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_all(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_all(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 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
685 #ifdef KVM_CAP_READONLY_MEM
686 valid_flags |= KVM_MEM_READONLY;
689 if (mem->flags & ~valid_flags)
696 * Allocate some memory and give it an address in the guest physical address
699 * Discontiguous memory is allowed, mostly for framebuffers.
701 * Must be called holding mmap_sem for write.
703 int __kvm_set_memory_region(struct kvm *kvm,
704 struct kvm_userspace_memory_region *mem,
709 unsigned long npages;
711 struct kvm_memory_slot *memslot;
712 struct kvm_memory_slot old, new;
713 struct kvm_memslots *slots, *old_memslots;
715 r = check_memory_region_flags(mem);
720 /* General sanity checks */
721 if (mem->memory_size & (PAGE_SIZE - 1))
723 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
725 /* We can read the guest memory with __xxx_user() later on. */
727 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
728 !access_ok(VERIFY_WRITE,
729 (void __user *)(unsigned long)mem->userspace_addr,
732 if (mem->slot >= KVM_MEM_SLOTS_NUM)
734 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
737 memslot = id_to_memslot(kvm->memslots, mem->slot);
738 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
739 npages = mem->memory_size >> PAGE_SHIFT;
742 if (npages > KVM_MEM_MAX_NR_PAGES)
746 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
748 new = old = *memslot;
751 new.base_gfn = base_gfn;
753 new.flags = mem->flags;
755 /* Disallow changing a memory slot's size. */
757 if (npages && old.npages && npages != old.npages)
760 /* Check for overlaps */
762 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
763 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
765 if (s == memslot || !s->npages)
767 if (!((base_gfn + npages <= s->base_gfn) ||
768 (base_gfn >= s->base_gfn + s->npages)))
772 /* Free page dirty bitmap if unneeded */
773 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
774 new.dirty_bitmap = NULL;
778 /* Allocate if a slot is being created */
779 if (npages && !old.npages) {
780 new.user_alloc = user_alloc;
781 new.userspace_addr = mem->userspace_addr;
783 if (kvm_arch_create_memslot(&new, npages))
787 /* Allocate page dirty bitmap if needed */
788 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
789 if (kvm_create_dirty_bitmap(&new) < 0)
791 /* destroy any largepage mappings for dirty tracking */
795 struct kvm_memory_slot *slot;
798 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
802 slot = id_to_memslot(slots, mem->slot);
803 slot->flags |= KVM_MEMSLOT_INVALID;
805 update_memslots(slots, NULL);
807 old_memslots = kvm->memslots;
808 rcu_assign_pointer(kvm->memslots, slots);
809 synchronize_srcu_expedited(&kvm->srcu);
810 /* From this point no new shadow pages pointing to a deleted
811 * memslot will be created.
813 * validation of sp->gfn happens in:
814 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
815 * - kvm_is_visible_gfn (mmu_check_roots)
817 kvm_arch_flush_shadow_memslot(kvm, slot);
821 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
825 /* map/unmap the pages in iommu page table */
827 r = kvm_iommu_map_pages(kvm, &new);
831 kvm_iommu_unmap_pages(kvm, &old);
834 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
839 /* actual memory is freed via old in kvm_free_physmem_slot below */
841 new.dirty_bitmap = NULL;
842 memset(&new.arch, 0, sizeof(new.arch));
845 update_memslots(slots, &new);
846 old_memslots = kvm->memslots;
847 rcu_assign_pointer(kvm->memslots, slots);
848 synchronize_srcu_expedited(&kvm->srcu);
850 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
853 * If the new memory slot is created, we need to clear all
856 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
857 kvm_arch_flush_shadow_all(kvm);
859 kvm_free_physmem_slot(&old, &new);
865 kvm_free_physmem_slot(&new, &old);
870 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
872 int kvm_set_memory_region(struct kvm *kvm,
873 struct kvm_userspace_memory_region *mem,
878 mutex_lock(&kvm->slots_lock);
879 r = __kvm_set_memory_region(kvm, mem, user_alloc);
880 mutex_unlock(&kvm->slots_lock);
883 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
885 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
887 kvm_userspace_memory_region *mem,
890 if (mem->slot >= KVM_MEMORY_SLOTS)
892 return kvm_set_memory_region(kvm, mem, user_alloc);
895 int kvm_get_dirty_log(struct kvm *kvm,
896 struct kvm_dirty_log *log, int *is_dirty)
898 struct kvm_memory_slot *memslot;
901 unsigned long any = 0;
904 if (log->slot >= KVM_MEMORY_SLOTS)
907 memslot = id_to_memslot(kvm->memslots, log->slot);
909 if (!memslot->dirty_bitmap)
912 n = kvm_dirty_bitmap_bytes(memslot);
914 for (i = 0; !any && i < n/sizeof(long); ++i)
915 any = memslot->dirty_bitmap[i];
918 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
929 bool kvm_largepages_enabled(void)
931 return largepages_enabled;
934 void kvm_disable_largepages(void)
936 largepages_enabled = false;
938 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
940 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
942 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
944 EXPORT_SYMBOL_GPL(gfn_to_memslot);
946 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
948 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
950 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
951 memslot->flags & KVM_MEMSLOT_INVALID)
956 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
958 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
960 struct vm_area_struct *vma;
961 unsigned long addr, size;
965 addr = gfn_to_hva(kvm, gfn);
966 if (kvm_is_error_hva(addr))
969 down_read(¤t->mm->mmap_sem);
970 vma = find_vma(current->mm, addr);
974 size = vma_kernel_pagesize(vma);
977 up_read(¤t->mm->mmap_sem);
982 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
984 return slot->flags & KVM_MEM_READONLY;
987 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
988 gfn_t *nr_pages, bool write)
990 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
991 return KVM_HVA_ERR_BAD;
993 if (memslot_is_readonly(slot) && write)
994 return KVM_HVA_ERR_RO_BAD;
997 *nr_pages = slot->npages - (gfn - slot->base_gfn);
999 return __gfn_to_hva_memslot(slot, gfn);
1002 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1005 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1008 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1011 return gfn_to_hva_many(slot, gfn, NULL);
1013 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1015 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1017 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1019 EXPORT_SYMBOL_GPL(gfn_to_hva);
1022 * The hva returned by this function is only allowed to be read.
