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 int vcpu_load(struct kvm_vcpu *vcpu)
138 if (mutex_lock_killable(&vcpu->mutex))
140 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141 /* The thread running this VCPU changed. */
142 struct pid *oldpid = vcpu->pid;
143 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144 rcu_assign_pointer(vcpu->pid, newpid);
149 preempt_notifier_register(&vcpu->preempt_notifier);
150 kvm_arch_vcpu_load(vcpu, cpu);
155 void vcpu_put(struct kvm_vcpu *vcpu)
158 kvm_arch_vcpu_put(vcpu);
159 preempt_notifier_unregister(&vcpu->preempt_notifier);
161 mutex_unlock(&vcpu->mutex);
164 static void ack_flush(void *_completed)
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 struct kvm_vcpu *vcpu;
175 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 kvm_make_request(req, vcpu);
182 /* Set ->requests bit before we read ->mode */
185 if (cpus != NULL && cpu != -1 && cpu != me &&
186 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187 cpumask_set_cpu(cpu, cpus);
189 if (unlikely(cpus == NULL))
190 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191 else if (!cpumask_empty(cpus))
192 smp_call_function_many(cpus, ack_flush, NULL, 1);
196 free_cpumask_var(cpus);
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
202 long dirty_count = kvm->tlbs_dirty;
205 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206 ++kvm->stat.remote_tlb_flush;
207 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
210 void kvm_reload_remote_mmus(struct kvm *kvm)
212 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
217 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
220 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
225 mutex_init(&vcpu->mutex);
230 init_waitqueue_head(&vcpu->wq);
231 kvm_async_pf_vcpu_init(vcpu);
233 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
238 vcpu->run = page_address(page);
240 kvm_vcpu_set_in_spin_loop(vcpu, false);
241 kvm_vcpu_set_dy_eligible(vcpu, false);
243 r = kvm_arch_vcpu_init(vcpu);
249 free_page((unsigned long)vcpu->run);
253 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
255 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
258 kvm_arch_vcpu_uninit(vcpu);
259 free_page((unsigned long)vcpu->run);
261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
266 return container_of(mn, struct kvm, mmu_notifier);
269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
270 struct mm_struct *mm,
271 unsigned long address)
273 struct kvm *kvm = mmu_notifier_to_kvm(mn);
274 int need_tlb_flush, idx;
277 * When ->invalidate_page runs, the linux pte has been zapped
278 * already but the page is still allocated until
279 * ->invalidate_page returns. So if we increase the sequence
280 * here the kvm page fault will notice if the spte can't be
281 * established because the page is going to be freed. If
282 * instead the kvm page fault establishes the spte before
283 * ->invalidate_page runs, kvm_unmap_hva will release it
286 * The sequence increase only need to be seen at spin_unlock
287 * time, and not at spin_lock time.
289 * Increasing the sequence after the spin_unlock would be
290 * unsafe because the kvm page fault could then establish the
291 * pte after kvm_unmap_hva returned, without noticing the page
292 * is going to be freed.
294 idx = srcu_read_lock(&kvm->srcu);
295 spin_lock(&kvm->mmu_lock);
297 kvm->mmu_notifier_seq++;
298 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
299 /* we've to flush the tlb before the pages can be freed */
301 kvm_flush_remote_tlbs(kvm);
303 spin_unlock(&kvm->mmu_lock);
304 srcu_read_unlock(&kvm->srcu, idx);
307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
308 struct mm_struct *mm,
309 unsigned long address,
312 struct kvm *kvm = mmu_notifier_to_kvm(mn);
315 idx = srcu_read_lock(&kvm->srcu);
316 spin_lock(&kvm->mmu_lock);
317 kvm->mmu_notifier_seq++;
318 kvm_set_spte_hva(kvm, address, pte);
319 spin_unlock(&kvm->mmu_lock);
320 srcu_read_unlock(&kvm->srcu, idx);
323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
324 struct mm_struct *mm,
328 struct kvm *kvm = mmu_notifier_to_kvm(mn);
329 int need_tlb_flush = 0, idx;
331 idx = srcu_read_lock(&kvm->srcu);
332 spin_lock(&kvm->mmu_lock);
334 * The count increase must become visible at unlock time as no
335 * spte can be established without taking the mmu_lock and
336 * count is also read inside the mmu_lock critical section.
338 kvm->mmu_notifier_count++;
339 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
340 need_tlb_flush |= kvm->tlbs_dirty;
341 /* we've to flush the tlb before the pages can be freed */
343 kvm_flush_remote_tlbs(kvm);
345 spin_unlock(&kvm->mmu_lock);
346 srcu_read_unlock(&kvm->srcu, idx);
349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
350 struct mm_struct *mm,
354 struct kvm *kvm = mmu_notifier_to_kvm(mn);
356 spin_lock(&kvm->mmu_lock);
358 * This sequence increase will notify the kvm page fault that
359 * the page that is going to be mapped in the spte could have
362 kvm->mmu_notifier_seq++;
365 * The above sequence increase must be visible before the
366 * below count decrease, which is ensured by the smp_wmb above
367 * in conjunction with the smp_rmb in mmu_notifier_retry().
