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 bool kvm_is_mmio_pfn(pfn_t pfn)
105 if (pfn_valid(pfn)) {
107 struct page *tail = pfn_to_page(pfn);
108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
125 return PageReserved(tail);
132 * Switches to specified vcpu, until a matching vcpu_put()
134 void vcpu_load(struct kvm_vcpu *vcpu)
138 mutex_lock(&vcpu->mutex);
139 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
140 /* The thread running this VCPU changed. */
141 struct pid *oldpid = vcpu->pid;
142 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
143 rcu_assign_pointer(vcpu->pid, newpid);
148 preempt_notifier_register(&vcpu->preempt_notifier);
149 kvm_arch_vcpu_load(vcpu, cpu);
153 void vcpu_put(struct kvm_vcpu *vcpu)
156 kvm_arch_vcpu_put(vcpu);
157 preempt_notifier_unregister(&vcpu->preempt_notifier);
159 mutex_unlock(&vcpu->mutex);
162 static void ack_flush(void *_completed)
166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
171 struct kvm_vcpu *vcpu;
173 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176 kvm_for_each_vcpu(i, vcpu, kvm) {
177 kvm_make_request(req, vcpu);
180 /* Set ->requests bit before we read ->mode */
183 if (cpus != NULL && cpu != -1 && cpu != me &&
184 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
185 cpumask_set_cpu(cpu, cpus);
187 if (unlikely(cpus == NULL))
188 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
189 else if (!cpumask_empty(cpus))
190 smp_call_function_many(cpus, ack_flush, NULL, 1);
194 free_cpumask_var(cpus);
198 void kvm_flush_remote_tlbs(struct kvm *kvm)
200 long dirty_count = kvm->tlbs_dirty;
203 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
204 ++kvm->stat.remote_tlb_flush;
205 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 void kvm_reload_remote_mmus(struct kvm *kvm)
210 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
218 mutex_init(&vcpu->mutex);
223 init_waitqueue_head(&vcpu->wq);
224 kvm_async_pf_vcpu_init(vcpu);
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
231 vcpu->run = page_address(page);
233 kvm_vcpu_set_in_spin_loop(vcpu, false);
234 kvm_vcpu_set_dy_eligible(vcpu, false);
236 r = kvm_arch_vcpu_init(vcpu);
242 free_page((unsigned long)vcpu->run);
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 kvm_arch_vcpu_uninit(vcpu);
252 free_page((unsigned long)vcpu->run);
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259 return container_of(mn, struct kvm, mmu_notifier);
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address)
266 struct kvm *kvm = mmu_notifier_to_kvm(mn);
267 int need_tlb_flush, idx;
270 * When ->invalidate_page runs, the linux pte has been zapped
271 * already but the page is still allocated until
272 * ->invalidate_page returns. So if we increase the sequence
273 * here the kvm page fault will notice if the spte can't be
274 * established because the page is going to be freed. If
275 * instead the kvm page fault establishes the spte before
276 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * The sequence increase only need to be seen at spin_unlock
280 * time, and not at spin_lock time.
282 * Increasing the sequence after the spin_unlock would be
283 * unsafe because the kvm page fault could then establish the
284 * pte after kvm_unmap_hva returned, without noticing the page
285 * is going to be freed.
287 idx = srcu_read_lock(&kvm->srcu);
288 spin_lock(&kvm->mmu_lock);
290 kvm->mmu_notifier_seq++;
291 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292 /* we've to flush the tlb before the pages can be freed */
294 kvm_flush_remote_tlbs(kvm);
296 spin_unlock(&kvm->mmu_lock);
297 srcu_read_unlock(&kvm->srcu, idx);
300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
301 struct mm_struct *mm,
302 unsigned long address,
305 struct kvm *kvm = mmu_notifier_to_kvm(mn);
308 idx = srcu_read_lock(&kvm->srcu);
309 spin_lock(&kvm->mmu_lock);
310 kvm->mmu_notifier_seq++;
311 kvm_set_spte_hva(kvm, address, pte);
312 spin_unlock(&kvm->mmu_lock);
313 srcu_read_unlock(&kvm->srcu, idx);
316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
317 struct mm_struct *mm,
321 struct kvm *kvm = mmu_notifier_to_kvm(mn);
322 int need_tlb_flush = 0, idx;
324 idx = srcu_read_lock(&kvm->srcu);
325 spin_lock(&kvm->mmu_lock);
327 * The count increase must become visible at unlock time as no
328 * spte can be established without taking the mmu_lock and
329 * count is also read inside the mmu_lock critical section.
331 kvm->mmu_notifier_count++;
332 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
333 need_tlb_flush |= kvm->tlbs_dirty;
334 /* we've to flush the tlb before the pages can be freed */
336 kvm_flush_remote_tlbs(kvm);
338 spin_unlock(&kvm->mmu_lock);
339 srcu_read_unlock(&kvm->srcu, idx);
342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
343 struct mm_struct *mm,
347 struct kvm *kvm = mmu_notifier_to_kvm(mn);
349 spin_lock(&kvm->mmu_lock);
351 * This sequence increase will notify the kvm page fault that
352 * the page that is going to be mapped in the spte could have
355 kvm->mmu_notifier_seq++;
358 * The above sequence increase must be visible before the
359 * below count decrease, which is ensured by the smp_wmb above
360 * in conjunction with the smp_rmb in mmu_notifier_retry().