1023 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1025 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1027 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1030 static int kvm_read_hva(void *data, void __user *hva, int len)
1032 return __copy_from_user(data, hva, len);
1035 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1037 return __copy_from_user_inatomic(data, hva, len);
1040 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1041 unsigned long start, int write, struct page **page)
1043 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1046 flags |= FOLL_WRITE;
1048 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1051 static inline int check_user_page_hwpoison(unsigned long addr)
1053 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1055 rc = __get_user_pages(current, current->mm, addr, 1,
1056 flags, NULL, NULL, NULL);
1057 return rc == -EHWPOISON;
1061 * The atomic path to get the writable pfn which will be stored in @pfn,
1062 * true indicates success, otherwise false is returned.
1064 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1065 bool write_fault, bool *writable, pfn_t *pfn)
1067 struct page *page[1];
1070 if (!(async || atomic))
1074 * Fast pin a writable pfn only if it is a write fault request
1075 * or the caller allows to map a writable pfn for a read fault
1078 if (!(write_fault || writable))
1081 npages = __get_user_pages_fast(addr, 1, 1, page);
1083 *pfn = page_to_pfn(page[0]);
1094 * The slow path to get the pfn of the specified host virtual address,
1095 * 1 indicates success, -errno is returned if error is detected.
1097 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1098 bool *writable, pfn_t *pfn)
1100 struct page *page[1];
1106 *writable = write_fault;
1109 down_read(¤t->mm->mmap_sem);
1110 npages = get_user_page_nowait(current, current->mm,
1111 addr, write_fault, page);
1112 up_read(¤t->mm->mmap_sem);
1114 npages = get_user_pages_fast(addr, 1, write_fault,
1119 /* map read fault as writable if possible */
1120 if (unlikely(!write_fault) && writable) {
1121 struct page *wpage[1];
1123 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1132 *pfn = page_to_pfn(page[0]);
1136 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1138 if (unlikely(!(vma->vm_flags & VM_READ)))
1141 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1148 * Pin guest page in memory and return its pfn.
1149 * @addr: host virtual address which maps memory to the guest
1150 * @atomic: whether this function can sleep
1151 * @async: whether this function need to wait IO complete if the
1152 * host page is not in the memory
1153 * @write_fault: whether we should get a writable host page
1154 * @writable: whether it allows to map a writable host page for !@write_fault
1156 * The function will map a writable host page for these two cases:
1157 * 1): @write_fault = true
1158 * 2): @write_fault = false && @writable, @writable will tell the caller
1159 * whether the mapping is writable.
