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 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(kvm);
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(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_memslot(struct kvm_memory_slot *slot,
990 return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
993 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
994 gfn_t *nr_pages, bool write)
996 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
997 return KVM_HVA_ERR_BAD;
999 if (memslot_is_readonly(slot) && write)
1000 return KVM_HVA_ERR_RO_BAD;
1003 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1005 return __gfn_to_hva_memslot(slot, gfn);
1008 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1011 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1014 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1017 return gfn_to_hva_many(slot, gfn, NULL);
1019 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1021 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1023 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1025 EXPORT_SYMBOL_GPL(gfn_to_hva);
1028 * The hva returned by this function is only allowed to be read.
1029 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1031 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1033 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1036 static int kvm_read_hva(void *data, void __user *hva, int len)
1038 return __copy_from_user(data, hva, len);
1041 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1043 return __copy_from_user_inatomic(data, hva, len);
1046 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1047 unsigned long start, int write, struct page **page)
1049 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1052 flags |= FOLL_WRITE;
1054 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1057 static inline int check_user_page_hwpoison(unsigned long addr)
1059 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1061 rc = __get_user_pages(current, current->mm, addr, 1,
1062 flags, NULL, NULL, NULL);
1063 return rc == -EHWPOISON;
1067 * The atomic path to get the writable pfn which will be stored in @pfn,
1068 * true indicates success, otherwise false is returned.
1070 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1071 bool write_fault, bool *writable, pfn_t *pfn)
1073 struct page *page[1];
1076 if (!(async || atomic))
1080 * Fast pin a writable pfn only if it is a write fault request
1081 * or the caller allows to map a writable pfn for a read fault
1084 if (!(write_fault || writable))
1087 npages = __get_user_pages_fast(addr, 1, 1, page);
1089 *pfn = page_to_pfn(page[0]);
1100 * The slow path to get the pfn of the specified host virtual address,
1101 * 1 indicates success, -errno is returned if error is detected.
1103 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1104 bool *writable, pfn_t *pfn)
1106 struct page *page[1];
1112 *writable = write_fault;
1115 down_read(¤t->mm->mmap_sem);
1116 npages = get_user_page_nowait(current, current->mm,
1117 addr, write_fault, page);
1118 up_read(¤t->mm->mmap_sem);
1120 npages = get_user_pages_fast(addr, 1, write_fault,
1125 /* map read fault as writable if possible */
1126 if (unlikely(!write_fault) && writable) {
1127 struct page *wpage[1];
1129 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1138 *pfn = page_to_pfn(page[0]);
1142 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1144 if (unlikely(!(vma->vm_flags & VM_READ)))
1147 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1154 * Pin guest page in memory and return its pfn.
1155 * @addr: host virtual address which maps memory to the guest
1156 * @atomic: whether this function can sleep
1157 * @async: whether this function need to wait IO complete if the
1158 * host page is not in the memory
1159 * @write_fault: whether we should get a writable host page
1160 * @writable: whether it allows to map a writable host page for !@write_fault
1162 * The function will map a writable host page for these two cases:
1163 * 1): @write_fault = true
1164 * 2): @write_fault = false && @writable, @writable will tell the caller
1165 * whether the mapping is writable.
