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 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
220 mutex_init(&vcpu->mutex);
225 init_waitqueue_head(&vcpu->wq);
226 kvm_async_pf_vcpu_init(vcpu);
228 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
233 vcpu->run = page_address(page);
235 kvm_vcpu_set_in_spin_loop(vcpu, false);
236 kvm_vcpu_set_dy_eligible(vcpu, false);
238 r = kvm_arch_vcpu_init(vcpu);
244 free_page((unsigned long)vcpu->run);
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
253 kvm_arch_vcpu_uninit(vcpu);
254 free_page((unsigned long)vcpu->run);
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 return container_of(mn, struct kvm, mmu_notifier);
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
269 int need_tlb_flush, idx;
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
289 idx = srcu_read_lock(&kvm->srcu);
290 spin_lock(&kvm->mmu_lock);
292 kvm->mmu_notifier_seq++;
293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294 /* we've to flush the tlb before the pages can be freed */
296 kvm_flush_remote_tlbs(kvm);
298 spin_unlock(&kvm->mmu_lock);
299 srcu_read_unlock(&kvm->srcu, idx);
302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long address,
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
310 idx = srcu_read_lock(&kvm->srcu);
311 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 kvm_set_spte_hva(kvm, address, pte);
314 spin_unlock(&kvm->mmu_lock);
315 srcu_read_unlock(&kvm->srcu, idx);
318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319 struct mm_struct *mm,
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
324 int need_tlb_flush = 0, idx;
326 idx = srcu_read_lock(&kvm->srcu);
327 spin_lock(&kvm->mmu_lock);
329 * The count increase must become visible at unlock time as no
330 * spte can be established without taking the mmu_lock and
331 * count is also read inside the mmu_lock critical section.
333 kvm->mmu_notifier_count++;
334 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335 need_tlb_flush |= kvm->tlbs_dirty;
336 /* we've to flush the tlb before the pages can be freed */
338 kvm_flush_remote_tlbs(kvm);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 spin_lock(&kvm->mmu_lock);
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
357 kvm->mmu_notifier_seq++;
360 * The above sequence increase must be visible before the
361 * below count decrease, which is ensured by the smp_wmb above
362 * in conjunction with the smp_rmb in mmu_notifier_retry().
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
367 BUG_ON(kvm->mmu_notifier_count < 0);
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long address)
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 idx = srcu_read_lock(&kvm->srcu);
378 spin_lock(&kvm->mmu_lock);
380 young = kvm_age_hva(kvm, address);
382 kvm_flush_remote_tlbs(kvm);
384 spin_unlock(&kvm->mmu_lock);
385 srcu_read_unlock(&kvm->srcu, idx);
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391 struct mm_struct *mm,
392 unsigned long address)
394 struct kvm *kvm = mmu_notifier_to_kvm(mn);
397 idx = srcu_read_lock(&kvm->srcu);
398 spin_lock(&kvm->mmu_lock);
399 young = kvm_test_age_hva(kvm, address);
400 spin_unlock(&kvm->mmu_lock);
401 srcu_read_unlock(&kvm->srcu, idx);
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407 struct mm_struct *mm)
409 struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 idx = srcu_read_lock(&kvm->srcu);
413 kvm_arch_flush_shadow_all(kvm);
414 srcu_read_unlock(&kvm->srcu, idx);
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418 .invalidate_page = kvm_mmu_notifier_invalidate_page,
419 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
421 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
422 .test_young = kvm_mmu_notifier_test_young,
423 .change_pte = kvm_mmu_notifier_change_pte,
424 .release = kvm_mmu_notifier_release,
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
433 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442 static void kvm_init_memslots_id(struct kvm *kvm)
445 struct kvm_memslots *slots = kvm->memslots;
447 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
448 slots->id_to_index[i] = slots->memslots[i].id = i;
451 static struct kvm *kvm_create_vm(unsigned long type)
454 struct kvm *kvm = kvm_arch_alloc_vm();
457 return ERR_PTR(-ENOMEM);
459 r = kvm_arch_init_vm(kvm, type);
461 goto out_err_nodisable;
463 r = hardware_enable_all();
465 goto out_err_nodisable;
467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
468 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
469 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
473 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
476 kvm_init_memslots_id(kvm);
477 if (init_srcu_struct(&kvm->srcu))
479 for (i = 0; i < KVM_NR_BUSES; i++) {
480 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
486 spin_lock_init(&kvm->mmu_lock);
487 kvm->mm = current->mm;
488 atomic_inc(&kvm->mm->mm_count);
489 kvm_eventfd_init(kvm);
490 mutex_init(&kvm->lock);
491 mutex_init(&kvm->irq_lock);
492 mutex_init(&kvm->slots_lock);
493 atomic_set(&kvm->users_count, 1);
495 r = kvm_init_mmu_notifier(kvm);
499 raw_spin_lock(&kvm_lock);
500 list_add(&kvm->vm_list, &vm_list);
501 raw_spin_unlock(&kvm_lock);
506 cleanup_srcu_struct(&kvm->srcu);
508 hardware_disable_all();
510 for (i = 0; i < KVM_NR_BUSES; i++)
511 kfree(kvm->buses[i]);
512 kfree(kvm->memslots);
513 kvm_arch_free_vm(kvm);
518 * Avoid using vmalloc for a small buffer.
519 * Should not be used when the size is statically known.
521 void *kvm_kvzalloc(unsigned long size)
523 if (size > PAGE_SIZE)
524 return vzalloc(size);
526 return kzalloc(size, GFP_KERNEL);
529 void kvm_kvfree(const void *addr)
531 if (is_vmalloc_addr(addr))
537 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
539 if (!memslot->dirty_bitmap)
542 kvm_kvfree(memslot->dirty_bitmap);
543 memslot->dirty_bitmap = NULL;
547 * Free any memory in @free but not in @dont.