362 kvm->mmu_notifier_count--;
363 spin_unlock(&kvm->mmu_lock);
365 BUG_ON(kvm->mmu_notifier_count < 0);
368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
369 struct mm_struct *mm,
370 unsigned long address)
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
375 idx = srcu_read_lock(&kvm->srcu);
376 spin_lock(&kvm->mmu_lock);
378 young = kvm_age_hva(kvm, address);
380 kvm_flush_remote_tlbs(kvm);
382 spin_unlock(&kvm->mmu_lock);
383 srcu_read_unlock(&kvm->srcu, idx);
388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
389 struct mm_struct *mm,
390 unsigned long address)
392 struct kvm *kvm = mmu_notifier_to_kvm(mn);
395 idx = srcu_read_lock(&kvm->srcu);
396 spin_lock(&kvm->mmu_lock);
397 young = kvm_test_age_hva(kvm, address);
398 spin_unlock(&kvm->mmu_lock);
399 srcu_read_unlock(&kvm->srcu, idx);
404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
405 struct mm_struct *mm)
407 struct kvm *kvm = mmu_notifier_to_kvm(mn);
410 idx = srcu_read_lock(&kvm->srcu);
411 kvm_arch_flush_shadow(kvm);
412 srcu_read_unlock(&kvm->srcu, idx);
415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
416 .invalidate_page = kvm_mmu_notifier_invalidate_page,
417 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
418 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
419 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
420 .test_young = kvm_mmu_notifier_test_young,
421 .change_pte = kvm_mmu_notifier_change_pte,
422 .release = kvm_mmu_notifier_release,
425 static int kvm_init_mmu_notifier(struct kvm *kvm)
427 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
428 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
431 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
433 static int kvm_init_mmu_notifier(struct kvm *kvm)
438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
440 static void kvm_init_memslots_id(struct kvm *kvm)
443 struct kvm_memslots *slots = kvm->memslots;
445 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
446 slots->id_to_index[i] = slots->memslots[i].id = i;
449 static struct kvm *kvm_create_vm(unsigned long type)
452 struct kvm *kvm = kvm_arch_alloc_vm();
455 return ERR_PTR(-ENOMEM);
457 r = kvm_arch_init_vm(kvm, type);
459 goto out_err_nodisable;
461 r = hardware_enable_all();
463 goto out_err_nodisable;
465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
466 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
467 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
474 kvm_init_memslots_id(kvm);
475 if (init_srcu_struct(&kvm->srcu))
477 for (i = 0; i < KVM_NR_BUSES; i++) {
478 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
484 spin_lock_init(&kvm->mmu_lock);
485 kvm->mm = current->mm;
486 atomic_inc(&kvm->mm->mm_count);
487 kvm_eventfd_init(kvm);
488 mutex_init(&kvm->lock);
489 mutex_init(&kvm->irq_lock);
490 mutex_init(&kvm->slots_lock);
491 atomic_set(&kvm->users_count, 1);
493 r = kvm_init_mmu_notifier(kvm);
497 raw_spin_lock(&kvm_lock);
498 list_add(&kvm->vm_list, &vm_list);
499 raw_spin_unlock(&kvm_lock);
504 cleanup_srcu_struct(&kvm->srcu);
506 hardware_disable_all();
508 for (i = 0; i < KVM_NR_BUSES; i++)
509 kfree(kvm->buses[i]);
510 kfree(kvm->memslots);
511 kvm_arch_free_vm(kvm);
516 * Avoid using vmalloc for a small buffer.
517 * Should not be used when the size is statically known.
519 void *kvm_kvzalloc(unsigned long size)
521 if (size > PAGE_SIZE)
522 return vzalloc(size);
524 return kzalloc(size, GFP_KERNEL);
527 void kvm_kvfree(const void *addr)
529 if (is_vmalloc_addr(addr))
535 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
537 if (!memslot->dirty_bitmap)
540 kvm_kvfree(memslot->dirty_bitmap);
541 memslot->dirty_bitmap = NULL;
545 * Free any memory in @free but not in @dont.
547 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
548 struct kvm_memory_slot *dont)
550 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
551 kvm_destroy_dirty_bitmap(free);
553 kvm_arch_free_memslot(free, dont);
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);
620 * Allocation size is twice as large as the actual dirty bitmap size.
621 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
623 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
628 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
629 if (!memslot->dirty_bitmap)
632 #endif /* !CONFIG_S390 */
636 static int cmp_memslot(const void *slot1, const void *slot2)
638 struct kvm_memory_slot *s1, *s2;
640 s1 = (struct kvm_memory_slot *)slot1;
641 s2 = (struct kvm_memory_slot *)slot2;
643 if (s1->npages < s2->npages)
645 if (s1->npages > s2->npages)
652 * Sort the memslots base on its size, so the larger slots
653 * will get better fit.
655 static void sort_memslots(struct kvm_memslots *slots)
659 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
660 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
662 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
663 slots->id_to_index[slots->memslots[i].id] = i;
666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
670 struct kvm_memory_slot *old = id_to_memslot(slots, id);
671 unsigned long npages = old->npages;
674 if (new->npages != npages)
675 sort_memslots(slots);
681 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
683 if (mem->flags & ~KVM_MEM_LOG_DIRTY_PAGES)
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;
709 r = check_memory_region_flags(mem);
714 /* General sanity checks */
715 if (mem->memory_size & (PAGE_SIZE - 1))
717 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
719 /* We can read the guest memory with __xxx_user() later on. */
721 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
722 !access_ok(VERIFY_WRITE,
723 (void __user *)(unsigned long)mem->userspace_addr,
726 if (mem->slot >= KVM_MEM_SLOTS_NUM)
728 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
731 memslot = id_to_memslot(kvm->memslots, mem->slot);
732 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
733 npages = mem->memory_size >> PAGE_SHIFT;
736 if (npages > KVM_MEM_MAX_NR_PAGES)
740 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
742 new = old = *memslot;
745 new.base_gfn = base_gfn;
747 new.flags = mem->flags;
749 /* Disallow changing a memory slot's size. */
751 if (npages && old.npages && npages != old.npages)
754 /* Check for overlaps */