369 kvm->mmu_notifier_count--;
370 spin_unlock(&kvm->mmu_lock);
372 BUG_ON(kvm->mmu_notifier_count < 0);
375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
376 struct mm_struct *mm,
377 unsigned long address)
379 struct kvm *kvm = mmu_notifier_to_kvm(mn);
382 idx = srcu_read_lock(&kvm->srcu);
383 spin_lock(&kvm->mmu_lock);
385 young = kvm_age_hva(kvm, address);
387 kvm_flush_remote_tlbs(kvm);
389 spin_unlock(&kvm->mmu_lock);
390 srcu_read_unlock(&kvm->srcu, idx);
395 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
396 struct mm_struct *mm,
397 unsigned long address)
399 struct kvm *kvm = mmu_notifier_to_kvm(mn);
402 idx = srcu_read_lock(&kvm->srcu);
403 spin_lock(&kvm->mmu_lock);
404 young = kvm_test_age_hva(kvm, address);
405 spin_unlock(&kvm->mmu_lock);
406 srcu_read_unlock(&kvm->srcu, idx);
411 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
412 struct mm_struct *mm)
414 struct kvm *kvm = mmu_notifier_to_kvm(mn);
417 idx = srcu_read_lock(&kvm->srcu);
418 kvm_arch_flush_shadow_all(kvm);
419 srcu_read_unlock(&kvm->srcu, idx);
422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
423 .invalidate_page = kvm_mmu_notifier_invalidate_page,
424 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
425 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
426 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
427 .test_young = kvm_mmu_notifier_test_young,
428 .change_pte = kvm_mmu_notifier_change_pte,
429 .release = kvm_mmu_notifier_release,
432 static int kvm_init_mmu_notifier(struct kvm *kvm)
434 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
435 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
438 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
440 static int kvm_init_mmu_notifier(struct kvm *kvm)
445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
447 static void kvm_init_memslots_id(struct kvm *kvm)
450 struct kvm_memslots *slots = kvm->memslots;
452 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
453 slots->id_to_index[i] = slots->memslots[i].id = i;
456 static struct kvm *kvm_create_vm(unsigned long type)
459 struct kvm *kvm = kvm_arch_alloc_vm();
462 return ERR_PTR(-ENOMEM);
464 r = kvm_arch_init_vm(kvm, type);
466 goto out_err_nodisable;
468 r = hardware_enable_all();
470 goto out_err_nodisable;
472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
473 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
474 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
477 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
480 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
483 kvm_init_memslots_id(kvm);
484 if (init_srcu_struct(&kvm->srcu))
486 for (i = 0; i < KVM_NR_BUSES; i++) {
487 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
493 spin_lock_init(&kvm->mmu_lock);
494 kvm->mm = current->mm;
495 atomic_inc(&kvm->mm->mm_count);
496 kvm_eventfd_init(kvm);
497 mutex_init(&kvm->lock);
498 mutex_init(&kvm->irq_lock);
499 mutex_init(&kvm->slots_lock);
500 atomic_set(&kvm->users_count, 1);
502 r = kvm_init_mmu_notifier(kvm);
506 raw_spin_lock(&kvm_lock);
507 list_add(&kvm->vm_list, &vm_list);
508 raw_spin_unlock(&kvm_lock);
513 cleanup_srcu_struct(&kvm->srcu);
515 hardware_disable_all();
517 for (i = 0; i < KVM_NR_BUSES; i++)
518 kfree(kvm->buses[i]);
519 kfree(kvm->memslots);
520 kvm_arch_free_vm(kvm);
525 * Avoid using vmalloc for a small buffer.
526 * Should not be used when the size is statically known.
528 void *kvm_kvzalloc(unsigned long size)
530 if (size > PAGE_SIZE)
531 return vzalloc(size);
533 return kzalloc(size, GFP_KERNEL);
536 void kvm_kvfree(const void *addr)
538 if (is_vmalloc_addr(addr))
544 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
546 if (!memslot->dirty_bitmap)
549 kvm_kvfree(memslot->dirty_bitmap);
550 memslot->dirty_bitmap = NULL;
554 * Free any memory in @free but not in @dont.
556 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
557 struct kvm_memory_slot *dont)
559 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
560 kvm_destroy_dirty_bitmap(free);
562 kvm_arch_free_memslot(free, dont);
567 void kvm_free_physmem(struct kvm *kvm)
569 struct kvm_memslots *slots = kvm->memslots;
570 struct kvm_memory_slot *memslot;
572 kvm_for_each_memslot(memslot, slots)
573 kvm_free_physmem_slot(memslot, NULL);
575 kfree(kvm->memslots);
578 static void kvm_destroy_vm(struct kvm *kvm)
581 struct mm_struct *mm = kvm->mm;
583 kvm_arch_sync_events(kvm);
584 raw_spin_lock(&kvm_lock);
585 list_del(&kvm->vm_list);
586 raw_spin_unlock(&kvm_lock);
587 kvm_free_irq_routing(kvm);
588 for (i = 0; i < KVM_NR_BUSES; i++)
589 kvm_io_bus_destroy(kvm->buses[i]);
590 kvm_coalesced_mmio_free(kvm);
591 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
594 kvm_arch_flush_shadow_all(kvm);
596 kvm_arch_destroy_vm(kvm);
597 kvm_free_physmem(kvm);
598 cleanup_srcu_struct(&kvm->srcu);
599 kvm_arch_free_vm(kvm);
600 hardware_disable_all();
604 void kvm_get_kvm(struct kvm *kvm)
606 atomic_inc(&kvm->users_count);
608 EXPORT_SYMBOL_GPL(kvm_get_kvm);
610 void kvm_put_kvm(struct kvm *kvm)
612 if (atomic_dec_and_test(&kvm->users_count))
615 EXPORT_SYMBOL_GPL(kvm_put_kvm);
618 static int kvm_vm_release(struct inode *inode, struct file *filp)
620 struct kvm *kvm = filp->private_data;
622 kvm_irqfd_release(kvm);
629 * Allocation size is twice as large as the actual dirty bitmap size.
630 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
632 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
635 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
637 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
638 if (!memslot->dirty_bitmap)
641 #endif /* !CONFIG_S390 */
645 static int cmp_memslot(const void *slot1, const void *slot2)
647 struct kvm_memory_slot *s1, *s2;
649 s1 = (struct kvm_memory_slot *)slot1;
650 s2 = (struct kvm_memory_slot *)slot2;
652 if (s1->npages < s2->npages)
654 if (s1->npages > s2->npages)
661 * Sort the memslots base on its size, so the larger slots
662 * will get better fit.
664 static void sort_memslots(struct kvm_memslots *slots)
668 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
669 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
671 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
672 slots->id_to_index[slots->memslots[i].id] = i;
675 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
679 struct kvm_memory_slot *old = id_to_memslot(slots, id);
680 unsigned long npages = old->npages;
683 if (new->npages != npages)
684 sort_memslots(slots);
690 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
692 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
694 #ifdef KVM_CAP_READONLY_MEM
695 valid_flags |= KVM_MEM_READONLY;
698 if (mem->flags & ~valid_flags)
705 * Allocate some memory and give it an address in the guest physical address
708 * Discontiguous memory is allowed, mostly for framebuffers.
710 * Must be called holding mmap_sem for write.
712 int __kvm_set_memory_region(struct kvm *kvm,
713 struct kvm_userspace_memory_region *mem,
718 unsigned long npages;
719 struct kvm_memory_slot *memslot, *slot;
720 struct kvm_memory_slot old, new;
721 struct kvm_memslots *slots = NULL, *old_memslots;
723 r = check_memory_region_flags(mem);
728 /* General sanity checks */
729 if (mem->memory_size & (PAGE_SIZE - 1))
731 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
733 /* We can read the guest memory with __xxx_user() later on. */
735 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
736 !access_ok(VERIFY_WRITE,
737 (void __user *)(unsigned long)mem->userspace_addr,
740 if (mem->slot >= KVM_MEM_SLOTS_NUM)
742 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
745 memslot = id_to_memslot(kvm->memslots, mem->slot);
746 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
747 npages = mem->memory_size >> PAGE_SHIFT;
750 if (npages > KVM_MEM_MAX_NR_PAGES)
754 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
756 new = old = *memslot;
759 new.base_gfn = base_gfn;
761 new.flags = mem->flags;
764 * Disallow changing a memory slot's size or changing anything about
765 * zero sized slots that doesn't involve making them non-zero.