1167 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1168 bool write_fault, bool *writable)
1170 struct vm_area_struct *vma;
1174 /* we can do it either atomically or asynchronously, not both */
1175 BUG_ON(atomic && async);
1177 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1181 return KVM_PFN_ERR_FAULT;
1183 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1187 down_read(¤t->mm->mmap_sem);
1188 if (npages == -EHWPOISON ||
1189 (!async && check_user_page_hwpoison(addr))) {
1190 pfn = KVM_PFN_ERR_HWPOISON;
1194 vma = find_vma_intersection(current->mm, addr, addr + 1);
1197 pfn = KVM_PFN_ERR_FAULT;
1198 else if ((vma->vm_flags & VM_PFNMAP)) {
1199 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1201 BUG_ON(!kvm_is_mmio_pfn(pfn));
1203 if (async && vma_is_valid(vma, write_fault))
1205 pfn = KVM_PFN_ERR_FAULT;
1208 up_read(¤t->mm->mmap_sem);
1213 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1214 bool *async, bool write_fault, bool *writable)
1216 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1218 if (addr == KVM_HVA_ERR_RO_BAD)
1219 return KVM_PFN_ERR_RO_FAULT;
1221 if (kvm_is_error_hva(addr))
1222 return KVM_PFN_ERR_BAD;
1224 /* Do not map writable pfn in the readonly memslot. */
1225 if (writable && memslot_is_readonly(slot)) {
1230 return hva_to_pfn(addr, atomic, async, write_fault,
1234 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1235 bool write_fault, bool *writable)
1237 struct kvm_memory_slot *slot;
1242 slot = gfn_to_memslot(kvm, gfn);
1244 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1248 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1250 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1252 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1254 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1255 bool write_fault, bool *writable)
1257 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1259 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1261 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1263 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1265 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1267 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1270 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1272 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1274 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1276 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1279 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1281 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1283 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1285 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1291 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1292 if (kvm_is_error_hva(addr))
1295 if (entry < nr_pages)
1298 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1300 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1302 static struct page *kvm_pfn_to_page(pfn_t pfn)
1304 if (is_error_pfn(pfn))
1305 return KVM_ERR_PTR_BAD_PAGE;
1307 if (kvm_is_mmio_pfn(pfn)) {
1309 return KVM_ERR_PTR_BAD_PAGE;
1312 return pfn_to_page(pfn);
1315 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1319 pfn = gfn_to_pfn(kvm, gfn);
1321 return kvm_pfn_to_page(pfn);
1324 EXPORT_SYMBOL_GPL(gfn_to_page);
1326 void kvm_release_page_clean(struct page *page)
1328 WARN_ON(is_error_page(page));
1330 kvm_release_pfn_clean(page_to_pfn(page));
1332 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1334 void kvm_release_pfn_clean(pfn_t pfn)
1336 WARN_ON(is_error_pfn(pfn));
1338 if (!kvm_is_mmio_pfn(pfn))
1339 put_page(pfn_to_page(pfn));
1341 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1343 void kvm_release_page_dirty(struct page *page)
1345 WARN_ON(is_error_page(page));
1347 kvm_release_pfn_dirty(page_to_pfn(page));
1349 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1351 void kvm_release_pfn_dirty(pfn_t pfn)
1353 kvm_set_pfn_dirty(pfn);
1354 kvm_release_pfn_clean(pfn);
1356 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1358 void kvm_set_page_dirty(struct page *page)
1360 kvm_set_pfn_dirty(page_to_pfn(page));
1362 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1364 void kvm_set_pfn_dirty(pfn_t pfn)
1366 if (!kvm_is_mmio_pfn(pfn)) {
1367 struct page *page = pfn_to_page(pfn);
1368 if (!PageReserved(page))
1372 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1374 void kvm_set_pfn_accessed(pfn_t pfn)
1376 if (!kvm_is_mmio_pfn(pfn))
1377 mark_page_accessed(pfn_to_page(pfn));
1379 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1381 void kvm_get_pfn(pfn_t pfn)
1383 if (!kvm_is_mmio_pfn(pfn))
1384 get_page(pfn_to_page(pfn));
1386 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1388 static int next_segment(unsigned long len, int offset)
1390 if (len > PAGE_SIZE - offset)
1391 return PAGE_SIZE - offset;
1396 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1402 addr = gfn_to_hva_read(kvm, gfn);
1403 if (kvm_is_error_hva(addr))
1405 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1410 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1412 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1414 gfn_t gfn = gpa >> PAGE_SHIFT;
1416 int offset = offset_in_page(gpa);
1419 while ((seg = next_segment(len, offset)) != 0) {
1420 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1430 EXPORT_SYMBOL_GPL(kvm_read_guest);
1432 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1437 gfn_t gfn = gpa >> PAGE_SHIFT;
1438 int offset = offset_in_page(gpa);
1440 addr = gfn_to_hva_read(kvm, gfn);
1441 if (kvm_is_error_hva(addr))
1443 pagefault_disable();
1444 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1450 EXPORT_SYMBOL(kvm_read_guest_atomic);
1452 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1453 int offset, int len)
1458 addr = gfn_to_hva(kvm, gfn);
1459 if (kvm_is_error_hva(addr))
1461 r = __copy_to_user((void __user *)addr + offset, data, len);
