
2118         kfree(fpu);
2119         kfree(kvm_sregs);
2120         return r;
2121 }
2122
2123 #ifdef CONFIG_COMPAT
2124 static long kvm_vcpu_compat_ioctl(struct file *filp,
2125                                   unsigned int ioctl, unsigned long arg)
2126 {
2127         struct kvm_vcpu *vcpu = filp->private_data;
2128         void __user *argp = compat_ptr(arg);
2129         int r;
2130
2131         if (vcpu->kvm->mm != current->mm)
2132                 return -EIO;
2133
2134         switch (ioctl) {
2135         case KVM_SET_SIGNAL_MASK: {
2136                 struct kvm_signal_mask __user *sigmask_arg = argp;
2137                 struct kvm_signal_mask kvm_sigmask;
2138                 compat_sigset_t csigset;
2139                 sigset_t sigset;
2140
2141                 if (argp) {
2142                         r = -EFAULT;
2143                         if (copy_from_user(&kvm_sigmask, argp,
2144                                            sizeof kvm_sigmask))
2145                                 goto out;
2146                         r = -EINVAL;
2147                         if (kvm_sigmask.len != sizeof csigset)
2148                                 goto out;
2149                         r = -EFAULT;
2150                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2151                                            sizeof csigset))
2152                                 goto out;
2153                         sigset_from_compat(&sigset, &csigset);
2154                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2155                 } else
2156                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2157                 break;
2158         }
2159         default:
2160                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2161         }
2162
2163 out:
2164         return r;
2165 }
2166 #endif
2167
2168 static long kvm_vm_ioctl(struct file *filp,
2169                            unsigned int ioctl, unsigned long arg)
2170 {
2171         struct kvm *kvm = filp->private_data;
2172         void __user *argp = (void __user *)arg;
2173         int r;
2174
2175         if (kvm->mm != current->mm)
2176                 return -EIO;
2177         switch (ioctl) {
2178         case KVM_CREATE_VCPU:
2179                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2180                 break;
2181         case KVM_SET_USER_MEMORY_REGION: {
2182                 struct kvm_userspace_memory_region kvm_userspace_mem;
2183
2184                 r = -EFAULT;
2185                 if (copy_from_user(&kvm_userspace_mem, argp,
2186                                                 sizeof kvm_userspace_mem))
2187                         goto out;
2188
2189                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2190                 break;
2191         }
2192         case KVM_GET_DIRTY_LOG: {
2193                 struct kvm_dirty_log log;
2194
2195                 r = -EFAULT;
2196                 if (copy_from_user(&log, argp, sizeof log))
2197                         goto out;
2198                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2199                 break;
2200         }
2201 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2202         case KVM_REGISTER_COALESCED_MMIO: {
2203                 struct kvm_coalesced_mmio_zone zone;
2204                 r = -EFAULT;
2205                 if (copy_from_user(&zone, argp, sizeof zone))
2206                         goto out;
2207                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2208                 break;
2209         }
2210         case KVM_UNREGISTER_COALESCED_MMIO: {
2211                 struct kvm_coalesced_mmio_zone zone;
2212                 r = -EFAULT;
2213                 if (copy_from_user(&zone, argp, sizeof zone))
2214                         goto out;
2215                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2216                 break;
2217         }
2218 #endif
2219         case KVM_IRQFD: {
2220                 struct kvm_irqfd data;
2221
2222                 r = -EFAULT;
2223                 if (copy_from_user(&data, argp, sizeof data))
2224                         goto out;
2225                 r = kvm_irqfd(kvm, &data);
2226                 break;
2227         }
2228         case KVM_IOEVENTFD: {
2229                 struct kvm_ioeventfd data;
2230
2231                 r = -EFAULT;
2232                 if (copy_from_user(&data, argp, sizeof data))
2233                         goto out;
2234                 r = kvm_ioeventfd(kvm, &data);
2235                 break;
2236         }
2237 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2238         case KVM_SET_BOOT_CPU_ID:
2239                 r = 0;
2240                 mutex_lock(&kvm->lock);
2241                 if (atomic_read(&kvm->online_vcpus) != 0)
2242                         r = -EBUSY;
