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>
45 #include <asm/processor.h>
47 #include <asm/uaccess.h>
48 #include <asm/pgtable.h>
50 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
51 #include "coalesced_mmio.h"
54 MODULE_AUTHOR("Qumranet");
55 MODULE_LICENSE("GPL");
57 DEFINE_SPINLOCK(kvm_lock);
60 static cpumask_t cpus_hardware_enabled;
62 struct kmem_cache *kvm_vcpu_cache;
63 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
65 static __read_mostly struct preempt_ops kvm_preempt_ops;
67 struct dentry *kvm_debugfs_dir;
69 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
74 static inline int valid_vcpu(int n)
76 return likely(n >= 0 && n < KVM_MAX_VCPUS);
79 static inline int is_mmio_pfn(pfn_t pfn)
82 return PageReserved(pfn_to_page(pfn));
88 * Switches to specified vcpu, until a matching vcpu_put()
90 void vcpu_load(struct kvm_vcpu *vcpu)
94 mutex_lock(&vcpu->mutex);
96 preempt_notifier_register(&vcpu->preempt_notifier);
97 kvm_arch_vcpu_load(vcpu, cpu);
101 void vcpu_put(struct kvm_vcpu *vcpu)
104 kvm_arch_vcpu_put(vcpu);
105 preempt_notifier_unregister(&vcpu->preempt_notifier);
107 mutex_unlock(&vcpu->mutex);
110 static void ack_flush(void *_completed)
114 void kvm_flush_remote_tlbs(struct kvm *kvm)
118 struct kvm_vcpu *vcpu;
122 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
123 vcpu = kvm->vcpus[i];
126 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
129 if (cpu != -1 && cpu != me)
132 if (cpus_empty(cpus))
134 ++kvm->stat.remote_tlb_flush;
135 smp_call_function_mask(cpus, ack_flush, NULL, 1);
140 void kvm_reload_remote_mmus(struct kvm *kvm)
144 struct kvm_vcpu *vcpu;
148 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
149 vcpu = kvm->vcpus[i];
152 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
155 if (cpu != -1 && cpu != me)
158 if (cpus_empty(cpus))
160 smp_call_function_mask(cpus, ack_flush, NULL, 1);
166 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
171 mutex_init(&vcpu->mutex);
175 init_waitqueue_head(&vcpu->wq);
177 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
182 vcpu->run = page_address(page);
184 r = kvm_arch_vcpu_init(vcpu);
190 free_page((unsigned long)vcpu->run);
194 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
196 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
198 kvm_arch_vcpu_uninit(vcpu);
199 free_page((unsigned long)vcpu->run);
201 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
203 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
204 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
206 return container_of(mn, struct kvm, mmu_notifier);
209 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
210 struct mm_struct *mm,
211 unsigned long address)
213 struct kvm *kvm = mmu_notifier_to_kvm(mn);
217 * When ->invalidate_page runs, the linux pte has been zapped
218 * already but the page is still allocated until
219 * ->invalidate_page returns. So if we increase the sequence
220 * here the kvm page fault will notice if the spte can't be
221 * established because the page is going to be freed. If
222 * instead the kvm page fault establishes the spte before
223 * ->invalidate_page runs, kvm_unmap_hva will release it
226 * The sequence increase only need to be seen at spin_unlock
227 * time, and not at spin_lock time.
229 * Increasing the sequence after the spin_unlock would be
230 * unsafe because the kvm page fault could then establish the
231 * pte after kvm_unmap_hva returned, without noticing the page
232 * is going to be freed.
234 spin_lock(&kvm->mmu_lock);
235 kvm->mmu_notifier_seq++;
236 need_tlb_flush = kvm_unmap_hva(kvm, address);
237 spin_unlock(&kvm->mmu_lock);
239 /* we've to flush the tlb before the pages can be freed */
241 kvm_flush_remote_tlbs(kvm);
245 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
246 struct mm_struct *mm,
250 struct kvm *kvm = mmu_notifier_to_kvm(mn);
251 int need_tlb_flush = 0;
253 spin_lock(&kvm->mmu_lock);
255 * The count increase must become visible at unlock time as no
256 * spte can be established without taking the mmu_lock and
257 * count is also read inside the mmu_lock critical section.