1464 mark_page_dirty(kvm, gfn);
1467 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1469 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1472 gfn_t gfn = gpa >> PAGE_SHIFT;
1474 int offset = offset_in_page(gpa);
1477 while ((seg = next_segment(len, offset)) != 0) {
1478 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1489 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1492 struct kvm_memslots *slots = kvm_memslots(kvm);
1493 int offset = offset_in_page(gpa);
1494 gfn_t gfn = gpa >> PAGE_SHIFT;
1497 ghc->generation = slots->generation;
1498 ghc->memslot = gfn_to_memslot(kvm, gfn);
1499 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1500 if (!kvm_is_error_hva(ghc->hva))
1507 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1509 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1510 void *data, unsigned long len)
1512 struct kvm_memslots *slots = kvm_memslots(kvm);
1515 if (slots->generation != ghc->generation)
1516 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1518 if (kvm_is_error_hva(ghc->hva))
1521 r = __copy_to_user((void __user *)ghc->hva, data, len);
1524 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1528 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1530 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1531 void *data, unsigned long len)
1533 struct kvm_memslots *slots = kvm_memslots(kvm);
1536 if (slots->generation != ghc->generation)
1537 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1539 if (kvm_is_error_hva(ghc->hva))
1542 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1548 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1550 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1552 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1555 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1557 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1559 gfn_t gfn = gpa >> PAGE_SHIFT;
1561 int offset = offset_in_page(gpa);
1564 while ((seg = next_segment(len, offset)) != 0) {
1565 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1574 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1576 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1579 if (memslot && memslot->dirty_bitmap) {
1580 unsigned long rel_gfn = gfn - memslot->base_gfn;
1582 /* TODO: introduce set_bit_le() and use it */
1583 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1587 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1589 struct kvm_memory_slot *memslot;
1591 memslot = gfn_to_memslot(kvm, gfn);
1592 mark_page_dirty_in_slot(kvm, memslot, gfn);
1596 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1598 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1603 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1605 if (kvm_arch_vcpu_runnable(vcpu)) {
1606 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1609 if (kvm_cpu_has_pending_timer(vcpu))
1611 if (signal_pending(current))
1617 finish_wait(&vcpu->wq, &wait);
1622 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1624 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1627 int cpu = vcpu->cpu;
1628 wait_queue_head_t *wqp;
1630 wqp = kvm_arch_vcpu_wq(vcpu);
1631 if (waitqueue_active(wqp)) {
1632 wake_up_interruptible(wqp);
1633 ++vcpu->stat.halt_wakeup;
1637 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1638 if (kvm_arch_vcpu_should_kick(vcpu))
1639 smp_send_reschedule(cpu);
1642 #endif /* !CONFIG_S390 */
1644 void kvm_resched(struct kvm_vcpu *vcpu)
1646 if (!need_resched())
1650 EXPORT_SYMBOL_GPL(kvm_resched);
1652 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1655 struct task_struct *task = NULL;
1658 pid = rcu_dereference(target->pid);
1660 task = get_pid_task(target->pid, PIDTYPE_PID);
1664 if (task->flags & PF_VCPU) {
1665 put_task_struct(task);
1668 if (yield_to(task, 1)) {
1669 put_task_struct(task);
1672 put_task_struct(task);
1675 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1677 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1679 * Helper that checks whether a VCPU is eligible for directed yield.
1680 * Most eligible candidate to yield is decided by following heuristics:
1682 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1683 * (preempted lock holder), indicated by @in_spin_loop.
1684 * Set at the beiginning and cleared at the end of interception/PLE handler.
1686 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1687 * chance last time (mostly it has become eligible now since we have probably
1688 * yielded to lockholder in last iteration. This is done by toggling
1689 * @dy_eligible each time a VCPU checked for eligibility.)
1691 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1692 * to preempted lock-holder could result in wrong VCPU selection and CPU
1693 * burning. Giving priority for a potential lock-holder increases lock
1696 * Since algorithm is based on heuristics, accessing another VCPU data without
1697 * locking does not harm. It may result in trying to yield to same VCPU, fail
1698 * and continue with next VCPU and so on.
1700 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1704 eligible = !vcpu->spin_loop.in_spin_loop ||
1705 (vcpu->spin_loop.in_spin_loop &&
1706 vcpu->spin_loop.dy_eligible);
1708 if (vcpu->spin_loop.in_spin_loop)
1709 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1714 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1716 struct kvm *kvm = me->kvm;
1717 struct kvm_vcpu *vcpu;
1718 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1723 kvm_vcpu_set_in_spin_loop(me, true);
1725 * We boost the priority of a VCPU that is runnable but not
1726 * currently running, because it got preempted by something
1727 * else and called schedule in __vcpu_run. Hopefully that
1728 * VCPU is holding the lock that we need and will release it.
1729 * We approximate round-robin by starting at the last boosted VCPU.
