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.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.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/intel-iommu.h>
46 #include <asm/processor.h>
48 #include <asm/uaccess.h>
49 #include <asm/pgtable.h>
51 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
52 #include "coalesced_mmio.h"
55 MODULE_AUTHOR("Qumranet");
56 MODULE_LICENSE("GPL");
58 DEFINE_SPINLOCK(kvm_lock);
61 static cpumask_t cpus_hardware_enabled;
63 struct kmem_cache *kvm_vcpu_cache;
64 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
66 static __read_mostly struct preempt_ops kvm_preempt_ops;
68 struct dentry *kvm_debugfs_dir;
70 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
75 static inline int valid_vcpu(int n)
77 return likely(n >= 0 && n < KVM_MAX_VCPUS);
80 inline int is_mmio_pfn(pfn_t pfn)
83 return PageReserved(pfn_to_page(pfn));
89 * Switches to specified vcpu, until a matching vcpu_put()
91 void vcpu_load(struct kvm_vcpu *vcpu)
95 mutex_lock(&vcpu->mutex);
97 preempt_notifier_register(&vcpu->preempt_notifier);
98 kvm_arch_vcpu_load(vcpu, cpu);
102 void vcpu_put(struct kvm_vcpu *vcpu)
105 kvm_arch_vcpu_put(vcpu);
106 preempt_notifier_unregister(&vcpu->preempt_notifier);
108 mutex_unlock(&vcpu->mutex);
111 static void ack_flush(void *_completed)
115 void kvm_flush_remote_tlbs(struct kvm *kvm)
119 struct kvm_vcpu *vcpu;
123 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
124 vcpu = kvm->vcpus[i];
127 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
130 if (cpu != -1 && cpu != me)
133 if (cpus_empty(cpus))
135 ++kvm->stat.remote_tlb_flush;
136 smp_call_function_mask(cpus, ack_flush, NULL, 1);
141 void kvm_reload_remote_mmus(struct kvm *kvm)
145 struct kvm_vcpu *vcpu;
149 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
150 vcpu = kvm->vcpus[i];
153 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
156 if (cpu != -1 && cpu != me)
159 if (cpus_empty(cpus))
161 smp_call_function_mask(cpus, ack_flush, NULL, 1);
167 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
172 mutex_init(&vcpu->mutex);
176 init_waitqueue_head(&vcpu->wq);
178 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
183 vcpu->run = page_address(page);
185 r = kvm_arch_vcpu_init(vcpu);
191 free_page((unsigned long)vcpu->run);
195 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
197 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
199 kvm_arch_vcpu_uninit(vcpu);
200 free_page((unsigned long)vcpu->run);
202 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
204 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
205 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
207 return container_of(mn, struct kvm, mmu_notifier);
210 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
211 struct mm_struct *mm,
212 unsigned long address)
214 struct kvm *kvm = mmu_notifier_to_kvm(mn);
218 * When ->invalidate_page runs, the linux pte has been zapped
219 * already but the page is still allocated until
220 * ->invalidate_page returns. So if we increase the sequence
221 * here the kvm page fault will notice if the spte can't be
222 * established because the page is going to be freed. If
223 * instead the kvm page fault establishes the spte before
224 * ->invalidate_page runs, kvm_unmap_hva will release it
227 * The sequence increase only need to be seen at spin_unlock
228 * time, and not at spin_lock time.
230 * Increasing the sequence after the spin_unlock would be
231 * unsafe because the kvm page fault could then establish the
232 * pte after kvm_unmap_hva returned, without noticing the page
233 * is going to be freed.
235 spin_lock(&kvm->mmu_lock);
236 kvm->mmu_notifier_seq++;
237 need_tlb_flush = kvm_unmap_hva(kvm, address);
238 spin_unlock(&kvm->mmu_lock);
240 /* we've to flush the tlb before the pages can be freed */
242 kvm_flush_remote_tlbs(kvm);
246 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
247 struct mm_struct *mm,
251 struct kvm *kvm = mmu_notifier_to_kvm(mn);
252 int need_tlb_flush = 0;
254 spin_lock(&kvm->mmu_lock);
256 * The count increase must become visible at unlock time as no
257 * spte can be established without taking the mmu_lock and
258 * count is also read inside the mmu_lock critical section.
