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 int 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);
292 kvm->mmu_notifier_seq++;
293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294 spin_unlock(&kvm->mmu_lock);
295 srcu_read_unlock(&kvm->srcu, idx);
297 /* we've to flush the tlb before the pages can be freed */
299 kvm_flush_remote_tlbs(kvm);
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 spin_unlock(&kvm->mmu_lock);
339 srcu_read_unlock(&kvm->srcu, idx);
341 /* we've to flush the tlb before the pages can be freed */
343 kvm_flush_remote_tlbs(kvm);
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++;
361 * The above sequence increase must be visible before the
362 * below count decrease but both values are read by the kvm
363 * page fault under mmu_lock spinlock so we don't need to add
364 * a smb_wmb() here in between the two.
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);
381 young = kvm_age_hva(kvm, address);
382 spin_unlock(&kvm->mmu_lock);
383 srcu_read_unlock(&kvm->srcu, idx);
386 kvm_flush_remote_tlbs(kvm);
391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392 struct mm_struct *mm,
393 unsigned long address)
395 struct kvm *kvm = mmu_notifier_to_kvm(mn);
398 idx = srcu_read_lock(&kvm->srcu);
399 spin_lock(&kvm->mmu_lock);
400 young = kvm_test_age_hva(kvm, address);
401 spin_unlock(&kvm->mmu_lock);
402 srcu_read_unlock(&kvm->srcu, idx);
407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408 struct mm_struct *mm)
410 struct kvm *kvm = mmu_notifier_to_kvm(mn);
413 idx = srcu_read_lock(&kvm->srcu);
414 kvm_arch_flush_shadow(kvm);
415 srcu_read_unlock(&kvm->srcu, idx);
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419 .invalidate_page = kvm_mmu_notifier_invalidate_page,
420 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
422 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
423 .test_young = kvm_mmu_notifier_test_young,
424 .change_pte = kvm_mmu_notifier_change_pte,
425 .release = kvm_mmu_notifier_release,
428 static int kvm_init_mmu_notifier(struct kvm *kvm)
430 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
434 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
436 static int kvm_init_mmu_notifier(struct kvm *kvm)
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
443 static void kvm_init_memslots_id(struct kvm *kvm)
446 struct kvm_memslots *slots = kvm->memslots;
448 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449 slots->id_to_index[i] = slots->memslots[i].id = i;
452 static struct kvm *kvm_create_vm(void)
455 struct kvm *kvm = kvm_arch_alloc_vm();
458 return ERR_PTR(-ENOMEM);
460 r = kvm_arch_init_vm(kvm);
462 goto out_err_nodisable;
464 r = hardware_enable_all();
466 goto out_err_nodisable;
468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
469 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
470 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
474 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
477 kvm_init_memslots_id(kvm);
478 if (init_srcu_struct(&kvm->srcu))
480 for (i = 0; i < KVM_NR_BUSES; i++) {
481 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
487 spin_lock_init(&kvm->mmu_lock);
488 kvm->mm = current->mm;
489 atomic_inc(&kvm->mm->mm_count);
490 kvm_eventfd_init(kvm);
491 mutex_init(&kvm->lock);
492 mutex_init(&kvm->irq_lock);
493 mutex_init(&kvm->slots_lock);
494 atomic_set(&kvm->users_count, 1);
496 r = kvm_init_mmu_notifier(kvm);
500 raw_spin_lock(&kvm_lock);
501 list_add(&kvm->vm_list, &vm_list);
502 raw_spin_unlock(&kvm_lock);
507 cleanup_srcu_struct(&kvm->srcu);
509 hardware_disable_all();
511 for (i = 0; i < KVM_NR_BUSES; i++)
512 kfree(kvm->buses[i]);
513 kfree(kvm->memslots);
514 kvm_arch_free_vm(kvm);
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
520 if (!memslot->dirty_bitmap)
523 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
524 vfree(memslot->dirty_bitmap_head);
526 kfree(memslot->dirty_bitmap_head);
528 memslot->dirty_bitmap = NULL;
529 memslot->dirty_bitmap_head = 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)
540 if (!dont || free->rmap != dont->rmap)
543 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
544 kvm_destroy_dirty_bitmap(free);
547 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
548 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
549 vfree(free->lpage_info[i]);
550 free->lpage_info[i] = NULL;
558 void kvm_free_physmem(struct kvm *kvm)
560 struct kvm_memslots *slots = kvm->memslots;
561 struct kvm_memory_slot *memslot;
563 kvm_for_each_memslot(memslot, slots)
564 kvm_free_physmem_slot(memslot, NULL);
566 kfree(kvm->memslots);
569 static void kvm_destroy_vm(struct kvm *kvm)
572 struct mm_struct *mm = kvm->mm;
574 kvm_arch_sync_events(kvm);
575 raw_spin_lock(&kvm_lock);
576 list_del(&kvm->vm_list);
577 raw_spin_unlock(&kvm_lock);
578 kvm_free_irq_routing(kvm);
579 for (i = 0; i < KVM_NR_BUSES; i++)
580 kvm_io_bus_destroy(kvm->buses[i]);
581 kvm_coalesced_mmio_free(kvm);
582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
583 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
585 kvm_arch_flush_shadow(kvm);
587 kvm_arch_destroy_vm(kvm);
588 kvm_free_physmem(kvm);
589 cleanup_srcu_struct(&kvm->srcu);
590 kvm_arch_free_vm(kvm);
591 hardware_disable_all();
595 void kvm_get_kvm(struct kvm *kvm)
597 atomic_inc(&kvm->users_count);
599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
601 void kvm_put_kvm(struct kvm *kvm)
603 if (atomic_dec_and_test(&kvm->users_count))
606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
609 static int kvm_vm_release(struct inode *inode, struct file *filp)
611 struct kvm *kvm = filp->private_data;
613 kvm_irqfd_release(kvm);
621 * Allocation size is twice as large as the actual dirty bitmap size.