1731 for (pass = 0; pass < 2 && !yielded; pass++) {
1732 kvm_for_each_vcpu(i, vcpu, kvm) {
1733 if (!pass && i <= last_boosted_vcpu) {
1734 i = last_boosted_vcpu;
1736 } else if (pass && i > last_boosted_vcpu)
1740 if (waitqueue_active(&vcpu->wq))
1742 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1744 if (kvm_vcpu_yield_to(vcpu)) {
1745 kvm->last_boosted_vcpu = i;
1751 kvm_vcpu_set_in_spin_loop(me, false);
1753 /* Ensure vcpu is not eligible during next spinloop */
1754 kvm_vcpu_set_dy_eligible(me, false);
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1758 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1760 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1763 if (vmf->pgoff == 0)
1764 page = virt_to_page(vcpu->run);
1766 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1767 page = virt_to_page(vcpu->arch.pio_data);
1769 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1770 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1771 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1774 return kvm_arch_vcpu_fault(vcpu, vmf);
1780 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1781 .fault = kvm_vcpu_fault,
1784 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1786 vma->vm_ops = &kvm_vcpu_vm_ops;
1790 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1792 struct kvm_vcpu *vcpu = filp->private_data;
1794 kvm_put_kvm(vcpu->kvm);
1798 static struct file_operations kvm_vcpu_fops = {
1799 .release = kvm_vcpu_release,
1800 .unlocked_ioctl = kvm_vcpu_ioctl,
1801 #ifdef CONFIG_COMPAT
1802 .compat_ioctl = kvm_vcpu_compat_ioctl,
1804 .mmap = kvm_vcpu_mmap,
1805 .llseek = noop_llseek,
1809 * Allocates an inode for the vcpu.
1811 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1813 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1817 * Creates some virtual cpus. Good luck creating more than one.
1819 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1822 struct kvm_vcpu *vcpu, *v;
1824 vcpu = kvm_arch_vcpu_create(kvm, id);
1826 return PTR_ERR(vcpu);
1828 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1830 r = kvm_arch_vcpu_setup(vcpu);
1834 mutex_lock(&kvm->lock);
1835 if (!kvm_vcpu_compatible(vcpu)) {
1837 goto unlock_vcpu_destroy;
1839 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1841 goto unlock_vcpu_destroy;
1844 kvm_for_each_vcpu(r, v, kvm)
1845 if (v->vcpu_id == id) {
1847 goto unlock_vcpu_destroy;
1850 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1852 /* Now it's all set up, let userspace reach it */
1854 r = create_vcpu_fd(vcpu);
1857 goto unlock_vcpu_destroy;
1860 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1862 atomic_inc(&kvm->online_vcpus);
1864 mutex_unlock(&kvm->lock);
1867 unlock_vcpu_destroy:
1868 mutex_unlock(&kvm->lock);
1870 kvm_arch_vcpu_destroy(vcpu);
1874 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1877 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1878 vcpu->sigset_active = 1;
1879 vcpu->sigset = *sigset;
1881 vcpu->sigset_active = 0;
1885 static long kvm_vcpu_ioctl(struct file *filp,
1886 unsigned int ioctl, unsigned long arg)
1888 struct kvm_vcpu *vcpu = filp->private_data;
1889 void __user *argp = (void __user *)arg;
1891 struct kvm_fpu *fpu = NULL;
1892 struct kvm_sregs *kvm_sregs = NULL;
1894 if (vcpu->kvm->mm != current->mm)
1897 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1899 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1900 * so vcpu_load() would break it.