260 kvm->mmu_notifier_count++;
261 for (; start < end; start += PAGE_SIZE)
262 need_tlb_flush |= kvm_unmap_hva(kvm, start);
263 spin_unlock(&kvm->mmu_lock);
265 /* we've to flush the tlb before the pages can be freed */
267 kvm_flush_remote_tlbs(kvm);
270 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
271 struct mm_struct *mm,
275 struct kvm *kvm = mmu_notifier_to_kvm(mn);
277 spin_lock(&kvm->mmu_lock);
279 * This sequence increase will notify the kvm page fault that
280 * the page that is going to be mapped in the spte could have
283 kvm->mmu_notifier_seq++;
285 * The above sequence increase must be visible before the
286 * below count decrease but both values are read by the kvm
287 * page fault under mmu_lock spinlock so we don't need to add
288 * a smb_wmb() here in between the two.
290 kvm->mmu_notifier_count--;
291 spin_unlock(&kvm->mmu_lock);
293 BUG_ON(kvm->mmu_notifier_count < 0);
296 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
297 struct mm_struct *mm,
298 unsigned long address)
300 struct kvm *kvm = mmu_notifier_to_kvm(mn);
303 spin_lock(&kvm->mmu_lock);
304 young = kvm_age_hva(kvm, address);
305 spin_unlock(&kvm->mmu_lock);
308 kvm_flush_remote_tlbs(kvm);
313 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
314 .invalidate_page = kvm_mmu_notifier_invalidate_page,
315 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
316 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
317 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
319 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
321 static struct kvm *kvm_create_vm(void)
323 struct kvm *kvm = kvm_arch_create_vm();
324 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
331 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
332 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
335 return ERR_PTR(-ENOMEM);
337 kvm->coalesced_mmio_ring =
338 (struct kvm_coalesced_mmio_ring *)page_address(page);
341 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
344 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
345 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
347 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
356 kvm->mm = current->mm;
357 atomic_inc(&kvm->mm->mm_count);
358 spin_lock_init(&kvm->mmu_lock);
359 kvm_io_bus_init(&kvm->pio_bus);
360 mutex_init(&kvm->lock);
361 kvm_io_bus_init(&kvm->mmio_bus);
362 init_rwsem(&kvm->slots_lock);
363 atomic_set(&kvm->users_count, 1);
364 spin_lock(&kvm_lock);
365 list_add(&kvm->vm_list, &vm_list);
366 spin_unlock(&kvm_lock);
367 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
368 kvm_coalesced_mmio_init(kvm);
375 * Free any memory in @free but not in @dont.
377 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
378 struct kvm_memory_slot *dont)
380 if (!dont || free->rmap != dont->rmap)
383 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
384 vfree(free->dirty_bitmap);
386 if (!dont || free->lpage_info != dont->lpage_info)
387 vfree(free->lpage_info);
390 free->dirty_bitmap = NULL;
392 free->lpage_info = NULL;
395 void kvm_free_physmem(struct kvm *kvm)
399 for (i = 0; i < kvm->nmemslots; ++i)
400 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
403 static void kvm_destroy_vm(struct kvm *kvm)
405 struct mm_struct *mm = kvm->mm;
407 spin_lock(&kvm_lock);
408 list_del(&kvm->vm_list);
409 spin_unlock(&kvm_lock);
410 kvm_io_bus_destroy(&kvm->pio_bus);
411 kvm_io_bus_destroy(&kvm->mmio_bus);
412 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
413 if (kvm->coalesced_mmio_ring != NULL)
414 free_page((unsigned long)kvm->coalesced_mmio_ring);
416 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
417 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
419 kvm_arch_destroy_vm(kvm);
423 void kvm_get_kvm(struct kvm *kvm)
425 atomic_inc(&kvm->users_count);
427 EXPORT_SYMBOL_GPL(kvm_get_kvm);
429 void kvm_put_kvm(struct kvm *kvm)
431 if (atomic_dec_and_test(&kvm->users_count))
434 EXPORT_SYMBOL_GPL(kvm_put_kvm);
437 static int kvm_vm_release(struct inode *inode, struct file *filp)
439 struct kvm *kvm = filp->private_data;
446 * Allocate some memory and give it an address in the guest physical address
449 * Discontiguous memory is allowed, mostly for framebuffers.
