2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
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>
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>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_RAW_SPINLOCK(kvm_lock);
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
85 struct dentry *kvm_debugfs_dir;
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
101 static bool largepages_enabled = true;
103 static struct page *hwpoison_page;
104 static pfn_t hwpoison_pfn;
106 struct page *fault_page;
109 inline int kvm_is_mmio_pfn(pfn_t pfn)
111 if (pfn_valid(pfn)) {
113 struct page *tail = pfn_to_page(pfn);
114 struct page *head = compound_trans_head(tail);
115 reserved = PageReserved(head);
118 * "head" is not a dangling pointer
119 * (compound_trans_head takes care of that)
120 * but the hugepage may have been splitted
121 * from under us (and we may not hold a
122 * reference count on the head page so it can
123 * be reused before we run PageReferenced), so
124 * we've to check PageTail before returning
131 return PageReserved(tail);
138 * Switches to specified vcpu, until a matching vcpu_put()
140 void vcpu_load(struct kvm_vcpu *vcpu)
144 mutex_lock(&vcpu->mutex);
145 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
146 /* The thread running this VCPU changed. */
147 struct pid *oldpid = vcpu->pid;
148 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
149 rcu_assign_pointer(vcpu->pid, newpid);
154 preempt_notifier_register(&vcpu->preempt_notifier);
155 kvm_arch_vcpu_load(vcpu, cpu);
159 void vcpu_put(struct kvm_vcpu *vcpu)
162 kvm_arch_vcpu_put(vcpu);
163 preempt_notifier_unregister(&vcpu->preempt_notifier);
165 mutex_unlock(&vcpu->mutex);
168 static void ack_flush(void *_completed)
172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
177 struct kvm_vcpu *vcpu;
179 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
182 kvm_for_each_vcpu(i, vcpu, kvm) {
183 kvm_make_request(req, vcpu);
186 /* Set ->requests bit before we read ->mode */
189 if (cpus != NULL && cpu != -1 && cpu != me &&
190 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191 cpumask_set_cpu(cpu, cpus);
193 if (unlikely(cpus == NULL))
194 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195 else if (!cpumask_empty(cpus))
196 smp_call_function_many(cpus, ack_flush, NULL, 1);
200 free_cpumask_var(cpus);
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
206 long dirty_count = kvm->tlbs_dirty;
209 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
210 ++kvm->stat.remote_tlb_flush;
211 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
214 void kvm_reload_remote_mmus(struct kvm *kvm)
216 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
224 mutex_init(&vcpu->mutex);
229 init_waitqueue_head(&vcpu->wq);
230 kvm_async_pf_vcpu_init(vcpu);
232 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
237 vcpu->run = page_address(page);
239 r = kvm_arch_vcpu_init(vcpu);
245 free_page((unsigned long)vcpu->run);
249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
254 kvm_arch_vcpu_uninit(vcpu);
255 free_page((unsigned long)vcpu->run);
257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
262 return container_of(mn, struct kvm, mmu_notifier);
265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
266 struct mm_struct *mm,
267 unsigned long address)
269 struct kvm *kvm = mmu_notifier_to_kvm(mn);
270 int need_tlb_flush, idx;
273 * When ->invalidate_page runs, the linux pte has been zapped
274 * already but the page is still allocated until
275 * ->invalidate_page returns. So if we increase the sequence
276 * here the kvm page fault will notice if the spte can't be
277 * established because the page is going to be freed. If
278 * instead the kvm page fault establishes the spte before
279 * ->invalidate_page runs, kvm_unmap_hva will release it
282 * The sequence increase only need to be seen at spin_unlock
283 * time, and not at spin_lock time.
285 * Increasing the sequence after the spin_unlock would be
286 * unsafe because the kvm page fault could then establish the
287 * pte after kvm_unmap_hva returned, without noticing the page
288 * is going to be freed.
290 idx = srcu_read_lock(&kvm->srcu);
291 spin_lock(&kvm->mmu_lock);
293 kvm->mmu_notifier_seq++;
294 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
295 /* we've to flush the tlb before the pages can be freed */
297 kvm_flush_remote_tlbs(kvm);
299 spin_unlock(&kvm->mmu_lock);
300 srcu_read_unlock(&kvm->srcu, idx);
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304 struct mm_struct *mm,
305 unsigned long address,
308 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 idx = srcu_read_lock(&kvm->srcu);
312 spin_lock(&kvm->mmu_lock);
313 kvm->mmu_notifier_seq++;
314 kvm_set_spte_hva(kvm, address, pte);
315 spin_unlock(&kvm->mmu_lock);
316 srcu_read_unlock(&kvm->srcu, idx);
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320 struct mm_struct *mm,
324 struct kvm *kvm = mmu_notifier_to_kvm(mn);
325 int need_tlb_flush = 0, idx;
327 idx = srcu_read_lock(&kvm->srcu);
328 spin_lock(&kvm->mmu_lock);
330 * The count increase must become visible at unlock time as no
331 * spte can be established without taking the mmu_lock and
332 * count is also read inside the mmu_lock critical section.
334 kvm->mmu_notifier_count++;
335 for (; start < end; start += PAGE_SIZE)
336 need_tlb_flush |= kvm_unmap_hva(kvm, start);
337 need_tlb_flush |= kvm->tlbs_dirty;
338 /* we've to flush the tlb before the pages can be freed */
340 kvm_flush_remote_tlbs(kvm);
342 spin_unlock(&kvm->mmu_lock);
343 srcu_read_unlock(&kvm->srcu, idx);
346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
347 struct mm_struct *mm,
351 struct kvm *kvm = mmu_notifier_to_kvm(mn);
353 spin_lock(&kvm->mmu_lock);
355 * This sequence increase will notify the kvm page fault that
356 * the page that is going to be mapped in the spte could have
359 kvm->mmu_notifier_seq++;
362 * The above sequence increase must be visible before the
363 * below count decrease, which is ensured by the smp_wmb above
364 * in conjunction with the smp_rmb in mmu_notifier_retry().
