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>
44 #include <asm/processor.h>
46 #include <asm/uaccess.h>
47 #include <asm/pgtable.h>
49 MODULE_AUTHOR("Qumranet");
50 MODULE_LICENSE("GPL");
52 DEFINE_SPINLOCK(kvm_lock);
55 static cpumask_t cpus_hardware_enabled;
57 struct kmem_cache *kvm_vcpu_cache;
58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
60 static __read_mostly struct preempt_ops kvm_preempt_ops;
62 static struct dentry *debugfs_dir;
64 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
67 static inline int valid_vcpu(int n)
69 return likely(n >= 0 && n < KVM_MAX_VCPUS);
73 * Switches to specified vcpu, until a matching vcpu_put()
75 void vcpu_load(struct kvm_vcpu *vcpu)
79 mutex_lock(&vcpu->mutex);
81 preempt_notifier_register(&vcpu->preempt_notifier);
82 kvm_arch_vcpu_load(vcpu, cpu);
86 void vcpu_put(struct kvm_vcpu *vcpu)
89 kvm_arch_vcpu_put(vcpu);
90 preempt_notifier_unregister(&vcpu->preempt_notifier);
92 mutex_unlock(&vcpu->mutex);
95 static void ack_flush(void *_completed)
99 void kvm_flush_remote_tlbs(struct kvm *kvm)
103 struct kvm_vcpu *vcpu;
106 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
107 vcpu = kvm->vcpus[i];
110 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
113 if (cpu != -1 && cpu != raw_smp_processor_id())
116 if (cpus_empty(cpus))
118 ++kvm->stat.remote_tlb_flush;
119 smp_call_function_mask(cpus, ack_flush, NULL, 1);
122 void kvm_reload_remote_mmus(struct kvm *kvm)
126 struct kvm_vcpu *vcpu;
129 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
130 vcpu = kvm->vcpus[i];
133 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
136 if (cpu != -1 && cpu != raw_smp_processor_id())
139 if (cpus_empty(cpus))
141 smp_call_function_mask(cpus, ack_flush, NULL, 1);
145 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
150 mutex_init(&vcpu->mutex);
154 init_waitqueue_head(&vcpu->wq);
156 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
161 vcpu->run = page_address(page);
163 r = kvm_arch_vcpu_init(vcpu);
169 free_page((unsigned long)vcpu->run);
173 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
175 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
177 kvm_arch_vcpu_uninit(vcpu);
178 free_page((unsigned long)vcpu->run);
180 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
182 static struct kvm *kvm_create_vm(void)
184 struct kvm *kvm = kvm_arch_create_vm();
189 kvm->mm = current->mm;
190 atomic_inc(&kvm->mm->mm_count);
191 spin_lock_init(&kvm->mmu_lock);
192 kvm_io_bus_init(&kvm->pio_bus);
193 mutex_init(&kvm->lock);
194 kvm_io_bus_init(&kvm->mmio_bus);
195 init_rwsem(&kvm->slots_lock);
196 spin_lock(&kvm_lock);
197 list_add(&kvm->vm_list, &vm_list);
198 spin_unlock(&kvm_lock);
204 * Free any memory in @free but not in @dont.
206 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
207 struct kvm_memory_slot *dont)
209 if (!dont || free->rmap != dont->rmap)
212 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
213 vfree(free->dirty_bitmap);
215 if (!dont || free->lpage_info != dont->lpage_info)
216 vfree(free->lpage_info);
219 free->dirty_bitmap = NULL;
221 free->lpage_info = NULL;
224 void kvm_free_physmem(struct kvm *kvm)
228 for (i = 0; i < kvm->nmemslots; ++i)
229 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
232 static void kvm_destroy_vm(struct kvm *kvm)
234 struct mm_struct *mm = kvm->mm;
236 spin_lock(&kvm_lock);
237 list_del(&kvm->vm_list);
238 spin_unlock(&kvm_lock);
239 kvm_io_bus_destroy(&kvm->pio_bus);
240 kvm_io_bus_destroy(&kvm->mmio_bus);
241 kvm_arch_destroy_vm(kvm);
245 static int kvm_vm_release(struct inode *inode, struct file *filp)
247 struct kvm *kvm = filp->private_data;
254 * Allocate some memory and give it an address in the guest physical address
257 * Discontiguous memory is allowed, mostly for framebuffers.