768 if (npages && old.npages && npages != old.npages)
770 if (!npages && !old.npages)
773 /* Check for overlaps */
775 kvm_for_each_memslot(slot, kvm->memslots) {
776 if (slot->id >= KVM_USER_MEM_SLOTS || slot == memslot)
778 if (!((base_gfn + npages <= slot->base_gfn) ||
779 (base_gfn >= slot->base_gfn + slot->npages)))
783 /* Free page dirty bitmap if unneeded */
784 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
785 new.dirty_bitmap = NULL;
790 * Allocate if a slot is being created. If modifying a slot,
791 * the userspace_addr cannot change.
794 new.user_alloc = user_alloc;
795 new.userspace_addr = mem->userspace_addr;
797 if (kvm_arch_create_memslot(&new, npages))
799 } else if (npages && mem->userspace_addr != old.userspace_addr) {
804 /* Allocate page dirty bitmap if needed */
805 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
806 if (kvm_create_dirty_bitmap(&new) < 0)
808 /* destroy any largepage mappings for dirty tracking */
811 if (!npages || base_gfn != old.base_gfn) {
812 struct kvm_memory_slot *slot;
815 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
819 slot = id_to_memslot(slots, mem->slot);
820 slot->flags |= KVM_MEMSLOT_INVALID;
822 update_memslots(slots, NULL);
824 old_memslots = kvm->memslots;
825 rcu_assign_pointer(kvm->memslots, slots);
826 synchronize_srcu_expedited(&kvm->srcu);
827 /* slot was deleted or moved, clear iommu mapping */
828 kvm_iommu_unmap_pages(kvm, &old);
829 /* From this point no new shadow pages pointing to a deleted,
830 * or moved, memslot will be created.
832 * validation of sp->gfn happens in:
833 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
834 * - kvm_is_visible_gfn (mmu_check_roots)
836 kvm_arch_flush_shadow_memslot(kvm, slot);
837 slots = old_memslots;
840 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
846 * We can re-use the old_memslots from above, the only difference
847 * from the currently installed memslots is the invalid flag. This
848 * will get overwritten by update_memslots anyway.
851 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
857 /* map new memory slot into the iommu */
859 r = kvm_iommu_map_pages(kvm, &new);
864 /* actual memory is freed via old in kvm_free_physmem_slot below */
866 new.dirty_bitmap = NULL;
867 memset(&new.arch, 0, sizeof(new.arch));
870 update_memslots(slots, &new);
871 old_memslots = kvm->memslots;
872 rcu_assign_pointer(kvm->memslots, slots);
873 synchronize_srcu_expedited(&kvm->srcu);
875 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
877 kvm_free_physmem_slot(&old, &new);
885 kvm_free_physmem_slot(&new, &old);
890 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
892 int kvm_set_memory_region(struct kvm *kvm,
893 struct kvm_userspace_memory_region *mem,
898 mutex_lock(&kvm->slots_lock);
899 r = __kvm_set_memory_region(kvm, mem, user_alloc);
900 mutex_unlock(&kvm->slots_lock);
903 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
905 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
907 kvm_userspace_memory_region *mem,
910 if (mem->slot >= KVM_USER_MEM_SLOTS)
912 return kvm_set_memory_region(kvm, mem, user_alloc);
915 int kvm_get_dirty_log(struct kvm *kvm,
916 struct kvm_dirty_log *log, int *is_dirty)
918 struct kvm_memory_slot *memslot;
921 unsigned long any = 0;
924 if (log->slot >= KVM_USER_MEM_SLOTS)
927 memslot = id_to_memslot(kvm->memslots, log->slot);
929 if (!memslot->dirty_bitmap)
932 n = kvm_dirty_bitmap_bytes(memslot);
934 for (i = 0; !any && i < n/sizeof(long); ++i)
935 any = memslot->dirty_bitmap[i];
938 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
949 bool kvm_largepages_enabled(void)
951 return largepages_enabled;
954 void kvm_disable_largepages(void)
956 largepages_enabled = false;
958 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
960 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
962 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
964 EXPORT_SYMBOL_GPL(gfn_to_memslot);
966 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
968 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
970 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
971 memslot->flags & KVM_MEMSLOT_INVALID)
976 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
978 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
980 struct vm_area_struct *vma;
981 unsigned long addr, size;
985 addr = gfn_to_hva(kvm, gfn);
986 if (kvm_is_error_hva(addr))
989 down_read(¤t->mm->mmap_sem);
990 vma = find_vma(current->mm, addr);
994 size = vma_kernel_pagesize(vma);
997 up_read(¤t->mm->mmap_sem);
1002 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1004 return slot->flags & KVM_MEM_READONLY;
1007 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1008 gfn_t *nr_pages, bool write)
1010 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1011 return KVM_HVA_ERR_BAD;
1013 if (memslot_is_readonly(slot) && write)
1014 return KVM_HVA_ERR_RO_BAD;
1017 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1019 return __gfn_to_hva_memslot(slot, gfn);
1022 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1025 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1028 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1031 return gfn_to_hva_many(slot, gfn, NULL);
1033 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1035 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1037 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1039 EXPORT_SYMBOL_GPL(gfn_to_hva);
1042 * The hva returned by this function is only allowed to be read.
1043 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1045 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1047 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1050 static int kvm_read_hva(void *data, void __user *hva, int len)
1052 return __copy_from_user(data, hva, len);
1055 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1057 return __copy_from_user_inatomic(data, hva, len);
1060 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1061 unsigned long start, int write, struct page **page)
1063 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1066 flags |= FOLL_WRITE;
1068 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1071 static inline int check_user_page_hwpoison(unsigned long addr)
1073 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1075 rc = __get_user_pages(current, current->mm, addr, 1,
1076 flags, NULL, NULL, NULL);
1077 return rc == -EHWPOISON;
1081 * The atomic path to get the writable pfn which will be stored in @pfn,
1082 * true indicates success, otherwise false is returned.
1084 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1085 bool write_fault, bool *writable, pfn_t *pfn)
1087 struct page *page[1];
1090 if (!(async || atomic))
1094 * Fast pin a writable pfn only if it is a write fault request
1095 * or the caller allows to map a writable pfn for a read fault
1098 if (!(write_fault || writable))
1101 npages = __get_user_pages_fast(addr, 1, 1, page);
1103 *pfn = page_to_pfn(page[0]);
1114 * The slow path to get the pfn of the specified host virtual address,
1115 * 1 indicates success, -errno is returned if error is detected.