366 kvm->mmu_notifier_count--;
367 spin_unlock(&kvm->mmu_lock);
369 BUG_ON(kvm->mmu_notifier_count < 0);
372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
373 struct mm_struct *mm,
374 unsigned long address)
376 struct kvm *kvm = mmu_notifier_to_kvm(mn);
379 idx = srcu_read_lock(&kvm->srcu);
380 spin_lock(&kvm->mmu_lock);
382 young = kvm_age_hva(kvm, address);
384 kvm_flush_remote_tlbs(kvm);
386 spin_unlock(&kvm->mmu_lock);
387 srcu_read_unlock(&kvm->srcu, idx);
392 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
393 struct mm_struct *mm,
394 unsigned long address)
396 struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 idx = srcu_read_lock(&kvm->srcu);
400 spin_lock(&kvm->mmu_lock);
401 young = kvm_test_age_hva(kvm, address);
402 spin_unlock(&kvm->mmu_lock);
403 srcu_read_unlock(&kvm->srcu, idx);
408 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
409 struct mm_struct *mm)
411 struct kvm *kvm = mmu_notifier_to_kvm(mn);
414 idx = srcu_read_lock(&kvm->srcu);
415 kvm_arch_flush_shadow(kvm);
416 srcu_read_unlock(&kvm->srcu, idx);
419 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
420 .invalidate_page = kvm_mmu_notifier_invalidate_page,
421 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
422 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
423 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
424 .test_young = kvm_mmu_notifier_test_young,
425 .change_pte = kvm_mmu_notifier_change_pte,
426 .release = kvm_mmu_notifier_release,
429 static int kvm_init_mmu_notifier(struct kvm *kvm)
431 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
432 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
435 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
437 static int kvm_init_mmu_notifier(struct kvm *kvm)
442 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
444 static void kvm_init_memslots_id(struct kvm *kvm)
447 struct kvm_memslots *slots = kvm->memslots;
449 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
450 slots->id_to_index[i] = slots->memslots[i].id = i;
453 static struct kvm *kvm_create_vm(unsigned long type)
456 struct kvm *kvm = kvm_arch_alloc_vm();
459 return ERR_PTR(-ENOMEM);
461 r = kvm_arch_init_vm(kvm, type);
463 goto out_err_nodisable;
465 r = hardware_enable_all();
467 goto out_err_nodisable;
469 #ifdef CONFIG_HAVE_KVM_IRQCHIP
470 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
471 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
475 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
478 kvm_init_memslots_id(kvm);
479 if (init_srcu_struct(&kvm->srcu))
481 for (i = 0; i < KVM_NR_BUSES; i++) {
482 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
488 spin_lock_init(&kvm->mmu_lock);
489 kvm->mm = current->mm;
490 atomic_inc(&kvm->mm->mm_count);
491 kvm_eventfd_init(kvm);
492 mutex_init(&kvm->lock);
493 mutex_init(&kvm->irq_lock);
494 mutex_init(&kvm->slots_lock);
495 atomic_set(&kvm->users_count, 1);
497 r = kvm_init_mmu_notifier(kvm);
501 raw_spin_lock(&kvm_lock);
502 list_add(&kvm->vm_list, &vm_list);
503 raw_spin_unlock(&kvm_lock);
508 cleanup_srcu_struct(&kvm->srcu);
510 hardware_disable_all();
512 for (i = 0; i < KVM_NR_BUSES; i++)
513 kfree(kvm->buses[i]);
514 kfree(kvm->memslots);
515 kvm_arch_free_vm(kvm);
519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
521 if (!memslot->dirty_bitmap)
524 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
525 vfree(memslot->dirty_bitmap_head);
527 kfree(memslot->dirty_bitmap_head);
529 memslot->dirty_bitmap = NULL;
530 memslot->dirty_bitmap_head = NULL;
534 * Free any memory in @free but not in @dont.
536 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
537 struct kvm_memory_slot *dont)
539 if (!dont || free->rmap != dont->rmap)
542 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
543 kvm_destroy_dirty_bitmap(free);
545 kvm_arch_free_memslot(free, dont);
551 void kvm_free_physmem(struct kvm *kvm)
553 struct kvm_memslots *slots = kvm->memslots;
554 struct kvm_memory_slot *memslot;
556 kvm_for_each_memslot(memslot, slots)
557 kvm_free_physmem_slot(memslot, NULL);
559 kfree(kvm->memslots);
562 static void kvm_destroy_vm(struct kvm *kvm)
565 struct mm_struct *mm = kvm->mm;
567 kvm_arch_sync_events(kvm);
568 raw_spin_lock(&kvm_lock);
569 list_del(&kvm->vm_list);
570 raw_spin_unlock(&kvm_lock);
571 kvm_free_irq_routing(kvm);
572 for (i = 0; i < KVM_NR_BUSES; i++)
573 kvm_io_bus_destroy(kvm->buses[i]);
574 kvm_coalesced_mmio_free(kvm);
575 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
576 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
578 kvm_arch_flush_shadow(kvm);
580 kvm_arch_destroy_vm(kvm);
581 kvm_free_physmem(kvm);
582 cleanup_srcu_struct(&kvm->srcu);
583 kvm_arch_free_vm(kvm);
584 hardware_disable_all();
588 void kvm_get_kvm(struct kvm *kvm)
590 atomic_inc(&kvm->users_count);
592 EXPORT_SYMBOL_GPL(kvm_get_kvm);
594 void kvm_put_kvm(struct kvm *kvm)
596 if (atomic_dec_and_test(&kvm->users_count))
599 EXPORT_SYMBOL_GPL(kvm_put_kvm);
602 static int kvm_vm_release(struct inode *inode, struct file *filp)
604 struct kvm *kvm = filp->private_data;
606 kvm_irqfd_release(kvm);
613 * Allocation size is twice as large as the actual dirty bitmap size.
614 * This makes it possible to do double buffering: see x86's
615 * kvm_vm_ioctl_get_dirty_log().
617 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
620 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
622 if (dirty_bytes > PAGE_SIZE)
623 memslot->dirty_bitmap = vzalloc(dirty_bytes);
625 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
627 if (!memslot->dirty_bitmap)
630 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
631 memslot->nr_dirty_pages = 0;
632 #endif /* !CONFIG_S390 */
636 static int cmp_memslot(const void *slot1, const void *slot2)
638 struct kvm_memory_slot *s1, *s2;
640 s1 = (struct kvm_memory_slot *)slot1;
641 s2 = (struct kvm_memory_slot *)slot2;
643 if (s1->npages < s2->npages)
645 if (s1->npages > s2->npages)
652 * Sort the memslots base on its size, so the larger slots
653 * will get better fit.