756 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
757 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
759 if (s == memslot || !s->npages)
761 if (!((base_gfn + npages <= s->base_gfn) ||
762 (base_gfn >= s->base_gfn + s->npages)))
766 /* Free page dirty bitmap if unneeded */
767 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
768 new.dirty_bitmap = NULL;
772 /* Allocate if a slot is being created */
773 if (npages && !old.npages) {
774 new.user_alloc = user_alloc;
775 new.userspace_addr = mem->userspace_addr;
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 */
835 new.dirty_bitmap = NULL;
836 memset(&new.arch, 0, sizeof(new.arch));
839 update_memslots(slots, &new);
840 old_memslots = kvm->memslots;
841 rcu_assign_pointer(kvm->memslots, slots);
842 synchronize_srcu_expedited(&kvm->srcu);
844 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
847 * If the new memory slot is created, we need to clear all
850 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
851 kvm_arch_flush_shadow(kvm);
853 kvm_free_physmem_slot(&old, &new);
859 kvm_free_physmem_slot(&new, &old);
864 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
866 int kvm_set_memory_region(struct kvm *kvm,
867 struct kvm_userspace_memory_region *mem,
872 mutex_lock(&kvm->slots_lock);
873 r = __kvm_set_memory_region(kvm, mem, user_alloc);
874 mutex_unlock(&kvm->slots_lock);
877 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
879 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
881 kvm_userspace_memory_region *mem,
884 if (mem->slot >= KVM_MEMORY_SLOTS)
886 return kvm_set_memory_region(kvm, mem, user_alloc);
889 int kvm_get_dirty_log(struct kvm *kvm,
890 struct kvm_dirty_log *log, int *is_dirty)
892 struct kvm_memory_slot *memslot;
895 unsigned long any = 0;
898 if (log->slot >= KVM_MEMORY_SLOTS)
901 memslot = id_to_memslot(kvm->memslots, log->slot);
903 if (!memslot->dirty_bitmap)
906 n = kvm_dirty_bitmap_bytes(memslot);
908 for (i = 0; !any && i < n/sizeof(long); ++i)
909 any = memslot->dirty_bitmap[i];
912 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
923 bool kvm_largepages_enabled(void)
925 return largepages_enabled;
928 void kvm_disable_largepages(void)
930 largepages_enabled = false;
932 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
934 static inline unsigned long bad_hva(void)
939 int kvm_is_error_hva(unsigned long addr)
941 return addr == bad_hva();
943 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
945 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
947 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
949 EXPORT_SYMBOL_GPL(gfn_to_memslot);
951 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
953 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
955 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
956 memslot->flags & KVM_MEMSLOT_INVALID)
961 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
963 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
965 struct vm_area_struct *vma;
966 unsigned long addr, size;
970 addr = gfn_to_hva(kvm, gfn);
971 if (kvm_is_error_hva(addr))
974 down_read(¤t->mm->mmap_sem);
975 vma = find_vma(current->mm, addr);
979 size = vma_kernel_pagesize(vma);
982 up_read(¤t->mm->mmap_sem);
987 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
990 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
994 *nr_pages = slot->npages - (gfn - slot->base_gfn);
996 return gfn_to_hva_memslot(slot, gfn);
999 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1001 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1003 EXPORT_SYMBOL_GPL(gfn_to_hva);
1006 * The hva returned by this function is only allowed to be read.
1007 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1009 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1011 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1014 static int kvm_read_hva(void *data, void __user *hva, int len)
1016 return __copy_from_user(data, hva, len);
1019 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1021 return __copy_from_user_inatomic(data, hva, len);
1024 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1025 unsigned long start, int write, struct page **page)
1027 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1030 flags |= FOLL_WRITE;
1032 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1035 static inline int check_user_page_hwpoison(unsigned long addr)
1037 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1039 rc = __get_user_pages(current, current->mm, addr, 1,
1040 flags, NULL, NULL, NULL);
1041 return rc == -EHWPOISON;
1045 * The atomic path to get the writable pfn which will be stored in @pfn,
1046 * true indicates success, otherwise false is returned.
1048 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1049 bool write_fault, bool *writable, pfn_t *pfn)
1051 struct page *page[1];
1054 if (!(async || atomic))
1058 * Fast pin a writable pfn only if it is a write fault request
1059 * or the caller allows to map a writable pfn for a read fault
1062 if (!(write_fault || writable))
1065 npages = __get_user_pages_fast(addr, 1, 1, page);
1067 *pfn = page_to_pfn(page[0]);
1078 * The slow path to get the pfn of the specified host virtual address,
1079 * 1 indicates success, -errno is returned if error is detected.
1081 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1082 bool *writable, pfn_t *pfn)
1084 struct page *page[1];
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,
1103 /* map read fault as writable if possible */
1104 if (unlikely(!write_fault) && writable) {
1105 struct page *wpage[1];
1107 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1116 *pfn = page_to_pfn(page[0]);
1121 * Pin guest page in memory and return its pfn.
1122 * @addr: host virtual address which maps memory to the guest
1123 * @atomic: whether this function can sleep
1124 * @async: whether this function need to wait IO complete if the
1125 * host page is not in the memory
1126 * @write_fault: whether we should get a writable host page
1127 * @writable: whether it allows to map a writable host page for !@write_fault
1129 * The function will map a writable host page for these two cases:
1130 * 1): @write_fault = true
1131 * 2): @write_fault = false && @writable, @writable will tell the caller
1132 * whether the mapping is writable.
