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,
680 struct kvm_memory_slot *old = id_to_memslot(slots, id);
681 unsigned long npages = old->npages;
684 if (new->npages != npages)
685 sort_memslots(slots);
688 slots->generation = last_generation + 1;
691 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
693 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
695 #ifdef KVM_CAP_READONLY_MEM
696 valid_flags |= KVM_MEM_READONLY;
699 if (mem->flags & ~valid_flags)
706 * Allocate some memory and give it an address in the guest physical address
709 * Discontiguous memory is allowed, mostly for framebuffers.
711 * Must be called holding mmap_sem for write.
713 int __kvm_set_memory_region(struct kvm *kvm,
714 struct kvm_userspace_memory_region *mem,
719 unsigned long npages;
720 struct kvm_memory_slot *memslot, *slot;
721 struct kvm_memory_slot old, new;
722 struct kvm_memslots *slots = NULL, *old_memslots;
724 r = check_memory_region_flags(mem);
729 /* General sanity checks */
730 if (mem->memory_size & (PAGE_SIZE - 1))
732 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
734 /* We can read the guest memory with __xxx_user() later on. */
736 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
737 !access_ok(VERIFY_WRITE,
738 (void __user *)(unsigned long)mem->userspace_addr,
741 if (mem->slot >= KVM_MEM_SLOTS_NUM)
743 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
746 memslot = id_to_memslot(kvm->memslots, mem->slot);
747 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
748 npages = mem->memory_size >> PAGE_SHIFT;
751 if (npages > KVM_MEM_MAX_NR_PAGES)
755 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
757 new = old = *memslot;
760 new.base_gfn = base_gfn;
762 new.flags = mem->flags;
765 * Disallow changing a memory slot's size or changing anything about
766 * zero sized slots that doesn't involve making them non-zero.
769 if (npages && old.npages && npages != old.npages)
771 if (!npages && !old.npages)
774 /* Check for overlaps */
776 kvm_for_each_memslot(slot, kvm->memslots) {
777 if (slot->id >= KVM_USER_MEM_SLOTS || slot == memslot)
779 if (!((base_gfn + npages <= slot->base_gfn) ||
780 (base_gfn >= slot->base_gfn + slot->npages)))
784 /* Free page dirty bitmap if unneeded */
785 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
786 new.dirty_bitmap = NULL;
791 * Allocate if a slot is being created. If modifying a slot,
792 * the userspace_addr cannot change.
795 new.user_alloc = user_alloc;
796 new.userspace_addr = mem->userspace_addr;
798 if (kvm_arch_create_memslot(&new, npages))
800 } else if (npages && mem->userspace_addr != old.userspace_addr) {
805 /* Allocate page dirty bitmap if needed */
806 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
807 if (kvm_create_dirty_bitmap(&new) < 0)
809 /* destroy any largepage mappings for dirty tracking */
812 if (!npages || base_gfn != old.base_gfn) {
813 struct kvm_memory_slot *slot;
816 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
820 slot = id_to_memslot(slots, mem->slot);
821 slot->flags |= KVM_MEMSLOT_INVALID;
823 update_memslots(slots, NULL, kvm->memslots->generation);
825 old_memslots = kvm->memslots;
826 rcu_assign_pointer(kvm->memslots, slots);
827 synchronize_srcu_expedited(&kvm->srcu);
828 /* slot was deleted or moved, clear iommu mapping */
829 kvm_iommu_unmap_pages(kvm, &old);
830 /* From this point no new shadow pages pointing to a deleted,
831 * or moved, memslot will be created.
833 * validation of sp->gfn happens in:
834 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
835 * - kvm_is_visible_gfn (mmu_check_roots)
837 kvm_arch_flush_shadow_memslot(kvm, slot);
838 slots = old_memslots;
841 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
847 * We can re-use the old_memslots from above, the only difference
848 * from the currently installed memslots is the invalid flag. This
849 * will get overwritten by update_memslots anyway.