549 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
550 struct kvm_memory_slot *dont)
552 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
553 kvm_destroy_dirty_bitmap(free);
555 kvm_arch_free_memslot(free, dont);
560 void kvm_free_physmem(struct kvm *kvm)
562 struct kvm_memslots *slots = kvm->memslots;
563 struct kvm_memory_slot *memslot;
565 kvm_for_each_memslot(memslot, slots)
566 kvm_free_physmem_slot(memslot, NULL);
568 kfree(kvm->memslots);
571 static void kvm_destroy_vm(struct kvm *kvm)
574 struct mm_struct *mm = kvm->mm;
576 kvm_arch_sync_events(kvm);
577 raw_spin_lock(&kvm_lock);
578 list_del(&kvm->vm_list);
579 raw_spin_unlock(&kvm_lock);
580 kvm_free_irq_routing(kvm);
581 for (i = 0; i < KVM_NR_BUSES; i++)
582 kvm_io_bus_destroy(kvm->buses[i]);
583 kvm_coalesced_mmio_free(kvm);
584 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
585 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
587 kvm_arch_flush_shadow_all(kvm);
589 kvm_arch_destroy_vm(kvm);
590 kvm_free_physmem(kvm);
591 cleanup_srcu_struct(&kvm->srcu);
592 kvm_arch_free_vm(kvm);
593 hardware_disable_all();
597 void kvm_get_kvm(struct kvm *kvm)
599 atomic_inc(&kvm->users_count);
601 EXPORT_SYMBOL_GPL(kvm_get_kvm);
603 void kvm_put_kvm(struct kvm *kvm)
605 if (atomic_dec_and_test(&kvm->users_count))
608 EXPORT_SYMBOL_GPL(kvm_put_kvm);
611 static int kvm_vm_release(struct inode *inode, struct file *filp)
613 struct kvm *kvm = filp->private_data;
615 kvm_irqfd_release(kvm);
622 * Allocation size is twice as large as the actual dirty bitmap size.
623 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
628 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
630 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
631 if (!memslot->dirty_bitmap)
634 #endif /* !CONFIG_S390 */
638 static int cmp_memslot(const void *slot1, const void *slot2)
640 struct kvm_memory_slot *s1, *s2;
642 s1 = (struct kvm_memory_slot *)slot1;
643 s2 = (struct kvm_memory_slot *)slot2;
645 if (s1->npages < s2->npages)
647 if (s1->npages > s2->npages)
654 * Sort the memslots base on its size, so the larger slots
655 * will get better fit.
657 static void sort_memslots(struct kvm_memslots *slots)
661 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
662 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
664 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
665 slots->id_to_index[slots->memslots[i].id] = i;
668 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
672 struct kvm_memory_slot *old = id_to_memslot(slots, id);
673 unsigned long npages = old->npages;
676 if (new->npages != npages)
677 sort_memslots(slots);
683 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
685 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
687 #ifdef KVM_CAP_READONLY_MEM
688 valid_flags |= KVM_MEM_READONLY;
691 if (mem->flags & ~valid_flags)
698 * Allocate some memory and give it an address in the guest physical address
701 * Discontiguous memory is allowed, mostly for framebuffers.
703 * Must be called holding mmap_sem for write.
705 int __kvm_set_memory_region(struct kvm *kvm,
706 struct kvm_userspace_memory_region *mem,
711 unsigned long npages;
713 struct kvm_memory_slot *memslot;
714 struct kvm_memory_slot old, new;
715 struct kvm_memslots *slots, *old_memslots;
717 r = check_memory_region_flags(mem);
722 /* General sanity checks */
723 if (mem->memory_size & (PAGE_SIZE - 1))
725 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
727 /* We can read the guest memory with __xxx_user() later on. */
729 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
730 !access_ok(VERIFY_WRITE,
731 (void __user *)(unsigned long)mem->userspace_addr,
734 if (mem->slot >= KVM_MEM_SLOTS_NUM)
736 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
739 memslot = id_to_memslot(kvm->memslots, mem->slot);
740 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
741 npages = mem->memory_size >> PAGE_SHIFT;
744 if (npages > KVM_MEM_MAX_NR_PAGES)
748 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
750 new = old = *memslot;
753 new.base_gfn = base_gfn;
755 new.flags = mem->flags;
757 /* Disallow changing a memory slot's size. */
759 if (npages && old.npages && npages != old.npages)
762 /* Check for overlaps */
764 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
765 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
767 if (s == memslot || !s->npages)
769 if (!((base_gfn + npages <= s->base_gfn) ||
770 (base_gfn >= s->base_gfn + s->npages)))
774 /* Free page dirty bitmap if unneeded */
775 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
776 new.dirty_bitmap = NULL;
780 /* Allocate if a slot is being created */
781 if (npages && !old.npages) {
782 new.user_alloc = user_alloc;
783 new.userspace_addr = mem->userspace_addr;
785 if (kvm_arch_create_memslot(&new, npages))
789 /* Allocate page dirty bitmap if needed */
790 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
791 if (kvm_create_dirty_bitmap(&new) < 0)
793 /* destroy any largepage mappings for dirty tracking */
796 if (!npages || base_gfn != old.base_gfn) {
797 struct kvm_memory_slot *slot;
800 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
804 slot = id_to_memslot(slots, mem->slot);
805 slot->flags |= KVM_MEMSLOT_INVALID;
807 update_memslots(slots, NULL);
809 old_memslots = kvm->memslots;
810 rcu_assign_pointer(kvm->memslots, slots);
811 synchronize_srcu_expedited(&kvm->srcu);
812 /* From this point no new shadow pages pointing to a deleted,
813 * or moved, memslot will be created.