259 * Must be called holding mmap_sem for write.
261 int __kvm_set_memory_region(struct kvm *kvm,
262 struct kvm_userspace_memory_region *mem,
267 unsigned long npages;
269 struct kvm_memory_slot *memslot;
270 struct kvm_memory_slot old, new;
273 /* General sanity checks */
274 if (mem->memory_size & (PAGE_SIZE - 1))
276 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
278 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
280 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
283 memslot = &kvm->memslots[mem->slot];
284 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
285 npages = mem->memory_size >> PAGE_SHIFT;
288 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
290 new = old = *memslot;
292 new.base_gfn = base_gfn;
294 new.flags = mem->flags;
296 /* Disallow changing a memory slot's size. */
298 if (npages && old.npages && npages != old.npages)
301 /* Check for overlaps */
303 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
304 struct kvm_memory_slot *s = &kvm->memslots[i];
308 if (!((base_gfn + npages <= s->base_gfn) ||
309 (base_gfn >= s->base_gfn + s->npages)))
313 /* Free page dirty bitmap if unneeded */
314 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
315 new.dirty_bitmap = NULL;
319 /* Allocate if a slot is being created */
320 if (npages && !new.rmap) {
321 new.rmap = vmalloc(npages * sizeof(struct page *));
326 memset(new.rmap, 0, npages * sizeof(*new.rmap));
328 new.user_alloc = user_alloc;
329 new.userspace_addr = mem->userspace_addr;
331 if (npages && !new.lpage_info) {
332 int largepages = npages / KVM_PAGES_PER_HPAGE;
333 if (npages % KVM_PAGES_PER_HPAGE)
335 if (base_gfn % KVM_PAGES_PER_HPAGE)
338 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
343 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
345 if (base_gfn % KVM_PAGES_PER_HPAGE)
346 new.lpage_info[0].write_count = 1;
347 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
348 new.lpage_info[largepages-1].write_count = 1;
351 /* Allocate page dirty bitmap if needed */
352 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
353 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
355 new.dirty_bitmap = vmalloc(dirty_bytes);
356 if (!new.dirty_bitmap)
358 memset(new.dirty_bitmap, 0, dirty_bytes);
361 if (mem->slot >= kvm->nmemslots)
362 kvm->nmemslots = mem->slot + 1;
366 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
372 kvm_free_physmem_slot(&old, &new);
376 kvm_free_physmem_slot(&new, &old);
381 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
383 int kvm_set_memory_region(struct kvm *kvm,
384 struct kvm_userspace_memory_region *mem,
389 down_write(&kvm->slots_lock);
390 r = __kvm_set_memory_region(kvm, mem, user_alloc);
391 up_write(&kvm->slots_lock);
394 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
396 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
398 kvm_userspace_memory_region *mem,
401 if (mem->slot >= KVM_MEMORY_SLOTS)
403 return kvm_set_memory_region(kvm, mem, user_alloc);
406 int kvm_get_dirty_log(struct kvm *kvm,
407 struct kvm_dirty_log *log, int *is_dirty)
409 struct kvm_memory_slot *memslot;
412 unsigned long any = 0;
415 if (log->slot >= KVM_MEMORY_SLOTS)
418 memslot = &kvm->memslots[log->slot];
420 if (!memslot->dirty_bitmap)
423 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
425 for (i = 0; !any && i < n/sizeof(long); ++i)
426 any = memslot->dirty_bitmap[i];
429 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
440 int is_error_page(struct page *page)
442 return page == bad_page;
444 EXPORT_SYMBOL_GPL(is_error_page);
446 static inline unsigned long bad_hva(void)
451 int kvm_is_error_hva(unsigned long addr)
453 return addr == bad_hva();
455 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
457 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
461 for (i = 0; i < kvm->nmemslots; ++i) {
462 struct kvm_memory_slot *memslot = &kvm->memslots[i];
464 if (gfn >= memslot->base_gfn
465 && gfn < memslot->base_gfn + memslot->npages)
471 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
473 gfn = unalias_gfn(kvm, gfn);
474 return __gfn_to_memslot(kvm, gfn);
477 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
481 gfn = unalias_gfn(kvm, gfn);
482 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
483 struct kvm_memory_slot *memslot = &kvm->memslots[i];
485 if (gfn >= memslot->base_gfn
486 && gfn < memslot->base_gfn + memslot->npages)
491 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