451 * Must be called holding mmap_sem for write.
453 int __kvm_set_memory_region(struct kvm *kvm,
454 struct kvm_userspace_memory_region *mem,
459 unsigned long npages;
461 struct kvm_memory_slot *memslot;
462 struct kvm_memory_slot old, new;
465 /* General sanity checks */
466 if (mem->memory_size & (PAGE_SIZE - 1))
468 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
470 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
472 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
475 memslot = &kvm->memslots[mem->slot];
476 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
477 npages = mem->memory_size >> PAGE_SHIFT;
480 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
482 new = old = *memslot;
484 new.base_gfn = base_gfn;
486 new.flags = mem->flags;
488 /* Disallow changing a memory slot's size. */
490 if (npages && old.npages && npages != old.npages)
493 /* Check for overlaps */
495 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
496 struct kvm_memory_slot *s = &kvm->memslots[i];
500 if (!((base_gfn + npages <= s->base_gfn) ||
501 (base_gfn >= s->base_gfn + s->npages)))
505 /* Free page dirty bitmap if unneeded */
506 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
507 new.dirty_bitmap = NULL;
511 /* Allocate if a slot is being created */
513 if (npages && !new.rmap) {
514 new.rmap = vmalloc(npages * sizeof(struct page *));
519 memset(new.rmap, 0, npages * sizeof(*new.rmap));
521 new.user_alloc = user_alloc;
523 * hva_to_rmmap() serialzies with the mmu_lock and to be
524 * safe it has to ignore memslots with !user_alloc &&
528 new.userspace_addr = mem->userspace_addr;
530 new.userspace_addr = 0;
532 if (npages && !new.lpage_info) {
533 int largepages = npages / KVM_PAGES_PER_HPAGE;
534 if (npages % KVM_PAGES_PER_HPAGE)
536 if (base_gfn % KVM_PAGES_PER_HPAGE)
539 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
544 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
546 if (base_gfn % KVM_PAGES_PER_HPAGE)
547 new.lpage_info[0].write_count = 1;
548 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
549 new.lpage_info[largepages-1].write_count = 1;
552 /* Allocate page dirty bitmap if needed */
553 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
554 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
556 new.dirty_bitmap = vmalloc(dirty_bytes);
557 if (!new.dirty_bitmap)
559 memset(new.dirty_bitmap, 0, dirty_bytes);
561 #endif /* not defined CONFIG_S390 */
564 kvm_arch_flush_shadow(kvm);
566 spin_lock(&kvm->mmu_lock);
567 if (mem->slot >= kvm->nmemslots)
568 kvm->nmemslots = mem->slot + 1;
571 spin_unlock(&kvm->mmu_lock);
573 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
575 spin_lock(&kvm->mmu_lock);
577 spin_unlock(&kvm->mmu_lock);
581 kvm_free_physmem_slot(&old, &new);
583 /* map the pages in iommu page table */
584 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
591 kvm_free_physmem_slot(&new, &old);
596 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
598 int kvm_set_memory_region(struct kvm *kvm,
599 struct kvm_userspace_memory_region *mem,
604 down_write(&kvm->slots_lock);
605 r = __kvm_set_memory_region(kvm, mem, user_alloc);
606 up_write(&kvm->slots_lock);
609 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
611 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
613 kvm_userspace_memory_region *mem,
616 if (mem->slot >= KVM_MEMORY_SLOTS)
618 return kvm_set_memory_region(kvm, mem, user_alloc);
621 int kvm_get_dirty_log(struct kvm *kvm,
622 struct kvm_dirty_log *log, int *is_dirty)
624 struct kvm_memory_slot *memslot;
627 unsigned long any = 0;
630 if (log->slot >= KVM_MEMORY_SLOTS)
633 memslot = &kvm->memslots[log->slot];
635 if (!memslot->dirty_bitmap)
638 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
640 for (i = 0; !any && i < n/sizeof(long); ++i)
641 any = memslot->dirty_bitmap[i];