259 kvm->mmu_notifier_count++;
260 for (; start < end; start += PAGE_SIZE)
261 need_tlb_flush |= kvm_unmap_hva(kvm, start);
262 spin_unlock(&kvm->mmu_lock);
264 /* we've to flush the tlb before the pages can be freed */
266 kvm_flush_remote_tlbs(kvm);
269 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
270 struct mm_struct *mm,
274 struct kvm *kvm = mmu_notifier_to_kvm(mn);
276 spin_lock(&kvm->mmu_lock);
278 * This sequence increase will notify the kvm page fault that
279 * the page that is going to be mapped in the spte could have
282 kvm->mmu_notifier_seq++;
284 * The above sequence increase must be visible before the
285 * below count decrease but both values are read by the kvm
286 * page fault under mmu_lock spinlock so we don't need to add
287 * a smb_wmb() here in between the two.
289 kvm->mmu_notifier_count--;
290 spin_unlock(&kvm->mmu_lock);
292 BUG_ON(kvm->mmu_notifier_count < 0);
295 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
296 struct mm_struct *mm,
297 unsigned long address)
299 struct kvm *kvm = mmu_notifier_to_kvm(mn);
302 spin_lock(&kvm->mmu_lock);
303 young = kvm_age_hva(kvm, address);
304 spin_unlock(&kvm->mmu_lock);
307 kvm_flush_remote_tlbs(kvm);
312 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
313 .invalidate_page = kvm_mmu_notifier_invalidate_page,
314 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
315 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
316 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
318 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
320 static struct kvm *kvm_create_vm(void)
322 struct kvm *kvm = kvm_arch_create_vm();
323 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
330 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
331 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
334 return ERR_PTR(-ENOMEM);
336 kvm->coalesced_mmio_ring =
337 (struct kvm_coalesced_mmio_ring *)page_address(page);
340 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
343 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
344 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
346 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
355 kvm->mm = current->mm;
356 atomic_inc(&kvm->mm->mm_count);
357 spin_lock_init(&kvm->mmu_lock);
358 kvm_io_bus_init(&kvm->pio_bus);
359 mutex_init(&kvm->lock);
360 kvm_io_bus_init(&kvm->mmio_bus);
361 init_rwsem(&kvm->slots_lock);
362 atomic_set(&kvm->users_count, 1);
363 spin_lock(&kvm_lock);
364 list_add(&kvm->vm_list, &vm_list);
365 spin_unlock(&kvm_lock);
366 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
367 kvm_coalesced_mmio_init(kvm);
374 * Free any memory in @free but not in @dont.
376 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
377 struct kvm_memory_slot *dont)
379 if (!dont || free->rmap != dont->rmap)
382 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
383 vfree(free->dirty_bitmap);
385 if (!dont || free->lpage_info != dont->lpage_info)
386 vfree(free->lpage_info);
389 free->dirty_bitmap = NULL;
391 free->lpage_info = NULL;
394 void kvm_free_physmem(struct kvm *kvm)
398 for (i = 0; i < kvm->nmemslots; ++i)
399 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
402 static void kvm_destroy_vm(struct kvm *kvm)
404 struct mm_struct *mm = kvm->mm;
406 spin_lock(&kvm_lock);
407 list_del(&kvm->vm_list);
408 spin_unlock(&kvm_lock);
409 kvm_io_bus_destroy(&kvm->pio_bus);
410 kvm_io_bus_destroy(&kvm->mmio_bus);
411 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
412 if (kvm->coalesced_mmio_ring != NULL)
413 free_page((unsigned long)kvm->coalesced_mmio_ring);
415 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
416 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
418 kvm_arch_destroy_vm(kvm);
422 void kvm_get_kvm(struct kvm *kvm)
424 atomic_inc(&kvm->users_count);
426 EXPORT_SYMBOL_GPL(kvm_get_kvm);
428 void kvm_put_kvm(struct kvm *kvm)
430 if (atomic_dec_and_test(&kvm->users_count))
433 EXPORT_SYMBOL_GPL(kvm_put_kvm);
436 static int kvm_vm_release(struct inode *inode, struct file *filp)
438 struct kvm *kvm = filp->private_data;
445 * Allocate some memory and give it an address in the guest physical address
448 * Discontiguous memory is allowed, mostly for framebuffers.
