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/sysdev.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>
51 #include <asm/processor.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
57 #include "coalesced_mmio.h"
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
68 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71 DEFINE_SPINLOCK(kvm_lock);
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
83 struct dentry *kvm_debugfs_dir;
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
92 static bool kvm_rebooting;
94 static bool largepages_enabled = true;
96 static struct page *hwpoison_page;
97 static pfn_t hwpoison_pfn;
99 static struct page *fault_page;
100 static pfn_t fault_pfn;
102 inline int kvm_is_mmio_pfn(pfn_t pfn)
104 if (pfn_valid(pfn)) {
105 struct page *page = compound_head(pfn_to_page(pfn));
106 return PageReserved(page);
113 * Switches to specified vcpu, until a matching vcpu_put()
115 void vcpu_load(struct kvm_vcpu *vcpu)
119 mutex_lock(&vcpu->mutex);
121 preempt_notifier_register(&vcpu->preempt_notifier);
122 kvm_arch_vcpu_load(vcpu, cpu);
126 void vcpu_put(struct kvm_vcpu *vcpu)
129 kvm_arch_vcpu_put(vcpu);
130 preempt_notifier_unregister(&vcpu->preempt_notifier);
132 mutex_unlock(&vcpu->mutex);
135 static void ack_flush(void *_completed)
139 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
144 struct kvm_vcpu *vcpu;
146 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
148 raw_spin_lock(&kvm->requests_lock);
149 me = smp_processor_id();
150 kvm_for_each_vcpu(i, vcpu, kvm) {
151 if (kvm_make_check_request(req, vcpu))
154 if (cpus != NULL && cpu != -1 && cpu != me)
155 cpumask_set_cpu(cpu, cpus);
157 if (unlikely(cpus == NULL))
158 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
159 else if (!cpumask_empty(cpus))
160 smp_call_function_many(cpus, ack_flush, NULL, 1);
163 raw_spin_unlock(&kvm->requests_lock);
164 free_cpumask_var(cpus);
168 void kvm_flush_remote_tlbs(struct kvm *kvm)
170 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
171 ++kvm->stat.remote_tlb_flush;
174 void kvm_reload_remote_mmus(struct kvm *kvm)
176 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
179 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
184 mutex_init(&vcpu->mutex);
188 init_waitqueue_head(&vcpu->wq);
190 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
195 vcpu->run = page_address(page);
197 r = kvm_arch_vcpu_init(vcpu);
203 free_page((unsigned long)vcpu->run);
207 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
209 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
211 kvm_arch_vcpu_uninit(vcpu);
212 free_page((unsigned long)vcpu->run);
214 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
216 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
217 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
219 return container_of(mn, struct kvm, mmu_notifier);
222 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
223 struct mm_struct *mm,
224 unsigned long address)
226 struct kvm *kvm = mmu_notifier_to_kvm(mn);
227 int need_tlb_flush, idx;
230 * When ->invalidate_page runs, the linux pte has been zapped
231 * already but the page is still allocated until
232 * ->invalidate_page returns. So if we increase the sequence
233 * here the kvm page fault will notice if the spte can't be
234 * established because the page is going to be freed. If
235 * instead the kvm page fault establishes the spte before
236 * ->invalidate_page runs, kvm_unmap_hva will release it
239 * The sequence increase only need to be seen at spin_unlock
240 * time, and not at spin_lock time.
242 * Increasing the sequence after the spin_unlock would be
243 * unsafe because the kvm page fault could then establish the
244 * pte after kvm_unmap_hva returned, without noticing the page
245 * is going to be freed.
247 idx = srcu_read_lock(&kvm->srcu);
248 spin_lock(&kvm->mmu_lock);
249 kvm->mmu_notifier_seq++;
250 need_tlb_flush = kvm_unmap_hva(kvm, address);
251 spin_unlock(&kvm->mmu_lock);
252 srcu_read_unlock(&kvm->srcu, idx);
254 /* we've to flush the tlb before the pages can be freed */
256 kvm_flush_remote_tlbs(kvm);
260 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address,
265 struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 idx = srcu_read_lock(&kvm->srcu);
269 spin_lock(&kvm->mmu_lock);
270 kvm->mmu_notifier_seq++;
271 kvm_set_spte_hva(kvm, address, pte);
272 spin_unlock(&kvm->mmu_lock);
273 srcu_read_unlock(&kvm->srcu, idx);
276 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
277 struct mm_struct *mm,
281 struct kvm *kvm = mmu_notifier_to_kvm(mn);
282 int need_tlb_flush = 0, idx;
284 idx = srcu_read_lock(&kvm->srcu);
285 spin_lock(&kvm->mmu_lock);
287 * The count increase must become visible at unlock time as no
288 * spte can be established without taking the mmu_lock and
289 * count is also read inside the mmu_lock critical section.
291 kvm->mmu_notifier_count++;
292 for (; start < end; start += PAGE_SIZE)
293 need_tlb_flush |= kvm_unmap_hva(kvm, start);
294 spin_unlock(&kvm->mmu_lock);
295 srcu_read_unlock(&kvm->srcu, idx);
297 /* we've to flush the tlb before the pages can be freed */
299 kvm_flush_remote_tlbs(kvm);
302 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
303 struct mm_struct *mm,
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 spin_lock(&kvm->mmu_lock);
311 * This sequence increase will notify the kvm page fault that
312 * the page that is going to be mapped in the spte could have
315 kvm->mmu_notifier_seq++;
317 * The above sequence increase must be visible before the
318 * below count decrease but both values are read by the kvm
319 * page fault under mmu_lock spinlock so we don't need to add
320 * a smb_wmb() here in between the two.