2243                 else
2244                         kvm->bsp_vcpu_id = arg;
2245                 mutex_unlock(&kvm->lock);
2246                 break;
2247 #endif
2248 #ifdef CONFIG_HAVE_KVM_MSI
2249         case KVM_SIGNAL_MSI: {
2250                 struct kvm_msi msi;
2251
2252                 r = -EFAULT;
2253                 if (copy_from_user(&msi, argp, sizeof msi))
2254                         goto out;
2255                 r = kvm_send_userspace_msi(kvm, &msi);
2256                 break;
2257         }
2258 #endif
2259 #ifdef __KVM_HAVE_IRQ_LINE
2260         case KVM_IRQ_LINE_STATUS:
2261         case KVM_IRQ_LINE: {
2262                 struct kvm_irq_level irq_event;
2263
2264                 r = -EFAULT;
2265                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2266                         goto out;
2267
2268                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2269                 if (r)
2270                         goto out;
2271
2272                 r = -EFAULT;
2273                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2274                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2275                                 goto out;
2276                 }
2277
2278                 r = 0;
2279                 break;
2280         }
2281 #endif
2282         default:
2283                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2284                 if (r == -ENOTTY)
2285                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2286         }
2287 out:
2288         return r;
2289 }
2290
2291 #ifdef CONFIG_COMPAT
2292 struct compat_kvm_dirty_log {
2293         __u32 slot;
2294         __u32 padding1;
2295         union {
2296                 compat_uptr_t dirty_bitmap; /* one bit per page */
2297                 __u64 padding2;
2298         };
2299 };
2300
2301 static long kvm_vm_compat_ioctl(struct file *filp,
2302                            unsigned int ioctl, unsigned long arg)
2303 {
2304         struct kvm *kvm = filp->private_data;
2305         int r;
2306
2307         if (kvm->mm != current->mm)
2308                 return -EIO;
2309         switch (ioctl) {
2310         case KVM_GET_DIRTY_LOG: {
2311                 struct compat_kvm_dirty_log compat_log;
2312                 struct kvm_dirty_log log;
2313
2314                 r = -EFAULT;
2315                 if (copy_from_user(&compat_log, (void __user *)arg,
2316                                    sizeof(compat_log)))
2317                         goto out;
2318                 log.slot         = compat_log.slot;
2319                 log.padding1     = compat_log.padding1;
2320                 log.padding2     = compat_log.padding2;
2321                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2322
2323                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2324                 break;
2325         }
2326         default:
2327                 r = kvm_vm_ioctl(filp, ioctl, arg);
2328         }
2329
2330 out:
2331         return r;
2332 }
2333 #endif
2334
2335 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2336 {
2337         struct page *page[1];
2338         unsigned long addr;
2339         int npages;
2340         gfn_t gfn = vmf->pgoff;
2341         struct kvm *kvm = vma->vm_file->private_data;
2342
2343         addr = gfn_to_hva(kvm, gfn);
2344         if (kvm_is_error_hva(addr))
2345                 return VM_FAULT_SIGBUS;
2346
2347         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2348                                 NULL);
2349         if (unlikely(npages != 1))
2350                 return VM_FAULT_SIGBUS;
2351
2352         vmf->page = page[0];
2353         return 0;
2354 }
2355
2356 static const struct vm_operations_struct kvm_vm_vm_ops = {
2357         .fault = kvm_vm_fault,
2358 };
2359
2360 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2361 {
2362         vma->vm_ops = &kvm_vm_vm_ops;
2363         return 0;
2364 }
2365
2366 static struct file_operations kvm_vm_fops = {
2367         .release        = kvm_vm_release,
2368         .unlocked_ioctl = kvm_vm_ioctl,
2369 #ifdef CONFIG_COMPAT
2370         .compat_ioctl   = kvm_vm_compat_ioctl,
2371 #endif
2372         .mmap           = kvm_vm_mmap,
2373         .llseek         = noop_llseek,
2374 };
2375
2376 static int kvm_dev_ioctl_create_vm(unsigned long type)
2377 {
2378         int r;
2379         struct kvm *kvm;
2380
2381         kvm = kvm_create_vm(type);
2382         if (IS_ERR(kvm))
2383                 return PTR_ERR(kvm);
2384 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2385         r = kvm_coalesced_mmio_init(kvm);
2386         if (r < 0) {
2387                 kvm_put_kvm(kvm);
2388                 return r;
2389         }
2390 #endif
2391         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2392         if (r < 0)
2393                 kvm_put_kvm(kvm);