1134 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1135 bool write_fault, bool *writable)
1137 struct vm_area_struct *vma;
1141 /* we can do it either atomically or asynchronously, not both */
1142 BUG_ON(atomic && async);
1144 BUG_ON(!write_fault && !writable);
1146 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1150 return KVM_PFN_ERR_FAULT;
1152 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1156 down_read(¤t->mm->mmap_sem);
1157 if (npages == -EHWPOISON ||
1158 (!async && check_user_page_hwpoison(addr))) {
1159 pfn = KVM_PFN_ERR_HWPOISON;
1163 vma = find_vma_intersection(current->mm, addr, addr + 1);
1166 pfn = KVM_PFN_ERR_FAULT;
1167 else if ((vma->vm_flags & VM_PFNMAP)) {
1168 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1170 BUG_ON(!kvm_is_mmio_pfn(pfn));
1172 if (async && (vma->vm_flags & VM_WRITE))
1174 pfn = KVM_PFN_ERR_FAULT;
1177 up_read(¤t->mm->mmap_sem);
1181 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1182 bool write_fault, bool *writable)
1189 addr = gfn_to_hva(kvm, gfn);
1190 if (kvm_is_error_hva(addr))
1191 return KVM_PFN_ERR_BAD;
1193 return hva_to_pfn(addr, atomic, async, write_fault, writable);
1196 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1198 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1200 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1202 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1203 bool write_fault, bool *writable)
1205 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1207 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1209 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1211 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1213 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1215 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1218 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1220 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1222 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1224 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1225 return hva_to_pfn(addr, false, NULL, true, NULL);
1228 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1230 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1232 return hva_to_pfn(addr, true, NULL, true, NULL);
1234 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1236 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1242 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1243 if (kvm_is_error_hva(addr))
1246 if (entry < nr_pages)
1249 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1251 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1253 static struct page *kvm_pfn_to_page(pfn_t pfn)
1255 if (is_error_pfn(pfn))
1256 return KVM_ERR_PTR_BAD_PAGE;
1258 if (kvm_is_mmio_pfn(pfn)) {
1260 return KVM_ERR_PTR_BAD_PAGE;
1263 return pfn_to_page(pfn);
1266 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1270 pfn = gfn_to_pfn(kvm, gfn);
1272 return kvm_pfn_to_page(pfn);
1275 EXPORT_SYMBOL_GPL(gfn_to_page);
1277 void kvm_release_page_clean(struct page *page)
1279 WARN_ON(is_error_page(page));
1281 kvm_release_pfn_clean(page_to_pfn(page));
1283 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1285 void kvm_release_pfn_clean(pfn_t pfn)
1287 WARN_ON(is_error_pfn(pfn));
1289 if (!kvm_is_mmio_pfn(pfn))
1290 put_page(pfn_to_page(pfn));
1292 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1294 void kvm_release_page_dirty(struct page *page)
1296 WARN_ON(is_error_page(page));
1298 kvm_release_pfn_dirty(page_to_pfn(page));
1300 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1302 void kvm_release_pfn_dirty(pfn_t pfn)
1304 kvm_set_pfn_dirty(pfn);
1305 kvm_release_pfn_clean(pfn);
1307 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1309 void kvm_set_page_dirty(struct page *page)
1311 kvm_set_pfn_dirty(page_to_pfn(page));
1313 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1315 void kvm_set_pfn_dirty(pfn_t pfn)
1317 if (!kvm_is_mmio_pfn(pfn)) {
1318 struct page *page = pfn_to_page(pfn);
1319 if (!PageReserved(page))
1323 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1325 void kvm_set_pfn_accessed(pfn_t pfn)
1327 if (!kvm_is_mmio_pfn(pfn))
1328 mark_page_accessed(pfn_to_page(pfn));
1330 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1332 void kvm_get_pfn(pfn_t pfn)
1334 if (!kvm_is_mmio_pfn(pfn))
1335 get_page(pfn_to_page(pfn));
1337 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1339 static int next_segment(unsigned long len, int offset)
1341 if (len > PAGE_SIZE - offset)
1342 return PAGE_SIZE - offset;
1347 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1353 addr = gfn_to_hva_read(kvm, gfn);
1354 if (kvm_is_error_hva(addr))
1356 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1361 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1363 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1365 gfn_t gfn = gpa >> PAGE_SHIFT;
1367 int offset = offset_in_page(gpa);
1370 while ((seg = next_segment(len, offset)) != 0) {
1371 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1381 EXPORT_SYMBOL_GPL(kvm_read_guest);
1383 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1388 gfn_t gfn = gpa >> PAGE_SHIFT;
1389 int offset = offset_in_page(gpa);
1391 addr = gfn_to_hva_read(kvm, gfn);
1392 if (kvm_is_error_hva(addr))
1394 pagefault_disable();
1395 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1401 EXPORT_SYMBOL(kvm_read_guest_atomic);
1403 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1404 int offset, int len)
1409 addr = gfn_to_hva(kvm, gfn);
1410 if (kvm_is_error_hva(addr))
1412 r = __copy_to_user((void __user *)addr + offset, data, len);
1415 mark_page_dirty(kvm, gfn);
1418 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1420 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1423 gfn_t gfn = gpa >> PAGE_SHIFT;
1425 int offset = offset_in_page(gpa);
1428 while ((seg = next_segment(len, offset)) != 0) {
1429 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1440 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1443 struct kvm_memslots *slots = kvm_memslots(kvm);
1444 int offset = offset_in_page(gpa);
1445 gfn_t gfn = gpa >> PAGE_SHIFT;
1448 ghc->generation = slots->generation;
1449 ghc->memslot = gfn_to_memslot(kvm, gfn);
1450 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1451 if (!kvm_is_error_hva(ghc->hva))
1458 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1460 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1461 void *data, unsigned long len)
1463 struct kvm_memslots *slots = kvm_memslots(kvm);
1466 if (slots->generation != ghc->generation)
1467 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1469 if (kvm_is_error_hva(ghc->hva))
1472 r = __copy_to_user((void __user *)ghc->hva, data, len);
1475 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1479 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1481 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1482 void *data, unsigned long len)
1484 struct kvm_memslots *slots = kvm_memslots(kvm);
1487 if (slots->generation != ghc->generation)
1488 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1490 if (kvm_is_error_hva(ghc->hva))
1493 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1499 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1501 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1503 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1506 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1508 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1510 gfn_t gfn = gpa >> PAGE_SHIFT;
1512 int offset = offset_in_page(gpa);
1515 while ((seg = next_segment(len, offset)) != 0) {
1516 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1525 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1527 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1530 if (memslot && memslot->dirty_bitmap) {
1531 unsigned long rel_gfn = gfn - memslot->base_gfn;
1533 /* TODO: introduce set_bit_le() and use it */
1534 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1538 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1540 struct kvm_memory_slot *memslot;
1542 memslot = gfn_to_memslot(kvm, gfn);
1543 mark_page_dirty_in_slot(kvm, memslot, gfn);
1547 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1549 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1554 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1556 if (kvm_arch_vcpu_runnable(vcpu)) {
1557 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1560 if (kvm_cpu_has_pending_timer(vcpu))
1562 if (signal_pending(current))
1568 finish_wait(&vcpu->wq, &wait);
1573 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1575 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1578 int cpu = vcpu->cpu;
1579 wait_queue_head_t *wqp;
1581 wqp = kvm_arch_vcpu_wq(vcpu);
1582 if (waitqueue_active(wqp)) {
1583 wake_up_interruptible(wqp);
1584 ++vcpu->stat.halt_wakeup;
1588 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1589 if (kvm_arch_vcpu_should_kick(vcpu))
1590 smp_send_reschedule(cpu);
1593 #endif /* !CONFIG_S390 */
1595 void kvm_resched(struct kvm_vcpu *vcpu)
1597 if (!need_resched())
1601 EXPORT_SYMBOL_GPL(kvm_resched);
1603 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1606 struct task_struct *task = NULL;
1609 pid = rcu_dereference(target->pid);
1611 task = get_pid_task(target->pid, PIDTYPE_PID);
1615 if (task->flags & PF_VCPU) {
1616 put_task_struct(task);
1619 if (yield_to(task, 1)) {
1620 put_task_struct(task);
1623 put_task_struct(task);
1626 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1628 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1630 * Helper that checks whether a VCPU is eligible for directed yield.