644 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
655 int is_error_page(struct page *page)
657 return page == bad_page;
659 EXPORT_SYMBOL_GPL(is_error_page);
661 int is_error_pfn(pfn_t pfn)
663 return pfn == bad_pfn;
665 EXPORT_SYMBOL_GPL(is_error_pfn);
667 static inline unsigned long bad_hva(void)
672 int kvm_is_error_hva(unsigned long addr)
674 return addr == bad_hva();
676 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
678 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
682 for (i = 0; i < kvm->nmemslots; ++i) {
683 struct kvm_memory_slot *memslot = &kvm->memslots[i];
685 if (gfn >= memslot->base_gfn
686 && gfn < memslot->base_gfn + memslot->npages)
692 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
694 gfn = unalias_gfn(kvm, gfn);
695 return __gfn_to_memslot(kvm, gfn);
698 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
702 gfn = unalias_gfn(kvm, gfn);
703 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
704 struct kvm_memory_slot *memslot = &kvm->memslots[i];
706 if (gfn >= memslot->base_gfn
707 && gfn < memslot->base_gfn + memslot->npages)
712 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
714 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
716 struct kvm_memory_slot *slot;
718 gfn = unalias_gfn(kvm, gfn);
719 slot = __gfn_to_memslot(kvm, gfn);
722 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
724 EXPORT_SYMBOL_GPL(gfn_to_hva);
726 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
728 struct page *page[1];
735 addr = gfn_to_hva(kvm, gfn);
736 if (kvm_is_error_hva(addr)) {
738 return page_to_pfn(bad_page);
741 npages = get_user_pages_fast(addr, 1, 1, page);
743 if (unlikely(npages != 1)) {
744 struct vm_area_struct *vma;
746 down_read(¤t->mm->mmap_sem);
747 vma = find_vma(current->mm, addr);
749 if (vma == NULL || addr < vma->vm_start ||
750 !(vma->vm_flags & VM_PFNMAP)) {
751 up_read(¤t->mm->mmap_sem);
753 return page_to_pfn(bad_page);
756 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
757 up_read(¤t->mm->mmap_sem);
758 BUG_ON(!is_mmio_pfn(pfn));
760 pfn = page_to_pfn(page[0]);
765 EXPORT_SYMBOL_GPL(gfn_to_pfn);
767 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
771 pfn = gfn_to_pfn(kvm, gfn);
772 if (!is_mmio_pfn(pfn))
773 return pfn_to_page(pfn);
775 WARN_ON(is_mmio_pfn(pfn));
781 EXPORT_SYMBOL_GPL(gfn_to_page);
783 void kvm_release_page_clean(struct page *page)
785 kvm_release_pfn_clean(page_to_pfn(page));
787 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
789 void kvm_release_pfn_clean(pfn_t pfn)
791 if (!is_mmio_pfn(pfn))
792 put_page(pfn_to_page(pfn));
794 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
796 void kvm_release_page_dirty(struct page *page)
798 kvm_release_pfn_dirty(page_to_pfn(page));
800 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
802 void kvm_release_pfn_dirty(pfn_t pfn)
804 kvm_set_pfn_dirty(pfn);
805 kvm_release_pfn_clean(pfn);
807 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
809 void kvm_set_page_dirty(struct page *page)
811 kvm_set_pfn_dirty(page_to_pfn(page));
813 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
815 void kvm_set_pfn_dirty(pfn_t pfn)
817 if (!is_mmio_pfn(pfn)) {
818 struct page *page = pfn_to_page(pfn);
819 if (!PageReserved(page))
823 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
825 void kvm_set_pfn_accessed(pfn_t pfn)
827 if (!is_mmio_pfn(pfn))
828 mark_page_accessed(pfn_to_page(pfn));
830 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
832 void kvm_get_pfn(pfn_t pfn)
834 if (!is_mmio_pfn(pfn))
835 get_page(pfn_to_page(pfn));
837 EXPORT_SYMBOL_GPL(kvm_get_pfn);
839 static int next_segment(unsigned long len, int offset)
841 if (len > PAGE_SIZE - offset)
842 return PAGE_SIZE - offset;
847 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
853 addr = gfn_to_hva(kvm, gfn);
