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);
520 * Avoid using vmalloc for a small buffer.
521 * Should not be used when the size is statically known.
523 void *kvm_kvzalloc(unsigned long size)
525 if (size > PAGE_SIZE)
526 return vzalloc(size);
528 return kzalloc(size, GFP_KERNEL);
531 void kvm_kvfree(const void *addr)
533 if (is_vmalloc_addr(addr))
539 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
541 if (!memslot->dirty_bitmap)
544 kvm_kvfree(memslot->dirty_bitmap);
545 memslot->dirty_bitmap = NULL;
549 * Free any memory in @free but not in @dont.
551 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
552 struct kvm_memory_slot *dont)
554 if (!dont || free->rmap != dont->rmap)
557 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
558 kvm_destroy_dirty_bitmap(free);
560 kvm_arch_free_memslot(free, dont);
566 void kvm_free_physmem(struct kvm *kvm)
568 struct kvm_memslots *slots = kvm->memslots;
569 struct kvm_memory_slot *memslot;
571 kvm_for_each_memslot(memslot, slots)
572 kvm_free_physmem_slot(memslot, NULL);
574 kfree(kvm->memslots);
577 static void kvm_destroy_vm(struct kvm *kvm)
580 struct mm_struct *mm = kvm->mm;
582 kvm_arch_sync_events(kvm);
583 raw_spin_lock(&kvm_lock);
584 list_del(&kvm->vm_list);
585 raw_spin_unlock(&kvm_lock);
586 kvm_free_irq_routing(kvm);
587 for (i = 0; i < KVM_NR_BUSES; i++)
588 kvm_io_bus_destroy(kvm->buses[i]);
589 kvm_coalesced_mmio_free(kvm);
590 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
591 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
593 kvm_arch_flush_shadow(kvm);
595 kvm_arch_destroy_vm(kvm);
596 kvm_free_physmem(kvm);
597 cleanup_srcu_struct(&kvm->srcu);
598 kvm_arch_free_vm(kvm);
599 hardware_disable_all();
603 void kvm_get_kvm(struct kvm *kvm)
605 atomic_inc(&kvm->users_count);
607 EXPORT_SYMBOL_GPL(kvm_get_kvm);
609 void kvm_put_kvm(struct kvm *kvm)
611 if (atomic_dec_and_test(&kvm->users_count))
614 EXPORT_SYMBOL_GPL(kvm_put_kvm);
617 static int kvm_vm_release(struct inode *inode, struct file *filp)
619 struct kvm *kvm = filp->private_data;
621 kvm_irqfd_release(kvm);
628 * Allocation size is twice as large as the actual dirty bitmap size.
629 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
631 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
634 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
636 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
637 if (!memslot->dirty_bitmap)
640 #endif /* !CONFIG_S390 */
644 static int cmp_memslot(const void *slot1, const void *slot2)
646 struct kvm_memory_slot *s1, *s2;
648 s1 = (struct kvm_memory_slot *)slot1;
649 s2 = (struct kvm_memory_slot *)slot2;
651 if (s1->npages < s2->npages)
653 if (s1->npages > s2->npages)
660 * Sort the memslots base on its size, so the larger slots
661 * will get better fit.
663 static void sort_memslots(struct kvm_memslots *slots)
667 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
668 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
670 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
671 slots->id_to_index[slots->memslots[i].id] = i;
674 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
678 struct kvm_memory_slot *old = id_to_memslot(slots, id);
679 unsigned long npages = old->npages;
682 if (new->npages != npages)
683 sort_memslots(slots);
690 * Allocate some memory and give it an address in the guest physical address
693 * Discontiguous memory is allowed, mostly for framebuffers.
695 * Must be called holding mmap_sem for write.
697 int __kvm_set_memory_region(struct kvm *kvm,
698 struct kvm_userspace_memory_region *mem,
703 unsigned long npages;
705 struct kvm_memory_slot *memslot;
706 struct kvm_memory_slot old, new;
707 struct kvm_memslots *slots, *old_memslots;
710 /* General sanity checks */
711 if (mem->memory_size & (PAGE_SIZE - 1))
713 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
715 /* We can read the guest memory with __xxx_user() later on. */
717 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
718 !access_ok(VERIFY_WRITE,
719 (void __user *)(unsigned long)mem->userspace_addr,
722 if (mem->slot >= KVM_MEM_SLOTS_NUM)
724 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
727 memslot = id_to_memslot(kvm->memslots, mem->slot);
728 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
729 npages = mem->memory_size >> PAGE_SHIFT;
732 if (npages > KVM_MEM_MAX_NR_PAGES)
736 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
738 new = old = *memslot;
741 new.base_gfn = base_gfn;
743 new.flags = mem->flags;
745 /* Disallow changing a memory slot's size. */
747 if (npages && old.npages && npages != old.npages)
750 /* Check for overlaps */
752 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
753 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
755 if (s == memslot || !s->npages)
757 if (!((base_gfn + npages <= s->base_gfn) ||
758 (base_gfn >= s->base_gfn + s->npages)))
762 /* Free page dirty bitmap if unneeded */
763 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
764 new.dirty_bitmap = NULL;
768 /* Allocate if a slot is being created */
769 if (npages && !old.npages) {
770 new.user_alloc = user_alloc;
771 new.userspace_addr = mem->userspace_addr;
773 new.rmap = vzalloc(npages * sizeof(*new.rmap));
776 #endif /* not defined CONFIG_S390 */
777 if (kvm_arch_create_memslot(&new, npages))
781 /* Allocate page dirty bitmap if needed */
782 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
783 if (kvm_create_dirty_bitmap(&new) < 0)
785 /* destroy any largepage mappings for dirty tracking */
789 struct kvm_memory_slot *slot;
792 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
796 slot = id_to_memslot(slots, mem->slot);
797 slot->flags |= KVM_MEMSLOT_INVALID;
799 update_memslots(slots, NULL);
801 old_memslots = kvm->memslots;
802 rcu_assign_pointer(kvm->memslots, slots);
803 synchronize_srcu_expedited(&kvm->srcu);
804 /* From this point no new shadow pages pointing to a deleted
805 * memslot will be created.
