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);
527 kfree(memslot->dirty_bitmap);
529 memslot->dirty_bitmap = NULL;
533 * Free any memory in @free but not in @dont.
535 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
536 struct kvm_memory_slot *dont)
538 if (!dont || free->rmap != dont->rmap)
541 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
542 kvm_destroy_dirty_bitmap(free);
544 kvm_arch_free_memslot(free, dont);
550 void kvm_free_physmem(struct kvm *kvm)
552 struct kvm_memslots *slots = kvm->memslots;
553 struct kvm_memory_slot *memslot;
555 kvm_for_each_memslot(memslot, slots)
556 kvm_free_physmem_slot(memslot, NULL);
558 kfree(kvm->memslots);
561 static void kvm_destroy_vm(struct kvm *kvm)
564 struct mm_struct *mm = kvm->mm;
566 kvm_arch_sync_events(kvm);
567 raw_spin_lock(&kvm_lock);
568 list_del(&kvm->vm_list);
569 raw_spin_unlock(&kvm_lock);
570 kvm_free_irq_routing(kvm);
571 for (i = 0; i < KVM_NR_BUSES; i++)
572 kvm_io_bus_destroy(kvm->buses[i]);
573 kvm_coalesced_mmio_free(kvm);
574 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
575 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
577 kvm_arch_flush_shadow(kvm);
579 kvm_arch_destroy_vm(kvm);
580 kvm_free_physmem(kvm);
581 cleanup_srcu_struct(&kvm->srcu);
582 kvm_arch_free_vm(kvm);
583 hardware_disable_all();
587 void kvm_get_kvm(struct kvm *kvm)
589 atomic_inc(&kvm->users_count);
591 EXPORT_SYMBOL_GPL(kvm_get_kvm);
593 void kvm_put_kvm(struct kvm *kvm)
595 if (atomic_dec_and_test(&kvm->users_count))
598 EXPORT_SYMBOL_GPL(kvm_put_kvm);
601 static int kvm_vm_release(struct inode *inode, struct file *filp)
603 struct kvm *kvm = filp->private_data;
605 kvm_irqfd_release(kvm);
612 * Allocation size is twice as large as the actual dirty bitmap size.
613 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
615 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
618 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
620 if (dirty_bytes > PAGE_SIZE)
621 memslot->dirty_bitmap = vzalloc(dirty_bytes);
623 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
625 if (!memslot->dirty_bitmap)
628 #endif /* !CONFIG_S390 */
632 static int cmp_memslot(const void *slot1, const void *slot2)
634 struct kvm_memory_slot *s1, *s2;
636 s1 = (struct kvm_memory_slot *)slot1;
637 s2 = (struct kvm_memory_slot *)slot2;
639 if (s1->npages < s2->npages)
641 if (s1->npages > s2->npages)
648 * Sort the memslots base on its size, so the larger slots
649 * will get better fit.
651 static void sort_memslots(struct kvm_memslots *slots)
655 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
656 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
658 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
659 slots->id_to_index[slots->memslots[i].id] = i;
662 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
666 struct kvm_memory_slot *old = id_to_memslot(slots, id);
667 unsigned long npages = old->npages;
670 if (new->npages != npages)
671 sort_memslots(slots);
678 * Allocate some memory and give it an address in the guest physical address
681 * Discontiguous memory is allowed, mostly for framebuffers.
683 * Must be called holding mmap_sem for write.
685 int __kvm_set_memory_region(struct kvm *kvm,
686 struct kvm_userspace_memory_region *mem,
691 unsigned long npages;
693 struct kvm_memory_slot *memslot;
694 struct kvm_memory_slot old, new;
695 struct kvm_memslots *slots, *old_memslots;
698 /* General sanity checks */
699 if (mem->memory_size & (PAGE_SIZE - 1))
701 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
703 /* We can read the guest memory with __xxx_user() later on. */
705 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
706 !access_ok(VERIFY_WRITE,
707 (void __user *)(unsigned long)mem->userspace_addr,
710 if (mem->slot >= KVM_MEM_SLOTS_NUM)
712 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
715 memslot = id_to_memslot(kvm->memslots, mem->slot);
716 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
717 npages = mem->memory_size >> PAGE_SHIFT;
720 if (npages > KVM_MEM_MAX_NR_PAGES)
724 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
726 new = old = *memslot;
729 new.base_gfn = base_gfn;
731 new.flags = mem->flags;
733 /* Disallow changing a memory slot's size. */
735 if (npages && old.npages && npages != old.npages)
738 /* Check for overlaps */
740 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
741 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
743 if (s == memslot || !s->npages)
745 if (!((base_gfn + npages <= s->base_gfn) ||
746 (base_gfn >= s->base_gfn + s->npages)))
750 /* Free page dirty bitmap if unneeded */
751 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
752 new.dirty_bitmap = NULL;
756 /* Allocate if a slot is being created */
757 if (npages && !old.npages) {
758 new.user_alloc = user_alloc;
759 new.userspace_addr = mem->userspace_addr;
761 new.rmap = vzalloc(npages * sizeof(*new.rmap));
764 #endif /* not defined CONFIG_S390 */
765 if (kvm_arch_create_memslot(&new, npages))
769 /* Allocate page dirty bitmap if needed */
770 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
771 if (kvm_create_dirty_bitmap(&new) < 0)
773 /* destroy any largepage mappings for dirty tracking */
777 struct kvm_memory_slot *slot;
780 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
784 slot = id_to_memslot(slots, mem->slot);
785 slot->flags |= KVM_MEMSLOT_INVALID;
787 update_memslots(slots, NULL);
789 old_memslots = kvm->memslots;
790 rcu_assign_pointer(kvm->memslots, slots);
791 synchronize_srcu_expedited(&kvm->srcu);
792 /* From this point no new shadow pages pointing to a deleted
793 * memslot will be created.