322 kvm->mmu_notifier_count--;
323 spin_unlock(&kvm->mmu_lock);
325 BUG_ON(kvm->mmu_notifier_count < 0);
328 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
329 struct mm_struct *mm,
330 unsigned long address)
332 struct kvm *kvm = mmu_notifier_to_kvm(mn);
335 idx = srcu_read_lock(&kvm->srcu);
336 spin_lock(&kvm->mmu_lock);
337 young = kvm_age_hva(kvm, address);
338 spin_unlock(&kvm->mmu_lock);
339 srcu_read_unlock(&kvm->srcu, idx);
342 kvm_flush_remote_tlbs(kvm);
347 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
348 struct mm_struct *mm)
350 struct kvm *kvm = mmu_notifier_to_kvm(mn);
353 idx = srcu_read_lock(&kvm->srcu);
354 kvm_arch_flush_shadow(kvm);
355 srcu_read_unlock(&kvm->srcu, idx);
358 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
359 .invalidate_page = kvm_mmu_notifier_invalidate_page,
360 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
361 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
362 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
363 .change_pte = kvm_mmu_notifier_change_pte,
364 .release = kvm_mmu_notifier_release,
367 static int kvm_init_mmu_notifier(struct kvm *kvm)
369 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
370 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
373 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
375 static int kvm_init_mmu_notifier(struct kvm *kvm)
380 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
382 static struct kvm *kvm_create_vm(void)
385 struct kvm *kvm = kvm_arch_create_vm();
390 r = hardware_enable_all();
392 goto out_err_nodisable;
394 #ifdef CONFIG_HAVE_KVM_IRQCHIP
395 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
396 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
400 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
403 if (init_srcu_struct(&kvm->srcu))
405 for (i = 0; i < KVM_NR_BUSES; i++) {
406 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
408 if (!kvm->buses[i]) {
409 cleanup_srcu_struct(&kvm->srcu);
414 r = kvm_init_mmu_notifier(kvm);
416 cleanup_srcu_struct(&kvm->srcu);
420 kvm->mm = current->mm;
421 atomic_inc(&kvm->mm->mm_count);
422 spin_lock_init(&kvm->mmu_lock);
423 raw_spin_lock_init(&kvm->requests_lock);
424 kvm_eventfd_init(kvm);
425 mutex_init(&kvm->lock);
426 mutex_init(&kvm->irq_lock);
427 mutex_init(&kvm->slots_lock);
428 atomic_set(&kvm->users_count, 1);
429 spin_lock(&kvm_lock);
430 list_add(&kvm->vm_list, &vm_list);
431 spin_unlock(&kvm_lock);
436 hardware_disable_all();
438 for (i = 0; i < KVM_NR_BUSES; i++)
439 kfree(kvm->buses[i]);
440 kfree(kvm->memslots);
446 * Free any memory in @free but not in @dont.
448 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
449 struct kvm_memory_slot *dont)
453 if (!dont || free->rmap != dont->rmap)
456 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
457 vfree(free->dirty_bitmap);
460 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
461 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
462 vfree(free->lpage_info[i]);
463 free->lpage_info[i] = NULL;
468 free->dirty_bitmap = NULL;
472 void kvm_free_physmem(struct kvm *kvm)
475 struct kvm_memslots *slots = kvm->memslots;
477 for (i = 0; i < slots->nmemslots; ++i)
478 kvm_free_physmem_slot(&slots->memslots[i], NULL);
480 kfree(kvm->memslots);
483 static void kvm_destroy_vm(struct kvm *kvm)
486 struct mm_struct *mm = kvm->mm;
488 kvm_arch_sync_events(kvm);
489 spin_lock(&kvm_lock);
490 list_del(&kvm->vm_list);
491 spin_unlock(&kvm_lock);
492 kvm_free_irq_routing(kvm);
493 for (i = 0; i < KVM_NR_BUSES; i++)
494 kvm_io_bus_destroy(kvm->buses[i]);
495 kvm_coalesced_mmio_free(kvm);
496 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
497 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
499 kvm_arch_flush_shadow(kvm);
501 kvm_arch_destroy_vm(kvm);
502 hardware_disable_all();
506 void kvm_get_kvm(struct kvm *kvm)
508 atomic_inc(&kvm->users_count);
510 EXPORT_SYMBOL_GPL(kvm_get_kvm);
512 void kvm_put_kvm(struct kvm *kvm)
514 if (atomic_dec_and_test(&kvm->users_count))
517 EXPORT_SYMBOL_GPL(kvm_put_kvm);
520 static int kvm_vm_release(struct inode *inode, struct file *filp)
522 struct kvm *kvm = filp->private_data;
524 kvm_irqfd_release(kvm);
531 * Allocate some memory and give it an address in the guest physical address
534 * Discontiguous memory is allowed, mostly for framebuffers.
536 * Must be called holding mmap_sem for write.