852 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
858 /* map new memory slot into the iommu */
860 r = kvm_iommu_map_pages(kvm, &new);
865 /* actual memory is freed via old in kvm_free_physmem_slot below */
867 new.dirty_bitmap = NULL;
868 memset(&new.arch, 0, sizeof(new.arch));
871 update_memslots(slots, &new, kvm->memslots->generation);
872 old_memslots = kvm->memslots;
873 rcu_assign_pointer(kvm->memslots, slots);
874 synchronize_srcu_expedited(&kvm->srcu);
876 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
878 kvm_free_physmem_slot(&old, &new);
886 kvm_free_physmem_slot(&new, &old);
891 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
893 int kvm_set_memory_region(struct kvm *kvm,
894 struct kvm_userspace_memory_region *mem,
899 mutex_lock(&kvm->slots_lock);
900 r = __kvm_set_memory_region(kvm, mem, user_alloc);
901 mutex_unlock(&kvm->slots_lock);
904 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
906 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
908 kvm_userspace_memory_region *mem,
911 if (mem->slot >= KVM_USER_MEM_SLOTS)
913 return kvm_set_memory_region(kvm, mem, user_alloc);
916 int kvm_get_dirty_log(struct kvm *kvm,
917 struct kvm_dirty_log *log, int *is_dirty)
919 struct kvm_memory_slot *memslot;
922 unsigned long any = 0;
925 if (log->slot >= KVM_USER_MEM_SLOTS)
928 memslot = id_to_memslot(kvm->memslots, log->slot);
930 if (!memslot->dirty_bitmap)
933 n = kvm_dirty_bitmap_bytes(memslot);
935 for (i = 0; !any && i < n/sizeof(long); ++i)
936 any = memslot->dirty_bitmap[i];
939 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
950 bool kvm_largepages_enabled(void)
952 return largepages_enabled;
955 void kvm_disable_largepages(void)
957 largepages_enabled = false;
959 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
961 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
963 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
965 EXPORT_SYMBOL_GPL(gfn_to_memslot);
967 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
969 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
971 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
972 memslot->flags & KVM_MEMSLOT_INVALID)
977 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
979 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
981 struct vm_area_struct *vma;
982 unsigned long addr, size;
986 addr = gfn_to_hva(kvm, gfn);
987 if (kvm_is_error_hva(addr))
990 down_read(¤t->mm->mmap_sem);
991 vma = find_vma(current->mm, addr);
995 size = vma_kernel_pagesize(vma);
998 up_read(¤t->mm->mmap_sem);
1003 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1005 return slot->flags & KVM_MEM_READONLY;
1008 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1009 gfn_t *nr_pages, bool write)
1011 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1012 return KVM_HVA_ERR_BAD;
1014 if (memslot_is_readonly(slot) && write)
1015 return KVM_HVA_ERR_RO_BAD;
1018 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1020 return __gfn_to_hva_memslot(slot, gfn);
1023 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1026 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1029 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1032 return gfn_to_hva_many(slot, gfn, NULL);
1034 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1036 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1038 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1040 EXPORT_SYMBOL_GPL(gfn_to_hva);
1043 * The hva returned by this function is only allowed to be read.
1044 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1046 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1048 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1051 static int kvm_read_hva(void *data, void __user *hva, int len)
1053 return __copy_from_user(data, hva, len);
1056 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1058 return __copy_from_user_inatomic(data, hva, len);
1061 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1062 unsigned long start, int write, struct page **page)
1064 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1067 flags |= FOLL_WRITE;
1069 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1072 static inline int check_user_page_hwpoison(unsigned long addr)
1074 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1076 rc = __get_user_pages(current, current->mm, addr, 1,
1077 flags, NULL, NULL, NULL);
1078 return rc == -EHWPOISON;
1082 * The atomic path to get the writable pfn which will be stored in @pfn,
1083 * true indicates success, otherwise false is returned.
1085 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1086 bool write_fault, bool *writable, pfn_t *pfn)
1088 struct page *page[1];
1091 if (!(async || atomic))
1095 * Fast pin a writable pfn only if it is a write fault request
1096 * or the caller allows to map a writable pfn for a read fault
1099 if (!(write_fault || writable))
1102 npages = __get_user_pages_fast(addr, 1, 1, page);
1104 *pfn = page_to_pfn(page[0]);
1115 * The slow path to get the pfn of the specified host virtual address,
1116 * 1 indicates success, -errno is returned if error is detected.
1118 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1119 bool *writable, pfn_t *pfn)
1121 struct page *page[1];
1127 *writable = write_fault;
1130 down_read(¤t->mm->mmap_sem);
1131 npages = get_user_page_nowait(current, current->mm,
1132 addr, write_fault, page);
1133 up_read(¤t->mm->mmap_sem);
1135 npages = get_user_pages_fast(addr, 1, write_fault,
1140 /* map read fault as writable if possible */
1141 if (unlikely(!write_fault) && writable) {
1142 struct page *wpage[1];
1144 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1153 *pfn = page_to_pfn(page[0]);
1157 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1159 if (unlikely(!(vma->vm_flags & VM_READ)))
1162 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1169 * Pin guest page in memory and return its pfn.
1170 * @addr: host virtual address which maps memory to the guest
1171 * @atomic: whether this function can sleep
1172 * @async: whether this function need to wait IO complete if the
1173 * host page is not in the memory
1174 * @write_fault: whether we should get a writable host page
1175 * @writable: whether it allows to map a writable host page for !@write_fault
1177 * The function will map a writable host page for these two cases:
1178 * 1): @write_fault = true
1179 * 2): @write_fault = false && @writable, @writable will tell the caller
1180 * whether the mapping is writable.
