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);
478 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
481 kvm_init_memslots_id(kvm);
482 if (init_srcu_struct(&kvm->srcu))
484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
491 spin_lock_init(&kvm->mmu_lock);
492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
494 kvm_eventfd_init(kvm);
495 mutex_init(&kvm->lock);
496 mutex_init(&kvm->irq_lock);
497 mutex_init(&kvm->slots_lock);
498 atomic_set(&kvm->users_count, 1);
500 r = kvm_init_mmu_notifier(kvm);
504 raw_spin_lock(&kvm_lock);
505 list_add(&kvm->vm_list, &vm_list);
506 raw_spin_unlock(&kvm_lock);
511 cleanup_srcu_struct(&kvm->srcu);
513 hardware_disable_all();
515 for (i = 0; i < KVM_NR_BUSES; i++)
516 kfree(kvm->buses[i]);
517 kfree(kvm->memslots);
518 kvm_arch_free_vm(kvm);
523 * Avoid using vmalloc for a small buffer.
524 * Should not be used when the size is statically known.
526 void *kvm_kvzalloc(unsigned long size)
528 if (size > PAGE_SIZE)
529 return vzalloc(size);
531 return kzalloc(size, GFP_KERNEL);
534 void kvm_kvfree(const void *addr)
536 if (is_vmalloc_addr(addr))
542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
544 if (!memslot->dirty_bitmap)
547 kvm_kvfree(memslot->dirty_bitmap);
548 memslot->dirty_bitmap = NULL;
552 * Free any memory in @free but not in @dont.
554 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
555 struct kvm_memory_slot *dont)
557 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
558 kvm_destroy_dirty_bitmap(free);
560 kvm_arch_free_memslot(free, dont);
565 void kvm_free_physmem(struct kvm *kvm)
567 struct kvm_memslots *slots = kvm->memslots;
568 struct kvm_memory_slot *memslot;
570 kvm_for_each_memslot(memslot, slots)
571 kvm_free_physmem_slot(memslot, NULL);
573 kfree(kvm->memslots);
576 static void kvm_destroy_vm(struct kvm *kvm)
579 struct mm_struct *mm = kvm->mm;
581 kvm_arch_sync_events(kvm);
582 raw_spin_lock(&kvm_lock);
583 list_del(&kvm->vm_list);
584 raw_spin_unlock(&kvm_lock);
585 kvm_free_irq_routing(kvm);
586 for (i = 0; i < KVM_NR_BUSES; i++)
587 kvm_io_bus_destroy(kvm->buses[i]);
588 kvm_coalesced_mmio_free(kvm);
589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
590 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
592 kvm_arch_flush_shadow_all(kvm);
594 kvm_arch_destroy_vm(kvm);
595 kvm_free_physmem(kvm);
596 cleanup_srcu_struct(&kvm->srcu);
597 kvm_arch_free_vm(kvm);
598 hardware_disable_all();
602 void kvm_get_kvm(struct kvm *kvm)
604 atomic_inc(&kvm->users_count);
606 EXPORT_SYMBOL_GPL(kvm_get_kvm);
608 void kvm_put_kvm(struct kvm *kvm)
610 if (atomic_dec_and_test(&kvm->users_count))
613 EXPORT_SYMBOL_GPL(kvm_put_kvm);
616 static int kvm_vm_release(struct inode *inode, struct file *filp)
618 struct kvm *kvm = filp->private_data;
620 kvm_irqfd_release(kvm);
627 * Allocation size is twice as large as the actual dirty bitmap size.
628 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
630 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
633 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
635 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
636 if (!memslot->dirty_bitmap)
639 #endif /* !CONFIG_S390 */
643 static int cmp_memslot(const void *slot1, const void *slot2)
645 struct kvm_memory_slot *s1, *s2;
647 s1 = (struct kvm_memory_slot *)slot1;
648 s2 = (struct kvm_memory_slot *)slot2;
650 if (s1->npages < s2->npages)
652 if (s1->npages > s2->npages)
659 * Sort the memslots base on its size, so the larger slots
660 * will get better fit.
662 static void sort_memslots(struct kvm_memslots *slots)
666 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
667 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
669 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
670 slots->id_to_index[slots->memslots[i].id] = i;
673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
677 struct kvm_memory_slot *old = id_to_memslot(slots, id);
678 unsigned long npages = old->npages;
681 if (new->npages != npages)
682 sort_memslots(slots);
688 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
690 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
692 #ifdef KVM_CAP_READONLY_MEM
693 valid_flags |= KVM_MEM_READONLY;
696 if (mem->flags & ~valid_flags)
703 * Allocate some memory and give it an address in the guest physical address
706 * Discontiguous memory is allowed, mostly for framebuffers.
708 * Must be called holding mmap_sem for write.
710 int __kvm_set_memory_region(struct kvm *kvm,
711 struct kvm_userspace_memory_region *mem,
716 unsigned long npages;
717 struct kvm_memory_slot *memslot, *slot;
718 struct kvm_memory_slot old, new;
719 struct kvm_memslots *slots = NULL, *old_memslots;
721 r = check_memory_region_flags(mem);
726 /* General sanity checks */
727 if (mem->memory_size & (PAGE_SIZE - 1))
729 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
731 /* We can read the guest memory with __xxx_user() later on. */
733 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
734 !access_ok(VERIFY_WRITE,
735 (void __user *)(unsigned long)mem->userspace_addr,
738 if (mem->slot >= KVM_MEM_SLOTS_NUM)
740 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
743 memslot = id_to_memslot(kvm->memslots, mem->slot);
744 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
745 npages = mem->memory_size >> PAGE_SHIFT;
748 if (npages > KVM_MEM_MAX_NR_PAGES)
752 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
754 new = old = *memslot;
757 new.base_gfn = base_gfn;
759 new.flags = mem->flags;
762 * Disallow changing a memory slot's size or changing anything about
763 * zero sized slots that doesn't involve making them non-zero.