1161 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1162 bool write_fault, bool *writable)
1164 struct vm_area_struct *vma;
1168 /* we can do it either atomically or asynchronously, not both */
1169 BUG_ON(atomic && async);
1171 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1175 return KVM_PFN_ERR_FAULT;
1177 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1181 down_read(¤t->mm->mmap_sem);
1182 if (npages == -EHWPOISON ||
1183 (!async && check_user_page_hwpoison(addr))) {
1184 pfn = KVM_PFN_ERR_HWPOISON;
1188 vma = find_vma_intersection(current->mm, addr, addr + 1);
1191 pfn = KVM_PFN_ERR_FAULT;
1192 else if ((vma->vm_flags & VM_PFNMAP)) {
1193 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1195 BUG_ON(!kvm_is_mmio_pfn(pfn));
1197 if (async && vma_is_valid(vma, write_fault))
1199 pfn = KVM_PFN_ERR_FAULT;
1202 up_read(¤t->mm->mmap_sem);
1207 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1208 bool *async, bool write_fault, bool *writable)
1210 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1212 if (addr == KVM_HVA_ERR_RO_BAD)
1213 return KVM_PFN_ERR_RO_FAULT;
1215 if (kvm_is_error_hva(addr))
1216 return KVM_PFN_ERR_BAD;
1218 /* Do not map writable pfn in the readonly memslot. */
1219 if (writable && memslot_is_readonly(slot)) {
1224 return hva_to_pfn(addr, atomic, async, write_fault,
1228 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1229 bool write_fault, bool *writable)
1231 struct kvm_memory_slot *slot;
1236 slot = gfn_to_memslot(kvm, gfn);
1238 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1242 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1244 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1246 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1248 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1249 bool write_fault, bool *writable)
1251 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1253 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1255 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1257 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1259 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1261 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1264 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1266 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1268 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1270 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1273 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1275 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1277 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1279 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1285 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1286 if (kvm_is_error_hva(addr))
1289 if (entry < nr_pages)
1292 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1294 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1296 static struct page *kvm_pfn_to_page(pfn_t pfn)
1298 if (is_error_pfn(pfn))
1299 return KVM_ERR_PTR_BAD_PAGE;
1301 if (kvm_is_mmio_pfn(pfn)) {
1303 return KVM_ERR_PTR_BAD_PAGE;
1306 return pfn_to_page(pfn);
1309 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1313 pfn = gfn_to_pfn(kvm, gfn);
1315 return kvm_pfn_to_page(pfn);
1318 EXPORT_SYMBOL_GPL(gfn_to_page);
1320 void kvm_release_page_clean(struct page *page)
1322 WARN_ON(is_error_page(page));
1324 kvm_release_pfn_clean(page_to_pfn(page));
1326 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1328 void kvm_release_pfn_clean(pfn_t pfn)
1330 WARN_ON(is_error_pfn(pfn));
1332 if (!kvm_is_mmio_pfn(pfn))
1333 put_page(pfn_to_page(pfn));
1335 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1337 void kvm_release_page_dirty(struct page *page)
1339 WARN_ON(is_error_page(page));
1341 kvm_release_pfn_dirty(page_to_pfn(page));
1343 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1345 void kvm_release_pfn_dirty(pfn_t pfn)
1347 kvm_set_pfn_dirty(pfn);
1348 kvm_release_pfn_clean(pfn);
1350 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1352 void kvm_set_page_dirty(struct page *page)
1354 kvm_set_pfn_dirty(page_to_pfn(page));
1356 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1358 void kvm_set_pfn_dirty(pfn_t pfn)
1360 if (!kvm_is_mmio_pfn(pfn)) {
1361 struct page *page = pfn_to_page(pfn);
1362 if (!PageReserved(page))
1366 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1368 void kvm_set_pfn_accessed(pfn_t pfn)
1370 if (!kvm_is_mmio_pfn(pfn))
1371 mark_page_accessed(pfn_to_page(pfn));
1373 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1375 void kvm_get_pfn(pfn_t pfn)
1377 if (!kvm_is_mmio_pfn(pfn))
1378 get_page(pfn_to_page(pfn));
1380 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1382 static int next_segment(unsigned long len, int offset)
1384 if (len > PAGE_SIZE - offset)
1385 return PAGE_SIZE - offset;
1390 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1396 addr = gfn_to_hva_read(kvm, gfn);
1397 if (kvm_is_error_hva(addr))
1399 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1404 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1406 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1408 gfn_t gfn = gpa >> PAGE_SHIFT;
1410 int offset = offset_in_page(gpa);
1413 while ((seg = next_segment(len, offset)) != 0) {
1414 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1424 EXPORT_SYMBOL_GPL(kvm_read_guest);
1426 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1431 gfn_t gfn = gpa >> PAGE_SHIFT;
1432 int offset = offset_in_page(gpa);
1434 addr = gfn_to_hva_read(kvm, gfn);
1435 if (kvm_is_error_hva(addr))
1437 pagefault_disable();
1438 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1444 EXPORT_SYMBOL(kvm_read_guest_atomic);
1446 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1447 int offset, int len)
1452 addr = gfn_to_hva(kvm, gfn);
1453 if (kvm_is_error_hva(addr))
1455 r = __copy_to_user((void __user *)addr + offset, data, len);
1458 mark_page_dirty(kvm, gfn);
1461 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1463 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1466 gfn_t gfn = gpa >> PAGE_SHIFT;
1468 int offset = offset_in_page(gpa);
1471 while ((seg = next_segment(len, offset)) != 0) {
1472 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1483 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1486 struct kvm_memslots *slots = kvm_memslots(kvm);
1487 int offset = offset_in_page(gpa);
1488 gfn_t gfn = gpa >> PAGE_SHIFT;
1491 ghc->generation = slots->generation;
1492 ghc->memslot = gfn_to_memslot(kvm, gfn);
1493 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1494 if (!kvm_is_error_hva(ghc->hva))
1501 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1503 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1504 void *data, unsigned long len)
1506 struct kvm_memslots *slots = kvm_memslots(kvm);