622 * This makes it possible to do double buffering: see x86's
623 * kvm_vm_ioctl_get_dirty_log().
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
627 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
629 if (dirty_bytes > PAGE_SIZE)
630 memslot->dirty_bitmap = vzalloc(dirty_bytes);
632 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
634 if (!memslot->dirty_bitmap)
637 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
638 memslot->nr_dirty_pages = 0;
641 #endif /* !CONFIG_S390 */
643 static struct kvm_memory_slot *
644 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
646 struct kvm_memory_slot *memslot;
648 kvm_for_each_memslot(memslot, slots)
649 if (gfn >= memslot->base_gfn &&
650 gfn < memslot->base_gfn + memslot->npages)
656 static int cmp_memslot(const void *slot1, const void *slot2)
658 struct kvm_memory_slot *s1, *s2;
660 s1 = (struct kvm_memory_slot *)slot1;
661 s2 = (struct kvm_memory_slot *)slot2;
663 if (s1->npages < s2->npages)
665 if (s1->npages > s2->npages)
672 * Sort the memslots base on its size, so the larger slots
673 * will get better fit.
675 static void sort_memslots(struct kvm_memslots *slots)
679 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
680 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
682 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
683 slots->id_to_index[slots->memslots[i].id] = i;
686 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
690 struct kvm_memory_slot *old = id_to_memslot(slots, id);
691 unsigned long npages = old->npages;
694 if (new->npages != npages)
695 sort_memslots(slots);
702 * Allocate some memory and give it an address in the guest physical address
705 * Discontiguous memory is allowed, mostly for framebuffers.
707 * Must be called holding mmap_sem for write.
709 int __kvm_set_memory_region(struct kvm *kvm,
710 struct kvm_userspace_memory_region *mem,
715 unsigned long npages;
717 struct kvm_memory_slot *memslot;
718 struct kvm_memory_slot old, new;
719 struct kvm_memslots *slots, *old_memslots;
722 /* General sanity checks */
723 if (mem->memory_size & (PAGE_SIZE - 1))
725 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
727 /* We can read the guest memory with __xxx_user() later on. */
729 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
730 !access_ok(VERIFY_WRITE,
731 (void __user *)(unsigned long)mem->userspace_addr,
734 if (mem->slot >= KVM_MEM_SLOTS_NUM)
736 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
739 memslot = id_to_memslot(kvm->memslots, mem->slot);
740 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
741 npages = mem->memory_size >> PAGE_SHIFT;
744 if (npages > KVM_MEM_MAX_NR_PAGES)
748 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
750 new = old = *memslot;
753 new.base_gfn = base_gfn;
755 new.flags = mem->flags;
757 /* Disallow changing a memory slot's size. */
759 if (npages && old.npages && npages != old.npages)
762 /* Check for overlaps */
764 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
765 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
767 if (s == memslot || !s->npages)
769 if (!((base_gfn + npages <= s->base_gfn) ||
770 (base_gfn >= s->base_gfn + s->npages)))
774 /* Free page dirty bitmap if unneeded */
775 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
776 new.dirty_bitmap = NULL;
780 /* Allocate if a slot is being created */
782 if (npages && !new.rmap) {
783 new.rmap = vzalloc(npages * sizeof(*new.rmap));
788 new.user_alloc = user_alloc;
789 new.userspace_addr = mem->userspace_addr;
794 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
800 /* Avoid unused variable warning if no large pages */
803 if (new.lpage_info[i])
806 lpages = 1 + ((base_gfn + npages - 1)
807 >> KVM_HPAGE_GFN_SHIFT(level));
808 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
810 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
812 if (!new.lpage_info[i])
815 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
816 new.lpage_info[i][0].write_count = 1;
817 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
818 new.lpage_info[i][lpages - 1].write_count = 1;
819 ugfn = new.userspace_addr >> PAGE_SHIFT;
821 * If the gfn and userspace address are not aligned wrt each
822 * other, or if explicitly asked to, disable large page
823 * support for this slot
825 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
827 for (j = 0; j < lpages; ++j)
828 new.lpage_info[i][j].write_count = 1;
833 /* Allocate page dirty bitmap if needed */
834 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
835 if (kvm_create_dirty_bitmap(&new) < 0)
837 /* destroy any largepage mappings for dirty tracking */
839 #else /* not defined CONFIG_S390 */
840 new.user_alloc = user_alloc;
842 new.userspace_addr = mem->userspace_addr;
843 #endif /* not defined CONFIG_S390 */
846 struct kvm_memory_slot *slot;
849 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
853 slot = id_to_memslot(slots, mem->slot);
854 slot->flags |= KVM_MEMSLOT_INVALID;
856 update_memslots(slots, NULL);
858 old_memslots = kvm->memslots;
859 rcu_assign_pointer(kvm->memslots, slots);
860 synchronize_srcu_expedited(&kvm->srcu);
861 /* From this point no new shadow pages pointing to a deleted
862 * memslot will be created.