766 if (npages && old.npages && npages != old.npages)
768 if (!npages && !old.npages)
771 /* Check for overlaps */
773 kvm_for_each_memslot(slot, kvm->memslots) {
774 if (slot->id >= KVM_USER_MEM_SLOTS || slot == memslot)
776 if (!((base_gfn + npages <= slot->base_gfn) ||
777 (base_gfn >= slot->base_gfn + slot->npages)))
781 /* Free page dirty bitmap if unneeded */
782 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
783 new.dirty_bitmap = NULL;
788 * Allocate if a slot is being created. If modifying a slot,
789 * the userspace_addr cannot change.
792 new.user_alloc = user_alloc;
793 new.userspace_addr = mem->userspace_addr;
795 if (kvm_arch_create_memslot(&new, npages))
797 } else if (npages && mem->userspace_addr != old.userspace_addr) {
802 /* Allocate page dirty bitmap if needed */
803 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
804 if (kvm_create_dirty_bitmap(&new) < 0)
806 /* destroy any largepage mappings for dirty tracking */
809 if (!npages || base_gfn != old.base_gfn) {
810 struct kvm_memory_slot *slot;
813 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
817 slot = id_to_memslot(slots, mem->slot);
818 slot->flags |= KVM_MEMSLOT_INVALID;
820 update_memslots(slots, NULL);
822 old_memslots = kvm->memslots;
823 rcu_assign_pointer(kvm->memslots, slots);
824 synchronize_srcu_expedited(&kvm->srcu);
825 /* slot was deleted or moved, clear iommu mapping */
826 kvm_iommu_unmap_pages(kvm, &old);
827 /* From this point no new shadow pages pointing to a deleted,
828 * or moved, memslot will be created.
830 * validation of sp->gfn happens in:
831 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
832 * - kvm_is_visible_gfn (mmu_check_roots)
834 kvm_arch_flush_shadow_memslot(kvm, slot);
835 slots = old_memslots;
838 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
844 * We can re-use the old_memslots from above, the only difference
845 * from the currently installed memslots is the invalid flag. This
846 * will get overwritten by update_memslots anyway.
849 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
855 /* map new memory slot into the iommu */
857 r = kvm_iommu_map_pages(kvm, &new);
862 /* actual memory is freed via old in kvm_free_physmem_slot below */
864 new.dirty_bitmap = NULL;
865 memset(&new.arch, 0, sizeof(new.arch));
868 update_memslots(slots, &new);
869 old_memslots = kvm->memslots;
870 rcu_assign_pointer(kvm->memslots, slots);
871 synchronize_srcu_expedited(&kvm->srcu);
873 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
875 kvm_free_physmem_slot(&old, &new);
883 kvm_free_physmem_slot(&new, &old);
888 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
890 int kvm_set_memory_region(struct kvm *kvm,
891 struct kvm_userspace_memory_region *mem,
896 mutex_lock(&kvm->slots_lock);
897 r = __kvm_set_memory_region(kvm, mem, user_alloc);
898 mutex_unlock(&kvm->slots_lock);
901 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
903 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
905 kvm_userspace_memory_region *mem,
908 if (mem->slot >= KVM_USER_MEM_SLOTS)
910 return kvm_set_memory_region(kvm, mem, user_alloc);
913 int kvm_get_dirty_log(struct kvm *kvm,
914 struct kvm_dirty_log *log, int *is_dirty)
916 struct kvm_memory_slot *memslot;
919 unsigned long any = 0;
922 if (log->slot >= KVM_USER_MEM_SLOTS)
925 memslot = id_to_memslot(kvm->memslots, log->slot);
927 if (!memslot->dirty_bitmap)
930 n = kvm_dirty_bitmap_bytes(memslot);
932 for (i = 0; !any && i < n/sizeof(long); ++i)
933 any = memslot->dirty_bitmap[i];
936 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
947 bool kvm_largepages_enabled(void)
949 return largepages_enabled;
952 void kvm_disable_largepages(void)
954 largepages_enabled = false;
956 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
958 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
960 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
962 EXPORT_SYMBOL_GPL(gfn_to_memslot);
964 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
966 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
968 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
969 memslot->flags & KVM_MEMSLOT_INVALID)
974 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
976 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
978 struct vm_area_struct *vma;
979 unsigned long addr, size;
983 addr = gfn_to_hva(kvm, gfn);
984 if (kvm_is_error_hva(addr))
987 down_read(¤t->mm->mmap_sem);
988 vma = find_vma(current->mm, addr);
992 size = vma_kernel_pagesize(vma);
995 up_read(¤t->mm->mmap_sem);
1000 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1002 return slot->flags & KVM_MEM_READONLY;
1005 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1006 gfn_t *nr_pages, bool write)
1008 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1009 return KVM_HVA_ERR_BAD;
1011 if (memslot_is_readonly(slot) && write)
1012 return KVM_HVA_ERR_RO_BAD;
1015 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1017 return __gfn_to_hva_memslot(slot, gfn);
1020 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1023 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1026 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1029 return gfn_to_hva_many(slot, gfn, NULL);
1031 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1033 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1035 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1037 EXPORT_SYMBOL_GPL(gfn_to_hva);
1040 * The hva returned by this function is only allowed to be read.