1182 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1183 bool write_fault, bool *writable)
1185 struct vm_area_struct *vma;
1189 /* we can do it either atomically or asynchronously, not both */
1190 BUG_ON(atomic && async);
1192 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1196 return KVM_PFN_ERR_FAULT;
1198 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1202 down_read(¤t->mm->mmap_sem);
1203 if (npages == -EHWPOISON ||
1204 (!async && check_user_page_hwpoison(addr))) {
1205 pfn = KVM_PFN_ERR_HWPOISON;
1209 vma = find_vma_intersection(current->mm, addr, addr + 1);
1212 pfn = KVM_PFN_ERR_FAULT;
1213 else if ((vma->vm_flags & VM_PFNMAP)) {
1214 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1216 BUG_ON(!kvm_is_mmio_pfn(pfn));
1218 if (async && vma_is_valid(vma, write_fault))
1220 pfn = KVM_PFN_ERR_FAULT;
1223 up_read(¤t->mm->mmap_sem);
1228 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1229 bool *async, bool write_fault, bool *writable)
1231 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1233 if (addr == KVM_HVA_ERR_RO_BAD)
1234 return KVM_PFN_ERR_RO_FAULT;
1236 if (kvm_is_error_hva(addr))
1237 return KVM_PFN_NOSLOT;
1239 /* Do not map writable pfn in the readonly memslot. */
1240 if (writable && memslot_is_readonly(slot)) {
1245 return hva_to_pfn(addr, atomic, async, write_fault,
1249 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1250 bool write_fault, bool *writable)
1252 struct kvm_memory_slot *slot;
1257 slot = gfn_to_memslot(kvm, gfn);
1259 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1263 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1265 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1267 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1269 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1270 bool write_fault, bool *writable)
1272 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1274 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1276 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1278 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1280 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1282 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1285 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1287 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1289 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1291 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1294 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1296 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1298 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1300 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1306 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1307 if (kvm_is_error_hva(addr))
1310 if (entry < nr_pages)
1313 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1315 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1317 static struct page *kvm_pfn_to_page(pfn_t pfn)
1319 if (is_error_noslot_pfn(pfn))
1320 return KVM_ERR_PTR_BAD_PAGE;
1322 if (kvm_is_mmio_pfn(pfn)) {
1324 return KVM_ERR_PTR_BAD_PAGE;
1327 return pfn_to_page(pfn);
1330 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1334 pfn = gfn_to_pfn(kvm, gfn);
1336 return kvm_pfn_to_page(pfn);
1339 EXPORT_SYMBOL_GPL(gfn_to_page);
1341 void kvm_release_page_clean(struct page *page)
1343 WARN_ON(is_error_page(page));
1345 kvm_release_pfn_clean(page_to_pfn(page));
1347 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1349 void kvm_release_pfn_clean(pfn_t pfn)
1351 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1352 put_page(pfn_to_page(pfn));
1354 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1356 void kvm_release_page_dirty(struct page *page)
1358 WARN_ON(is_error_page(page));
1360 kvm_release_pfn_dirty(page_to_pfn(page));
1362 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1364 void kvm_release_pfn_dirty(pfn_t pfn)
1366 kvm_set_pfn_dirty(pfn);
1367 kvm_release_pfn_clean(pfn);
1369 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1371 void kvm_set_page_dirty(struct page *page)
1373 kvm_set_pfn_dirty(page_to_pfn(page));
1375 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1377 void kvm_set_pfn_dirty(pfn_t pfn)
1379 if (!kvm_is_mmio_pfn(pfn)) {
1380 struct page *page = pfn_to_page(pfn);
1381 if (!PageReserved(page))
1385 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1387 void kvm_set_pfn_accessed(pfn_t pfn)
1389 if (!kvm_is_mmio_pfn(pfn))
1390 mark_page_accessed(pfn_to_page(pfn));
1392 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1394 void kvm_get_pfn(pfn_t pfn)
1396 if (!kvm_is_mmio_pfn(pfn))
1397 get_page(pfn_to_page(pfn));
1399 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1401 static int next_segment(unsigned long len, int offset)
1403 if (len > PAGE_SIZE - offset)
1404 return PAGE_SIZE - offset;
1409 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1415 addr = gfn_to_hva_read(kvm, gfn);
1416 if (kvm_is_error_hva(addr))
1418 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1423 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1425 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1427 gfn_t gfn = gpa >> PAGE_SHIFT;
1429 int offset = offset_in_page(gpa);
1432 while ((seg = next_segment(len, offset)) != 0) {
1433 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1443 EXPORT_SYMBOL_GPL(kvm_read_guest);
1445 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1450 gfn_t gfn = gpa >> PAGE_SHIFT;
1451 int offset = offset_in_page(gpa);
1453 addr = gfn_to_hva_read(kvm, gfn);
1454 if (kvm_is_error_hva(addr))
1456 pagefault_disable();
1457 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1463 EXPORT_SYMBOL(kvm_read_guest_atomic);
1465 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1466 int offset, int len)
1471 addr = gfn_to_hva(kvm, gfn);
1472 if (kvm_is_error_hva(addr))
1474 r = __copy_to_user((void __user *)addr + offset, data, len);
1477 mark_page_dirty(kvm, gfn);
1480 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1482 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1485 gfn_t gfn = gpa >> PAGE_SHIFT;
1487 int offset = offset_in_page(gpa);
1490 while ((seg = next_segment(len, offset)) != 0) {
1491 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1502 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1505 struct kvm_memslots *slots = kvm_memslots(kvm);
1506 int offset = offset_in_page(gpa);
1507 gfn_t gfn = gpa >> PAGE_SHIFT;
1510 ghc->generation = slots->generation;
1511 ghc->memslot = gfn_to_memslot(kvm, gfn);
1512 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1513 if (!kvm_is_error_hva(ghc->hva))
1520 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1522 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1523 void *data, unsigned long len)