1117 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1118 bool *writable, pfn_t *pfn)
1120 struct page *page[1];
1126 *writable = write_fault;
1129 down_read(¤t->mm->mmap_sem);
1130 npages = get_user_page_nowait(current, current->mm,
1131 addr, write_fault, page);
1132 up_read(¤t->mm->mmap_sem);
1134 npages = get_user_pages_fast(addr, 1, write_fault,
1139 /* map read fault as writable if possible */
1140 if (unlikely(!write_fault) && writable) {
1141 struct page *wpage[1];
1143 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1152 *pfn = page_to_pfn(page[0]);
1156 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1158 if (unlikely(!(vma->vm_flags & VM_READ)))
1161 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1168 * Pin guest page in memory and return its pfn.
1169 * @addr: host virtual address which maps memory to the guest
1170 * @atomic: whether this function can sleep
1171 * @async: whether this function need to wait IO complete if the
1172 * host page is not in the memory
1173 * @write_fault: whether we should get a writable host page
1174 * @writable: whether it allows to map a writable host page for !@write_fault
1176 * The function will map a writable host page for these two cases:
1177 * 1): @write_fault = true
1178 * 2): @write_fault = false && @writable, @writable will tell the caller
1179 * whether the mapping is writable.
1181 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1182 bool write_fault, bool *writable)
1184 struct vm_area_struct *vma;
1188 /* we can do it either atomically or asynchronously, not both */
1189 BUG_ON(atomic && async);
1191 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1195 return KVM_PFN_ERR_FAULT;
1197 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1201 down_read(¤t->mm->mmap_sem);
1202 if (npages == -EHWPOISON ||
1203 (!async && check_user_page_hwpoison(addr))) {
1204 pfn = KVM_PFN_ERR_HWPOISON;
1208 vma = find_vma_intersection(current->mm, addr, addr + 1);
1211 pfn = KVM_PFN_ERR_FAULT;
1212 else if ((vma->vm_flags & VM_PFNMAP)) {
1213 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1215 BUG_ON(!kvm_is_mmio_pfn(pfn));
1217 if (async && vma_is_valid(vma, write_fault))
1219 pfn = KVM_PFN_ERR_FAULT;
1222 up_read(¤t->mm->mmap_sem);
1227 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1228 bool *async, bool write_fault, bool *writable)
1230 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1232 if (addr == KVM_HVA_ERR_RO_BAD)
1233 return KVM_PFN_ERR_RO_FAULT;
1235 if (kvm_is_error_hva(addr))
1236 return KVM_PFN_NOSLOT;
1238 /* Do not map writable pfn in the readonly memslot. */
1239 if (writable && memslot_is_readonly(slot)) {
1244 return hva_to_pfn(addr, atomic, async, write_fault,
1248 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1249 bool write_fault, bool *writable)
1251 struct kvm_memory_slot *slot;
1256 slot = gfn_to_memslot(kvm, gfn);
1258 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1262 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1264 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1266 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1268 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1269 bool write_fault, bool *writable)
1271 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1273 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1275 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1277 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1279 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1281 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1284 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1286 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1288 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1290 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1293 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1295 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1297 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1299 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1305 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1306 if (kvm_is_error_hva(addr))
1309 if (entry < nr_pages)
1312 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1314 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1316 static struct page *kvm_pfn_to_page(pfn_t pfn)
1318 if (is_error_noslot_pfn(pfn))
1319 return KVM_ERR_PTR_BAD_PAGE;
1321 if (kvm_is_mmio_pfn(pfn)) {
1323 return KVM_ERR_PTR_BAD_PAGE;
1326 return pfn_to_page(pfn);
1329 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1333 pfn = gfn_to_pfn(kvm, gfn);
1335 return kvm_pfn_to_page(pfn);
1338 EXPORT_SYMBOL_GPL(gfn_to_page);
1340 void kvm_release_page_clean(struct page *page)
1342 WARN_ON(is_error_page(page));
1344 kvm_release_pfn_clean(page_to_pfn(page));
1346 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1348 void kvm_release_pfn_clean(pfn_t pfn)
1350 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1351 put_page(pfn_to_page(pfn));
1353 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1355 void kvm_release_page_dirty(struct page *page)
1357 WARN_ON(is_error_page(page));
1359 kvm_release_pfn_dirty(page_to_pfn(page));
1361 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1363 void kvm_release_pfn_dirty(pfn_t pfn)
1365 kvm_set_pfn_dirty(pfn);
1366 kvm_release_pfn_clean(pfn);
1368 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1370 void kvm_set_page_dirty(struct page *page)
1372 kvm_set_pfn_dirty(page_to_pfn(page));
1374 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1376 void kvm_set_pfn_dirty(pfn_t pfn)
1378 if (!kvm_is_mmio_pfn(pfn)) {
1379 struct page *page = pfn_to_page(pfn);
1380 if (!PageReserved(page))
1384 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1386 void kvm_set_pfn_accessed(pfn_t pfn)
1388 if (!kvm_is_mmio_pfn(pfn))
1389 mark_page_accessed(pfn_to_page(pfn));
1391 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1393 void kvm_get_pfn(pfn_t pfn)
1395 if (!kvm_is_mmio_pfn(pfn))
1396 get_page(pfn_to_page(pfn));
1398 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1400 static int next_segment(unsigned long len, int offset)
1402 if (len > PAGE_SIZE - offset)
1403 return PAGE_SIZE - offset;
1408 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1414 addr = gfn_to_hva_read(kvm, gfn);
1415 if (kvm_is_error_hva(addr))
1417 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1422 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1424 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1426 gfn_t gfn = gpa >> PAGE_SHIFT;
1428 int offset = offset_in_page(gpa);
1431 while ((seg = next_segment(len, offset)) != 0) {
1432 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1442 EXPORT_SYMBOL_GPL(kvm_read_guest);
1444 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1449 gfn_t gfn = gpa >> PAGE_SHIFT;
1450 int offset = offset_in_page(gpa);
1452 addr = gfn_to_hva_read(kvm, gfn);
1453 if (kvm_is_error_hva(addr))
1455 pagefault_disable();
1456 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1462 EXPORT_SYMBOL(kvm_read_guest_atomic);
1464 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1465 int offset, int len)
1470 addr = gfn_to_hva(kvm, gfn);
1471 if (kvm_is_error_hva(addr))
1473 r = __copy_to_user((void __user *)addr + offset, data, len);
1476 mark_page_dirty(kvm, gfn);
1479 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1481 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1484 gfn_t gfn = gpa >> PAGE_SHIFT;