815 * validation of sp->gfn happens in:
816 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
817 * - kvm_is_visible_gfn (mmu_check_roots)
819 kvm_arch_flush_shadow_memslot(kvm, slot);
823 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
827 /* map/unmap the pages in iommu page table */
829 r = kvm_iommu_map_pages(kvm, &new);
833 kvm_iommu_unmap_pages(kvm, &old);
836 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
841 /* actual memory is freed via old in kvm_free_physmem_slot below */
843 new.dirty_bitmap = NULL;
844 memset(&new.arch, 0, sizeof(new.arch));
847 update_memslots(slots, &new);
848 old_memslots = kvm->memslots;
849 rcu_assign_pointer(kvm->memslots, slots);
850 synchronize_srcu_expedited(&kvm->srcu);
852 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
854 kvm_free_physmem_slot(&old, &new);
860 kvm_free_physmem_slot(&new, &old);
865 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
867 int kvm_set_memory_region(struct kvm *kvm,
868 struct kvm_userspace_memory_region *mem,
873 mutex_lock(&kvm->slots_lock);
874 r = __kvm_set_memory_region(kvm, mem, user_alloc);
875 mutex_unlock(&kvm->slots_lock);
878 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
880 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
882 kvm_userspace_memory_region *mem,
885 if (mem->slot >= KVM_MEMORY_SLOTS)
887 return kvm_set_memory_region(kvm, mem, user_alloc);
890 int kvm_get_dirty_log(struct kvm *kvm,
891 struct kvm_dirty_log *log, int *is_dirty)
893 struct kvm_memory_slot *memslot;
896 unsigned long any = 0;
899 if (log->slot >= KVM_MEMORY_SLOTS)
902 memslot = id_to_memslot(kvm->memslots, log->slot);
904 if (!memslot->dirty_bitmap)
907 n = kvm_dirty_bitmap_bytes(memslot);
909 for (i = 0; !any && i < n/sizeof(long); ++i)
910 any = memslot->dirty_bitmap[i];
913 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
924 bool kvm_largepages_enabled(void)
926 return largepages_enabled;
929 void kvm_disable_largepages(void)
931 largepages_enabled = false;
933 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
935 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
937 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
939 EXPORT_SYMBOL_GPL(gfn_to_memslot);
941 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
943 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
945 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
946 memslot->flags & KVM_MEMSLOT_INVALID)
951 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
953 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
955 struct vm_area_struct *vma;
956 unsigned long addr, size;
960 addr = gfn_to_hva(kvm, gfn);
961 if (kvm_is_error_hva(addr))
964 down_read(¤t->mm->mmap_sem);
965 vma = find_vma(current->mm, addr);
969 size = vma_kernel_pagesize(vma);
972 up_read(¤t->mm->mmap_sem);
977 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
979 return slot->flags & KVM_MEM_READONLY;
982 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
983 gfn_t *nr_pages, bool write)
985 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
986 return KVM_HVA_ERR_BAD;
988 if (memslot_is_readonly(slot) && write)
989 return KVM_HVA_ERR_RO_BAD;
992 *nr_pages = slot->npages - (gfn - slot->base_gfn);
994 return __gfn_to_hva_memslot(slot, gfn);
997 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1000 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1003 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1006 return gfn_to_hva_many(slot, gfn, NULL);
1008 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1010 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1012 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1014 EXPORT_SYMBOL_GPL(gfn_to_hva);
1017 * The hva returned by this function is only allowed to be read.
1018 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1020 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1022 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1025 static int kvm_read_hva(void *data, void __user *hva, int len)
1027 return __copy_from_user(data, hva, len);
1030 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1032 return __copy_from_user_inatomic(data, hva, len);
1035 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1036 unsigned long start, int write, struct page **page)
1038 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1041 flags |= FOLL_WRITE;
1043 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1046 static inline int check_user_page_hwpoison(unsigned long addr)
1048 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1050 rc = __get_user_pages(current, current->mm, addr, 1,
1051 flags, NULL, NULL, NULL);
1052 return rc == -EHWPOISON;
1056 * The atomic path to get the writable pfn which will be stored in @pfn,
1057 * true indicates success, otherwise false is returned.
1059 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1060 bool write_fault, bool *writable, pfn_t *pfn)
1062 struct page *page[1];
1065 if (!(async || atomic))
1069 * Fast pin a writable pfn only if it is a write fault request
1070 * or the caller allows to map a writable pfn for a read fault
1073 if (!(write_fault || writable))
1076 npages = __get_user_pages_fast(addr, 1, 1, page);
1078 *pfn = page_to_pfn(page[0]);
1089 * The slow path to get the pfn of the specified host virtual address,
1090 * 1 indicates success, -errno is returned if error is detected.
1092 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1093 bool *writable, pfn_t *pfn)
1095 struct page *page[1];
1101 *writable = write_fault;
1104 down_read(¤t->mm->mmap_sem);
1105 npages = get_user_page_nowait(current, current->mm,
1106 addr, write_fault, page);
1107 up_read(¤t->mm->mmap_sem);
1109 npages = get_user_pages_fast(addr, 1, write_fault,
1114 /* map read fault as writable if possible */
1115 if (unlikely(!write_fault) && writable) {
1116 struct page *wpage[1];
1118 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1127 *pfn = page_to_pfn(page[0]);
1131 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1133 if (unlikely(!(vma->vm_flags & VM_READ)))
1136 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1143 * Pin guest page in memory and return its pfn.
1144 * @addr: host virtual address which maps memory to the guest
1145 * @atomic: whether this function can sleep
1146 * @async: whether this function need to wait IO complete if the
1147 * host page is not in the memory
1148 * @write_fault: whether we should get a writable host page
1149 * @writable: whether it allows to map a writable host page for !@write_fault
1151 * The function will map a writable host page for these two cases:
1152 * 1): @write_fault = true
1153 * 2): @write_fault = false && @writable, @writable will tell the caller
1154 * whether the mapping is writable.