1525 struct kvm_memslots *slots = kvm_memslots(kvm);
1528 if (slots->generation != ghc->generation)
1529 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1531 if (kvm_is_error_hva(ghc->hva))
1534 r = __copy_to_user((void __user *)ghc->hva, data, len);
1537 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1541 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1543 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1544 void *data, unsigned long len)
1546 struct kvm_memslots *slots = kvm_memslots(kvm);
1549 if (slots->generation != ghc->generation)
1550 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1552 if (kvm_is_error_hva(ghc->hva))
1555 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1561 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1563 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1565 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1568 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1570 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1572 gfn_t gfn = gpa >> PAGE_SHIFT;
1574 int offset = offset_in_page(gpa);
1577 while ((seg = next_segment(len, offset)) != 0) {
1578 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1587 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1589 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1592 if (memslot && memslot->dirty_bitmap) {
1593 unsigned long rel_gfn = gfn - memslot->base_gfn;
1595 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1599 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1601 struct kvm_memory_slot *memslot;
1603 memslot = gfn_to_memslot(kvm, gfn);
1604 mark_page_dirty_in_slot(kvm, memslot, gfn);
1608 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1610 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1615 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1617 if (kvm_arch_vcpu_runnable(vcpu)) {
1618 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1621 if (kvm_cpu_has_pending_timer(vcpu))
1623 if (signal_pending(current))
1629 finish_wait(&vcpu->wq, &wait);
1634 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1636 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1639 int cpu = vcpu->cpu;
1640 wait_queue_head_t *wqp;
1642 wqp = kvm_arch_vcpu_wq(vcpu);
1643 if (waitqueue_active(wqp)) {
1644 wake_up_interruptible(wqp);
1645 ++vcpu->stat.halt_wakeup;
1649 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1650 if (kvm_arch_vcpu_should_kick(vcpu))
1651 smp_send_reschedule(cpu);
1654 #endif /* !CONFIG_S390 */
1656 void kvm_resched(struct kvm_vcpu *vcpu)
1658 if (!need_resched())
1662 EXPORT_SYMBOL_GPL(kvm_resched);
1664 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1667 struct task_struct *task = NULL;
1670 pid = rcu_dereference(target->pid);
1672 task = get_pid_task(target->pid, PIDTYPE_PID);
1676 if (task->flags & PF_VCPU) {
1677 put_task_struct(task);
1680 if (yield_to(task, 1)) {
1681 put_task_struct(task);
1684 put_task_struct(task);
1687 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1689 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1691 * Helper that checks whether a VCPU is eligible for directed yield.
1692 * Most eligible candidate to yield is decided by following heuristics:
1694 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1695 * (preempted lock holder), indicated by @in_spin_loop.
1696 * Set at the beiginning and cleared at the end of interception/PLE handler.
1698 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1699 * chance last time (mostly it has become eligible now since we have probably
1700 * yielded to lockholder in last iteration. This is done by toggling
1701 * @dy_eligible each time a VCPU checked for eligibility.)
1703 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1704 * to preempted lock-holder could result in wrong VCPU selection and CPU
1705 * burning. Giving priority for a potential lock-holder increases lock
1708 * Since algorithm is based on heuristics, accessing another VCPU data without
1709 * locking does not harm. It may result in trying to yield to same VCPU, fail
1710 * and continue with next VCPU and so on.
1712 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1716 eligible = !vcpu->spin_loop.in_spin_loop ||
1717 (vcpu->spin_loop.in_spin_loop &&
1718 vcpu->spin_loop.dy_eligible);
1720 if (vcpu->spin_loop.in_spin_loop)
1721 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1726 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1728 struct kvm *kvm = me->kvm;
1729 struct kvm_vcpu *vcpu;
1730 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1735 kvm_vcpu_set_in_spin_loop(me, true);
1737 * We boost the priority of a VCPU that is runnable but not
1738 * currently running, because it got preempted by something
1739 * else and called schedule in __vcpu_run. Hopefully that
1740 * VCPU is holding the lock that we need and will release it.
1741 * We approximate round-robin by starting at the last boosted VCPU.
1743 for (pass = 0; pass < 2 && !yielded; pass++) {
1744 kvm_for_each_vcpu(i, vcpu, kvm) {
1745 if (!pass && i <= last_boosted_vcpu) {
1746 i = last_boosted_vcpu;
1748 } else if (pass && i > last_boosted_vcpu)
1752 if (waitqueue_active(&vcpu->wq))
1754 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1756 if (kvm_vcpu_yield_to(vcpu)) {
1757 kvm->last_boosted_vcpu = i;
1763 kvm_vcpu_set_in_spin_loop(me, false);
1765 /* Ensure vcpu is not eligible during next spinloop */
1766 kvm_vcpu_set_dy_eligible(me, false);
1768 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1770 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1772 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1775 if (vmf->pgoff == 0)
1776 page = virt_to_page(vcpu->run);
1778 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1779 page = virt_to_page(vcpu->arch.pio_data);
1781 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1782 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1783 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1786 return kvm_arch_vcpu_fault(vcpu, vmf);
1792 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1793 .fault = kvm_vcpu_fault,
1796 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1798 vma->vm_ops = &kvm_vcpu_vm_ops;
1802 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1804 struct kvm_vcpu *vcpu = filp->private_data;
1806 kvm_put_kvm(vcpu->kvm);
1810 static struct file_operations kvm_vcpu_fops = {
1811 .release = kvm_vcpu_release,
1812 .unlocked_ioctl = kvm_vcpu_ioctl,
1813 #ifdef CONFIG_COMPAT
1814 .compat_ioctl = kvm_vcpu_compat_ioctl,
1816 .mmap = kvm_vcpu_mmap,
1817 .llseek = noop_llseek,
1821 * Allocates an inode for the vcpu.
