3caf8162eb6b191f74a2cf2f012184717b7ab163
[linux-3.10.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 bool kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 int vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         if (mutex_lock_killable(&vcpu->mutex))
139                 return -EINTR;
140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141                 /* The thread running this VCPU changed. */
142                 struct pid *oldpid = vcpu->pid;
143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144                 rcu_assign_pointer(vcpu->pid, newpid);
145                 synchronize_rcu();
146                 put_pid(oldpid);
147         }
148         cpu = get_cpu();
149         preempt_notifier_register(&vcpu->preempt_notifier);
150         kvm_arch_vcpu_load(vcpu, cpu);
151         put_cpu();
152         return 0;
153 }
154
155 void vcpu_put(struct kvm_vcpu *vcpu)
156 {
157         preempt_disable();
158         kvm_arch_vcpu_put(vcpu);
159         preempt_notifier_unregister(&vcpu->preempt_notifier);
160         preempt_enable();
161         mutex_unlock(&vcpu->mutex);
162 }
163
164 static void ack_flush(void *_completed)
165 {
166 }
167
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
169 {
170         int i, cpu, me;
171         cpumask_var_t cpus;
172         bool called = true;
173         struct kvm_vcpu *vcpu;
174
175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176
177         me = get_cpu();
178         kvm_for_each_vcpu(i, vcpu, kvm) {
179                 kvm_make_request(req, vcpu);
180                 cpu = vcpu->cpu;
181
182                 /* Set ->requests bit before we read ->mode */
183                 smp_mb();
184
185                 if (cpus != NULL && cpu != -1 && cpu != me &&
186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187                         cpumask_set_cpu(cpu, cpus);
188         }
189         if (unlikely(cpus == NULL))
190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191         else if (!cpumask_empty(cpus))
192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
193         else
194                 called = false;
195         put_cpu();
196         free_cpumask_var(cpus);
197         return called;
198 }
199
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
201 {
202         long dirty_count = kvm->tlbs_dirty;
203
204         smp_mb();
205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206                 ++kvm->stat.remote_tlb_flush;
207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 }
209
210 void kvm_reload_remote_mmus(struct kvm *kvm)
211 {
212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 }
214
215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216 {
217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218 }
219
220 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
221 {
222         struct page *page;
223         int r;
224
225         mutex_init(&vcpu->mutex);
226         vcpu->cpu = -1;
227         vcpu->kvm = kvm;
228         vcpu->vcpu_id = id;
229         vcpu->pid = NULL;
230         init_waitqueue_head(&vcpu->wq);
231         kvm_async_pf_vcpu_init(vcpu);
232
233         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
234         if (!page) {
235                 r = -ENOMEM;
236                 goto fail;
237         }
238         vcpu->run = page_address(page);
239
240         kvm_vcpu_set_in_spin_loop(vcpu, false);
241         kvm_vcpu_set_dy_eligible(vcpu, false);
242
243         r = kvm_arch_vcpu_init(vcpu);
244         if (r < 0)
245                 goto fail_free_run;
246         return 0;
247
248 fail_free_run:
249         free_page((unsigned long)vcpu->run);
250 fail:
251         return r;
252 }
253 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
254
255 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
256 {
257         put_pid(vcpu->pid);
258         kvm_arch_vcpu_uninit(vcpu);
259         free_page((unsigned long)vcpu->run);
260 }
261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
262
263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
265 {
266         return container_of(mn, struct kvm, mmu_notifier);
267 }
268
269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
270                                              struct mm_struct *mm,
271                                              unsigned long address)
272 {
273         struct kvm *kvm = mmu_notifier_to_kvm(mn);
274         int need_tlb_flush, idx;
275
276         /*
277          * When ->invalidate_page runs, the linux pte has been zapped
278          * already but the page is still allocated until
279          * ->invalidate_page returns. So if we increase the sequence
280          * here the kvm page fault will notice if the spte can't be
281          * established because the page is going to be freed. If
282          * instead the kvm page fault establishes the spte before
283          * ->invalidate_page runs, kvm_unmap_hva will release it
284          * before returning.
285          *
286          * The sequence increase only need to be seen at spin_unlock
287          * time, and not at spin_lock time.
288          *
289          * Increasing the sequence after the spin_unlock would be
290          * unsafe because the kvm page fault could then establish the
291          * pte after kvm_unmap_hva returned, without noticing the page
292          * is going to be freed.
293          */
294         idx = srcu_read_lock(&kvm->srcu);
295         spin_lock(&kvm->mmu_lock);
296
297         kvm->mmu_notifier_seq++;
298         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
299         /* we've to flush the tlb before the pages can be freed */
300         if (need_tlb_flush)
301                 kvm_flush_remote_tlbs(kvm);
302
303         spin_unlock(&kvm->mmu_lock);
304         srcu_read_unlock(&kvm->srcu, idx);
305 }
306
307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
308                                         struct mm_struct *mm,
309                                         unsigned long address,
310                                         pte_t pte)
311 {
312         struct kvm *kvm = mmu_notifier_to_kvm(mn);
313         int idx;
314
315         idx = srcu_read_lock(&kvm->srcu);
316         spin_lock(&kvm->mmu_lock);
317         kvm->mmu_notifier_seq++;
318         kvm_set_spte_hva(kvm, address, pte);
319         spin_unlock(&kvm->mmu_lock);
320         srcu_read_unlock(&kvm->srcu, idx);
321 }
322
323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
324                                                     struct mm_struct *mm,
325                                                     unsigned long start,
326                                                     unsigned long end)
327 {
328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
329         int need_tlb_flush = 0, idx;
330
331         idx = srcu_read_lock(&kvm->srcu);
332         spin_lock(&kvm->mmu_lock);
333         /*
334          * The count increase must become visible at unlock time as no
335          * spte can be established without taking the mmu_lock and
336          * count is also read inside the mmu_lock critical section.
337          */
338         kvm->mmu_notifier_count++;
339         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
340         need_tlb_flush |= kvm->tlbs_dirty;
341         /* we've to flush the tlb before the pages can be freed */
342         if (need_tlb_flush)
343                 kvm_flush_remote_tlbs(kvm);
344
345         spin_unlock(&kvm->mmu_lock);
346         srcu_read_unlock(&kvm->srcu, idx);
347 }
348
349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
350                                                   struct mm_struct *mm,
351                                                   unsigned long start,
352                                                   unsigned long end)
353 {
354         struct kvm *kvm = mmu_notifier_to_kvm(mn);
355
356         spin_lock(&kvm->mmu_lock);
357         /*
358          * This sequence increase will notify the kvm page fault that
359          * the page that is going to be mapped in the spte could have
360          * been freed.
361          */
362         kvm->mmu_notifier_seq++;
363         smp_wmb();
364         /*
365          * The above sequence increase must be visible before the
366          * below count decrease, which is ensured by the smp_wmb above
367          * in conjunction with the smp_rmb in mmu_notifier_retry().
368          */
369         kvm->mmu_notifier_count--;
370         spin_unlock(&kvm->mmu_lock);
371
372         BUG_ON(kvm->mmu_notifier_count < 0);
373 }
374
375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
376                                               struct mm_struct *mm,
377                                               unsigned long address)
378 {
379         struct kvm *kvm = mmu_notifier_to_kvm(mn);
380         int young, idx;
381
382         idx = srcu_read_lock(&kvm->srcu);
383         spin_lock(&kvm->mmu_lock);
384
385         young = kvm_age_hva(kvm, address);
386         if (young)
387                 kvm_flush_remote_tlbs(kvm);
388
389         spin_unlock(&kvm->mmu_lock);
390         srcu_read_unlock(&kvm->srcu, idx);
391
392         return young;
393 }
394
395 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
396                                        struct mm_struct *mm,
397                                        unsigned long address)
398 {
399         struct kvm *kvm = mmu_notifier_to_kvm(mn);
400         int young, idx;
401
402         idx = srcu_read_lock(&kvm->srcu);
403         spin_lock(&kvm->mmu_lock);
404         young = kvm_test_age_hva(kvm, address);
405         spin_unlock(&kvm->mmu_lock);
406         srcu_read_unlock(&kvm->srcu, idx);
407
408         return young;
409 }
410
411 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
412                                      struct mm_struct *mm)
413 {
414         struct kvm *kvm = mmu_notifier_to_kvm(mn);
415         int idx;
416
417         idx = srcu_read_lock(&kvm->srcu);
418         kvm_arch_flush_shadow_all(kvm);
419         srcu_read_unlock(&kvm->srcu, idx);
420 }
421
422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
423         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
424         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
425         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
426         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
427         .test_young             = kvm_mmu_notifier_test_young,
428         .change_pte             = kvm_mmu_notifier_change_pte,
429         .release                = kvm_mmu_notifier_release,
430 };
431
432 static int kvm_init_mmu_notifier(struct kvm *kvm)
433 {
434         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
435         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
436 }
437
438 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
439
440 static int kvm_init_mmu_notifier(struct kvm *kvm)
441 {
442         return 0;
443 }
444
445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
446
447 static void kvm_init_memslots_id(struct kvm *kvm)
448 {
449         int i;
450         struct kvm_memslots *slots = kvm->memslots;
451
452         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
453                 slots->id_to_index[i] = slots->memslots[i].id = i;
454 }
455
456 static struct kvm *kvm_create_vm(unsigned long type)
457 {
458         int r, i;
459         struct kvm *kvm = kvm_arch_alloc_vm();
460
461         if (!kvm)
462                 return ERR_PTR(-ENOMEM);
463
464         r = kvm_arch_init_vm(kvm, type);
465         if (r)
466                 goto out_err_nodisable;
467
468         r = hardware_enable_all();
469         if (r)
470                 goto out_err_nodisable;
471
472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
473         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
474         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
475 #endif
476
477         r = -ENOMEM;
478         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
479         if (!kvm->memslots)
480                 goto out_err_nosrcu;
481         kvm_init_memslots_id(kvm);
482         if (init_srcu_struct(&kvm->srcu))
483                 goto out_err_nosrcu;
484         for (i = 0; i < KVM_NR_BUSES; i++) {
485                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
486                                         GFP_KERNEL);
487                 if (!kvm->buses[i])
488                         goto out_err;
489         }
490
491         spin_lock_init(&kvm->mmu_lock);
492         kvm->mm = current->mm;
493         atomic_inc(&kvm->mm->mm_count);
494         kvm_eventfd_init(kvm);
495         mutex_init(&kvm->lock);
496         mutex_init(&kvm->irq_lock);
497         mutex_init(&kvm->slots_lock);
498         atomic_set(&kvm->users_count, 1);
499
500         r = kvm_init_mmu_notifier(kvm);
501         if (r)
502                 goto out_err;
503
504         raw_spin_lock(&kvm_lock);
505         list_add(&kvm->vm_list, &vm_list);
506         raw_spin_unlock(&kvm_lock);
507
508         return kvm;
509
510 out_err:
511         cleanup_srcu_struct(&kvm->srcu);
512 out_err_nosrcu:
513         hardware_disable_all();
514 out_err_nodisable:
515         for (i = 0; i < KVM_NR_BUSES; i++)
516                 kfree(kvm->buses[i]);
517         kfree(kvm->memslots);
518         kvm_arch_free_vm(kvm);
519         return ERR_PTR(r);
520 }
521
522 /*
523  * Avoid using vmalloc for a small buffer.