864 * validation of sp->gfn happens in:
865 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
866 * - kvm_is_visible_gfn (mmu_check_roots)
868 kvm_arch_flush_shadow(kvm);
872 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
876 /* map the pages in iommu page table */
878 r = kvm_iommu_map_pages(kvm, &new);
884 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
889 /* actual memory is freed via old in kvm_free_physmem_slot below */
892 new.dirty_bitmap = NULL;
893 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
894 new.lpage_info[i] = NULL;
897 update_memslots(slots, &new);
898 old_memslots = kvm->memslots;
899 rcu_assign_pointer(kvm->memslots, slots);
900 synchronize_srcu_expedited(&kvm->srcu);
902 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
905 * If the new memory slot is created, we need to clear all
908 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
909 kvm_arch_flush_shadow(kvm);
911 kvm_free_physmem_slot(&old, &new);
917 kvm_free_physmem_slot(&new, &old);
922 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
924 int kvm_set_memory_region(struct kvm *kvm,
925 struct kvm_userspace_memory_region *mem,
930 mutex_lock(&kvm->slots_lock);
931 r = __kvm_set_memory_region(kvm, mem, user_alloc);
932 mutex_unlock(&kvm->slots_lock);
935 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
937 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
939 kvm_userspace_memory_region *mem,
942 if (mem->slot >= KVM_MEMORY_SLOTS)
944 return kvm_set_memory_region(kvm, mem, user_alloc);
947 int kvm_get_dirty_log(struct kvm *kvm,
948 struct kvm_dirty_log *log, int *is_dirty)
950 struct kvm_memory_slot *memslot;
953 unsigned long any = 0;
956 if (log->slot >= KVM_MEMORY_SLOTS)
959 memslot = id_to_memslot(kvm->memslots, log->slot);
961 if (!memslot->dirty_bitmap)
964 n = kvm_dirty_bitmap_bytes(memslot);
966 for (i = 0; !any && i < n/sizeof(long); ++i)
967 any = memslot->dirty_bitmap[i];
970 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
981 void kvm_disable_largepages(void)
983 largepages_enabled = false;
985 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
987 int is_error_page(struct page *page)
989 return page == bad_page || page == hwpoison_page || page == fault_page;
991 EXPORT_SYMBOL_GPL(is_error_page);
993 int is_error_pfn(pfn_t pfn)
995 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
997 EXPORT_SYMBOL_GPL(is_error_pfn);
999 int is_hwpoison_pfn(pfn_t pfn)
1001 return pfn == hwpoison_pfn;
1003 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
1005 int is_fault_pfn(pfn_t pfn)
1007 return pfn == fault_pfn;
1009 EXPORT_SYMBOL_GPL(is_fault_pfn);
1011 int is_noslot_pfn(pfn_t pfn)
1013 return pfn == bad_pfn;
1015 EXPORT_SYMBOL_GPL(is_noslot_pfn);
1017 int is_invalid_pfn(pfn_t pfn)
1019 return pfn == hwpoison_pfn || pfn == fault_pfn;
1021 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1023 static inline unsigned long bad_hva(void)
1028 int kvm_is_error_hva(unsigned long addr)
1030 return addr == bad_hva();
1032 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1034 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
1037 return search_memslots(slots, gfn);
1040 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1042 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1044 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1046 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1048 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1050 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1051 memslot->flags & KVM_MEMSLOT_INVALID)
1056 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1058 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1060 struct vm_area_struct *vma;
1061 unsigned long addr, size;
1065 addr = gfn_to_hva(kvm, gfn);
1066 if (kvm_is_error_hva(addr))
1069 down_read(¤t->mm->mmap_sem);
1070 vma = find_vma(current->mm, addr);
1074 size = vma_kernel_pagesize(vma);
1077 up_read(¤t->mm->mmap_sem);
1082 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1085 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1089 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1091 return gfn_to_hva_memslot(slot, gfn);
1094 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1096 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1098 EXPORT_SYMBOL_GPL(gfn_to_hva);
1100 static pfn_t get_fault_pfn(void)
1102 get_page(fault_page);
1106 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1107 unsigned long start, int write, struct page **page)
1109 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1112 flags |= FOLL_WRITE;
1114 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1117 static inline int check_user_page_hwpoison(unsigned long addr)
1119 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1121 rc = __get_user_pages(current, current->mm, addr, 1,
1122 flags, NULL, NULL, NULL);
1123 return rc == -EHWPOISON;
1126 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1127 bool *async, bool write_fault, bool *writable)
1129 struct page *page[1];
1133 /* we can do it either atomically or asynchronously, not both */
1134 BUG_ON(atomic && async);
1136 BUG_ON(!write_fault && !writable);
1141 if (atomic || async)
1142 npages = __get_user_pages_fast(addr, 1, 1, page);
1144 if (unlikely(npages != 1) && !atomic) {
1148 *writable = write_fault;
1151 down_read(¤t->mm->mmap_sem);
1152 npages = get_user_page_nowait(current, current->mm,
1153 addr, write_fault, page);
1154 up_read(¤t->mm->mmap_sem);
1156 npages = get_user_pages_fast(addr, 1, write_fault,
1159 /* map read fault as writable if possible */
1160 if (unlikely(!write_fault) && npages == 1) {
1161 struct page *wpage[1];
1163 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1173 if (unlikely(npages != 1)) {
1174 struct vm_area_struct *vma;
1177 return get_fault_pfn();
1179 down_read(¤t->mm->mmap_sem);
1180 if (npages == -EHWPOISON ||
1181 (!async && check_user_page_hwpoison(addr))) {
1182 up_read(¤t->mm->mmap_sem);
1183 get_page(hwpoison_page);
1184 return page_to_pfn(hwpoison_page);
1187 vma = find_vma_intersection(current->mm, addr, addr+1);
1190 pfn = get_fault_pfn();
1191 else if ((vma->vm_flags & VM_PFNMAP)) {
1192 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1194 BUG_ON(!kvm_is_mmio_pfn(pfn));
1196 if (async && (vma->vm_flags & VM_WRITE))
1198 pfn = get_fault_pfn();
1200 up_read(¤t->mm->mmap_sem);
1202 pfn = page_to_pfn(page[0]);
1207 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1209 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1211 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1213 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1214 bool write_fault, bool *writable)
1221 addr = gfn_to_hva(kvm, gfn);
1222 if (kvm_is_error_hva(addr)) {
1224 return page_to_pfn(bad_page);
1227 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1230 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1232 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1234 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1236 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1237 bool write_fault, bool *writable)