1509 if (slots->generation != ghc->generation)
1510 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1512 if (kvm_is_error_hva(ghc->hva))
1515 r = __copy_to_user((void __user *)ghc->hva, data, len);
1518 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1522 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1524 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1525 void *data, unsigned long len)
1527 struct kvm_memslots *slots = kvm_memslots(kvm);
1530 if (slots->generation != ghc->generation)
1531 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1533 if (kvm_is_error_hva(ghc->hva))
1536 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1542 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1544 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1546 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1549 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1551 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1553 gfn_t gfn = gpa >> PAGE_SHIFT;
1555 int offset = offset_in_page(gpa);
1558 while ((seg = next_segment(len, offset)) != 0) {
1559 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1568 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1570 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1573 if (memslot && memslot->dirty_bitmap) {
1574 unsigned long rel_gfn = gfn - memslot->base_gfn;
1576 /* TODO: introduce set_bit_le() and use it */
1577 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1581 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1583 struct kvm_memory_slot *memslot;
1585 memslot = gfn_to_memslot(kvm, gfn);
1586 mark_page_dirty_in_slot(kvm, memslot, gfn);
1590 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1592 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1597 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1599 if (kvm_arch_vcpu_runnable(vcpu)) {
1600 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1603 if (kvm_cpu_has_pending_timer(vcpu))
1605 if (signal_pending(current))
1611 finish_wait(&vcpu->wq, &wait);
1616 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1618 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1621 int cpu = vcpu->cpu;
1622 wait_queue_head_t *wqp;
1624 wqp = kvm_arch_vcpu_wq(vcpu);
1625 if (waitqueue_active(wqp)) {
1626 wake_up_interruptible(wqp);
1627 ++vcpu->stat.halt_wakeup;
1631 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1632 if (kvm_arch_vcpu_should_kick(vcpu))
1633 smp_send_reschedule(cpu);
1636 #endif /* !CONFIG_S390 */
1638 void kvm_resched(struct kvm_vcpu *vcpu)
1640 if (!need_resched())
1644 EXPORT_SYMBOL_GPL(kvm_resched);
1646 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1649 struct task_struct *task = NULL;
1652 pid = rcu_dereference(target->pid);
1654 task = get_pid_task(target->pid, PIDTYPE_PID);
1658 if (task->flags & PF_VCPU) {
1659 put_task_struct(task);
1662 if (yield_to(task, 1)) {
1663 put_task_struct(task);
1666 put_task_struct(task);
1669 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1671 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1673 * Helper that checks whether a VCPU is eligible for directed yield.
1674 * Most eligible candidate to yield is decided by following heuristics:
1676 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1677 * (preempted lock holder), indicated by @in_spin_loop.
1678 * Set at the beiginning and cleared at the end of interception/PLE handler.
1680 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1681 * chance last time (mostly it has become eligible now since we have probably
1682 * yielded to lockholder in last iteration. This is done by toggling
1683 * @dy_eligible each time a VCPU checked for eligibility.)
1685 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1686 * to preempted lock-holder could result in wrong VCPU selection and CPU
1687 * burning. Giving priority for a potential lock-holder increases lock
1690 * Since algorithm is based on heuristics, accessing another VCPU data without
1691 * locking does not harm. It may result in trying to yield to same VCPU, fail
1692 * and continue with next VCPU and so on.
1694 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1698 eligible = !vcpu->spin_loop.in_spin_loop ||
1699 (vcpu->spin_loop.in_spin_loop &&
1700 vcpu->spin_loop.dy_eligible);
1702 if (vcpu->spin_loop.in_spin_loop)
1703 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1708 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1710 struct kvm *kvm = me->kvm;
1711 struct kvm_vcpu *vcpu;
1712 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1717 kvm_vcpu_set_in_spin_loop(me, true);
1719 * We boost the priority of a VCPU that is runnable but not
1720 * currently running, because it got preempted by something
1721 * else and called schedule in __vcpu_run. Hopefully that
1722 * VCPU is holding the lock that we need and will release it.
1723 * We approximate round-robin by starting at the last boosted VCPU.
1725 for (pass = 0; pass < 2 && !yielded; pass++) {
1726 kvm_for_each_vcpu(i, vcpu, kvm) {
1727 if (!pass && i <= last_boosted_vcpu) {
1728 i = last_boosted_vcpu;
1730 } else if (pass && i > last_boosted_vcpu)
1734 if (waitqueue_active(&vcpu->wq))
1736 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1738 if (kvm_vcpu_yield_to(vcpu)) {
1739 kvm->last_boosted_vcpu = i;
1745 kvm_vcpu_set_in_spin_loop(me, false);
1747 /* Ensure vcpu is not eligible during next spinloop */
1748 kvm_vcpu_set_dy_eligible(me, false);
1750 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1752 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1754 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1757 if (vmf->pgoff == 0)
1758 page = virt_to_page(vcpu->run);
1760 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1761 page = virt_to_page(vcpu->arch.pio_data);
1763 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1764 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1765 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1768 return kvm_arch_vcpu_fault(vcpu, vmf);
1774 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1775 .fault = kvm_vcpu_fault,
1778 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1780 vma->vm_ops = &kvm_vcpu_vm_ops;
1784 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1786 struct kvm_vcpu *vcpu = filp->private_data;
1788 kvm_put_kvm(vcpu->kvm);
1792 static struct file_operations kvm_vcpu_fops = {
1793 .release = kvm_vcpu_release,
1794 .unlocked_ioctl = kvm_vcpu_ioctl,
1795 #ifdef CONFIG_COMPAT
1796 .compat_ioctl = kvm_vcpu_compat_ioctl,
1798 .mmap = kvm_vcpu_mmap,
1799 .llseek = noop_llseek,
1803 * Allocates an inode for the vcpu.
1805 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1807 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1811 * Creates some virtual cpus. Good luck creating more than one.