1902 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1903 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1913 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1914 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1916 case KVM_GET_REGS: {
1917 struct kvm_regs *kvm_regs;
1920 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1923 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1927 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1934 case KVM_SET_REGS: {
1935 struct kvm_regs *kvm_regs;
1938 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1939 if (IS_ERR(kvm_regs)) {
1940 r = PTR_ERR(kvm_regs);
1943 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1951 case KVM_GET_SREGS: {
1952 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1956 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1960 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1965 case KVM_SET_SREGS: {
1966 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1967 if (IS_ERR(kvm_sregs)) {
1968 r = PTR_ERR(kvm_sregs);
1971 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1977 case KVM_GET_MP_STATE: {
1978 struct kvm_mp_state mp_state;
1980 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1984 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1989 case KVM_SET_MP_STATE: {
1990 struct kvm_mp_state mp_state;
1993 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1995 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2001 case KVM_TRANSLATE: {
2002 struct kvm_translation tr;
2005 if (copy_from_user(&tr, argp, sizeof tr))
2007 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2011 if (copy_to_user(argp, &tr, sizeof tr))
2016 case KVM_SET_GUEST_DEBUG: {
2017 struct kvm_guest_debug dbg;
2020 if (copy_from_user(&dbg, argp, sizeof dbg))
2022 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2028 case KVM_SET_SIGNAL_MASK: {
2029 struct kvm_signal_mask __user *sigmask_arg = argp;
2030 struct kvm_signal_mask kvm_sigmask;
2031 sigset_t sigset, *p;
2036 if (copy_from_user(&kvm_sigmask, argp,
2037 sizeof kvm_sigmask))
2040 if (kvm_sigmask.len != sizeof sigset)
2043 if (copy_from_user(&sigset, sigmask_arg->sigset,
2048 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2052 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2056 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2060 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2066 fpu = memdup_user(argp, sizeof(*fpu));
2071 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2078 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2087 #ifdef CONFIG_COMPAT
2088 static long kvm_vcpu_compat_ioctl(struct file *filp,
2089 unsigned int ioctl, unsigned long arg)
2091 struct kvm_vcpu *vcpu = filp->private_data;
2092 void __user *argp = compat_ptr(arg);
2095 if (vcpu->kvm->mm != current->mm)
2099 case KVM_SET_SIGNAL_MASK: {
2100 struct kvm_signal_mask __user *sigmask_arg = argp;
2101 struct kvm_signal_mask kvm_sigmask;
2102 compat_sigset_t csigset;
2107 if (copy_from_user(&kvm_sigmask, argp,
2108 sizeof kvm_sigmask))
2111 if (kvm_sigmask.len != sizeof csigset)
2114 if (copy_from_user(&csigset, sigmask_arg->sigset,
2118 sigset_from_compat(&sigset, &csigset);
2119 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2123 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2131 static long kvm_vm_ioctl(struct file *filp,
2132 unsigned int ioctl, unsigned long arg)
2134 struct kvm *kvm = filp->private_data;
2135 void __user *argp = (void __user *)arg;
2138 if (kvm->mm != current->mm)
2141 case KVM_CREATE_VCPU:
2142 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2146 case KVM_SET_USER_MEMORY_REGION: {
2147 struct kvm_userspace_memory_region kvm_userspace_mem;
2150 if (copy_from_user(&kvm_userspace_mem, argp,
2151 sizeof kvm_userspace_mem))
2154 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2159 case KVM_GET_DIRTY_LOG: {
2160 struct kvm_dirty_log log;
2163 if (copy_from_user(&log, argp, sizeof log))
2165 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2170 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2171 case KVM_REGISTER_COALESCED_MMIO: {
2172 struct kvm_coalesced_mmio_zone zone;
2174 if (copy_from_user(&zone, argp, sizeof zone))
2176 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2182 case KVM_UNREGISTER_COALESCED_MMIO: {
2183 struct kvm_coalesced_mmio_zone zone;
2185 if (copy_from_user(&zone, argp, sizeof zone))
2187 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2195 struct kvm_irqfd data;
2198 if (copy_from_user(&data, argp, sizeof data))
2200 r = kvm_irqfd(kvm, &data);
2203 case KVM_IOEVENTFD: {
2204 struct kvm_ioeventfd data;
2207 if (copy_from_user(&data, argp, sizeof data))