1239 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1241 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1243 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1245 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1247 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1249 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1252 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1254 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1256 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1257 struct kvm_memory_slot *slot, gfn_t gfn)
1259 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1260 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1263 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1269 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1270 if (kvm_is_error_hva(addr))
1273 if (entry < nr_pages)
1276 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1278 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1280 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1284 pfn = gfn_to_pfn(kvm, gfn);
1285 if (!kvm_is_mmio_pfn(pfn))
1286 return pfn_to_page(pfn);
1288 WARN_ON(kvm_is_mmio_pfn(pfn));
1294 EXPORT_SYMBOL_GPL(gfn_to_page);
1296 void kvm_release_page_clean(struct page *page)
1298 kvm_release_pfn_clean(page_to_pfn(page));
1300 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1302 void kvm_release_pfn_clean(pfn_t pfn)
1304 if (!kvm_is_mmio_pfn(pfn))
1305 put_page(pfn_to_page(pfn));
1307 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1309 void kvm_release_page_dirty(struct page *page)
1311 kvm_release_pfn_dirty(page_to_pfn(page));
1313 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1315 void kvm_release_pfn_dirty(pfn_t pfn)
1317 kvm_set_pfn_dirty(pfn);
1318 kvm_release_pfn_clean(pfn);
1320 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1322 void kvm_set_page_dirty(struct page *page)
1324 kvm_set_pfn_dirty(page_to_pfn(page));
1326 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1328 void kvm_set_pfn_dirty(pfn_t pfn)
1330 if (!kvm_is_mmio_pfn(pfn)) {
1331 struct page *page = pfn_to_page(pfn);
1332 if (!PageReserved(page))
1336 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1338 void kvm_set_pfn_accessed(pfn_t pfn)
1340 if (!kvm_is_mmio_pfn(pfn))
1341 mark_page_accessed(pfn_to_page(pfn));
1343 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1345 void kvm_get_pfn(pfn_t pfn)
1347 if (!kvm_is_mmio_pfn(pfn))
1348 get_page(pfn_to_page(pfn));
1350 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1352 static int next_segment(unsigned long len, int offset)
1354 if (len > PAGE_SIZE - offset)
1355 return PAGE_SIZE - offset;
1360 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1366 addr = gfn_to_hva(kvm, gfn);
1367 if (kvm_is_error_hva(addr))
1369 r = __copy_from_user(data, (void __user *)addr + offset, len);
1374 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1376 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
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_read_guest_page(kvm, gfn, data, offset, seg);
1394 EXPORT_SYMBOL_GPL(kvm_read_guest);
1396 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1401 gfn_t gfn = gpa >> PAGE_SHIFT;
1402 int offset = offset_in_page(gpa);
1404 addr = gfn_to_hva(kvm, gfn);
1405 if (kvm_is_error_hva(addr))
1407 pagefault_disable();
1408 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1414 EXPORT_SYMBOL(kvm_read_guest_atomic);
1416 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1417 int offset, int len)
1422 addr = gfn_to_hva(kvm, gfn);
1423 if (kvm_is_error_hva(addr))
1425 r = __copy_to_user((void __user *)addr + offset, data, len);
1428 mark_page_dirty(kvm, gfn);
1431 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1433 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1436 gfn_t gfn = gpa >> PAGE_SHIFT;
1438 int offset = offset_in_page(gpa);
1441 while ((seg = next_segment(len, offset)) != 0) {
1442 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1453 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1456 struct kvm_memslots *slots = kvm_memslots(kvm);
1457 int offset = offset_in_page(gpa);
1458 gfn_t gfn = gpa >> PAGE_SHIFT;
1461 ghc->generation = slots->generation;
1462 ghc->memslot = __gfn_to_memslot(slots, gfn);
1463 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1464 if (!kvm_is_error_hva(ghc->hva))
1471 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1473 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1474 void *data, unsigned long len)
1476 struct kvm_memslots *slots = kvm_memslots(kvm);
1479 if (slots->generation != ghc->generation)
1480 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1482 if (kvm_is_error_hva(ghc->hva))
1485 r = __copy_to_user((void __user *)ghc->hva, data, len);
1488 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1492 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1494 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1495 void *data, unsigned long len)
1497 struct kvm_memslots *slots = kvm_memslots(kvm);
1500 if (slots->generation != ghc->generation)
1501 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1503 if (kvm_is_error_hva(ghc->hva))
1506 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1512 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1514 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1516 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1519 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1521 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1523 gfn_t gfn = gpa >> PAGE_SHIFT;
1525 int offset = offset_in_page(gpa);
1528 while ((seg = next_segment(len, offset)) != 0) {
1529 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1538 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1540 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1543 if (memslot && memslot->dirty_bitmap) {
1544 unsigned long rel_gfn = gfn - memslot->base_gfn;
1546 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1547 memslot->nr_dirty_pages++;
1551 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1553 struct kvm_memory_slot *memslot;
1555 memslot = gfn_to_memslot(kvm, gfn);
1556 mark_page_dirty_in_slot(kvm, memslot, gfn);
1560 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1562 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1567 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1569 if (kvm_arch_vcpu_runnable(vcpu)) {
1570 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1573 if (kvm_cpu_has_pending_timer(vcpu))
1575 if (signal_pending(current))
1581 finish_wait(&vcpu->wq, &wait);
1584 void kvm_resched(struct kvm_vcpu *vcpu)
1586 if (!need_resched())
1590 EXPORT_SYMBOL_GPL(kvm_resched);
1592 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1594 struct kvm *kvm = me->kvm;
1595 struct kvm_vcpu *vcpu;
1596 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1602 * We boost the priority of a VCPU that is runnable but not
1603 * currently running, because it got preempted by something
1604 * else and called schedule in __vcpu_run. Hopefully that
1605 * VCPU is holding the lock that we need and will release it.