524  * Should not be used when the size is statically known.
525  */
526 void *kvm_kvzalloc(unsigned long size)
527 {
528         if (size > PAGE_SIZE)
529                 return vzalloc(size);
530         else
531                 return kzalloc(size, GFP_KERNEL);
532 }
533
534 void kvm_kvfree(const void *addr)
535 {
536         if (is_vmalloc_addr(addr))
537                 vfree(addr);
538         else
539                 kfree(addr);
540 }
541
542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
543 {
544         if (!memslot->dirty_bitmap)
545                 return;
546
547         kvm_kvfree(memslot->dirty_bitmap);
548         memslot->dirty_bitmap = NULL;
549 }
550
551 /*
552  * Free any memory in @free but not in @dont.
553  */
554 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
555                                   struct kvm_memory_slot *dont)
556 {
557         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
558                 kvm_destroy_dirty_bitmap(free);
559
560         kvm_arch_free_memslot(free, dont);
561
562         free->npages = 0;
563 }
564
565 void kvm_free_physmem(struct kvm *kvm)
566 {
567         struct kvm_memslots *slots = kvm->memslots;
568         struct kvm_memory_slot *memslot;
569
570         kvm_for_each_memslot(memslot, slots)
571                 kvm_free_physmem_slot(memslot, NULL);
572
573         kfree(kvm->memslots);
574 }
575
576 static void kvm_destroy_vm(struct kvm *kvm)
577 {
578         int i;
579         struct mm_struct *mm = kvm->mm;
580
581         kvm_arch_sync_events(kvm);
582         raw_spin_lock(&kvm_lock);
583         list_del(&kvm->vm_list);
584         raw_spin_unlock(&kvm_lock);
585         kvm_free_irq_routing(kvm);
586         for (i = 0; i < KVM_NR_BUSES; i++)
587                 kvm_io_bus_destroy(kvm->buses[i]);
588         kvm_coalesced_mmio_free(kvm);
589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
590         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
591 #else
592         kvm_arch_flush_shadow_all(kvm);
593 #endif
594         kvm_arch_destroy_vm(kvm);
595         kvm_free_physmem(kvm);
596         cleanup_srcu_struct(&kvm->srcu);
597         kvm_arch_free_vm(kvm);
598         hardware_disable_all();
599         mmdrop(mm);
600 }
601
602 void kvm_get_kvm(struct kvm *kvm)
603 {
604         atomic_inc(&kvm->users_count);
605 }
606 EXPORT_SYMBOL_GPL(kvm_get_kvm);
607
608 void kvm_put_kvm(struct kvm *kvm)
609 {
610         if (atomic_dec_and_test(&kvm->users_count))
611                 kvm_destroy_vm(kvm);
612 }
613 EXPORT_SYMBOL_GPL(kvm_put_kvm);
614
615
616 static int kvm_vm_release(struct inode *inode, struct file *filp)
617 {
618         struct kvm *kvm = filp->private_data;
619
620         kvm_irqfd_release(kvm);
621
622         kvm_put_kvm(kvm);
623         return 0;
624 }
625
626 /*
627  * Allocation size is twice as large as the actual dirty bitmap size.
628  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
629  */
630 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
631 {
632 #ifndef CONFIG_S390
633         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
634
635         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
636         if (!memslot->dirty_bitmap)
637                 return -ENOMEM;
638
639 #endif /* !CONFIG_S390 */
640         return 0;
641 }
642
643 static int cmp_memslot(const void *slot1, const void *slot2)
644 {
645         struct kvm_memory_slot *s1, *s2;
646
647         s1 = (struct kvm_memory_slot *)slot1;
648         s2 = (struct kvm_memory_slot *)slot2;
649
650         if (s1->npages < s2->npages)
651                 return 1;
652         if (s1->npages > s2->npages)
653                 return -1;
654
655         return 0;
656 }
657
658 /*
659  * Sort the memslots base on its size, so the larger slots
660  * will get better fit.
661  */
662 static void sort_memslots(struct kvm_memslots *slots)
663 {
664         int i;
665
666         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
667               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
668
669         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
670                 slots->id_to_index[slots->memslots[i].id] = i;
671 }
672
673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
674 {
675         if (new) {
676                 int id = new->id;
677                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
678                 unsigned long npages = old->npages;
679
680                 *old = *new;
681                 if (new->npages != npages)
682                         sort_memslots(slots);
683         }
684
685         slots->generation++;
686 }
687
688 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
689 {
690         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
691
692 #ifdef KVM_CAP_READONLY_MEM
693         valid_flags |= KVM_MEM_READONLY;
694 #endif
695
696         if (mem->flags & ~valid_flags)
697                 return -EINVAL;
698
699         return 0;
700 }
701
702 /*
703  * Allocate some memory and give it an address in the guest physical address
704  * space.
705  *
706  * Discontiguous memory is allowed, mostly for framebuffers.
707  *
708  * Must be called holding mmap_sem for write.
709  */
710 int __kvm_set_memory_region(struct kvm *kvm,
711                             struct kvm_userspace_memory_region *mem,
712                             int user_alloc)
713 {
714         int r;
715         gfn_t base_gfn;
716         unsigned long npages;
717         struct kvm_memory_slot *memslot, *slot;
718         struct kvm_memory_slot old, new;
719         struct kvm_memslots *slots, *old_memslots;
720
721         r = check_memory_region_flags(mem);
722         if (r)
723                 goto out;
724
725         r = -EINVAL;
726         /* General sanity checks */
727         if (mem->memory_size & (PAGE_SIZE - 1))
728                 goto out;
729         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
730                 goto out;
731         /* We can read the guest memory with __xxx_user() later on. */
732         if (user_alloc &&
733             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
734              !access_ok(VERIFY_WRITE,
735                         (void __user *)(unsigned long)mem->userspace_addr,
736                         mem->memory_size)))
737                 goto out;
738         if (mem->slot >= KVM_MEM_SLOTS_NUM)
739                 goto out;
740         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
741                 goto out;
742
743         memslot = id_to_memslot(kvm->memslots, mem->slot);
744         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
745         npages = mem->memory_size >> PAGE_SHIFT;
746
747         r = -EINVAL;
748         if (npages > KVM_MEM_MAX_NR_PAGES)
749                 goto out;
750
751         if (!npages)
752                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
753
754         new = old = *memslot;
755
756         new.id = mem->slot;
757         new.base_gfn = base_gfn;
758         new.npages = npages;
759         new.flags = mem->flags;
760
761         /*
762          * Disallow changing a memory slot's size or changing anything about
763          * zero sized slots that doesn't involve making them non-zero.
764          */
765         r = -EINVAL;
766         if (npages && old.npages && npages != old.npages)
767                 goto out_free;
768         if (!npages && !old.npages)
769                 goto out_free;
770
771         /* Check for overlaps */
772         r = -EEXIST;
773         kvm_for_each_memslot(slot, kvm->memslots) {
774                 if (slot->id >= KVM_MEMORY_SLOTS || slot == memslot)
775                         continue;
776                 if (!((base_gfn + npages <= slot->base_gfn) ||
777                       (base_gfn >= slot->base_gfn + slot->npages)))
778                         goto out_free;
779         }
780
781         /* Free page dirty bitmap if unneeded */
782         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
783                 new.dirty_bitmap = NULL;
784
785         r = -ENOMEM;
786
787         /* Allocate if a slot is being created */
788         if (!old.npages) {
789                 new.user_alloc = user_alloc;
790                 new.userspace_addr = mem->userspace_addr;
791
792                 if (kvm_arch_create_memslot(&new, npages))
793                         goto out_free;
794         }
795
796         /* Allocate page dirty bitmap if needed */
797         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
798                 if (kvm_create_dirty_bitmap(&new) < 0)
799                         goto out_free;
800                 /* destroy any largepage mappings for dirty tracking */
801         }
802
803         if (!npages || base_gfn != old.base_gfn) {
804                 struct kvm_memory_slot *slot;
805
806                 r = -ENOMEM;
807                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
808                                 GFP_KERNEL);
809                 if (!slots)
810                         goto out_free;
811                 slot = id_to_memslot(slots, mem->slot);
812                 slot->flags |= KVM_MEMSLOT_INVALID;
813
814                 update_memslots(slots, NULL);
815
816                 old_memslots = kvm->memslots;
817                 rcu_assign_pointer(kvm->memslots, slots);
818                 synchronize_srcu_expedited(&kvm->srcu);
819                 /* From this point no new shadow pages pointing to a deleted,
820                  * or moved, memslot will be created.