2209 r = kvm_ioeventfd(kvm, &data);
2212 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2213 case KVM_SET_BOOT_CPU_ID:
2215 mutex_lock(&kvm->lock);
2216 if (atomic_read(&kvm->online_vcpus) != 0)
2219 kvm->bsp_vcpu_id = arg;
2220 mutex_unlock(&kvm->lock);
2223 #ifdef CONFIG_HAVE_KVM_MSI
2224 case KVM_SIGNAL_MSI: {
2228 if (copy_from_user(&msi, argp, sizeof msi))
2230 r = kvm_send_userspace_msi(kvm, &msi);
2234 #ifdef __KVM_HAVE_IRQ_LINE
2235 case KVM_IRQ_LINE_STATUS:
2236 case KVM_IRQ_LINE: {
2237 struct kvm_irq_level irq_event;
2240 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2243 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2248 if (ioctl == KVM_IRQ_LINE_STATUS) {
2249 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2258 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2260 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2266 #ifdef CONFIG_COMPAT
2267 struct compat_kvm_dirty_log {
2271 compat_uptr_t dirty_bitmap; /* one bit per page */
2276 static long kvm_vm_compat_ioctl(struct file *filp,
2277 unsigned int ioctl, unsigned long arg)
2279 struct kvm *kvm = filp->private_data;
2282 if (kvm->mm != current->mm)
2285 case KVM_GET_DIRTY_LOG: {
2286 struct compat_kvm_dirty_log compat_log;
2287 struct kvm_dirty_log log;
2290 if (copy_from_user(&compat_log, (void __user *)arg,
2291 sizeof(compat_log)))
2293 log.slot = compat_log.slot;
2294 log.padding1 = compat_log.padding1;
2295 log.padding2 = compat_log.padding2;
2296 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2298 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2304 r = kvm_vm_ioctl(filp, ioctl, arg);
2312 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2314 struct page *page[1];
2317 gfn_t gfn = vmf->pgoff;
2318 struct kvm *kvm = vma->vm_file->private_data;
2320 addr = gfn_to_hva(kvm, gfn);
2321 if (kvm_is_error_hva(addr))
2322 return VM_FAULT_SIGBUS;
2324 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2326 if (unlikely(npages != 1))
2327 return VM_FAULT_SIGBUS;
2329 vmf->page = page[0];
2333 static const struct vm_operations_struct kvm_vm_vm_ops = {
2334 .fault = kvm_vm_fault,
2337 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2339 vma->vm_ops = &kvm_vm_vm_ops;
2343 static struct file_operations kvm_vm_fops = {
2344 .release = kvm_vm_release,
2345 .unlocked_ioctl = kvm_vm_ioctl,
2346 #ifdef CONFIG_COMPAT
2347 .compat_ioctl = kvm_vm_compat_ioctl,
2349 .mmap = kvm_vm_mmap,
2350 .llseek = noop_llseek,
2353 static int kvm_dev_ioctl_create_vm(unsigned long type)
2358 kvm = kvm_create_vm(type);
2360 return PTR_ERR(kvm);
2361 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2362 r = kvm_coalesced_mmio_init(kvm);
2368 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2375 static long kvm_dev_ioctl_check_extension_generic(long arg)
2378 case KVM_CAP_USER_MEMORY:
2379 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2380 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2381 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2382 case KVM_CAP_SET_BOOT_CPU_ID:
2384 case KVM_CAP_INTERNAL_ERROR_DATA:
2385 #ifdef CONFIG_HAVE_KVM_MSI
2386 case KVM_CAP_SIGNAL_MSI:
2389 #ifdef KVM_CAP_IRQ_ROUTING
2390 case KVM_CAP_IRQ_ROUTING:
2391 return KVM_MAX_IRQ_ROUTES;
2396 return kvm_dev_ioctl_check_extension(arg);
2399 static long kvm_dev_ioctl(struct file *filp,
2400 unsigned int ioctl, unsigned long arg)
2405 case KVM_GET_API_VERSION:
2409 r = KVM_API_VERSION;
2412 r = kvm_dev_ioctl_create_vm(arg);
2414 case KVM_CHECK_EXTENSION:
2415 r = kvm_dev_ioctl_check_extension_generic(arg);
2417 case KVM_GET_VCPU_MMAP_SIZE:
2421 r = PAGE_SIZE; /* struct kvm_run */
2423 r += PAGE_SIZE; /* pio data page */
2425 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2426 r += PAGE_SIZE; /* coalesced mmio ring page */
2429 case KVM_TRACE_ENABLE:
2430 case KVM_TRACE_PAUSE:
2431 case KVM_TRACE_DISABLE:
2435 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2441 static struct file_operations kvm_chardev_ops = {
2442 .unlocked_ioctl = kvm_dev_ioctl,
2443 .compat_ioctl = kvm_dev_ioctl,
2444 .llseek = noop_llseek,
2447 static struct miscdevice kvm_dev = {
2453 static void hardware_enable_nolock(void *junk)
2455 int cpu = raw_smp_processor_id();
2458 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2461 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2463 r = kvm_arch_hardware_enable(NULL);
2466 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2467 atomic_inc(&hardware_enable_failed);
2468 printk(KERN_INFO "kvm: enabling virtualization on "
2469 "CPU%d failed\n", cpu);
2473 static void hardware_enable(void *junk)
2475 raw_spin_lock(&kvm_lock);
2476 hardware_enable_nolock(junk);