655 static void sort_memslots(struct kvm_memslots *slots)
659 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
660 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
662 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
663 slots->id_to_index[slots->memslots[i].id] = i;
666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
670 struct kvm_memory_slot *old = id_to_memslot(slots, id);
671 unsigned long npages = old->npages;
674 if (new->npages != npages)
675 sort_memslots(slots);
682 * Allocate some memory and give it an address in the guest physical address
685 * Discontiguous memory is allowed, mostly for framebuffers.
687 * Must be called holding mmap_sem for write.
689 int __kvm_set_memory_region(struct kvm *kvm,
690 struct kvm_userspace_memory_region *mem,
695 unsigned long npages;
697 struct kvm_memory_slot *memslot;
698 struct kvm_memory_slot old, new;
699 struct kvm_memslots *slots, *old_memslots;
702 /* General sanity checks */
703 if (mem->memory_size & (PAGE_SIZE - 1))
705 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
707 /* We can read the guest memory with __xxx_user() later on. */
709 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
710 !access_ok(VERIFY_WRITE,
711 (void __user *)(unsigned long)mem->userspace_addr,
714 if (mem->slot >= KVM_MEM_SLOTS_NUM)
716 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
719 memslot = id_to_memslot(kvm->memslots, mem->slot);
720 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
721 npages = mem->memory_size >> PAGE_SHIFT;
724 if (npages > KVM_MEM_MAX_NR_PAGES)
728 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
730 new = old = *memslot;
733 new.base_gfn = base_gfn;
735 new.flags = mem->flags;
737 /* Disallow changing a memory slot's size. */
739 if (npages && old.npages && npages != old.npages)
742 /* Check for overlaps */
744 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
745 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
747 if (s == memslot || !s->npages)
749 if (!((base_gfn + npages <= s->base_gfn) ||
750 (base_gfn >= s->base_gfn + s->npages)))
754 /* Free page dirty bitmap if unneeded */
755 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
756 new.dirty_bitmap = NULL;
760 /* Allocate if a slot is being created */
761 if (npages && !old.npages) {
762 new.user_alloc = user_alloc;
763 new.userspace_addr = mem->userspace_addr;
765 new.rmap = vzalloc(npages * sizeof(*new.rmap));
768 #endif /* not defined CONFIG_S390 */
769 if (kvm_arch_create_memslot(&new, npages))
773 /* Allocate page dirty bitmap if needed */
774 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
775 if (kvm_create_dirty_bitmap(&new) < 0)
777 /* destroy any largepage mappings for dirty tracking */
781 struct kvm_memory_slot *slot;
784 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
788 slot = id_to_memslot(slots, mem->slot);
789 slot->flags |= KVM_MEMSLOT_INVALID;
791 update_memslots(slots, NULL);
793 old_memslots = kvm->memslots;
794 rcu_assign_pointer(kvm->memslots, slots);
795 synchronize_srcu_expedited(&kvm->srcu);
796 /* From this point no new shadow pages pointing to a deleted
797 * memslot will be created.
799 * validation of sp->gfn happens in:
800 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
801 * - kvm_is_visible_gfn (mmu_check_roots)
803 kvm_arch_flush_shadow(kvm);
807 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
811 /* map the pages in iommu page table */
813 r = kvm_iommu_map_pages(kvm, &new);
819 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
824 /* actual memory is freed via old in kvm_free_physmem_slot below */
827 new.dirty_bitmap = NULL;
828 memset(&new.arch, 0, sizeof(new.arch));
831 update_memslots(slots, &new);
832 old_memslots = kvm->memslots;
833 rcu_assign_pointer(kvm->memslots, slots);
834 synchronize_srcu_expedited(&kvm->srcu);
836 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
839 * If the new memory slot is created, we need to clear all
842 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
843 kvm_arch_flush_shadow(kvm);
845 kvm_free_physmem_slot(&old, &new);
851 kvm_free_physmem_slot(&new, &old);
856 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
858 int kvm_set_memory_region(struct kvm *kvm,
859 struct kvm_userspace_memory_region *mem,
864 mutex_lock(&kvm->slots_lock);
865 r = __kvm_set_memory_region(kvm, mem, user_alloc);
866 mutex_unlock(&kvm->slots_lock);
869 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
871 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
873 kvm_userspace_memory_region *mem,
876 if (mem->slot >= KVM_MEMORY_SLOTS)
878 return kvm_set_memory_region(kvm, mem, user_alloc);
881 int kvm_get_dirty_log(struct kvm *kvm,
882 struct kvm_dirty_log *log, int *is_dirty)
884 struct kvm_memory_slot *memslot;
887 unsigned long any = 0;
890 if (log->slot >= KVM_MEMORY_SLOTS)
893 memslot = id_to_memslot(kvm->memslots, log->slot);
895 if (!memslot->dirty_bitmap)
898 n = kvm_dirty_bitmap_bytes(memslot);
900 for (i = 0; !any && i < n/sizeof(long); ++i)
901 any = memslot->dirty_bitmap[i];
904 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
915 bool kvm_largepages_enabled(void)
917 return largepages_enabled;
920 void kvm_disable_largepages(void)
922 largepages_enabled = false;
924 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
926 int is_error_page(struct page *page)
928 return page == bad_page || page == hwpoison_page || page == fault_page;
930 EXPORT_SYMBOL_GPL(is_error_page);
932 int is_error_pfn(pfn_t pfn)
934 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
936 EXPORT_SYMBOL_GPL(is_error_pfn);
938 int is_hwpoison_pfn(pfn_t pfn)
940 return pfn == hwpoison_pfn;
942 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
944 int is_fault_pfn(pfn_t pfn)
946 return pfn == fault_pfn;
948 EXPORT_SYMBOL_GPL(is_fault_pfn);
950 int is_noslot_pfn(pfn_t pfn)
952 return pfn == bad_pfn;
954 EXPORT_SYMBOL_GPL(is_noslot_pfn);
956 int is_invalid_pfn(pfn_t pfn)
958 return pfn == hwpoison_pfn || pfn == fault_pfn;
960 EXPORT_SYMBOL_GPL(is_invalid_pfn);
962 static inline unsigned long bad_hva(void)
967 int kvm_is_error_hva(unsigned long addr)
969 return addr == bad_hva();
971 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
973 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
975 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
977 EXPORT_SYMBOL_GPL(gfn_to_memslot);
979 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
981 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
983 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
984 memslot->flags & KVM_MEMSLOT_INVALID)
989 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
991 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
993 struct vm_area_struct *vma;
994 unsigned long addr, size;
998 addr = gfn_to_hva(kvm, gfn);
999 if (kvm_is_error_hva(addr))
1002 down_read(¤t->mm->mmap_sem);
1003 vma = find_vma(current->mm, addr);
1007 size = vma_kernel_pagesize(vma);
1010 up_read(¤t->mm->mmap_sem);