795 * validation of sp->gfn happens in:
796 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
797 * - kvm_is_visible_gfn (mmu_check_roots)
799 kvm_arch_flush_shadow(kvm);
803 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
807 /* map/unmap the pages in iommu page table */
809 r = kvm_iommu_map_pages(kvm, &new);
813 kvm_iommu_unmap_pages(kvm, &old);
816 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
821 /* actual memory is freed via old in kvm_free_physmem_slot below */
824 new.dirty_bitmap = NULL;
825 memset(&new.arch, 0, sizeof(new.arch));
828 update_memslots(slots, &new);
829 old_memslots = kvm->memslots;
830 rcu_assign_pointer(kvm->memslots, slots);
831 synchronize_srcu_expedited(&kvm->srcu);
833 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
836 * If the new memory slot is created, we need to clear all
839 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
840 kvm_arch_flush_shadow(kvm);
842 kvm_free_physmem_slot(&old, &new);
848 kvm_free_physmem_slot(&new, &old);
853 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
855 int kvm_set_memory_region(struct kvm *kvm,
856 struct kvm_userspace_memory_region *mem,
861 mutex_lock(&kvm->slots_lock);
862 r = __kvm_set_memory_region(kvm, mem, user_alloc);
863 mutex_unlock(&kvm->slots_lock);
866 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
868 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
870 kvm_userspace_memory_region *mem,
873 if (mem->slot >= KVM_MEMORY_SLOTS)
875 return kvm_set_memory_region(kvm, mem, user_alloc);
878 int kvm_get_dirty_log(struct kvm *kvm,
879 struct kvm_dirty_log *log, int *is_dirty)
881 struct kvm_memory_slot *memslot;
884 unsigned long any = 0;
887 if (log->slot >= KVM_MEMORY_SLOTS)
890 memslot = id_to_memslot(kvm->memslots, log->slot);
892 if (!memslot->dirty_bitmap)
895 n = kvm_dirty_bitmap_bytes(memslot);
897 for (i = 0; !any && i < n/sizeof(long); ++i)
898 any = memslot->dirty_bitmap[i];
901 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
912 bool kvm_largepages_enabled(void)
914 return largepages_enabled;
917 void kvm_disable_largepages(void)
919 largepages_enabled = false;
921 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
923 int is_error_page(struct page *page)
925 return page == bad_page || page == hwpoison_page || page == fault_page;
927 EXPORT_SYMBOL_GPL(is_error_page);
929 int is_error_pfn(pfn_t pfn)
931 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
933 EXPORT_SYMBOL_GPL(is_error_pfn);
935 int is_hwpoison_pfn(pfn_t pfn)
937 return pfn == hwpoison_pfn;
939 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
941 int is_fault_pfn(pfn_t pfn)
943 return pfn == fault_pfn;
945 EXPORT_SYMBOL_GPL(is_fault_pfn);
947 int is_noslot_pfn(pfn_t pfn)
949 return pfn == bad_pfn;
951 EXPORT_SYMBOL_GPL(is_noslot_pfn);
953 int is_invalid_pfn(pfn_t pfn)
955 return pfn == hwpoison_pfn || pfn == fault_pfn;
957 EXPORT_SYMBOL_GPL(is_invalid_pfn);
959 static inline unsigned long bad_hva(void)
964 int kvm_is_error_hva(unsigned long addr)
966 return addr == bad_hva();
968 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
970 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
972 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
974 EXPORT_SYMBOL_GPL(gfn_to_memslot);
976 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
978 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
980 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
981 memslot->flags & KVM_MEMSLOT_INVALID)
986 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
988 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
990 struct vm_area_struct *vma;
991 unsigned long addr, size;
995 addr = gfn_to_hva(kvm, gfn);
996 if (kvm_is_error_hva(addr))
999 down_read(¤t->mm->mmap_sem);
1000 vma = find_vma(current->mm, addr);
1004 size = vma_kernel_pagesize(vma);
1007 up_read(¤t->mm->mmap_sem);
1012 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1015 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1019 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1021 return gfn_to_hva_memslot(slot, gfn);
1024 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1026 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1028 EXPORT_SYMBOL_GPL(gfn_to_hva);
1030 static pfn_t get_fault_pfn(void)
1032 get_page(fault_page);
1036 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1037 unsigned long start, int write, struct page **page)
1039 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1042 flags |= FOLL_WRITE;
1044 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1047 static inline int check_user_page_hwpoison(unsigned long addr)
1049 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1051 rc = __get_user_pages(current, current->mm, addr, 1,
1052 flags, NULL, NULL, NULL);
1053 return rc == -EHWPOISON;
1056 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1057 bool *async, bool write_fault, bool *writable)
1059 struct page *page[1];
1063 /* we can do it either atomically or asynchronously, not both */
1064 BUG_ON(atomic && async);
1066 BUG_ON(!write_fault && !writable);
1071 if (atomic || async)
1072 npages = __get_user_pages_fast(addr, 1, 1, page);
1074 if (unlikely(npages != 1) && !atomic) {
1078 *writable = write_fault;
1081 down_read(¤t->mm->mmap_sem);
1082 npages = get_user_page_nowait(current, current->mm,
1083 addr, write_fault, page);
1084 up_read(¤t->mm->mmap_sem);
1086 npages = get_user_pages_fast(addr, 1, write_fault,
1089 /* map read fault as writable if possible */
1090 if (unlikely(!write_fault) && npages == 1) {
1091 struct page *wpage[1];
