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);
216 free->dirty_bitmap = NULL;
220 void kvm_free_physmem(struct kvm *kvm)
224 for (i = 0; i < kvm->nmemslots; ++i)
225 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
228 static void kvm_destroy_vm(struct kvm *kvm)
230 struct mm_struct *mm = kvm->mm;
232 spin_lock(&kvm_lock);
233 list_del(&kvm->vm_list);
234 spin_unlock(&kvm_lock);
235 kvm_io_bus_destroy(&kvm->pio_bus);
236 kvm_io_bus_destroy(&kvm->mmio_bus);
237 kvm_arch_destroy_vm(kvm);
241 static int kvm_vm_release(struct inode *inode, struct file *filp)
243 struct kvm *kvm = filp->private_data;
250 * Allocate some memory and give it an address in the guest physical address
253 * Discontiguous memory is allowed, mostly for framebuffers.
255 * Must be called holding mmap_sem for write.
257 int __kvm_set_memory_region(struct kvm *kvm,
258 struct kvm_userspace_memory_region *mem,
263 unsigned long npages;
265 struct kvm_memory_slot *memslot;
266 struct kvm_memory_slot old, new;
269 /* General sanity checks */
270 if (mem->memory_size & (PAGE_SIZE - 1))
272 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
274 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
276 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
279 memslot = &kvm->memslots[mem->slot];
280 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
281 npages = mem->memory_size >> PAGE_SHIFT;
284 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
286 new = old = *memslot;
288 new.base_gfn = base_gfn;
290 new.flags = mem->flags;
292 /* Disallow changing a memory slot's size. */
294 if (npages && old.npages && npages != old.npages)
297 /* Check for overlaps */
299 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
300 struct kvm_memory_slot *s = &kvm->memslots[i];
304 if (!((base_gfn + npages <= s->base_gfn) ||
305 (base_gfn >= s->base_gfn + s->npages)))
309 /* Free page dirty bitmap if unneeded */
310 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
311 new.dirty_bitmap = NULL;
315 /* Allocate if a slot is being created */
316 if (npages && !new.rmap) {
317 new.rmap = vmalloc(npages * sizeof(struct page *));
322 memset(new.rmap, 0, npages * sizeof(*new.rmap));
324 new.user_alloc = user_alloc;
325 new.userspace_addr = mem->userspace_addr;
328 /* Allocate page dirty bitmap if needed */
329 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
330 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
332 new.dirty_bitmap = vmalloc(dirty_bytes);
333 if (!new.dirty_bitmap)
335 memset(new.dirty_bitmap, 0, dirty_bytes);
338 if (mem->slot >= kvm->nmemslots)
339 kvm->nmemslots = mem->slot + 1;
343 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
349 kvm_free_physmem_slot(&old, &new);
353 kvm_free_physmem_slot(&new, &old);
358 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
360 int kvm_set_memory_region(struct kvm *kvm,
361 struct kvm_userspace_memory_region *mem,
366 down_write(&kvm->slots_lock);
367 r = __kvm_set_memory_region(kvm, mem, user_alloc);
368 up_write(&kvm->slots_lock);
371 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
373 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
375 kvm_userspace_memory_region *mem,
378 if (mem->slot >= KVM_MEMORY_SLOTS)
380 return kvm_set_memory_region(kvm, mem, user_alloc);
383 int kvm_get_dirty_log(struct kvm *kvm,
384 struct kvm_dirty_log *log, int *is_dirty)
386 struct kvm_memory_slot *memslot;
389 unsigned long any = 0;
392 if (log->slot >= KVM_MEMORY_SLOTS)
395 memslot = &kvm->memslots[log->slot];
397 if (!memslot->dirty_bitmap)
400 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
402 for (i = 0; !any && i < n/sizeof(long); ++i)
403 any = memslot->dirty_bitmap[i];
406 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
417 int is_error_page(struct page *page)
419 return page == bad_page;
421 EXPORT_SYMBOL_GPL(is_error_page);
423 static inline unsigned long bad_hva(void)
428 int kvm_is_error_hva(unsigned long addr)
430 return addr == bad_hva();
432 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
434 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
438 for (i = 0; i < kvm->nmemslots; ++i) {
439 struct kvm_memory_slot *memslot = &kvm->memslots[i];
441 if (gfn >= memslot->base_gfn
442 && gfn < memslot->base_gfn + memslot->npages)
448 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
450 gfn = unalias_gfn(kvm, gfn);
451 return __gfn_to_memslot(kvm, gfn);
454 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
458 gfn = unalias_gfn(kvm, gfn);
459 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
460 struct kvm_memory_slot *memslot = &kvm->memslots[i];