1631 * Most eligible candidate to yield is decided by following heuristics:
1633 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1634 * (preempted lock holder), indicated by @in_spin_loop.
1635 * Set at the beiginning and cleared at the end of interception/PLE handler.
1637 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1638 * chance last time (mostly it has become eligible now since we have probably
1639 * yielded to lockholder in last iteration. This is done by toggling
1640 * @dy_eligible each time a VCPU checked for eligibility.)
1642 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1643 * to preempted lock-holder could result in wrong VCPU selection and CPU
1644 * burning. Giving priority for a potential lock-holder increases lock
1647 * Since algorithm is based on heuristics, accessing another VCPU data without
1648 * locking does not harm. It may result in trying to yield to same VCPU, fail
1649 * and continue with next VCPU and so on.
1651 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1655 eligible = !vcpu->spin_loop.in_spin_loop ||
1656 (vcpu->spin_loop.in_spin_loop &&
1657 vcpu->spin_loop.dy_eligible);
1659 if (vcpu->spin_loop.in_spin_loop)
1660 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1665 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1667 struct kvm *kvm = me->kvm;
1668 struct kvm_vcpu *vcpu;
1669 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1674 kvm_vcpu_set_in_spin_loop(me, true);
1676 * We boost the priority of a VCPU that is runnable but not
1677 * currently running, because it got preempted by something
1678 * else and called schedule in __vcpu_run. Hopefully that
1679 * VCPU is holding the lock that we need and will release it.
1680 * We approximate round-robin by starting at the last boosted VCPU.
1682 for (pass = 0; pass < 2 && !yielded; pass++) {
1683 kvm_for_each_vcpu(i, vcpu, kvm) {
1684 if (!pass && i <= last_boosted_vcpu) {
1685 i = last_boosted_vcpu;
1687 } else if (pass && i > last_boosted_vcpu)
1691 if (waitqueue_active(&vcpu->wq))
1693 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1695 if (kvm_vcpu_yield_to(vcpu)) {
1696 kvm->last_boosted_vcpu = i;
1702 kvm_vcpu_set_in_spin_loop(me, false);
1704 /* Ensure vcpu is not eligible during next spinloop */
1705 kvm_vcpu_set_dy_eligible(me, false);
1707 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1709 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1711 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1714 if (vmf->pgoff == 0)
1715 page = virt_to_page(vcpu->run);
1717 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1718 page = virt_to_page(vcpu->arch.pio_data);
1720 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1721 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1722 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1725 return kvm_arch_vcpu_fault(vcpu, vmf);
1731 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1732 .fault = kvm_vcpu_fault,
1735 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1737 vma->vm_ops = &kvm_vcpu_vm_ops;
1741 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1743 struct kvm_vcpu *vcpu = filp->private_data;
1745 kvm_put_kvm(vcpu->kvm);
1749 static struct file_operations kvm_vcpu_fops = {
1750 .release = kvm_vcpu_release,
1751 .unlocked_ioctl = kvm_vcpu_ioctl,
1752 #ifdef CONFIG_COMPAT
1753 .compat_ioctl = kvm_vcpu_compat_ioctl,
1755 .mmap = kvm_vcpu_mmap,
1756 .llseek = noop_llseek,
1760 * Allocates an inode for the vcpu.
1762 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1764 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1768 * Creates some virtual cpus. Good luck creating more than one.
1770 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1773 struct kvm_vcpu *vcpu, *v;
1775 vcpu = kvm_arch_vcpu_create(kvm, id);
1777 return PTR_ERR(vcpu);
1779 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1781 r = kvm_arch_vcpu_setup(vcpu);
1785 mutex_lock(&kvm->lock);
1786 if (!kvm_vcpu_compatible(vcpu)) {
1788 goto unlock_vcpu_destroy;
1790 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1792 goto unlock_vcpu_destroy;
1795 kvm_for_each_vcpu(r, v, kvm)
1796 if (v->vcpu_id == id) {
1798 goto unlock_vcpu_destroy;
1801 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1803 /* Now it's all set up, let userspace reach it */
1805 r = create_vcpu_fd(vcpu);
1808 goto unlock_vcpu_destroy;
1811 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1813 atomic_inc(&kvm->online_vcpus);
1815 mutex_unlock(&kvm->lock);
1818 unlock_vcpu_destroy:
1819 mutex_unlock(&kvm->lock);
1821 kvm_arch_vcpu_destroy(vcpu);
1825 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1828 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1829 vcpu->sigset_active = 1;
1830 vcpu->sigset = *sigset;
1832 vcpu->sigset_active = 0;
1836 static long kvm_vcpu_ioctl(struct file *filp,
1837 unsigned int ioctl, unsigned long arg)
1839 struct kvm_vcpu *vcpu = filp->private_data;
1840 void __user *argp = (void __user *)arg;
1842 struct kvm_fpu *fpu = NULL;
1843 struct kvm_sregs *kvm_sregs = NULL;
1845 if (vcpu->kvm->mm != current->mm)
1848 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1850 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1851 * so vcpu_load() would break it.