1093 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1103 if (unlikely(npages != 1)) {
1104 struct vm_area_struct *vma;
1107 return get_fault_pfn();
1109 down_read(¤t->mm->mmap_sem);
1110 if (npages == -EHWPOISON ||
1111 (!async && check_user_page_hwpoison(addr))) {
1112 up_read(¤t->mm->mmap_sem);
1113 get_page(hwpoison_page);
1114 return page_to_pfn(hwpoison_page);
1117 vma = find_vma_intersection(current->mm, addr, addr+1);
1120 pfn = get_fault_pfn();
1121 else if ((vma->vm_flags & VM_PFNMAP)) {
1122 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1124 BUG_ON(!kvm_is_mmio_pfn(pfn));
1126 if (async && (vma->vm_flags & VM_WRITE))
1128 pfn = get_fault_pfn();
1130 up_read(¤t->mm->mmap_sem);
1132 pfn = page_to_pfn(page[0]);
1137 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1139 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1141 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1143 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1144 bool write_fault, bool *writable)
1151 addr = gfn_to_hva(kvm, gfn);
1152 if (kvm_is_error_hva(addr)) {
1154 return page_to_pfn(bad_page);
1157 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1160 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1162 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1164 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1166 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1167 bool write_fault, bool *writable)
1169 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1171 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1173 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1175 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1177 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1179 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1182 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1184 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1186 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1187 struct kvm_memory_slot *slot, gfn_t gfn)
1189 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1190 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1193 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1199 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1200 if (kvm_is_error_hva(addr))
1203 if (entry < nr_pages)
1206 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1208 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1210 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1214 pfn = gfn_to_pfn(kvm, gfn);
1215 if (!kvm_is_mmio_pfn(pfn))
1216 return pfn_to_page(pfn);
1218 WARN_ON(kvm_is_mmio_pfn(pfn));
1224 EXPORT_SYMBOL_GPL(gfn_to_page);
1226 void kvm_release_page_clean(struct page *page)
1228 kvm_release_pfn_clean(page_to_pfn(page));
1230 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1232 void kvm_release_pfn_clean(pfn_t pfn)
1234 if (!kvm_is_mmio_pfn(pfn))
1235 put_page(pfn_to_page(pfn));
1237 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1239 void kvm_release_page_dirty(struct page *page)
1241 kvm_release_pfn_dirty(page_to_pfn(page));
1243 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1245 void kvm_release_pfn_dirty(pfn_t pfn)
1247 kvm_set_pfn_dirty(pfn);
1248 kvm_release_pfn_clean(pfn);
1250 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1252 void kvm_set_page_dirty(struct page *page)
1254 kvm_set_pfn_dirty(page_to_pfn(page));
1256 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1258 void kvm_set_pfn_dirty(pfn_t pfn)
1260 if (!kvm_is_mmio_pfn(pfn)) {
1261 struct page *page = pfn_to_page(pfn);
1262 if (!PageReserved(page))
1266 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1268 void kvm_set_pfn_accessed(pfn_t pfn)
1270 if (!kvm_is_mmio_pfn(pfn))
1271 mark_page_accessed(pfn_to_page(pfn));
1273 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1275 void kvm_get_pfn(pfn_t pfn)
1277 if (!kvm_is_mmio_pfn(pfn))
1278 get_page(pfn_to_page(pfn));
1280 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1282 static int next_segment(unsigned long len, int offset)
1284 if (len > PAGE_SIZE - offset)
1285 return PAGE_SIZE - offset;
1290 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1296 addr = gfn_to_hva(kvm, gfn);
1297 if (kvm_is_error_hva(addr))
1299 r = __copy_from_user(data, (void __user *)addr + offset, len);
1304 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1306 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1308 gfn_t gfn = gpa >> PAGE_SHIFT;
1310 int offset = offset_in_page(gpa);
1313 while ((seg = next_segment(len, offset)) != 0) {
1314 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1324 EXPORT_SYMBOL_GPL(kvm_read_guest);
1326 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1331 gfn_t gfn = gpa >> PAGE_SHIFT;
1332 int offset = offset_in_page(gpa);
1334 addr = gfn_to_hva(kvm, gfn);
1335 if (kvm_is_error_hva(addr))
1337 pagefault_disable();
1338 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1344 EXPORT_SYMBOL(kvm_read_guest_atomic);
1346 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1347 int offset, int len)
1352 addr = gfn_to_hva(kvm, gfn);
1353 if (kvm_is_error_hva(addr))
1355 r = __copy_to_user((void __user *)addr + offset, data, len);
1358 mark_page_dirty(kvm, gfn);
1361 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1363 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1366 gfn_t gfn = gpa >> PAGE_SHIFT;
1368 int offset = offset_in_page(gpa);
1371 while ((seg = next_segment(len, offset)) != 0) {