821                  *
822                  * validation of sp->gfn happens in:
823                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
824                  *      - kvm_is_visible_gfn (mmu_check_roots)
825                  */
826                 kvm_arch_flush_shadow_memslot(kvm, slot);
827                 kfree(old_memslots);
828         }
829
830         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
831         if (r)
832                 goto out_free;
833
834         /* map/unmap the pages in iommu page table */
835         if (npages) {
836                 r = kvm_iommu_map_pages(kvm, &new);
837                 if (r)
838                         goto out_free;
839         } else
840                 kvm_iommu_unmap_pages(kvm, &old);
841
842         r = -ENOMEM;
843         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
844                         GFP_KERNEL);
845         if (!slots)
846                 goto out_free;
847
848         /* actual memory is freed via old in kvm_free_physmem_slot below */
849         if (!npages) {
850                 new.dirty_bitmap = NULL;
851                 memset(&new.arch, 0, sizeof(new.arch));
852         }
853
854         update_memslots(slots, &new);
855         old_memslots = kvm->memslots;
856         rcu_assign_pointer(kvm->memslots, slots);
857         synchronize_srcu_expedited(&kvm->srcu);
858
859         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
860
861         kvm_free_physmem_slot(&old, &new);
862         kfree(old_memslots);
863
864         return 0;
865
866 out_free:
867         kvm_free_physmem_slot(&new, &old);
868 out:
869         return r;
870
871 }
872 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
873
874 int kvm_set_memory_region(struct kvm *kvm,
875                           struct kvm_userspace_memory_region *mem,
876                           int user_alloc)
877 {
878         int r;
879
880         mutex_lock(&kvm->slots_lock);
881         r = __kvm_set_memory_region(kvm, mem, user_alloc);
882         mutex_unlock(&kvm->slots_lock);
883         return r;
884 }
885 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
886
887 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
888                                    struct
889                                    kvm_userspace_memory_region *mem,
890                                    int user_alloc)
891 {
892         if (mem->slot >= KVM_MEMORY_SLOTS)
893                 return -EINVAL;
894         return kvm_set_memory_region(kvm, mem, user_alloc);
895 }
896
897 int kvm_get_dirty_log(struct kvm *kvm,
898                         struct kvm_dirty_log *log, int *is_dirty)
899 {
900         struct kvm_memory_slot *memslot;
901         int r, i;
902         unsigned long n;
903         unsigned long any = 0;
904
905         r = -EINVAL;
906         if (log->slot >= KVM_MEMORY_SLOTS)
907                 goto out;
908
909         memslot = id_to_memslot(kvm->memslots, log->slot);
910         r = -ENOENT;
911         if (!memslot->dirty_bitmap)
912                 goto out;
913
914         n = kvm_dirty_bitmap_bytes(memslot);
915
916         for (i = 0; !any && i < n/sizeof(long); ++i)
917                 any = memslot->dirty_bitmap[i];
918
919         r = -EFAULT;
920         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
921                 goto out;
922
923         if (any)
924                 *is_dirty = 1;
925
926         r = 0;
927 out:
928         return r;
929 }
930
931 bool kvm_largepages_enabled(void)
932 {
933         return largepages_enabled;
934 }
935
936 void kvm_disable_largepages(void)
937 {
938         largepages_enabled = false;
939 }
940 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
941
942 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
943 {
944         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
945 }
946 EXPORT_SYMBOL_GPL(gfn_to_memslot);
947
948 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
949 {
950         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
951
952         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
953               memslot->flags & KVM_MEMSLOT_INVALID)
954                 return 0;
955
956         return 1;
957 }
958 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
959
960 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
961 {
962         struct vm_area_struct *vma;
963         unsigned long addr, size;
964
965         size = PAGE_SIZE;
966
967         addr = gfn_to_hva(kvm, gfn);
968         if (kvm_is_error_hva(addr))
969                 return PAGE_SIZE;
970
971         down_read(&current->mm->mmap_sem);
972         vma = find_vma(current->mm, addr);
973         if (!vma)
974                 goto out;
975
976         size = vma_kernel_pagesize(vma);
977
978 out:
979         up_read(&current->mm->mmap_sem);
980
981         return size;
982 }
983
984 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
985 {
986         return slot->flags & KVM_MEM_READONLY;
987 }
988
989 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
990                                        gfn_t *nr_pages, bool write)
991 {
992         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
993                 return KVM_HVA_ERR_BAD;
994
995         if (memslot_is_readonly(slot) && write)
996                 return KVM_HVA_ERR_RO_BAD;
997
998         if (nr_pages)
999                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1000
1001         return __gfn_to_hva_memslot(slot, gfn);
1002 }
1003
1004 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1005                                      gfn_t *nr_pages)
1006 {
1007         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1008 }
1009
1010 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1011                                  gfn_t gfn)
1012 {
1013         return gfn_to_hva_many(slot, gfn, NULL);
1014 }
1015 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1016
1017 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1018 {
1019         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1020 }
1021 EXPORT_SYMBOL_GPL(gfn_to_hva);
1022
1023 /*
1024  * The hva returned by this function is only allowed to be read.
1025  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1026  */
1027 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1028 {
1029         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1030 }
1031
1032 static int kvm_read_hva(void *data, void __user *hva, int len)
1033 {
1034         return __copy_from_user(data, hva, len);
1035 }
1036
1037 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1038 {
1039         return __copy_from_user_inatomic(data, hva, len);
1040 }
1041
1042 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1043         unsigned long start, int write, struct page **page)
1044 {
1045         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1046
1047         if (write)
1048                 flags |= FOLL_WRITE;
1049
1050         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1051 }
1052
1053 static inline int check_user_page_hwpoison(unsigned long addr)
1054 {
1055         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1056
1057         rc = __get_user_pages(current, current->mm, addr, 1,
1058                               flags, NULL, NULL, NULL);
1059         return rc == -EHWPOISON;
1060 }
1061
1062 /*
1063  * The atomic path to get the writable pfn which will be stored in @pfn,
1064  * true indicates success, otherwise false is returned.
1065  */
1066 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1067                             bool write_fault, bool *writable, pfn_t *pfn)
1068 {
1069         struct page *page[1];
1070         int npages;
1071
1072         if (!(async || atomic))
1073                 return false;
1074
1075         /*
1076          * Fast pin a writable pfn only if it is a write fault request
1077          * or the caller allows to map a writable pfn for a read fault
1078          * request.
1079          */
1080         if (!(write_fault || writable))
1081                 return false;
1082
1083         npages = __get_user_pages_fast(addr, 1, 1, page);
1084         if (npages == 1) {
1085                 *pfn = page_to_pfn(page[0]);
1086
1087                 if (writable)
1088                         *writable = true;
1089                 return true;
1090         }
1091
1092         return false;
1093 }
1094
1095 /*
1096  * The slow path to get the pfn of the specified host virtual address,
1097  * 1 indicates success, -errno is returned if error is detected.
1098  */
1099 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1100                            bool *writable, pfn_t *pfn)
1101 {
1102         struct page *page[1];
1103         int npages = 0;
1104
1105         might_sleep();
1106
1107         if (writable)
1108                 *writable = write_fault;
1109
1110         if (async) {
1111                 down_read(&current->mm->mmap_sem);
1112                 npages = get_user_page_nowait(current, current->mm,
1113                                               addr, write_fault, page);
1114                 up_read(&current->mm->mmap_sem);
1115         } else
1116                 npages = get_user_pages_fast(addr, 1, write_fault,
1117                                              page);
1118         if (npages != 1)
1119                 return npages;
1120
1121         /* map read fault as writable if possible */
1122         if (unlikely(!write_fault) && writable) {
1123                 struct page *wpage[1];
1124
1125                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1126                 if (npages == 1) {
1127                         *writable = true;
1128                         put_page(page[0]);
1129                         page[0] = wpage[0];
1130                 }
1131
1132                 npages = 1;
1133         }
1134         *pfn = page_to_pfn(page[0]);
1135         return npages;
1136 }
1137
1138 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1139 {
1140         if (unlikely(!(vma->vm_flags & VM_READ)))
1141                 return false;
1142
1143         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1144                 return false;
1145
1146         return true;
1147 }
1148
1149 /*
1150  * Pin guest page in memory and return its pfn.
1151  * @addr: host virtual address which maps memory to the guest
1152  * @atomic: whether this function can sleep
1153  * @async: whether this function need to wait IO complete if the
1154  *         host page is not in the memory
1155  * @write_fault: whether we should get a writable host page
1156  * @writable: whether it allows to map a writable host page for !@write_fault
1157  *
1158  * The function will map a writable host page for these two cases:
1159  * 1): @write_fault = true
1160  * 2): @write_fault = false && @writable, @writable will tell the caller
1161  *     whether the mapping is writable.