807 * validation of sp->gfn happens in:
808 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
809 * - kvm_is_visible_gfn (mmu_check_roots)
811 kvm_arch_flush_shadow(kvm);
815 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
819 /* map/unmap the pages in iommu page table */
821 r = kvm_iommu_map_pages(kvm, &new);
825 kvm_iommu_unmap_pages(kvm, &old);
828 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
833 /* actual memory is freed via old in kvm_free_physmem_slot below */
836 new.dirty_bitmap = NULL;
837 memset(&new.arch, 0, sizeof(new.arch));
840 update_memslots(slots, &new);
841 old_memslots = kvm->memslots;
842 rcu_assign_pointer(kvm->memslots, slots);
843 synchronize_srcu_expedited(&kvm->srcu);
845 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
848 * If the new memory slot is created, we need to clear all
851 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
852 kvm_arch_flush_shadow(kvm);
854 kvm_free_physmem_slot(&old, &new);
860 kvm_free_physmem_slot(&new, &old);
865 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
867 int kvm_set_memory_region(struct kvm *kvm,
868 struct kvm_userspace_memory_region *mem,
873 mutex_lock(&kvm->slots_lock);
874 r = __kvm_set_memory_region(kvm, mem, user_alloc);
875 mutex_unlock(&kvm->slots_lock);
878 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
880 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
882 kvm_userspace_memory_region *mem,
885 if (mem->slot >= KVM_MEMORY_SLOTS)
887 return kvm_set_memory_region(kvm, mem, user_alloc);
890 int kvm_get_dirty_log(struct kvm *kvm,
891 struct kvm_dirty_log *log, int *is_dirty)
893 struct kvm_memory_slot *memslot;
896 unsigned long any = 0;
899 if (log->slot >= KVM_MEMORY_SLOTS)
902 memslot = id_to_memslot(kvm->memslots, log->slot);
904 if (!memslot->dirty_bitmap)
907 n = kvm_dirty_bitmap_bytes(memslot);
909 for (i = 0; !any && i < n/sizeof(long); ++i)
910 any = memslot->dirty_bitmap[i];
913 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
924 bool kvm_largepages_enabled(void)
926 return largepages_enabled;
929 void kvm_disable_largepages(void)
931 largepages_enabled = false;
933 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
935 int is_error_page(struct page *page)
937 return page == bad_page || page == hwpoison_page || page == fault_page;
939 EXPORT_SYMBOL_GPL(is_error_page);
941 int is_error_pfn(pfn_t pfn)
943 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
945 EXPORT_SYMBOL_GPL(is_error_pfn);
947 int is_hwpoison_pfn(pfn_t pfn)
949 return pfn == hwpoison_pfn;
951 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
953 int is_fault_pfn(pfn_t pfn)
955 return pfn == fault_pfn;
957 EXPORT_SYMBOL_GPL(is_fault_pfn);
959 int is_noslot_pfn(pfn_t pfn)
961 return pfn == bad_pfn;
963 EXPORT_SYMBOL_GPL(is_noslot_pfn);
965 int is_invalid_pfn(pfn_t pfn)
967 return pfn == hwpoison_pfn || pfn == fault_pfn;
969 EXPORT_SYMBOL_GPL(is_invalid_pfn);
971 static inline unsigned long bad_hva(void)
976 int kvm_is_error_hva(unsigned long addr)
978 return addr == bad_hva();
980 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
982 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
984 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
986 EXPORT_SYMBOL_GPL(gfn_to_memslot);
988 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
990 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
992 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
993 memslot->flags & KVM_MEMSLOT_INVALID)
998 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1000 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1002 struct vm_area_struct *vma;
1003 unsigned long addr, size;
1007 addr = gfn_to_hva(kvm, gfn);
1008 if (kvm_is_error_hva(addr))
1011 down_read(¤t->mm->mmap_sem);
1012 vma = find_vma(current->mm, addr);
1016 size = vma_kernel_pagesize(vma);
1019 up_read(¤t->mm->mmap_sem);
1024 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1027 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1031 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1033 return gfn_to_hva_memslot(slot, gfn);
1036 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1038 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1040 EXPORT_SYMBOL_GPL(gfn_to_hva);
1042 static pfn_t get_fault_pfn(void)
1044 get_page(fault_page);
1048 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1049 unsigned long start, int write, struct page **page)
1051 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1054 flags |= FOLL_WRITE;
1056 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1059 static inline int check_user_page_hwpoison(unsigned long addr)
1061 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1063 rc = __get_user_pages(current, current->mm, addr, 1,
1064 flags, NULL, NULL, NULL);
1065 return rc == -EHWPOISON;
1068 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1069 bool *async, bool write_fault, bool *writable)
1071 struct page *page[1];
1075 /* we can do it either atomically or asynchronously, not both */
1076 BUG_ON(atomic && async);
1078 BUG_ON(!write_fault && !writable);
1083 if (atomic || async)
1084 npages = __get_user_pages_fast(addr, 1, 1, page);
1086 if (unlikely(npages != 1) && !atomic) {
1090 *writable = write_fault;
1093 down_read(¤t->mm->mmap_sem);
1094 npages = get_user_page_nowait(current, current->mm,
1095 addr, write_fault, page);
1096 up_read(¤t->mm->mmap_sem);
1098 npages = get_user_pages_fast(addr, 1, write_fault,
1101 /* map read fault as writable if possible */
1102 if (unlikely(!write_fault) && npages == 1) {