1372 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1383 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1386 struct kvm_memslots *slots = kvm_memslots(kvm);
1387 int offset = offset_in_page(gpa);
1388 gfn_t gfn = gpa >> PAGE_SHIFT;
1391 ghc->generation = slots->generation;
1392 ghc->memslot = gfn_to_memslot(kvm, gfn);
1393 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1394 if (!kvm_is_error_hva(ghc->hva))
1401 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1403 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1404 void *data, unsigned long len)
1406 struct kvm_memslots *slots = kvm_memslots(kvm);
1409 if (slots->generation != ghc->generation)
1410 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1412 if (kvm_is_error_hva(ghc->hva))
1415 r = __copy_to_user((void __user *)ghc->hva, data, len);
1418 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1422 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1424 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1425 void *data, unsigned long len)
1427 struct kvm_memslots *slots = kvm_memslots(kvm);
1430 if (slots->generation != ghc->generation)
1431 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1433 if (kvm_is_error_hva(ghc->hva))
1436 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1442 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1444 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1446 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1449 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1451 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1453 gfn_t gfn = gpa >> PAGE_SHIFT;
1455 int offset = offset_in_page(gpa);
1458 while ((seg = next_segment(len, offset)) != 0) {
1459 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1468 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1470 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1473 if (memslot && memslot->dirty_bitmap) {
1474 unsigned long rel_gfn = gfn - memslot->base_gfn;
1476 /* TODO: introduce set_bit_le() and use it */
1477 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1481 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1483 struct kvm_memory_slot *memslot;
1485 memslot = gfn_to_memslot(kvm, gfn);
1486 mark_page_dirty_in_slot(kvm, memslot, gfn);
1490 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1492 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1497 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1499 if (kvm_arch_vcpu_runnable(vcpu)) {
1500 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1503 if (kvm_cpu_has_pending_timer(vcpu))
1505 if (signal_pending(current))
1511 finish_wait(&vcpu->wq, &wait);
1516 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1518 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1521 int cpu = vcpu->cpu;
1522 wait_queue_head_t *wqp;
1524 wqp = kvm_arch_vcpu_wq(vcpu);
1525 if (waitqueue_active(wqp)) {
1526 wake_up_interruptible(wqp);
1527 ++vcpu->stat.halt_wakeup;
1531 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1532 if (kvm_arch_vcpu_should_kick(vcpu))
1533 smp_send_reschedule(cpu);
1536 #endif /* !CONFIG_S390 */
1538 void kvm_resched(struct kvm_vcpu *vcpu)
1540 if (!need_resched())
1544 EXPORT_SYMBOL_GPL(kvm_resched);
1546 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1549 struct task_struct *task = NULL;
1552 pid = rcu_dereference(target->pid);
1554 task = get_pid_task(target->pid, PIDTYPE_PID);
1558 if (task->flags & PF_VCPU) {
1559 put_task_struct(task);
1562 if (yield_to(task, 1)) {
1563 put_task_struct(task);
1566 put_task_struct(task);
1569 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1571 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1573 struct kvm *kvm = me->kvm;
1574 struct kvm_vcpu *vcpu;
1575 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1581 * We boost the priority of a VCPU that is runnable but not
1582 * currently running, because it got preempted by something
1583 * else and called schedule in __vcpu_run. Hopefully that
1584 * VCPU is holding the lock that we need and will release it.
1585 * We approximate round-robin by starting at the last boosted VCPU.
1587 for (pass = 0; pass < 2 && !yielded; pass++) {
1588 kvm_for_each_vcpu(i, vcpu, kvm) {
1589 if (!pass && i < last_boosted_vcpu) {
1590 i = last_boosted_vcpu;
1592 } else if (pass && i > last_boosted_vcpu)
1596 if (waitqueue_active(&vcpu->wq))
1598 if (kvm_vcpu_yield_to(vcpu)) {
1599 kvm->last_boosted_vcpu = i;
1606 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1608 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1610 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1613 if (vmf->pgoff == 0)
1614 page = virt_to_page(vcpu->run);
1616 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1617 page = virt_to_page(vcpu->arch.pio_data);
1619 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1620 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1621 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1624 return kvm_arch_vcpu_fault(vcpu, vmf);
1630 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1631 .fault = kvm_vcpu_fault,
1634 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1636 vma->vm_ops = &kvm_vcpu_vm_ops;
1640 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1642 struct kvm_vcpu *vcpu = filp->private_data;
1644 kvm_put_kvm(vcpu->kvm);
1648 static struct file_operations kvm_vcpu_fops = {
1649 .release = kvm_vcpu_release,
1650 .unlocked_ioctl = kvm_vcpu_ioctl,
1651 #ifdef CONFIG_COMPAT
1652 .compat_ioctl = kvm_vcpu_compat_ioctl,
1654 .mmap = kvm_vcpu_mmap,
1655 .llseek = noop_llseek,
1659 * Allocates an inode for the vcpu.