1162  */
1163 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1164                         bool write_fault, bool *writable)
1165 {
1166         struct vm_area_struct *vma;
1167         pfn_t pfn = 0;
1168         int npages;
1169
1170         /* we can do it either atomically or asynchronously, not both */
1171         BUG_ON(atomic && async);
1172
1173         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1174                 return pfn;
1175
1176         if (atomic)
1177                 return KVM_PFN_ERR_FAULT;
1178
1179         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1180         if (npages == 1)
1181                 return pfn;
1182
1183         down_read(&current->mm->mmap_sem);
1184         if (npages == -EHWPOISON ||
1185               (!async && check_user_page_hwpoison(addr))) {
1186                 pfn = KVM_PFN_ERR_HWPOISON;
1187                 goto exit;
1188         }
1189
1190         vma = find_vma_intersection(current->mm, addr, addr + 1);
1191
1192         if (vma == NULL)
1193                 pfn = KVM_PFN_ERR_FAULT;
1194         else if ((vma->vm_flags & VM_PFNMAP)) {
1195                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1196                         vma->vm_pgoff;
1197                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1198         } else {
1199                 if (async && vma_is_valid(vma, write_fault))
1200                         *async = true;
1201                 pfn = KVM_PFN_ERR_FAULT;
1202         }
1203 exit:
1204         up_read(&current->mm->mmap_sem);
1205         return pfn;
1206 }
1207
1208 static pfn_t
1209 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1210                      bool *async, bool write_fault, bool *writable)
1211 {
1212         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1213
1214         if (addr == KVM_HVA_ERR_RO_BAD)
1215                 return KVM_PFN_ERR_RO_FAULT;
1216
1217         if (kvm_is_error_hva(addr))
1218                 return KVM_PFN_NOSLOT;
1219
1220         /* Do not map writable pfn in the readonly memslot. */
1221         if (writable && memslot_is_readonly(slot)) {
1222                 *writable = false;
1223                 writable = NULL;
1224         }
1225
1226         return hva_to_pfn(addr, atomic, async, write_fault,
1227                           writable);
1228 }
1229
1230 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1231                           bool write_fault, bool *writable)
1232 {
1233         struct kvm_memory_slot *slot;
1234
1235         if (async)
1236                 *async = false;
1237
1238         slot = gfn_to_memslot(kvm, gfn);
1239
1240         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1241                                     writable);
1242 }
1243
1244 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1245 {
1246         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1247 }
1248 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1249
1250 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1251                        bool write_fault, bool *writable)
1252 {
1253         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1254 }
1255 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1256
1257 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1258 {
1259         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1260 }
1261 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1262
1263 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1264                       bool *writable)
1265 {
1266         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1267 }
1268 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1269
1270 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1271 {
1272         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1273 }
1274
1275 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1276 {
1277         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1278 }
1279 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1280
1281 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1282                                                                   int nr_pages)
1283 {
1284         unsigned long addr;
1285         gfn_t entry;
1286
1287         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1288         if (kvm_is_error_hva(addr))
1289                 return -1;
1290
1291         if (entry < nr_pages)
1292                 return 0;
1293
1294         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1295 }
1296 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1297
1298 static struct page *kvm_pfn_to_page(pfn_t pfn)
1299 {
1300         if (is_error_noslot_pfn(pfn))
1301                 return KVM_ERR_PTR_BAD_PAGE;
1302
1303         if (kvm_is_mmio_pfn(pfn)) {
1304                 WARN_ON(1);
1305                 return KVM_ERR_PTR_BAD_PAGE;
1306         }
1307
1308         return pfn_to_page(pfn);
1309 }
1310
1311 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1312 {
1313         pfn_t pfn;
1314
1315         pfn = gfn_to_pfn(kvm, gfn);
1316
1317         return kvm_pfn_to_page(pfn);
1318 }
1319
1320 EXPORT_SYMBOL_GPL(gfn_to_page);
1321
1322 void kvm_release_page_clean(struct page *page)
1323 {
1324         WARN_ON(is_error_page(page));
1325
1326         kvm_release_pfn_clean(page_to_pfn(page));
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1329
1330 void kvm_release_pfn_clean(pfn_t pfn)
1331 {
1332         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1333                 put_page(pfn_to_page(pfn));
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1336
1337 void kvm_release_page_dirty(struct page *page)
1338 {
1339         WARN_ON(is_error_page(page));
1340
1341         kvm_release_pfn_dirty(page_to_pfn(page));
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1344
1345 void kvm_release_pfn_dirty(pfn_t pfn)
1346 {
1347         kvm_set_pfn_dirty(pfn);
1348         kvm_release_pfn_clean(pfn);
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1351
1352 void kvm_set_page_dirty(struct page *page)
1353 {
1354         kvm_set_pfn_dirty(page_to_pfn(page));
1355 }
1356 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1357
1358 void kvm_set_pfn_dirty(pfn_t pfn)
1359 {
1360         if (!kvm_is_mmio_pfn(pfn)) {
1361                 struct page *page = pfn_to_page(pfn);
1362                 if (!PageReserved(page))
1363                         SetPageDirty(page);
1364         }
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1367
1368 void kvm_set_pfn_accessed(pfn_t pfn)
1369 {
1370         if (!kvm_is_mmio_pfn(pfn))
1371                 mark_page_accessed(pfn_to_page(pfn));
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1374
1375 void kvm_get_pfn(pfn_t pfn)
1376 {
1377         if (!kvm_is_mmio_pfn(pfn))
1378                 get_page(pfn_to_page(pfn));
1379 }
1380 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1381
1382 static int next_segment(unsigned long len, int offset)
1383 {
1384         if (len > PAGE_SIZE - offset)
1385                 return PAGE_SIZE - offset;
1386         else
1387                 return len;
1388 }
1389
1390 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1391                         int len)
1392 {
1393         int r;
1394         unsigned long addr;
1395
1396         addr = gfn_to_hva_read(kvm, gfn);
1397         if (kvm_is_error_hva(addr))
1398                 return -EFAULT;
1399         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1400         if (r)
1401                 return -EFAULT;
1402         return 0;
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1405
1406 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1407 {
1408         gfn_t gfn = gpa >> PAGE_SHIFT;
1409         int seg;
1410         int offset = offset_in_page(gpa);
1411         int ret;
1412
1413         while ((seg = next_segment(len, offset)) != 0) {
1414                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1415                 if (ret < 0)
1416                         return ret;
1417                 offset = 0;
1418                 len -= seg;
1419                 data += seg;
1420                 ++gfn;
1421         }
1422         return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_read_guest);
1425
1426 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1427                           unsigned long len)
1428 {
1429         int r;
1430         unsigned long addr;
1431         gfn_t gfn = gpa >> PAGE_SHIFT;
1432         int offset = offset_in_page(gpa);
1433
1434         addr = gfn_to_hva_read(kvm, gfn);
1435         if (kvm_is_error_hva(addr))
1436                 return -EFAULT;
1437         pagefault_disable();
1438         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1439         pagefault_enable();
1440         if (r)
1441                 return -EFAULT;
1442         return 0;
1443 }
1444 EXPORT_SYMBOL(kvm_read_guest_atomic);
1445
1446 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1447                          int offset, int len)
1448 {
1449         int r;
1450         unsigned long addr;
1451
1452         addr = gfn_to_hva(kvm, gfn);
1453         if (kvm_is_error_hva(addr))
1454                 return -EFAULT;
1455         r = __copy_to_user((void __user *)addr + offset, data, len);
1456         if (r)
1457                 return -EFAULT;
1458         mark_page_dirty(kvm, gfn);
1459         return 0;
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1462
1463 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1464                     unsigned long len)
1465 {
1466         gfn_t gfn = gpa >> PAGE_SHIFT;
1467         int seg;
1468         int offset = offset_in_page(gpa);
1469         int ret;
1470
1471         while ((seg = next_segment(len, offset)) != 0) {
1472                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1473                 if (ret < 0)
1474                         return ret;
1475                 offset = 0;
1476                 len -= seg;
1477                 data += seg;
1478                 ++gfn;
1479         }
1480         return 0;
1481 }
1482
1483 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1484                               gpa_t gpa)
1485 {
1486         struct kvm_memslots *slots = kvm_memslots(kvm);
1487         int offset = offset_in_page(gpa);
1488         gfn_t gfn = gpa >> PAGE_SHIFT;
1489
1490         ghc->gpa = gpa;
1491         ghc->generation = slots->generation;
1492         ghc->memslot = gfn_to_memslot(kvm, gfn);
1493         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1494         if (!kvm_is_error_hva(ghc->hva))
1495                 ghc->hva += offset;
1496         else
1497                 return -EFAULT;
1498
1499         return 0;
1500 }
1501 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1502
1503 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1504                            void *data, unsigned long len)
1505 {
1506         struct kvm_memslots *slots = kvm_memslots(kvm);
1507         int r;
1508
1509         if (slots->generation != ghc->generation)
1510                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1511
1512         if (kvm_is_error_hva(ghc->hva))
1513                 return -EFAULT;
1514
1515         r = __copy_to_user((void __user *)ghc->hva, data, len);
1516         if (r)
1517                 return -EFAULT;
1518         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1519
1520         return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1523
1524 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1525                            void *data, unsigned long len)
1526 {
1527         struct kvm_memslots *slots = kvm_memslots(kvm);
1528         int r;
1529
1530         if (slots->generation != ghc->generation)
1531                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1532
1533         if (kvm_is_error_hva(ghc->hva))
1534                 return -EFAULT;
1535
1536         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1537         if (r)
1538                 return -EFAULT;
1539
1540         return 0;
1541 }
1542 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1543
1544 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1545 {
1546         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1547                                     offset, len);
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1550
1551 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1552 {
1553         gfn_t gfn = gpa >> PAGE_SHIFT;
1554         int seg;
1555         int offset = offset_in_page(gpa);
1556         int ret;
1557
1558         while ((seg = next_segment(len, offset)) != 0) {
1559                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1560                 if (ret < 0)
1561                         return ret;
1562                 offset = 0;
1563                 len -= seg;
1564                 ++gfn;
1565         }
1566         return 0;
1567 }
1568 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1569
1570 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1571                              gfn_t gfn)
1572 {
1573         if (memslot && memslot->dirty_bitmap) {
1574                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1575
1576                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1577         }
1578 }
1579
1580 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1581 {
1582         struct kvm_memory_slot *memslot;
1583
1584         memslot = gfn_to_memslot(kvm, gfn);
1585         mark_page_dirty_in_slot(kvm, memslot, gfn);
1586 }
1587
1588 /*
1589  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1590  */
1591 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1592 {
1593         DEFINE_WAIT(wait);
1594
1595         for (;;) {
1596                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1597
1598                 if (kvm_arch_vcpu_runnable(vcpu)) {
1599                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1600                         break;
1601                 }
1602                 if (kvm_cpu_has_pending_timer(vcpu))
1603                         break;
1604                 if (signal_pending(current))
1605                         break;
1606
1607                 schedule();
1608         }
1609
1610         finish_wait(&vcpu->wq, &wait);
1611 }
1612
1613 #ifndef CONFIG_S390
1614 /*
1615  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1616  */
1617 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1618 {
1619         int me;
1620         int cpu = vcpu->cpu;
1621         wait_queue_head_t *wqp;
1622
1623         wqp = kvm_arch_vcpu_wq(vcpu);
1624         if (waitqueue_active(wqp)) {
1625                 wake_up_interruptible(wqp);
1626                 ++vcpu->stat.halt_wakeup;
1627         }
1628
1629         me = get_cpu();
1630         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1631                 if (kvm_arch_vcpu_should_kick(vcpu))
1632                         smp_send_reschedule(cpu);
1633         put_cpu();
1634 }
1635 #endif /* !CONFIG_S390 */
1636
1637 void kvm_resched(struct kvm_vcpu *vcpu)
1638 {
1639         if (!need_resched())
1640                 return;
1641         cond_resched();
1642 }
1643 EXPORT_SYMBOL_GPL(kvm_resched);
1644
1645 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1646 {
1647         struct pid *pid;
1648         struct task_struct *task = NULL;
1649
1650         rcu_read_lock();
1651         pid = rcu_dereference(target->pid);
1652         if (pid)
1653                 task = get_pid_task(target->pid, PIDTYPE_PID);
1654         rcu_read_unlock();
1655         if (!task)
1656                 return false;
1657         if (task->flags & PF_VCPU) {
1658                 put_task_struct(task);
1659                 return false;
1660         }
1661         if (yield_to(task, 1)) {
1662                 put_task_struct(task);
1663                 return true;
1664         }
1665         put_task_struct(task);
1666         return false;
1667 }
1668 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1669
1670 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1671 /*
1672  * Helper that checks whether a VCPU is eligible for directed yield.