1103 struct page *wpage[1];
1105 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1115 if (unlikely(npages != 1)) {
1116 struct vm_area_struct *vma;
1119 return get_fault_pfn();
1121 down_read(¤t->mm->mmap_sem);
1122 if (npages == -EHWPOISON ||
1123 (!async && check_user_page_hwpoison(addr))) {
1124 up_read(¤t->mm->mmap_sem);
1125 get_page(hwpoison_page);
1126 return page_to_pfn(hwpoison_page);
1129 vma = find_vma_intersection(current->mm, addr, addr+1);
1132 pfn = get_fault_pfn();
1133 else if ((vma->vm_flags & VM_PFNMAP)) {
1134 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1136 BUG_ON(!kvm_is_mmio_pfn(pfn));
1138 if (async && (vma->vm_flags & VM_WRITE))
1140 pfn = get_fault_pfn();
1142 up_read(¤t->mm->mmap_sem);
1144 pfn = page_to_pfn(page[0]);
1149 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1151 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1153 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1155 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1156 bool write_fault, bool *writable)
1163 addr = gfn_to_hva(kvm, gfn);
1164 if (kvm_is_error_hva(addr)) {
1166 return page_to_pfn(bad_page);
1169 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1172 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1174 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1176 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1178 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1179 bool write_fault, bool *writable)
1181 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1183 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1185 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1187 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1189 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1191 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1194 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1196 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1198 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1199 struct kvm_memory_slot *slot, gfn_t gfn)
1201 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1202 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1205 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1211 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1212 if (kvm_is_error_hva(addr))
1215 if (entry < nr_pages)
1218 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1220 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1222 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1226 pfn = gfn_to_pfn(kvm, gfn);
1227 if (!kvm_is_mmio_pfn(pfn))
1228 return pfn_to_page(pfn);
1230 WARN_ON(kvm_is_mmio_pfn(pfn));
1236 EXPORT_SYMBOL_GPL(gfn_to_page);
1238 void kvm_release_page_clean(struct page *page)
1240 kvm_release_pfn_clean(page_to_pfn(page));
1242 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1244 void kvm_release_pfn_clean(pfn_t pfn)
1246 if (!kvm_is_mmio_pfn(pfn))
1247 put_page(pfn_to_page(pfn));
1249 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1251 void kvm_release_page_dirty(struct page *page)
1253 kvm_release_pfn_dirty(page_to_pfn(page));
1255 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1257 void kvm_release_pfn_dirty(pfn_t pfn)
1259 kvm_set_pfn_dirty(pfn);
1260 kvm_release_pfn_clean(pfn);
1262 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1264 void kvm_set_page_dirty(struct page *page)
1266 kvm_set_pfn_dirty(page_to_pfn(page));
1268 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1270 void kvm_set_pfn_dirty(pfn_t pfn)
1272 if (!kvm_is_mmio_pfn(pfn)) {
1273 struct page *page = pfn_to_page(pfn);
1274 if (!PageReserved(page))
1278 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1280 void kvm_set_pfn_accessed(pfn_t pfn)
1282 if (!kvm_is_mmio_pfn(pfn))
1283 mark_page_accessed(pfn_to_page(pfn));
1285 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1287 void kvm_get_pfn(pfn_t pfn)
1289 if (!kvm_is_mmio_pfn(pfn))
1290 get_page(pfn_to_page(pfn));
1292 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1294 static int next_segment(unsigned long len, int offset)
1296 if (len > PAGE_SIZE - offset)
1297 return PAGE_SIZE - offset;
1302 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1308 addr = gfn_to_hva(kvm, gfn);
1309 if (kvm_is_error_hva(addr))
1311 r = __copy_from_user(data, (void __user *)addr + offset, len);
1316 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1318 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1320 gfn_t gfn = gpa >> PAGE_SHIFT;
1322 int offset = offset_in_page(gpa);
1325 while ((seg = next_segment(len, offset)) != 0) {
1326 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1336 EXPORT_SYMBOL_GPL(kvm_read_guest);
1338 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1343 gfn_t gfn = gpa >> PAGE_SHIFT;
1344 int offset = offset_in_page(gpa);
1346 addr = gfn_to_hva(kvm, gfn);
1347 if (kvm_is_error_hva(addr))
1349 pagefault_disable();
1350 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1356 EXPORT_SYMBOL(kvm_read_guest_atomic);
1358 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1359 int offset, int len)
1364 addr = gfn_to_hva(kvm, gfn);
1365 if (kvm_is_error_hva(addr))
1367 r = __copy_to_user((void __user *)addr + offset, data, len);
1370 mark_page_dirty(kvm, gfn);
1373 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1375 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1378 gfn_t gfn = gpa >> PAGE_SHIFT;
1380 int offset = offset_in_page(gpa);
1383 while ((seg = next_segment(len, offset)) != 0) {