1156 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1157 bool write_fault, bool *writable)
1159 struct vm_area_struct *vma;
1163 /* we can do it either atomically or asynchronously, not both */
1164 BUG_ON(atomic && async);
1166 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1170 return KVM_PFN_ERR_FAULT;
1172 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1176 down_read(¤t->mm->mmap_sem);
1177 if (npages == -EHWPOISON ||
1178 (!async && check_user_page_hwpoison(addr))) {
1179 pfn = KVM_PFN_ERR_HWPOISON;
1183 vma = find_vma_intersection(current->mm, addr, addr + 1);
1186 pfn = KVM_PFN_ERR_FAULT;
1187 else if ((vma->vm_flags & VM_PFNMAP)) {
1188 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1190 BUG_ON(!kvm_is_mmio_pfn(pfn));
1192 if (async && vma_is_valid(vma, write_fault))
1194 pfn = KVM_PFN_ERR_FAULT;
1197 up_read(¤t->mm->mmap_sem);
1202 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1203 bool *async, bool write_fault, bool *writable)
1205 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1207 if (addr == KVM_HVA_ERR_RO_BAD)
1208 return KVM_PFN_ERR_RO_FAULT;
1210 if (kvm_is_error_hva(addr))
1211 return KVM_PFN_ERR_BAD;
1213 /* Do not map writable pfn in the readonly memslot. */
1214 if (writable && memslot_is_readonly(slot)) {
1219 return hva_to_pfn(addr, atomic, async, write_fault,
1223 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1224 bool write_fault, bool *writable)
1226 struct kvm_memory_slot *slot;
1231 slot = gfn_to_memslot(kvm, gfn);
1233 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1237 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1239 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1241 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1243 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1244 bool write_fault, bool *writable)
1246 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1248 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1250 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1252 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1254 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1256 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1259 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1261 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1263 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1265 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1268 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1270 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1272 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1274 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1280 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1281 if (kvm_is_error_hva(addr))
1284 if (entry < nr_pages)
1287 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1289 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1291 static struct page *kvm_pfn_to_page(pfn_t pfn)
1293 if (is_error_pfn(pfn))
1294 return KVM_ERR_PTR_BAD_PAGE;
1296 if (kvm_is_mmio_pfn(pfn)) {
1298 return KVM_ERR_PTR_BAD_PAGE;
1301 return pfn_to_page(pfn);
1304 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1308 pfn = gfn_to_pfn(kvm, gfn);
1310 return kvm_pfn_to_page(pfn);
1313 EXPORT_SYMBOL_GPL(gfn_to_page);
1315 void kvm_release_page_clean(struct page *page)
1317 WARN_ON(is_error_page(page));
1319 kvm_release_pfn_clean(page_to_pfn(page));
1321 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1323 void kvm_release_pfn_clean(pfn_t pfn)
1325 WARN_ON(is_error_pfn(pfn));
1327 if (!kvm_is_mmio_pfn(pfn))
1328 put_page(pfn_to_page(pfn));
1330 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1332 void kvm_release_page_dirty(struct page *page)
1334 WARN_ON(is_error_page(page));
1336 kvm_release_pfn_dirty(page_to_pfn(page));
1338 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1340 void kvm_release_pfn_dirty(pfn_t pfn)
1342 kvm_set_pfn_dirty(pfn);
1343 kvm_release_pfn_clean(pfn);
1345 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1347 void kvm_set_page_dirty(struct page *page)
1349 kvm_set_pfn_dirty(page_to_pfn(page));
1351 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1353 void kvm_set_pfn_dirty(pfn_t pfn)
1355 if (!kvm_is_mmio_pfn(pfn)) {
1356 struct page *page = pfn_to_page(pfn);
1357 if (!PageReserved(page))
1361 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1363 void kvm_set_pfn_accessed(pfn_t pfn)
1365 if (!kvm_is_mmio_pfn(pfn))
1366 mark_page_accessed(pfn_to_page(pfn));
1368 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1370 void kvm_get_pfn(pfn_t pfn)
1372 if (!kvm_is_mmio_pfn(pfn))
1373 get_page(pfn_to_page(pfn));
1375 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1377 static int next_segment(unsigned long len, int offset)
1379 if (len > PAGE_SIZE - offset)
1380 return PAGE_SIZE - offset;
1385 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1391 addr = gfn_to_hva_read(kvm, gfn);
1392 if (kvm_is_error_hva(addr))
1394 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1399 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1401 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1403 gfn_t gfn = gpa >> PAGE_SHIFT;
1405 int offset = offset_in_page(gpa);
1408 while ((seg = next_segment(len, offset)) != 0) {
1409 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1419 EXPORT_SYMBOL_GPL(kvm_read_guest);
1421 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1426 gfn_t gfn = gpa >> PAGE_SHIFT;
1427 int offset = offset_in_page(gpa);
1429 addr = gfn_to_hva_read(kvm, gfn);
1430 if (kvm_is_error_hva(addr))
1432 pagefault_disable();
1433 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1439 EXPORT_SYMBOL(kvm_read_guest_atomic);
1441 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1442 int offset, int len)
1447 addr = gfn_to_hva(kvm, gfn);
1448 if (kvm_is_error_hva(addr))
1450 r = __copy_to_user((void __user *)addr + offset, data, len);
1453 mark_page_dirty(kvm, gfn);
1456 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1458 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1461 gfn_t gfn = gpa >> PAGE_SHIFT;
1463 int offset = offset_in_page(gpa);
1466 while ((seg = next_segment(len, offset)) != 0) {
1467 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1478 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1481 struct kvm_memslots *slots = kvm_memslots(kvm);
1482 int offset = offset_in_page(gpa);
1483 gfn_t gfn = gpa >> PAGE_SHIFT;
1486 ghc->generation = slots->generation;
1487 ghc->memslot = gfn_to_memslot(kvm, gfn);
1488 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1489 if (!kvm_is_error_hva(ghc->hva))
1496 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1498 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1499 void *data, unsigned long len)
1501 struct kvm_memslots *slots = kvm_memslots(kvm);
1504 if (slots->generation != ghc->generation)
1505 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1507 if (kvm_is_error_hva(ghc->hva))
1510 r = __copy_to_user((void __user *)ghc->hva, data, len);
1513 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1517 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1519 int kvm_read_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_from_user(data, (void __user *)ghc->hva, len);
1537 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1539 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1541 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1544 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1546 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1548 gfn_t gfn = gpa >> PAGE_SHIFT;
1550 int offset = offset_in_page(gpa);
1553 while ((seg = next_segment(len, offset)) != 0) {
1554 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1563 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1565 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1568 if (memslot && memslot->dirty_bitmap) {
1569 unsigned long rel_gfn = gfn - memslot->base_gfn;
1571 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1575 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1577 struct kvm_memory_slot *memslot;
1579 memslot = gfn_to_memslot(kvm, gfn);
1580 mark_page_dirty_in_slot(kvm, memslot, gfn);
1584 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1586 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1591 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1593 if (kvm_arch_vcpu_runnable(vcpu)) {
1594 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1597 if (kvm_cpu_has_pending_timer(vcpu))
1599 if (signal_pending(current))
1605 finish_wait(&vcpu->wq, &wait);
1610 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1612 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1615 int cpu = vcpu->cpu;
1616 wait_queue_head_t *wqp;
1618 wqp = kvm_arch_vcpu_wq(vcpu);
1619 if (waitqueue_active(wqp)) {
1620 wake_up_interruptible(wqp);
1621 ++vcpu->stat.halt_wakeup;
1625 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1626 if (kvm_arch_vcpu_should_kick(vcpu))
1627 smp_send_reschedule(cpu);
1630 #endif /* !CONFIG_S390 */
1632 void kvm_resched(struct kvm_vcpu *vcpu)
1634 if (!need_resched())
1638 EXPORT_SYMBOL_GPL(kvm_resched);
1640 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1643 struct task_struct *task = NULL;
1646 pid = rcu_dereference(target->pid);
1648 task = get_pid_task(target->pid, PIDTYPE_PID);
1652 if (task->flags & PF_VCPU) {
1653 put_task_struct(task);
1656 if (yield_to(task, 1)) {
1657 put_task_struct(task);
1660 put_task_struct(task);
1663 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1665 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1667 * Helper that checks whether a VCPU is eligible for directed yield.