1823 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1825 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1829 * Creates some virtual cpus. Good luck creating more than one.
1831 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1834 struct kvm_vcpu *vcpu, *v;
1836 vcpu = kvm_arch_vcpu_create(kvm, id);
1838 return PTR_ERR(vcpu);
1840 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1842 r = kvm_arch_vcpu_setup(vcpu);
1846 mutex_lock(&kvm->lock);
1847 if (!kvm_vcpu_compatible(vcpu)) {
1849 goto unlock_vcpu_destroy;
1851 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1853 goto unlock_vcpu_destroy;
1856 kvm_for_each_vcpu(r, v, kvm)
1857 if (v->vcpu_id == id) {
1859 goto unlock_vcpu_destroy;
1862 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1864 /* Now it's all set up, let userspace reach it */
1866 r = create_vcpu_fd(vcpu);
1869 goto unlock_vcpu_destroy;
1872 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1874 atomic_inc(&kvm->online_vcpus);
1876 mutex_unlock(&kvm->lock);
1877 kvm_arch_vcpu_postcreate(vcpu);
1880 unlock_vcpu_destroy:
1881 mutex_unlock(&kvm->lock);
1883 kvm_arch_vcpu_destroy(vcpu);
1887 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1890 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1891 vcpu->sigset_active = 1;
1892 vcpu->sigset = *sigset;
1894 vcpu->sigset_active = 0;
1898 static long kvm_vcpu_ioctl(struct file *filp,
1899 unsigned int ioctl, unsigned long arg)
1901 struct kvm_vcpu *vcpu = filp->private_data;
1902 void __user *argp = (void __user *)arg;
1904 struct kvm_fpu *fpu = NULL;
1905 struct kvm_sregs *kvm_sregs = NULL;
1907 if (vcpu->kvm->mm != current->mm)
1910 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1912 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1913 * so vcpu_load() would break it.
1915 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1916 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1920 r = vcpu_load(vcpu);
1928 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1929 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1931 case KVM_GET_REGS: {
1932 struct kvm_regs *kvm_regs;
1935 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1938 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1942 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1949 case KVM_SET_REGS: {
1950 struct kvm_regs *kvm_regs;
1953 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1954 if (IS_ERR(kvm_regs)) {
1955 r = PTR_ERR(kvm_regs);
1958 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1962 case KVM_GET_SREGS: {
1963 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1967 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1971 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1976 case KVM_SET_SREGS: {
1977 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1978 if (IS_ERR(kvm_sregs)) {
1979 r = PTR_ERR(kvm_sregs);
1983 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1986 case KVM_GET_MP_STATE: {
1987 struct kvm_mp_state mp_state;
1989 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1993 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1998 case KVM_SET_MP_STATE: {
1999 struct kvm_mp_state mp_state;
2002 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2004 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2007 case KVM_TRANSLATE: {
2008 struct kvm_translation tr;
2011 if (copy_from_user(&tr, argp, sizeof tr))
2013 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2017 if (copy_to_user(argp, &tr, sizeof tr))
2022 case KVM_SET_GUEST_DEBUG: {
2023 struct kvm_guest_debug dbg;
2026 if (copy_from_user(&dbg, argp, sizeof dbg))
2028 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2031 case KVM_SET_SIGNAL_MASK: {
2032 struct kvm_signal_mask __user *sigmask_arg = argp;
2033 struct kvm_signal_mask kvm_sigmask;
2034 sigset_t sigset, *p;
2039 if (copy_from_user(&kvm_sigmask, argp,
2040 sizeof kvm_sigmask))
2043 if (kvm_sigmask.len != sizeof sigset)
2046 if (copy_from_user(&sigset, sigmask_arg->sigset,
2051 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2055 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2059 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2063 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2069 fpu = memdup_user(argp, sizeof(*fpu));
2075 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2079 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2088 #ifdef CONFIG_COMPAT
2089 static long kvm_vcpu_compat_ioctl(struct file *filp,
2090 unsigned int ioctl, unsigned long arg)
2092 struct kvm_vcpu *vcpu = filp->private_data;
2093 void __user *argp = compat_ptr(arg);
2096 if (vcpu->kvm->mm != current->mm)
2100 case KVM_SET_SIGNAL_MASK: {
2101 struct kvm_signal_mask __user *sigmask_arg = argp;
2102 struct kvm_signal_mask kvm_sigmask;
2103 compat_sigset_t csigset;
2108 if (copy_from_user(&kvm_sigmask, argp,
2109 sizeof kvm_sigmask))
2112 if (kvm_sigmask.len != sizeof csigset)
2115 if (copy_from_user(&csigset, sigmask_arg->sigset,
2118 sigset_from_compat(&sigset, &csigset);
2119 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2121 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2125 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2133 static long kvm_vm_ioctl(struct file *filp,