1384 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1395 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1398 struct kvm_memslots *slots = kvm_memslots(kvm);
1399 int offset = offset_in_page(gpa);
1400 gfn_t gfn = gpa >> PAGE_SHIFT;
1403 ghc->generation = slots->generation;
1404 ghc->memslot = gfn_to_memslot(kvm, gfn);
1405 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1406 if (!kvm_is_error_hva(ghc->hva))
1413 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1415 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1416 void *data, unsigned long len)
1418 struct kvm_memslots *slots = kvm_memslots(kvm);
1421 if (slots->generation != ghc->generation)
1422 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1424 if (kvm_is_error_hva(ghc->hva))
1427 r = __copy_to_user((void __user *)ghc->hva, data, len);
1430 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1434 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1436 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1437 void *data, unsigned long len)
1439 struct kvm_memslots *slots = kvm_memslots(kvm);
1442 if (slots->generation != ghc->generation)
1443 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1445 if (kvm_is_error_hva(ghc->hva))
1448 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1454 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1456 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1458 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1461 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1463 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1465 gfn_t gfn = gpa >> PAGE_SHIFT;
1467 int offset = offset_in_page(gpa);
1470 while ((seg = next_segment(len, offset)) != 0) {
1471 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1480 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1482 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1485 if (memslot && memslot->dirty_bitmap) {
1486 unsigned long rel_gfn = gfn - memslot->base_gfn;
1488 /* TODO: introduce set_bit_le() and use it */
1489 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1493 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1495 struct kvm_memory_slot *memslot;
1497 memslot = gfn_to_memslot(kvm, gfn);
1498 mark_page_dirty_in_slot(kvm, memslot, gfn);
1502 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1504 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1509 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1511 if (kvm_arch_vcpu_runnable(vcpu)) {
1512 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1515 if (kvm_cpu_has_pending_timer(vcpu))
1517 if (signal_pending(current))
1523 finish_wait(&vcpu->wq, &wait);
1528 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1530 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1533 int cpu = vcpu->cpu;
1534 wait_queue_head_t *wqp;
1536 wqp = kvm_arch_vcpu_wq(vcpu);
1537 if (waitqueue_active(wqp)) {
1538 wake_up_interruptible(wqp);
1539 ++vcpu->stat.halt_wakeup;
1543 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1544 if (kvm_arch_vcpu_should_kick(vcpu))
1545 smp_send_reschedule(cpu);
1548 #endif /* !CONFIG_S390 */
1550 void kvm_resched(struct kvm_vcpu *vcpu)
1552 if (!need_resched())
1556 EXPORT_SYMBOL_GPL(kvm_resched);
1558 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1561 struct task_struct *task = NULL;
1564 pid = rcu_dereference(target->pid);
1566 task = get_pid_task(target->pid, PIDTYPE_PID);
1570 if (task->flags & PF_VCPU) {
1571 put_task_struct(task);
1574 if (yield_to(task, 1)) {
1575 put_task_struct(task);
1578 put_task_struct(task);
1581 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1583 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1585 struct kvm *kvm = me->kvm;
1586 struct kvm_vcpu *vcpu;
1587 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1593 * We boost the priority of a VCPU that is runnable but not
1594 * currently running, because it got preempted by something
1595 * else and called schedule in __vcpu_run. Hopefully that
1596 * VCPU is holding the lock that we need and will release it.
1597 * We approximate round-robin by starting at the last boosted VCPU.
1599 for (pass = 0; pass < 2 && !yielded; pass++) {
1600 kvm_for_each_vcpu(i, vcpu, kvm) {
1601 if (!pass && i <= last_boosted_vcpu) {
1602 i = last_boosted_vcpu;
1604 } else if (pass && i > last_boosted_vcpu)
1608 if (waitqueue_active(&vcpu->wq))
1610 if (kvm_vcpu_yield_to(vcpu)) {
1611 kvm->last_boosted_vcpu = i;
1618 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1620 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1622 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1625 if (vmf->pgoff == 0)
1626 page = virt_to_page(vcpu->run);
1628 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1629 page = virt_to_page(vcpu->arch.pio_data);
1631 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1632 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1633 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1636 return kvm_arch_vcpu_fault(vcpu, vmf);
1642 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1643 .fault = kvm_vcpu_fault,
1646 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1648 vma->vm_ops = &kvm_vcpu_vm_ops;
1652 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1654 struct kvm_vcpu *vcpu = filp->private_data;
1656 kvm_put_kvm(vcpu->kvm);
1660 static struct file_operations kvm_vcpu_fops = {
1661 .release = kvm_vcpu_release,
1662 .unlocked_ioctl = kvm_vcpu_ioctl,
1663 #ifdef CONFIG_COMPAT
1664 .compat_ioctl = kvm_vcpu_compat_ioctl,
1666 .mmap = kvm_vcpu_mmap,
1667 .llseek = noop_llseek,
1671 * Allocates an inode for the vcpu.