462 if (gfn >= memslot->base_gfn
463 && gfn < memslot->base_gfn + memslot->npages)
468 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
470 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
472 struct kvm_memory_slot *slot;
474 gfn = unalias_gfn(kvm, gfn);
475 slot = __gfn_to_memslot(kvm, gfn);
478 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
482 * Requires current->mm->mmap_sem to be held
484 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
486 struct page *page[1];
492 addr = gfn_to_hva(kvm, gfn);
493 if (kvm_is_error_hva(addr)) {
498 npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
509 EXPORT_SYMBOL_GPL(gfn_to_page);
511 void kvm_release_page_clean(struct page *page)
515 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
517 void kvm_release_page_dirty(struct page *page)
519 if (!PageReserved(page))
523 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
525 static int next_segment(unsigned long len, int offset)
527 if (len > PAGE_SIZE - offset)
528 return PAGE_SIZE - offset;
533 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
539 addr = gfn_to_hva(kvm, gfn);
540 if (kvm_is_error_hva(addr))
542 r = copy_from_user(data, (void __user *)addr + offset, len);
547 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
549 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
551 gfn_t gfn = gpa >> PAGE_SHIFT;
553 int offset = offset_in_page(gpa);
556 while ((seg = next_segment(len, offset)) != 0) {
557 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
567 EXPORT_SYMBOL_GPL(kvm_read_guest);
569 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
574 gfn_t gfn = gpa >> PAGE_SHIFT;
575 int offset = offset_in_page(gpa);
577 addr = gfn_to_hva(kvm, gfn);
578 if (kvm_is_error_hva(addr))
581 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
587 EXPORT_SYMBOL(kvm_read_guest_atomic);
589 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
595 addr = gfn_to_hva(kvm, gfn);
596 if (kvm_is_error_hva(addr))
598 r = copy_to_user((void __user *)addr + offset, data, len);
601 mark_page_dirty(kvm, gfn);
604 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
606 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
609 gfn_t gfn = gpa >> PAGE_SHIFT;
611 int offset = offset_in_page(gpa);
614 while ((seg = next_segment(len, offset)) != 0) {
615 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
626 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
628 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
630 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
632 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
634 gfn_t gfn = gpa >> PAGE_SHIFT;
636 int offset = offset_in_page(gpa);
639 while ((seg = next_segment(len, offset)) != 0) {
640 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
649 EXPORT_SYMBOL_GPL(kvm_clear_guest);
651 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
653 struct kvm_memory_slot *memslot;
655 gfn = unalias_gfn(kvm, gfn);
656 memslot = __gfn_to_memslot(kvm, gfn);
657 if (memslot && memslot->dirty_bitmap) {
658 unsigned long rel_gfn = gfn - memslot->base_gfn;
661 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
662 set_bit(rel_gfn, memslot->dirty_bitmap);
667 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
669 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
671 DECLARE_WAITQUEUE(wait, current);
673 add_wait_queue(&vcpu->wq, &wait);
676 * We will block until either an interrupt or a signal wakes us up
678 while (!kvm_cpu_has_interrupt(vcpu)
679 && !signal_pending(current)
680 && !kvm_arch_vcpu_runnable(vcpu)) {
681 set_current_state(TASK_INTERRUPTIBLE);
687 __set_current_state(TASK_RUNNING);
688 remove_wait_queue(&vcpu->wq, &wait);
691 void kvm_resched(struct kvm_vcpu *vcpu)
697 EXPORT_SYMBOL_GPL(kvm_resched);
699 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
701 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
705 page = virt_to_page(vcpu->run);
707 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
708 page = virt_to_page(vcpu->arch.pio_data);
711 return VM_FAULT_SIGBUS;
717 static struct vm_operations_struct kvm_vcpu_vm_ops = {
718 .fault = kvm_vcpu_fault,
721 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
723 vma->vm_ops = &kvm_vcpu_vm_ops;
727 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
729 struct kvm_vcpu *vcpu = filp->private_data;
731 fput(vcpu->kvm->filp);
735 static const struct file_operations kvm_vcpu_fops = {
736 .release = kvm_vcpu_release,
737 .unlocked_ioctl = kvm_vcpu_ioctl,
738 .compat_ioctl = kvm_vcpu_ioctl,
739 .mmap = kvm_vcpu_mmap,
743 * Allocates an inode for the vcpu.