538 int __kvm_set_memory_region(struct kvm *kvm,
539 struct kvm_userspace_memory_region *mem,
542 int r, flush_shadow = 0;
544 unsigned long npages;
546 struct kvm_memory_slot *memslot;
547 struct kvm_memory_slot old, new;
548 struct kvm_memslots *slots, *old_memslots;
551 /* General sanity checks */
552 if (mem->memory_size & (PAGE_SIZE - 1))
554 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
556 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
558 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
560 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
563 memslot = &kvm->memslots->memslots[mem->slot];
564 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
565 npages = mem->memory_size >> PAGE_SHIFT;
568 if (npages > KVM_MEM_MAX_NR_PAGES)
572 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
574 new = old = *memslot;
577 new.base_gfn = base_gfn;
579 new.flags = mem->flags;
581 /* Disallow changing a memory slot's size. */
583 if (npages && old.npages && npages != old.npages)
586 /* Check for overlaps */
588 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
589 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
591 if (s == memslot || !s->npages)
593 if (!((base_gfn + npages <= s->base_gfn) ||
594 (base_gfn >= s->base_gfn + s->npages)))
598 /* Free page dirty bitmap if unneeded */
599 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
600 new.dirty_bitmap = NULL;
604 /* Allocate if a slot is being created */
606 if (npages && !new.rmap) {
607 new.rmap = vmalloc(npages * sizeof(*new.rmap));
612 memset(new.rmap, 0, npages * sizeof(*new.rmap));
614 new.user_alloc = user_alloc;
615 new.userspace_addr = mem->userspace_addr;
620 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
626 /* Avoid unused variable warning if no large pages */
629 if (new.lpage_info[i])
632 lpages = 1 + ((base_gfn + npages - 1)
633 >> KVM_HPAGE_GFN_SHIFT(level));
634 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
636 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
638 if (!new.lpage_info[i])
641 memset(new.lpage_info[i], 0,
642 lpages * sizeof(*new.lpage_info[i]));
644 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
645 new.lpage_info[i][0].write_count = 1;
646 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
647 new.lpage_info[i][lpages - 1].write_count = 1;
648 ugfn = new.userspace_addr >> PAGE_SHIFT;
650 * If the gfn and userspace address are not aligned wrt each
651 * other, or if explicitly asked to, disable large page
652 * support for this slot
654 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
656 for (j = 0; j < lpages; ++j)
657 new.lpage_info[i][j].write_count = 1;
662 /* Allocate page dirty bitmap if needed */
663 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
664 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
666 new.dirty_bitmap = vmalloc(dirty_bytes);
667 if (!new.dirty_bitmap)
669 memset(new.dirty_bitmap, 0, dirty_bytes);
670 /* destroy any largepage mappings for dirty tracking */
674 #else /* not defined CONFIG_S390 */
675 new.user_alloc = user_alloc;
677 new.userspace_addr = mem->userspace_addr;
678 #endif /* not defined CONFIG_S390 */
682 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
685 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
686 if (mem->slot >= slots->nmemslots)
687 slots->nmemslots = mem->slot + 1;
688 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
690 old_memslots = kvm->memslots;
691 rcu_assign_pointer(kvm->memslots, slots);
692 synchronize_srcu_expedited(&kvm->srcu);
693 /* From this point no new shadow pages pointing to a deleted
694 * memslot will be created.
696 * validation of sp->gfn happens in:
697 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
698 * - kvm_is_visible_gfn (mmu_check_roots)
700 kvm_arch_flush_shadow(kvm);
704 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
708 /* map the pages in iommu page table */
710 r = kvm_iommu_map_pages(kvm, &new);
716 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
719 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
720 if (mem->slot >= slots->nmemslots)
721 slots->nmemslots = mem->slot + 1;
723 /* actual memory is freed via old in kvm_free_physmem_slot below */
726 new.dirty_bitmap = NULL;
727 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
728 new.lpage_info[i] = NULL;
731 slots->memslots[mem->slot] = new;
732 old_memslots = kvm->memslots;
733 rcu_assign_pointer(kvm->memslots, slots);
734 synchronize_srcu_expedited(&kvm->srcu);
736 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
738 kvm_free_physmem_slot(&old, &new);
742 kvm_arch_flush_shadow(kvm);
747 kvm_free_physmem_slot(&new, &old);
752 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
754 int kvm_set_memory_region(struct kvm *kvm,
755 struct kvm_userspace_memory_region *mem,
760 mutex_lock(&kvm->slots_lock);
761 r = __kvm_set_memory_region(kvm, mem, user_alloc);
762 mutex_unlock(&kvm->slots_lock);
765 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
767 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
769 kvm_userspace_memory_region *mem,
772 if (mem->slot >= KVM_MEMORY_SLOTS)
774 return kvm_set_memory_region(kvm, mem, user_alloc);
777 int kvm_get_dirty_log(struct kvm *kvm,
778 struct kvm_dirty_log *log, int *is_dirty)
780 struct kvm_memory_slot *memslot;
783 unsigned long any = 0;
786 if (log->slot >= KVM_MEMORY_SLOTS)
789 memslot = &kvm->memslots->memslots[log->slot];
791 if (!memslot->dirty_bitmap)
794 n = kvm_dirty_bitmap_bytes(memslot);
796 for (i = 0; !any && i < n/sizeof(long); ++i)
797 any = memslot->dirty_bitmap[i];
800 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
811 void kvm_disable_largepages(void)
813 largepages_enabled = false;
815 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
817 int is_error_page(struct page *page)
819 return page == bad_page || page == hwpoison_page || page == fault_page;
821 EXPORT_SYMBOL_GPL(is_error_page);