1668 * Most eligible candidate to yield is decided by following heuristics:
1670 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1671 * (preempted lock holder), indicated by @in_spin_loop.
1672 * Set at the beiginning and cleared at the end of interception/PLE handler.
1674 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1675 * chance last time (mostly it has become eligible now since we have probably
1676 * yielded to lockholder in last iteration. This is done by toggling
1677 * @dy_eligible each time a VCPU checked for eligibility.)
1679 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1680 * to preempted lock-holder could result in wrong VCPU selection and CPU
1681 * burning. Giving priority for a potential lock-holder increases lock
1684 * Since algorithm is based on heuristics, accessing another VCPU data without
1685 * locking does not harm. It may result in trying to yield to same VCPU, fail
1686 * and continue with next VCPU and so on.
1688 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1692 eligible = !vcpu->spin_loop.in_spin_loop ||
1693 (vcpu->spin_loop.in_spin_loop &&
1694 vcpu->spin_loop.dy_eligible);
1696 if (vcpu->spin_loop.in_spin_loop)
1697 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1702 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1704 struct kvm *kvm = me->kvm;
1705 struct kvm_vcpu *vcpu;
1706 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1711 kvm_vcpu_set_in_spin_loop(me, true);
1713 * We boost the priority of a VCPU that is runnable but not
1714 * currently running, because it got preempted by something
1715 * else and called schedule in __vcpu_run. Hopefully that
1716 * VCPU is holding the lock that we need and will release it.
1717 * We approximate round-robin by starting at the last boosted VCPU.
1719 for (pass = 0; pass < 2 && !yielded; pass++) {
1720 kvm_for_each_vcpu(i, vcpu, kvm) {
1721 if (!pass && i <= last_boosted_vcpu) {
1722 i = last_boosted_vcpu;
1724 } else if (pass && i > last_boosted_vcpu)
1728 if (waitqueue_active(&vcpu->wq))
1730 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1732 if (kvm_vcpu_yield_to(vcpu)) {
1733 kvm->last_boosted_vcpu = i;
1739 kvm_vcpu_set_in_spin_loop(me, false);
1741 /* Ensure vcpu is not eligible during next spinloop */
1742 kvm_vcpu_set_dy_eligible(me, false);
1744 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1746 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1748 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1751 if (vmf->pgoff == 0)
1752 page = virt_to_page(vcpu->run);
1754 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1755 page = virt_to_page(vcpu->arch.pio_data);
1757 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1758 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1759 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1762 return kvm_arch_vcpu_fault(vcpu, vmf);
1768 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1769 .fault = kvm_vcpu_fault,
1772 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1774 vma->vm_ops = &kvm_vcpu_vm_ops;
1778 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1780 struct kvm_vcpu *vcpu = filp->private_data;
1782 kvm_put_kvm(vcpu->kvm);
1786 static struct file_operations kvm_vcpu_fops = {
1787 .release = kvm_vcpu_release,
1788 .unlocked_ioctl = kvm_vcpu_ioctl,
1789 #ifdef CONFIG_COMPAT
1790 .compat_ioctl = kvm_vcpu_compat_ioctl,
1792 .mmap = kvm_vcpu_mmap,
1793 .llseek = noop_llseek,
1797 * Allocates an inode for the vcpu.
1799 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1801 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1805 * Creates some virtual cpus. Good luck creating more than one.
1807 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1810 struct kvm_vcpu *vcpu, *v;
1812 vcpu = kvm_arch_vcpu_create(kvm, id);
1814 return PTR_ERR(vcpu);
1816 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1818 r = kvm_arch_vcpu_setup(vcpu);
1822 mutex_lock(&kvm->lock);
1823 if (!kvm_vcpu_compatible(vcpu)) {
1825 goto unlock_vcpu_destroy;
1827 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1829 goto unlock_vcpu_destroy;
1832 kvm_for_each_vcpu(r, v, kvm)
1833 if (v->vcpu_id == id) {
1835 goto unlock_vcpu_destroy;
1838 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1840 /* Now it's all set up, let userspace reach it */
1842 r = create_vcpu_fd(vcpu);
1845 goto unlock_vcpu_destroy;
1848 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1850 atomic_inc(&kvm->online_vcpus);
1852 mutex_unlock(&kvm->lock);
1855 unlock_vcpu_destroy:
1856 mutex_unlock(&kvm->lock);
1858 kvm_arch_vcpu_destroy(vcpu);
1862 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1865 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1866 vcpu->sigset_active = 1;
1867 vcpu->sigset = *sigset;
1869 vcpu->sigset_active = 0;
1873 static long kvm_vcpu_ioctl(struct file *filp,
1874 unsigned int ioctl, unsigned long arg)
1876 struct kvm_vcpu *vcpu = filp->private_data;
1877 void __user *argp = (void __user *)arg;
1879 struct kvm_fpu *fpu = NULL;
1880 struct kvm_sregs *kvm_sregs = NULL;
1882 if (vcpu->kvm->mm != current->mm)
1885 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1887 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1888 * so vcpu_load() would break it.