2394
2395         return r;
2396 }
2397
2398 static long kvm_dev_ioctl_check_extension_generic(long arg)
2399 {
2400         switch (arg) {
2401         case KVM_CAP_USER_MEMORY:
2402         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2403         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2404 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2405         case KVM_CAP_SET_BOOT_CPU_ID:
2406 #endif
2407         case KVM_CAP_INTERNAL_ERROR_DATA:
2408 #ifdef CONFIG_HAVE_KVM_MSI
2409         case KVM_CAP_SIGNAL_MSI:
2410 #endif
2411                 return 1;
2412 #ifdef KVM_CAP_IRQ_ROUTING
2413         case KVM_CAP_IRQ_ROUTING:
2414                 return KVM_MAX_IRQ_ROUTES;
2415 #endif
2416         default:
2417                 break;
2418         }
2419         return kvm_dev_ioctl_check_extension(arg);
2420 }
2421
2422 static long kvm_dev_ioctl(struct file *filp,
2423                           unsigned int ioctl, unsigned long arg)
2424 {
2425         long r = -EINVAL;
2426
2427         switch (ioctl) {
2428         case KVM_GET_API_VERSION:
2429                 r = -EINVAL;
2430                 if (arg)
2431                         goto out;
2432                 r = KVM_API_VERSION;
2433                 break;
2434         case KVM_CREATE_VM:
2435                 r = kvm_dev_ioctl_create_vm(arg);
2436                 break;
2437         case KVM_CHECK_EXTENSION:
2438                 r = kvm_dev_ioctl_check_extension_generic(arg);
2439                 break;
2440         case KVM_GET_VCPU_MMAP_SIZE:
2441                 r = -EINVAL;
2442                 if (arg)
2443                         goto out;
2444                 r = PAGE_SIZE;     /* struct kvm_run */
2445 #ifdef CONFIG_X86
2446                 r += PAGE_SIZE;    /* pio data page */
2447 #endif
2448 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2449                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2450 #endif
2451                 break;
2452         case KVM_TRACE_ENABLE:
2453         case KVM_TRACE_PAUSE:
2454         case KVM_TRACE_DISABLE:
2455                 r = -EOPNOTSUPP;
2456                 break;
2457         default:
2458                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2459         }
2460 out:
2461         return r;
2462 }
2463
2464 static struct file_operations kvm_chardev_ops = {
2465         .unlocked_ioctl = kvm_dev_ioctl,
2466         .compat_ioctl   = kvm_dev_ioctl,
2467         .llseek         = noop_llseek,
2468 };
2469
2470 static struct miscdevice kvm_dev = {
2471         KVM_MINOR,
2472         "kvm",
2473         &kvm_chardev_ops,
2474 };
2475
2476 static void hardware_enable_nolock(void *junk)
2477 {
2478         int cpu = raw_smp_processor_id();
2479         int r;
2480
2481         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2482                 return;
2483
2484         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2485
2486         r = kvm_arch_hardware_enable(NULL);
2487
2488         if (r) {
2489                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2490                 atomic_inc(&hardware_enable_failed);
2491                 printk(KERN_INFO "kvm: enabling virtualization on "
2492                                  "CPU%d failed\n", cpu);
2493         }
2494 }
2495
2496 static void hardware_enable(void *junk)
2497 {
2498         raw_spin_lock(&kvm_lock);
2499         hardware_enable_nolock(junk);
2500         raw_spin_unlock(&kvm_lock);
2501 }
2502
2503 static void hardware_disable_nolock(void *junk)
2504 {
2505         int cpu = raw_smp_processor_id();
2506
2507         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2508                 return;
2509         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2510         kvm_arch_hardware_disable(NULL);
2511 }
2512
2513 static void hardware_disable(void *junk)
2514 {
2515         raw_spin_lock(&kvm_lock);
2516         hardware_disable_nolock(junk);
2517         raw_spin_unlock(&kvm_lock);
2518 }
2519
2520 static void hardware_disable_all_nolock(void)
2521 {
2522         BUG_ON(!kvm_usage_count);
2523
2524         kvm_usage_count--;
2525         if (!kvm_usage_count)
2526                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2527 }
2528
2529 static void hardware_disable_all(void)
2530 {
2531         raw_spin_lock(&kvm_lock);
2532         hardware_disable_all_nolock();
2533         raw_spin_unlock(&kvm_lock);
2534 }
2535
2536 static int hardware_enable_all(void)
2537 {
2538         int r = 0;
2539
2540         raw_spin_lock(&kvm_lock);
2541
2542         kvm_usage_count++;
2543         if (kvm_usage_count == 1) {
2544                 atomic_set(&hardware_enable_failed, 0);
2545                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2546
2547                 if (atomic_read(&hardware_enable_failed)) {