823 int is_error_pfn(pfn_t pfn)
825 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
827 EXPORT_SYMBOL_GPL(is_error_pfn);
829 int is_hwpoison_pfn(pfn_t pfn)
831 return pfn == hwpoison_pfn;
833 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
835 int is_fault_pfn(pfn_t pfn)
837 return pfn == fault_pfn;
839 EXPORT_SYMBOL_GPL(is_fault_pfn);
841 static inline unsigned long bad_hva(void)
846 int kvm_is_error_hva(unsigned long addr)
848 return addr == bad_hva();
850 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
852 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
855 struct kvm_memslots *slots = kvm_memslots(kvm);
857 for (i = 0; i < slots->nmemslots; ++i) {
858 struct kvm_memory_slot *memslot = &slots->memslots[i];
860 if (gfn >= memslot->base_gfn
861 && gfn < memslot->base_gfn + memslot->npages)
866 EXPORT_SYMBOL_GPL(gfn_to_memslot);
868 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
871 struct kvm_memslots *slots = kvm_memslots(kvm);
873 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
874 struct kvm_memory_slot *memslot = &slots->memslots[i];
876 if (memslot->flags & KVM_MEMSLOT_INVALID)
879 if (gfn >= memslot->base_gfn
880 && gfn < memslot->base_gfn + memslot->npages)
885 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
887 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
889 struct vm_area_struct *vma;
890 unsigned long addr, size;
894 addr = gfn_to_hva(kvm, gfn);
895 if (kvm_is_error_hva(addr))
898 down_read(¤t->mm->mmap_sem);
899 vma = find_vma(current->mm, addr);
903 size = vma_kernel_pagesize(vma);
906 up_read(¤t->mm->mmap_sem);
911 int memslot_id(struct kvm *kvm, gfn_t gfn)
914 struct kvm_memslots *slots = kvm_memslots(kvm);
915 struct kvm_memory_slot *memslot = NULL;
917 for (i = 0; i < slots->nmemslots; ++i) {
918 memslot = &slots->memslots[i];
920 if (gfn >= memslot->base_gfn
921 && gfn < memslot->base_gfn + memslot->npages)
925 return memslot - slots->memslots;
928 static unsigned long gfn_to_hva_many(struct kvm *kvm, gfn_t gfn,
931 struct kvm_memory_slot *slot;
933 slot = gfn_to_memslot(kvm, gfn);
934 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
938 *nr_pages = slot->npages - (gfn - slot->base_gfn);
940 return gfn_to_hva_memslot(slot, gfn);
943 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
945 return gfn_to_hva_many(kvm, gfn, NULL);
947 EXPORT_SYMBOL_GPL(gfn_to_hva);
949 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic)
951 struct page *page[1];
956 npages = __get_user_pages_fast(addr, 1, 1, page);
959 npages = get_user_pages_fast(addr, 1, 1, page);
962 if (unlikely(npages != 1)) {
963 struct vm_area_struct *vma;
966 goto return_fault_page;
968 down_read(¤t->mm->mmap_sem);
969 if (is_hwpoison_address(addr)) {
970 up_read(¤t->mm->mmap_sem);
971 get_page(hwpoison_page);
972 return page_to_pfn(hwpoison_page);
975 vma = find_vma(current->mm, addr);
977 if (vma == NULL || addr < vma->vm_start ||
978 !(vma->vm_flags & VM_PFNMAP)) {
979 up_read(¤t->mm->mmap_sem);
981 get_page(fault_page);
982 return page_to_pfn(fault_page);
985 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
986 up_read(¤t->mm->mmap_sem);
987 BUG_ON(!kvm_is_mmio_pfn(pfn));
989 pfn = page_to_pfn(page[0]);
994 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
996 return hva_to_pfn(kvm, addr, true);
998 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1000 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic)
1004 addr = gfn_to_hva(kvm, gfn);
1005 if (kvm_is_error_hva(addr)) {
1007 return page_to_pfn(bad_page);
1010 return hva_to_pfn(kvm, addr, atomic);
1013 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1015 return __gfn_to_pfn(kvm, gfn, true);
1017 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1019 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1021 return __gfn_to_pfn(kvm, gfn, false);
1023 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1025 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1026 struct kvm_memory_slot *slot, gfn_t gfn)
1028 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1029 return hva_to_pfn(kvm, addr, false);
1032 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1038 addr = gfn_to_hva_many(kvm, gfn, &entry);
1039 if (kvm_is_error_hva(addr))
1042 if (entry < nr_pages)
1045 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1047 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1049 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1053 pfn = gfn_to_pfn(kvm, gfn);
1054 if (!kvm_is_mmio_pfn(pfn))
1055 return pfn_to_page(pfn);
1057 WARN_ON(kvm_is_mmio_pfn(pfn));
1063 EXPORT_SYMBOL_GPL(gfn_to_page);
1065 void kvm_release_page_clean(struct page *page)
1067 kvm_release_pfn_clean(page_to_pfn(page));
1069 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1071 void kvm_release_pfn_clean(pfn_t pfn)
1073 if (!kvm_is_mmio_pfn(pfn))
1074 put_page(pfn_to_page(pfn));
1076 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1078 void kvm_release_page_dirty(struct page *page)
1080 kvm_release_pfn_dirty(page_to_pfn(page));
1082 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1084 void kvm_release_pfn_dirty(pfn_t pfn)
1086 kvm_set_pfn_dirty(pfn);
1087 kvm_release_pfn_clean(pfn);
1089 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1091 void kvm_set_page_dirty(struct page *page)
1093 kvm_set_pfn_dirty(page_to_pfn(page));
1095 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1097 void kvm_set_pfn_dirty(pfn_t pfn)
1099 if (!kvm_is_mmio_pfn(pfn)) {
1100 struct page *page = pfn_to_page(pfn);
1101 if (!PageReserved(page))
1105 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1107 void kvm_set_pfn_accessed(pfn_t pfn)
1109 if (!kvm_is_mmio_pfn(pfn))
1110 mark_page_accessed(pfn_to_page(pfn));
1112 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1114 void kvm_get_pfn(pfn_t pfn)
1116 if (!kvm_is_mmio_pfn(pfn))
1117 get_page(pfn_to_page(pfn));
1119 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1121 static int next_segment(unsigned long len, int offset)