450 * Must be called holding mmap_sem for write.
452 int __kvm_set_memory_region(struct kvm *kvm,
453 struct kvm_userspace_memory_region *mem,
458 unsigned long npages;
460 struct kvm_memory_slot *memslot;
461 struct kvm_memory_slot old, new;
464 /* General sanity checks */
465 if (mem->memory_size & (PAGE_SIZE - 1))
467 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
469 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
471 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
474 memslot = &kvm->memslots[mem->slot];
475 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
476 npages = mem->memory_size >> PAGE_SHIFT;
479 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
481 new = old = *memslot;
483 new.base_gfn = base_gfn;
485 new.flags = mem->flags;
487 /* Disallow changing a memory slot's size. */
489 if (npages && old.npages && npages != old.npages)
492 /* Check for overlaps */
494 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
495 struct kvm_memory_slot *s = &kvm->memslots[i];
499 if (!((base_gfn + npages <= s->base_gfn) ||
500 (base_gfn >= s->base_gfn + s->npages)))
504 /* Free page dirty bitmap if unneeded */
505 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
506 new.dirty_bitmap = NULL;
510 /* Allocate if a slot is being created */
512 if (npages && !new.rmap) {
513 new.rmap = vmalloc(npages * sizeof(struct page *));
518 memset(new.rmap, 0, npages * sizeof(*new.rmap));
520 new.user_alloc = user_alloc;
522 * hva_to_rmmap() serialzies with the mmu_lock and to be
523 * safe it has to ignore memslots with !user_alloc &&
527 new.userspace_addr = mem->userspace_addr;
529 new.userspace_addr = 0;
531 if (npages && !new.lpage_info) {
532 int largepages = npages / KVM_PAGES_PER_HPAGE;
533 if (npages % KVM_PAGES_PER_HPAGE)
535 if (base_gfn % KVM_PAGES_PER_HPAGE)
538 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
543 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
545 if (base_gfn % KVM_PAGES_PER_HPAGE)
546 new.lpage_info[0].write_count = 1;
547 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
548 new.lpage_info[largepages-1].write_count = 1;
551 /* Allocate page dirty bitmap if needed */
552 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
553 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
555 new.dirty_bitmap = vmalloc(dirty_bytes);
556 if (!new.dirty_bitmap)
558 memset(new.dirty_bitmap, 0, dirty_bytes);
560 #endif /* not defined CONFIG_S390 */
563 kvm_arch_flush_shadow(kvm);
565 spin_lock(&kvm->mmu_lock);
566 if (mem->slot >= kvm->nmemslots)
567 kvm->nmemslots = mem->slot + 1;
570 spin_unlock(&kvm->mmu_lock);
572 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
574 spin_lock(&kvm->mmu_lock);
576 spin_unlock(&kvm->mmu_lock);
580 kvm_free_physmem_slot(&old, &new);
584 kvm_free_physmem_slot(&new, &old);
589 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
591 int kvm_set_memory_region(struct kvm *kvm,
592 struct kvm_userspace_memory_region *mem,
597 down_write(&kvm->slots_lock);
598 r = __kvm_set_memory_region(kvm, mem, user_alloc);
599 up_write(&kvm->slots_lock);
602 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
604 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
606 kvm_userspace_memory_region *mem,
609 if (mem->slot >= KVM_MEMORY_SLOTS)
611 return kvm_set_memory_region(kvm, mem, user_alloc);
614 int kvm_get_dirty_log(struct kvm *kvm,
615 struct kvm_dirty_log *log, int *is_dirty)
617 struct kvm_memory_slot *memslot;
620 unsigned long any = 0;
623 if (log->slot >= KVM_MEMORY_SLOTS)
626 memslot = &kvm->memslots[log->slot];
628 if (!memslot->dirty_bitmap)
631 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
633 for (i = 0; !any && i < n/sizeof(long); ++i)
634 any = memslot->dirty_bitmap[i];
637 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