2477 raw_spin_unlock(&kvm_lock);
2480 static void hardware_disable_nolock(void *junk)
2482 int cpu = raw_smp_processor_id();
2484 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2486 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2487 kvm_arch_hardware_disable(NULL);
2490 static void hardware_disable(void *junk)
2492 raw_spin_lock(&kvm_lock);
2493 hardware_disable_nolock(junk);
2494 raw_spin_unlock(&kvm_lock);
2497 static void hardware_disable_all_nolock(void)
2499 BUG_ON(!kvm_usage_count);
2502 if (!kvm_usage_count)
2503 on_each_cpu(hardware_disable_nolock, NULL, 1);
2506 static void hardware_disable_all(void)
2508 raw_spin_lock(&kvm_lock);
2509 hardware_disable_all_nolock();
2510 raw_spin_unlock(&kvm_lock);
2513 static int hardware_enable_all(void)
2517 raw_spin_lock(&kvm_lock);
2520 if (kvm_usage_count == 1) {
2521 atomic_set(&hardware_enable_failed, 0);
2522 on_each_cpu(hardware_enable_nolock, NULL, 1);
2524 if (atomic_read(&hardware_enable_failed)) {
2525 hardware_disable_all_nolock();
2530 raw_spin_unlock(&kvm_lock);
2535 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2540 if (!kvm_usage_count)
2543 val &= ~CPU_TASKS_FROZEN;
2546 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2548 hardware_disable(NULL);
2551 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2553 hardware_enable(NULL);
2560 asmlinkage void kvm_spurious_fault(void)
2562 /* Fault while not rebooting. We want the trace. */
2565 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2567 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2571 * Some (well, at least mine) BIOSes hang on reboot if
2574 * And Intel TXT required VMX off for all cpu when system shutdown.
2576 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2577 kvm_rebooting = true;
2578 on_each_cpu(hardware_disable_nolock, NULL, 1);
2582 static struct notifier_block kvm_reboot_notifier = {
2583 .notifier_call = kvm_reboot,
2587 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2591 for (i = 0; i < bus->dev_count; i++) {
2592 struct kvm_io_device *pos = bus->range[i].dev;
2594 kvm_iodevice_destructor(pos);
2599 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2601 const struct kvm_io_range *r1 = p1;
2602 const struct kvm_io_range *r2 = p2;
2604 if (r1->addr < r2->addr)
2606 if (r1->addr + r1->len > r2->addr + r2->len)
2611 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2612 gpa_t addr, int len)
2614 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2620 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2621 kvm_io_bus_sort_cmp, NULL);
2626 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2627 gpa_t addr, int len)
2629 struct kvm_io_range *range, key;
2632 key = (struct kvm_io_range) {
2637 range = bsearch(&key, bus->range, bus->dev_count,
2638 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2642 off = range - bus->range;
2644 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2650 /* kvm_io_bus_write - called under kvm->slots_lock */
2651 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2652 int len, const void *val)
2655 struct kvm_io_bus *bus;
2656 struct kvm_io_range range;
2658 range = (struct kvm_io_range) {
2663 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2664 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2668 while (idx < bus->dev_count &&
2669 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2670 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2678 /* kvm_io_bus_read - called under kvm->slots_lock */
2679 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2683 struct kvm_io_bus *bus;
2684 struct kvm_io_range range;
2686 range = (struct kvm_io_range) {
2691 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2692 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2696 while (idx < bus->dev_count &&
2697 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2698 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2706 /* Caller must hold slots_lock. */
2707 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2708 int len, struct kvm_io_device *dev)
2710 struct kvm_io_bus *new_bus, *bus;
2712 bus = kvm->buses[bus_idx];
2713 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2716 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2717 sizeof(struct kvm_io_range)), GFP_KERNEL);
2720 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2721 sizeof(struct kvm_io_range)));
2722 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2723 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2724 synchronize_srcu_expedited(&kvm->srcu);
2730 /* Caller must hold slots_lock. */
2731 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2732 struct kvm_io_device *dev)