1890 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1891 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1895 r = vcpu_load(vcpu);
1903 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1904 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1906 case KVM_GET_REGS: {
1907 struct kvm_regs *kvm_regs;
1910 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1913 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1917 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1924 case KVM_SET_REGS: {
1925 struct kvm_regs *kvm_regs;
1928 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1929 if (IS_ERR(kvm_regs)) {
1930 r = PTR_ERR(kvm_regs);
1933 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1941 case KVM_GET_SREGS: {
1942 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1946 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1950 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1955 case KVM_SET_SREGS: {
1956 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1957 if (IS_ERR(kvm_sregs)) {
1958 r = PTR_ERR(kvm_sregs);
1961 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1967 case KVM_GET_MP_STATE: {
1968 struct kvm_mp_state mp_state;
1970 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1974 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1979 case KVM_SET_MP_STATE: {
1980 struct kvm_mp_state mp_state;
1983 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1985 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1991 case KVM_TRANSLATE: {
1992 struct kvm_translation tr;
1995 if (copy_from_user(&tr, argp, sizeof tr))
1997 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2001 if (copy_to_user(argp, &tr, sizeof tr))
2006 case KVM_SET_GUEST_DEBUG: {
2007 struct kvm_guest_debug dbg;
2010 if (copy_from_user(&dbg, argp, sizeof dbg))
2012 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2018 case KVM_SET_SIGNAL_MASK: {
2019 struct kvm_signal_mask __user *sigmask_arg = argp;
2020 struct kvm_signal_mask kvm_sigmask;
2021 sigset_t sigset, *p;
2026 if (copy_from_user(&kvm_sigmask, argp,
2027 sizeof kvm_sigmask))
2030 if (kvm_sigmask.len != sizeof sigset)
2033 if (copy_from_user(&sigset, sigmask_arg->sigset,
2038 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2042 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2046 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2050 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2056 fpu = memdup_user(argp, sizeof(*fpu));
2061 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2068 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2077 #ifdef CONFIG_COMPAT
2078 static long kvm_vcpu_compat_ioctl(struct file *filp,
2079 unsigned int ioctl, unsigned long arg)
2081 struct kvm_vcpu *vcpu = filp->private_data;
2082 void __user *argp = compat_ptr(arg);
2085 if (vcpu->kvm->mm != current->mm)
2089 case KVM_SET_SIGNAL_MASK: {
2090 struct kvm_signal_mask __user *sigmask_arg = argp;
2091 struct kvm_signal_mask kvm_sigmask;
2092 compat_sigset_t csigset;
2097 if (copy_from_user(&kvm_sigmask, argp,
2098 sizeof kvm_sigmask))
2101 if (kvm_sigmask.len != sizeof csigset)
2104 if (copy_from_user(&csigset, sigmask_arg->sigset,
2107 sigset_from_compat(&sigset, &csigset);
2108 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2110 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2114 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2122 static long kvm_vm_ioctl(struct file *filp,
2123 unsigned int ioctl, unsigned long arg)
2125 struct kvm *kvm = filp->private_data;
2126 void __user *argp = (void __user *)arg;
2129 if (kvm->mm != current->mm)
2132 case KVM_CREATE_VCPU:
2133 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2137 case KVM_SET_USER_MEMORY_REGION: {
2138 struct kvm_userspace_memory_region kvm_userspace_mem;
2141 if (copy_from_user(&kvm_userspace_mem, argp,
2142 sizeof kvm_userspace_mem))
2145 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2150 case KVM_GET_DIRTY_LOG: {
2151 struct kvm_dirty_log log;
2154 if (copy_from_user(&log, argp, sizeof log))
2156 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2161 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2162 case KVM_REGISTER_COALESCED_MMIO: {
2163 struct kvm_coalesced_mmio_zone zone;
2165 if (copy_from_user(&zone, argp, sizeof zone))
2167 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2173 case KVM_UNREGISTER_COALESCED_MMIO: {
2174 struct kvm_coalesced_mmio_zone zone;
2176 if (copy_from_user(&zone, argp, sizeof zone))
2178 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2186 struct kvm_irqfd data;
2189 if (copy_from_user(&data, argp, sizeof data))
2191 r = kvm_irqfd(kvm, &data);
2194 case KVM_IOEVENTFD: {
2195 struct kvm_ioeventfd data;
2198 if (copy_from_user(&data, argp, sizeof data))
2200 r = kvm_ioeventfd(kvm, &data);
2203 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2204 case KVM_SET_BOOT_CPU_ID:
2206 mutex_lock(&kvm->lock);
2207 if (atomic_read(&kvm->online_vcpus) != 0)
2210 kvm->bsp_vcpu_id = arg;
2211 mutex_unlock(&kvm->lock);
2214 #ifdef CONFIG_HAVE_KVM_MSI
2215 case KVM_SIGNAL_MSI: {
2219 if (copy_from_user(&msi, argp, sizeof msi))
2221 r = kvm_send_userspace_msi(kvm, &msi);
2225 #ifdef __KVM_HAVE_IRQ_LINE
2226 case KVM_IRQ_LINE_STATUS:
2227 case KVM_IRQ_LINE: {
2228 struct kvm_irq_level irq_event;
2231 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2234 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2239 if (ioctl == KVM_IRQ_LINE_STATUS) {