493 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
495 struct kvm_memory_slot *slot;
497 gfn = unalias_gfn(kvm, gfn);
498 slot = __gfn_to_memslot(kvm, gfn);
501 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
505 * Requires current->mm->mmap_sem to be held
507 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
509 struct page *page[1];
515 addr = gfn_to_hva(kvm, gfn);
516 if (kvm_is_error_hva(addr)) {
521 npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
532 EXPORT_SYMBOL_GPL(gfn_to_page);
534 void kvm_release_page_clean(struct page *page)
538 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
540 void kvm_release_page_dirty(struct page *page)
542 if (!PageReserved(page))
546 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
548 static int next_segment(unsigned long len, int offset)
550 if (len > PAGE_SIZE - offset)
551 return PAGE_SIZE - offset;
556 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
562 addr = gfn_to_hva(kvm, gfn);
563 if (kvm_is_error_hva(addr))
565 r = copy_from_user(data, (void __user *)addr + offset, len);
570 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
572 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
574 gfn_t gfn = gpa >> PAGE_SHIFT;
576 int offset = offset_in_page(gpa);
579 while ((seg = next_segment(len, offset)) != 0) {
580 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
590 EXPORT_SYMBOL_GPL(kvm_read_guest);
592 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
597 gfn_t gfn = gpa >> PAGE_SHIFT;
598 int offset = offset_in_page(gpa);
600 addr = gfn_to_hva(kvm, gfn);
601 if (kvm_is_error_hva(addr))
604 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
610 EXPORT_SYMBOL(kvm_read_guest_atomic);
612 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
618 addr = gfn_to_hva(kvm, gfn);
619 if (kvm_is_error_hva(addr))
621 r = copy_to_user((void __user *)addr + offset, data, len);
624 mark_page_dirty(kvm, gfn);
627 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
629 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
632 gfn_t gfn = gpa >> PAGE_SHIFT;
634 int offset = offset_in_page(gpa);
637 while ((seg = next_segment(len, offset)) != 0) {
638 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
649 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
651 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
653 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
655 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
657 gfn_t gfn = gpa >> PAGE_SHIFT;
659 int offset = offset_in_page(gpa);
662 while ((seg = next_segment(len, offset)) != 0) {
663 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
672 EXPORT_SYMBOL_GPL(kvm_clear_guest);
674 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
676 struct kvm_memory_slot *memslot;
678 gfn = unalias_gfn(kvm, gfn);
679 memslot = __gfn_to_memslot(kvm, gfn);
680 if (memslot && memslot->dirty_bitmap) {
681 unsigned long rel_gfn = gfn - memslot->base_gfn;
684 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
685 set_bit(rel_gfn, memslot->dirty_bitmap);
690 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
692 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
694 DECLARE_WAITQUEUE(wait, current);
696 add_wait_queue(&vcpu->wq, &wait);
699 * We will block until either an interrupt or a signal wakes us up
701 while (!kvm_cpu_has_interrupt(vcpu)
702 && !signal_pending(current)
703 && !kvm_arch_vcpu_runnable(vcpu)) {
704 set_current_state(TASK_INTERRUPTIBLE);
710 __set_current_state(TASK_RUNNING);
711 remove_wait_queue(&vcpu->wq, &wait);
714 void kvm_resched(struct kvm_vcpu *vcpu)
720 EXPORT_SYMBOL_GPL(kvm_resched);
722 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
724 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
728 page = virt_to_page(vcpu->run);
730 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
731 page = virt_to_page(vcpu->arch.pio_data);
734 return VM_FAULT_SIGBUS;
740 static struct vm_operations_struct kvm_vcpu_vm_ops = {
741 .fault = kvm_vcpu_fault,
744 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
746 vma->vm_ops = &kvm_vcpu_vm_ops;
750 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
752 struct kvm_vcpu *vcpu = filp->private_data;
754 fput(vcpu->kvm->filp);
758 static const struct file_operations kvm_vcpu_fops = {
759 .release = kvm_vcpu_release,
760 .unlocked_ioctl = kvm_vcpu_ioctl,
761 .compat_ioctl = kvm_vcpu_ioctl,
762 .mmap = kvm_vcpu_mmap,
766 * Allocates an inode for the vcpu.