1606 * We approximate round-robin by starting at the last boosted VCPU.
1608 for (pass = 0; pass < 2 && !yielded; pass++) {
1609 kvm_for_each_vcpu(i, vcpu, kvm) {
1610 struct task_struct *task = NULL;
1612 if (!pass && i < last_boosted_vcpu) {
1613 i = last_boosted_vcpu;
1615 } else if (pass && i > last_boosted_vcpu)
1619 if (waitqueue_active(&vcpu->wq))
1622 pid = rcu_dereference(vcpu->pid);
1624 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1628 if (task->flags & PF_VCPU) {
1629 put_task_struct(task);
1632 if (yield_to(task, 1)) {
1633 put_task_struct(task);
1634 kvm->last_boosted_vcpu = i;
1638 put_task_struct(task);
1642 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1644 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1646 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1649 if (vmf->pgoff == 0)
1650 page = virt_to_page(vcpu->run);
1652 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1653 page = virt_to_page(vcpu->arch.pio_data);
1655 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1656 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1657 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1660 return VM_FAULT_SIGBUS;
1666 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1667 .fault = kvm_vcpu_fault,
1670 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1672 vma->vm_ops = &kvm_vcpu_vm_ops;
1676 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1678 struct kvm_vcpu *vcpu = filp->private_data;
1680 kvm_put_kvm(vcpu->kvm);
1684 static struct file_operations kvm_vcpu_fops = {
1685 .release = kvm_vcpu_release,
1686 .unlocked_ioctl = kvm_vcpu_ioctl,
1687 #ifdef CONFIG_COMPAT
1688 .compat_ioctl = kvm_vcpu_compat_ioctl,
1690 .mmap = kvm_vcpu_mmap,
1691 .llseek = noop_llseek,
1695 * Allocates an inode for the vcpu.
1697 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1699 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1703 * Creates some virtual cpus. Good luck creating more than one.
1705 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1708 struct kvm_vcpu *vcpu, *v;
1710 vcpu = kvm_arch_vcpu_create(kvm, id);
1712 return PTR_ERR(vcpu);
1714 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1716 r = kvm_arch_vcpu_setup(vcpu);
1720 mutex_lock(&kvm->lock);
1721 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1723 goto unlock_vcpu_destroy;
1726 kvm_for_each_vcpu(r, v, kvm)
1727 if (v->vcpu_id == id) {
1729 goto unlock_vcpu_destroy;
1732 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1734 /* Now it's all set up, let userspace reach it */
1736 r = create_vcpu_fd(vcpu);
1739 goto unlock_vcpu_destroy;
1742 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1744 atomic_inc(&kvm->online_vcpus);
1746 mutex_unlock(&kvm->lock);
1749 unlock_vcpu_destroy:
1750 mutex_unlock(&kvm->lock);
1752 kvm_arch_vcpu_destroy(vcpu);
1756 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1759 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1760 vcpu->sigset_active = 1;
1761 vcpu->sigset = *sigset;
1763 vcpu->sigset_active = 0;
1767 static long kvm_vcpu_ioctl(struct file *filp,
1768 unsigned int ioctl, unsigned long arg)
1770 struct kvm_vcpu *vcpu = filp->private_data;
1771 void __user *argp = (void __user *)arg;
1773 struct kvm_fpu *fpu = NULL;
1774 struct kvm_sregs *kvm_sregs = NULL;
1776 if (vcpu->kvm->mm != current->mm)
1779 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1781 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1782 * so vcpu_load() would break it.
1784 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1785 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1795 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1796 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1798 case KVM_GET_REGS: {
1799 struct kvm_regs *kvm_regs;
1802 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1805 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1809 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1816 case KVM_SET_REGS: {
1817 struct kvm_regs *kvm_regs;
1820 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1821 if (IS_ERR(kvm_regs)) {
1822 r = PTR_ERR(kvm_regs);
1825 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1833 case KVM_GET_SREGS: {
1834 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1838 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1842 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1847 case KVM_SET_SREGS: {
1848 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1849 if (IS_ERR(kvm_sregs)) {
1850 r = PTR_ERR(kvm_sregs);
1853 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1859 case KVM_GET_MP_STATE: {
1860 struct kvm_mp_state mp_state;
1862 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1866 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1871 case KVM_SET_MP_STATE: {
1872 struct kvm_mp_state mp_state;
1875 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1877 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1883 case KVM_TRANSLATE: {
1884 struct kvm_translation tr;
1887 if (copy_from_user(&tr, argp, sizeof tr))
1889 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1893 if (copy_to_user(argp, &tr, sizeof tr))
1898 case KVM_SET_GUEST_DEBUG: {
1899 struct kvm_guest_debug dbg;
1902 if (copy_from_user(&dbg, argp, sizeof dbg))
1904 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1910 case KVM_SET_SIGNAL_MASK: {
1911 struct kvm_signal_mask __user *sigmask_arg = argp;
1912 struct kvm_signal_mask kvm_sigmask;
1913 sigset_t sigset, *p;
1918 if (copy_from_user(&kvm_sigmask, argp,