2240 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2249 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2251 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2257 #ifdef CONFIG_COMPAT
2258 struct compat_kvm_dirty_log {
2262 compat_uptr_t dirty_bitmap; /* one bit per page */
2267 static long kvm_vm_compat_ioctl(struct file *filp,
2268 unsigned int ioctl, unsigned long arg)
2270 struct kvm *kvm = filp->private_data;
2273 if (kvm->mm != current->mm)
2276 case KVM_GET_DIRTY_LOG: {
2277 struct compat_kvm_dirty_log compat_log;
2278 struct kvm_dirty_log log;
2281 if (copy_from_user(&compat_log, (void __user *)arg,
2282 sizeof(compat_log)))
2284 log.slot = compat_log.slot;
2285 log.padding1 = compat_log.padding1;
2286 log.padding2 = compat_log.padding2;
2287 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2289 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2295 r = kvm_vm_ioctl(filp, ioctl, arg);
2303 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2305 struct page *page[1];
2308 gfn_t gfn = vmf->pgoff;
2309 struct kvm *kvm = vma->vm_file->private_data;
2311 addr = gfn_to_hva(kvm, gfn);
2312 if (kvm_is_error_hva(addr))
2313 return VM_FAULT_SIGBUS;
2315 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2317 if (unlikely(npages != 1))
2318 return VM_FAULT_SIGBUS;
2320 vmf->page = page[0];
2324 static const struct vm_operations_struct kvm_vm_vm_ops = {
2325 .fault = kvm_vm_fault,
2328 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2330 vma->vm_ops = &kvm_vm_vm_ops;
2334 static struct file_operations kvm_vm_fops = {
2335 .release = kvm_vm_release,
2336 .unlocked_ioctl = kvm_vm_ioctl,
2337 #ifdef CONFIG_COMPAT
2338 .compat_ioctl = kvm_vm_compat_ioctl,
2340 .mmap = kvm_vm_mmap,
2341 .llseek = noop_llseek,
2344 static int kvm_dev_ioctl_create_vm(unsigned long type)
2349 kvm = kvm_create_vm(type);
2351 return PTR_ERR(kvm);
2352 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2353 r = kvm_coalesced_mmio_init(kvm);
2359 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2366 static long kvm_dev_ioctl_check_extension_generic(long arg)
2369 case KVM_CAP_USER_MEMORY:
2370 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2371 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2372 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID:
2375 case KVM_CAP_INTERNAL_ERROR_DATA:
2376 #ifdef CONFIG_HAVE_KVM_MSI
2377 case KVM_CAP_SIGNAL_MSI:
2380 #ifdef KVM_CAP_IRQ_ROUTING
2381 case KVM_CAP_IRQ_ROUTING:
2382 return KVM_MAX_IRQ_ROUTES;
2387 return kvm_dev_ioctl_check_extension(arg);
2390 static long kvm_dev_ioctl(struct file *filp,
2391 unsigned int ioctl, unsigned long arg)
2396 case KVM_GET_API_VERSION:
2400 r = KVM_API_VERSION;
2403 r = kvm_dev_ioctl_create_vm(arg);
2405 case KVM_CHECK_EXTENSION:
2406 r = kvm_dev_ioctl_check_extension_generic(arg);
2408 case KVM_GET_VCPU_MMAP_SIZE:
2412 r = PAGE_SIZE; /* struct kvm_run */
2414 r += PAGE_SIZE; /* pio data page */
2416 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2417 r += PAGE_SIZE; /* coalesced mmio ring page */
2420 case KVM_TRACE_ENABLE:
2421 case KVM_TRACE_PAUSE:
2422 case KVM_TRACE_DISABLE:
2426 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2432 static struct file_operations kvm_chardev_ops = {
2433 .unlocked_ioctl = kvm_dev_ioctl,
2434 .compat_ioctl = kvm_dev_ioctl,
2435 .llseek = noop_llseek,
2438 static struct miscdevice kvm_dev = {
2444 static void hardware_enable_nolock(void *junk)
2446 int cpu = raw_smp_processor_id();
2449 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2452 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2454 r = kvm_arch_hardware_enable(NULL);
2457 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2458 atomic_inc(&hardware_enable_failed);
2459 printk(KERN_INFO "kvm: enabling virtualization on "
2460 "CPU%d failed\n", cpu);
2464 static void hardware_enable(void *junk)
2466 raw_spin_lock(&kvm_lock);
2467 hardware_enable_nolock(junk);
2468 raw_spin_unlock(&kvm_lock);
2471 static void hardware_disable_nolock(void *junk)
2473 int cpu = raw_smp_processor_id();
2475 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2477 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2478 kvm_arch_hardware_disable(NULL);
2481 static void hardware_disable(void *junk)
2483 raw_spin_lock(&kvm_lock);
2484 hardware_disable_nolock(junk);
2485 raw_spin_unlock(&kvm_lock);
2488 static void hardware_disable_all_nolock(void)
2490 BUG_ON(!kvm_usage_count);
2493 if (!kvm_usage_count)
2494 on_each_cpu(hardware_disable_nolock, NULL, 1);
2497 static void hardware_disable_all(void)
2499 raw_spin_lock(&kvm_lock);
2500 hardware_disable_all_nolock();
2501 raw_spin_unlock(&kvm_lock);
2504 static int hardware_enable_all(void)
2508 raw_spin_lock(&kvm_lock);
2511 if (kvm_usage_count == 1) {
2512 atomic_set(&hardware_enable_failed, 0);
2513 on_each_cpu(hardware_enable_nolock, NULL, 1);
2515 if (atomic_read(&hardware_enable_failed)) {
2516 hardware_disable_all_nolock();
2521 raw_spin_unlock(&kvm_lock);
2526 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2531 if (!kvm_usage_count)
2534 val &= ~CPU_TASKS_FROZEN;
2537 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2539 hardware_disable(NULL);
2542 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2544 hardware_enable(NULL);
2551 asmlinkage void kvm_spurious_fault(void)
2553 /* Fault while not rebooting. We want the trace. */
2556 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2558 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2562 * Some (well, at least mine) BIOSes hang on reboot if
2565 * And Intel TXT required VMX off for all cpu when system shutdown.