1673  * Most eligible candidate to yield is decided by following heuristics:
1674  *
1675  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1676  *  (preempted lock holder), indicated by @in_spin_loop.
1677  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1678  *
1679  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1680  *  chance last time (mostly it has become eligible now since we have probably
1681  *  yielded to lockholder in last iteration. This is done by toggling
1682  *  @dy_eligible each time a VCPU checked for eligibility.)
1683  *
1684  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1685  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1686  *  burning. Giving priority for a potential lock-holder increases lock
1687  *  progress.
1688  *
1689  *  Since algorithm is based on heuristics, accessing another VCPU data without
1690  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1691  *  and continue with next VCPU and so on.
1692  */
1693 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1694 {
1695         bool eligible;
1696
1697         eligible = !vcpu->spin_loop.in_spin_loop ||
1698                         (vcpu->spin_loop.in_spin_loop &&
1699                          vcpu->spin_loop.dy_eligible);
1700
1701         if (vcpu->spin_loop.in_spin_loop)
1702                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1703
1704         return eligible;
1705 }
1706 #endif
1707 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1708 {
1709         struct kvm *kvm = me->kvm;
1710         struct kvm_vcpu *vcpu;
1711         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1712         int yielded = 0;
1713         int pass;
1714         int i;
1715
1716         kvm_vcpu_set_in_spin_loop(me, true);
1717         /*
1718          * We boost the priority of a VCPU that is runnable but not
1719          * currently running, because it got preempted by something
1720          * else and called schedule in __vcpu_run.  Hopefully that
1721          * VCPU is holding the lock that we need and will release it.
1722          * We approximate round-robin by starting at the last boosted VCPU.
1723          */
1724         for (pass = 0; pass < 2 && !yielded; pass++) {
1725                 kvm_for_each_vcpu(i, vcpu, kvm) {
1726                         if (!pass && i <= last_boosted_vcpu) {
1727                                 i = last_boosted_vcpu;
1728                                 continue;
1729                         } else if (pass && i > last_boosted_vcpu)
1730                                 break;
1731                         if (vcpu == me)
1732                                 continue;
1733                         if (waitqueue_active(&vcpu->wq))
1734                                 continue;
1735                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1736                                 continue;
1737                         if (kvm_vcpu_yield_to(vcpu)) {
1738                                 kvm->last_boosted_vcpu = i;
1739                                 yielded = 1;
1740                                 break;
1741                         }
1742                 }
1743         }
1744         kvm_vcpu_set_in_spin_loop(me, false);
1745
1746         /* Ensure vcpu is not eligible during next spinloop */
1747         kvm_vcpu_set_dy_eligible(me, false);
1748 }
1749 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1750
1751 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1752 {
1753         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1754         struct page *page;
1755
1756         if (vmf->pgoff == 0)
1757                 page = virt_to_page(vcpu->run);
1758 #ifdef CONFIG_X86
1759         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1760                 page = virt_to_page(vcpu->arch.pio_data);
1761 #endif
1762 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1763         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1764                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1765 #endif
1766         else
1767                 return kvm_arch_vcpu_fault(vcpu, vmf);
1768         get_page(page);
1769         vmf->page = page;
1770         return 0;
1771 }
1772
1773 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1774         .fault = kvm_vcpu_fault,
1775 };
1776
1777 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1778 {
1779         vma->vm_ops = &kvm_vcpu_vm_ops;
1780         return 0;
1781 }
1782
1783 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1784 {
1785         struct kvm_vcpu *vcpu = filp->private_data;
1786
1787         kvm_put_kvm(vcpu->kvm);
1788         return 0;
1789 }
1790
1791 static struct file_operations kvm_vcpu_fops = {
1792         .release        = kvm_vcpu_release,
1793         .unlocked_ioctl = kvm_vcpu_ioctl,
1794 #ifdef CONFIG_COMPAT
1795         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1796 #endif
1797         .mmap           = kvm_vcpu_mmap,
1798         .llseek         = noop_llseek,
1799 };
1800
1801 /*
1802  * Allocates an inode for the vcpu.
1803  */
1804 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1805 {
1806         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1807 }
1808
1809 /*
1810  * Creates some virtual cpus.  Good luck creating more than one.
1811  */
1812 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1813 {
1814         int r;
1815         struct kvm_vcpu *vcpu, *v;
1816
1817         vcpu = kvm_arch_vcpu_create(kvm, id);
1818         if (IS_ERR(vcpu))
1819                 return PTR_ERR(vcpu);
1820
1821         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1822
1823         r = kvm_arch_vcpu_setup(vcpu);
1824         if (r)
1825                 goto vcpu_destroy;
1826
1827         mutex_lock(&kvm->lock);
1828         if (!kvm_vcpu_compatible(vcpu)) {
1829                 r = -EINVAL;
1830                 goto unlock_vcpu_destroy;
1831         }
1832         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1833                 r = -EINVAL;
1834                 goto unlock_vcpu_destroy;
1835         }
1836
1837         kvm_for_each_vcpu(r, v, kvm)
1838                 if (v->vcpu_id == id) {
1839                         r = -EEXIST;
1840                         goto unlock_vcpu_destroy;
1841                 }
1842
1843         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1844
1845         /* Now it's all set up, let userspace reach it */
1846         kvm_get_kvm(kvm);
1847         r = create_vcpu_fd(vcpu);
1848         if (r < 0) {
1849                 kvm_put_kvm(kvm);
1850                 goto unlock_vcpu_destroy;
1851         }
1852
1853         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1854         smp_wmb();
1855         atomic_inc(&kvm->online_vcpus);
1856
1857         mutex_unlock(&kvm->lock);
1858         kvm_arch_vcpu_postcreate(vcpu);
1859         return r;
1860
1861 unlock_vcpu_destroy:
1862         mutex_unlock(&kvm->lock);
1863 vcpu_destroy:
1864         kvm_arch_vcpu_destroy(vcpu);
1865         return r;
1866 }
1867
1868 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1869 {
1870         if (sigset) {
1871                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1872                 vcpu->sigset_active = 1;
1873                 vcpu->sigset = *sigset;
1874         } else
1875                 vcpu->sigset_active = 0;
1876         return 0;
1877 }
1878
1879 static long kvm_vcpu_ioctl(struct file *filp,
1880                            unsigned int ioctl, unsigned long arg)
1881 {
1882         struct kvm_vcpu *vcpu = filp->private_data;
1883         void __user *argp = (void __user *)arg;
1884         int r;
1885         struct kvm_fpu *fpu = NULL;
1886         struct kvm_sregs *kvm_sregs = NULL;
1887
1888         if (vcpu->kvm->mm != current->mm)
1889                 return -EIO;
1890
1891 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1892         /*
1893          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1894          * so vcpu_load() would break it.