1015 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1018 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1022 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1024 return gfn_to_hva_memslot(slot, gfn);
1027 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1029 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1031 EXPORT_SYMBOL_GPL(gfn_to_hva);
1033 static pfn_t get_fault_pfn(void)
1035 get_page(fault_page);
1039 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1040 unsigned long start, int write, struct page **page)
1042 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1045 flags |= FOLL_WRITE;
1047 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1050 static inline int check_user_page_hwpoison(unsigned long addr)
1052 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1054 rc = __get_user_pages(current, current->mm, addr, 1,
1055 flags, NULL, NULL, NULL);
1056 return rc == -EHWPOISON;
1059 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1060 bool *async, bool write_fault, bool *writable)
1062 struct page *page[1];
1066 /* we can do it either atomically or asynchronously, not both */
1067 BUG_ON(atomic && async);
1069 BUG_ON(!write_fault && !writable);
1074 if (atomic || async)
1075 npages = __get_user_pages_fast(addr, 1, 1, page);
1077 if (unlikely(npages != 1) && !atomic) {
1081 *writable = write_fault;
1084 down_read(¤t->mm->mmap_sem);
1085 npages = get_user_page_nowait(current, current->mm,
1086 addr, write_fault, page);
1087 up_read(¤t->mm->mmap_sem);
1089 npages = get_user_pages_fast(addr, 1, write_fault,
1092 /* map read fault as writable if possible */
1093 if (unlikely(!write_fault) && npages == 1) {
1094 struct page *wpage[1];
1096 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1106 if (unlikely(npages != 1)) {
1107 struct vm_area_struct *vma;
1110 return get_fault_pfn();
1112 down_read(¤t->mm->mmap_sem);
1113 if (npages == -EHWPOISON ||
1114 (!async && check_user_page_hwpoison(addr))) {
1115 up_read(¤t->mm->mmap_sem);
1116 get_page(hwpoison_page);
1117 return page_to_pfn(hwpoison_page);
1120 vma = find_vma_intersection(current->mm, addr, addr+1);
1123 pfn = get_fault_pfn();
1124 else if ((vma->vm_flags & VM_PFNMAP)) {
1125 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1127 BUG_ON(!kvm_is_mmio_pfn(pfn));
1129 if (async && (vma->vm_flags & VM_WRITE))
1131 pfn = get_fault_pfn();
1133 up_read(¤t->mm->mmap_sem);
1135 pfn = page_to_pfn(page[0]);
1140 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1142 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1144 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1146 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1147 bool write_fault, bool *writable)
1154 addr = gfn_to_hva(kvm, gfn);
1155 if (kvm_is_error_hva(addr)) {
1157 return page_to_pfn(bad_page);
1160 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1163 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1165 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1167 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1169 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1170 bool write_fault, bool *writable)
1172 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1174 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1176 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1178 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1180 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1182 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1185 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1187 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1189 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1190 struct kvm_memory_slot *slot, gfn_t gfn)
1192 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1193 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1196 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1202 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1203 if (kvm_is_error_hva(addr))
1206 if (entry < nr_pages)
1209 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1211 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1213 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1217 pfn = gfn_to_pfn(kvm, gfn);
1218 if (!kvm_is_mmio_pfn(pfn))
1219 return pfn_to_page(pfn);
1221 WARN_ON(kvm_is_mmio_pfn(pfn));
1227 EXPORT_SYMBOL_GPL(gfn_to_page);
1229 void kvm_release_page_clean(struct page *page)
1231 kvm_release_pfn_clean(page_to_pfn(page));
1233 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1235 void kvm_release_pfn_clean(pfn_t pfn)
1237 if (!kvm_is_mmio_pfn(pfn))
1238 put_page(pfn_to_page(pfn));
1240 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1242 void kvm_release_page_dirty(struct page *page)
1244 kvm_release_pfn_dirty(page_to_pfn(page));
1246 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1248 void kvm_release_pfn_dirty(pfn_t pfn)
1250 kvm_set_pfn_dirty(pfn);
1251 kvm_release_pfn_clean(pfn);
1253 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1255 void kvm_set_page_dirty(struct page *page)
1257 kvm_set_pfn_dirty(page_to_pfn(page));
1259 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1261 void kvm_set_pfn_dirty(pfn_t pfn)
1263 if (!kvm_is_mmio_pfn(pfn)) {
1264 struct page *page = pfn_to_page(pfn);
1265 if (!PageReserved(page))
1269 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1271 void kvm_set_pfn_accessed(pfn_t pfn)
1273 if (!kvm_is_mmio_pfn(pfn))
1274 mark_page_accessed(pfn_to_page(pfn));
1276 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1278 void kvm_get_pfn(pfn_t pfn)
1280 if (!kvm_is_mmio_pfn(pfn))
1281 get_page(pfn_to_page(pfn));
1283 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1285 static int next_segment(unsigned long len, int offset)
1287 if (len > PAGE_SIZE - offset)
1288 return PAGE_SIZE - offset;
1293 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1299 addr = gfn_to_hva(kvm, gfn);
1300 if (kvm_is_error_hva(addr))
1302 r = __copy_from_user(data, (void __user *)addr + offset, len);
1307 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1309 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1311 gfn_t gfn = gpa >> PAGE_SHIFT;
1313 int offset = offset_in_page(gpa);
1316 while ((seg = next_segment(len, offset)) != 0) {
1317 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1327 EXPORT_SYMBOL_GPL(kvm_read_guest);
1329 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1334 gfn_t gfn = gpa >> PAGE_SHIFT;
1335 int offset = offset_in_page(gpa);
1337 addr = gfn_to_hva(kvm, gfn);
1338 if (kvm_is_error_hva(addr))
1340 pagefault_disable();
1341 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1347 EXPORT_SYMBOL(kvm_read_guest_atomic);
1349 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1350 int offset, int len)
1355 addr = gfn_to_hva(kvm, gfn);
1356 if (kvm_is_error_hva(addr))
1358 r = __copy_to_user((void __user *)addr + offset, data, len);
1361 mark_page_dirty(kvm, gfn);