1919 sizeof kvm_sigmask))
1922 if (kvm_sigmask.len != sizeof sigset)
1925 if (copy_from_user(&sigset, sigmask_arg->sigset,
1930 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1934 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1938 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1942 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1948 fpu = memdup_user(argp, sizeof(*fpu));
1953 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1960 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1969 #ifdef CONFIG_COMPAT
1970 static long kvm_vcpu_compat_ioctl(struct file *filp,
1971 unsigned int ioctl, unsigned long arg)
1973 struct kvm_vcpu *vcpu = filp->private_data;
1974 void __user *argp = compat_ptr(arg);
1977 if (vcpu->kvm->mm != current->mm)
1981 case KVM_SET_SIGNAL_MASK: {
1982 struct kvm_signal_mask __user *sigmask_arg = argp;
1983 struct kvm_signal_mask kvm_sigmask;
1984 compat_sigset_t csigset;
1989 if (copy_from_user(&kvm_sigmask, argp,
1990 sizeof kvm_sigmask))
1993 if (kvm_sigmask.len != sizeof csigset)
1996 if (copy_from_user(&csigset, sigmask_arg->sigset,
2000 sigset_from_compat(&sigset, &csigset);
2001 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2005 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2013 static long kvm_vm_ioctl(struct file *filp,
2014 unsigned int ioctl, unsigned long arg)
2016 struct kvm *kvm = filp->private_data;
2017 void __user *argp = (void __user *)arg;
2020 if (kvm->mm != current->mm)
2023 case KVM_CREATE_VCPU:
2024 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2028 case KVM_SET_USER_MEMORY_REGION: {
2029 struct kvm_userspace_memory_region kvm_userspace_mem;
2032 if (copy_from_user(&kvm_userspace_mem, argp,
2033 sizeof kvm_userspace_mem))
2036 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2041 case KVM_GET_DIRTY_LOG: {
2042 struct kvm_dirty_log log;
2045 if (copy_from_user(&log, argp, sizeof log))
2047 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2052 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2053 case KVM_REGISTER_COALESCED_MMIO: {
2054 struct kvm_coalesced_mmio_zone zone;
2056 if (copy_from_user(&zone, argp, sizeof zone))
2058 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2064 case KVM_UNREGISTER_COALESCED_MMIO: {
2065 struct kvm_coalesced_mmio_zone zone;
2067 if (copy_from_user(&zone, argp, sizeof zone))
2069 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2077 struct kvm_irqfd data;
2080 if (copy_from_user(&data, argp, sizeof data))
2082 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2085 case KVM_IOEVENTFD: {
2086 struct kvm_ioeventfd data;
2089 if (copy_from_user(&data, argp, sizeof data))
2091 r = kvm_ioeventfd(kvm, &data);
2094 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2095 case KVM_SET_BOOT_CPU_ID:
2097 mutex_lock(&kvm->lock);
2098 if (atomic_read(&kvm->online_vcpus) != 0)
2101 kvm->bsp_vcpu_id = arg;
2102 mutex_unlock(&kvm->lock);
2106 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2108 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2114 #ifdef CONFIG_COMPAT
2115 struct compat_kvm_dirty_log {
2119 compat_uptr_t dirty_bitmap; /* one bit per page */
2124 static long kvm_vm_compat_ioctl(struct file *filp,
2125 unsigned int ioctl, unsigned long arg)
2127 struct kvm *kvm = filp->private_data;
2130 if (kvm->mm != current->mm)
2133 case KVM_GET_DIRTY_LOG: {
2134 struct compat_kvm_dirty_log compat_log;
2135 struct kvm_dirty_log log;
2138 if (copy_from_user(&compat_log, (void __user *)arg,
2139 sizeof(compat_log)))
2141 log.slot = compat_log.slot;
2142 log.padding1 = compat_log.padding1;
2143 log.padding2 = compat_log.padding2;
2144 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2146 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2152 r = kvm_vm_ioctl(filp, ioctl, arg);
2160 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2162 struct page *page[1];
2165 gfn_t gfn = vmf->pgoff;
2166 struct kvm *kvm = vma->vm_file->private_data;
2168 addr = gfn_to_hva(kvm, gfn);
2169 if (kvm_is_error_hva(addr))
2170 return VM_FAULT_SIGBUS;
2172 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2174 if (unlikely(npages != 1))
2175 return VM_FAULT_SIGBUS;
2177 vmf->page = page[0];
2181 static const struct vm_operations_struct kvm_vm_vm_ops = {
2182 .fault = kvm_vm_fault,
2185 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2187 vma->vm_ops = &kvm_vm_vm_ops;
2191 static struct file_operations kvm_vm_fops = {
2192 .release = kvm_vm_release,
2193 .unlocked_ioctl = kvm_vm_ioctl,
2194 #ifdef CONFIG_COMPAT
2195 .compat_ioctl = kvm_vm_compat_ioctl,
2197 .mmap = kvm_vm_mmap,
2198 .llseek = noop_llseek,
2201 static int kvm_dev_ioctl_create_vm(void)
2206 kvm = kvm_create_vm();
2208 return PTR_ERR(kvm);
2209 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2210 r = kvm_coalesced_mmio_init(kvm);
2216 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2223 static long kvm_dev_ioctl_check_extension_generic(long arg)
2226 case KVM_CAP_USER_MEMORY:
2227 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2228 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2229 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2230 case KVM_CAP_SET_BOOT_CPU_ID:
2232 case KVM_CAP_INTERNAL_ERROR_DATA:
2234 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2235 case KVM_CAP_IRQ_ROUTING:
2236 return KVM_MAX_IRQ_ROUTES;
2241 return kvm_dev_ioctl_check_extension(arg);
2244 static long kvm_dev_ioctl(struct file *filp,
2245 unsigned int ioctl, unsigned long arg)
2250 case KVM_GET_API_VERSION:
2254 r = KVM_API_VERSION;
2260 r = kvm_dev_ioctl_create_vm();
2262 case KVM_CHECK_EXTENSION:
2263 r = kvm_dev_ioctl_check_extension_generic(arg);
2265 case KVM_GET_VCPU_MMAP_SIZE:
2269 r = PAGE_SIZE; /* struct kvm_run */
2271 r += PAGE_SIZE; /* pio data page */
2273 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2274 r += PAGE_SIZE; /* coalesced mmio ring page */
2277 case KVM_TRACE_ENABLE:
2278 case KVM_TRACE_PAUSE:
2279 case KVM_TRACE_DISABLE:
2283 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2289 static struct file_operations kvm_chardev_ops = {
2290 .unlocked_ioctl = kvm_dev_ioctl,
2291 .compat_ioctl = kvm_dev_ioctl,
2292 .llseek = noop_llseek,
2295 static struct miscdevice kvm_dev = {
2301 static void hardware_enable_nolock(void *junk)
2303 int cpu = raw_smp_processor_id();
2306 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2309 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2311 r = kvm_arch_hardware_enable(NULL);
2314 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2315 atomic_inc(&hardware_enable_failed);
2316 printk(KERN_INFO "kvm: enabling virtualization on "
2317 "CPU%d failed\n", cpu);
2321 static void hardware_enable(void *junk)
2323 raw_spin_lock(&kvm_lock);
2324 hardware_enable_nolock(junk);
2325 raw_spin_unlock(&kvm_lock);
2328 static void hardware_disable_nolock(void *junk)
2330 int cpu = raw_smp_processor_id();
2332 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2334 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2335 kvm_arch_hardware_disable(NULL);
2338 static void hardware_disable(void *junk)
2340 raw_spin_lock(&kvm_lock);
2341 hardware_disable_nolock(junk);
2342 raw_spin_unlock(&kvm_lock);
2345 static void hardware_disable_all_nolock(void)
2347 BUG_ON(!kvm_usage_count);
2350 if (!kvm_usage_count)
2351 on_each_cpu(hardware_disable_nolock, NULL, 1);
2354 static void hardware_disable_all(void)
2356 raw_spin_lock(&kvm_lock);
2357 hardware_disable_all_nolock();
2358 raw_spin_unlock(&kvm_lock);
2361 static int hardware_enable_all(void)
2365 raw_spin_lock(&kvm_lock);
2368 if (kvm_usage_count == 1) {
2369 atomic_set(&hardware_enable_failed, 0);
2370 on_each_cpu(hardware_enable_nolock, NULL, 1);
2372 if (atomic_read(&hardware_enable_failed)) {
2373 hardware_disable_all_nolock();
2378 raw_spin_unlock(&kvm_lock);
2383 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2388 if (!kvm_usage_count)
2391 val &= ~CPU_TASKS_FROZEN;
2394 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2396 hardware_disable(NULL);
2399 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2401 hardware_enable(NULL);
2408 asmlinkage void kvm_spurious_fault(void)
2410 /* Fault while not rebooting. We want the trace. */
2413 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2415 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2419 * Some (well, at least mine) BIOSes hang on reboot if
2422 * And Intel TXT required VMX off for all cpu when system shutdown.
2424 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2425 kvm_rebooting = true;
2426 on_each_cpu(hardware_disable_nolock, NULL, 1);
2430 static struct notifier_block kvm_reboot_notifier = {
2431 .notifier_call = kvm_reboot,
2435 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2439 for (i = 0; i < bus->dev_count; i++) {
2440 struct kvm_io_device *pos = bus->range[i].dev;
2442 kvm_iodevice_destructor(pos);
2447 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2449 const struct kvm_io_range *r1 = p1;
2450 const struct kvm_io_range *r2 = p2;
2452 if (r1->addr < r2->addr)
2454 if (r1->addr + r1->len > r2->addr + r2->len)
2459 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2460 gpa_t addr, int len)
2462 if (bus->dev_count == NR_IOBUS_DEVS)
2465 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2471 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2472 kvm_io_bus_sort_cmp, NULL);
2477 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2478 gpa_t addr, int len)
2480 struct kvm_io_range *range, key;
2483 key = (struct kvm_io_range) {
2488 range = bsearch(&key, bus->range, bus->dev_count,
2489 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2493 off = range - bus->range;
2495 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2501 /* kvm_io_bus_write - called under kvm->slots_lock */
2502 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2503 int len, const void *val)
2506 struct kvm_io_bus *bus;
2507 struct kvm_io_range range;
2509 range = (struct kvm_io_range) {
2514 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2515 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2519 while (idx < bus->dev_count &&
2520 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2521 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2529 /* kvm_io_bus_read - called under kvm->slots_lock */
2530 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2534 struct kvm_io_bus *bus;
2535 struct kvm_io_range range;
2537 range = (struct kvm_io_range) {
2542 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2543 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2547 while (idx < bus->dev_count &&
2548 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2549 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2557 /* Caller must hold slots_lock. */
2558 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2559 int len, struct kvm_io_device *dev)
2561 struct kvm_io_bus *new_bus, *bus;
2563 bus = kvm->buses[bus_idx];
2564 if (bus->dev_count > NR_IOBUS_DEVS-1)