1661 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1663 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1667 * Creates some virtual cpus. Good luck creating more than one.
1669 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1672 struct kvm_vcpu *vcpu, *v;
1674 vcpu = kvm_arch_vcpu_create(kvm, id);
1676 return PTR_ERR(vcpu);
1678 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1680 r = kvm_arch_vcpu_setup(vcpu);
1684 mutex_lock(&kvm->lock);
1685 if (!kvm_vcpu_compatible(vcpu)) {
1687 goto unlock_vcpu_destroy;
1689 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1691 goto unlock_vcpu_destroy;
1694 kvm_for_each_vcpu(r, v, kvm)
1695 if (v->vcpu_id == id) {
1697 goto unlock_vcpu_destroy;
1700 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1702 /* Now it's all set up, let userspace reach it */
1704 r = create_vcpu_fd(vcpu);
1707 goto unlock_vcpu_destroy;
1710 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1712 atomic_inc(&kvm->online_vcpus);
1714 mutex_unlock(&kvm->lock);
1717 unlock_vcpu_destroy:
1718 mutex_unlock(&kvm->lock);
1720 kvm_arch_vcpu_destroy(vcpu);
1724 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1727 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1728 vcpu->sigset_active = 1;
1729 vcpu->sigset = *sigset;
1731 vcpu->sigset_active = 0;
1735 static long kvm_vcpu_ioctl(struct file *filp,
1736 unsigned int ioctl, unsigned long arg)
1738 struct kvm_vcpu *vcpu = filp->private_data;
1739 void __user *argp = (void __user *)arg;
1741 struct kvm_fpu *fpu = NULL;
1742 struct kvm_sregs *kvm_sregs = NULL;
1744 if (vcpu->kvm->mm != current->mm)
1747 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1749 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1750 * so vcpu_load() would break it.
1752 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1753 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1763 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1764 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1766 case KVM_GET_REGS: {
1767 struct kvm_regs *kvm_regs;
1770 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1773 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1777 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1784 case KVM_SET_REGS: {
1785 struct kvm_regs *kvm_regs;
1788 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1789 if (IS_ERR(kvm_regs)) {
1790 r = PTR_ERR(kvm_regs);
1793 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1801 case KVM_GET_SREGS: {
1802 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1806 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1810 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1815 case KVM_SET_SREGS: {
1816 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1817 if (IS_ERR(kvm_sregs)) {
1818 r = PTR_ERR(kvm_sregs);
1821 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1827 case KVM_GET_MP_STATE: {
1828 struct kvm_mp_state mp_state;
1830 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1834 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1839 case KVM_SET_MP_STATE: {
1840 struct kvm_mp_state mp_state;
1843 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1845 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1851 case KVM_TRANSLATE: {
1852 struct kvm_translation tr;
1855 if (copy_from_user(&tr, argp, sizeof tr))
1857 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1861 if (copy_to_user(argp, &tr, sizeof tr))
1866 case KVM_SET_GUEST_DEBUG: {
1867 struct kvm_guest_debug dbg;
1870 if (copy_from_user(&dbg, argp, sizeof dbg))
1872 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1878 case KVM_SET_SIGNAL_MASK: {
1879 struct kvm_signal_mask __user *sigmask_arg = argp;
1880 struct kvm_signal_mask kvm_sigmask;
1881 sigset_t sigset, *p;
1886 if (copy_from_user(&kvm_sigmask, argp,
1887 sizeof kvm_sigmask))
1890 if (kvm_sigmask.len != sizeof sigset)
1893 if (copy_from_user(&sigset, sigmask_arg->sigset,
1898 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1902 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1906 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1910 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1916 fpu = memdup_user(argp, sizeof(*fpu));
1921 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1928 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1937 #ifdef CONFIG_COMPAT
1938 static long kvm_vcpu_compat_ioctl(struct file *filp,
1939 unsigned int ioctl, unsigned long arg)
1941 struct kvm_vcpu *vcpu = filp->private_data;
1942 void __user *argp = compat_ptr(arg);
1945 if (vcpu->kvm->mm != current->mm)
1949 case KVM_SET_SIGNAL_MASK: {
1950 struct kvm_signal_mask __user *sigmask_arg = argp;
1951 struct kvm_signal_mask kvm_sigmask;
1952 compat_sigset_t csigset;
1957 if (copy_from_user(&kvm_sigmask, argp,
1958 sizeof kvm_sigmask))
1961 if (kvm_sigmask.len != sizeof csigset)
1964 if (copy_from_user(&csigset, sigmask_arg->sigset,
1968 sigset_from_compat(&sigset, &csigset);
1969 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1973 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1981 static long kvm_vm_ioctl(struct file *filp,
1982 unsigned int ioctl, unsigned long arg)
1984 struct kvm *kvm = filp->private_data;