1041 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1043 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1045 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1048 static int kvm_read_hva(void *data, void __user *hva, int len)
1050 return __copy_from_user(data, hva, len);
1053 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1055 return __copy_from_user_inatomic(data, hva, len);
1058 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1059 unsigned long start, int write, struct page **page)
1061 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1064 flags |= FOLL_WRITE;
1066 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1069 static inline int check_user_page_hwpoison(unsigned long addr)
1071 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1073 rc = __get_user_pages(current, current->mm, addr, 1,
1074 flags, NULL, NULL, NULL);
1075 return rc == -EHWPOISON;
1079 * The atomic path to get the writable pfn which will be stored in @pfn,
1080 * true indicates success, otherwise false is returned.
1082 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1083 bool write_fault, bool *writable, pfn_t *pfn)
1085 struct page *page[1];
1088 if (!(async || atomic))
1092 * Fast pin a writable pfn only if it is a write fault request
1093 * or the caller allows to map a writable pfn for a read fault
1096 if (!(write_fault || writable))
1099 npages = __get_user_pages_fast(addr, 1, 1, page);
1101 *pfn = page_to_pfn(page[0]);
1112 * The slow path to get the pfn of the specified host virtual address,
1113 * 1 indicates success, -errno is returned if error is detected.
1115 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1116 bool *writable, pfn_t *pfn)
1118 struct page *page[1];
1124 *writable = write_fault;
1127 down_read(¤t->mm->mmap_sem);
1128 npages = get_user_page_nowait(current, current->mm,
1129 addr, write_fault, page);
1130 up_read(¤t->mm->mmap_sem);
1132 npages = get_user_pages_fast(addr, 1, write_fault,
1137 /* map read fault as writable if possible */
1138 if (unlikely(!write_fault) && writable) {
1139 struct page *wpage[1];
1141 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1150 *pfn = page_to_pfn(page[0]);
1154 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1156 if (unlikely(!(vma->vm_flags & VM_READ)))
1159 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1166 * Pin guest page in memory and return its pfn.
1167 * @addr: host virtual address which maps memory to the guest
1168 * @atomic: whether this function can sleep
1169 * @async: whether this function need to wait IO complete if the
1170 * host page is not in the memory
1171 * @write_fault: whether we should get a writable host page
1172 * @writable: whether it allows to map a writable host page for !@write_fault
1174 * The function will map a writable host page for these two cases:
1175 * 1): @write_fault = true
1176 * 2): @write_fault = false && @writable, @writable will tell the caller
1177 * whether the mapping is writable.
1179 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1180 bool write_fault, bool *writable)
1182 struct vm_area_struct *vma;
1186 /* we can do it either atomically or asynchronously, not both */
1187 BUG_ON(atomic && async);
1189 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1193 return KVM_PFN_ERR_FAULT;
1195 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1199 down_read(¤t->mm->mmap_sem);
1200 if (npages == -EHWPOISON ||
1201 (!async && check_user_page_hwpoison(addr))) {
1202 pfn = KVM_PFN_ERR_HWPOISON;
1206 vma = find_vma_intersection(current->mm, addr, addr + 1);
1209 pfn = KVM_PFN_ERR_FAULT;
1210 else if ((vma->vm_flags & VM_PFNMAP)) {
1211 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1213 BUG_ON(!kvm_is_mmio_pfn(pfn));
1215 if (async && vma_is_valid(vma, write_fault))
1217 pfn = KVM_PFN_ERR_FAULT;
1220 up_read(¤t->mm->mmap_sem);
1225 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1226 bool *async, bool write_fault, bool *writable)
1228 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1230 if (addr == KVM_HVA_ERR_RO_BAD)
1231 return KVM_PFN_ERR_RO_FAULT;
1233 if (kvm_is_error_hva(addr))
1234 return KVM_PFN_NOSLOT;
1236 /* Do not map writable pfn in the readonly memslot. */
1237 if (writable && memslot_is_readonly(slot)) {
1242 return hva_to_pfn(addr, atomic, async, write_fault,
1246 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1247 bool write_fault, bool *writable)
1249 struct kvm_memory_slot *slot;
1254 slot = gfn_to_memslot(kvm, gfn);
1256 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1260 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1262 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1264 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1266 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1267 bool write_fault, bool *writable)
1269 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1271 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1273 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1275 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1277 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1279 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1282 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1284 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1286 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1288 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1291 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1293 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1295 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1297 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1303 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1304 if (kvm_is_error_hva(addr))
1307 if (entry < nr_pages)
1310 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1312 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1314 static struct page *kvm_pfn_to_page(pfn_t pfn)
1316 if (is_error_noslot_pfn(pfn))
1317 return KVM_ERR_PTR_BAD_PAGE;
1319 if (kvm_is_mmio_pfn(pfn)) {
1321 return KVM_ERR_PTR_BAD_PAGE;
1324 return pfn_to_page(pfn);
1327 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1331 pfn = gfn_to_pfn(kvm, gfn);
1333 return kvm_pfn_to_page(pfn);
1336 EXPORT_SYMBOL_GPL(gfn_to_page);
1338 void kvm_release_page_clean(struct page *page)
1340 WARN_ON(is_error_page(page));
1342 kvm_release_pfn_clean(page_to_pfn(page));
1344 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1346 void kvm_release_pfn_clean(pfn_t pfn)
1348 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1349 put_page(pfn_to_page(pfn));
1351 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1353 void kvm_release_page_dirty(struct page *page)
1355 WARN_ON(is_error_page(page));
1357 kvm_release_pfn_dirty(page_to_pfn(page));
1359 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1361 void kvm_release_pfn_dirty(pfn_t pfn)
1363 kvm_set_pfn_dirty(pfn);
1364 kvm_release_pfn_clean(pfn);