1364 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1366 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1369 gfn_t gfn = gpa >> PAGE_SHIFT;
1371 int offset = offset_in_page(gpa);
1374 while ((seg = next_segment(len, offset)) != 0) {
1375 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1386 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1389 struct kvm_memslots *slots = kvm_memslots(kvm);
1390 int offset = offset_in_page(gpa);
1391 gfn_t gfn = gpa >> PAGE_SHIFT;
1394 ghc->generation = slots->generation;
1395 ghc->memslot = gfn_to_memslot(kvm, gfn);
1396 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1397 if (!kvm_is_error_hva(ghc->hva))
1404 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1406 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1407 void *data, unsigned long len)
1409 struct kvm_memslots *slots = kvm_memslots(kvm);
1412 if (slots->generation != ghc->generation)
1413 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1415 if (kvm_is_error_hva(ghc->hva))
1418 r = __copy_to_user((void __user *)ghc->hva, data, len);
1421 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1425 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1427 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1428 void *data, unsigned long len)
1430 struct kvm_memslots *slots = kvm_memslots(kvm);
1433 if (slots->generation != ghc->generation)
1434 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1436 if (kvm_is_error_hva(ghc->hva))
1439 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1445 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1447 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1449 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1452 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1454 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1456 gfn_t gfn = gpa >> PAGE_SHIFT;
1458 int offset = offset_in_page(gpa);
1461 while ((seg = next_segment(len, offset)) != 0) {
1462 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1471 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1473 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1476 if (memslot && memslot->dirty_bitmap) {
1477 unsigned long rel_gfn = gfn - memslot->base_gfn;
1479 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1480 memslot->nr_dirty_pages++;
1484 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1486 struct kvm_memory_slot *memslot;
1488 memslot = gfn_to_memslot(kvm, gfn);
1489 mark_page_dirty_in_slot(kvm, memslot, gfn);
1493 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1495 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1500 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1502 if (kvm_arch_vcpu_runnable(vcpu)) {
1503 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1506 if (kvm_cpu_has_pending_timer(vcpu))
1508 if (signal_pending(current))
1514 finish_wait(&vcpu->wq, &wait);
1517 void kvm_resched(struct kvm_vcpu *vcpu)
1519 if (!need_resched())
1523 EXPORT_SYMBOL_GPL(kvm_resched);
1525 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1527 struct kvm *kvm = me->kvm;
1528 struct kvm_vcpu *vcpu;
1529 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1535 * We boost the priority of a VCPU that is runnable but not
1536 * currently running, because it got preempted by something
1537 * else and called schedule in __vcpu_run. Hopefully that
1538 * VCPU is holding the lock that we need and will release it.
1539 * We approximate round-robin by starting at the last boosted VCPU.
1541 for (pass = 0; pass < 2 && !yielded; pass++) {
1542 kvm_for_each_vcpu(i, vcpu, kvm) {
1543 struct task_struct *task = NULL;
1545 if (!pass && i < last_boosted_vcpu) {
1546 i = last_boosted_vcpu;
1548 } else if (pass && i > last_boosted_vcpu)
1552 if (waitqueue_active(&vcpu->wq))
1555 pid = rcu_dereference(vcpu->pid);
1557 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1561 if (task->flags & PF_VCPU) {
1562 put_task_struct(task);
1565 if (yield_to(task, 1)) {
1566 put_task_struct(task);
1567 kvm->last_boosted_vcpu = i;
1571 put_task_struct(task);
1575 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1577 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1579 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1582 if (vmf->pgoff == 0)
1583 page = virt_to_page(vcpu->run);
1585 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1586 page = virt_to_page(vcpu->arch.pio_data);
1588 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1589 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1590 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1593 return kvm_arch_vcpu_fault(vcpu, vmf);
1599 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1600 .fault = kvm_vcpu_fault,
1603 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1605 vma->vm_ops = &kvm_vcpu_vm_ops;
1609 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1611 struct kvm_vcpu *vcpu = filp->private_data;
1613 kvm_put_kvm(vcpu->kvm);
1617 static struct file_operations kvm_vcpu_fops = {
1618 .release = kvm_vcpu_release,
1619 .unlocked_ioctl = kvm_vcpu_ioctl,
1620 #ifdef CONFIG_COMPAT
1621 .compat_ioctl = kvm_vcpu_compat_ioctl,
1623 .mmap = kvm_vcpu_mmap,
1624 .llseek = noop_llseek,
1628 * Allocates an inode for the vcpu.
1630 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1632 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1636 * Creates some virtual cpus. Good luck creating more than one.
1638 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1641 struct kvm_vcpu *vcpu, *v;
1643 vcpu = kvm_arch_vcpu_create(kvm, id);
1645 return PTR_ERR(vcpu);
1647 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1649 r = kvm_arch_vcpu_setup(vcpu);
1653 mutex_lock(&kvm->lock);
1654 if (!kvm_vcpu_compatible(vcpu)) {
1656 goto unlock_vcpu_destroy;
1658 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1660 goto unlock_vcpu_destroy;
1663 kvm_for_each_vcpu(r, v, kvm)
1664 if (v->vcpu_id == id) {
1666 goto unlock_vcpu_destroy;
1669 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1671 /* Now it's all set up, let userspace reach it */
1673 r = create_vcpu_fd(vcpu);
1676 goto unlock_vcpu_destroy;
1679 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1681 atomic_inc(&kvm->online_vcpus);
1683 mutex_unlock(&kvm->lock);
1686 unlock_vcpu_destroy:
1687 mutex_unlock(&kvm->lock);
1689 kvm_arch_vcpu_destroy(vcpu);
1693 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1696 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1697 vcpu->sigset_active = 1;
1698 vcpu->sigset = *sigset;
1700 vcpu->sigset_active = 0;
1704 static long kvm_vcpu_ioctl(struct file *filp,
1705 unsigned int ioctl, unsigned long arg)
1707 struct kvm_vcpu *vcpu = filp->private_data;
1708 void __user *argp = (void __user *)arg;
1710 struct kvm_fpu *fpu = NULL;
1711 struct kvm_sregs *kvm_sregs = NULL;
1713 if (vcpu->kvm->mm != current->mm)
1716 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1718 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1719 * so vcpu_load() would break it.