1985 void __user *argp = (void __user *)arg;
1988 if (kvm->mm != current->mm)
1991 case KVM_CREATE_VCPU:
1992 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1996 case KVM_SET_USER_MEMORY_REGION: {
1997 struct kvm_userspace_memory_region kvm_userspace_mem;
2000 if (copy_from_user(&kvm_userspace_mem, argp,
2001 sizeof kvm_userspace_mem))
2004 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2009 case KVM_GET_DIRTY_LOG: {
2010 struct kvm_dirty_log log;
2013 if (copy_from_user(&log, argp, sizeof log))
2015 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2020 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2021 case KVM_REGISTER_COALESCED_MMIO: {
2022 struct kvm_coalesced_mmio_zone zone;
2024 if (copy_from_user(&zone, argp, sizeof zone))
2026 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2032 case KVM_UNREGISTER_COALESCED_MMIO: {
2033 struct kvm_coalesced_mmio_zone zone;
2035 if (copy_from_user(&zone, argp, sizeof zone))
2037 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2045 struct kvm_irqfd data;
2048 if (copy_from_user(&data, argp, sizeof data))
2050 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2053 case KVM_IOEVENTFD: {
2054 struct kvm_ioeventfd data;
2057 if (copy_from_user(&data, argp, sizeof data))
2059 r = kvm_ioeventfd(kvm, &data);
2062 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2063 case KVM_SET_BOOT_CPU_ID:
2065 mutex_lock(&kvm->lock);
2066 if (atomic_read(&kvm->online_vcpus) != 0)
2069 kvm->bsp_vcpu_id = arg;
2070 mutex_unlock(&kvm->lock);
2073 #ifdef CONFIG_HAVE_KVM_MSI
2074 case KVM_SIGNAL_MSI: {
2078 if (copy_from_user(&msi, argp, sizeof msi))
2080 r = kvm_send_userspace_msi(kvm, &msi);
2085 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2087 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2093 #ifdef CONFIG_COMPAT
2094 struct compat_kvm_dirty_log {
2098 compat_uptr_t dirty_bitmap; /* one bit per page */
2103 static long kvm_vm_compat_ioctl(struct file *filp,
2104 unsigned int ioctl, unsigned long arg)
2106 struct kvm *kvm = filp->private_data;
2109 if (kvm->mm != current->mm)
2112 case KVM_GET_DIRTY_LOG: {
2113 struct compat_kvm_dirty_log compat_log;
2114 struct kvm_dirty_log log;
2117 if (copy_from_user(&compat_log, (void __user *)arg,
2118 sizeof(compat_log)))
2120 log.slot = compat_log.slot;
2121 log.padding1 = compat_log.padding1;
2122 log.padding2 = compat_log.padding2;
2123 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2125 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2131 r = kvm_vm_ioctl(filp, ioctl, arg);
2139 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2141 struct page *page[1];
2144 gfn_t gfn = vmf->pgoff;
2145 struct kvm *kvm = vma->vm_file->private_data;
2147 addr = gfn_to_hva(kvm, gfn);
2148 if (kvm_is_error_hva(addr))
2149 return VM_FAULT_SIGBUS;
2151 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2153 if (unlikely(npages != 1))
2154 return VM_FAULT_SIGBUS;
2156 vmf->page = page[0];
2160 static const struct vm_operations_struct kvm_vm_vm_ops = {
2161 .fault = kvm_vm_fault,
2164 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2166 vma->vm_ops = &kvm_vm_vm_ops;
2170 static struct file_operations kvm_vm_fops = {
2171 .release = kvm_vm_release,
2172 .unlocked_ioctl = kvm_vm_ioctl,
2173 #ifdef CONFIG_COMPAT
2174 .compat_ioctl = kvm_vm_compat_ioctl,
2176 .mmap = kvm_vm_mmap,
2177 .llseek = noop_llseek,
2180 static int kvm_dev_ioctl_create_vm(unsigned long type)
2185 kvm = kvm_create_vm(type);
2187 return PTR_ERR(kvm);
2188 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2189 r = kvm_coalesced_mmio_init(kvm);
2195 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2202 static long kvm_dev_ioctl_check_extension_generic(long arg)
2205 case KVM_CAP_USER_MEMORY:
2206 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2207 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2208 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2209 case KVM_CAP_SET_BOOT_CPU_ID:
2211 case KVM_CAP_INTERNAL_ERROR_DATA:
2212 #ifdef CONFIG_HAVE_KVM_MSI
2213 case KVM_CAP_SIGNAL_MSI:
2216 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2217 case KVM_CAP_IRQ_ROUTING:
2218 return KVM_MAX_IRQ_ROUTES;
2223 return kvm_dev_ioctl_check_extension(arg);
2226 static long kvm_dev_ioctl(struct file *filp,
2227 unsigned int ioctl, unsigned long arg)
2232 case KVM_GET_API_VERSION:
2236 r = KVM_API_VERSION;
2239 r = kvm_dev_ioctl_create_vm(arg);
2241 case KVM_CHECK_EXTENSION:
2242 r = kvm_dev_ioctl_check_extension_generic(arg);
2244 case KVM_GET_VCPU_MMAP_SIZE:
2248 r = PAGE_SIZE; /* struct kvm_run */
2250 r += PAGE_SIZE; /* pio data page */
2252 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2253 r += PAGE_SIZE; /* coalesced mmio ring page */
2256 case KVM_TRACE_ENABLE:
2257 case KVM_TRACE_PAUSE:
2258 case KVM_TRACE_DISABLE:
2262 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2268 static struct file_operations kvm_chardev_ops = {
2269 .unlocked_ioctl = kvm_dev_ioctl,
2270 .compat_ioctl = kvm_dev_ioctl,
2271 .llseek = noop_llseek,
2274 static struct miscdevice kvm_dev = {
2280 static void hardware_enable_nolock(void *junk)
2282 int cpu = raw_smp_processor_id();