1813 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1816 struct kvm_vcpu *vcpu, *v;
1818 vcpu = kvm_arch_vcpu_create(kvm, id);
1820 return PTR_ERR(vcpu);
1822 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1824 r = kvm_arch_vcpu_setup(vcpu);
1828 mutex_lock(&kvm->lock);
1829 if (!kvm_vcpu_compatible(vcpu)) {
1831 goto unlock_vcpu_destroy;
1833 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1835 goto unlock_vcpu_destroy;
1838 kvm_for_each_vcpu(r, v, kvm)
1839 if (v->vcpu_id == id) {
1841 goto unlock_vcpu_destroy;
1844 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1846 /* Now it's all set up, let userspace reach it */
1848 r = create_vcpu_fd(vcpu);
1851 goto unlock_vcpu_destroy;
1854 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1856 atomic_inc(&kvm->online_vcpus);
1858 mutex_unlock(&kvm->lock);
1861 unlock_vcpu_destroy:
1862 mutex_unlock(&kvm->lock);
1864 kvm_arch_vcpu_destroy(vcpu);
1868 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1871 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1872 vcpu->sigset_active = 1;
1873 vcpu->sigset = *sigset;
1875 vcpu->sigset_active = 0;
1879 static long kvm_vcpu_ioctl(struct file *filp,
1880 unsigned int ioctl, unsigned long arg)
1882 struct kvm_vcpu *vcpu = filp->private_data;
1883 void __user *argp = (void __user *)arg;
1885 struct kvm_fpu *fpu = NULL;
1886 struct kvm_sregs *kvm_sregs = NULL;
1888 if (vcpu->kvm->mm != current->mm)
1891 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1893 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1894 * so vcpu_load() would break it.
1896 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1897 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1907 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1908 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1910 case KVM_GET_REGS: {
1911 struct kvm_regs *kvm_regs;
1914 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1917 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1921 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1928 case KVM_SET_REGS: {
1929 struct kvm_regs *kvm_regs;
1932 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1933 if (IS_ERR(kvm_regs)) {
1934 r = PTR_ERR(kvm_regs);
1937 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1945 case KVM_GET_SREGS: {
1946 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1950 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1954 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1959 case KVM_SET_SREGS: {
1960 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1961 if (IS_ERR(kvm_sregs)) {
1962 r = PTR_ERR(kvm_sregs);
1965 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1971 case KVM_GET_MP_STATE: {
1972 struct kvm_mp_state mp_state;
1974 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1978 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1983 case KVM_SET_MP_STATE: {
1984 struct kvm_mp_state mp_state;
1987 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1989 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1995 case KVM_TRANSLATE: {
1996 struct kvm_translation tr;
1999 if (copy_from_user(&tr, argp, sizeof tr))
2001 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2005 if (copy_to_user(argp, &tr, sizeof tr))
2010 case KVM_SET_GUEST_DEBUG: {
2011 struct kvm_guest_debug dbg;
2014 if (copy_from_user(&dbg, argp, sizeof dbg))
2016 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2022 case KVM_SET_SIGNAL_MASK: {
2023 struct kvm_signal_mask __user *sigmask_arg = argp;
2024 struct kvm_signal_mask kvm_sigmask;
2025 sigset_t sigset, *p;
2030 if (copy_from_user(&kvm_sigmask, argp,
2031 sizeof kvm_sigmask))
2034 if (kvm_sigmask.len != sizeof sigset)
2037 if (copy_from_user(&sigset, sigmask_arg->sigset,
2042 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2046 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2050 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2054 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2060 fpu = memdup_user(argp, sizeof(*fpu));
2065 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2072 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2081 #ifdef CONFIG_COMPAT
2082 static long kvm_vcpu_compat_ioctl(struct file *filp,
2083 unsigned int ioctl, unsigned long arg)
2085 struct kvm_vcpu *vcpu = filp->private_data;
2086 void __user *argp = compat_ptr(arg);
2089 if (vcpu->kvm->mm != current->mm)
2093 case KVM_SET_SIGNAL_MASK: {
2094 struct kvm_signal_mask __user *sigmask_arg = argp;
2095 struct kvm_signal_mask kvm_sigmask;
2096 compat_sigset_t csigset;
2101 if (copy_from_user(&kvm_sigmask, argp,
2102 sizeof kvm_sigmask))
2105 if (kvm_sigmask.len != sizeof csigset)
2108 if (copy_from_user(&csigset, sigmask_arg->sigset,
2112 sigset_from_compat(&sigset, &csigset);
2113 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2117 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2125 static long kvm_vm_ioctl(struct file *filp,
2126 unsigned int ioctl, unsigned long arg)
2128 struct kvm *kvm = filp->private_data;
2129 void __user *argp = (void __user *)arg;