648 int is_error_page(struct page *page)
650 return page == bad_page;
652 EXPORT_SYMBOL_GPL(is_error_page);
654 int is_error_pfn(pfn_t pfn)
656 return pfn == bad_pfn;
658 EXPORT_SYMBOL_GPL(is_error_pfn);
660 static inline unsigned long bad_hva(void)
665 int kvm_is_error_hva(unsigned long addr)
667 return addr == bad_hva();
669 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
671 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
675 for (i = 0; i < kvm->nmemslots; ++i) {
676 struct kvm_memory_slot *memslot = &kvm->memslots[i];
678 if (gfn >= memslot->base_gfn
679 && gfn < memslot->base_gfn + memslot->npages)
685 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
687 gfn = unalias_gfn(kvm, gfn);
688 return __gfn_to_memslot(kvm, gfn);
691 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
695 gfn = unalias_gfn(kvm, gfn);
696 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
697 struct kvm_memory_slot *memslot = &kvm->memslots[i];
699 if (gfn >= memslot->base_gfn
700 && gfn < memslot->base_gfn + memslot->npages)
705 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
707 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
709 struct kvm_memory_slot *slot;
711 gfn = unalias_gfn(kvm, gfn);
712 slot = __gfn_to_memslot(kvm, gfn);
715 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
717 EXPORT_SYMBOL_GPL(gfn_to_hva);
720 * Requires current->mm->mmap_sem to be held
722 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
724 struct page *page[1];
731 addr = gfn_to_hva(kvm, gfn);
732 if (kvm_is_error_hva(addr)) {
734 return page_to_pfn(bad_page);
737 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
740 if (unlikely(npages != 1)) {
741 struct vm_area_struct *vma;
743 vma = find_vma(current->mm, addr);
744 if (vma == NULL || addr < vma->vm_start ||
745 !(vma->vm_flags & VM_PFNMAP)) {
747 return page_to_pfn(bad_page);
750 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
751 BUG_ON(!is_mmio_pfn(pfn));
753 pfn = page_to_pfn(page[0]);
758 EXPORT_SYMBOL_GPL(gfn_to_pfn);
760 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
764 pfn = gfn_to_pfn(kvm, gfn);
765 if (!is_mmio_pfn(pfn))
766 return pfn_to_page(pfn);
768 WARN_ON(is_mmio_pfn(pfn));
774 EXPORT_SYMBOL_GPL(gfn_to_page);
776 void kvm_release_page_clean(struct page *page)
778 kvm_release_pfn_clean(page_to_pfn(page));
780 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
782 void kvm_release_pfn_clean(pfn_t pfn)
784 if (!is_mmio_pfn(pfn))
785 put_page(pfn_to_page(pfn));
787 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
789 void kvm_release_page_dirty(struct page *page)
791 kvm_release_pfn_dirty(page_to_pfn(page));
793 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
795 void kvm_release_pfn_dirty(pfn_t pfn)
797 kvm_set_pfn_dirty(pfn);
798 kvm_release_pfn_clean(pfn);
800 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
802 void kvm_set_page_dirty(struct page *page)
804 kvm_set_pfn_dirty(page_to_pfn(page));
806 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
808 void kvm_set_pfn_dirty(pfn_t pfn)
810 if (!is_mmio_pfn(pfn)) {
811 struct page *page = pfn_to_page(pfn);
812 if (!PageReserved(page))
816 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
818 void kvm_set_pfn_accessed(pfn_t pfn)
820 if (!is_mmio_pfn(pfn))
821 mark_page_accessed(pfn_to_page(pfn));
823 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
825 void kvm_get_pfn(pfn_t pfn)
827 if (!is_mmio_pfn(pfn))
828 get_page(pfn_to_page(pfn));
830 EXPORT_SYMBOL_GPL(kvm_get_pfn);
832 static int next_segment(unsigned long len, int offset)
834 if (len > PAGE_SIZE - offset)
835 return PAGE_SIZE - offset;
840 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
846 addr = gfn_to_hva(kvm, gfn);
847 if (kvm_is_error_hva(addr))