768 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
774 r = anon_inode_getfd(&fd, &inode, &file,
775 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
778 atomic_inc(&vcpu->kvm->filp->f_count);
783 * Creates some virtual cpus. Good luck creating more than one.
785 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
788 struct kvm_vcpu *vcpu;
793 vcpu = kvm_arch_vcpu_create(kvm, n);
795 return PTR_ERR(vcpu);
797 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
799 r = kvm_arch_vcpu_setup(vcpu);
803 mutex_lock(&kvm->lock);
806 mutex_unlock(&kvm->lock);
809 kvm->vcpus[n] = vcpu;
810 mutex_unlock(&kvm->lock);
812 /* Now it's all set up, let userspace reach it */
813 r = create_vcpu_fd(vcpu);
819 mutex_lock(&kvm->lock);
820 kvm->vcpus[n] = NULL;
821 mutex_unlock(&kvm->lock);
823 kvm_arch_vcpu_destroy(vcpu);
827 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
830 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
831 vcpu->sigset_active = 1;
832 vcpu->sigset = *sigset;
834 vcpu->sigset_active = 0;
838 static long kvm_vcpu_ioctl(struct file *filp,
839 unsigned int ioctl, unsigned long arg)
841 struct kvm_vcpu *vcpu = filp->private_data;
842 void __user *argp = (void __user *)arg;
845 if (vcpu->kvm->mm != current->mm)
852 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
855 struct kvm_regs kvm_regs;
857 memset(&kvm_regs, 0, sizeof kvm_regs);
858 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
862 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
868 struct kvm_regs kvm_regs;
871 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
873 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
879 case KVM_GET_SREGS: {
880 struct kvm_sregs kvm_sregs;
882 memset(&kvm_sregs, 0, sizeof kvm_sregs);
883 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
887 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
892 case KVM_SET_SREGS: {
893 struct kvm_sregs kvm_sregs;
896 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
898 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
904 case KVM_TRANSLATE: {
905 struct kvm_translation tr;
908 if (copy_from_user(&tr, argp, sizeof tr))
910 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
914 if (copy_to_user(argp, &tr, sizeof tr))
919 case KVM_DEBUG_GUEST: {
920 struct kvm_debug_guest dbg;
923 if (copy_from_user(&dbg, argp, sizeof dbg))
925 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
931 case KVM_SET_SIGNAL_MASK: {
932 struct kvm_signal_mask __user *sigmask_arg = argp;
933 struct kvm_signal_mask kvm_sigmask;
939 if (copy_from_user(&kvm_sigmask, argp,
943 if (kvm_sigmask.len != sizeof sigset)
946 if (copy_from_user(&sigset, sigmask_arg->sigset,
951 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
957 memset(&fpu, 0, sizeof fpu);
958 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
962 if (copy_to_user(argp, &fpu, sizeof fpu))
971 if (copy_from_user(&fpu, argp, sizeof fpu))
973 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
980 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
986 static long kvm_vm_ioctl(struct file *filp,
987 unsigned int ioctl, unsigned long arg)
989 struct kvm *kvm = filp->private_data;
990 void __user *argp = (void __user *)arg;
993 if (kvm->mm != current->mm)
996 case KVM_CREATE_VCPU:
997 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1001 case KVM_SET_USER_MEMORY_REGION: {
1002 struct kvm_userspace_memory_region kvm_userspace_mem;
1005 if (copy_from_user(&kvm_userspace_mem, argp,
1006 sizeof kvm_userspace_mem))
1009 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1014 case KVM_GET_DIRTY_LOG: {
1015 struct kvm_dirty_log log;
1018 if (copy_from_user(&log, argp, sizeof log))