1895          */
1896         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1897                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1898 #endif
1899
1900
1901         r = vcpu_load(vcpu);
1902         if (r)
1903                 return r;
1904         switch (ioctl) {
1905         case KVM_RUN:
1906                 r = -EINVAL;
1907                 if (arg)
1908                         goto out;
1909                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1910                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1911                 break;
1912         case KVM_GET_REGS: {
1913                 struct kvm_regs *kvm_regs;
1914
1915                 r = -ENOMEM;
1916                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1917                 if (!kvm_regs)
1918                         goto out;
1919                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1920                 if (r)
1921                         goto out_free1;
1922                 r = -EFAULT;
1923                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1924                         goto out_free1;
1925                 r = 0;
1926 out_free1:
1927                 kfree(kvm_regs);
1928                 break;
1929         }
1930         case KVM_SET_REGS: {
1931                 struct kvm_regs *kvm_regs;
1932
1933                 r = -ENOMEM;
1934                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1935                 if (IS_ERR(kvm_regs)) {
1936                         r = PTR_ERR(kvm_regs);
1937                         goto out;
1938                 }
1939                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1940                 kfree(kvm_regs);
1941                 break;
1942         }
1943         case KVM_GET_SREGS: {
1944                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1945                 r = -ENOMEM;
1946                 if (!kvm_sregs)
1947                         goto out;
1948                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1949                 if (r)
1950                         goto out;
1951                 r = -EFAULT;
1952                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1953                         goto out;
1954                 r = 0;
1955                 break;
1956         }
1957         case KVM_SET_SREGS: {
1958                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1959                 if (IS_ERR(kvm_sregs)) {
1960                         r = PTR_ERR(kvm_sregs);
1961                         kvm_sregs = NULL;
1962                         goto out;
1963                 }
1964                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1965                 break;
1966         }
1967         case KVM_GET_MP_STATE: {
1968                 struct kvm_mp_state mp_state;
1969
1970                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1971                 if (r)
1972                         goto out;
1973                 r = -EFAULT;
1974                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1975                         goto out;
1976                 r = 0;
1977                 break;
1978         }
1979         case KVM_SET_MP_STATE: {
1980                 struct kvm_mp_state mp_state;
1981
1982                 r = -EFAULT;
1983                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1984                         goto out;
1985                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1986                 break;
1987         }
1988         case KVM_TRANSLATE: {
1989                 struct kvm_translation tr;
1990
1991                 r = -EFAULT;
1992                 if (copy_from_user(&tr, argp, sizeof tr))
1993                         goto out;
1994                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1995                 if (r)
1996                         goto out;
1997                 r = -EFAULT;
1998                 if (copy_to_user(argp, &tr, sizeof tr))
1999                         goto out;
2000                 r = 0;
2001                 break;
2002         }
2003         case KVM_SET_GUEST_DEBUG: {
2004                 struct kvm_guest_debug dbg;
2005
2006                 r = -EFAULT;
2007                 if (copy_from_user(&dbg, argp, sizeof dbg))
2008                         goto out;
2009                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2010                 break;
2011         }
2012         case KVM_SET_SIGNAL_MASK: {
2013                 struct kvm_signal_mask __user *sigmask_arg = argp;
2014                 struct kvm_signal_mask kvm_sigmask;
2015                 sigset_t sigset, *p;
2016
2017                 p = NULL;
2018                 if (argp) {
2019                         r = -EFAULT;
2020                         if (copy_from_user(&kvm_sigmask, argp,
2021                                            sizeof kvm_sigmask))
2022                                 goto out;
2023                         r = -EINVAL;
2024                         if (kvm_sigmask.len != sizeof sigset)
2025                                 goto out;
2026                         r = -EFAULT;
2027                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2028                                            sizeof sigset))
2029                                 goto out;
2030                         p = &sigset;
2031                 }
2032                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2033                 break;
2034         }
2035         case KVM_GET_FPU: {
2036                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2037                 r = -ENOMEM;
2038                 if (!fpu)
2039                         goto out;
2040                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2041                 if (r)
2042                         goto out;
2043                 r = -EFAULT;
2044                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2045                         goto out;
2046                 r = 0;
2047                 break;
2048         }
2049         case KVM_SET_FPU: {
2050                 fpu = memdup_user(argp, sizeof(*fpu));
2051                 if (IS_ERR(fpu)) {
2052                         r = PTR_ERR(fpu);
2053                         fpu = NULL;
2054                         goto out;
2055                 }
2056                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2057                 break;
2058         }
2059         default:
2060                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2061         }
2062 out:
2063         vcpu_put(vcpu);
2064         kfree(fpu);
2065         kfree(kvm_sregs);
2066         return r;
2067 }
2068
2069 #ifdef CONFIG_COMPAT
2070 static long kvm_vcpu_compat_ioctl(struct file *filp,
2071                                   unsigned int ioctl, unsigned long arg)
2072 {
2073         struct kvm_vcpu *vcpu = filp->private_data;
2074         void __user *argp = compat_ptr(arg);
2075         int r;
2076
2077         if (vcpu->kvm->mm != current->mm)
2078                 return -EIO;
2079
2080         switch (ioctl) {
2081         case KVM_SET_SIGNAL_MASK: {
2082                 struct kvm_signal_mask __user *sigmask_arg = argp;
2083                 struct kvm_signal_mask kvm_sigmask;
2084                 compat_sigset_t csigset;
2085                 sigset_t sigset;
2086
2087                 if (argp) {
2088                         r = -EFAULT;
2089                         if (copy_from_user(&kvm_sigmask, argp,
2090                                            sizeof kvm_sigmask))
2091                                 goto out;
2092                         r = -EINVAL;
2093                         if (kvm_sigmask.len != sizeof csigset)
2094                                 goto out;
2095                         r = -EFAULT;
2096                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2097                                            sizeof csigset))
2098                                 goto out;
2099                         sigset_from_compat(&sigset, &csigset);
2100                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2101                 } else
2102                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2103                 break;
2104         }
2105         default:
2106                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2107         }
2108
2109 out:
2110         return r;
2111 }
2112 #endif
2113
2114 static long kvm_vm_ioctl(struct file *filp,
2115                            unsigned int ioctl, unsigned long arg)
2116 {
2117         struct kvm *kvm = filp->private_data;
2118         void __user *argp = (void __user *)arg;
2119         int r;
2120
2121         if (kvm->mm != current->mm)
2122                 return -EIO;
2123         switch (ioctl) {
2124         case KVM_CREATE_VCPU:
2125                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2126                 break;
2127         case KVM_SET_USER_MEMORY_REGION: {
2128                 struct kvm_userspace_memory_region kvm_userspace_mem;
2129
2130                 r = -EFAULT;
2131                 if (copy_from_user(&kvm_userspace_mem, argp,
2132                                                 sizeof kvm_userspace_mem))
2133                         goto out;
2134
2135                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2136                 break;
2137         }
2138         case KVM_GET_DIRTY_LOG: {
2139                 struct kvm_dirty_log log;
2140
2141                 r = -EFAULT;
2142                 if (copy_from_user(&log, argp, sizeof log))
2143                         goto out;
2144                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2145                 break;
2146         }
2147 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2148         case KVM_REGISTER_COALESCED_MMIO: {
2149                 struct kvm_coalesced_mmio_zone zone;
2150                 r = -EFAULT;
2151                 if (copy_from_user(&zone, argp, sizeof zone))
2152                         goto out;
2153                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2154                 break;
2155         }
2156         case KVM_UNREGISTER_COALESCED_MMIO: {
2157                 struct kvm_coalesced_mmio_zone zone;
2158                 r = -EFAULT;
2159                 if (copy_from_user(&zone, argp, sizeof zone))
2160                         goto out;
2161                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2162                 break;
2163         }
2164 #endif
2165         case KVM_IRQFD: {
2166                 struct kvm_irqfd data;
2167
2168                 r = -EFAULT;
2169                 if (copy_from_user(&data, argp, sizeof data))
2170                         goto out;
2171                 r = kvm_irqfd(kvm, &data);
2172                 break;
2173         }
2174         case KVM_IOEVENTFD: {
2175                 struct kvm_ioeventfd data;
2176
2177                 r = -EFAULT;
2178                 if (copy_from_user(&data, argp, sizeof data))
2179                         goto out;
2180                 r = kvm_ioeventfd(kvm, &data);
2181                 break;
2182         }
2183 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2184         case KVM_SET_BOOT_CPU_ID:
2185                 r = 0;