1486 int offset = offset_in_page(gpa);
1489 while ((seg = next_segment(len, offset)) != 0) {
1490 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1501 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1504 struct kvm_memslots *slots = kvm_memslots(kvm);
1505 int offset = offset_in_page(gpa);
1506 gfn_t gfn = gpa >> PAGE_SHIFT;
1509 ghc->generation = slots->generation;
1510 ghc->memslot = gfn_to_memslot(kvm, gfn);
1511 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1512 if (!kvm_is_error_hva(ghc->hva))
1519 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1521 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1522 void *data, unsigned long len)
1524 struct kvm_memslots *slots = kvm_memslots(kvm);
1527 if (slots->generation != ghc->generation)
1528 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1530 if (kvm_is_error_hva(ghc->hva))
1533 r = __copy_to_user((void __user *)ghc->hva, data, len);
1536 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1540 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1542 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1543 void *data, unsigned long len)
1545 struct kvm_memslots *slots = kvm_memslots(kvm);
1548 if (slots->generation != ghc->generation)
1549 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1551 if (kvm_is_error_hva(ghc->hva))
1554 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1560 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1562 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1564 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1567 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1569 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1571 gfn_t gfn = gpa >> PAGE_SHIFT;
1573 int offset = offset_in_page(gpa);
1576 while ((seg = next_segment(len, offset)) != 0) {
1577 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1586 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1588 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1591 if (memslot && memslot->dirty_bitmap) {
1592 unsigned long rel_gfn = gfn - memslot->base_gfn;
1594 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1598 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1600 struct kvm_memory_slot *memslot;
1602 memslot = gfn_to_memslot(kvm, gfn);
1603 mark_page_dirty_in_slot(kvm, memslot, gfn);
1607 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1609 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1614 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1616 if (kvm_arch_vcpu_runnable(vcpu)) {
1617 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1620 if (kvm_cpu_has_pending_timer(vcpu))
1622 if (signal_pending(current))
1628 finish_wait(&vcpu->wq, &wait);
1633 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1635 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1638 int cpu = vcpu->cpu;
1639 wait_queue_head_t *wqp;
1641 wqp = kvm_arch_vcpu_wq(vcpu);
1642 if (waitqueue_active(wqp)) {
1643 wake_up_interruptible(wqp);
1644 ++vcpu->stat.halt_wakeup;
1648 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1649 if (kvm_arch_vcpu_should_kick(vcpu))
1650 smp_send_reschedule(cpu);
1653 #endif /* !CONFIG_S390 */
1655 void kvm_resched(struct kvm_vcpu *vcpu)
1657 if (!need_resched())
1661 EXPORT_SYMBOL_GPL(kvm_resched);
1663 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1666 struct task_struct *task = NULL;
1669 pid = rcu_dereference(target->pid);
1671 task = get_pid_task(target->pid, PIDTYPE_PID);
1675 if (task->flags & PF_VCPU) {
1676 put_task_struct(task);
1679 if (yield_to(task, 1)) {
1680 put_task_struct(task);
1683 put_task_struct(task);
1686 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1688 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1690 * Helper that checks whether a VCPU is eligible for directed yield.
1691 * Most eligible candidate to yield is decided by following heuristics:
1693 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1694 * (preempted lock holder), indicated by @in_spin_loop.
1695 * Set at the beiginning and cleared at the end of interception/PLE handler.
1697 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1698 * chance last time (mostly it has become eligible now since we have probably
1699 * yielded to lockholder in last iteration. This is done by toggling
1700 * @dy_eligible each time a VCPU checked for eligibility.)
1702 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1703 * to preempted lock-holder could result in wrong VCPU selection and CPU
1704 * burning. Giving priority for a potential lock-holder increases lock
1707 * Since algorithm is based on heuristics, accessing another VCPU data without
1708 * locking does not harm. It may result in trying to yield to same VCPU, fail
1709 * and continue with next VCPU and so on.
1711 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1715 eligible = !vcpu->spin_loop.in_spin_loop ||
1716 (vcpu->spin_loop.in_spin_loop &&
1717 vcpu->spin_loop.dy_eligible);
1719 if (vcpu->spin_loop.in_spin_loop)
1720 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1725 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1727 struct kvm *kvm = me->kvm;
1728 struct kvm_vcpu *vcpu;
1729 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1734 kvm_vcpu_set_in_spin_loop(me, true);
1736 * We boost the priority of a VCPU that is runnable but not
1737 * currently running, because it got preempted by something
1738 * else and called schedule in __vcpu_run. Hopefully that
1739 * VCPU is holding the lock that we need and will release it.
1740 * We approximate round-robin by starting at the last boosted VCPU.
1742 for (pass = 0; pass < 2 && !yielded; pass++) {
1743 kvm_for_each_vcpu(i, vcpu, kvm) {
1744 if (!pass && i <= last_boosted_vcpu) {
1745 i = last_boosted_vcpu;
1747 } else if (pass && i > last_boosted_vcpu)
1751 if (waitqueue_active(&vcpu->wq))
1753 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1755 if (kvm_vcpu_yield_to(vcpu)) {
1756 kvm->last_boosted_vcpu = i;
1762 kvm_vcpu_set_in_spin_loop(me, false);
1764 /* Ensure vcpu is not eligible during next spinloop */
1765 kvm_vcpu_set_dy_eligible(me, false);
1767 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1769 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1771 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1774 if (vmf->pgoff == 0)
1775 page = virt_to_page(vcpu->run);
1777 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1778 page = virt_to_page(vcpu->arch.pio_data);
1780 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1781 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1782 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1785 return kvm_arch_vcpu_fault(vcpu, vmf);
1791 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1792 .fault = kvm_vcpu_fault,
1795 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1797 vma->vm_ops = &kvm_vcpu_vm_ops;
1801 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1803 struct kvm_vcpu *vcpu = filp->private_data;
1805 kvm_put_kvm(vcpu->kvm);
1809 static struct file_operations kvm_vcpu_fops = {
1810 .release = kvm_vcpu_release,
1811 .unlocked_ioctl = kvm_vcpu_ioctl,
1812 #ifdef CONFIG_COMPAT
1813 .compat_ioctl = kvm_vcpu_compat_ioctl,
1815 .mmap = kvm_vcpu_mmap,
1816 .llseek = noop_llseek,
1820 * Allocates an inode for the vcpu.
1822 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1824 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1828 * Creates some virtual cpus. Good luck creating more than one.