1673 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1675 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1679 * Creates some virtual cpus. Good luck creating more than one.
1681 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1684 struct kvm_vcpu *vcpu, *v;
1686 vcpu = kvm_arch_vcpu_create(kvm, id);
1688 return PTR_ERR(vcpu);
1690 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1692 r = kvm_arch_vcpu_setup(vcpu);
1696 mutex_lock(&kvm->lock);
1697 if (!kvm_vcpu_compatible(vcpu)) {
1699 goto unlock_vcpu_destroy;
1701 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1703 goto unlock_vcpu_destroy;
1706 kvm_for_each_vcpu(r, v, kvm)
1707 if (v->vcpu_id == id) {
1709 goto unlock_vcpu_destroy;
1712 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1714 /* Now it's all set up, let userspace reach it */
1716 r = create_vcpu_fd(vcpu);
1719 goto unlock_vcpu_destroy;
1722 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1724 atomic_inc(&kvm->online_vcpus);
1726 mutex_unlock(&kvm->lock);
1729 unlock_vcpu_destroy:
1730 mutex_unlock(&kvm->lock);
1732 kvm_arch_vcpu_destroy(vcpu);
1736 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1739 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1740 vcpu->sigset_active = 1;
1741 vcpu->sigset = *sigset;
1743 vcpu->sigset_active = 0;
1747 static long kvm_vcpu_ioctl(struct file *filp,
1748 unsigned int ioctl, unsigned long arg)
1750 struct kvm_vcpu *vcpu = filp->private_data;
1751 void __user *argp = (void __user *)arg;
1753 struct kvm_fpu *fpu = NULL;
1754 struct kvm_sregs *kvm_sregs = NULL;
1756 if (vcpu->kvm->mm != current->mm)
1759 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1761 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1762 * so vcpu_load() would break it.
1764 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1765 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1775 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1776 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1778 case KVM_GET_REGS: {
1779 struct kvm_regs *kvm_regs;
1782 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1785 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1789 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1796 case KVM_SET_REGS: {
1797 struct kvm_regs *kvm_regs;
1800 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1801 if (IS_ERR(kvm_regs)) {
1802 r = PTR_ERR(kvm_regs);
1805 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1813 case KVM_GET_SREGS: {
1814 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1818 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1822 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1827 case KVM_SET_SREGS: {
1828 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1829 if (IS_ERR(kvm_sregs)) {
1830 r = PTR_ERR(kvm_sregs);
1833 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1839 case KVM_GET_MP_STATE: {
1840 struct kvm_mp_state mp_state;
1842 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1846 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1851 case KVM_SET_MP_STATE: {
1852 struct kvm_mp_state mp_state;
1855 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1857 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1863 case KVM_TRANSLATE: {
1864 struct kvm_translation tr;
1867 if (copy_from_user(&tr, argp, sizeof tr))
1869 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1873 if (copy_to_user(argp, &tr, sizeof tr))
1878 case KVM_SET_GUEST_DEBUG: {
1879 struct kvm_guest_debug dbg;
1882 if (copy_from_user(&dbg, argp, sizeof dbg))
1884 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1890 case KVM_SET_SIGNAL_MASK: {
1891 struct kvm_signal_mask __user *sigmask_arg = argp;
1892 struct kvm_signal_mask kvm_sigmask;
1893 sigset_t sigset, *p;
1898 if (copy_from_user(&kvm_sigmask, argp,
1899 sizeof kvm_sigmask))
1902 if (kvm_sigmask.len != sizeof sigset)
1905 if (copy_from_user(&sigset, sigmask_arg->sigset,
1910 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1914 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1918 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1922 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1928 fpu = memdup_user(argp, sizeof(*fpu));
1933 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1940 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1949 #ifdef CONFIG_COMPAT
1950 static long kvm_vcpu_compat_ioctl(struct file *filp,
1951 unsigned int ioctl, unsigned long arg)
1953 struct kvm_vcpu *vcpu = filp->private_data;
1954 void __user *argp = compat_ptr(arg);
1957 if (vcpu->kvm->mm != current->mm)
1961 case KVM_SET_SIGNAL_MASK: {
1962 struct kvm_signal_mask __user *sigmask_arg = argp;
1963 struct kvm_signal_mask kvm_sigmask;
1964 compat_sigset_t csigset;
1969 if (copy_from_user(&kvm_sigmask, argp,
1970 sizeof kvm_sigmask))
1973 if (kvm_sigmask.len != sizeof csigset)
1976 if (copy_from_user(&csigset, sigmask_arg->sigset,
1980 sigset_from_compat(&sigset, &csigset);
1981 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1985 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1993 static long kvm_vm_ioctl(struct file *filp,
1994 unsigned int ioctl, unsigned long arg)