2132 if (kvm->mm != current->mm)
2135 case KVM_CREATE_VCPU:
2136 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2140 case KVM_SET_USER_MEMORY_REGION: {
2141 struct kvm_userspace_memory_region kvm_userspace_mem;
2144 if (copy_from_user(&kvm_userspace_mem, argp,
2145 sizeof kvm_userspace_mem))
2148 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2153 case KVM_GET_DIRTY_LOG: {
2154 struct kvm_dirty_log log;
2157 if (copy_from_user(&log, argp, sizeof log))
2159 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2164 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2165 case KVM_REGISTER_COALESCED_MMIO: {
2166 struct kvm_coalesced_mmio_zone zone;
2168 if (copy_from_user(&zone, argp, sizeof zone))
2170 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2176 case KVM_UNREGISTER_COALESCED_MMIO: {
2177 struct kvm_coalesced_mmio_zone zone;
2179 if (copy_from_user(&zone, argp, sizeof zone))
2181 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2189 struct kvm_irqfd data;
2192 if (copy_from_user(&data, argp, sizeof data))
2194 r = kvm_irqfd(kvm, &data);
2197 case KVM_IOEVENTFD: {
2198 struct kvm_ioeventfd data;
2201 if (copy_from_user(&data, argp, sizeof data))
2203 r = kvm_ioeventfd(kvm, &data);
2206 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2207 case KVM_SET_BOOT_CPU_ID:
2209 mutex_lock(&kvm->lock);
2210 if (atomic_read(&kvm->online_vcpus) != 0)
2213 kvm->bsp_vcpu_id = arg;
2214 mutex_unlock(&kvm->lock);
2217 #ifdef CONFIG_HAVE_KVM_MSI
2218 case KVM_SIGNAL_MSI: {
2222 if (copy_from_user(&msi, argp, sizeof msi))
2224 r = kvm_send_userspace_msi(kvm, &msi);
2228 #ifdef __KVM_HAVE_IRQ_LINE
2229 case KVM_IRQ_LINE_STATUS:
2230 case KVM_IRQ_LINE: {
2231 struct kvm_irq_level irq_event;
2234 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2237 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2242 if (ioctl == KVM_IRQ_LINE_STATUS) {
2243 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2252 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2254 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2260 #ifdef CONFIG_COMPAT
2261 struct compat_kvm_dirty_log {
2265 compat_uptr_t dirty_bitmap; /* one bit per page */
2270 static long kvm_vm_compat_ioctl(struct file *filp,
2271 unsigned int ioctl, unsigned long arg)
2273 struct kvm *kvm = filp->private_data;
2276 if (kvm->mm != current->mm)
2279 case KVM_GET_DIRTY_LOG: {
2280 struct compat_kvm_dirty_log compat_log;
2281 struct kvm_dirty_log log;
2284 if (copy_from_user(&compat_log, (void __user *)arg,
2285 sizeof(compat_log)))
2287 log.slot = compat_log.slot;
2288 log.padding1 = compat_log.padding1;
2289 log.padding2 = compat_log.padding2;
2290 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2292 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2298 r = kvm_vm_ioctl(filp, ioctl, arg);
2306 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2308 struct page *page[1];
2311 gfn_t gfn = vmf->pgoff;
2312 struct kvm *kvm = vma->vm_file->private_data;
2314 addr = gfn_to_hva(kvm, gfn);
2315 if (kvm_is_error_hva(addr))
2316 return VM_FAULT_SIGBUS;
2318 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2320 if (unlikely(npages != 1))
2321 return VM_FAULT_SIGBUS;
2323 vmf->page = page[0];
2327 static const struct vm_operations_struct kvm_vm_vm_ops = {
2328 .fault = kvm_vm_fault,
2331 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2333 vma->vm_ops = &kvm_vm_vm_ops;
2337 static struct file_operations kvm_vm_fops = {
2338 .release = kvm_vm_release,
2339 .unlocked_ioctl = kvm_vm_ioctl,
2340 #ifdef CONFIG_COMPAT
2341 .compat_ioctl = kvm_vm_compat_ioctl,
2343 .mmap = kvm_vm_mmap,
2344 .llseek = noop_llseek,
2347 static int kvm_dev_ioctl_create_vm(unsigned long type)
2352 kvm = kvm_create_vm(type);
2354 return PTR_ERR(kvm);
2355 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2356 r = kvm_coalesced_mmio_init(kvm);
2362 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2369 static long kvm_dev_ioctl_check_extension_generic(long arg)
2372 case KVM_CAP_USER_MEMORY:
2373 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2374 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2375 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2376 case KVM_CAP_SET_BOOT_CPU_ID:
2378 case KVM_CAP_INTERNAL_ERROR_DATA:
2379 #ifdef CONFIG_HAVE_KVM_MSI
2380 case KVM_CAP_SIGNAL_MSI:
2383 #ifdef KVM_CAP_IRQ_ROUTING
2384 case KVM_CAP_IRQ_ROUTING:
2385 return KVM_MAX_IRQ_ROUTES;
2390 return kvm_dev_ioctl_check_extension(arg);
2393 static long kvm_dev_ioctl(struct file *filp,
2394 unsigned int ioctl, unsigned long arg)
2399 case KVM_GET_API_VERSION:
2403 r = KVM_API_VERSION;
2406 r = kvm_dev_ioctl_create_vm(arg);
2408 case KVM_CHECK_EXTENSION:
2409 r = kvm_dev_ioctl_check_extension_generic(arg);
2411 case KVM_GET_VCPU_MMAP_SIZE:
2415 r = PAGE_SIZE; /* struct kvm_run */
2417 r += PAGE_SIZE; /* pio data page */
2419 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2420 r += PAGE_SIZE; /* coalesced mmio ring page */
2423 case KVM_TRACE_ENABLE:
2424 case KVM_TRACE_PAUSE:
2425 case KVM_TRACE_DISABLE:
2429 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2435 static struct file_operations kvm_chardev_ops = {
2436 .unlocked_ioctl = kvm_dev_ioctl,
2437 .compat_ioctl = kvm_dev_ioctl,
2438 .llseek = noop_llseek,
2441 static struct miscdevice kvm_dev = {
2447 static void hardware_enable_nolock(void *junk)
2449 int cpu = raw_smp_processor_id();
2452 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2455 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2457 r = kvm_arch_hardware_enable(NULL);
2460 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2461 atomic_inc(&hardware_enable_failed);
2462 printk(KERN_INFO "kvm: enabling virtualization on "
2463 "CPU%d failed\n", cpu);
2467 static void hardware_enable(void *junk)
2469 raw_spin_lock(&kvm_lock);
2470 hardware_enable_nolock(junk);
2471 raw_spin_unlock(&kvm_lock);
2474 static void hardware_disable_nolock(void *junk)
2476 int cpu = raw_smp_processor_id();
2478 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2480 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2481 kvm_arch_hardware_disable(NULL);
2484 static void hardware_disable(void *junk)
2486 raw_spin_lock(&kvm_lock);
2487 hardware_disable_nolock(junk);
2488 raw_spin_unlock(&kvm_lock);
2491 static void hardware_disable_all_nolock(void)
2493 BUG_ON(!kvm_usage_count);
2496 if (!kvm_usage_count)
2497 on_each_cpu(hardware_disable_nolock, NULL, 1);
2500 static void hardware_disable_all(void)
2502 raw_spin_lock(&kvm_lock);
2503 hardware_disable_all_nolock();
2504 raw_spin_unlock(&kvm_lock);
2507 static int hardware_enable_all(void)
2511 raw_spin_lock(&kvm_lock);
2514 if (kvm_usage_count == 1) {
2515 atomic_set(&hardware_enable_failed, 0);
2516 on_each_cpu(hardware_enable_nolock, NULL, 1);
2518 if (atomic_read(&hardware_enable_failed)) {
2519 hardware_disable_all_nolock();
2524 raw_spin_unlock(&kvm_lock);
2529 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2534 if (!kvm_usage_count)
2537 val &= ~CPU_TASKS_FROZEN;
2540 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2542 hardware_disable(NULL);
2545 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2547 hardware_enable(NULL);
2554 asmlinkage void kvm_spurious_fault(void)
2556 /* Fault while not rebooting. We want the trace. */
2559 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2561 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2565 * Some (well, at least mine) BIOSes hang on reboot if
2568 * And Intel TXT required VMX off for all cpu when system shutdown.
2570 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2571 kvm_rebooting = true;
2572 on_each_cpu(hardware_disable_nolock, NULL, 1);
2576 static struct notifier_block kvm_reboot_notifier = {
2577 .notifier_call = kvm_reboot,
2581 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2585 for (i = 0; i < bus->dev_count; i++) {
2586 struct kvm_io_device *pos = bus->range[i].dev;
2588 kvm_iodevice_destructor(pos);
2593 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2595 const struct kvm_io_range *r1 = p1;
2596 const struct kvm_io_range *r2 = p2;
2598 if (r1->addr < r2->addr)
2600 if (r1->addr + r1->len > r2->addr + r2->len)
2605 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2606 gpa_t addr, int len)
2608 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2614 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2615 kvm_io_bus_sort_cmp, NULL);
2620 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2621 gpa_t addr, int len)
2623 struct kvm_io_range *range, key;
2626 key = (struct kvm_io_range) {
2631 range = bsearch(&key, bus->range, bus->dev_count,
2632 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2636 off = range - bus->range;
2638 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2644 /* kvm_io_bus_write - called under kvm->slots_lock */
2645 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2646 int len, const void *val)
2649 struct kvm_io_bus *bus;
2650 struct kvm_io_range range;
2652 range = (struct kvm_io_range) {
2657 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2658 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2662 while (idx < bus->dev_count &&
2663 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2664 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2672 /* kvm_io_bus_read - called under kvm->slots_lock */
2673 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2677 struct kvm_io_bus *bus;
2678 struct kvm_io_range range;
2680 range = (struct kvm_io_range) {
2685 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2686 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2690 while (idx < bus->dev_count &&
2691 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2692 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2700 /* Caller must hold slots_lock. */
2701 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2702 int len, struct kvm_io_device *dev)
2704 struct kvm_io_bus *new_bus, *bus;
2706 bus = kvm->buses[bus_idx];
2707 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2710 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2711 sizeof(struct kvm_io_range)), GFP_KERNEL);