2186                 mutex_lock(&kvm->lock);
2187                 if (atomic_read(&kvm->online_vcpus) != 0)
2188                         r = -EBUSY;
2189                 else
2190                         kvm->bsp_vcpu_id = arg;
2191                 mutex_unlock(&kvm->lock);
2192                 break;
2193 #endif
2194 #ifdef CONFIG_HAVE_KVM_MSI
2195         case KVM_SIGNAL_MSI: {
2196                 struct kvm_msi msi;
2197
2198                 r = -EFAULT;
2199                 if (copy_from_user(&msi, argp, sizeof msi))
2200                         goto out;
2201                 r = kvm_send_userspace_msi(kvm, &msi);
2202                 break;
2203         }
2204 #endif
2205 #ifdef __KVM_HAVE_IRQ_LINE
2206         case KVM_IRQ_LINE_STATUS:
2207         case KVM_IRQ_LINE: {
2208                 struct kvm_irq_level irq_event;
2209
2210                 r = -EFAULT;
2211                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2212                         goto out;
2213
2214                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2215                 if (r)
2216                         goto out;
2217
2218                 r = -EFAULT;
2219                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2220                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2221                                 goto out;
2222                 }
2223
2224                 r = 0;
2225                 break;
2226         }
2227 #endif
2228         default:
2229                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2230                 if (r == -ENOTTY)
2231                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2232         }
2233 out:
2234         return r;
2235 }
2236
2237 #ifdef CONFIG_COMPAT
2238 struct compat_kvm_dirty_log {
2239         __u32 slot;
2240         __u32 padding1;
2241         union {
2242                 compat_uptr_t dirty_bitmap; /* one bit per page */
2243                 __u64 padding2;
2244         };
2245 };
2246
2247 static long kvm_vm_compat_ioctl(struct file *filp,
2248                            unsigned int ioctl, unsigned long arg)
2249 {
2250         struct kvm *kvm = filp->private_data;
2251         int r;
2252
2253         if (kvm->mm != current->mm)
2254                 return -EIO;
2255         switch (ioctl) {
2256         case KVM_GET_DIRTY_LOG: {
2257                 struct compat_kvm_dirty_log compat_log;
2258                 struct kvm_dirty_log log;
2259
2260                 r = -EFAULT;
2261                 if (copy_from_user(&compat_log, (void __user *)arg,
2262                                    sizeof(compat_log)))
2263                         goto out;
2264                 log.slot         = compat_log.slot;
2265                 log.padding1     = compat_log.padding1;
2266                 log.padding2     = compat_log.padding2;
2267                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2268
2269                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2270                 break;
2271         }
2272         default:
2273                 r = kvm_vm_ioctl(filp, ioctl, arg);
2274         }
2275
2276 out:
2277         return r;
2278 }
2279 #endif
2280
2281 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2282 {
2283         struct page *page[1];
2284         unsigned long addr;
2285         int npages;
2286         gfn_t gfn = vmf->pgoff;
2287         struct kvm *kvm = vma->vm_file->private_data;
2288
2289         addr = gfn_to_hva(kvm, gfn);
2290         if (kvm_is_error_hva(addr))
2291                 return VM_FAULT_SIGBUS;
2292
2293         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2294                                 NULL);
2295         if (unlikely(npages != 1))
2296                 return VM_FAULT_SIGBUS;
2297
2298         vmf->page = page[0];
2299         return 0;
2300 }
2301
2302 static const struct vm_operations_struct kvm_vm_vm_ops = {
2303         .fault = kvm_vm_fault,
2304 };
2305
2306 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2307 {
2308         vma->vm_ops = &kvm_vm_vm_ops;
2309         return 0;
2310 }
2311
2312 static struct file_operations kvm_vm_fops = {
2313         .release        = kvm_vm_release,
2314         .unlocked_ioctl = kvm_vm_ioctl,
2315 #ifdef CONFIG_COMPAT
2316         .compat_ioctl   = kvm_vm_compat_ioctl,
2317 #endif
2318         .mmap           = kvm_vm_mmap,
2319         .llseek         = noop_llseek,
2320 };
2321
2322 static int kvm_dev_ioctl_create_vm(unsigned long type)
2323 {
2324         int r;
2325         struct kvm *kvm;
2326
2327         kvm = kvm_create_vm(type);
2328         if (IS_ERR(kvm))
2329                 return PTR_ERR(kvm);
2330 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2331         r = kvm_coalesced_mmio_init(kvm);
2332         if (r < 0) {
2333                 kvm_put_kvm(kvm);
2334                 return r;
2335         }
2336 #endif
2337         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2338         if (r < 0)
2339                 kvm_put_kvm(kvm);
2340
2341         return r;
2342 }
2343
2344 static long kvm_dev_ioctl_check_extension_generic(long arg)
2345 {
2346         switch (arg) {
2347         case KVM_CAP_USER_MEMORY:
2348         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2349         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2350 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2351         case KVM_CAP_SET_BOOT_CPU_ID:
2352 #endif
2353         case KVM_CAP_INTERNAL_ERROR_DATA:
2354 #ifdef CONFIG_HAVE_KVM_MSI
2355         case KVM_CAP_SIGNAL_MSI:
2356 #endif
2357                 return 1;
2358 #ifdef KVM_CAP_IRQ_ROUTING
2359         case KVM_CAP_IRQ_ROUTING:
2360                 return KVM_MAX_IRQ_ROUTES;
2361 #endif
2362         default:
2363                 break;
2364         }
2365         return kvm_dev_ioctl_check_extension(arg);
2366 }
2367
2368 static long kvm_dev_ioctl(struct file *filp,
2369                           unsigned int ioctl, unsigned long arg)
2370 {
2371         long r = -EINVAL;
2372
2373         switch (ioctl) {
2374         case KVM_GET_API_VERSION:
2375                 r = -EINVAL;
2376                 if (arg)
2377                         goto out;
2378                 r = KVM_API_VERSION;
2379                 break;
2380         case KVM_CREATE_VM:
2381                 r = kvm_dev_ioctl_create_vm(arg);
2382                 break;
2383         case KVM_CHECK_EXTENSION:
2384                 r = kvm_dev_ioctl_check_extension_generic(arg);
2385                 break;
2386         case KVM_GET_VCPU_MMAP_SIZE:
2387                 r = -EINVAL;
2388                 if (arg)
2389                         goto out;
2390                 r = PAGE_SIZE;     /* struct kvm_run */
2391 #ifdef CONFIG_X86
2392                 r += PAGE_SIZE;    /* pio data page */
2393 #endif
2394 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2395                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2396 #endif
2397                 break;
2398         case KVM_TRACE_ENABLE:
2399         case KVM_TRACE_PAUSE:
2400         case KVM_TRACE_DISABLE:
2401                 r = -EOPNOTSUPP;
2402                 break;
2403         default:
2404                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2405         }
2406 out:
2407         return r;
2408 }
2409
2410 static struct file_operations kvm_chardev_ops = {
2411         .unlocked_ioctl = kvm_dev_ioctl,
2412         .compat_ioctl   = kvm_dev_ioctl,
2413         .llseek         = noop_llseek,
2414 };
2415
2416 static struct miscdevice kvm_dev = {
2417         KVM_MINOR,
2418         "kvm",
2419         &kvm_chardev_ops,
2420 };
2421
2422 static void hardware_enable_nolock(void *junk)
2423 {
2424         int cpu = raw_smp_processor_id();
2425         int r;
2426
2427         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2428                 return;
2429
2430         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2431
2432         r = kvm_arch_hardware_enable(NULL);
2433
2434         if (r) {
2435                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2436                 atomic_inc(&hardware_enable_failed);
2437                 printk(KERN_INFO "kvm: enabling virtualization on "
2438                                  "CPU%d failed\n", cpu);
2439         }
2440 }
2441
2442 static void hardware_enable(void *junk)
2443 {
2444         raw_spin_lock(&kvm_lock);
2445         hardware_enable_nolock(junk);
2446         raw_spin_unlock(&kvm_lock);
2447 }
2448
2449 static void hardware_disable_nolock(void *junk)
2450 {
2451         int cpu = raw_smp_processor_id();
2452
2453         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2454                 return;
2455         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2456         kvm_arch_hardware_disable(NULL);
2457 }
2458
2459 static void hardware_disable(void *junk)
2460 {
2461         raw_spin_lock(&kvm_lock);
2462         hardware_disable_nolock(junk);
2463         raw_spin_unlock(&kvm_lock);
2464 }
2465
2466 static void hardware_disable_all_nolock(void)
2467 {
2468         BUG_ON(!kvm_usage_count);
2469
2470         kvm_usage_count--;
2471         if (!kvm_usage_count)
2472                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2473 }
2474
2475 static void hardware_disable_all(void)
2476 {
2477         raw_spin_lock(&kvm_lock);
2478         hardware_disable_all_nolock();
2479         raw_spin_unlock(&kvm_lock);
2480 }
2481
2482 static int hardware_enable_all(void)
2483 {
2484         int r = 0;
2485
2486         raw_spin_lock(&kvm_lock);
2487
2488         kvm_usage_count++;
2489         if (kvm_usage_count == 1) {
2490                 atomic_set(&hardware_enable_failed, 0);
2491                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2492
2493                 if (atomic_read(&hardware_enable_failed)) {
2494                         hardware_disable_all_nolock();
2495                         r = -EBUSY;
2496                 }
2497         }
2498
2499         raw_spin_unlock(&kvm_lock);
2500
2501         return r;
2502 }
2503
2504 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2505                            void *v)
2506 {
2507         int cpu = (long)v;
2508
2509         if (!kvm_usage_count)
2510                 return NOTIFY_OK;
2511
2512         val &= ~CPU_TASKS_FROZEN;
2513         switch (val) {
2514         case CPU_DYING:
2515                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2516                        cpu);
2517                 hardware_disable(NULL);
2518                 break;
2519         case CPU_STARTING:
2520                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2521                        cpu);
2522                 hardware_enable(NULL);
2523                 break;
2524         }
2525         return NOTIFY_OK;
2526 }
2527
2528
2529 asmlinkage void kvm_spurious_fault(void)
2530 {
2531         /* Fault while not rebooting.  We want the trace. */
2532         BUG();
2533 }
2534 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2535
2536 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2537                       void *v)
2538 {
2539         /*
2540          * Some (well, at least mine) BIOSes hang on reboot if
2541          * in vmx root mode.
2542          *
2543          * And Intel TXT required VMX off for all cpu when system shutdown.