1996 struct kvm *kvm = filp->private_data;
1997 void __user *argp = (void __user *)arg;
2000 if (kvm->mm != current->mm)
2003 case KVM_CREATE_VCPU:
2004 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2008 case KVM_SET_USER_MEMORY_REGION: {
2009 struct kvm_userspace_memory_region kvm_userspace_mem;
2012 if (copy_from_user(&kvm_userspace_mem, argp,
2013 sizeof kvm_userspace_mem))
2016 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2021 case KVM_GET_DIRTY_LOG: {
2022 struct kvm_dirty_log log;
2025 if (copy_from_user(&log, argp, sizeof log))
2027 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2032 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2033 case KVM_REGISTER_COALESCED_MMIO: {
2034 struct kvm_coalesced_mmio_zone zone;
2036 if (copy_from_user(&zone, argp, sizeof zone))
2038 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2044 case KVM_UNREGISTER_COALESCED_MMIO: {
2045 struct kvm_coalesced_mmio_zone zone;
2047 if (copy_from_user(&zone, argp, sizeof zone))
2049 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2057 struct kvm_irqfd data;
2060 if (copy_from_user(&data, argp, sizeof data))
2062 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2065 case KVM_IOEVENTFD: {
2066 struct kvm_ioeventfd data;
2069 if (copy_from_user(&data, argp, sizeof data))
2071 r = kvm_ioeventfd(kvm, &data);
2074 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2075 case KVM_SET_BOOT_CPU_ID:
2077 mutex_lock(&kvm->lock);
2078 if (atomic_read(&kvm->online_vcpus) != 0)
2081 kvm->bsp_vcpu_id = arg;
2082 mutex_unlock(&kvm->lock);
2085 #ifdef CONFIG_HAVE_KVM_MSI
2086 case KVM_SIGNAL_MSI: {
2090 if (copy_from_user(&msi, argp, sizeof msi))
2092 r = kvm_send_userspace_msi(kvm, &msi);
2097 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2099 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2105 #ifdef CONFIG_COMPAT
2106 struct compat_kvm_dirty_log {
2110 compat_uptr_t dirty_bitmap; /* one bit per page */
2115 static long kvm_vm_compat_ioctl(struct file *filp,
2116 unsigned int ioctl, unsigned long arg)
2118 struct kvm *kvm = filp->private_data;
2121 if (kvm->mm != current->mm)
2124 case KVM_GET_DIRTY_LOG: {
2125 struct compat_kvm_dirty_log compat_log;
2126 struct kvm_dirty_log log;
2129 if (copy_from_user(&compat_log, (void __user *)arg,
2130 sizeof(compat_log)))
2132 log.slot = compat_log.slot;
2133 log.padding1 = compat_log.padding1;
2134 log.padding2 = compat_log.padding2;
2135 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2137 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2143 r = kvm_vm_ioctl(filp, ioctl, arg);
2151 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2153 struct page *page[1];
2156 gfn_t gfn = vmf->pgoff;
2157 struct kvm *kvm = vma->vm_file->private_data;
2159 addr = gfn_to_hva(kvm, gfn);
2160 if (kvm_is_error_hva(addr))
2161 return VM_FAULT_SIGBUS;
2163 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2165 if (unlikely(npages != 1))
2166 return VM_FAULT_SIGBUS;
2168 vmf->page = page[0];
2172 static const struct vm_operations_struct kvm_vm_vm_ops = {
2173 .fault = kvm_vm_fault,
2176 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2178 vma->vm_ops = &kvm_vm_vm_ops;
2182 static struct file_operations kvm_vm_fops = {
2183 .release = kvm_vm_release,
2184 .unlocked_ioctl = kvm_vm_ioctl,
2185 #ifdef CONFIG_COMPAT
2186 .compat_ioctl = kvm_vm_compat_ioctl,
2188 .mmap = kvm_vm_mmap,
2189 .llseek = noop_llseek,
2192 static int kvm_dev_ioctl_create_vm(unsigned long type)
2197 kvm = kvm_create_vm(type);
2199 return PTR_ERR(kvm);
2200 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2201 r = kvm_coalesced_mmio_init(kvm);
2207 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2214 static long kvm_dev_ioctl_check_extension_generic(long arg)
2217 case KVM_CAP_USER_MEMORY:
2218 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2219 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2220 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2221 case KVM_CAP_SET_BOOT_CPU_ID:
2223 case KVM_CAP_INTERNAL_ERROR_DATA:
2224 #ifdef CONFIG_HAVE_KVM_MSI
2225 case KVM_CAP_SIGNAL_MSI:
2228 #ifdef KVM_CAP_IRQ_ROUTING
2229 case KVM_CAP_IRQ_ROUTING:
2230 return KVM_MAX_IRQ_ROUTES;
2235 return kvm_dev_ioctl_check_extension(arg);
2238 static long kvm_dev_ioctl(struct file *filp,
2239 unsigned int ioctl, unsigned long arg)
2244 case KVM_GET_API_VERSION:
2248 r = KVM_API_VERSION;
2251 r = kvm_dev_ioctl_create_vm(arg);
2253 case KVM_CHECK_EXTENSION:
2254 r = kvm_dev_ioctl_check_extension_generic(arg);
2256 case KVM_GET_VCPU_MMAP_SIZE:
2260 r = PAGE_SIZE; /* struct kvm_run */
2262 r += PAGE_SIZE; /* pio data page */
2264 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2265 r += PAGE_SIZE; /* coalesced mmio ring page */
2268 case KVM_TRACE_ENABLE:
2269 case KVM_TRACE_PAUSE:
2270 case KVM_TRACE_DISABLE:
2274 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2280 static struct file_operations kvm_chardev_ops = {
2281 .unlocked_ioctl = kvm_dev_ioctl,
2282 .compat_ioctl = kvm_dev_ioctl,
2283 .llseek = noop_llseek,
2286 static struct miscdevice kvm_dev = {
2292 static void hardware_enable_nolock(void *junk)
2294 int cpu = raw_smp_processor_id();