1020 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1026 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1032 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1034 struct kvm *kvm = vma->vm_file->private_data;
1037 if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1038 return VM_FAULT_SIGBUS;
1039 page = gfn_to_page(kvm, vmf->pgoff);
1040 if (is_error_page(page)) {
1041 kvm_release_page_clean(page);
1042 return VM_FAULT_SIGBUS;
1048 static struct vm_operations_struct kvm_vm_vm_ops = {
1049 .fault = kvm_vm_fault,
1052 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1054 vma->vm_ops = &kvm_vm_vm_ops;
1058 static const struct file_operations kvm_vm_fops = {
1059 .release = kvm_vm_release,
1060 .unlocked_ioctl = kvm_vm_ioctl,
1061 .compat_ioctl = kvm_vm_ioctl,
1062 .mmap = kvm_vm_mmap,
1065 static int kvm_dev_ioctl_create_vm(void)
1068 struct inode *inode;
1072 kvm = kvm_create_vm();
1074 return PTR_ERR(kvm);
1075 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1077 kvm_destroy_vm(kvm);
1086 static long kvm_dev_ioctl(struct file *filp,
1087 unsigned int ioctl, unsigned long arg)
1089 void __user *argp = (void __user *)arg;
1093 case KVM_GET_API_VERSION:
1097 r = KVM_API_VERSION;
1103 r = kvm_dev_ioctl_create_vm();
1105 case KVM_CHECK_EXTENSION:
1106 r = kvm_dev_ioctl_check_extension((long)argp);
1108 case KVM_GET_VCPU_MMAP_SIZE:
1112 r = PAGE_SIZE; /* struct kvm_run */
1114 r += PAGE_SIZE; /* pio data page */
1118 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1124 static struct file_operations kvm_chardev_ops = {
1125 .unlocked_ioctl = kvm_dev_ioctl,
1126 .compat_ioctl = kvm_dev_ioctl,
1129 static struct miscdevice kvm_dev = {
1135 static void hardware_enable(void *junk)
1137 int cpu = raw_smp_processor_id();
1139 if (cpu_isset(cpu, cpus_hardware_enabled))
1141 cpu_set(cpu, cpus_hardware_enabled);
1142 kvm_arch_hardware_enable(NULL);
1145 static void hardware_disable(void *junk)
1147 int cpu = raw_smp_processor_id();
1149 if (!cpu_isset(cpu, cpus_hardware_enabled))
1151 cpu_clear(cpu, cpus_hardware_enabled);
1152 decache_vcpus_on_cpu(cpu);
1153 kvm_arch_hardware_disable(NULL);
1156 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1161 val &= ~CPU_TASKS_FROZEN;
1164 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1166 hardware_disable(NULL);
1168 case CPU_UP_CANCELED:
1169 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1171 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1174 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1176 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1182 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1185 if (val == SYS_RESTART) {
1187 * Some (well, at least mine) BIOSes hang on reboot if
1190 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1191 on_each_cpu(hardware_disable, NULL, 0, 1);
1196 static struct notifier_block kvm_reboot_notifier = {
1197 .notifier_call = kvm_reboot,
1201 void kvm_io_bus_init(struct kvm_io_bus *bus)
1203 memset(bus, 0, sizeof(*bus));
1206 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1210 for (i = 0; i < bus->dev_count; i++) {
1211 struct kvm_io_device *pos = bus->devs[i];
1213 kvm_iodevice_destructor(pos);
1217 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1221 for (i = 0; i < bus->dev_count; i++) {
1222 struct kvm_io_device *pos = bus->devs[i];
1224 if (pos->in_range(pos, addr))
1231 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1233 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1235 bus->devs[bus->dev_count++] = dev;
1238 static struct notifier_block kvm_cpu_notifier = {
1239 .notifier_call = kvm_cpu_hotplug,
1240 .priority = 20, /* must be > scheduler priority */
1243 static int vm_stat_get(void *_offset, u64 *val)
1245 unsigned offset = (long)_offset;
1249 spin_lock(&kvm_lock);
1250 list_for_each_entry(kvm, &vm_list, vm_list)