2544          */
2545         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2546         kvm_rebooting = true;
2547         on_each_cpu(hardware_disable_nolock, NULL, 1);
2548         return NOTIFY_OK;
2549 }
2550
2551 static struct notifier_block kvm_reboot_notifier = {
2552         .notifier_call = kvm_reboot,
2553         .priority = 0,
2554 };
2555
2556 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2557 {
2558         int i;
2559
2560         for (i = 0; i < bus->dev_count; i++) {
2561                 struct kvm_io_device *pos = bus->range[i].dev;
2562
2563                 kvm_iodevice_destructor(pos);
2564         }
2565         kfree(bus);
2566 }
2567
2568 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2569 {
2570         const struct kvm_io_range *r1 = p1;
2571         const struct kvm_io_range *r2 = p2;
2572
2573         if (r1->addr < r2->addr)
2574                 return -1;
2575         if (r1->addr + r1->len > r2->addr + r2->len)
2576                 return 1;
2577         return 0;
2578 }
2579
2580 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2581                           gpa_t addr, int len)
2582 {
2583         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2584                 .addr = addr,
2585                 .len = len,
2586                 .dev = dev,
2587         };
2588
2589         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2590                 kvm_io_bus_sort_cmp, NULL);
2591
2592         return 0;
2593 }
2594
2595 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2596                              gpa_t addr, int len)
2597 {
2598         struct kvm_io_range *range, key;
2599         int off;
2600
2601         key = (struct kvm_io_range) {
2602                 .addr = addr,
2603                 .len = len,
2604         };
2605
2606         range = bsearch(&key, bus->range, bus->dev_count,
2607                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2608         if (range == NULL)
2609                 return -ENOENT;
2610
2611         off = range - bus->range;
2612
2613         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2614                 off--;
2615
2616         return off;
2617 }
2618
2619 /* kvm_io_bus_write - called under kvm->slots_lock */
2620 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2621                      int len, const void *val)
2622 {
2623         int idx;
2624         struct kvm_io_bus *bus;
2625         struct kvm_io_range range;
2626
2627         range = (struct kvm_io_range) {
2628                 .addr = addr,
2629                 .len = len,
2630         };
2631
2632         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2633         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2634         if (idx < 0)
2635                 return -EOPNOTSUPP;
2636
2637         while (idx < bus->dev_count &&
2638                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2639                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2640                         return 0;
2641                 idx++;
2642         }
2643
2644         return -EOPNOTSUPP;
2645 }
2646
2647 /* kvm_io_bus_read - called under kvm->slots_lock */
2648 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2649                     int len, void *val)
2650 {
2651         int idx;
2652         struct kvm_io_bus *bus;
2653         struct kvm_io_range range;
2654
2655         range = (struct kvm_io_range) {
2656                 .addr = addr,
2657                 .len = len,
2658         };
2659
2660         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2661         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2662         if (idx < 0)
2663                 return -EOPNOTSUPP;
2664
2665         while (idx < bus->dev_count &&
2666                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2667                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2668                         return 0;
2669                 idx++;
2670         }
2671
2672         return -EOPNOTSUPP;
2673 }
2674
2675 /* Caller must hold slots_lock. */
2676 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2677                             int len, struct kvm_io_device *dev)
2678 {
2679         struct kvm_io_bus *new_bus, *bus;
2680
2681         bus = kvm->buses[bus_idx];
2682         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2683                 return -ENOSPC;
2684
2685         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2686                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2687         if (!new_bus)
2688                 return -ENOMEM;
2689         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2690                sizeof(struct kvm_io_range)));
2691         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2692         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2693         synchronize_srcu_expedited(&kvm->srcu);
2694         kfree(bus);
2695
2696         return 0;
2697 }
2698
2699 /* Caller must hold slots_lock. */
2700 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2701                               struct kvm_io_device *dev)
2702 {
2703         int i, r;
2704         struct kvm_io_bus *new_bus, *bus;
2705
2706         bus = kvm->buses[bus_idx];
2707         r = -ENOENT;
2708         for (i = 0; i < bus->dev_count; i++)
2709                 if (bus->range[i].dev == dev) {
2710                         r = 0;
2711                         break;
2712                 }
2713
2714         if (r)
2715                 return r;
2716
2717         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2718                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2719         if (!new_bus)
2720                 return -ENOMEM;
2721
2722         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2723         new_bus->dev_count--;
2724         memcpy(new_bus->range + i, bus->range + i + 1,
2725                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2726
2727         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2728         synchronize_srcu_expedited(&kvm->srcu);
2729         kfree(bus);
2730         return r;
2731 }
2732
2733 static struct notifier_block kvm_cpu_notifier = {
2734         .notifier_call = kvm_cpu_hotplug,
2735 };
2736
2737 static int vm_stat_get(void *_offset, u64 *val)
2738 {
2739         unsigned offset = (long)_offset;
2740         struct kvm *kvm;
2741
2742         *val = 0;
2743         raw_spin_lock(&kvm_lock);
2744         list_for_each_entry(kvm, &vm_list, vm_list)
2745                 *val += *(u32 *)((void *)kvm + offset);
2746         raw_spin_unlock(&kvm_lock);
2747         return 0;
2748 }
2749
2750 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2751
2752 static int vcpu_stat_get(void *_offset, u64 *val)
2753 {
2754         unsigned offset = (long)_offset;
2755         struct kvm *kvm;
2756         struct kvm_vcpu *vcpu;
2757         int i;
2758
2759         *val = 0;
2760         raw_spin_lock(&kvm_lock);
2761         list_for_each_entry(kvm, &vm_list, vm_list)
2762                 kvm_for_each_vcpu(i, vcpu, kvm)
2763                         *val += *(u32 *)((void *)vcpu + offset);
2764
2765         raw_spin_unlock(&kvm_lock);
2766         return 0;
2767 }
2768
2769 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2770
2771 static const struct file_operations *stat_fops[] = {
2772         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2773         [KVM_STAT_VM]   = &vm_stat_fops,
2774 };
2775
2776 static int kvm_init_debug(void)
2777 {
2778         int r = -EFAULT;
2779         struct kvm_stats_debugfs_item *p;
2780
2781         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2782         if (kvm_debugfs_dir == NULL)
2783                 goto out;
2784
2785         for (p = debugfs_entries; p->name; ++p) {
2786                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2787                                                 (void *)(long)p->offset,
2788                                                 stat_fops[p->kind]);
2789                 if (p->dentry == NULL)
2790                         goto out_dir;
2791         }
2792
2793         return 0;
2794
2795 out_dir:
2796         debugfs_remove_recursive(kvm_debugfs_dir);
2797 out:
2798         return r;
2799 }
2800
2801 static void kvm_exit_debug(void)
2802 {
2803         struct kvm_stats_debugfs_item *p;
2804
2805         for (p = debugfs_entries; p->name; ++p)
2806                 debugfs_remove(p->dentry);
2807         debugfs_remove(kvm_debugfs_dir);
2808 }
2809
2810 static int kvm_suspend(void)
2811 {
2812         if (kvm_usage_count)
2813                 hardware_disable_nolock(NULL);
2814         return 0;
2815 }
2816
2817 static void kvm_resume(void)
2818 {
2819         if (kvm_usage_count) {
2820                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2821                 hardware_enable_nolock(NULL);
2822         }
2823 }
2824
2825 static struct syscore_ops kvm_syscore_ops = {
2826         .suspend = kvm_suspend,
2827         .resume = kvm_resume,
2828 };
2829
2830 static inline
2831 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2832 {
2833         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2834 }
2835
2836 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2837 {
2838         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2839
2840         kvm_arch_vcpu_load(vcpu, cpu);
2841 }
2842
2843 static void kvm_sched_out(struct preempt_notifier *pn,
2844                           struct task_struct *next)
2845 {
2846         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2847
2848         kvm_arch_vcpu_put(vcpu);
2849 }
2850
2851 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2852                   struct module *module)
2853 {
2854         int r;
2855         int cpu;
2856
2857         r = kvm_arch_init(opaque);
2858         if (r)
2859                 goto out_fail;
2860
2861         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2862                 r = -ENOMEM;
2863                 goto out_free_0;
2864         }
2865
2866         r = kvm_arch_hardware_setup();
2867         if (r < 0)
2868                 goto out_free_0a;
2869
2870         for_each_online_cpu(cpu) {
2871                 smp_call_function_single(cpu,
2872                                 kvm_arch_check_processor_compat,
2873                                 &r, 1);
2874                 if (r < 0)
2875                         goto out_free_1;
2876         }
2877
2878         r = register_cpu_notifier(&kvm_cpu_notifier);
2879         if (r)
2880                 goto out_free_2;
2881         register_reboot_notifier(&kvm_reboot_notifier);
2882
2883         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2884         if (!vcpu_align)
2885                 vcpu_align = __alignof__(struct kvm_vcpu);
2886         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2887                                            0, NULL);
2888         if (!kvm_vcpu_cache) {
2889                 r = -ENOMEM;
2890                 goto out_free_3;
2891         }
2892
2893         r = kvm_async_pf_init();
2894         if (r)
2895                 goto out_free;
2896
2897         kvm_chardev_ops.owner = module;
2898         kvm_vm_fops.owner = module;
2899         kvm_vcpu_fops.owner = module;
2900
2901         r = misc_register(&kvm_dev);
2902         if (r) {
2903                 printk(KERN_ERR "kvm: misc device register failed\n");
2904                 goto out_unreg;
2905         }
2906
2907         register_syscore_ops(&kvm_syscore_ops);
2908
2909         kvm_preempt_ops.sched_in = kvm_sched_in;
2910         kvm_preempt_ops.sched_out = kvm_sched_out;
2911
2912         r = kvm_init_debug();
2913         if (r) {
2914                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2915                 goto out_undebugfs;
2916         }
2917
2918         return 0;
2919
2920 out_undebugfs:
2921         unregister_syscore_ops(&kvm_syscore_ops);
2922 out_unreg:
2923         kvm_async_pf_deinit();
2924 out_free:
2925         kmem_cache_destroy(kvm_vcpu_cache);
2926 out_free_3:
2927         unregister_reboot_notifier(&kvm_reboot_notifier);
2928         unregister_cpu_notifier(&kvm_cpu_notifier);
2929 out_free_2:
2930 out_free_1:
2931         kvm_arch_hardware_unsetup();
2932 out_free_0a:
2933         free_cpumask_var(cpus_hardware_enabled);
2934 out_free_0:
2935         kvm_arch_exit();
2936 out_fail:
2937         return r;
2938 }
2939 EXPORT_SYMBOL_GPL(kvm_init);
2940
2941 void kvm_exit(void)
2942 {
2943         kvm_exit_debug();
2944         misc_deregister(&kvm_dev);
2945         kmem_cache_destroy(kvm_vcpu_cache);
2946         kvm_async_pf_deinit();
2947         unregister_syscore_ops(&kvm_syscore_ops);
2948         unregister_reboot_notifier(&kvm_reboot_notifier);
2949         unregister_cpu_notifier(&kvm_cpu_notifier);
2950         on_each_cpu(hardware_disable_nolock, NULL, 1);
2951         kvm_arch_hardware_unsetup();
2952         kvm_arch_exit();
2953         free_cpumask_var(cpus_hardware_enabled);
2954 }
2955 EXPORT_SYMBOL_GPL(kvm_exit);