2548                         hardware_disable_all_nolock();
2549                         r = -EBUSY;
2550                 }
2551         }
2552
2553         raw_spin_unlock(&kvm_lock);
2554
2555         return r;
2556 }
2557
2558 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2559                            void *v)
2560 {
2561         int cpu = (long)v;
2562
2563         if (!kvm_usage_count)
2564                 return NOTIFY_OK;
2565
2566         val &= ~CPU_TASKS_FROZEN;
2567         switch (val) {
2568         case CPU_DYING:
2569                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2570                        cpu);
2571                 hardware_disable(NULL);
2572                 break;
2573         case CPU_STARTING:
2574                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2575                        cpu);
2576                 hardware_enable(NULL);
2577                 break;
2578         }
2579         return NOTIFY_OK;
2580 }
2581
2582
2583 asmlinkage void kvm_spurious_fault(void)
2584 {
2585         /* Fault while not rebooting.  We want the trace. */
2586         BUG();
2587 }
2588 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2589
2590 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2591                       void *v)
2592 {
2593         /*
2594          * Some (well, at least mine) BIOSes hang on reboot if
2595          * in vmx root mode.
2596          *
2597          * And Intel TXT required VMX off for all cpu when system shutdown.
2598          */
2599         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2600         kvm_rebooting = true;
2601         on_each_cpu(hardware_disable_nolock, NULL, 1);
2602         return NOTIFY_OK;
2603 }
2604
2605 static struct notifier_block kvm_reboot_notifier = {
2606         .notifier_call = kvm_reboot,
2607         .priority = 0,
2608 };
2609
2610 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2611 {
2612         int i;
2613
2614         for (i = 0; i < bus->dev_count; i++) {
2615                 struct kvm_io_device *pos = bus->range[i].dev;
2616
2617                 kvm_iodevice_destructor(pos);
2618         }
2619         kfree(bus);
2620 }
2621
2622 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2623 {
2624         const struct kvm_io_range *r1 = p1;
2625         const struct kvm_io_range *r2 = p2;
2626
2627         if (r1->addr < r2->addr)
2628                 return -1;
2629         if (r1->addr + r1->len > r2->addr + r2->len)
2630                 return 1;
2631         return 0;
2632 }
2633
2634 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2635                           gpa_t addr, int len)
2636 {
2637         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2638                 .addr = addr,
2639                 .len = len,
2640                 .dev = dev,
2641         };
2642
2643         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2644                 kvm_io_bus_sort_cmp, NULL);
2645
2646         return 0;
2647 }
2648
2649 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2650                              gpa_t addr, int len)
2651 {
2652         struct kvm_io_range *range, key;
2653         int off;
2654
2655         key = (struct kvm_io_range) {
2656                 .addr = addr,
2657                 .len = len,
2658         };
2659
2660         range = bsearch(&key, bus->range, bus->dev_count,
2661                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2662         if (range == NULL)
2663                 return -ENOENT;
2664
2665         off = range - bus->range;
2666
2667         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2668                 off--;
2669
2670         return off;
2671 }
2672
2673 /* kvm_io_bus_write - called under kvm->slots_lock */
2674 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2675                      int len, const void *val)
2676 {
2677         int idx;
2678         struct kvm_io_bus *bus;
2679         struct kvm_io_range range;
2680
2681         range = (struct kvm_io_range) {
2682                 .addr = addr,
2683                 .len = len,
2684         };
2685
2686         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2687         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2688         if (idx < 0)
2689                 return -EOPNOTSUPP;
2690
2691         while (idx < bus->dev_count &&
2692                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2693                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2694                         return 0;
2695                 idx++;
2696         }
2697
2698         return -EOPNOTSUPP;
2699 }
2700
2701 /* kvm_io_bus_read - called under kvm->slots_lock */
2702 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2703                     int len, void *val)
2704 {
2705         int idx;
2706         struct kvm_io_bus *bus;