1123 if (len > PAGE_SIZE - offset)
1124 return PAGE_SIZE - offset;
1129 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1135 addr = gfn_to_hva(kvm, gfn);
1136 if (kvm_is_error_hva(addr))
1138 r = copy_from_user(data, (void __user *)addr + offset, len);
1143 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1145 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1147 gfn_t gfn = gpa >> PAGE_SHIFT;
1149 int offset = offset_in_page(gpa);
1152 while ((seg = next_segment(len, offset)) != 0) {
1153 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1163 EXPORT_SYMBOL_GPL(kvm_read_guest);
1165 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1170 gfn_t gfn = gpa >> PAGE_SHIFT;
1171 int offset = offset_in_page(gpa);
1173 addr = gfn_to_hva(kvm, gfn);
1174 if (kvm_is_error_hva(addr))
1176 pagefault_disable();
1177 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1183 EXPORT_SYMBOL(kvm_read_guest_atomic);
1185 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1186 int offset, int len)
1191 addr = gfn_to_hva(kvm, gfn);
1192 if (kvm_is_error_hva(addr))
1194 r = copy_to_user((void __user *)addr + offset, data, len);
1197 mark_page_dirty(kvm, gfn);
1200 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1202 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1205 gfn_t gfn = gpa >> PAGE_SHIFT;
1207 int offset = offset_in_page(gpa);
1210 while ((seg = next_segment(len, offset)) != 0) {
1211 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1222 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1224 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1226 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1228 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1230 gfn_t gfn = gpa >> PAGE_SHIFT;
1232 int offset = offset_in_page(gpa);
1235 while ((seg = next_segment(len, offset)) != 0) {
1236 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1245 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1247 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1249 struct kvm_memory_slot *memslot;
1251 memslot = gfn_to_memslot(kvm, gfn);
1252 if (memslot && memslot->dirty_bitmap) {
1253 unsigned long rel_gfn = gfn - memslot->base_gfn;
1255 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1260 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1262 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1267 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1269 if (kvm_arch_vcpu_runnable(vcpu)) {
1270 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1273 if (kvm_cpu_has_pending_timer(vcpu))
1275 if (signal_pending(current))
1281 finish_wait(&vcpu->wq, &wait);
1284 void kvm_resched(struct kvm_vcpu *vcpu)
1286 if (!need_resched())
1290 EXPORT_SYMBOL_GPL(kvm_resched);
1292 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1297 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1299 /* Sleep for 100 us, and hope lock-holder got scheduled */
1300 expires = ktime_add_ns(ktime_get(), 100000UL);
1301 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1303 finish_wait(&vcpu->wq, &wait);
1305 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1307 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1309 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1312 if (vmf->pgoff == 0)
1313 page = virt_to_page(vcpu->run);
1315 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1316 page = virt_to_page(vcpu->arch.pio_data);
1318 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1319 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1320 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1323 return VM_FAULT_SIGBUS;
1329 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1330 .fault = kvm_vcpu_fault,
1333 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1335 vma->vm_ops = &kvm_vcpu_vm_ops;
1339 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1341 struct kvm_vcpu *vcpu = filp->private_data;
1343 kvm_put_kvm(vcpu->kvm);
1347 static struct file_operations kvm_vcpu_fops = {
1348 .release = kvm_vcpu_release,
1349 .unlocked_ioctl = kvm_vcpu_ioctl,
1350 .compat_ioctl = kvm_vcpu_ioctl,
1351 .mmap = kvm_vcpu_mmap,
1352 .llseek = noop_llseek,
1356 * Allocates an inode for the vcpu.
1358 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1360 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1364 * Creates some virtual cpus. Good luck creating more than one.
1366 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1369 struct kvm_vcpu *vcpu, *v;
1371 vcpu = kvm_arch_vcpu_create(kvm, id);
1373 return PTR_ERR(vcpu);
1375 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1377 r = kvm_arch_vcpu_setup(vcpu);
1381 mutex_lock(&kvm->lock);
1382 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1387 kvm_for_each_vcpu(r, v, kvm)
1388 if (v->vcpu_id == id) {
1393 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1395 /* Now it's all set up, let userspace reach it */
1397 r = create_vcpu_fd(vcpu);
1403 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1405 atomic_inc(&kvm->online_vcpus);
1407 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1408 if (kvm->bsp_vcpu_id == id)
1409 kvm->bsp_vcpu = vcpu;
1411 mutex_unlock(&kvm->lock);
1415 mutex_unlock(&kvm->lock);
1416 kvm_arch_vcpu_destroy(vcpu);
1420 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1423 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1424 vcpu->sigset_active = 1;
1425 vcpu->sigset = *sigset;
1427 vcpu->sigset_active = 0;
1431 static long kvm_vcpu_ioctl(struct file *filp,
1432 unsigned int ioctl, unsigned long arg)
1434 struct kvm_vcpu *vcpu = filp->private_data;
1435 void __user *argp = (void __user *)arg;
1437 struct kvm_fpu *fpu = NULL;
1438 struct kvm_sregs *kvm_sregs = NULL;
1440 if (vcpu->kvm->mm != current->mm)
1443 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1445 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1446 * so vcpu_load() would break it.