2134 unsigned int ioctl, unsigned long arg)
2136 struct kvm *kvm = filp->private_data;
2137 void __user *argp = (void __user *)arg;
2140 if (kvm->mm != current->mm)
2143 case KVM_CREATE_VCPU:
2144 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2146 case KVM_SET_USER_MEMORY_REGION: {
2147 struct kvm_userspace_memory_region kvm_userspace_mem;
2150 if (copy_from_user(&kvm_userspace_mem, argp,
2151 sizeof kvm_userspace_mem))
2154 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2157 case KVM_GET_DIRTY_LOG: {
2158 struct kvm_dirty_log log;
2161 if (copy_from_user(&log, argp, sizeof log))
2163 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2166 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2167 case KVM_REGISTER_COALESCED_MMIO: {
2168 struct kvm_coalesced_mmio_zone zone;
2170 if (copy_from_user(&zone, argp, sizeof zone))
2172 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2175 case KVM_UNREGISTER_COALESCED_MMIO: {
2176 struct kvm_coalesced_mmio_zone zone;
2178 if (copy_from_user(&zone, argp, sizeof zone))
2180 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2185 struct kvm_irqfd data;
2188 if (copy_from_user(&data, argp, sizeof data))
2190 r = kvm_irqfd(kvm, &data);
2193 case KVM_IOEVENTFD: {
2194 struct kvm_ioeventfd data;
2197 if (copy_from_user(&data, argp, sizeof data))
2199 r = kvm_ioeventfd(kvm, &data);
2202 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2203 case KVM_SET_BOOT_CPU_ID:
2205 mutex_lock(&kvm->lock);
2206 if (atomic_read(&kvm->online_vcpus) != 0)
2209 kvm->bsp_vcpu_id = arg;
2210 mutex_unlock(&kvm->lock);
2213 #ifdef CONFIG_HAVE_KVM_MSI
2214 case KVM_SIGNAL_MSI: {
2218 if (copy_from_user(&msi, argp, sizeof msi))
2220 r = kvm_send_userspace_msi(kvm, &msi);
2224 #ifdef __KVM_HAVE_IRQ_LINE
2225 case KVM_IRQ_LINE_STATUS:
2226 case KVM_IRQ_LINE: {
2227 struct kvm_irq_level irq_event;
2230 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2233 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2238 if (ioctl == KVM_IRQ_LINE_STATUS) {
2239 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2248 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2250 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2256 #ifdef CONFIG_COMPAT
2257 struct compat_kvm_dirty_log {
2261 compat_uptr_t dirty_bitmap; /* one bit per page */
2266 static long kvm_vm_compat_ioctl(struct file *filp,
2267 unsigned int ioctl, unsigned long arg)
2269 struct kvm *kvm = filp->private_data;
2272 if (kvm->mm != current->mm)
2275 case KVM_GET_DIRTY_LOG: {
2276 struct compat_kvm_dirty_log compat_log;
2277 struct kvm_dirty_log log;
2280 if (copy_from_user(&compat_log, (void __user *)arg,
2281 sizeof(compat_log)))
2283 log.slot = compat_log.slot;
2284 log.padding1 = compat_log.padding1;
2285 log.padding2 = compat_log.padding2;
2286 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2288 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2292 r = kvm_vm_ioctl(filp, ioctl, arg);
2300 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2302 struct page *page[1];
2305 gfn_t gfn = vmf->pgoff;
2306 struct kvm *kvm = vma->vm_file->private_data;
2308 addr = gfn_to_hva(kvm, gfn);
2309 if (kvm_is_error_hva(addr))
2310 return VM_FAULT_SIGBUS;
2312 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2314 if (unlikely(npages != 1))
2315 return VM_FAULT_SIGBUS;
2317 vmf->page = page[0];
2321 static const struct vm_operations_struct kvm_vm_vm_ops = {
2322 .fault = kvm_vm_fault,
2325 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2327 vma->vm_ops = &kvm_vm_vm_ops;
2331 static struct file_operations kvm_vm_fops = {
2332 .release = kvm_vm_release,
2333 .unlocked_ioctl = kvm_vm_ioctl,
2334 #ifdef CONFIG_COMPAT
2335 .compat_ioctl = kvm_vm_compat_ioctl,
2337 .mmap = kvm_vm_mmap,
2338 .llseek = noop_llseek,
2341 static int kvm_dev_ioctl_create_vm(unsigned long type)
2346 kvm = kvm_create_vm(type);
2348 return PTR_ERR(kvm);
2349 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2350 r = kvm_coalesced_mmio_init(kvm);
2356 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2363 static long kvm_dev_ioctl_check_extension_generic(long arg)
2366 case KVM_CAP_USER_MEMORY:
2367 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2368 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2369 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2370 case KVM_CAP_SET_BOOT_CPU_ID:
2372 case KVM_CAP_INTERNAL_ERROR_DATA:
2373 #ifdef CONFIG_HAVE_KVM_MSI
2374 case KVM_CAP_SIGNAL_MSI:
2377 #ifdef KVM_CAP_IRQ_ROUTING
2378 case KVM_CAP_IRQ_ROUTING:
2379 return KVM_MAX_IRQ_ROUTES;
2384 return kvm_dev_ioctl_check_extension(arg);
2387 static long kvm_dev_ioctl(struct file *filp,
2388 unsigned int ioctl, unsigned long arg)
2393 case KVM_GET_API_VERSION:
2397 r = KVM_API_VERSION;
2400 r = kvm_dev_ioctl_create_vm(arg);
2402 case KVM_CHECK_EXTENSION:
2403 r = kvm_dev_ioctl_check_extension_generic(arg);
2405 case KVM_GET_VCPU_MMAP_SIZE:
2409 r = PAGE_SIZE; /* struct kvm_run */
2411 r += PAGE_SIZE; /* pio data page */
2413 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2414 r += PAGE_SIZE; /* coalesced mmio ring page */