1721 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1722 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1732 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1733 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1735 case KVM_GET_REGS: {
1736 struct kvm_regs *kvm_regs;
1739 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1742 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1746 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1753 case KVM_SET_REGS: {
1754 struct kvm_regs *kvm_regs;
1757 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1758 if (IS_ERR(kvm_regs)) {
1759 r = PTR_ERR(kvm_regs);
1762 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1770 case KVM_GET_SREGS: {
1771 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1775 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1779 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1784 case KVM_SET_SREGS: {
1785 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1786 if (IS_ERR(kvm_sregs)) {
1787 r = PTR_ERR(kvm_sregs);
1790 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1796 case KVM_GET_MP_STATE: {
1797 struct kvm_mp_state mp_state;
1799 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1803 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1808 case KVM_SET_MP_STATE: {
1809 struct kvm_mp_state mp_state;
1812 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1814 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1820 case KVM_TRANSLATE: {
1821 struct kvm_translation tr;
1824 if (copy_from_user(&tr, argp, sizeof tr))
1826 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1830 if (copy_to_user(argp, &tr, sizeof tr))
1835 case KVM_SET_GUEST_DEBUG: {
1836 struct kvm_guest_debug dbg;
1839 if (copy_from_user(&dbg, argp, sizeof dbg))
1841 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1847 case KVM_SET_SIGNAL_MASK: {
1848 struct kvm_signal_mask __user *sigmask_arg = argp;
1849 struct kvm_signal_mask kvm_sigmask;
1850 sigset_t sigset, *p;
1855 if (copy_from_user(&kvm_sigmask, argp,
1856 sizeof kvm_sigmask))
1859 if (kvm_sigmask.len != sizeof sigset)
1862 if (copy_from_user(&sigset, sigmask_arg->sigset,
1867 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1871 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1875 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1879 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1885 fpu = memdup_user(argp, sizeof(*fpu));
1890 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1897 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1906 #ifdef CONFIG_COMPAT
1907 static long kvm_vcpu_compat_ioctl(struct file *filp,
1908 unsigned int ioctl, unsigned long arg)
1910 struct kvm_vcpu *vcpu = filp->private_data;
1911 void __user *argp = compat_ptr(arg);
1914 if (vcpu->kvm->mm != current->mm)
1918 case KVM_SET_SIGNAL_MASK: {
1919 struct kvm_signal_mask __user *sigmask_arg = argp;
1920 struct kvm_signal_mask kvm_sigmask;
1921 compat_sigset_t csigset;
1926 if (copy_from_user(&kvm_sigmask, argp,
1927 sizeof kvm_sigmask))
1930 if (kvm_sigmask.len != sizeof csigset)
1933 if (copy_from_user(&csigset, sigmask_arg->sigset,
1937 sigset_from_compat(&sigset, &csigset);
1938 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1942 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1950 static long kvm_vm_ioctl(struct file *filp,
1951 unsigned int ioctl, unsigned long arg)
1953 struct kvm *kvm = filp->private_data;
1954 void __user *argp = (void __user *)arg;
1957 if (kvm->mm != current->mm)
1960 case KVM_CREATE_VCPU:
1961 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1965 case KVM_SET_USER_MEMORY_REGION: {
1966 struct kvm_userspace_memory_region kvm_userspace_mem;
1969 if (copy_from_user(&kvm_userspace_mem, argp,
1970 sizeof kvm_userspace_mem))
1973 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1978 case KVM_GET_DIRTY_LOG: {
1979 struct kvm_dirty_log log;
1982 if (copy_from_user(&log, argp, sizeof log))
1984 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1989 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1990 case KVM_REGISTER_COALESCED_MMIO: {
1991 struct kvm_coalesced_mmio_zone zone;
1993 if (copy_from_user(&zone, argp, sizeof zone))
1995 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2001 case KVM_UNREGISTER_COALESCED_MMIO: {
2002 struct kvm_coalesced_mmio_zone zone;
2004 if (copy_from_user(&zone, argp, sizeof zone))
2006 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2014 struct kvm_irqfd data;
2017 if (copy_from_user(&data, argp, sizeof data))
2019 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2022 case KVM_IOEVENTFD: {
2023 struct kvm_ioeventfd data;
2026 if (copy_from_user(&data, argp, sizeof data))
2028 r = kvm_ioeventfd(kvm, &data);
2031 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2032 case KVM_SET_BOOT_CPU_ID:
2034 mutex_lock(&kvm->lock);
2035 if (atomic_read(&kvm->online_vcpus) != 0)
2038 kvm->bsp_vcpu_id = arg;
2039 mutex_unlock(&kvm->lock);
2043 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2045 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2051 #ifdef CONFIG_COMPAT
2052 struct compat_kvm_dirty_log {
2056 compat_uptr_t dirty_bitmap; /* one bit per page */
2061 static long kvm_vm_compat_ioctl(struct file *filp,