2285 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2288 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2290 r = kvm_arch_hardware_enable(NULL);
2293 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2294 atomic_inc(&hardware_enable_failed);
2295 printk(KERN_INFO "kvm: enabling virtualization on "
2296 "CPU%d failed\n", cpu);
2300 static void hardware_enable(void *junk)
2302 raw_spin_lock(&kvm_lock);
2303 hardware_enable_nolock(junk);
2304 raw_spin_unlock(&kvm_lock);
2307 static void hardware_disable_nolock(void *junk)
2309 int cpu = raw_smp_processor_id();
2311 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2313 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2314 kvm_arch_hardware_disable(NULL);
2317 static void hardware_disable(void *junk)
2319 raw_spin_lock(&kvm_lock);
2320 hardware_disable_nolock(junk);
2321 raw_spin_unlock(&kvm_lock);
2324 static void hardware_disable_all_nolock(void)
2326 BUG_ON(!kvm_usage_count);
2329 if (!kvm_usage_count)
2330 on_each_cpu(hardware_disable_nolock, NULL, 1);
2333 static void hardware_disable_all(void)
2335 raw_spin_lock(&kvm_lock);
2336 hardware_disable_all_nolock();
2337 raw_spin_unlock(&kvm_lock);
2340 static int hardware_enable_all(void)
2344 raw_spin_lock(&kvm_lock);
2347 if (kvm_usage_count == 1) {
2348 atomic_set(&hardware_enable_failed, 0);
2349 on_each_cpu(hardware_enable_nolock, NULL, 1);
2351 if (atomic_read(&hardware_enable_failed)) {
2352 hardware_disable_all_nolock();
2357 raw_spin_unlock(&kvm_lock);
2362 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2367 if (!kvm_usage_count)
2370 val &= ~CPU_TASKS_FROZEN;
2373 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2375 hardware_disable(NULL);
2378 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2380 hardware_enable(NULL);
2387 asmlinkage void kvm_spurious_fault(void)
2389 /* Fault while not rebooting. We want the trace. */
2392 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2394 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2398 * Some (well, at least mine) BIOSes hang on reboot if
2401 * And Intel TXT required VMX off for all cpu when system shutdown.
2403 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2404 kvm_rebooting = true;
2405 on_each_cpu(hardware_disable_nolock, NULL, 1);
2409 static struct notifier_block kvm_reboot_notifier = {
2410 .notifier_call = kvm_reboot,
2414 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2418 for (i = 0; i < bus->dev_count; i++) {
2419 struct kvm_io_device *pos = bus->range[i].dev;
2421 kvm_iodevice_destructor(pos);
2426 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2428 const struct kvm_io_range *r1 = p1;
2429 const struct kvm_io_range *r2 = p2;
2431 if (r1->addr < r2->addr)
2433 if (r1->addr + r1->len > r2->addr + r2->len)
2438 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2439 gpa_t addr, int len)
2441 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2447 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2448 kvm_io_bus_sort_cmp, NULL);
2453 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2454 gpa_t addr, int len)
2456 struct kvm_io_range *range, key;
2459 key = (struct kvm_io_range) {
2464 range = bsearch(&key, bus->range, bus->dev_count,
2465 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2469 off = range - bus->range;
2471 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2477 /* kvm_io_bus_write - called under kvm->slots_lock */
2478 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2479 int len, const void *val)
2482 struct kvm_io_bus *bus;
2483 struct kvm_io_range range;
2485 range = (struct kvm_io_range) {
2490 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2491 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2495 while (idx < bus->dev_count &&
2496 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2497 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2505 /* kvm_io_bus_read - called under kvm->slots_lock */
2506 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2510 struct kvm_io_bus *bus;
2511 struct kvm_io_range range;
2513 range = (struct kvm_io_range) {
2518 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2519 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2523 while (idx < bus->dev_count &&
2524 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2525 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2533 /* Caller must hold slots_lock. */
2534 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2535 int len, struct kvm_io_device *dev)
2537 struct kvm_io_bus *new_bus, *bus;
2539 bus = kvm->buses[bus_idx];
2540 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2543 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2544 sizeof(struct kvm_io_range)), GFP_KERNEL);
2547 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2548 sizeof(struct kvm_io_range)));
2549 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2550 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2551 synchronize_srcu_expedited(&kvm->srcu);
2557 /* Caller must hold slots_lock. */
2558 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2559 struct kvm_io_device *dev)
2562 struct kvm_io_bus *new_bus, *bus;
2564 bus = kvm->buses[bus_idx];