1830 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1833 struct kvm_vcpu *vcpu, *v;
1835 vcpu = kvm_arch_vcpu_create(kvm, id);
1837 return PTR_ERR(vcpu);
1839 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1841 r = kvm_arch_vcpu_setup(vcpu);
1845 mutex_lock(&kvm->lock);
1846 if (!kvm_vcpu_compatible(vcpu)) {
1848 goto unlock_vcpu_destroy;
1850 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1852 goto unlock_vcpu_destroy;
1855 kvm_for_each_vcpu(r, v, kvm)
1856 if (v->vcpu_id == id) {
1858 goto unlock_vcpu_destroy;
1861 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1863 /* Now it's all set up, let userspace reach it */
1865 r = create_vcpu_fd(vcpu);
1868 goto unlock_vcpu_destroy;
1871 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1873 atomic_inc(&kvm->online_vcpus);
1875 mutex_unlock(&kvm->lock);
1876 kvm_arch_vcpu_postcreate(vcpu);
1879 unlock_vcpu_destroy:
1880 mutex_unlock(&kvm->lock);
1882 kvm_arch_vcpu_destroy(vcpu);
1886 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1889 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1890 vcpu->sigset_active = 1;
1891 vcpu->sigset = *sigset;
1893 vcpu->sigset_active = 0;
1897 static long kvm_vcpu_ioctl(struct file *filp,
1898 unsigned int ioctl, unsigned long arg)
1900 struct kvm_vcpu *vcpu = filp->private_data;
1901 void __user *argp = (void __user *)arg;
1903 struct kvm_fpu *fpu = NULL;
1904 struct kvm_sregs *kvm_sregs = NULL;
1906 if (vcpu->kvm->mm != current->mm)
1909 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1911 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1912 * so vcpu_load() would break it.
1914 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1915 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1919 r = vcpu_load(vcpu);
1927 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1928 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1930 case KVM_GET_REGS: {
1931 struct kvm_regs *kvm_regs;
1934 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1937 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1941 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1948 case KVM_SET_REGS: {
1949 struct kvm_regs *kvm_regs;
1952 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1953 if (IS_ERR(kvm_regs)) {
1954 r = PTR_ERR(kvm_regs);
1957 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1961 case KVM_GET_SREGS: {
1962 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1966 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1970 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1975 case KVM_SET_SREGS: {
1976 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1977 if (IS_ERR(kvm_sregs)) {
1978 r = PTR_ERR(kvm_sregs);
1982 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1985 case KVM_GET_MP_STATE: {
1986 struct kvm_mp_state mp_state;
1988 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1992 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1997 case KVM_SET_MP_STATE: {
1998 struct kvm_mp_state mp_state;
2001 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2003 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2006 case KVM_TRANSLATE: {
2007 struct kvm_translation tr;
2010 if (copy_from_user(&tr, argp, sizeof tr))
2012 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2016 if (copy_to_user(argp, &tr, sizeof tr))
2021 case KVM_SET_GUEST_DEBUG: {
2022 struct kvm_guest_debug dbg;
2025 if (copy_from_user(&dbg, argp, sizeof dbg))
2027 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2030 case KVM_SET_SIGNAL_MASK: {
2031 struct kvm_signal_mask __user *sigmask_arg = argp;
2032 struct kvm_signal_mask kvm_sigmask;
2033 sigset_t sigset, *p;
2038 if (copy_from_user(&kvm_sigmask, argp,
2039 sizeof kvm_sigmask))
2042 if (kvm_sigmask.len != sizeof sigset)
2045 if (copy_from_user(&sigset, sigmask_arg->sigset,
2050 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2054 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2058 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2062 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2068 fpu = memdup_user(argp, sizeof(*fpu));
2074 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,
2117 sigset_from_compat(&sigset, &csigset);
2118 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2120 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2124 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2132 static long kvm_vm_ioctl(struct file *filp,
2133 unsigned int ioctl, unsigned long arg)
2135 struct kvm *kvm = filp->private_data;
2136 void __user *argp = (void __user *)arg;
2139 if (kvm->mm != current->mm)
2142 case KVM_CREATE_VCPU:
2143 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2145 case KVM_SET_USER_MEMORY_REGION: {
2146 struct kvm_userspace_memory_region kvm_userspace_mem;
2149 if (copy_from_user(&kvm_userspace_mem, argp,
2150 sizeof kvm_userspace_mem))
2153 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2156 case KVM_GET_DIRTY_LOG: {
2157 struct kvm_dirty_log log;
2160 if (copy_from_user(&log, argp, sizeof log))
2162 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2165 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2166 case KVM_REGISTER_COALESCED_MMIO: {
2167 struct kvm_coalesced_mmio_zone zone;
2169 if (copy_from_user(&zone, argp, sizeof zone))
2171 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2174 case KVM_UNREGISTER_COALESCED_MMIO: {
2175 struct kvm_coalesced_mmio_zone zone;
2177 if (copy_from_user(&zone, argp, sizeof zone))
2179 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2184 struct kvm_irqfd data;
2187 if (copy_from_user(&data, argp, sizeof data))
2189 r = kvm_irqfd(kvm, &data);
2192 case KVM_IOEVENTFD: {
2193 struct kvm_ioeventfd data;
2196 if (copy_from_user(&data, argp, sizeof data))
2198 r = kvm_ioeventfd(kvm, &data);
2201 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2202 case KVM_SET_BOOT_CPU_ID:
2204 mutex_lock(&kvm->lock);
2205 if (atomic_read(&kvm->online_vcpus) != 0)
2208 kvm->bsp_vcpu_id = arg;
2209 mutex_unlock(&kvm->lock);
2212 #ifdef CONFIG_HAVE_KVM_MSI