1853 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1854 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1864 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1865 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1867 case KVM_GET_REGS: {
1868 struct kvm_regs *kvm_regs;
1871 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1874 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1878 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1885 case KVM_SET_REGS: {
1886 struct kvm_regs *kvm_regs;
1889 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1890 if (IS_ERR(kvm_regs)) {
1891 r = PTR_ERR(kvm_regs);
1894 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1902 case KVM_GET_SREGS: {
1903 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1907 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1911 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1916 case KVM_SET_SREGS: {
1917 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1918 if (IS_ERR(kvm_sregs)) {
1919 r = PTR_ERR(kvm_sregs);
1922 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1928 case KVM_GET_MP_STATE: {
1929 struct kvm_mp_state mp_state;
1931 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1935 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1940 case KVM_SET_MP_STATE: {
1941 struct kvm_mp_state mp_state;
1944 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1946 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1952 case KVM_TRANSLATE: {
1953 struct kvm_translation tr;
1956 if (copy_from_user(&tr, argp, sizeof tr))
1958 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1962 if (copy_to_user(argp, &tr, sizeof tr))
1967 case KVM_SET_GUEST_DEBUG: {
1968 struct kvm_guest_debug dbg;
1971 if (copy_from_user(&dbg, argp, sizeof dbg))
1973 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1979 case KVM_SET_SIGNAL_MASK: {
1980 struct kvm_signal_mask __user *sigmask_arg = argp;
1981 struct kvm_signal_mask kvm_sigmask;
1982 sigset_t sigset, *p;
1987 if (copy_from_user(&kvm_sigmask, argp,
1988 sizeof kvm_sigmask))
1991 if (kvm_sigmask.len != sizeof sigset)
1994 if (copy_from_user(&sigset, sigmask_arg->sigset,
1999 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2003 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2007 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2011 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2017 fpu = memdup_user(argp, sizeof(*fpu));
2022 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2029 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2038 #ifdef CONFIG_COMPAT
2039 static long kvm_vcpu_compat_ioctl(struct file *filp,
2040 unsigned int ioctl, unsigned long arg)
2042 struct kvm_vcpu *vcpu = filp->private_data;
2043 void __user *argp = compat_ptr(arg);
2046 if (vcpu->kvm->mm != current->mm)
2050 case KVM_SET_SIGNAL_MASK: {
2051 struct kvm_signal_mask __user *sigmask_arg = argp;
2052 struct kvm_signal_mask kvm_sigmask;
2053 compat_sigset_t csigset;
2058 if (copy_from_user(&kvm_sigmask, argp,
2059 sizeof kvm_sigmask))
2062 if (kvm_sigmask.len != sizeof csigset)
2065 if (copy_from_user(&csigset, sigmask_arg->sigset,
2069 sigset_from_compat(&sigset, &csigset);
2070 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2074 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2082 static long kvm_vm_ioctl(struct file *filp,
2083 unsigned int ioctl, unsigned long arg)
2085 struct kvm *kvm = filp->private_data;
2086 void __user *argp = (void __user *)arg;
2089 if (kvm->mm != current->mm)
2092 case KVM_CREATE_VCPU:
2093 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2097 case KVM_SET_USER_MEMORY_REGION: {
2098 struct kvm_userspace_memory_region kvm_userspace_mem;
2101 if (copy_from_user(&kvm_userspace_mem, argp,
2102 sizeof kvm_userspace_mem))
2105 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2110 case KVM_GET_DIRTY_LOG: {
2111 struct kvm_dirty_log log;
2114 if (copy_from_user(&log, argp, sizeof log))
2116 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2121 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2122 case KVM_REGISTER_COALESCED_MMIO: {
2123 struct kvm_coalesced_mmio_zone zone;
2125 if (copy_from_user(&zone, argp, sizeof zone))
2127 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2133 case KVM_UNREGISTER_COALESCED_MMIO: {
2134 struct kvm_coalesced_mmio_zone zone;
2136 if (copy_from_user(&zone, argp, sizeof zone))
2138 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2146 struct kvm_irqfd data;
2149 if (copy_from_user(&data, argp, sizeof data))
2151 r = kvm_irqfd(kvm, &data);
2154 case KVM_IOEVENTFD: {
2155 struct kvm_ioeventfd data;
2158 if (copy_from_user(&data, argp, sizeof data))
2160 r = kvm_ioeventfd(kvm, &data);
2163 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2164 case KVM_SET_BOOT_CPU_ID:
2166 mutex_lock(&kvm->lock);
2167 if (atomic_read(&kvm->online_vcpus) != 0)
2170 kvm->bsp_vcpu_id = arg;
2171 mutex_unlock(&kvm->lock);
2174 #ifdef CONFIG_HAVE_KVM_MSI
2175 case KVM_SIGNAL_MSI: {
2179 if (copy_from_user(&msi, argp, sizeof msi))
2181 r = kvm_send_userspace_msi(kvm, &msi);
2185 #ifdef __KVM_HAVE_IRQ_LINE
2186 case KVM_IRQ_LINE_STATUS:
2187 case KVM_IRQ_LINE: {
2188 struct kvm_irq_level irq_event;
2191 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2194 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2199 if (ioctl == KVM_IRQ_LINE_STATUS) {
2200 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2209 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2211 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2217 #ifdef CONFIG_COMPAT