1448 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1449 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1459 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1461 case KVM_GET_REGS: {
1462 struct kvm_regs *kvm_regs;
1465 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1468 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1472 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1479 case KVM_SET_REGS: {
1480 struct kvm_regs *kvm_regs;
1483 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1487 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1489 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1497 case KVM_GET_SREGS: {
1498 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1502 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1506 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1511 case KVM_SET_SREGS: {
1512 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1517 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1519 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1525 case KVM_GET_MP_STATE: {
1526 struct kvm_mp_state mp_state;
1528 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1532 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1537 case KVM_SET_MP_STATE: {
1538 struct kvm_mp_state mp_state;
1541 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1543 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1549 case KVM_TRANSLATE: {
1550 struct kvm_translation tr;
1553 if (copy_from_user(&tr, argp, sizeof tr))
1555 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1559 if (copy_to_user(argp, &tr, sizeof tr))
1564 case KVM_SET_GUEST_DEBUG: {
1565 struct kvm_guest_debug dbg;
1568 if (copy_from_user(&dbg, argp, sizeof dbg))
1570 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1576 case KVM_SET_SIGNAL_MASK: {
1577 struct kvm_signal_mask __user *sigmask_arg = argp;
1578 struct kvm_signal_mask kvm_sigmask;
1579 sigset_t sigset, *p;
1584 if (copy_from_user(&kvm_sigmask, argp,
1585 sizeof kvm_sigmask))
1588 if (kvm_sigmask.len != sizeof sigset)
1591 if (copy_from_user(&sigset, sigmask_arg->sigset,
1596 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1600 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1604 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1608 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1614 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1619 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1621 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1628 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1637 static long kvm_vm_ioctl(struct file *filp,
1638 unsigned int ioctl, unsigned long arg)
1640 struct kvm *kvm = filp->private_data;
1641 void __user *argp = (void __user *)arg;
1644 if (kvm->mm != current->mm)
1647 case KVM_CREATE_VCPU:
1648 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1652 case KVM_SET_USER_MEMORY_REGION: {
1653 struct kvm_userspace_memory_region kvm_userspace_mem;
1656 if (copy_from_user(&kvm_userspace_mem, argp,
1657 sizeof kvm_userspace_mem))
1660 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1665 case KVM_GET_DIRTY_LOG: {
1666 struct kvm_dirty_log log;
1669 if (copy_from_user(&log, argp, sizeof log))
1671 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1676 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1677 case KVM_REGISTER_COALESCED_MMIO: {
1678 struct kvm_coalesced_mmio_zone zone;
1680 if (copy_from_user(&zone, argp, sizeof zone))
1682 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1688 case KVM_UNREGISTER_COALESCED_MMIO: {
1689 struct kvm_coalesced_mmio_zone zone;
1691 if (copy_from_user(&zone, argp, sizeof zone))
1693 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1701 struct kvm_irqfd data;
1704 if (copy_from_user(&data, argp, sizeof data))
1706 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1709 case KVM_IOEVENTFD: {
1710 struct kvm_ioeventfd data;
1713 if (copy_from_user(&data, argp, sizeof data))
1715 r = kvm_ioeventfd(kvm, &data);
1718 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1719 case KVM_SET_BOOT_CPU_ID:
1721 mutex_lock(&kvm->lock);
1722 if (atomic_read(&kvm->online_vcpus) != 0)
1725 kvm->bsp_vcpu_id = arg;
1726 mutex_unlock(&kvm->lock);
1730 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1732 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1738 #ifdef CONFIG_COMPAT
1739 struct compat_kvm_dirty_log {
1743 compat_uptr_t dirty_bitmap; /* one bit per page */
1748 static long kvm_vm_compat_ioctl(struct file *filp,
1749 unsigned int ioctl, unsigned long arg)
1751 struct kvm *kvm = filp->private_data;
1754 if (kvm->mm != current->mm)
1757 case KVM_GET_DIRTY_LOG: {
1758 struct compat_kvm_dirty_log compat_log;
1759 struct kvm_dirty_log log;
1762 if (copy_from_user(&compat_log, (void __user *)arg,