849 r = copy_from_user(data, (void __user *)addr + offset, len);
854 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
856 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
858 gfn_t gfn = gpa >> PAGE_SHIFT;
860 int offset = offset_in_page(gpa);
863 while ((seg = next_segment(len, offset)) != 0) {
864 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
874 EXPORT_SYMBOL_GPL(kvm_read_guest);
876 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
881 gfn_t gfn = gpa >> PAGE_SHIFT;
882 int offset = offset_in_page(gpa);
884 addr = gfn_to_hva(kvm, gfn);
885 if (kvm_is_error_hva(addr))
888 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
894 EXPORT_SYMBOL(kvm_read_guest_atomic);
896 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
902 addr = gfn_to_hva(kvm, gfn);
903 if (kvm_is_error_hva(addr))
905 r = copy_to_user((void __user *)addr + offset, data, len);
908 mark_page_dirty(kvm, gfn);
911 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
913 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
916 gfn_t gfn = gpa >> PAGE_SHIFT;
918 int offset = offset_in_page(gpa);
921 while ((seg = next_segment(len, offset)) != 0) {
922 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
933 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
935 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
937 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
939 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
941 gfn_t gfn = gpa >> PAGE_SHIFT;
943 int offset = offset_in_page(gpa);
946 while ((seg = next_segment(len, offset)) != 0) {
947 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
956 EXPORT_SYMBOL_GPL(kvm_clear_guest);
958 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
960 struct kvm_memory_slot *memslot;
962 gfn = unalias_gfn(kvm, gfn);
963 memslot = __gfn_to_memslot(kvm, gfn);
964 if (memslot && memslot->dirty_bitmap) {
965 unsigned long rel_gfn = gfn - memslot->base_gfn;
968 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
969 set_bit(rel_gfn, memslot->dirty_bitmap);
974 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
976 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
981 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
983 if (kvm_cpu_has_interrupt(vcpu))
985 if (kvm_cpu_has_pending_timer(vcpu))
987 if (kvm_arch_vcpu_runnable(vcpu))
989 if (signal_pending(current))
997 finish_wait(&vcpu->wq, &wait);
1000 void kvm_resched(struct kvm_vcpu *vcpu)
1002 if (!need_resched())
1006 EXPORT_SYMBOL_GPL(kvm_resched);
1008 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1010 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1013 if (vmf->pgoff == 0)
1014 page = virt_to_page(vcpu->run);
1016 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1017 page = virt_to_page(vcpu->arch.pio_data);
1019 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1020 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1021 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1024 return VM_FAULT_SIGBUS;
1030 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1031 .fault = kvm_vcpu_fault,
1034 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1036 vma->vm_ops = &kvm_vcpu_vm_ops;
1040 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1042 struct kvm_vcpu *vcpu = filp->private_data;
1044 kvm_put_kvm(vcpu->kvm);
1048 static const struct file_operations kvm_vcpu_fops = {
1049 .release = kvm_vcpu_release,
1050 .unlocked_ioctl = kvm_vcpu_ioctl,
1051 .compat_ioctl = kvm_vcpu_ioctl,
1052 .mmap = kvm_vcpu_mmap,
1056 * Allocates an inode for the vcpu.
1058 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1060 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1062 kvm_put_kvm(vcpu->kvm);
1067 * Creates some virtual cpus. Good luck creating more than one.