745 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
751 r = anon_inode_getfd(&fd, &inode, &file,
752 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
755 atomic_inc(&vcpu->kvm->filp->f_count);
760 * Creates some virtual cpus. Good luck creating more than one.
762 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
765 struct kvm_vcpu *vcpu;
770 vcpu = kvm_arch_vcpu_create(kvm, n);
772 return PTR_ERR(vcpu);
774 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
776 r = kvm_arch_vcpu_setup(vcpu);
780 mutex_lock(&kvm->lock);
783 mutex_unlock(&kvm->lock);
786 kvm->vcpus[n] = vcpu;
787 mutex_unlock(&kvm->lock);
789 /* Now it's all set up, let userspace reach it */
790 r = create_vcpu_fd(vcpu);
796 mutex_lock(&kvm->lock);
797 kvm->vcpus[n] = NULL;
798 mutex_unlock(&kvm->lock);
800 kvm_arch_vcpu_destroy(vcpu);
804 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
807 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
808 vcpu->sigset_active = 1;
809 vcpu->sigset = *sigset;
811 vcpu->sigset_active = 0;
815 static long kvm_vcpu_ioctl(struct file *filp,
816 unsigned int ioctl, unsigned long arg)
818 struct kvm_vcpu *vcpu = filp->private_data;
819 void __user *argp = (void __user *)arg;
822 if (vcpu->kvm->mm != current->mm)
829 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
832 struct kvm_regs kvm_regs;
834 memset(&kvm_regs, 0, sizeof kvm_regs);
835 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
839 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
845 struct kvm_regs kvm_regs;
848 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
850 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
856 case KVM_GET_SREGS: {
857 struct kvm_sregs kvm_sregs;
859 memset(&kvm_sregs, 0, sizeof kvm_sregs);
860 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
864 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
869 case KVM_SET_SREGS: {
870 struct kvm_sregs kvm_sregs;
873 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
875 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
881 case KVM_TRANSLATE: {
882 struct kvm_translation tr;
885 if (copy_from_user(&tr, argp, sizeof tr))
887 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
891 if (copy_to_user(argp, &tr, sizeof tr))
896 case KVM_DEBUG_GUEST: {
897 struct kvm_debug_guest dbg;
900 if (copy_from_user(&dbg, argp, sizeof dbg))
902 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
908 case KVM_SET_SIGNAL_MASK: {
909 struct kvm_signal_mask __user *sigmask_arg = argp;
910 struct kvm_signal_mask kvm_sigmask;
916 if (copy_from_user(&kvm_sigmask, argp,
920 if (kvm_sigmask.len != sizeof sigset)
923 if (copy_from_user(&sigset, sigmask_arg->sigset,
928 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
934 memset(&fpu, 0, sizeof fpu);
935 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
939 if (copy_to_user(argp, &fpu, sizeof fpu))
948 if (copy_from_user(&fpu, argp, sizeof fpu))
950 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
957 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
963 static long kvm_vm_ioctl(struct file *filp,
964 unsigned int ioctl, unsigned long arg)
966 struct kvm *kvm = filp->private_data;
967 void __user *argp = (void __user *)arg;
970 if (kvm->mm != current->mm)
973 case KVM_CREATE_VCPU:
974 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
978 case KVM_SET_USER_MEMORY_REGION: {
979 struct kvm_userspace_memory_region kvm_userspace_mem;
982 if (copy_from_user(&kvm_userspace_mem, argp,
983 sizeof kvm_userspace_mem))
986 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
991 case KVM_GET_DIRTY_LOG: {
992 struct kvm_dirty_log log;
995 if (copy_from_user(&log, argp, sizeof log))
997 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1003 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1009 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1011 struct kvm *kvm = vma->vm_file->private_data;
1014 if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1015 return VM_FAULT_SIGBUS;