1366 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1368 void kvm_set_page_dirty(struct page *page)
1370 kvm_set_pfn_dirty(page_to_pfn(page));
1372 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1374 void kvm_set_pfn_dirty(pfn_t pfn)
1376 if (!kvm_is_mmio_pfn(pfn)) {
1377 struct page *page = pfn_to_page(pfn);
1378 if (!PageReserved(page))
1382 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1384 void kvm_set_pfn_accessed(pfn_t pfn)
1386 if (!kvm_is_mmio_pfn(pfn))
1387 mark_page_accessed(pfn_to_page(pfn));
1389 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1391 void kvm_get_pfn(pfn_t pfn)
1393 if (!kvm_is_mmio_pfn(pfn))
1394 get_page(pfn_to_page(pfn));
1396 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1398 static int next_segment(unsigned long len, int offset)
1400 if (len > PAGE_SIZE - offset)
1401 return PAGE_SIZE - offset;
1406 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1412 addr = gfn_to_hva_read(kvm, gfn);
1413 if (kvm_is_error_hva(addr))
1415 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1420 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1422 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1424 gfn_t gfn = gpa >> PAGE_SHIFT;
1426 int offset = offset_in_page(gpa);
1429 while ((seg = next_segment(len, offset)) != 0) {
1430 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1440 EXPORT_SYMBOL_GPL(kvm_read_guest);
1442 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1447 gfn_t gfn = gpa >> PAGE_SHIFT;
1448 int offset = offset_in_page(gpa);
1450 addr = gfn_to_hva_read(kvm, gfn);
1451 if (kvm_is_error_hva(addr))
1453 pagefault_disable();
1454 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1460 EXPORT_SYMBOL(kvm_read_guest_atomic);
1462 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1463 int offset, int len)
1468 addr = gfn_to_hva(kvm, gfn);
1469 if (kvm_is_error_hva(addr))
1471 r = __copy_to_user((void __user *)addr + offset, data, len);
1474 mark_page_dirty(kvm, gfn);
1477 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1479 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1482 gfn_t gfn = gpa >> PAGE_SHIFT;
1484 int offset = offset_in_page(gpa);
1487 while ((seg = next_segment(len, offset)) != 0) {
1488 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1499 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1502 struct kvm_memslots *slots = kvm_memslots(kvm);
1503 int offset = offset_in_page(gpa);
1504 gfn_t gfn = gpa >> PAGE_SHIFT;
1507 ghc->generation = slots->generation;
1508 ghc->memslot = gfn_to_memslot(kvm, gfn);
1509 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1510 if (!kvm_is_error_hva(ghc->hva))
1517 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1519 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1520 void *data, unsigned long len)
1522 struct kvm_memslots *slots = kvm_memslots(kvm);
1525 if (slots->generation != ghc->generation)
1526 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1528 if (kvm_is_error_hva(ghc->hva))
1531 r = __copy_to_user((void __user *)ghc->hva, data, len);
1534 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1538 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1540 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1541 void *data, unsigned long len)
1543 struct kvm_memslots *slots = kvm_memslots(kvm);
1546 if (slots->generation != ghc->generation)
1547 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1549 if (kvm_is_error_hva(ghc->hva))
1552 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1558 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1560 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1562 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1565 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1567 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1569 gfn_t gfn = gpa >> PAGE_SHIFT;
1571 int offset = offset_in_page(gpa);
1574 while ((seg = next_segment(len, offset)) != 0) {
1575 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1584 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1586 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1589 if (memslot && memslot->dirty_bitmap) {
1590 unsigned long rel_gfn = gfn - memslot->base_gfn;
1592 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1596 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1598 struct kvm_memory_slot *memslot;
1600 memslot = gfn_to_memslot(kvm, gfn);
1601 mark_page_dirty_in_slot(kvm, memslot, gfn);
1605 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1607 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1612 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1614 if (kvm_arch_vcpu_runnable(vcpu)) {
1615 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1618 if (kvm_cpu_has_pending_timer(vcpu))
1620 if (signal_pending(current))
1626 finish_wait(&vcpu->wq, &wait);
1631 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1633 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1636 int cpu = vcpu->cpu;
1637 wait_queue_head_t *wqp;
1639 wqp = kvm_arch_vcpu_wq(vcpu);
1640 if (waitqueue_active(wqp)) {
1641 wake_up_interruptible(wqp);
1642 ++vcpu->stat.halt_wakeup;
1646 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1647 if (kvm_arch_vcpu_should_kick(vcpu))
1648 smp_send_reschedule(cpu);
1651 #endif /* !CONFIG_S390 */
1653 void kvm_resched(struct kvm_vcpu *vcpu)
1655 if (!need_resched())
1659 EXPORT_SYMBOL_GPL(kvm_resched);
1661 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1664 struct task_struct *task = NULL;
1667 pid = rcu_dereference(target->pid);
1669 task = get_pid_task(target->pid, PIDTYPE_PID);
1673 if (task->flags & PF_VCPU) {
1674 put_task_struct(task);
1677 if (yield_to(task, 1)) {
1678 put_task_struct(task);
1681 put_task_struct(task);
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1686 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1688 * Helper that checks whether a VCPU is eligible for directed yield.
1689 * Most eligible candidate to yield is decided by following heuristics:
1691 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1692 * (preempted lock holder), indicated by @in_spin_loop.
1693 * Set at the beiginning and cleared at the end of interception/PLE handler.
1695 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1696 * chance last time (mostly it has become eligible now since we have probably
1697 * yielded to lockholder in last iteration. This is done by toggling
1698 * @dy_eligible each time a VCPU checked for eligibility.)
1700 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1701 * to preempted lock-holder could result in wrong VCPU selection and CPU
1702 * burning. Giving priority for a potential lock-holder increases lock
1705 * Since algorithm is based on heuristics, accessing another VCPU data without
1706 * locking does not harm. It may result in trying to yield to same VCPU, fail
1707 * and continue with next VCPU and so on.