1069 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1072 struct kvm_vcpu *vcpu;
1077 vcpu = kvm_arch_vcpu_create(kvm, n);
1079 return PTR_ERR(vcpu);
1081 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1083 r = kvm_arch_vcpu_setup(vcpu);
1087 mutex_lock(&kvm->lock);
1088 if (kvm->vcpus[n]) {
1090 mutex_unlock(&kvm->lock);
1093 kvm->vcpus[n] = vcpu;
1094 mutex_unlock(&kvm->lock);
1096 /* Now it's all set up, let userspace reach it */
1098 r = create_vcpu_fd(vcpu);
1104 mutex_lock(&kvm->lock);
1105 kvm->vcpus[n] = NULL;
1106 mutex_unlock(&kvm->lock);
1108 kvm_arch_vcpu_destroy(vcpu);
1112 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1115 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1116 vcpu->sigset_active = 1;
1117 vcpu->sigset = *sigset;
1119 vcpu->sigset_active = 0;
1123 static long kvm_vcpu_ioctl(struct file *filp,
1124 unsigned int ioctl, unsigned long arg)
1126 struct kvm_vcpu *vcpu = filp->private_data;
1127 void __user *argp = (void __user *)arg;
1129 struct kvm_fpu *fpu = NULL;
1130 struct kvm_sregs *kvm_sregs = NULL;
1132 if (vcpu->kvm->mm != current->mm)
1139 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1141 case KVM_GET_REGS: {
1142 struct kvm_regs *kvm_regs;
1145 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1148 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1152 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1159 case KVM_SET_REGS: {
1160 struct kvm_regs *kvm_regs;
1163 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1167 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1169 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1177 case KVM_GET_SREGS: {
1178 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1182 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1186 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1191 case KVM_SET_SREGS: {
1192 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1197 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1199 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1205 case KVM_GET_MP_STATE: {
1206 struct kvm_mp_state mp_state;
1208 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1212 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1217 case KVM_SET_MP_STATE: {
1218 struct kvm_mp_state mp_state;
1221 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1223 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1229 case KVM_TRANSLATE: {
1230 struct kvm_translation tr;
1233 if (copy_from_user(&tr, argp, sizeof tr))
1235 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1239 if (copy_to_user(argp, &tr, sizeof tr))
1244 case KVM_DEBUG_GUEST: {
1245 struct kvm_debug_guest dbg;
1248 if (copy_from_user(&dbg, argp, sizeof dbg))
1250 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1256 case KVM_SET_SIGNAL_MASK: {
1257 struct kvm_signal_mask __user *sigmask_arg = argp;
1258 struct kvm_signal_mask kvm_sigmask;
1259 sigset_t sigset, *p;
1264 if (copy_from_user(&kvm_sigmask, argp,
1265 sizeof kvm_sigmask))
1268 if (kvm_sigmask.len != sizeof sigset)
1271 if (copy_from_user(&sigset, sigmask_arg->sigset,
1276 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1280 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1284 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1288 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1294 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1299 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1301 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1308 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1316 static long kvm_vm_ioctl(struct file *filp,
1317 unsigned int ioctl, unsigned long arg)
1319 struct kvm *kvm = filp->private_data;
1320 void __user *argp = (void __user *)arg;
1323 if (kvm->mm != current->mm)
1326 case KVM_CREATE_VCPU:
1327 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1331 case KVM_SET_USER_MEMORY_REGION: {
1332 struct kvm_userspace_memory_region kvm_userspace_mem;
1335 if (copy_from_user(&kvm_userspace_mem, argp,
1336 sizeof kvm_userspace_mem))
1339 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1344 case KVM_GET_DIRTY_LOG: {
1345 struct kvm_dirty_log log;
1348 if (copy_from_user(&log, argp, sizeof log))