2062 unsigned int ioctl, unsigned long arg)
2064 struct kvm *kvm = filp->private_data;
2067 if (kvm->mm != current->mm)
2070 case KVM_GET_DIRTY_LOG: {
2071 struct compat_kvm_dirty_log compat_log;
2072 struct kvm_dirty_log log;
2075 if (copy_from_user(&compat_log, (void __user *)arg,
2076 sizeof(compat_log)))
2078 log.slot = compat_log.slot;
2079 log.padding1 = compat_log.padding1;
2080 log.padding2 = compat_log.padding2;
2081 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2083 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2089 r = kvm_vm_ioctl(filp, ioctl, arg);
2097 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2099 struct page *page[1];
2102 gfn_t gfn = vmf->pgoff;
2103 struct kvm *kvm = vma->vm_file->private_data;
2105 addr = gfn_to_hva(kvm, gfn);
2106 if (kvm_is_error_hva(addr))
2107 return VM_FAULT_SIGBUS;
2109 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2111 if (unlikely(npages != 1))
2112 return VM_FAULT_SIGBUS;
2114 vmf->page = page[0];
2118 static const struct vm_operations_struct kvm_vm_vm_ops = {
2119 .fault = kvm_vm_fault,
2122 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2124 vma->vm_ops = &kvm_vm_vm_ops;
2128 static struct file_operations kvm_vm_fops = {
2129 .release = kvm_vm_release,
2130 .unlocked_ioctl = kvm_vm_ioctl,
2131 #ifdef CONFIG_COMPAT
2132 .compat_ioctl = kvm_vm_compat_ioctl,
2134 .mmap = kvm_vm_mmap,
2135 .llseek = noop_llseek,
2138 static int kvm_dev_ioctl_create_vm(unsigned long type)
2143 kvm = kvm_create_vm(type);
2145 return PTR_ERR(kvm);
2146 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2147 r = kvm_coalesced_mmio_init(kvm);
2153 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2160 static long kvm_dev_ioctl_check_extension_generic(long arg)
2163 case KVM_CAP_USER_MEMORY:
2164 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2165 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2166 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2167 case KVM_CAP_SET_BOOT_CPU_ID:
2169 case KVM_CAP_INTERNAL_ERROR_DATA:
2171 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2172 case KVM_CAP_IRQ_ROUTING:
2173 return KVM_MAX_IRQ_ROUTES;
2178 return kvm_dev_ioctl_check_extension(arg);
2181 static long kvm_dev_ioctl(struct file *filp,
2182 unsigned int ioctl, unsigned long arg)
2187 case KVM_GET_API_VERSION:
2191 r = KVM_API_VERSION;
2194 r = kvm_dev_ioctl_create_vm(arg);
2196 case KVM_CHECK_EXTENSION:
2197 r = kvm_dev_ioctl_check_extension_generic(arg);
2199 case KVM_GET_VCPU_MMAP_SIZE:
2203 r = PAGE_SIZE; /* struct kvm_run */
2205 r += PAGE_SIZE; /* pio data page */
2207 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2208 r += PAGE_SIZE; /* coalesced mmio ring page */
2211 case KVM_TRACE_ENABLE:
2212 case KVM_TRACE_PAUSE:
2213 case KVM_TRACE_DISABLE:
2217 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2223 static struct file_operations kvm_chardev_ops = {
2224 .unlocked_ioctl = kvm_dev_ioctl,
2225 .compat_ioctl = kvm_dev_ioctl,
2226 .llseek = noop_llseek,
2229 static struct miscdevice kvm_dev = {
2235 static void hardware_enable_nolock(void *junk)
2237 int cpu = raw_smp_processor_id();
2240 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2243 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2245 r = kvm_arch_hardware_enable(NULL);
2248 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2249 atomic_inc(&hardware_enable_failed);
2250 printk(KERN_INFO "kvm: enabling virtualization on "
2251 "CPU%d failed\n", cpu);
2255 static void hardware_enable(void *junk)
2257 raw_spin_lock(&kvm_lock);
2258 hardware_enable_nolock(junk);
2259 raw_spin_unlock(&kvm_lock);
2262 static void hardware_disable_nolock(void *junk)
2264 int cpu = raw_smp_processor_id();
2266 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2268 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2269 kvm_arch_hardware_disable(NULL);
2272 static void hardware_disable(void *junk)
2274 raw_spin_lock(&kvm_lock);
2275 hardware_disable_nolock(junk);
2276 raw_spin_unlock(&kvm_lock);
2279 static void hardware_disable_all_nolock(void)
2281 BUG_ON(!kvm_usage_count);
2284 if (!kvm_usage_count)
2285 on_each_cpu(hardware_disable_nolock, NULL, 1);
2288 static void hardware_disable_all(void)
2290 raw_spin_lock(&kvm_lock);
2291 hardware_disable_all_nolock();
2292 raw_spin_unlock(&kvm_lock);
2295 static int hardware_enable_all(void)
2299 raw_spin_lock(&kvm_lock);
2302 if (kvm_usage_count == 1) {
2303 atomic_set(&hardware_enable_failed, 0);
2304 on_each_cpu(hardware_enable_nolock, NULL, 1);
2306 if (atomic_read(&hardware_enable_failed)) {
2307 hardware_disable_all_nolock();
2312 raw_spin_unlock(&kvm_lock);
2317 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2322 if (!kvm_usage_count)
2325 val &= ~CPU_TASKS_FROZEN;
2328 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2330 hardware_disable(NULL);
2333 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2335 hardware_enable(NULL);
2342 asmlinkage void kvm_spurious_fault(void)
2344 /* Fault while not rebooting. We want the trace. */
2347 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2349 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2353 * Some (well, at least mine) BIOSes hang on reboot if
2356 * And Intel TXT required VMX off for all cpu when system shutdown.