2213 case KVM_SIGNAL_MSI: {
2217 if (copy_from_user(&msi, argp, sizeof msi))
2219 r = kvm_send_userspace_msi(kvm, &msi);
2223 #ifdef __KVM_HAVE_IRQ_LINE
2224 case KVM_IRQ_LINE_STATUS:
2225 case KVM_IRQ_LINE: {
2226 struct kvm_irq_level irq_event;
2229 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2232 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2237 if (ioctl == KVM_IRQ_LINE_STATUS) {
2238 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2247 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2249 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2255 #ifdef CONFIG_COMPAT
2256 struct compat_kvm_dirty_log {
2260 compat_uptr_t dirty_bitmap; /* one bit per page */
2265 static long kvm_vm_compat_ioctl(struct file *filp,
2266 unsigned int ioctl, unsigned long arg)
2268 struct kvm *kvm = filp->private_data;
2271 if (kvm->mm != current->mm)
2274 case KVM_GET_DIRTY_LOG: {
2275 struct compat_kvm_dirty_log compat_log;
2276 struct kvm_dirty_log log;
2279 if (copy_from_user(&compat_log, (void __user *)arg,
2280 sizeof(compat_log)))
2282 log.slot = compat_log.slot;
2283 log.padding1 = compat_log.padding1;
2284 log.padding2 = compat_log.padding2;
2285 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2287 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2291 r = kvm_vm_ioctl(filp, ioctl, arg);
2299 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2301 struct page *page[1];
2304 gfn_t gfn = vmf->pgoff;
2305 struct kvm *kvm = vma->vm_file->private_data;
2307 addr = gfn_to_hva(kvm, gfn);
2308 if (kvm_is_error_hva(addr))
2309 return VM_FAULT_SIGBUS;
2311 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2313 if (unlikely(npages != 1))
2314 return VM_FAULT_SIGBUS;
2316 vmf->page = page[0];
2320 static const struct vm_operations_struct kvm_vm_vm_ops = {
2321 .fault = kvm_vm_fault,
2324 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2326 vma->vm_ops = &kvm_vm_vm_ops;
2330 static struct file_operations kvm_vm_fops = {
2331 .release = kvm_vm_release,
2332 .unlocked_ioctl = kvm_vm_ioctl,
2333 #ifdef CONFIG_COMPAT
2334 .compat_ioctl = kvm_vm_compat_ioctl,
2336 .mmap = kvm_vm_mmap,
2337 .llseek = noop_llseek,
2340 static int kvm_dev_ioctl_create_vm(unsigned long type)
2345 kvm = kvm_create_vm(type);
2347 return PTR_ERR(kvm);
2348 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2349 r = kvm_coalesced_mmio_init(kvm);
2355 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2362 static long kvm_dev_ioctl_check_extension_generic(long arg)
2365 case KVM_CAP_USER_MEMORY:
2366 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2367 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2368 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2369 case KVM_CAP_SET_BOOT_CPU_ID:
2371 case KVM_CAP_INTERNAL_ERROR_DATA:
2372 #ifdef CONFIG_HAVE_KVM_MSI
2373 case KVM_CAP_SIGNAL_MSI:
2376 #ifdef KVM_CAP_IRQ_ROUTING
2377 case KVM_CAP_IRQ_ROUTING:
2378 return KVM_MAX_IRQ_ROUTES;
2383 return kvm_dev_ioctl_check_extension(arg);
2386 static long kvm_dev_ioctl(struct file *filp,
2387 unsigned int ioctl, unsigned long arg)
2392 case KVM_GET_API_VERSION:
2396 r = KVM_API_VERSION;
2399 r = kvm_dev_ioctl_create_vm(arg);
2401 case KVM_CHECK_EXTENSION:
2402 r = kvm_dev_ioctl_check_extension_generic(arg);
2404 case KVM_GET_VCPU_MMAP_SIZE:
2408 r = PAGE_SIZE; /* struct kvm_run */
2410 r += PAGE_SIZE; /* pio data page */
2412 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2413 r += PAGE_SIZE; /* coalesced mmio ring page */
2416 case KVM_TRACE_ENABLE:
2417 case KVM_TRACE_PAUSE:
2418 case KVM_TRACE_DISABLE:
2422 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2428 static struct file_operations kvm_chardev_ops = {
2429 .unlocked_ioctl = kvm_dev_ioctl,
2430 .compat_ioctl = kvm_dev_ioctl,
2431 .llseek = noop_llseek,
2434 static struct miscdevice kvm_dev = {
2440 static void hardware_enable_nolock(void *junk)
2442 int cpu = raw_smp_processor_id();
2445 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2448 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2450 r = kvm_arch_hardware_enable(NULL);
2453 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2454 atomic_inc(&hardware_enable_failed);
2455 printk(KERN_INFO "kvm: enabling virtualization on "
2456 "CPU%d failed\n", cpu);
2460 static void hardware_enable(void *junk)
2462 raw_spin_lock(&kvm_lock);
2463 hardware_enable_nolock(junk);
2464 raw_spin_unlock(&kvm_lock);
2467 static void hardware_disable_nolock(void *junk)
2469 int cpu = raw_smp_processor_id();
2471 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2473 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2474 kvm_arch_hardware_disable(NULL);
2477 static void hardware_disable(void *junk)
2479 raw_spin_lock(&kvm_lock);
2480 hardware_disable_nolock(junk);
2481 raw_spin_unlock(&kvm_lock);
2484 static void hardware_disable_all_nolock(void)
2486 BUG_ON(!kvm_usage_count);
2489 if (!kvm_usage_count)
2490 on_each_cpu(hardware_disable_nolock, NULL, 1);
2493 static void hardware_disable_all(void)
2495 raw_spin_lock(&kvm_lock);
2496 hardware_disable_all_nolock();
2497 raw_spin_unlock(&kvm_lock);
2500 static int hardware_enable_all(void)
2504 raw_spin_lock(&kvm_lock);
2507 if (kvm_usage_count == 1) {
2508 atomic_set(&hardware_enable_failed, 0);
2509 on_each_cpu(hardware_enable_nolock, NULL, 1);
2511 if (atomic_read(&hardware_enable_failed)) {
2512 hardware_disable_all_nolock();
2517 raw_spin_unlock(&kvm_lock);
2522 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2527 if (!kvm_usage_count)
2530 val &= ~CPU_TASKS_FROZEN;
2533 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2535 hardware_disable(NULL);
2538 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2540 hardware_enable(NULL);
2547 asmlinkage void kvm_spurious_fault(void)
2549 /* Fault while not rebooting. We want the trace. */
2552 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2554 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2558 * Some (well, at least mine) BIOSes hang on reboot if
2561 * And Intel TXT required VMX off for all cpu when system shutdown.