1350 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1355 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1356 case KVM_REGISTER_COALESCED_MMIO: {
1357 struct kvm_coalesced_mmio_zone zone;
1359 if (copy_from_user(&zone, argp, sizeof zone))
1362 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1368 case KVM_UNREGISTER_COALESCED_MMIO: {
1369 struct kvm_coalesced_mmio_zone zone;
1371 if (copy_from_user(&zone, argp, sizeof zone))
1374 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1382 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1388 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1390 struct kvm *kvm = vma->vm_file->private_data;
1393 if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1394 return VM_FAULT_SIGBUS;
1395 page = gfn_to_page(kvm, vmf->pgoff);
1396 if (is_error_page(page)) {
1397 kvm_release_page_clean(page);
1398 return VM_FAULT_SIGBUS;
1404 static struct vm_operations_struct kvm_vm_vm_ops = {
1405 .fault = kvm_vm_fault,
1408 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1410 vma->vm_ops = &kvm_vm_vm_ops;
1414 static const struct file_operations kvm_vm_fops = {
1415 .release = kvm_vm_release,
1416 .unlocked_ioctl = kvm_vm_ioctl,
1417 .compat_ioctl = kvm_vm_ioctl,
1418 .mmap = kvm_vm_mmap,
1421 static int kvm_dev_ioctl_create_vm(void)
1426 kvm = kvm_create_vm();
1428 return PTR_ERR(kvm);
1429 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1436 static long kvm_dev_ioctl(struct file *filp,
1437 unsigned int ioctl, unsigned long arg)
1442 case KVM_GET_API_VERSION:
1446 r = KVM_API_VERSION;
1452 r = kvm_dev_ioctl_create_vm();
1454 case KVM_CHECK_EXTENSION:
1455 r = kvm_dev_ioctl_check_extension(arg);
1457 case KVM_GET_VCPU_MMAP_SIZE:
1461 r = PAGE_SIZE; /* struct kvm_run */
1463 r += PAGE_SIZE; /* pio data page */
1465 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1466 r += PAGE_SIZE; /* coalesced mmio ring page */
1469 case KVM_TRACE_ENABLE:
1470 case KVM_TRACE_PAUSE:
1471 case KVM_TRACE_DISABLE:
1472 r = kvm_trace_ioctl(ioctl, arg);
1475 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1481 static struct file_operations kvm_chardev_ops = {
1482 .unlocked_ioctl = kvm_dev_ioctl,
1483 .compat_ioctl = kvm_dev_ioctl,
1486 static struct miscdevice kvm_dev = {
1492 static void hardware_enable(void *junk)
1494 int cpu = raw_smp_processor_id();
1496 if (cpu_isset(cpu, cpus_hardware_enabled))
1498 cpu_set(cpu, cpus_hardware_enabled);
1499 kvm_arch_hardware_enable(NULL);
1502 static void hardware_disable(void *junk)
1504 int cpu = raw_smp_processor_id();
1506 if (!cpu_isset(cpu, cpus_hardware_enabled))
1508 cpu_clear(cpu, cpus_hardware_enabled);
1509 kvm_arch_hardware_disable(NULL);
1512 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1517 val &= ~CPU_TASKS_FROZEN;
1520 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1522 hardware_disable(NULL);
1524 case CPU_UP_CANCELED:
1525 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1527 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1530 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1532 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1539 asmlinkage void kvm_handle_fault_on_reboot(void)
1542 /* spin while reset goes on */
1545 /* Fault while not rebooting. We want the trace. */
1548 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1550 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1553 if (val == SYS_RESTART) {
1555 * Some (well, at least mine) BIOSes hang on reboot if
1558 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1559 kvm_rebooting = true;
1560 on_each_cpu(hardware_disable, NULL, 1);
1565 static struct notifier_block kvm_reboot_notifier = {
1566 .notifier_call = kvm_reboot,
1570 void kvm_io_bus_init(struct kvm_io_bus *bus)
1572 memset(bus, 0, sizeof(*bus));
1575 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1579 for (i = 0; i < bus->dev_count; i++) {
1580 struct kvm_io_device *pos = bus->devs[i];
1582 kvm_iodevice_destructor(pos);
1586 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1587 gpa_t addr, int len, int is_write)
1591 for (i = 0; i < bus->dev_count; i++) {
1592 struct kvm_io_device *pos = bus->devs[i];
1594 if (pos->in_range(pos, addr, len, is_write))