2218 struct compat_kvm_dirty_log {
2222 compat_uptr_t dirty_bitmap; /* one bit per page */
2227 static long kvm_vm_compat_ioctl(struct file *filp,
2228 unsigned int ioctl, unsigned long arg)
2230 struct kvm *kvm = filp->private_data;
2233 if (kvm->mm != current->mm)
2236 case KVM_GET_DIRTY_LOG: {
2237 struct compat_kvm_dirty_log compat_log;
2238 struct kvm_dirty_log log;
2241 if (copy_from_user(&compat_log, (void __user *)arg,
2242 sizeof(compat_log)))
2244 log.slot = compat_log.slot;
2245 log.padding1 = compat_log.padding1;
2246 log.padding2 = compat_log.padding2;
2247 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2249 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2255 r = kvm_vm_ioctl(filp, ioctl, arg);
2263 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2265 struct page *page[1];
2268 gfn_t gfn = vmf->pgoff;
2269 struct kvm *kvm = vma->vm_file->private_data;
2271 addr = gfn_to_hva(kvm, gfn);
2272 if (kvm_is_error_hva(addr))
2273 return VM_FAULT_SIGBUS;
2275 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2277 if (unlikely(npages != 1))
2278 return VM_FAULT_SIGBUS;
2280 vmf->page = page[0];
2284 static const struct vm_operations_struct kvm_vm_vm_ops = {
2285 .fault = kvm_vm_fault,
2288 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2290 vma->vm_ops = &kvm_vm_vm_ops;
2294 static struct file_operations kvm_vm_fops = {
2295 .release = kvm_vm_release,
2296 .unlocked_ioctl = kvm_vm_ioctl,
2297 #ifdef CONFIG_COMPAT
2298 .compat_ioctl = kvm_vm_compat_ioctl,
2300 .mmap = kvm_vm_mmap,
2301 .llseek = noop_llseek,
2304 static int kvm_dev_ioctl_create_vm(unsigned long type)
2309 kvm = kvm_create_vm(type);
2311 return PTR_ERR(kvm);
2312 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2313 r = kvm_coalesced_mmio_init(kvm);
2319 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2326 static long kvm_dev_ioctl_check_extension_generic(long arg)
2329 case KVM_CAP_USER_MEMORY:
2330 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2331 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2332 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2333 case KVM_CAP_SET_BOOT_CPU_ID:
2335 case KVM_CAP_INTERNAL_ERROR_DATA:
2336 #ifdef CONFIG_HAVE_KVM_MSI
2337 case KVM_CAP_SIGNAL_MSI:
2340 #ifdef KVM_CAP_IRQ_ROUTING
2341 case KVM_CAP_IRQ_ROUTING:
2342 return KVM_MAX_IRQ_ROUTES;
2347 return kvm_dev_ioctl_check_extension(arg);
2350 static long kvm_dev_ioctl(struct file *filp,
2351 unsigned int ioctl, unsigned long arg)
2356 case KVM_GET_API_VERSION:
2360 r = KVM_API_VERSION;
2363 r = kvm_dev_ioctl_create_vm(arg);
2365 case KVM_CHECK_EXTENSION:
2366 r = kvm_dev_ioctl_check_extension_generic(arg);
2368 case KVM_GET_VCPU_MMAP_SIZE:
2372 r = PAGE_SIZE; /* struct kvm_run */
2374 r += PAGE_SIZE; /* pio data page */
2376 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2377 r += PAGE_SIZE; /* coalesced mmio ring page */
2380 case KVM_TRACE_ENABLE:
2381 case KVM_TRACE_PAUSE:
2382 case KVM_TRACE_DISABLE:
2386 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2392 static struct file_operations kvm_chardev_ops = {
2393 .unlocked_ioctl = kvm_dev_ioctl,
2394 .compat_ioctl = kvm_dev_ioctl,
2395 .llseek = noop_llseek,
2398 static struct miscdevice kvm_dev = {
2404 static void hardware_enable_nolock(void *junk)
2406 int cpu = raw_smp_processor_id();
2409 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2412 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2414 r = kvm_arch_hardware_enable(NULL);
2417 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2418 atomic_inc(&hardware_enable_failed);
2419 printk(KERN_INFO "kvm: enabling virtualization on "
2420 "CPU%d failed\n", cpu);
2424 static void hardware_enable(void *junk)
2426 raw_spin_lock(&kvm_lock);
2427 hardware_enable_nolock(junk);
2428 raw_spin_unlock(&kvm_lock);
2431 static void hardware_disable_nolock(void *junk)
2433 int cpu = raw_smp_processor_id();
2435 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2437 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2438 kvm_arch_hardware_disable(NULL);
2441 static void hardware_disable(void *junk)
2443 raw_spin_lock(&kvm_lock);
2444 hardware_disable_nolock(junk);
2445 raw_spin_unlock(&kvm_lock);
2448 static void hardware_disable_all_nolock(void)
2450 BUG_ON(!kvm_usage_count);
2453 if (!kvm_usage_count)
2454 on_each_cpu(hardware_disable_nolock, NULL, 1);
2457 static void hardware_disable_all(void)
2459 raw_spin_lock(&kvm_lock);
2460 hardware_disable_all_nolock();
2461 raw_spin_unlock(&kvm_lock);
2464 static int hardware_enable_all(void)
2468 raw_spin_lock(&kvm_lock);
2471 if (kvm_usage_count == 1) {
2472 atomic_set(&hardware_enable_failed, 0);
2473 on_each_cpu(hardware_enable_nolock, NULL, 1);
2475 if (atomic_read(&hardware_enable_failed)) {
2476 hardware_disable_all_nolock();
2481 raw_spin_unlock(&kvm_lock);
2486 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2491 if (!kvm_usage_count)
2494 val &= ~CPU_TASKS_FROZEN;
2497 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2499 hardware_disable(NULL);
2502 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2504 hardware_enable(NULL);
2511 asmlinkage void kvm_spurious_fault(void)
2513 /* Fault while not rebooting. We want the trace. */
2516 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2518 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2522 * Some (well, at least mine) BIOSes hang on reboot if
2525 * And Intel TXT required VMX off for all cpu when system shutdown.