854 if (kvm_is_error_hva(addr))
856 r = copy_from_user(data, (void __user *)addr + offset, len);
861 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
863 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
865 gfn_t gfn = gpa >> PAGE_SHIFT;
867 int offset = offset_in_page(gpa);
870 while ((seg = next_segment(len, offset)) != 0) {
871 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
881 EXPORT_SYMBOL_GPL(kvm_read_guest);
883 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
888 gfn_t gfn = gpa >> PAGE_SHIFT;
889 int offset = offset_in_page(gpa);
891 addr = gfn_to_hva(kvm, gfn);
892 if (kvm_is_error_hva(addr))
895 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
901 EXPORT_SYMBOL(kvm_read_guest_atomic);
903 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
909 addr = gfn_to_hva(kvm, gfn);
910 if (kvm_is_error_hva(addr))
912 r = copy_to_user((void __user *)addr + offset, data, len);
915 mark_page_dirty(kvm, gfn);
918 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
920 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
923 gfn_t gfn = gpa >> PAGE_SHIFT;
925 int offset = offset_in_page(gpa);
928 while ((seg = next_segment(len, offset)) != 0) {
929 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
940 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
942 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
944 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
946 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
948 gfn_t gfn = gpa >> PAGE_SHIFT;
950 int offset = offset_in_page(gpa);
953 while ((seg = next_segment(len, offset)) != 0) {
954 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
963 EXPORT_SYMBOL_GPL(kvm_clear_guest);
965 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
967 struct kvm_memory_slot *memslot;
969 gfn = unalias_gfn(kvm, gfn);
970 memslot = __gfn_to_memslot(kvm, gfn);
971 if (memslot && memslot->dirty_bitmap) {
972 unsigned long rel_gfn = gfn - memslot->base_gfn;
975 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
976 set_bit(rel_gfn, memslot->dirty_bitmap);
981 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
983 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
988 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
990 if (kvm_cpu_has_interrupt(vcpu) ||
991 kvm_cpu_has_pending_timer(vcpu) ||
992 kvm_arch_vcpu_runnable(vcpu)) {
993 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
996 if (signal_pending(current))
1004 finish_wait(&vcpu->wq, &wait);
1007 void kvm_resched(struct kvm_vcpu *vcpu)
1009 if (!need_resched())
1013 EXPORT_SYMBOL_GPL(kvm_resched);
1015 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1017 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1020 if (vmf->pgoff == 0)
1021 page = virt_to_page(vcpu->run);
1023 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1024 page = virt_to_page(vcpu->arch.pio_data);
1026 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1027 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1028 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1031 return VM_FAULT_SIGBUS;
1037 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1038 .fault = kvm_vcpu_fault,
1041 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1043 vma->vm_ops = &kvm_vcpu_vm_ops;
1047 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1049 struct kvm_vcpu *vcpu = filp->private_data;
1051 kvm_put_kvm(vcpu->kvm);
1055 static const struct file_operations kvm_vcpu_fops = {
1056 .release = kvm_vcpu_release,
1057 .unlocked_ioctl = kvm_vcpu_ioctl,
1058 .compat_ioctl = kvm_vcpu_ioctl,
1059 .mmap = kvm_vcpu_mmap,
1063 * Allocates an inode for the vcpu.
1065 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1067 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1069 kvm_put_kvm(vcpu->kvm);
1074 * Creates some virtual cpus. Good luck creating more than one.