2297 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2300 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2302 r = kvm_arch_hardware_enable(NULL);
2305 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2306 atomic_inc(&hardware_enable_failed);
2307 printk(KERN_INFO "kvm: enabling virtualization on "
2308 "CPU%d failed\n", cpu);
2312 static void hardware_enable(void *junk)
2314 raw_spin_lock(&kvm_lock);
2315 hardware_enable_nolock(junk);
2316 raw_spin_unlock(&kvm_lock);
2319 static void hardware_disable_nolock(void *junk)
2321 int cpu = raw_smp_processor_id();
2323 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2325 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2326 kvm_arch_hardware_disable(NULL);
2329 static void hardware_disable(void *junk)
2331 raw_spin_lock(&kvm_lock);
2332 hardware_disable_nolock(junk);
2333 raw_spin_unlock(&kvm_lock);
2336 static void hardware_disable_all_nolock(void)
2338 BUG_ON(!kvm_usage_count);
2341 if (!kvm_usage_count)
2342 on_each_cpu(hardware_disable_nolock, NULL, 1);
2345 static void hardware_disable_all(void)
2347 raw_spin_lock(&kvm_lock);
2348 hardware_disable_all_nolock();
2349 raw_spin_unlock(&kvm_lock);
2352 static int hardware_enable_all(void)
2356 raw_spin_lock(&kvm_lock);
2359 if (kvm_usage_count == 1) {
2360 atomic_set(&hardware_enable_failed, 0);
2361 on_each_cpu(hardware_enable_nolock, NULL, 1);
2363 if (atomic_read(&hardware_enable_failed)) {
2364 hardware_disable_all_nolock();
2369 raw_spin_unlock(&kvm_lock);
2374 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2379 if (!kvm_usage_count)
2382 val &= ~CPU_TASKS_FROZEN;
2385 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2387 hardware_disable(NULL);
2390 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2392 hardware_enable(NULL);
2399 asmlinkage void kvm_spurious_fault(void)
2401 /* Fault while not rebooting. We want the trace. */
2404 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2406 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2410 * Some (well, at least mine) BIOSes hang on reboot if
2413 * And Intel TXT required VMX off for all cpu when system shutdown.
2415 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2416 kvm_rebooting = true;
2417 on_each_cpu(hardware_disable_nolock, NULL, 1);
2421 static struct notifier_block kvm_reboot_notifier = {
2422 .notifier_call = kvm_reboot,
2426 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2430 for (i = 0; i < bus->dev_count; i++) {
2431 struct kvm_io_device *pos = bus->range[i].dev;
2433 kvm_iodevice_destructor(pos);
2438 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2440 const struct kvm_io_range *r1 = p1;
2441 const struct kvm_io_range *r2 = p2;
2443 if (r1->addr < r2->addr)
2445 if (r1->addr + r1->len > r2->addr + r2->len)
2450 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2451 gpa_t addr, int len)
2453 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2459 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2460 kvm_io_bus_sort_cmp, NULL);
2465 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2466 gpa_t addr, int len)
2468 struct kvm_io_range *range, key;
2471 key = (struct kvm_io_range) {
2476 range = bsearch(&key, bus->range, bus->dev_count,
2477 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2481 off = range - bus->range;
2483 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2489 /* kvm_io_bus_write - called under kvm->slots_lock */
2490 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2491 int len, const void *val)
2494 struct kvm_io_bus *bus;
2495 struct kvm_io_range range;
2497 range = (struct kvm_io_range) {
2502 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2503 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2507 while (idx < bus->dev_count &&
2508 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2509 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2517 /* kvm_io_bus_read - called under kvm->slots_lock */
2518 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2522 struct kvm_io_bus *bus;
2523 struct kvm_io_range range;
2525 range = (struct kvm_io_range) {
2530 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2531 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2535 while (idx < bus->dev_count &&
2536 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2537 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2545 /* Caller must hold slots_lock. */
2546 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2547 int len, struct kvm_io_device *dev)
2549 struct kvm_io_bus *new_bus, *bus;
2551 bus = kvm->buses[bus_idx];
2552 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2555 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2556 sizeof(struct kvm_io_range)), GFP_KERNEL);
2559 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2560 sizeof(struct kvm_io_range)));
2561 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2562 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2563 synchronize_srcu_expedited(&kvm->srcu);
2569 /* Caller must hold slots_lock. */
2570 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2571 struct kvm_io_device *dev)
2574 struct kvm_io_bus *new_bus, *bus;
2576 bus = kvm->buses[bus_idx];