2527 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2528 kvm_rebooting = true;
2529 on_each_cpu(hardware_disable_nolock, NULL, 1);
2533 static struct notifier_block kvm_reboot_notifier = {
2534 .notifier_call = kvm_reboot,
2538 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2542 for (i = 0; i < bus->dev_count; i++) {
2543 struct kvm_io_device *pos = bus->range[i].dev;
2545 kvm_iodevice_destructor(pos);
2550 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2552 const struct kvm_io_range *r1 = p1;
2553 const struct kvm_io_range *r2 = p2;
2555 if (r1->addr < r2->addr)
2557 if (r1->addr + r1->len > r2->addr + r2->len)
2562 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2563 gpa_t addr, int len)
2565 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2571 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2572 kvm_io_bus_sort_cmp, NULL);
2577 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2578 gpa_t addr, int len)
2580 struct kvm_io_range *range, key;
2583 key = (struct kvm_io_range) {
2588 range = bsearch(&key, bus->range, bus->dev_count,
2589 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2593 off = range - bus->range;
2595 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2601 /* kvm_io_bus_write - called under kvm->slots_lock */
2602 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2603 int len, const void *val)
2606 struct kvm_io_bus *bus;
2607 struct kvm_io_range range;
2609 range = (struct kvm_io_range) {
2614 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2615 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2619 while (idx < bus->dev_count &&
2620 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2621 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2629 /* kvm_io_bus_read - called under kvm->slots_lock */
2630 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2634 struct kvm_io_bus *bus;
2635 struct kvm_io_range range;
2637 range = (struct kvm_io_range) {
2642 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2643 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2647 while (idx < bus->dev_count &&
2648 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2649 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2657 /* Caller must hold slots_lock. */
2658 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2659 int len, struct kvm_io_device *dev)
2661 struct kvm_io_bus *new_bus, *bus;
2663 bus = kvm->buses[bus_idx];
2664 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2667 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2668 sizeof(struct kvm_io_range)), GFP_KERNEL);
2671 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2672 sizeof(struct kvm_io_range)));
2673 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2674 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2675 synchronize_srcu_expedited(&kvm->srcu);
2681 /* Caller must hold slots_lock. */
2682 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2683 struct kvm_io_device *dev)
2686 struct kvm_io_bus *new_bus, *bus;
2688 bus = kvm->buses[bus_idx];
2690 for (i = 0; i < bus->dev_count; i++)
2691 if (bus->range[i].dev == dev) {
2699 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2700 sizeof(struct kvm_io_range)), GFP_KERNEL);
2704 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2705 new_bus->dev_count--;
2706 memcpy(new_bus->range + i, bus->range + i + 1,
2707 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2709 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2710 synchronize_srcu_expedited(&kvm->srcu);
2715 static struct notifier_block kvm_cpu_notifier = {
2716 .notifier_call = kvm_cpu_hotplug,
2719 static int vm_stat_get(void *_offset, u64 *val)
2721 unsigned offset = (long)_offset;
2725 raw_spin_lock(&kvm_lock);
2726 list_for_each_entry(kvm, &vm_list, vm_list)
2727 *val += *(u32 *)((void *)kvm + offset);
2728 raw_spin_unlock(&kvm_lock);
2732 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2734 static int vcpu_stat_get(void *_offset, u64 *val)
2736 unsigned offset = (long)_offset;
2738 struct kvm_vcpu *vcpu;
2742 raw_spin_lock(&kvm_lock);
2743 list_for_each_entry(kvm, &vm_list, vm_list)
2744 kvm_for_each_vcpu(i, vcpu, kvm)
2745 *val += *(u32 *)((void *)vcpu + offset);
2747 raw_spin_unlock(&kvm_lock);
2751 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2753 static const struct file_operations *stat_fops[] = {
2754 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2755 [KVM_STAT_VM] = &vm_stat_fops,
2758 static int kvm_init_debug(void)
2761 struct kvm_stats_debugfs_item *p;
2763 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2764 if (kvm_debugfs_dir == NULL)
2767 for (p = debugfs_entries; p->name; ++p) {
2768 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2769 (void *)(long)p->offset,
2770 stat_fops[p->kind]);
2771 if (p->dentry == NULL)
2778 debugfs_remove_recursive(kvm_debugfs_dir);
2783 static void kvm_exit_debug(void)
2785 struct kvm_stats_debugfs_item *p;
2787 for (p = debugfs_entries; p->name; ++p)
2788 debugfs_remove(p->dentry);
2789 debugfs_remove(kvm_debugfs_dir);
2792 static int kvm_suspend(void)
2794 if (kvm_usage_count)
2795 hardware_disable_nolock(NULL);
2799 static void kvm_resume(void)
2801 if (kvm_usage_count) {
2802 WARN_ON(raw_spin_is_locked(&kvm_lock));
2803 hardware_enable_nolock(NULL);
2807 static struct syscore_ops kvm_syscore_ops = {
2808 .suspend = kvm_suspend,
2809 .resume = kvm_resume,
2813 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2815 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2818 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2820 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2822 kvm_arch_vcpu_load(vcpu, cpu);
2825 static void kvm_sched_out(struct preempt_notifier *pn,
2826 struct task_struct *next)
2828 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2830 kvm_arch_vcpu_put(vcpu);
2833 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2834 struct module *module)
2839 r = kvm_arch_init(opaque);
2843 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2848 r = kvm_arch_hardware_setup();
2852 for_each_online_cpu(cpu) {
2853 smp_call_function_single(cpu,
2854 kvm_arch_check_processor_compat,
2860 r = register_cpu_notifier(&kvm_cpu_notifier);
2863 register_reboot_notifier(&kvm_reboot_notifier);
2865 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2867 vcpu_align = __alignof__(struct kvm_vcpu);
2868 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2870 if (!kvm_vcpu_cache) {
2875 r = kvm_async_pf_init();
2879 kvm_chardev_ops.owner = module;
2880 kvm_vm_fops.owner = module;
2881 kvm_vcpu_fops.owner = module;
2883 r = misc_register(&kvm_dev);
2885 printk(KERN_ERR "kvm: misc device register failed\n");
2889 register_syscore_ops(&kvm_syscore_ops);
2891 kvm_preempt_ops.sched_in = kvm_sched_in;
2892 kvm_preempt_ops.sched_out = kvm_sched_out;
2894 r = kvm_init_debug();
2896 printk(KERN_ERR "kvm: create debugfs files failed\n");
2903 unregister_syscore_ops(&kvm_syscore_ops);
2905 kvm_async_pf_deinit();
2907 kmem_cache_destroy(kvm_vcpu_cache);
2909 unregister_reboot_notifier(&kvm_reboot_notifier);
2910 unregister_cpu_notifier(&kvm_cpu_notifier);
2913 kvm_arch_hardware_unsetup();
2915 free_cpumask_var(cpus_hardware_enabled);
2921 EXPORT_SYMBOL_GPL(kvm_init);
2926 misc_deregister(&kvm_dev);
2927 kmem_cache_destroy(kvm_vcpu_cache);
2928 kvm_async_pf_deinit();
2929 unregister_syscore_ops(&kvm_syscore_ops);
2930 unregister_reboot_notifier(&kvm_reboot_notifier);
2931 unregister_cpu_notifier(&kvm_cpu_notifier);
2932 on_each_cpu(hardware_disable_nolock, NULL, 1);
2933 kvm_arch_hardware_unsetup();
2935 free_cpumask_var(cpus_hardware_enabled);
2937 EXPORT_SYMBOL_GPL(kvm_exit);