1076 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1079 struct kvm_vcpu *vcpu;
1084 vcpu = kvm_arch_vcpu_create(kvm, n);
1086 return PTR_ERR(vcpu);
1088 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1090 r = kvm_arch_vcpu_setup(vcpu);
1094 mutex_lock(&kvm->lock);
1095 if (kvm->vcpus[n]) {
1099 kvm->vcpus[n] = vcpu;
1100 mutex_unlock(&kvm->lock);
1102 /* Now it's all set up, let userspace reach it */
1104 r = create_vcpu_fd(vcpu);
1110 mutex_lock(&kvm->lock);
1111 kvm->vcpus[n] = NULL;
1113 mutex_unlock(&kvm->lock);
1114 kvm_arch_vcpu_destroy(vcpu);
1118 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1121 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1122 vcpu->sigset_active = 1;
1123 vcpu->sigset = *sigset;
1125 vcpu->sigset_active = 0;
1129 static long kvm_vcpu_ioctl(struct file *filp,
1130 unsigned int ioctl, unsigned long arg)
1132 struct kvm_vcpu *vcpu = filp->private_data;
1133 void __user *argp = (void __user *)arg;
1135 struct kvm_fpu *fpu = NULL;
1136 struct kvm_sregs *kvm_sregs = NULL;
1138 if (vcpu->kvm->mm != current->mm)
1145 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1147 case KVM_GET_REGS: {
1148 struct kvm_regs *kvm_regs;
1151 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1154 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1158 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1165 case KVM_SET_REGS: {
1166 struct kvm_regs *kvm_regs;
1169 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1173 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1175 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1183 case KVM_GET_SREGS: {
1184 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1188 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1192 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1197 case KVM_SET_SREGS: {
1198 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1203 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1205 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1211 case KVM_GET_MP_STATE: {
1212 struct kvm_mp_state mp_state;
1214 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1218 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1223 case KVM_SET_MP_STATE: {
1224 struct kvm_mp_state mp_state;
1227 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1229 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1235 case KVM_TRANSLATE: {
1236 struct kvm_translation tr;
1239 if (copy_from_user(&tr, argp, sizeof tr))
1241 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1245 if (copy_to_user(argp, &tr, sizeof tr))
1250 case KVM_DEBUG_GUEST: {
1251 struct kvm_debug_guest dbg;
1254 if (copy_from_user(&dbg, argp, sizeof dbg))
1256 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1262 case KVM_SET_SIGNAL_MASK: {
1263 struct kvm_signal_mask __user *sigmask_arg = argp;
1264 struct kvm_signal_mask kvm_sigmask;
1265 sigset_t sigset, *p;
1270 if (copy_from_user(&kvm_sigmask, argp,
1271 sizeof kvm_sigmask))
1274 if (kvm_sigmask.len != sizeof sigset)
1277 if (copy_from_user(&sigset, sigmask_arg->sigset,
1282 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1286 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1290 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1294 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1300 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1305 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1307 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1314 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1322 static long kvm_vm_ioctl(struct file *filp,
1323 unsigned int ioctl, unsigned long arg)
1325 struct kvm *kvm = filp->private_data;
1326 void __user *argp = (void __user *)arg;
1329 if (kvm->mm != current->mm)
1332 case KVM_CREATE_VCPU:
1333 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1337 case KVM_SET_USER_MEMORY_REGION: {
1338 struct kvm_userspace_memory_region kvm_userspace_mem;
1341 if (copy_from_user(&kvm_userspace_mem, argp,
1342 sizeof kvm_userspace_mem))
1345 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1350 case KVM_GET_DIRTY_LOG: {
1351 struct kvm_dirty_log log;
1354 if (copy_from_user(&log, argp, sizeof log))