2707         struct kvm_io_range range;
2708
2709         range = (struct kvm_io_range) {
2710                 .addr = addr,
2711                 .len = len,
2712         };
2713
2714         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2715         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2716         if (idx < 0)
2717                 return -EOPNOTSUPP;
2718
2719         while (idx < bus->dev_count &&
2720                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2721                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2722                         return 0;
2723                 idx++;
2724         }
2725
2726         return -EOPNOTSUPP;
2727 }
2728
2729 /* Caller must hold slots_lock. */
2730 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2731                             int len, struct kvm_io_device *dev)
2732 {
2733         struct kvm_io_bus *new_bus, *bus;
2734
2735         bus = kvm->buses[bus_idx];
2736         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2737                 return -ENOSPC;
2738
2739         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2740                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2741         if (!new_bus)
2742                 return -ENOMEM;
2743         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2744                sizeof(struct kvm_io_range)));
2745         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2746         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2747         synchronize_srcu_expedited(&kvm->srcu);
2748         kfree(bus);
2749
2750         return 0;
2751 }
2752
2753 /* Caller must hold slots_lock. */
2754 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2755                               struct kvm_io_device *dev)
2756 {
2757         int i, r;
2758         struct kvm_io_bus *new_bus, *bus;
2759
2760         bus = kvm->buses[bus_idx];
2761         r = -ENOENT;
2762         for (i = 0; i < bus->dev_count; i++)
2763                 if (bus->range[i].dev == dev) {
2764                         r = 0;
2765                         break;
2766                 }
2767
2768         if (r)
2769                 return r;
2770
2771         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2772                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2773         if (!new_bus)
2774                 return -ENOMEM;
2775
2776         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2777         new_bus->dev_count--;
2778         memcpy(new_bus->range + i, bus->range + i + 1,
2779                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2780
2781         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2782         synchronize_srcu_expedited(&kvm->srcu);
2783         kfree(bus);
2784         return r;
2785 }
2786
2787 static struct notifier_block kvm_cpu_notifier = {
2788         .notifier_call = kvm_cpu_hotplug,
2789 };
2790
2791 static int vm_stat_get(void *_offset, u64 *val)
2792 {
2793         unsigned offset = (long)_offset;
2794         struct kvm *kvm;
2795
2796         *val = 0;
2797         raw_spin_lock(&kvm_lock);
2798         list_for_each_entry(kvm, &vm_list, vm_list)
2799                 *val += *(u32 *)((void *)kvm + offset);
2800         raw_spin_unlock(&kvm_lock);
2801         return 0;
2802 }
2803
2804 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2805
2806 static int vcpu_stat_get(void *_offset, u64 *val)
2807 {
2808         unsigned offset = (long)_offset;
2809         struct kvm *kvm;
2810         struct kvm_vcpu *vcpu;
2811         int i;
2812
2813         *val = 0;
2814         raw_spin_lock(&kvm_lock);
2815         list_for_each_entry(kvm, &vm_list, vm_list)
2816                 kvm_for_each_vcpu(i, vcpu, kvm)
2817                         *val += *(u32 *)((void *)vcpu + offset);
2818
2819         raw_spin_unlock(&kvm_lock);
2820         return 0;
2821 }
2822
2823 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2824
2825 static const struct file_operations *stat_fops[] = {
2826         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2827         [KVM_STAT_VM]   = &vm_stat_fops,
2828 };
2829
2830 static int kvm_init_debug(void)
2831 {
2832         int r = -EFAULT;
2833         struct kvm_stats_debugfs_item *p;
2834
2835         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2836         if (kvm_debugfs_dir == NULL)
2837                 goto out;
2838
2839         for (p = debugfs_entries; p->name; ++p) {
2840                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2841                                                 (void *)(long)p->offset,
2842                                                 stat_fops[p->kind]);
2843                 if (p->dentry == NULL)
2844                         goto out_dir;
2845         }
2846
2847         return 0;
2848
2849 out_dir:
2850         debugfs_remove_recursive(kvm_debugfs_dir);