1709 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1713 eligible = !vcpu->spin_loop.in_spin_loop ||
1714 (vcpu->spin_loop.in_spin_loop &&
1715 vcpu->spin_loop.dy_eligible);
1717 if (vcpu->spin_loop.in_spin_loop)
1718 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1723 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1725 struct kvm *kvm = me->kvm;
1726 struct kvm_vcpu *vcpu;
1727 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1732 kvm_vcpu_set_in_spin_loop(me, true);
1734 * We boost the priority of a VCPU that is runnable but not
1735 * currently running, because it got preempted by something
1736 * else and called schedule in __vcpu_run. Hopefully that
1737 * VCPU is holding the lock that we need and will release it.
1738 * We approximate round-robin by starting at the last boosted VCPU.
1740 for (pass = 0; pass < 2 && !yielded; pass++) {
1741 kvm_for_each_vcpu(i, vcpu, kvm) {
1742 if (!pass && i <= last_boosted_vcpu) {
1743 i = last_boosted_vcpu;
1745 } else if (pass && i > last_boosted_vcpu)
1749 if (waitqueue_active(&vcpu->wq))
1751 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1753 if (kvm_vcpu_yield_to(vcpu)) {
1754 kvm->last_boosted_vcpu = i;
1760 kvm_vcpu_set_in_spin_loop(me, false);
1762 /* Ensure vcpu is not eligible during next spinloop */
1763 kvm_vcpu_set_dy_eligible(me, false);
1765 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1767 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1769 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1772 if (vmf->pgoff == 0)
1773 page = virt_to_page(vcpu->run);
1775 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1776 page = virt_to_page(vcpu->arch.pio_data);
1778 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1779 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1780 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1783 return kvm_arch_vcpu_fault(vcpu, vmf);
1789 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1790 .fault = kvm_vcpu_fault,
1793 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1795 vma->vm_ops = &kvm_vcpu_vm_ops;
1799 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1801 struct kvm_vcpu *vcpu = filp->private_data;
1803 kvm_put_kvm(vcpu->kvm);
1807 static struct file_operations kvm_vcpu_fops = {
1808 .release = kvm_vcpu_release,
1809 .unlocked_ioctl = kvm_vcpu_ioctl,
1810 #ifdef CONFIG_COMPAT
1811 .compat_ioctl = kvm_vcpu_compat_ioctl,
1813 .mmap = kvm_vcpu_mmap,
1814 .llseek = noop_llseek,
1818 * Allocates an inode for the vcpu.
1820 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1822 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1826 * Creates some virtual cpus. Good luck creating more than one.
1828 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1831 struct kvm_vcpu *vcpu, *v;
1833 vcpu = kvm_arch_vcpu_create(kvm, id);
1835 return PTR_ERR(vcpu);
1837 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1839 r = kvm_arch_vcpu_setup(vcpu);
1843 mutex_lock(&kvm->lock);
1844 if (!kvm_vcpu_compatible(vcpu)) {
1846 goto unlock_vcpu_destroy;
1848 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1850 goto unlock_vcpu_destroy;
1853 kvm_for_each_vcpu(r, v, kvm)
1854 if (v->vcpu_id == id) {
1856 goto unlock_vcpu_destroy;
1859 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1861 /* Now it's all set up, let userspace reach it */
1863 r = create_vcpu_fd(vcpu);
1866 goto unlock_vcpu_destroy;
1869 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1871 atomic_inc(&kvm->online_vcpus);
1873 mutex_unlock(&kvm->lock);
1874 kvm_arch_vcpu_postcreate(vcpu);
1877 unlock_vcpu_destroy:
1878 mutex_unlock(&kvm->lock);
1880 kvm_arch_vcpu_destroy(vcpu);
1884 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1887 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1888 vcpu->sigset_active = 1;
1889 vcpu->sigset = *sigset;
1891 vcpu->sigset_active = 0;
1895 static long kvm_vcpu_ioctl(struct file *filp,
1896 unsigned int ioctl, unsigned long arg)
1898 struct kvm_vcpu *vcpu = filp->private_data;
1899 void __user *argp = (void __user *)arg;
1901 struct kvm_fpu *fpu = NULL;
1902 struct kvm_sregs *kvm_sregs = NULL;
1904 if (vcpu->kvm->mm != current->mm)
1907 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1909 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1910 * so vcpu_load() would break it.