1356 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1361 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1362 case KVM_REGISTER_COALESCED_MMIO: {
1363 struct kvm_coalesced_mmio_zone zone;
1365 if (copy_from_user(&zone, argp, sizeof zone))
1368 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1374 case KVM_UNREGISTER_COALESCED_MMIO: {
1375 struct kvm_coalesced_mmio_zone zone;
1377 if (copy_from_user(&zone, argp, sizeof zone))
1380 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1388 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1394 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1396 struct page *page[1];
1399 gfn_t gfn = vmf->pgoff;
1400 struct kvm *kvm = vma->vm_file->private_data;
1402 addr = gfn_to_hva(kvm, gfn);
1403 if (kvm_is_error_hva(addr))
1404 return VM_FAULT_SIGBUS;
1406 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1408 if (unlikely(npages != 1))
1409 return VM_FAULT_SIGBUS;
1411 vmf->page = page[0];
1415 static struct vm_operations_struct kvm_vm_vm_ops = {
1416 .fault = kvm_vm_fault,
1419 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1421 vma->vm_ops = &kvm_vm_vm_ops;
1425 static const struct file_operations kvm_vm_fops = {
1426 .release = kvm_vm_release,
1427 .unlocked_ioctl = kvm_vm_ioctl,
1428 .compat_ioctl = kvm_vm_ioctl,
1429 .mmap = kvm_vm_mmap,
1432 static int kvm_dev_ioctl_create_vm(void)
1437 kvm = kvm_create_vm();
1439 return PTR_ERR(kvm);
1440 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1447 static long kvm_dev_ioctl(struct file *filp,
1448 unsigned int ioctl, unsigned long arg)
1453 case KVM_GET_API_VERSION:
1457 r = KVM_API_VERSION;
1463 r = kvm_dev_ioctl_create_vm();
1465 case KVM_CHECK_EXTENSION:
1466 r = kvm_dev_ioctl_check_extension(arg);
1468 case KVM_GET_VCPU_MMAP_SIZE:
1472 r = PAGE_SIZE; /* struct kvm_run */
1474 r += PAGE_SIZE; /* pio data page */
1476 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1477 r += PAGE_SIZE; /* coalesced mmio ring page */
1480 case KVM_TRACE_ENABLE:
1481 case KVM_TRACE_PAUSE:
1482 case KVM_TRACE_DISABLE:
1483 r = kvm_trace_ioctl(ioctl, arg);
1486 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1492 static struct file_operations kvm_chardev_ops = {
1493 .unlocked_ioctl = kvm_dev_ioctl,
1494 .compat_ioctl = kvm_dev_ioctl,
1497 static struct miscdevice kvm_dev = {
1503 static void hardware_enable(void *junk)
1505 int cpu = raw_smp_processor_id();
1507 if (cpu_isset(cpu, cpus_hardware_enabled))
1509 cpu_set(cpu, cpus_hardware_enabled);
1510 kvm_arch_hardware_enable(NULL);
1513 static void hardware_disable(void *junk)
1515 int cpu = raw_smp_processor_id();
1517 if (!cpu_isset(cpu, cpus_hardware_enabled))
1519 cpu_clear(cpu, cpus_hardware_enabled);
1520 kvm_arch_hardware_disable(NULL);
1523 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1528 val &= ~CPU_TASKS_FROZEN;
1531 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1533 hardware_disable(NULL);
1535 case CPU_UP_CANCELED:
1536 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1538 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1541 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1543 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1550 asmlinkage void kvm_handle_fault_on_reboot(void)
1553 /* spin while reset goes on */
1556 /* Fault while not rebooting. We want the trace. */
1559 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1561 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1564 if (val == SYS_RESTART) {
1566 * Some (well, at least mine) BIOSes hang on reboot if
1569 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1570 kvm_rebooting = true;
1571 on_each_cpu(hardware_disable, NULL, 1);
1576 static struct notifier_block kvm_reboot_notifier = {
1577 .notifier_call = kvm_reboot,
1581 void kvm_io_bus_init(struct kvm_io_bus *bus)
1583 memset(bus, 0, sizeof(*bus));
1586 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1590 for (i = 0; i < bus->dev_count; i++) {
1591 struct kvm_io_device *pos = bus->devs[i];
1593 kvm_iodevice_destructor(pos);
1597 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1598 gpa_t addr, int len, int is_write)
1602 for (i = 0; i < bus->dev_count; i++) {
1603 struct kvm_io_device *pos = bus->devs[i];
1605 if (pos->in_range(pos, addr, len, is_write))