2417 case KVM_TRACE_ENABLE:
2418 case KVM_TRACE_PAUSE:
2419 case KVM_TRACE_DISABLE:
2423 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2429 static struct file_operations kvm_chardev_ops = {
2430 .unlocked_ioctl = kvm_dev_ioctl,
2431 .compat_ioctl = kvm_dev_ioctl,
2432 .llseek = noop_llseek,
2435 static struct miscdevice kvm_dev = {
2441 static void hardware_enable_nolock(void *junk)
2443 int cpu = raw_smp_processor_id();
2446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2449 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2451 r = kvm_arch_hardware_enable(NULL);
2454 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2455 atomic_inc(&hardware_enable_failed);
2456 printk(KERN_INFO "kvm: enabling virtualization on "
2457 "CPU%d failed\n", cpu);
2461 static void hardware_enable(void *junk)
2463 raw_spin_lock(&kvm_lock);
2464 hardware_enable_nolock(junk);
2465 raw_spin_unlock(&kvm_lock);
2468 static void hardware_disable_nolock(void *junk)
2470 int cpu = raw_smp_processor_id();
2472 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2474 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2475 kvm_arch_hardware_disable(NULL);
2478 static void hardware_disable(void *junk)
2480 raw_spin_lock(&kvm_lock);
2481 hardware_disable_nolock(junk);
2482 raw_spin_unlock(&kvm_lock);
2485 static void hardware_disable_all_nolock(void)
2487 BUG_ON(!kvm_usage_count);
2490 if (!kvm_usage_count)
2491 on_each_cpu(hardware_disable_nolock, NULL, 1);
2494 static void hardware_disable_all(void)
2496 raw_spin_lock(&kvm_lock);
2497 hardware_disable_all_nolock();
2498 raw_spin_unlock(&kvm_lock);
2501 static int hardware_enable_all(void)
2505 raw_spin_lock(&kvm_lock);
2508 if (kvm_usage_count == 1) {
2509 atomic_set(&hardware_enable_failed, 0);
2510 on_each_cpu(hardware_enable_nolock, NULL, 1);
2512 if (atomic_read(&hardware_enable_failed)) {
2513 hardware_disable_all_nolock();
2518 raw_spin_unlock(&kvm_lock);
2523 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2528 if (!kvm_usage_count)
2531 val &= ~CPU_TASKS_FROZEN;
2534 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2536 hardware_disable(NULL);
2539 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2541 hardware_enable(NULL);
2548 asmlinkage void kvm_spurious_fault(void)
2550 /* Fault while not rebooting. We want the trace. */
2553 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2555 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2559 * Some (well, at least mine) BIOSes hang on reboot if
2562 * And Intel TXT required VMX off for all cpu when system shutdown.
2564 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2565 kvm_rebooting = true;
2566 on_each_cpu(hardware_disable_nolock, NULL, 1);
2570 static struct notifier_block kvm_reboot_notifier = {
2571 .notifier_call = kvm_reboot,
2575 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2579 for (i = 0; i < bus->dev_count; i++) {
2580 struct kvm_io_device *pos = bus->range[i].dev;
2582 kvm_iodevice_destructor(pos);
2587 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2589 const struct kvm_io_range *r1 = p1;
2590 const struct kvm_io_range *r2 = p2;
2592 if (r1->addr < r2->addr)
2594 if (r1->addr + r1->len > r2->addr + r2->len)
2599 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2600 gpa_t addr, int len)
2602 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2608 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2609 kvm_io_bus_sort_cmp, NULL);
2614 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2615 gpa_t addr, int len)
2617 struct kvm_io_range *range, key;
2620 key = (struct kvm_io_range) {
2625 range = bsearch(&key, bus->range, bus->dev_count,
2626 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2630 off = range - bus->range;
2632 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2638 /* kvm_io_bus_write - called under kvm->slots_lock */
2639 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2640 int len, const void *val)
2643 struct kvm_io_bus *bus;
2644 struct kvm_io_range range;
2646 range = (struct kvm_io_range) {
2651 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2652 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2656 while (idx < bus->dev_count &&
2657 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2658 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2666 /* kvm_io_bus_read - called under kvm->slots_lock */
2667 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2671 struct kvm_io_bus *bus;
2672 struct kvm_io_range range;
2674 range = (struct kvm_io_range) {
2679 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2680 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2684 while (idx < bus->dev_count &&
2685 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2686 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2694 /* Caller must hold slots_lock. */
2695 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2696 int len, struct kvm_io_device *dev)
2698 struct kvm_io_bus *new_bus, *bus;
2700 bus = kvm->buses[bus_idx];
2701 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2704 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2705 sizeof(struct kvm_io_range)), GFP_KERNEL);