2567 new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2570 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2571 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2572 synchronize_srcu_expedited(&kvm->srcu);
2578 /* Caller must hold slots_lock. */
2579 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2580 struct kvm_io_device *dev)
2583 struct kvm_io_bus *new_bus, *bus;
2585 bus = kvm->buses[bus_idx];
2587 new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2592 for (i = 0; i < new_bus->dev_count; i++)
2593 if (new_bus->range[i].dev == dev) {
2595 new_bus->dev_count--;
2596 new_bus->range[i] = new_bus->range[new_bus->dev_count];
2597 sort(new_bus->range, new_bus->dev_count,
2598 sizeof(struct kvm_io_range),
2599 kvm_io_bus_sort_cmp, NULL);
2608 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2609 synchronize_srcu_expedited(&kvm->srcu);
2614 static struct notifier_block kvm_cpu_notifier = {
2615 .notifier_call = kvm_cpu_hotplug,
2618 static int vm_stat_get(void *_offset, u64 *val)
2620 unsigned offset = (long)_offset;
2624 raw_spin_lock(&kvm_lock);
2625 list_for_each_entry(kvm, &vm_list, vm_list)
2626 *val += *(u32 *)((void *)kvm + offset);
2627 raw_spin_unlock(&kvm_lock);
2631 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2633 static int vcpu_stat_get(void *_offset, u64 *val)
2635 unsigned offset = (long)_offset;
2637 struct kvm_vcpu *vcpu;
2641 raw_spin_lock(&kvm_lock);
2642 list_for_each_entry(kvm, &vm_list, vm_list)
2643 kvm_for_each_vcpu(i, vcpu, kvm)
2644 *val += *(u32 *)((void *)vcpu + offset);
2646 raw_spin_unlock(&kvm_lock);
2650 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2652 static const struct file_operations *stat_fops[] = {
2653 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2654 [KVM_STAT_VM] = &vm_stat_fops,
2657 static int kvm_init_debug(void)
2660 struct kvm_stats_debugfs_item *p;
2662 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2663 if (kvm_debugfs_dir == NULL)
2666 for (p = debugfs_entries; p->name; ++p) {
2667 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2668 (void *)(long)p->offset,
2669 stat_fops[p->kind]);
2670 if (p->dentry == NULL)
2677 debugfs_remove_recursive(kvm_debugfs_dir);
2682 static void kvm_exit_debug(void)
2684 struct kvm_stats_debugfs_item *p;
2686 for (p = debugfs_entries; p->name; ++p)
2687 debugfs_remove(p->dentry);
2688 debugfs_remove(kvm_debugfs_dir);
2691 static int kvm_suspend(void)
2693 if (kvm_usage_count)
2694 hardware_disable_nolock(NULL);
2698 static void kvm_resume(void)
2700 if (kvm_usage_count) {
2701 WARN_ON(raw_spin_is_locked(&kvm_lock));
2702 hardware_enable_nolock(NULL);
2706 static struct syscore_ops kvm_syscore_ops = {
2707 .suspend = kvm_suspend,
2708 .resume = kvm_resume,
2711 struct page *bad_page;
2715 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2717 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2720 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2722 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2724 kvm_arch_vcpu_load(vcpu, cpu);
2727 static void kvm_sched_out(struct preempt_notifier *pn,
2728 struct task_struct *next)
2730 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2732 kvm_arch_vcpu_put(vcpu);
2735 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2736 struct module *module)
2741 r = kvm_arch_init(opaque);
2745 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2747 if (bad_page == NULL) {
2752 bad_pfn = page_to_pfn(bad_page);
2754 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2756 if (hwpoison_page == NULL) {
2761 hwpoison_pfn = page_to_pfn(hwpoison_page);
2763 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2765 if (fault_page == NULL) {
2770 fault_pfn = page_to_pfn(fault_page);
2772 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2777 r = kvm_arch_hardware_setup();
2781 for_each_online_cpu(cpu) {
2782 smp_call_function_single(cpu,
2783 kvm_arch_check_processor_compat,
2789 r = register_cpu_notifier(&kvm_cpu_notifier);
2792 register_reboot_notifier(&kvm_reboot_notifier);
2794 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2796 vcpu_align = __alignof__(struct kvm_vcpu);
2797 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2799 if (!kvm_vcpu_cache) {
2804 r = kvm_async_pf_init();
2808 kvm_chardev_ops.owner = module;
2809 kvm_vm_fops.owner = module;
2810 kvm_vcpu_fops.owner = module;
2812 r = misc_register(&kvm_dev);
2814 printk(KERN_ERR "kvm: misc device register failed\n");
2818 register_syscore_ops(&kvm_syscore_ops);
2820 kvm_preempt_ops.sched_in = kvm_sched_in;
2821 kvm_preempt_ops.sched_out = kvm_sched_out;
2823 r = kvm_init_debug();
2825 printk(KERN_ERR "kvm: create debugfs files failed\n");
2832 unregister_syscore_ops(&kvm_syscore_ops);
2834 kvm_async_pf_deinit();
2836 kmem_cache_destroy(kvm_vcpu_cache);
2838 unregister_reboot_notifier(&kvm_reboot_notifier);
2839 unregister_cpu_notifier(&kvm_cpu_notifier);
2842 kvm_arch_hardware_unsetup();
2844 free_cpumask_var(cpus_hardware_enabled);
2847 __free_page(fault_page);
2849 __free_page(hwpoison_page);
2850 __free_page(bad_page);
2856 EXPORT_SYMBOL_GPL(kvm_init);
2861 misc_deregister(&kvm_dev);
2862 kmem_cache_destroy(kvm_vcpu_cache);
2863 kvm_async_pf_deinit();
2864 unregister_syscore_ops(&kvm_syscore_ops);
2865 unregister_reboot_notifier(&kvm_reboot_notifier);
2866 unregister_cpu_notifier(&kvm_cpu_notifier);
2867 on_each_cpu(hardware_disable_nolock, NULL, 1);
2868 kvm_arch_hardware_unsetup();
2870 free_cpumask_var(cpus_hardware_enabled);
2871 __free_page(hwpoison_page);
2872 __free_page(bad_page);
2874 EXPORT_SYMBOL_GPL(kvm_exit);