2714 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2715 sizeof(struct kvm_io_range)));
2716 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2717 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2718 synchronize_srcu_expedited(&kvm->srcu);
2724 /* Caller must hold slots_lock. */
2725 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2726 struct kvm_io_device *dev)
2729 struct kvm_io_bus *new_bus, *bus;
2731 bus = kvm->buses[bus_idx];
2733 for (i = 0; i < bus->dev_count; i++)
2734 if (bus->range[i].dev == dev) {
2742 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2743 sizeof(struct kvm_io_range)), GFP_KERNEL);
2747 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2748 new_bus->dev_count--;
2749 memcpy(new_bus->range + i, bus->range + i + 1,
2750 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2752 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2753 synchronize_srcu_expedited(&kvm->srcu);
2758 static struct notifier_block kvm_cpu_notifier = {
2759 .notifier_call = kvm_cpu_hotplug,
2762 static int vm_stat_get(void *_offset, u64 *val)
2764 unsigned offset = (long)_offset;
2768 raw_spin_lock(&kvm_lock);
2769 list_for_each_entry(kvm, &vm_list, vm_list)
2770 *val += *(u32 *)((void *)kvm + offset);
2771 raw_spin_unlock(&kvm_lock);
2775 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2777 static int vcpu_stat_get(void *_offset, u64 *val)
2779 unsigned offset = (long)_offset;
2781 struct kvm_vcpu *vcpu;
2785 raw_spin_lock(&kvm_lock);
2786 list_for_each_entry(kvm, &vm_list, vm_list)
2787 kvm_for_each_vcpu(i, vcpu, kvm)
2788 *val += *(u32 *)((void *)vcpu + offset);
2790 raw_spin_unlock(&kvm_lock);
2794 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2796 static const struct file_operations *stat_fops[] = {
2797 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2798 [KVM_STAT_VM] = &vm_stat_fops,
2801 static int kvm_init_debug(void)
2804 struct kvm_stats_debugfs_item *p;
2806 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2807 if (kvm_debugfs_dir == NULL)
2810 for (p = debugfs_entries; p->name; ++p) {
2811 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2812 (void *)(long)p->offset,
2813 stat_fops[p->kind]);
2814 if (p->dentry == NULL)
2821 debugfs_remove_recursive(kvm_debugfs_dir);
2826 static void kvm_exit_debug(void)
2828 struct kvm_stats_debugfs_item *p;
2830 for (p = debugfs_entries; p->name; ++p)
2831 debugfs_remove(p->dentry);
2832 debugfs_remove(kvm_debugfs_dir);
2835 static int kvm_suspend(void)
2837 if (kvm_usage_count)
2838 hardware_disable_nolock(NULL);
2842 static void kvm_resume(void)
2844 if (kvm_usage_count) {
2845 WARN_ON(raw_spin_is_locked(&kvm_lock));
2846 hardware_enable_nolock(NULL);
2850 static struct syscore_ops kvm_syscore_ops = {
2851 .suspend = kvm_suspend,
2852 .resume = kvm_resume,
2856 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2858 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2861 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2863 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2865 kvm_arch_vcpu_load(vcpu, cpu);
2868 static void kvm_sched_out(struct preempt_notifier *pn,
2869 struct task_struct *next)
2871 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2873 kvm_arch_vcpu_put(vcpu);
2876 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2877 struct module *module)
2882 r = kvm_arch_init(opaque);
2886 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2891 r = kvm_arch_hardware_setup();
2895 for_each_online_cpu(cpu) {
2896 smp_call_function_single(cpu,
2897 kvm_arch_check_processor_compat,
2903 r = register_cpu_notifier(&kvm_cpu_notifier);
2906 register_reboot_notifier(&kvm_reboot_notifier);
2908 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2910 vcpu_align = __alignof__(struct kvm_vcpu);
2911 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2913 if (!kvm_vcpu_cache) {
2918 r = kvm_async_pf_init();
2922 kvm_chardev_ops.owner = module;
2923 kvm_vm_fops.owner = module;
2924 kvm_vcpu_fops.owner = module;
2926 r = misc_register(&kvm_dev);
2928 printk(KERN_ERR "kvm: misc device register failed\n");
2932 register_syscore_ops(&kvm_syscore_ops);
2934 kvm_preempt_ops.sched_in = kvm_sched_in;
2935 kvm_preempt_ops.sched_out = kvm_sched_out;
2937 r = kvm_init_debug();
2939 printk(KERN_ERR "kvm: create debugfs files failed\n");
2946 unregister_syscore_ops(&kvm_syscore_ops);
2948 kvm_async_pf_deinit();
2950 kmem_cache_destroy(kvm_vcpu_cache);
2952 unregister_reboot_notifier(&kvm_reboot_notifier);
2953 unregister_cpu_notifier(&kvm_cpu_notifier);
2956 kvm_arch_hardware_unsetup();
2958 free_cpumask_var(cpus_hardware_enabled);
2964 EXPORT_SYMBOL_GPL(kvm_init);
2969 misc_deregister(&kvm_dev);
2970 kmem_cache_destroy(kvm_vcpu_cache);
2971 kvm_async_pf_deinit();
2972 unregister_syscore_ops(&kvm_syscore_ops);
2973 unregister_reboot_notifier(&kvm_reboot_notifier);
2974 unregister_cpu_notifier(&kvm_cpu_notifier);
2975 on_each_cpu(hardware_disable_nolock, NULL, 1);
2976 kvm_arch_hardware_unsetup();
2978 free_cpumask_var(cpus_hardware_enabled);
2980 EXPORT_SYMBOL_GPL(kvm_exit);