2358 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2359 kvm_rebooting = true;
2360 on_each_cpu(hardware_disable_nolock, NULL, 1);
2364 static struct notifier_block kvm_reboot_notifier = {
2365 .notifier_call = kvm_reboot,
2369 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2373 for (i = 0; i < bus->dev_count; i++) {
2374 struct kvm_io_device *pos = bus->range[i].dev;
2376 kvm_iodevice_destructor(pos);
2381 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2383 const struct kvm_io_range *r1 = p1;
2384 const struct kvm_io_range *r2 = p2;
2386 if (r1->addr < r2->addr)
2388 if (r1->addr + r1->len > r2->addr + r2->len)
2393 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2394 gpa_t addr, int len)
2396 if (bus->dev_count == NR_IOBUS_DEVS)
2399 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2405 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2406 kvm_io_bus_sort_cmp, NULL);
2411 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2412 gpa_t addr, int len)
2414 struct kvm_io_range *range, key;
2417 key = (struct kvm_io_range) {
2422 range = bsearch(&key, bus->range, bus->dev_count,
2423 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2427 off = range - bus->range;
2429 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2435 /* kvm_io_bus_write - called under kvm->slots_lock */
2436 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2437 int len, const void *val)
2440 struct kvm_io_bus *bus;
2441 struct kvm_io_range range;
2443 range = (struct kvm_io_range) {
2448 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2449 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2453 while (idx < bus->dev_count &&
2454 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2455 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2463 /* kvm_io_bus_read - called under kvm->slots_lock */
2464 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2468 struct kvm_io_bus *bus;
2469 struct kvm_io_range range;
2471 range = (struct kvm_io_range) {
2476 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2477 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2481 while (idx < bus->dev_count &&
2482 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2483 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2491 /* Caller must hold slots_lock. */
2492 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2493 int len, struct kvm_io_device *dev)
2495 struct kvm_io_bus *new_bus, *bus;
2497 bus = kvm->buses[bus_idx];
2498 if (bus->dev_count > NR_IOBUS_DEVS-1)
2501 new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2504 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2505 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2506 synchronize_srcu_expedited(&kvm->srcu);
2512 /* Caller must hold slots_lock. */
2513 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2514 struct kvm_io_device *dev)
2517 struct kvm_io_bus *new_bus, *bus;
2519 bus = kvm->buses[bus_idx];
2521 new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2526 for (i = 0; i < new_bus->dev_count; i++)
2527 if (new_bus->range[i].dev == dev) {
2529 new_bus->dev_count--;
2530 new_bus->range[i] = new_bus->range[new_bus->dev_count];
2531 sort(new_bus->range, new_bus->dev_count,
2532 sizeof(struct kvm_io_range),
2533 kvm_io_bus_sort_cmp, NULL);
2542 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2543 synchronize_srcu_expedited(&kvm->srcu);
2548 static struct notifier_block kvm_cpu_notifier = {
2549 .notifier_call = kvm_cpu_hotplug,
2552 static int vm_stat_get(void *_offset, u64 *val)
2554 unsigned offset = (long)_offset;
2558 raw_spin_lock(&kvm_lock);
2559 list_for_each_entry(kvm, &vm_list, vm_list)
2560 *val += *(u32 *)((void *)kvm + offset);
2561 raw_spin_unlock(&kvm_lock);
2565 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2567 static int vcpu_stat_get(void *_offset, u64 *val)
2569 unsigned offset = (long)_offset;
2571 struct kvm_vcpu *vcpu;
2575 raw_spin_lock(&kvm_lock);
2576 list_for_each_entry(kvm, &vm_list, vm_list)
2577 kvm_for_each_vcpu(i, vcpu, kvm)
2578 *val += *(u32 *)((void *)vcpu + offset);
2580 raw_spin_unlock(&kvm_lock);
2584 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2586 static const struct file_operations *stat_fops[] = {
2587 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2588 [KVM_STAT_VM] = &vm_stat_fops,
2591 static int kvm_init_debug(void)
2594 struct kvm_stats_debugfs_item *p;
2596 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2597 if (kvm_debugfs_dir == NULL)
2600 for (p = debugfs_entries; p->name; ++p) {
2601 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2602 (void *)(long)p->offset,
2603 stat_fops[p->kind]);
2604 if (p->dentry == NULL)
2611 debugfs_remove_recursive(kvm_debugfs_dir);
2616 static void kvm_exit_debug(void)
2618 struct kvm_stats_debugfs_item *p;
2620 for (p = debugfs_entries; p->name; ++p)
2621 debugfs_remove(p->dentry);
2622 debugfs_remove(kvm_debugfs_dir);
2625 static int kvm_suspend(void)
2627 if (kvm_usage_count)
2628 hardware_disable_nolock(NULL);
2632 static void kvm_resume(void)
2634 if (kvm_usage_count) {
2635 WARN_ON(raw_spin_is_locked(&kvm_lock));
2636 hardware_enable_nolock(NULL);
2640 static struct syscore_ops kvm_syscore_ops = {
2641 .suspend = kvm_suspend,
2642 .resume = kvm_resume,
2645 struct page *bad_page;
2649 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2651 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2654 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2656 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2658 kvm_arch_vcpu_load(vcpu, cpu);
2661 static void kvm_sched_out(struct preempt_notifier *pn,
2662 struct task_struct *next)
2664 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2666 kvm_arch_vcpu_put(vcpu);
2669 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2670 struct module *module)
2675 r = kvm_arch_init(opaque);
2679 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2681 if (bad_page == NULL) {
2686 bad_pfn = page_to_pfn(bad_page);
2688 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2690 if (hwpoison_page == NULL) {
2695 hwpoison_pfn = page_to_pfn(hwpoison_page);
2697 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2699 if (fault_page == NULL) {
2704 fault_pfn = page_to_pfn(fault_page);
2706 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2711 r = kvm_arch_hardware_setup();
2715 for_each_online_cpu(cpu) {
2716 smp_call_function_single(cpu,
2717 kvm_arch_check_processor_compat,
2723 r = register_cpu_notifier(&kvm_cpu_notifier);
2726 register_reboot_notifier(&kvm_reboot_notifier);
2728 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2730 vcpu_align = __alignof__(struct kvm_vcpu);
2731 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2733 if (!kvm_vcpu_cache) {
2738 r = kvm_async_pf_init();
2742 kvm_chardev_ops.owner = module;
2743 kvm_vm_fops.owner = module;
2744 kvm_vcpu_fops.owner = module;
2746 r = misc_register(&kvm_dev);
2748 printk(KERN_ERR "kvm: misc device register failed\n");
2752 register_syscore_ops(&kvm_syscore_ops);
2754 kvm_preempt_ops.sched_in = kvm_sched_in;
2755 kvm_preempt_ops.sched_out = kvm_sched_out;
2757 r = kvm_init_debug();
2759 printk(KERN_ERR "kvm: create debugfs files failed\n");
2766 unregister_syscore_ops(&kvm_syscore_ops);
2768 kvm_async_pf_deinit();
2770 kmem_cache_destroy(kvm_vcpu_cache);
2772 unregister_reboot_notifier(&kvm_reboot_notifier);
2773 unregister_cpu_notifier(&kvm_cpu_notifier);
2776 kvm_arch_hardware_unsetup();
2778 free_cpumask_var(cpus_hardware_enabled);
2781 __free_page(fault_page);
2783 __free_page(hwpoison_page);
2784 __free_page(bad_page);
2790 EXPORT_SYMBOL_GPL(kvm_init);
2795 misc_deregister(&kvm_dev);
2796 kmem_cache_destroy(kvm_vcpu_cache);
2797 kvm_async_pf_deinit();
2798 unregister_syscore_ops(&kvm_syscore_ops);
2799 unregister_reboot_notifier(&kvm_reboot_notifier);
2800 unregister_cpu_notifier(&kvm_cpu_notifier);
2801 on_each_cpu(hardware_disable_nolock, NULL, 1);
2802 kvm_arch_hardware_unsetup();
2804 free_cpumask_var(cpus_hardware_enabled);
2805 __free_page(hwpoison_page);
2806 __free_page(bad_page);
2808 EXPORT_SYMBOL_GPL(kvm_exit);