2735 struct kvm_io_bus *new_bus, *bus;
2737 bus = kvm->buses[bus_idx];
2739 for (i = 0; i < bus->dev_count; i++)
2740 if (bus->range[i].dev == dev) {
2748 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2749 sizeof(struct kvm_io_range)), GFP_KERNEL);
2753 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2754 new_bus->dev_count--;
2755 memcpy(new_bus->range + i, bus->range + i + 1,
2756 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2758 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2759 synchronize_srcu_expedited(&kvm->srcu);
2764 static struct notifier_block kvm_cpu_notifier = {
2765 .notifier_call = kvm_cpu_hotplug,
2768 static int vm_stat_get(void *_offset, u64 *val)
2770 unsigned offset = (long)_offset;
2774 raw_spin_lock(&kvm_lock);
2775 list_for_each_entry(kvm, &vm_list, vm_list)
2776 *val += *(u32 *)((void *)kvm + offset);
2777 raw_spin_unlock(&kvm_lock);
2781 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2783 static int vcpu_stat_get(void *_offset, u64 *val)
2785 unsigned offset = (long)_offset;
2787 struct kvm_vcpu *vcpu;
2791 raw_spin_lock(&kvm_lock);
2792 list_for_each_entry(kvm, &vm_list, vm_list)
2793 kvm_for_each_vcpu(i, vcpu, kvm)
2794 *val += *(u32 *)((void *)vcpu + offset);
2796 raw_spin_unlock(&kvm_lock);
2800 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2802 static const struct file_operations *stat_fops[] = {
2803 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2804 [KVM_STAT_VM] = &vm_stat_fops,
2807 static int kvm_init_debug(void)
2810 struct kvm_stats_debugfs_item *p;
2812 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2813 if (kvm_debugfs_dir == NULL)
2816 for (p = debugfs_entries; p->name; ++p) {
2817 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2818 (void *)(long)p->offset,
2819 stat_fops[p->kind]);
2820 if (p->dentry == NULL)
2827 debugfs_remove_recursive(kvm_debugfs_dir);
2832 static void kvm_exit_debug(void)
2834 struct kvm_stats_debugfs_item *p;
2836 for (p = debugfs_entries; p->name; ++p)
2837 debugfs_remove(p->dentry);
2838 debugfs_remove(kvm_debugfs_dir);
2841 static int kvm_suspend(void)
2843 if (kvm_usage_count)
2844 hardware_disable_nolock(NULL);
2848 static void kvm_resume(void)
2850 if (kvm_usage_count) {
2851 WARN_ON(raw_spin_is_locked(&kvm_lock));
2852 hardware_enable_nolock(NULL);
2856 static struct syscore_ops kvm_syscore_ops = {
2857 .suspend = kvm_suspend,
2858 .resume = kvm_resume,
2862 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2864 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2867 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2869 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2871 kvm_arch_vcpu_load(vcpu, cpu);
2874 static void kvm_sched_out(struct preempt_notifier *pn,
2875 struct task_struct *next)
2877 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2879 kvm_arch_vcpu_put(vcpu);
2882 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2883 struct module *module)
2888 r = kvm_arch_init(opaque);
2892 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2897 r = kvm_arch_hardware_setup();
2901 for_each_online_cpu(cpu) {
2902 smp_call_function_single(cpu,
2903 kvm_arch_check_processor_compat,
2909 r = register_cpu_notifier(&kvm_cpu_notifier);
2912 register_reboot_notifier(&kvm_reboot_notifier);
2914 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2916 vcpu_align = __alignof__(struct kvm_vcpu);
2917 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2919 if (!kvm_vcpu_cache) {
2924 r = kvm_async_pf_init();
2928 kvm_chardev_ops.owner = module;
2929 kvm_vm_fops.owner = module;
2930 kvm_vcpu_fops.owner = module;
2932 r = misc_register(&kvm_dev);
2934 printk(KERN_ERR "kvm: misc device register failed\n");
2938 register_syscore_ops(&kvm_syscore_ops);
2940 kvm_preempt_ops.sched_in = kvm_sched_in;
2941 kvm_preempt_ops.sched_out = kvm_sched_out;
2943 r = kvm_init_debug();
2945 printk(KERN_ERR "kvm: create debugfs files failed\n");
2952 unregister_syscore_ops(&kvm_syscore_ops);
2954 kvm_async_pf_deinit();
2956 kmem_cache_destroy(kvm_vcpu_cache);
2958 unregister_reboot_notifier(&kvm_reboot_notifier);
2959 unregister_cpu_notifier(&kvm_cpu_notifier);
2962 kvm_arch_hardware_unsetup();
2964 free_cpumask_var(cpus_hardware_enabled);
2970 EXPORT_SYMBOL_GPL(kvm_init);
2975 misc_deregister(&kvm_dev);
2976 kmem_cache_destroy(kvm_vcpu_cache);
2977 kvm_async_pf_deinit();
2978 unregister_syscore_ops(&kvm_syscore_ops);
2979 unregister_reboot_notifier(&kvm_reboot_notifier);
2980 unregister_cpu_notifier(&kvm_cpu_notifier);
2981 on_each_cpu(hardware_disable_nolock, NULL, 1);
2982 kvm_arch_hardware_unsetup();
2984 free_cpumask_var(cpus_hardware_enabled);
2986 EXPORT_SYMBOL_GPL(kvm_exit);