1763 sizeof(compat_log)))
1765 log.slot = compat_log.slot;
1766 log.padding1 = compat_log.padding1;
1767 log.padding2 = compat_log.padding2;
1768 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1770 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1776 r = kvm_vm_ioctl(filp, ioctl, arg);
1784 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1786 struct page *page[1];
1789 gfn_t gfn = vmf->pgoff;
1790 struct kvm *kvm = vma->vm_file->private_data;
1792 addr = gfn_to_hva(kvm, gfn);
1793 if (kvm_is_error_hva(addr))
1794 return VM_FAULT_SIGBUS;
1796 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1798 if (unlikely(npages != 1))
1799 return VM_FAULT_SIGBUS;
1801 vmf->page = page[0];
1805 static const struct vm_operations_struct kvm_vm_vm_ops = {
1806 .fault = kvm_vm_fault,
1809 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1811 vma->vm_ops = &kvm_vm_vm_ops;
1815 static struct file_operations kvm_vm_fops = {
1816 .release = kvm_vm_release,
1817 .unlocked_ioctl = kvm_vm_ioctl,
1818 #ifdef CONFIG_COMPAT
1819 .compat_ioctl = kvm_vm_compat_ioctl,
1821 .mmap = kvm_vm_mmap,
1822 .llseek = noop_llseek,
1825 static int kvm_dev_ioctl_create_vm(void)
1830 kvm = kvm_create_vm();
1832 return PTR_ERR(kvm);
1833 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1834 r = kvm_coalesced_mmio_init(kvm);
1840 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1847 static long kvm_dev_ioctl_check_extension_generic(long arg)
1850 case KVM_CAP_USER_MEMORY:
1851 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1852 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1853 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1854 case KVM_CAP_SET_BOOT_CPU_ID:
1856 case KVM_CAP_INTERNAL_ERROR_DATA:
1858 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1859 case KVM_CAP_IRQ_ROUTING:
1860 return KVM_MAX_IRQ_ROUTES;
1865 return kvm_dev_ioctl_check_extension(arg);
1868 static long kvm_dev_ioctl(struct file *filp,
1869 unsigned int ioctl, unsigned long arg)
1874 case KVM_GET_API_VERSION:
1878 r = KVM_API_VERSION;
1884 r = kvm_dev_ioctl_create_vm();
1886 case KVM_CHECK_EXTENSION:
1887 r = kvm_dev_ioctl_check_extension_generic(arg);
1889 case KVM_GET_VCPU_MMAP_SIZE:
1893 r = PAGE_SIZE; /* struct kvm_run */
1895 r += PAGE_SIZE; /* pio data page */
1897 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1898 r += PAGE_SIZE; /* coalesced mmio ring page */
1901 case KVM_TRACE_ENABLE:
1902 case KVM_TRACE_PAUSE:
1903 case KVM_TRACE_DISABLE:
1907 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1913 static struct file_operations kvm_chardev_ops = {
1914 .unlocked_ioctl = kvm_dev_ioctl,
1915 .compat_ioctl = kvm_dev_ioctl,
1916 .llseek = noop_llseek,
1919 static struct miscdevice kvm_dev = {
1925 static void hardware_enable(void *junk)
1927 int cpu = raw_smp_processor_id();
1930 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1933 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1935 r = kvm_arch_hardware_enable(NULL);
1938 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1939 atomic_inc(&hardware_enable_failed);
1940 printk(KERN_INFO "kvm: enabling virtualization on "
1941 "CPU%d failed\n", cpu);
1945 static void hardware_disable(void *junk)
1947 int cpu = raw_smp_processor_id();
1949 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1951 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1952 kvm_arch_hardware_disable(NULL);
1955 static void hardware_disable_all_nolock(void)
1957 BUG_ON(!kvm_usage_count);
1960 if (!kvm_usage_count)
1961 on_each_cpu(hardware_disable, NULL, 1);
1964 static void hardware_disable_all(void)
1966 spin_lock(&kvm_lock);
1967 hardware_disable_all_nolock();
1968 spin_unlock(&kvm_lock);
1971 static int hardware_enable_all(void)
1975 spin_lock(&kvm_lock);
1978 if (kvm_usage_count == 1) {
1979 atomic_set(&hardware_enable_failed, 0);
1980 on_each_cpu(hardware_enable, NULL, 1);
1982 if (atomic_read(&hardware_enable_failed)) {
1983 hardware_disable_all_nolock();
1988 spin_unlock(&kvm_lock);
1993 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1998 if (!kvm_usage_count)
2001 val &= ~CPU_TASKS_FROZEN;
2004 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2006 hardware_disable(NULL);
2009 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2011 spin_lock(&kvm_lock);
2012 hardware_enable(NULL);
2013 spin_unlock(&kvm_lock);
2020 asmlinkage void kvm_handle_fault_on_reboot(void)
2022 if (kvm_rebooting) {
2023 /* spin while reset goes on */
2028 /* Fault while not rebooting. We want the trace. */
2031 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2033 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2037 * Some (well, at least mine) BIOSes hang on reboot if
2040 * And Intel TXT required VMX off for all cpu when system shutdown.