2566 for (i = 0; i < bus->dev_count; i++)
2567 if (bus->range[i].dev == dev) {
2575 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2576 sizeof(struct kvm_io_range)), GFP_KERNEL);
2580 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2581 new_bus->dev_count--;
2582 memcpy(new_bus->range + i, bus->range + i + 1,
2583 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2585 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2586 synchronize_srcu_expedited(&kvm->srcu);
2591 static struct notifier_block kvm_cpu_notifier = {
2592 .notifier_call = kvm_cpu_hotplug,
2595 static int vm_stat_get(void *_offset, u64 *val)
2597 unsigned offset = (long)_offset;
2601 raw_spin_lock(&kvm_lock);
2602 list_for_each_entry(kvm, &vm_list, vm_list)
2603 *val += *(u32 *)((void *)kvm + offset);
2604 raw_spin_unlock(&kvm_lock);
2608 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2610 static int vcpu_stat_get(void *_offset, u64 *val)
2612 unsigned offset = (long)_offset;
2614 struct kvm_vcpu *vcpu;
2618 raw_spin_lock(&kvm_lock);
2619 list_for_each_entry(kvm, &vm_list, vm_list)
2620 kvm_for_each_vcpu(i, vcpu, kvm)
2621 *val += *(u32 *)((void *)vcpu + offset);
2623 raw_spin_unlock(&kvm_lock);
2627 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2629 static const struct file_operations *stat_fops[] = {
2630 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2631 [KVM_STAT_VM] = &vm_stat_fops,
2634 static int kvm_init_debug(void)
2637 struct kvm_stats_debugfs_item *p;
2639 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2640 if (kvm_debugfs_dir == NULL)
2643 for (p = debugfs_entries; p->name; ++p) {
2644 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2645 (void *)(long)p->offset,
2646 stat_fops[p->kind]);
2647 if (p->dentry == NULL)
2654 debugfs_remove_recursive(kvm_debugfs_dir);
2659 static void kvm_exit_debug(void)
2661 struct kvm_stats_debugfs_item *p;
2663 for (p = debugfs_entries; p->name; ++p)
2664 debugfs_remove(p->dentry);
2665 debugfs_remove(kvm_debugfs_dir);
2668 static int kvm_suspend(void)
2670 if (kvm_usage_count)
2671 hardware_disable_nolock(NULL);
2675 static void kvm_resume(void)
2677 if (kvm_usage_count) {
2678 WARN_ON(raw_spin_is_locked(&kvm_lock));
2679 hardware_enable_nolock(NULL);
2683 static struct syscore_ops kvm_syscore_ops = {
2684 .suspend = kvm_suspend,
2685 .resume = kvm_resume,
2688 struct page *bad_page;
2692 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2694 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2697 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2699 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2701 kvm_arch_vcpu_load(vcpu, cpu);
2704 static void kvm_sched_out(struct preempt_notifier *pn,
2705 struct task_struct *next)
2707 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2709 kvm_arch_vcpu_put(vcpu);
2712 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2713 struct module *module)
2718 r = kvm_arch_init(opaque);
2722 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2724 if (bad_page == NULL) {
2729 bad_pfn = page_to_pfn(bad_page);
2731 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2733 if (hwpoison_page == NULL) {
2738 hwpoison_pfn = page_to_pfn(hwpoison_page);
2740 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2742 if (fault_page == NULL) {
2747 fault_pfn = page_to_pfn(fault_page);
2749 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2754 r = kvm_arch_hardware_setup();
2758 for_each_online_cpu(cpu) {
2759 smp_call_function_single(cpu,
2760 kvm_arch_check_processor_compat,
2766 r = register_cpu_notifier(&kvm_cpu_notifier);
2769 register_reboot_notifier(&kvm_reboot_notifier);
2771 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2773 vcpu_align = __alignof__(struct kvm_vcpu);
2774 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2776 if (!kvm_vcpu_cache) {
2781 r = kvm_async_pf_init();
2785 kvm_chardev_ops.owner = module;
2786 kvm_vm_fops.owner = module;
2787 kvm_vcpu_fops.owner = module;
2789 r = misc_register(&kvm_dev);
2791 printk(KERN_ERR "kvm: misc device register failed\n");
2795 register_syscore_ops(&kvm_syscore_ops);
2797 kvm_preempt_ops.sched_in = kvm_sched_in;
2798 kvm_preempt_ops.sched_out = kvm_sched_out;
2800 r = kvm_init_debug();
2802 printk(KERN_ERR "kvm: create debugfs files failed\n");
2809 unregister_syscore_ops(&kvm_syscore_ops);
2811 kvm_async_pf_deinit();
2813 kmem_cache_destroy(kvm_vcpu_cache);
2815 unregister_reboot_notifier(&kvm_reboot_notifier);
2816 unregister_cpu_notifier(&kvm_cpu_notifier);
2819 kvm_arch_hardware_unsetup();
2821 free_cpumask_var(cpus_hardware_enabled);
2824 __free_page(fault_page);
2826 __free_page(hwpoison_page);
2827 __free_page(bad_page);
2833 EXPORT_SYMBOL_GPL(kvm_init);
2838 misc_deregister(&kvm_dev);
2839 kmem_cache_destroy(kvm_vcpu_cache);
2840 kvm_async_pf_deinit();
2841 unregister_syscore_ops(&kvm_syscore_ops);
2842 unregister_reboot_notifier(&kvm_reboot_notifier);
2843 unregister_cpu_notifier(&kvm_cpu_notifier);
2844 on_each_cpu(hardware_disable_nolock, NULL, 1);
2845 kvm_arch_hardware_unsetup();
2847 free_cpumask_var(cpus_hardware_enabled);
2848 __free_page(hwpoison_page);
2849 __free_page(bad_page);
2851 EXPORT_SYMBOL_GPL(kvm_exit);