2708 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2709 sizeof(struct kvm_io_range)));
2710 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2711 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2712 synchronize_srcu_expedited(&kvm->srcu);
2718 /* Caller must hold slots_lock. */
2719 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2720 struct kvm_io_device *dev)
2723 struct kvm_io_bus *new_bus, *bus;
2725 bus = kvm->buses[bus_idx];
2727 for (i = 0; i < bus->dev_count; i++)
2728 if (bus->range[i].dev == dev) {
2736 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2737 sizeof(struct kvm_io_range)), GFP_KERNEL);
2741 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2742 new_bus->dev_count--;
2743 memcpy(new_bus->range + i, bus->range + i + 1,
2744 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2746 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2747 synchronize_srcu_expedited(&kvm->srcu);
2752 static struct notifier_block kvm_cpu_notifier = {
2753 .notifier_call = kvm_cpu_hotplug,
2756 static int vm_stat_get(void *_offset, u64 *val)
2758 unsigned offset = (long)_offset;
2762 raw_spin_lock(&kvm_lock);
2763 list_for_each_entry(kvm, &vm_list, vm_list)
2764 *val += *(u32 *)((void *)kvm + offset);
2765 raw_spin_unlock(&kvm_lock);
2769 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2771 static int vcpu_stat_get(void *_offset, u64 *val)
2773 unsigned offset = (long)_offset;
2775 struct kvm_vcpu *vcpu;
2779 raw_spin_lock(&kvm_lock);
2780 list_for_each_entry(kvm, &vm_list, vm_list)
2781 kvm_for_each_vcpu(i, vcpu, kvm)
2782 *val += *(u32 *)((void *)vcpu + offset);
2784 raw_spin_unlock(&kvm_lock);
2788 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2790 static const struct file_operations *stat_fops[] = {
2791 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2792 [KVM_STAT_VM] = &vm_stat_fops,
2795 static int kvm_init_debug(void)
2798 struct kvm_stats_debugfs_item *p;
2800 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2801 if (kvm_debugfs_dir == NULL)
2804 for (p = debugfs_entries; p->name; ++p) {
2805 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2806 (void *)(long)p->offset,
2807 stat_fops[p->kind]);
2808 if (p->dentry == NULL)
2815 debugfs_remove_recursive(kvm_debugfs_dir);
2820 static void kvm_exit_debug(void)
2822 struct kvm_stats_debugfs_item *p;
2824 for (p = debugfs_entries; p->name; ++p)
2825 debugfs_remove(p->dentry);
2826 debugfs_remove(kvm_debugfs_dir);
2829 static int kvm_suspend(void)
2831 if (kvm_usage_count)
2832 hardware_disable_nolock(NULL);
2836 static void kvm_resume(void)
2838 if (kvm_usage_count) {
2839 WARN_ON(raw_spin_is_locked(&kvm_lock));
2840 hardware_enable_nolock(NULL);
2844 static struct syscore_ops kvm_syscore_ops = {
2845 .suspend = kvm_suspend,
2846 .resume = kvm_resume,
2850 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2852 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2855 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2857 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2859 kvm_arch_vcpu_load(vcpu, cpu);
2862 static void kvm_sched_out(struct preempt_notifier *pn,
2863 struct task_struct *next)
2865 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2867 kvm_arch_vcpu_put(vcpu);
2870 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2871 struct module *module)
2876 r = kvm_arch_init(opaque);
2880 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2885 r = kvm_arch_hardware_setup();
2889 for_each_online_cpu(cpu) {
2890 smp_call_function_single(cpu,
2891 kvm_arch_check_processor_compat,
2897 r = register_cpu_notifier(&kvm_cpu_notifier);
2900 register_reboot_notifier(&kvm_reboot_notifier);
2902 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2904 vcpu_align = __alignof__(struct kvm_vcpu);
2905 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2907 if (!kvm_vcpu_cache) {
2912 r = kvm_async_pf_init();
2916 kvm_chardev_ops.owner = module;
2917 kvm_vm_fops.owner = module;
2918 kvm_vcpu_fops.owner = module;
2920 r = misc_register(&kvm_dev);
2922 printk(KERN_ERR "kvm: misc device register failed\n");
2926 register_syscore_ops(&kvm_syscore_ops);
2928 kvm_preempt_ops.sched_in = kvm_sched_in;
2929 kvm_preempt_ops.sched_out = kvm_sched_out;
2931 r = kvm_init_debug();
2933 printk(KERN_ERR "kvm: create debugfs files failed\n");
2940 unregister_syscore_ops(&kvm_syscore_ops);
2942 kvm_async_pf_deinit();
2944 kmem_cache_destroy(kvm_vcpu_cache);
2946 unregister_reboot_notifier(&kvm_reboot_notifier);
2947 unregister_cpu_notifier(&kvm_cpu_notifier);
2950 kvm_arch_hardware_unsetup();
2952 free_cpumask_var(cpus_hardware_enabled);
2958 EXPORT_SYMBOL_GPL(kvm_init);
2963 misc_deregister(&kvm_dev);
2964 kmem_cache_destroy(kvm_vcpu_cache);
2965 kvm_async_pf_deinit();
2966 unregister_syscore_ops(&kvm_syscore_ops);
2967 unregister_reboot_notifier(&kvm_reboot_notifier);
2968 unregister_cpu_notifier(&kvm_cpu_notifier);
2969 on_each_cpu(hardware_disable_nolock, NULL, 1);
2970 kvm_arch_hardware_unsetup();
2972 free_cpumask_var(cpus_hardware_enabled);
2974 EXPORT_SYMBOL_GPL(kvm_exit);