1912 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1913 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1917 r = vcpu_load(vcpu);
1925 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1926 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1928 case KVM_GET_REGS: {
1929 struct kvm_regs *kvm_regs;
1932 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1935 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1939 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1946 case KVM_SET_REGS: {
1947 struct kvm_regs *kvm_regs;
1950 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1951 if (IS_ERR(kvm_regs)) {
1952 r = PTR_ERR(kvm_regs);
1955 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1959 case KVM_GET_SREGS: {
1960 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1964 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1968 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1973 case KVM_SET_SREGS: {
1974 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1975 if (IS_ERR(kvm_sregs)) {
1976 r = PTR_ERR(kvm_sregs);
1980 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1983 case KVM_GET_MP_STATE: {
1984 struct kvm_mp_state mp_state;
1986 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1990 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1995 case KVM_SET_MP_STATE: {
1996 struct kvm_mp_state mp_state;
1999 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2001 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2004 case KVM_TRANSLATE: {
2005 struct kvm_translation tr;
2008 if (copy_from_user(&tr, argp, sizeof tr))
2010 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2014 if (copy_to_user(argp, &tr, sizeof tr))
2019 case KVM_SET_GUEST_DEBUG: {
2020 struct kvm_guest_debug dbg;
2023 if (copy_from_user(&dbg, argp, sizeof dbg))
2025 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));
2072 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2076 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2085 #ifdef CONFIG_COMPAT
2086 static long kvm_vcpu_compat_ioctl(struct file *filp,
2087 unsigned int ioctl, unsigned long arg)
2089 struct kvm_vcpu *vcpu = filp->private_data;
2090 void __user *argp = compat_ptr(arg);
2093 if (vcpu->kvm->mm != current->mm)
2097 case KVM_SET_SIGNAL_MASK: {
2098 struct kvm_signal_mask __user *sigmask_arg = argp;
2099 struct kvm_signal_mask kvm_sigmask;
2100 compat_sigset_t csigset;
2105 if (copy_from_user(&kvm_sigmask, argp,
2106 sizeof kvm_sigmask))
2109 if (kvm_sigmask.len != sizeof csigset)
2112 if (copy_from_user(&csigset, sigmask_arg->sigset,
2115 sigset_from_compat(&sigset, &csigset);
2116 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2118 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2122 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2130 static long kvm_vm_ioctl(struct file *filp,
2131 unsigned int ioctl, unsigned long arg)
2133 struct kvm *kvm = filp->private_data;
2134 void __user *argp = (void __user *)arg;
2137 if (kvm->mm != current->mm)
2140 case KVM_CREATE_VCPU:
2141 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2143 case KVM_SET_USER_MEMORY_REGION: {
2144 struct kvm_userspace_memory_region kvm_userspace_mem;
2147 if (copy_from_user(&kvm_userspace_mem, argp,
2148 sizeof kvm_userspace_mem))
2151 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2154 case KVM_GET_DIRTY_LOG: {
2155 struct kvm_dirty_log log;
2158 if (copy_from_user(&log, argp, sizeof log))
2160 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2163 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2164 case KVM_REGISTER_COALESCED_MMIO: {
2165 struct kvm_coalesced_mmio_zone zone;
2167 if (copy_from_user(&zone, argp, sizeof zone))
2169 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2172 case KVM_UNREGISTER_COALESCED_MMIO: {
2173 struct kvm_coalesced_mmio_zone zone;
2175 if (copy_from_user(&zone, argp, sizeof zone))
2177 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2182 struct kvm_irqfd data;
2185 if (copy_from_user(&data, argp, sizeof data))
2187 r = kvm_irqfd(kvm, &data);
2190 case KVM_IOEVENTFD: {
2191 struct kvm_ioeventfd data;
2194 if (copy_from_user(&data, argp, sizeof data))
2196 r = kvm_ioeventfd(kvm, &data);
2199 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2200 case KVM_SET_BOOT_CPU_ID:
2202 mutex_lock(&kvm->lock);
2203 if (atomic_read(&kvm->online_vcpus) != 0)
2206 kvm->bsp_vcpu_id = arg;
2207 mutex_unlock(&kvm->lock);
2210 #ifdef CONFIG_HAVE_KVM_MSI
2211 case KVM_SIGNAL_MSI: {
2215 if (copy_from_user(&msi, argp, sizeof msi))
2217 r = kvm_send_userspace_msi(kvm, &msi);
2221 #ifdef __KVM_HAVE_IRQ_LINE
2222 case KVM_IRQ_LINE_STATUS:
2223 case KVM_IRQ_LINE: {
2224 struct kvm_irq_level irq_event;
2227 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2230 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2235 if (ioctl == KVM_IRQ_LINE_STATUS) {
2236 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2245 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2247 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2253 #ifdef CONFIG_COMPAT
2254 struct compat_kvm_dirty_log {
2258 compat_uptr_t dirty_bitmap; /* one bit per page */
2263 static long kvm_vm_compat_ioctl(struct file *filp,
2264 unsigned int ioctl, unsigned long arg)
2266 struct kvm *kvm = filp->private_data;
2269 if (kvm->mm != current->mm)
2272 case KVM_GET_DIRTY_LOG: {
2273 struct compat_kvm_dirty_log compat_log;
2274 struct kvm_dirty_log log;
2277 if (copy_from_user(&compat_log, (void __user *)arg,
2278 sizeof(compat_log)))
2280 log.slot = compat_log.slot;
2281 log.padding1 = compat_log.padding1;
2282 log.padding2 = compat_log.padding2;
2283 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2285 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2289 r = kvm_vm_ioctl(filp, ioctl, arg);
2297 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2299 struct page *page[1];
2302 gfn_t gfn = vmf->pgoff;
2303 struct kvm *kvm = vma->vm_file->private_data;
2305 addr = gfn_to_hva(kvm, gfn);
2306 if (kvm_is_error_hva(addr))
2307 return VM_FAULT_SIGBUS;