1601 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1603 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1605 bus->devs[bus->dev_count++] = dev;
1608 static struct notifier_block kvm_cpu_notifier = {
1609 .notifier_call = kvm_cpu_hotplug,
1610 .priority = 20, /* must be > scheduler priority */
1613 static int vm_stat_get(void *_offset, u64 *val)
1615 unsigned offset = (long)_offset;
1619 spin_lock(&kvm_lock);
1620 list_for_each_entry(kvm, &vm_list, vm_list)
1621 *val += *(u32 *)((void *)kvm + offset);
1622 spin_unlock(&kvm_lock);
1626 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1628 static int vcpu_stat_get(void *_offset, u64 *val)
1630 unsigned offset = (long)_offset;
1632 struct kvm_vcpu *vcpu;
1636 spin_lock(&kvm_lock);
1637 list_for_each_entry(kvm, &vm_list, vm_list)
1638 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1639 vcpu = kvm->vcpus[i];
1641 *val += *(u32 *)((void *)vcpu + offset);
1643 spin_unlock(&kvm_lock);
1647 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1649 static struct file_operations *stat_fops[] = {
1650 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1651 [KVM_STAT_VM] = &vm_stat_fops,
1654 static void kvm_init_debug(void)
1656 struct kvm_stats_debugfs_item *p;
1658 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1659 for (p = debugfs_entries; p->name; ++p)
1660 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1661 (void *)(long)p->offset,
1662 stat_fops[p->kind]);
1665 static void kvm_exit_debug(void)
1667 struct kvm_stats_debugfs_item *p;
1669 for (p = debugfs_entries; p->name; ++p)
1670 debugfs_remove(p->dentry);
1671 debugfs_remove(kvm_debugfs_dir);
1674 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1676 hardware_disable(NULL);
1680 static int kvm_resume(struct sys_device *dev)
1682 hardware_enable(NULL);
1686 static struct sysdev_class kvm_sysdev_class = {
1688 .suspend = kvm_suspend,
1689 .resume = kvm_resume,
1692 static struct sys_device kvm_sysdev = {
1694 .cls = &kvm_sysdev_class,
1697 struct page *bad_page;
1701 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1703 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1706 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1708 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1710 kvm_arch_vcpu_load(vcpu, cpu);
1713 static void kvm_sched_out(struct preempt_notifier *pn,
1714 struct task_struct *next)
1716 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1718 kvm_arch_vcpu_put(vcpu);
1721 int kvm_init(void *opaque, unsigned int vcpu_size,
1722 struct module *module)
1729 r = kvm_arch_init(opaque);
1733 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1735 if (bad_page == NULL) {
1740 bad_pfn = page_to_pfn(bad_page);
1742 r = kvm_arch_hardware_setup();
1746 for_each_online_cpu(cpu) {
1747 smp_call_function_single(cpu,
1748 kvm_arch_check_processor_compat,
1754 on_each_cpu(hardware_enable, NULL, 1);
1755 r = register_cpu_notifier(&kvm_cpu_notifier);
1758 register_reboot_notifier(&kvm_reboot_notifier);
1760 r = sysdev_class_register(&kvm_sysdev_class);
1764 r = sysdev_register(&kvm_sysdev);
1768 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1769 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1770 __alignof__(struct kvm_vcpu),
1772 if (!kvm_vcpu_cache) {
1777 kvm_chardev_ops.owner = module;
1779 r = misc_register(&kvm_dev);
1781 printk(KERN_ERR "kvm: misc device register failed\n");
1785 kvm_preempt_ops.sched_in = kvm_sched_in;
1786 kvm_preempt_ops.sched_out = kvm_sched_out;
1791 kmem_cache_destroy(kvm_vcpu_cache);
1793 sysdev_unregister(&kvm_sysdev);
1795 sysdev_class_unregister(&kvm_sysdev_class);
1797 unregister_reboot_notifier(&kvm_reboot_notifier);
1798 unregister_cpu_notifier(&kvm_cpu_notifier);
1800 on_each_cpu(hardware_disable, NULL, 1);
1802 kvm_arch_hardware_unsetup();
1804 __free_page(bad_page);
1811 EXPORT_SYMBOL_GPL(kvm_init);
1815 kvm_trace_cleanup();
1816 misc_deregister(&kvm_dev);
1817 kmem_cache_destroy(kvm_vcpu_cache);
1818 sysdev_unregister(&kvm_sysdev);
1819 sysdev_class_unregister(&kvm_sysdev_class);
1820 unregister_reboot_notifier(&kvm_reboot_notifier);
1821 unregister_cpu_notifier(&kvm_cpu_notifier);
1822 on_each_cpu(hardware_disable, NULL, 1);
1823 kvm_arch_hardware_unsetup();
1826 __free_page(bad_page);
1828 EXPORT_SYMBOL_GPL(kvm_exit);