1612 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1614 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1616 bus->devs[bus->dev_count++] = dev;
1619 static struct notifier_block kvm_cpu_notifier = {
1620 .notifier_call = kvm_cpu_hotplug,
1621 .priority = 20, /* must be > scheduler priority */
1624 static int vm_stat_get(void *_offset, u64 *val)
1626 unsigned offset = (long)_offset;
1630 spin_lock(&kvm_lock);
1631 list_for_each_entry(kvm, &vm_list, vm_list)
1632 *val += *(u32 *)((void *)kvm + offset);
1633 spin_unlock(&kvm_lock);
1637 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1639 static int vcpu_stat_get(void *_offset, u64 *val)
1641 unsigned offset = (long)_offset;
1643 struct kvm_vcpu *vcpu;
1647 spin_lock(&kvm_lock);
1648 list_for_each_entry(kvm, &vm_list, vm_list)
1649 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1650 vcpu = kvm->vcpus[i];
1652 *val += *(u32 *)((void *)vcpu + offset);
1654 spin_unlock(&kvm_lock);
1658 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1660 static struct file_operations *stat_fops[] = {
1661 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1662 [KVM_STAT_VM] = &vm_stat_fops,
1665 static void kvm_init_debug(void)
1667 struct kvm_stats_debugfs_item *p;
1669 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1670 for (p = debugfs_entries; p->name; ++p)
1671 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1672 (void *)(long)p->offset,
1673 stat_fops[p->kind]);
1676 static void kvm_exit_debug(void)
1678 struct kvm_stats_debugfs_item *p;
1680 for (p = debugfs_entries; p->name; ++p)
1681 debugfs_remove(p->dentry);
1682 debugfs_remove(kvm_debugfs_dir);
1685 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1687 hardware_disable(NULL);
1691 static int kvm_resume(struct sys_device *dev)
1693 hardware_enable(NULL);
1697 static struct sysdev_class kvm_sysdev_class = {
1699 .suspend = kvm_suspend,
1700 .resume = kvm_resume,
1703 static struct sys_device kvm_sysdev = {
1705 .cls = &kvm_sysdev_class,
1708 struct page *bad_page;
1712 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1714 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1717 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1719 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1721 kvm_arch_vcpu_load(vcpu, cpu);
1724 static void kvm_sched_out(struct preempt_notifier *pn,
1725 struct task_struct *next)
1727 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1729 kvm_arch_vcpu_put(vcpu);
1732 int kvm_init(void *opaque, unsigned int vcpu_size,
1733 struct module *module)
1740 r = kvm_arch_init(opaque);
1744 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1746 if (bad_page == NULL) {
1751 bad_pfn = page_to_pfn(bad_page);
1753 r = kvm_arch_hardware_setup();
1757 for_each_online_cpu(cpu) {
1758 smp_call_function_single(cpu,
1759 kvm_arch_check_processor_compat,
1765 on_each_cpu(hardware_enable, NULL, 1);
1766 r = register_cpu_notifier(&kvm_cpu_notifier);
1769 register_reboot_notifier(&kvm_reboot_notifier);
1771 r = sysdev_class_register(&kvm_sysdev_class);
1775 r = sysdev_register(&kvm_sysdev);
1779 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1780 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1781 __alignof__(struct kvm_vcpu),
1783 if (!kvm_vcpu_cache) {
1788 kvm_chardev_ops.owner = module;
1790 r = misc_register(&kvm_dev);
1792 printk(KERN_ERR "kvm: misc device register failed\n");
1796 kvm_preempt_ops.sched_in = kvm_sched_in;
1797 kvm_preempt_ops.sched_out = kvm_sched_out;
1802 kmem_cache_destroy(kvm_vcpu_cache);
1804 sysdev_unregister(&kvm_sysdev);
1806 sysdev_class_unregister(&kvm_sysdev_class);
1808 unregister_reboot_notifier(&kvm_reboot_notifier);
1809 unregister_cpu_notifier(&kvm_cpu_notifier);
1811 on_each_cpu(hardware_disable, NULL, 1);
1813 kvm_arch_hardware_unsetup();
1815 __free_page(bad_page);
1822 EXPORT_SYMBOL_GPL(kvm_init);
1826 kvm_trace_cleanup();
1827 misc_deregister(&kvm_dev);
1828 kmem_cache_destroy(kvm_vcpu_cache);
1829 sysdev_unregister(&kvm_sysdev);
1830 sysdev_class_unregister(&kvm_sysdev_class);
1831 unregister_reboot_notifier(&kvm_reboot_notifier);
1832 unregister_cpu_notifier(&kvm_cpu_notifier);
1833 on_each_cpu(hardware_disable, NULL, 1);
1834 kvm_arch_hardware_unsetup();
1837 __free_page(bad_page);
1839 EXPORT_SYMBOL_GPL(kvm_exit);