2309 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2311 if (unlikely(npages != 1))
2312 return VM_FAULT_SIGBUS;
2314 vmf->page = page[0];
2318 static const struct vm_operations_struct kvm_vm_vm_ops = {
2319 .fault = kvm_vm_fault,
2322 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2324 vma->vm_ops = &kvm_vm_vm_ops;
2328 static struct file_operations kvm_vm_fops = {
2329 .release = kvm_vm_release,
2330 .unlocked_ioctl = kvm_vm_ioctl,
2331 #ifdef CONFIG_COMPAT
2332 .compat_ioctl = kvm_vm_compat_ioctl,
2334 .mmap = kvm_vm_mmap,
2335 .llseek = noop_llseek,
2338 static int kvm_dev_ioctl_create_vm(unsigned long type)
2343 kvm = kvm_create_vm(type);
2345 return PTR_ERR(kvm);
2346 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2347 r = kvm_coalesced_mmio_init(kvm);
2353 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2360 static long kvm_dev_ioctl_check_extension_generic(long arg)
2363 case KVM_CAP_USER_MEMORY:
2364 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2365 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2366 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2367 case KVM_CAP_SET_BOOT_CPU_ID:
2369 case KVM_CAP_INTERNAL_ERROR_DATA:
2370 #ifdef CONFIG_HAVE_KVM_MSI
2371 case KVM_CAP_SIGNAL_MSI:
2374 #ifdef KVM_CAP_IRQ_ROUTING
2375 case KVM_CAP_IRQ_ROUTING:
2376 return KVM_MAX_IRQ_ROUTES;
2381 return kvm_dev_ioctl_check_extension(arg);
2384 static long kvm_dev_ioctl(struct file *filp,
2385 unsigned int ioctl, unsigned long arg)
2390 case KVM_GET_API_VERSION:
2394 r = KVM_API_VERSION;
2397 r = kvm_dev_ioctl_create_vm(arg);
2399 case KVM_CHECK_EXTENSION:
2400 r = kvm_dev_ioctl_check_extension_generic(arg);
2402 case KVM_GET_VCPU_MMAP_SIZE:
2406 r = PAGE_SIZE; /* struct kvm_run */
2408 r += PAGE_SIZE; /* pio data page */
2410 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2411 r += PAGE_SIZE; /* coalesced mmio ring page */
2414 case KVM_TRACE_ENABLE:
2415 case KVM_TRACE_PAUSE:
2416 case KVM_TRACE_DISABLE:
2420 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2426 static struct file_operations kvm_chardev_ops = {
2427 .unlocked_ioctl = kvm_dev_ioctl,
2428 .compat_ioctl = kvm_dev_ioctl,
2429 .llseek = noop_llseek,
2432 static struct miscdevice kvm_dev = {
2438 static void hardware_enable_nolock(void *junk)
2440 int cpu = raw_smp_processor_id();
2443 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2446 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2448 r = kvm_arch_hardware_enable(NULL);
2451 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2452 atomic_inc(&hardware_enable_failed);
2453 printk(KERN_INFO "kvm: enabling virtualization on "
2454 "CPU%d failed\n", cpu);
2458 static void hardware_enable(void *junk)
2460 raw_spin_lock(&kvm_lock);
2461 hardware_enable_nolock(junk);
2462 raw_spin_unlock(&kvm_lock);
2465 static void hardware_disable_nolock(void *junk)
2467 int cpu = raw_smp_processor_id();
2469 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2471 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2472 kvm_arch_hardware_disable(NULL);
2475 static void hardware_disable(void *junk)
2477 raw_spin_lock(&kvm_lock);
2478 hardware_disable_nolock(junk);
2479 raw_spin_unlock(&kvm_lock);
2482 static void hardware_disable_all_nolock(void)
2484 BUG_ON(!kvm_usage_count);
2487 if (!kvm_usage_count)
2488 on_each_cpu(hardware_disable_nolock, NULL, 1);
2491 static void hardware_disable_all(void)
2493 raw_spin_lock(&kvm_lock);
2494 hardware_disable_all_nolock();
2495 raw_spin_unlock(&kvm_lock);
2498 static int hardware_enable_all(void)
2502 raw_spin_lock(&kvm_lock);
2505 if (kvm_usage_count == 1) {
2506 atomic_set(&hardware_enable_failed, 0);
2507 on_each_cpu(hardware_enable_nolock, NULL, 1);
2509 if (atomic_read(&hardware_enable_failed)) {
2510 hardware_disable_all_nolock();
2515 raw_spin_unlock(&kvm_lock);
2520 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2525 if (!kvm_usage_count)
2528 val &= ~CPU_TASKS_FROZEN;
2531 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2533 hardware_disable(NULL);
2536 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2538 hardware_enable(NULL);
2545 asmlinkage void kvm_spurious_fault(void)
2547 /* Fault while not rebooting. We want the trace. */
2550 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2552 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2556 * Some (well, at least mine) BIOSes hang on reboot if
2559 * And Intel TXT required VMX off for all cpu when system shutdown.
2561 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2562 kvm_rebooting = true;
2563 on_each_cpu(hardware_disable_nolock, NULL, 1);
2567 static struct notifier_block kvm_reboot_notifier = {
2568 .notifier_call = kvm_reboot,
2572 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2576 for (i = 0; i < bus->dev_count; i++) {
2577 struct kvm_io_device *pos = bus->range[i].dev;
2579 kvm_iodevice_destructor(pos);
2584 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2586 const struct kvm_io_range *r1 = p1;
2587 const struct kvm_io_range *r2 = p2;
2589 if (r1->addr < r2->addr)
2591 if (r1->addr + r1->len > r2->addr + r2->len)
2596 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2597 gpa_t addr, int len)
2599 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2605 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2606 kvm_io_bus_sort_cmp, NULL);
2611 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2612 gpa_t addr, int len)
2614 struct kvm_io_range *range, key;
2617 key = (struct kvm_io_range) {
2622 range = bsearch(&key, bus->range, bus->dev_count,
2623 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2627 off = range - bus->range;
2629 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2635 /* kvm_io_bus_write - called under kvm->slots_lock */
2636 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2637 int len, const void *val)
2640 struct kvm_io_bus *bus;
2641 struct kvm_io_range range;
2643 range = (struct kvm_io_range) {
2648 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2649 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2653 while (idx < bus->dev_count &&