2578 for (i = 0; i < bus->dev_count; i++)
2579 if (bus->range[i].dev == dev) {
2587 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2588 sizeof(struct kvm_io_range)), GFP_KERNEL);
2592 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2593 new_bus->dev_count--;
2594 memcpy(new_bus->range + i, bus->range + i + 1,
2595 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2597 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2598 synchronize_srcu_expedited(&kvm->srcu);
2603 static struct notifier_block kvm_cpu_notifier = {
2604 .notifier_call = kvm_cpu_hotplug,
2607 static int vm_stat_get(void *_offset, u64 *val)
2609 unsigned offset = (long)_offset;
2613 raw_spin_lock(&kvm_lock);
2614 list_for_each_entry(kvm, &vm_list, vm_list)
2615 *val += *(u32 *)((void *)kvm + offset);
2616 raw_spin_unlock(&kvm_lock);
2620 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2622 static int vcpu_stat_get(void *_offset, u64 *val)
2624 unsigned offset = (long)_offset;
2626 struct kvm_vcpu *vcpu;
2630 raw_spin_lock(&kvm_lock);
2631 list_for_each_entry(kvm, &vm_list, vm_list)
2632 kvm_for_each_vcpu(i, vcpu, kvm)
2633 *val += *(u32 *)((void *)vcpu + offset);
2635 raw_spin_unlock(&kvm_lock);
2639 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2641 static const struct file_operations *stat_fops[] = {
2642 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2643 [KVM_STAT_VM] = &vm_stat_fops,
2646 static int kvm_init_debug(void)
2649 struct kvm_stats_debugfs_item *p;
2651 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2652 if (kvm_debugfs_dir == NULL)
2655 for (p = debugfs_entries; p->name; ++p) {
2656 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2657 (void *)(long)p->offset,
2658 stat_fops[p->kind]);
2659 if (p->dentry == NULL)
2666 debugfs_remove_recursive(kvm_debugfs_dir);
2671 static void kvm_exit_debug(void)
2673 struct kvm_stats_debugfs_item *p;
2675 for (p = debugfs_entries; p->name; ++p)
2676 debugfs_remove(p->dentry);
2677 debugfs_remove(kvm_debugfs_dir);
2680 static int kvm_suspend(void)
2682 if (kvm_usage_count)
2683 hardware_disable_nolock(NULL);
2687 static void kvm_resume(void)
2689 if (kvm_usage_count) {
2690 WARN_ON(raw_spin_is_locked(&kvm_lock));
2691 hardware_enable_nolock(NULL);
2695 static struct syscore_ops kvm_syscore_ops = {
2696 .suspend = kvm_suspend,
2697 .resume = kvm_resume,
2700 struct page *bad_page;
2704 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2706 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2709 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2711 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2713 kvm_arch_vcpu_load(vcpu, cpu);
2716 static void kvm_sched_out(struct preempt_notifier *pn,
2717 struct task_struct *next)
2719 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2721 kvm_arch_vcpu_put(vcpu);
2724 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2725 struct module *module)
2730 r = kvm_arch_init(opaque);
2734 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2736 if (bad_page == NULL) {
2741 bad_pfn = page_to_pfn(bad_page);
2743 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2745 if (hwpoison_page == NULL) {
2750 hwpoison_pfn = page_to_pfn(hwpoison_page);
2752 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2754 if (fault_page == NULL) {
2759 fault_pfn = page_to_pfn(fault_page);
2761 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2766 r = kvm_arch_hardware_setup();
2770 for_each_online_cpu(cpu) {
2771 smp_call_function_single(cpu,
2772 kvm_arch_check_processor_compat,
2778 r = register_cpu_notifier(&kvm_cpu_notifier);
2781 register_reboot_notifier(&kvm_reboot_notifier);
2783 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2785 vcpu_align = __alignof__(struct kvm_vcpu);
2786 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2788 if (!kvm_vcpu_cache) {
2793 r = kvm_async_pf_init();
2797 kvm_chardev_ops.owner = module;
2798 kvm_vm_fops.owner = module;
2799 kvm_vcpu_fops.owner = module;
2801 r = misc_register(&kvm_dev);
2803 printk(KERN_ERR "kvm: misc device register failed\n");
2807 register_syscore_ops(&kvm_syscore_ops);
2809 kvm_preempt_ops.sched_in = kvm_sched_in;
2810 kvm_preempt_ops.sched_out = kvm_sched_out;
2812 r = kvm_init_debug();
2814 printk(KERN_ERR "kvm: create debugfs files failed\n");
2821 unregister_syscore_ops(&kvm_syscore_ops);
2823 kvm_async_pf_deinit();
2825 kmem_cache_destroy(kvm_vcpu_cache);
2827 unregister_reboot_notifier(&kvm_reboot_notifier);
2828 unregister_cpu_notifier(&kvm_cpu_notifier);
2831 kvm_arch_hardware_unsetup();
2833 free_cpumask_var(cpus_hardware_enabled);
2836 __free_page(fault_page);
2838 __free_page(hwpoison_page);
2839 __free_page(bad_page);
2845 EXPORT_SYMBOL_GPL(kvm_init);
2850 misc_deregister(&kvm_dev);
2851 kmem_cache_destroy(kvm_vcpu_cache);
2852 kvm_async_pf_deinit();
2853 unregister_syscore_ops(&kvm_syscore_ops);
2854 unregister_reboot_notifier(&kvm_reboot_notifier);
2855 unregister_cpu_notifier(&kvm_cpu_notifier);
2856 on_each_cpu(hardware_disable_nolock, NULL, 1);
2857 kvm_arch_hardware_unsetup();
2859 free_cpumask_var(cpus_hardware_enabled);
2860 __free_page(hwpoison_page);
2861 __free_page(bad_page);
2863 EXPORT_SYMBOL_GPL(kvm_exit);