1251 *val += *(u32 *)((void *)kvm + offset);
1252 spin_unlock(&kvm_lock);
1256 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1258 static int vcpu_stat_get(void *_offset, u64 *val)
1260 unsigned offset = (long)_offset;
1262 struct kvm_vcpu *vcpu;
1266 spin_lock(&kvm_lock);
1267 list_for_each_entry(kvm, &vm_list, vm_list)
1268 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1269 vcpu = kvm->vcpus[i];
1271 *val += *(u32 *)((void *)vcpu + offset);
1273 spin_unlock(&kvm_lock);
1277 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1279 static struct file_operations *stat_fops[] = {
1280 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1281 [KVM_STAT_VM] = &vm_stat_fops,
1284 static void kvm_init_debug(void)
1286 struct kvm_stats_debugfs_item *p;
1288 debugfs_dir = debugfs_create_dir("kvm", NULL);
1289 for (p = debugfs_entries; p->name; ++p)
1290 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1291 (void *)(long)p->offset,
1292 stat_fops[p->kind]);
1295 static void kvm_exit_debug(void)
1297 struct kvm_stats_debugfs_item *p;
1299 for (p = debugfs_entries; p->name; ++p)
1300 debugfs_remove(p->dentry);
1301 debugfs_remove(debugfs_dir);
1304 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1306 hardware_disable(NULL);
1310 static int kvm_resume(struct sys_device *dev)
1312 hardware_enable(NULL);
1316 static struct sysdev_class kvm_sysdev_class = {
1318 .suspend = kvm_suspend,
1319 .resume = kvm_resume,
1322 static struct sys_device kvm_sysdev = {
1324 .cls = &kvm_sysdev_class,
1327 struct page *bad_page;
1330 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1332 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1335 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1337 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1339 kvm_arch_vcpu_load(vcpu, cpu);
1342 static void kvm_sched_out(struct preempt_notifier *pn,
1343 struct task_struct *next)
1345 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1347 kvm_arch_vcpu_put(vcpu);
1350 int kvm_init(void *opaque, unsigned int vcpu_size,
1351 struct module *module)
1358 r = kvm_arch_init(opaque);
1362 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1364 if (bad_page == NULL) {
1369 r = kvm_arch_hardware_setup();
1373 for_each_online_cpu(cpu) {
1374 smp_call_function_single(cpu,
1375 kvm_arch_check_processor_compat,
1381 on_each_cpu(hardware_enable, NULL, 0, 1);
1382 r = register_cpu_notifier(&kvm_cpu_notifier);
1385 register_reboot_notifier(&kvm_reboot_notifier);
1387 r = sysdev_class_register(&kvm_sysdev_class);
1391 r = sysdev_register(&kvm_sysdev);
1395 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1396 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1397 __alignof__(struct kvm_vcpu),
1399 if (!kvm_vcpu_cache) {
1404 kvm_chardev_ops.owner = module;
1406 r = misc_register(&kvm_dev);
1408 printk(KERN_ERR "kvm: misc device register failed\n");
1412 kvm_preempt_ops.sched_in = kvm_sched_in;
1413 kvm_preempt_ops.sched_out = kvm_sched_out;
1418 kmem_cache_destroy(kvm_vcpu_cache);
1420 sysdev_unregister(&kvm_sysdev);
1422 sysdev_class_unregister(&kvm_sysdev_class);
1424 unregister_reboot_notifier(&kvm_reboot_notifier);
1425 unregister_cpu_notifier(&kvm_cpu_notifier);
1427 on_each_cpu(hardware_disable, NULL, 0, 1);
1429 kvm_arch_hardware_unsetup();
1431 __free_page(bad_page);
1438 EXPORT_SYMBOL_GPL(kvm_init);
1442 misc_deregister(&kvm_dev);
1443 kmem_cache_destroy(kvm_vcpu_cache);
1444 sysdev_unregister(&kvm_sysdev);
1445 sysdev_class_unregister(&kvm_sysdev_class);
1446 unregister_reboot_notifier(&kvm_reboot_notifier);
1447 unregister_cpu_notifier(&kvm_cpu_notifier);
1448 on_each_cpu(hardware_disable, NULL, 0, 1);
1449 kvm_arch_hardware_unsetup();
1452 __free_page(bad_page);
1454 EXPORT_SYMBOL_GPL(kvm_exit);