2654 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2655 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2663 /* kvm_io_bus_read - called under kvm->slots_lock */
2664 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2668 struct kvm_io_bus *bus;
2669 struct kvm_io_range range;
2671 range = (struct kvm_io_range) {
2676 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2677 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2681 while (idx < bus->dev_count &&
2682 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2683 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2691 /* Caller must hold slots_lock. */
2692 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2693 int len, struct kvm_io_device *dev)
2695 struct kvm_io_bus *new_bus, *bus;
2697 bus = kvm->buses[bus_idx];
2698 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2701 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2702 sizeof(struct kvm_io_range)), GFP_KERNEL);
2705 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2706 sizeof(struct kvm_io_range)));
2707 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2708 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2709 synchronize_srcu_expedited(&kvm->srcu);
2715 /* Caller must hold slots_lock. */
2716 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2717 struct kvm_io_device *dev)
2720 struct kvm_io_bus *new_bus, *bus;
2722 bus = kvm->buses[bus_idx];
2724 for (i = 0; i < bus->dev_count; i++)
2725 if (bus->range[i].dev == dev) {
2733 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2734 sizeof(struct kvm_io_range)), GFP_KERNEL);
2738 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2739 new_bus->dev_count--;
2740 memcpy(new_bus->range + i, bus->range + i + 1,
2741 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2743 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2744 synchronize_srcu_expedited(&kvm->srcu);
2749 static struct notifier_block kvm_cpu_notifier = {
2750 .notifier_call = kvm_cpu_hotplug,
2753 static int vm_stat_get(void *_offset, u64 *val)
2755 unsigned offset = (long)_offset;
2759 raw_spin_lock(&kvm_lock);
2760 list_for_each_entry(kvm, &vm_list, vm_list)
2761 *val += *(u32 *)((void *)kvm + offset);
2762 raw_spin_unlock(&kvm_lock);
2766 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2768 static int vcpu_stat_get(void *_offset, u64 *val)
2770 unsigned offset = (long)_offset;
2772 struct kvm_vcpu *vcpu;
2776 raw_spin_lock(&kvm_lock);
2777 list_for_each_entry(kvm, &vm_list, vm_list)
2778 kvm_for_each_vcpu(i, vcpu, kvm)
2779 *val += *(u32 *)((void *)vcpu + offset);
2781 raw_spin_unlock(&kvm_lock);
2785 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2787 static const struct file_operations *stat_fops[] = {
2788 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2789 [KVM_STAT_VM] = &vm_stat_fops,
2792 static int kvm_init_debug(void)
2795 struct kvm_stats_debugfs_item *p;
2797 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2798 if (kvm_debugfs_dir == NULL)
2801 for (p = debugfs_entries; p->name; ++p) {
2802 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2803 (void *)(long)p->offset,
2804 stat_fops[p->kind]);
2805 if (p->dentry == NULL)
2812 debugfs_remove_recursive(kvm_debugfs_dir);
2817 static void kvm_exit_debug(void)
2819 struct kvm_stats_debugfs_item *p;
2821 for (p = debugfs_entries; p->name; ++p)
2822 debugfs_remove(p->dentry);
2823 debugfs_remove(kvm_debugfs_dir);
2826 static int kvm_suspend(void)
2828 if (kvm_usage_count)
2829 hardware_disable_nolock(NULL);
2833 static void kvm_resume(void)
2835 if (kvm_usage_count) {
2836 WARN_ON(raw_spin_is_locked(&kvm_lock));
2837 hardware_enable_nolock(NULL);
2841 static struct syscore_ops kvm_syscore_ops = {
2842 .suspend = kvm_suspend,
2843 .resume = kvm_resume,
2847 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2849 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2852 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2854 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2856 kvm_arch_vcpu_load(vcpu, cpu);
2859 static void kvm_sched_out(struct preempt_notifier *pn,
2860 struct task_struct *next)
2862 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2864 kvm_arch_vcpu_put(vcpu);
2867 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2868 struct module *module)
2873 r = kvm_arch_init(opaque);
2877 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2882 r = kvm_arch_hardware_setup();
2886 for_each_online_cpu(cpu) {
2887 smp_call_function_single(cpu,
2888 kvm_arch_check_processor_compat,
2894 r = register_cpu_notifier(&kvm_cpu_notifier);
2897 register_reboot_notifier(&kvm_reboot_notifier);
2899 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2901 vcpu_align = __alignof__(struct kvm_vcpu);
2902 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2904 if (!kvm_vcpu_cache) {
2909 r = kvm_async_pf_init();
2913 kvm_chardev_ops.owner = module;
2914 kvm_vm_fops.owner = module;
2915 kvm_vcpu_fops.owner = module;
2917 r = misc_register(&kvm_dev);
2919 printk(KERN_ERR "kvm: misc device register failed\n");
2923 register_syscore_ops(&kvm_syscore_ops);
2925 kvm_preempt_ops.sched_in = kvm_sched_in;
2926 kvm_preempt_ops.sched_out = kvm_sched_out;
2928 r = kvm_init_debug();
2930 printk(KERN_ERR "kvm: create debugfs files failed\n");
2937 unregister_syscore_ops(&kvm_syscore_ops);
2939 kvm_async_pf_deinit();
2941 kmem_cache_destroy(kvm_vcpu_cache);
2943 unregister_reboot_notifier(&kvm_reboot_notifier);
2944 unregister_cpu_notifier(&kvm_cpu_notifier);
2947 kvm_arch_hardware_unsetup();
2949 free_cpumask_var(cpus_hardware_enabled);
2955 EXPORT_SYMBOL_GPL(kvm_init);
2960 misc_deregister(&kvm_dev);
2961 kmem_cache_destroy(kvm_vcpu_cache);
2962 kvm_async_pf_deinit();
2963 unregister_syscore_ops(&kvm_syscore_ops);
2964 unregister_reboot_notifier(&kvm_reboot_notifier);
2965 unregister_cpu_notifier(&kvm_cpu_notifier);
2966 on_each_cpu(hardware_disable_nolock, NULL, 1);
2967 kvm_arch_hardware_unsetup();
2969 free_cpumask_var(cpus_hardware_enabled);
2971 EXPORT_SYMBOL_GPL(kvm_exit);