2042 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2043 kvm_rebooting = true;
2044 on_each_cpu(hardware_disable, NULL, 1);
2048 static struct notifier_block kvm_reboot_notifier = {
2049 .notifier_call = kvm_reboot,
2053 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2057 for (i = 0; i < bus->dev_count; i++) {
2058 struct kvm_io_device *pos = bus->devs[i];
2060 kvm_iodevice_destructor(pos);
2065 /* kvm_io_bus_write - called under kvm->slots_lock */
2066 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2067 int len, const void *val)
2070 struct kvm_io_bus *bus;
2072 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2073 for (i = 0; i < bus->dev_count; i++)
2074 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2079 /* kvm_io_bus_read - called under kvm->slots_lock */
2080 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2084 struct kvm_io_bus *bus;
2086 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2087 for (i = 0; i < bus->dev_count; i++)
2088 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2093 /* Caller must hold slots_lock. */
2094 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2095 struct kvm_io_device *dev)
2097 struct kvm_io_bus *new_bus, *bus;
2099 bus = kvm->buses[bus_idx];
2100 if (bus->dev_count > NR_IOBUS_DEVS-1)
2103 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2106 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2107 new_bus->devs[new_bus->dev_count++] = dev;
2108 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2109 synchronize_srcu_expedited(&kvm->srcu);
2115 /* Caller must hold slots_lock. */
2116 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2117 struct kvm_io_device *dev)
2120 struct kvm_io_bus *new_bus, *bus;
2122 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2126 bus = kvm->buses[bus_idx];
2127 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2130 for (i = 0; i < new_bus->dev_count; i++)
2131 if (new_bus->devs[i] == dev) {
2133 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2142 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2143 synchronize_srcu_expedited(&kvm->srcu);
2148 static struct notifier_block kvm_cpu_notifier = {
2149 .notifier_call = kvm_cpu_hotplug,
2152 static int vm_stat_get(void *_offset, u64 *val)
2154 unsigned offset = (long)_offset;
2158 spin_lock(&kvm_lock);
2159 list_for_each_entry(kvm, &vm_list, vm_list)
2160 *val += *(u32 *)((void *)kvm + offset);
2161 spin_unlock(&kvm_lock);
2165 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2167 static int vcpu_stat_get(void *_offset, u64 *val)
2169 unsigned offset = (long)_offset;
2171 struct kvm_vcpu *vcpu;
2175 spin_lock(&kvm_lock);
2176 list_for_each_entry(kvm, &vm_list, vm_list)
2177 kvm_for_each_vcpu(i, vcpu, kvm)
2178 *val += *(u32 *)((void *)vcpu + offset);
2180 spin_unlock(&kvm_lock);
2184 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2186 static const struct file_operations *stat_fops[] = {
2187 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2188 [KVM_STAT_VM] = &vm_stat_fops,
2191 static void kvm_init_debug(void)
2193 struct kvm_stats_debugfs_item *p;
2195 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2196 for (p = debugfs_entries; p->name; ++p)
2197 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2198 (void *)(long)p->offset,
2199 stat_fops[p->kind]);
2202 static void kvm_exit_debug(void)
2204 struct kvm_stats_debugfs_item *p;
2206 for (p = debugfs_entries; p->name; ++p)
2207 debugfs_remove(p->dentry);
2208 debugfs_remove(kvm_debugfs_dir);
2211 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2213 if (kvm_usage_count)
2214 hardware_disable(NULL);
2218 static int kvm_resume(struct sys_device *dev)
2220 if (kvm_usage_count) {
2221 WARN_ON(spin_is_locked(&kvm_lock));
2222 hardware_enable(NULL);
2227 static struct sysdev_class kvm_sysdev_class = {
2229 .suspend = kvm_suspend,
2230 .resume = kvm_resume,
2233 static struct sys_device kvm_sysdev = {
2235 .cls = &kvm_sysdev_class,
2238 struct page *bad_page;
2242 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2244 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2247 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2249 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2251 kvm_arch_vcpu_load(vcpu, cpu);
2254 static void kvm_sched_out(struct preempt_notifier *pn,
2255 struct task_struct *next)
2257 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2259 kvm_arch_vcpu_put(vcpu);
2262 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2263 struct module *module)
2268 r = kvm_arch_init(opaque);
2272 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2274 if (bad_page == NULL) {
2279 bad_pfn = page_to_pfn(bad_page);
2281 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2283 if (hwpoison_page == NULL) {
2288 hwpoison_pfn = page_to_pfn(hwpoison_page);
2290 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2292 if (fault_page == NULL) {
2297 fault_pfn = page_to_pfn(fault_page);
2299 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2304 r = kvm_arch_hardware_setup();
2308 for_each_online_cpu(cpu) {
2309 smp_call_function_single(cpu,
2310 kvm_arch_check_processor_compat,
2316 r = register_cpu_notifier(&kvm_cpu_notifier);
2319 register_reboot_notifier(&kvm_reboot_notifier);
2321 r = sysdev_class_register(&kvm_sysdev_class);
2325 r = sysdev_register(&kvm_sysdev);
2329 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2331 vcpu_align = __alignof__(struct kvm_vcpu);
2332 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2334 if (!kvm_vcpu_cache) {
2339 kvm_chardev_ops.owner = module;
2340 kvm_vm_fops.owner = module;
2341 kvm_vcpu_fops.owner = module;
2343 r = misc_register(&kvm_dev);
2345 printk(KERN_ERR "kvm: misc device register failed\n");
2349 kvm_preempt_ops.sched_in = kvm_sched_in;
2350 kvm_preempt_ops.sched_out = kvm_sched_out;
2357 kmem_cache_destroy(kvm_vcpu_cache);
2359 sysdev_unregister(&kvm_sysdev);
2361 sysdev_class_unregister(&kvm_sysdev_class);
2363 unregister_reboot_notifier(&kvm_reboot_notifier);
2364 unregister_cpu_notifier(&kvm_cpu_notifier);
2367 kvm_arch_hardware_unsetup();
2369 free_cpumask_var(cpus_hardware_enabled);
2372 __free_page(fault_page);
2374 __free_page(hwpoison_page);
2375 __free_page(bad_page);
2381 EXPORT_SYMBOL_GPL(kvm_init);
2386 misc_deregister(&kvm_dev);
2387 kmem_cache_destroy(kvm_vcpu_cache);
2388 sysdev_unregister(&kvm_sysdev);
2389 sysdev_class_unregister(&kvm_sysdev_class);
2390 unregister_reboot_notifier(&kvm_reboot_notifier);
2391 unregister_cpu_notifier(&kvm_cpu_notifier);
2392 on_each_cpu(hardware_disable, NULL, 1);
2393 kvm_arch_hardware_unsetup();
2395 free_cpumask_var(cpus_hardware_enabled);
2396 __free_page(hwpoison_page);
2397 __free_page(bad_page);
2399 EXPORT_SYMBOL_GPL(kvm_exit);