2567 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2568 kvm_rebooting = true;
2569 on_each_cpu(hardware_disable_nolock, NULL, 1);
2573 static struct notifier_block kvm_reboot_notifier = {
2574 .notifier_call = kvm_reboot,
2578 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2582 for (i = 0; i < bus->dev_count; i++) {
2583 struct kvm_io_device *pos = bus->range[i].dev;
2585 kvm_iodevice_destructor(pos);
2590 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2592 const struct kvm_io_range *r1 = p1;
2593 const struct kvm_io_range *r2 = p2;
2595 if (r1->addr < r2->addr)
2597 if (r1->addr + r1->len > r2->addr + r2->len)
2602 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2603 gpa_t addr, int len)
2605 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2611 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2612 kvm_io_bus_sort_cmp, NULL);
2617 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2618 gpa_t addr, int len)
2620 struct kvm_io_range *range, key;
2623 key = (struct kvm_io_range) {
2628 range = bsearch(&key, bus->range, bus->dev_count,
2629 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2633 off = range - bus->range;
2635 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2641 /* kvm_io_bus_write - called under kvm->slots_lock */
2642 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2643 int len, const void *val)
2646 struct kvm_io_bus *bus;
2647 struct kvm_io_range range;
2649 range = (struct kvm_io_range) {
2654 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2655 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2659 while (idx < bus->dev_count &&
2660 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2661 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2669 /* kvm_io_bus_read - called under kvm->slots_lock */
2670 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2674 struct kvm_io_bus *bus;
2675 struct kvm_io_range range;
2677 range = (struct kvm_io_range) {
2682 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2683 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2687 while (idx < bus->dev_count &&
2688 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2689 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2697 /* Caller must hold slots_lock. */
2698 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2699 int len, struct kvm_io_device *dev)
2701 struct kvm_io_bus *new_bus, *bus;
2703 bus = kvm->buses[bus_idx];
2704 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2707 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2708 sizeof(struct kvm_io_range)), GFP_KERNEL);
2711 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2712 sizeof(struct kvm_io_range)));
2713 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2714 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2715 synchronize_srcu_expedited(&kvm->srcu);
2721 /* Caller must hold slots_lock. */
2722 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2723 struct kvm_io_device *dev)
2726 struct kvm_io_bus *new_bus, *bus;
2728 bus = kvm->buses[bus_idx];
2730 for (i = 0; i < bus->dev_count; i++)
2731 if (bus->range[i].dev == dev) {
2739 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2740 sizeof(struct kvm_io_range)), GFP_KERNEL);
2744 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2745 new_bus->dev_count--;
2746 memcpy(new_bus->range + i, bus->range + i + 1,
2747 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2749 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2750 synchronize_srcu_expedited(&kvm->srcu);
2755 static struct notifier_block kvm_cpu_notifier = {
2756 .notifier_call = kvm_cpu_hotplug,
2759 static int vm_stat_get(void *_offset, u64 *val)
2761 unsigned offset = (long)_offset;
2765 raw_spin_lock(&kvm_lock);
2766 list_for_each_entry(kvm, &vm_list, vm_list)
2767 *val += *(u32 *)((void *)kvm + offset);
2768 raw_spin_unlock(&kvm_lock);
2772 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2774 static int vcpu_stat_get(void *_offset, u64 *val)
2776 unsigned offset = (long)_offset;
2778 struct kvm_vcpu *vcpu;
2782 raw_spin_lock(&kvm_lock);
2783 list_for_each_entry(kvm, &vm_list, vm_list)
2784 kvm_for_each_vcpu(i, vcpu, kvm)
2785 *val += *(u32 *)((void *)vcpu + offset);
2787 raw_spin_unlock(&kvm_lock);
2791 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2793 static const struct file_operations *stat_fops[] = {
2794 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2795 [KVM_STAT_VM] = &vm_stat_fops,
2798 static int kvm_init_debug(void)
2801 struct kvm_stats_debugfs_item *p;
2803 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2804 if (kvm_debugfs_dir == NULL)
2807 for (p = debugfs_entries; p->name; ++p) {
2808 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2809 (void *)(long)p->offset,
2810 stat_fops[p->kind]);
2811 if (p->dentry == NULL)
2818 debugfs_remove_recursive(kvm_debugfs_dir);
2823 static void kvm_exit_debug(void)
2825 struct kvm_stats_debugfs_item *p;
2827 for (p = debugfs_entries; p->name; ++p)
2828 debugfs_remove(p->dentry);
2829 debugfs_remove(kvm_debugfs_dir);
2832 static int kvm_suspend(void)
2834 if (kvm_usage_count)
2835 hardware_disable_nolock(NULL);
2839 static void kvm_resume(void)
2841 if (kvm_usage_count) {
2842 WARN_ON(raw_spin_is_locked(&kvm_lock));
2843 hardware_enable_nolock(NULL);
2847 static struct syscore_ops kvm_syscore_ops = {
2848 .suspend = kvm_suspend,
2849 .resume = kvm_resume,
2853 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2855 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2858 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2860 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2862 kvm_arch_vcpu_load(vcpu, cpu);
2865 static void kvm_sched_out(struct preempt_notifier *pn,
2866 struct task_struct *next)
2868 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2870 kvm_arch_vcpu_put(vcpu);
2873 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2874 struct module *module)
2879 r = kvm_arch_init(opaque);
2883 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2888 r = kvm_arch_hardware_setup();
2892 for_each_online_cpu(cpu) {
2893 smp_call_function_single(cpu,
2894 kvm_arch_check_processor_compat,
2900 r = register_cpu_notifier(&kvm_cpu_notifier);
2903 register_reboot_notifier(&kvm_reboot_notifier);
2905 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2907 vcpu_align = __alignof__(struct kvm_vcpu);
2908 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2910 if (!kvm_vcpu_cache) {
2915 r = kvm_async_pf_init();
2919 kvm_chardev_ops.owner = module;
2920 kvm_vm_fops.owner = module;
2921 kvm_vcpu_fops.owner = module;
2923 r = misc_register(&kvm_dev);
2925 printk(KERN_ERR "kvm: misc device register failed\n");
2929 register_syscore_ops(&kvm_syscore_ops);
2931 kvm_preempt_ops.sched_in = kvm_sched_in;
2932 kvm_preempt_ops.sched_out = kvm_sched_out;
2934 r = kvm_init_debug();
2936 printk(KERN_ERR "kvm: create debugfs files failed\n");
2943 unregister_syscore_ops(&kvm_syscore_ops);
2945 kvm_async_pf_deinit();
2947 kmem_cache_destroy(kvm_vcpu_cache);
2949 unregister_reboot_notifier(&kvm_reboot_notifier);
2950 unregister_cpu_notifier(&kvm_cpu_notifier);
2953 kvm_arch_hardware_unsetup();
2955 free_cpumask_var(cpus_hardware_enabled);
2961 EXPORT_SYMBOL_GPL(kvm_init);
2966 misc_deregister(&kvm_dev);
2967 kmem_cache_destroy(kvm_vcpu_cache);
2968 kvm_async_pf_deinit();
2969 unregister_syscore_ops(&kvm_syscore_ops);
2970 unregister_reboot_notifier(&kvm_reboot_notifier);
2971 unregister_cpu_notifier(&kvm_cpu_notifier);
2972 on_each_cpu(hardware_disable_nolock, NULL, 1);
2973 kvm_arch_hardware_unsetup();
2975 free_cpumask_var(cpus_hardware_enabled);
2977 EXPORT_SYMBOL_GPL(kvm_exit);