2563 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2564 kvm_rebooting = true;
2565 on_each_cpu(hardware_disable_nolock, NULL, 1);
2569 static struct notifier_block kvm_reboot_notifier = {
2570 .notifier_call = kvm_reboot,
2574 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2578 for (i = 0; i < bus->dev_count; i++) {
2579 struct kvm_io_device *pos = bus->range[i].dev;
2581 kvm_iodevice_destructor(pos);
2586 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2588 const struct kvm_io_range *r1 = p1;
2589 const struct kvm_io_range *r2 = p2;
2591 if (r1->addr < r2->addr)
2593 if (r1->addr + r1->len > r2->addr + r2->len)
2598 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2599 gpa_t addr, int len)
2601 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2607 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2608 kvm_io_bus_sort_cmp, NULL);
2613 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2614 gpa_t addr, int len)
2616 struct kvm_io_range *range, key;
2619 key = (struct kvm_io_range) {
2624 range = bsearch(&key, bus->range, bus->dev_count,
2625 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2629 off = range - bus->range;
2631 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2637 /* kvm_io_bus_write - called under kvm->slots_lock */
2638 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2639 int len, const void *val)
2642 struct kvm_io_bus *bus;
2643 struct kvm_io_range range;
2645 range = (struct kvm_io_range) {
2650 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2651 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2655 while (idx < bus->dev_count &&
2656 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2657 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2665 /* kvm_io_bus_read - called under kvm->slots_lock */
2666 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2670 struct kvm_io_bus *bus;
2671 struct kvm_io_range range;
2673 range = (struct kvm_io_range) {
2678 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2679 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2683 while (idx < bus->dev_count &&
2684 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2685 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2693 /* Caller must hold slots_lock. */
2694 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2695 int len, struct kvm_io_device *dev)
2697 struct kvm_io_bus *new_bus, *bus;
2699 bus = kvm->buses[bus_idx];
2700 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2703 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2704 sizeof(struct kvm_io_range)), GFP_KERNEL);
2707 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2708 sizeof(struct kvm_io_range)));
2709 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2710 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2711 synchronize_srcu_expedited(&kvm->srcu);
2717 /* Caller must hold slots_lock. */
2718 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2719 struct kvm_io_device *dev)
2722 struct kvm_io_bus *new_bus, *bus;
2724 bus = kvm->buses[bus_idx];
2726 for (i = 0; i < bus->dev_count; i++)
2727 if (bus->range[i].dev == dev) {
2735 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2736 sizeof(struct kvm_io_range)), GFP_KERNEL);
2740 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2741 new_bus->dev_count--;
2742 memcpy(new_bus->range + i, bus->range + i + 1,
2743 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2745 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2746 synchronize_srcu_expedited(&kvm->srcu);
2751 static struct notifier_block kvm_cpu_notifier = {
2752 .notifier_call = kvm_cpu_hotplug,
2755 static int vm_stat_get(void *_offset, u64 *val)
2757 unsigned offset = (long)_offset;
2761 raw_spin_lock(&kvm_lock);
2762 list_for_each_entry(kvm, &vm_list, vm_list)
2763 *val += *(u32 *)((void *)kvm + offset);
2764 raw_spin_unlock(&kvm_lock);
2768 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2770 static int vcpu_stat_get(void *_offset, u64 *val)
2772 unsigned offset = (long)_offset;
2774 struct kvm_vcpu *vcpu;
2778 raw_spin_lock(&kvm_lock);
2779 list_for_each_entry(kvm, &vm_list, vm_list)
2780 kvm_for_each_vcpu(i, vcpu, kvm)
2781 *val += *(u32 *)((void *)vcpu + offset);
2783 raw_spin_unlock(&kvm_lock);
2787 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2789 static const struct file_operations *stat_fops[] = {
2790 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2791 [KVM_STAT_VM] = &vm_stat_fops,
2794 static int kvm_init_debug(void)
2797 struct kvm_stats_debugfs_item *p;
2799 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2800 if (kvm_debugfs_dir == NULL)
2803 for (p = debugfs_entries; p->name; ++p) {
2804 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2805 (void *)(long)p->offset,
2806 stat_fops[p->kind]);
2807 if (p->dentry == NULL)
2814 debugfs_remove_recursive(kvm_debugfs_dir);
2819 static void kvm_exit_debug(void)
2821 struct kvm_stats_debugfs_item *p;
2823 for (p = debugfs_entries; p->name; ++p)
2824 debugfs_remove(p->dentry);
2825 debugfs_remove(kvm_debugfs_dir);
2828 static int kvm_suspend(void)
2830 if (kvm_usage_count)
2831 hardware_disable_nolock(NULL);
2835 static void kvm_resume(void)
2837 if (kvm_usage_count) {
2838 WARN_ON(raw_spin_is_locked(&kvm_lock));
2839 hardware_enable_nolock(NULL);
2843 static struct syscore_ops kvm_syscore_ops = {
2844 .suspend = kvm_suspend,
2845 .resume = kvm_resume,
2849 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2851 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2854 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2856 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2858 kvm_arch_vcpu_load(vcpu, cpu);
2861 static void kvm_sched_out(struct preempt_notifier *pn,
2862 struct task_struct *next)
2864 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2866 kvm_arch_vcpu_put(vcpu);
2869 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2870 struct module *module)
2875 r = kvm_arch_init(opaque);
2879 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2884 r = kvm_arch_hardware_setup();
2888 for_each_online_cpu(cpu) {
2889 smp_call_function_single(cpu,
2890 kvm_arch_check_processor_compat,
2896 r = register_cpu_notifier(&kvm_cpu_notifier);
2899 register_reboot_notifier(&kvm_reboot_notifier);
2901 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2903 vcpu_align = __alignof__(struct kvm_vcpu);
2904 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2906 if (!kvm_vcpu_cache) {
2911 r = kvm_async_pf_init();
2915 kvm_chardev_ops.owner = module;
2916 kvm_vm_fops.owner = module;
2917 kvm_vcpu_fops.owner = module;
2919 r = misc_register(&kvm_dev);
2921 printk(KERN_ERR "kvm: misc device register failed\n");
2925 register_syscore_ops(&kvm_syscore_ops);
2927 kvm_preempt_ops.sched_in = kvm_sched_in;
2928 kvm_preempt_ops.sched_out = kvm_sched_out;
2930 r = kvm_init_debug();
2932 printk(KERN_ERR "kvm: create debugfs files failed\n");
2939 unregister_syscore_ops(&kvm_syscore_ops);
2941 kvm_async_pf_deinit();
2943 kmem_cache_destroy(kvm_vcpu_cache);
2945 unregister_reboot_notifier(&kvm_reboot_notifier);
2946 unregister_cpu_notifier(&kvm_cpu_notifier);
2949 kvm_arch_hardware_unsetup();
2951 free_cpumask_var(cpus_hardware_enabled);
2957 EXPORT_SYMBOL_GPL(kvm_init);
2962 misc_deregister(&kvm_dev);
2963 kmem_cache_destroy(kvm_vcpu_cache);
2964 kvm_async_pf_deinit();
2965 unregister_syscore_ops(&kvm_syscore_ops);
2966 unregister_reboot_notifier(&kvm_reboot_notifier);
2967 unregister_cpu_notifier(&kvm_cpu_notifier);
2968 on_each_cpu(hardware_disable_nolock, NULL, 1);
2969 kvm_arch_hardware_unsetup();
2971 free_cpumask_var(cpus_hardware_enabled);
2973 EXPORT_SYMBOL_GPL(kvm_exit);