2851 out:
2852         return r;
2853 }
2854
2855 static void kvm_exit_debug(void)
2856 {
2857         struct kvm_stats_debugfs_item *p;
2858
2859         for (p = debugfs_entries; p->name; ++p)
2860                 debugfs_remove(p->dentry);
2861         debugfs_remove(kvm_debugfs_dir);
2862 }
2863
2864 static int kvm_suspend(void)
2865 {
2866         if (kvm_usage_count)
2867                 hardware_disable_nolock(NULL);
2868         return 0;
2869 }
2870
2871 static void kvm_resume(void)
2872 {
2873         if (kvm_usage_count) {
2874                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2875                 hardware_enable_nolock(NULL);
2876         }
2877 }
2878
2879 static struct syscore_ops kvm_syscore_ops = {
2880         .suspend = kvm_suspend,
2881         .resume = kvm_resume,
2882 };
2883
2884 static inline
2885 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2886 {
2887         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2888 }
2889
2890 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2891 {
2892         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2893
2894         kvm_arch_vcpu_load(vcpu, cpu);
2895 }
2896
2897 static void kvm_sched_out(struct preempt_notifier *pn,
2898                           struct task_struct *next)
2899 {
2900         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2901
2902         kvm_arch_vcpu_put(vcpu);
2903 }
2904
2905 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2906                   struct module *module)
2907 {
2908         int r;
2909         int cpu;
2910
2911         r = kvm_arch_init(opaque);
2912         if (r)
2913                 goto out_fail;
2914
2915         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2916                 r = -ENOMEM;
2917                 goto out_free_0;
2918         }
2919
2920         r = kvm_arch_hardware_setup();
2921         if (r < 0)
2922                 goto out_free_0a;
2923
2924         for_each_online_cpu(cpu) {
2925                 smp_call_function_single(cpu,
2926                                 kvm_arch_check_processor_compat,
2927                                 &r, 1);
2928                 if (r < 0)
2929                         goto out_free_1;
2930         }
2931
2932         r = register_cpu_notifier(&kvm_cpu_notifier);
2933         if (r)
2934                 goto out_free_2;
2935         register_reboot_notifier(&kvm_reboot_notifier);
2936
2937         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2938         if (!vcpu_align)
2939                 vcpu_align = __alignof__(struct kvm_vcpu);
2940         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2941                                            0, NULL);
2942         if (!kvm_vcpu_cache) {
2943                 r = -ENOMEM;
2944                 goto out_free_3;
2945         }
2946
2947         r = kvm_async_pf_init();
2948         if (r)
2949                 goto out_free;
2950
2951         kvm_chardev_ops.owner = module;
2952         kvm_vm_fops.owner = module;
2953         kvm_vcpu_fops.owner = module;
2954
2955         r = misc_register(&kvm_dev);
2956         if (r) {
2957                 printk(KERN_ERR "kvm: misc device register failed\n");
2958                 goto out_unreg;
2959         }
2960
2961         register_syscore_ops(&kvm_syscore_ops);
2962
2963         kvm_preempt_ops.sched_in = kvm_sched_in;
2964         kvm_preempt_ops.sched_out = kvm_sched_out;
2965
2966         r = kvm_init_debug();
2967         if (r) {
2968                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2969                 goto out_undebugfs;
2970         }
2971
2972         return 0;
2973
2974 out_undebugfs:
2975         unregister_syscore_ops(&kvm_syscore_ops);
2976 out_unreg:
2977         kvm_async_pf_deinit();
2978 out_free:
2979         kmem_cache_destroy(kvm_vcpu_cache);
2980 out_free_3:
2981         unregister_reboot_notifier(&kvm_reboot_notifier);
2982         unregister_cpu_notifier(&kvm_cpu_notifier);
2983 out_free_2:
2984 out_free_1:
2985         kvm_arch_hardware_unsetup();
2986 out_free_0a:
2987         free_cpumask_var(cpus_hardware_enabled);
2988 out_free_0:
2989         kvm_arch_exit();
2990 out_fail:
2991         return r;
2992 }
2993 EXPORT_SYMBOL_GPL(kvm_init);
2994
2995 void kvm_exit(void)
2996 {
2997         kvm_exit_debug();
2998         misc_deregister(&kvm_dev);
2999         kmem_cache_destroy(kvm_vcpu_cache);
3000         kvm_async_pf_deinit();
3001         unregister_syscore_ops(&kvm_syscore_ops);
3002         unregister_reboot_notifier(&kvm_reboot_notifier);
3003         unregister_cpu_notifier(&kvm_cpu_notifier);
3004         on_each_cpu(hardware_disable_nolock, NULL, 1);
3005         kvm_arch_hardware_unsetup();
3006         kvm_arch_exit();
3007         free_cpumask_var(cpus_hardware_enabled);
3008 }
3009 EXPORT_SYMBOL_GPL(kvm_exit);