1016 page = gfn_to_page(kvm, vmf->pgoff);
1017 if (is_error_page(page)) {
1018 kvm_release_page_clean(page);
1019 return VM_FAULT_SIGBUS;
1025 static struct vm_operations_struct kvm_vm_vm_ops = {
1026 .fault = kvm_vm_fault,
1029 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1031 vma->vm_ops = &kvm_vm_vm_ops;
1035 static const struct file_operations kvm_vm_fops = {
1036 .release = kvm_vm_release,
1037 .unlocked_ioctl = kvm_vm_ioctl,
1038 .compat_ioctl = kvm_vm_ioctl,
1039 .mmap = kvm_vm_mmap,
1042 static int kvm_dev_ioctl_create_vm(void)
1045 struct inode *inode;
1049 kvm = kvm_create_vm();
1051 return PTR_ERR(kvm);
1052 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1054 kvm_destroy_vm(kvm);
1063 static long kvm_dev_ioctl(struct file *filp,
1064 unsigned int ioctl, unsigned long arg)
1066 void __user *argp = (void __user *)arg;
1070 case KVM_GET_API_VERSION:
1074 r = KVM_API_VERSION;
1080 r = kvm_dev_ioctl_create_vm();
1082 case KVM_CHECK_EXTENSION:
1083 r = kvm_dev_ioctl_check_extension((long)argp);
1085 case KVM_GET_VCPU_MMAP_SIZE:
1089 r = PAGE_SIZE; /* struct kvm_run */
1091 r += PAGE_SIZE; /* pio data page */
1095 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1101 static struct file_operations kvm_chardev_ops = {
1102 .unlocked_ioctl = kvm_dev_ioctl,
1103 .compat_ioctl = kvm_dev_ioctl,
1106 static struct miscdevice kvm_dev = {
1112 static void hardware_enable(void *junk)
1114 int cpu = raw_smp_processor_id();
1116 if (cpu_isset(cpu, cpus_hardware_enabled))
1118 cpu_set(cpu, cpus_hardware_enabled);
1119 kvm_arch_hardware_enable(NULL);
1122 static void hardware_disable(void *junk)
1124 int cpu = raw_smp_processor_id();
1126 if (!cpu_isset(cpu, cpus_hardware_enabled))
1128 cpu_clear(cpu, cpus_hardware_enabled);
1129 decache_vcpus_on_cpu(cpu);
1130 kvm_arch_hardware_disable(NULL);
1133 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1138 val &= ~CPU_TASKS_FROZEN;
1141 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1143 hardware_disable(NULL);
1145 case CPU_UP_CANCELED:
1146 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1148 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1151 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1153 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1159 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1162 if (val == SYS_RESTART) {
1164 * Some (well, at least mine) BIOSes hang on reboot if
1167 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1168 on_each_cpu(hardware_disable, NULL, 0, 1);
1173 static struct notifier_block kvm_reboot_notifier = {
1174 .notifier_call = kvm_reboot,
1178 void kvm_io_bus_init(struct kvm_io_bus *bus)
1180 memset(bus, 0, sizeof(*bus));
1183 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1187 for (i = 0; i < bus->dev_count; i++) {
1188 struct kvm_io_device *pos = bus->devs[i];
1190 kvm_iodevice_destructor(pos);
1194 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1198 for (i = 0; i < bus->dev_count; i++) {
1199 struct kvm_io_device *pos = bus->devs[i];
1201 if (pos->in_range(pos, addr))
1208 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1210 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1212 bus->devs[bus->dev_count++] = dev;
1215 static struct notifier_block kvm_cpu_notifier = {
1216 .notifier_call = kvm_cpu_hotplug,
1217 .priority = 20, /* must be > scheduler priority */
1220 static int vm_stat_get(void *_offset, u64 *val)
1222 unsigned offset = (long)_offset;
1226 spin_lock(&kvm_lock);
1227 list_for_each_entry(kvm, &vm_list, vm_list)
1228 *val += *(u32 *)((void *)kvm + offset);
1229 spin_unlock(&kvm_lock);
1233 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1235 static int vcpu_stat_get(void *_offset, u64 *val)
1237 unsigned offset = (long)_offset;
1239 struct kvm_vcpu *vcpu;
1243 spin_lock(&kvm_lock);
1244 list_for_each_entry(kvm, &vm_list, vm_list)
1245 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1246 vcpu = kvm->vcpus[i];
1248 *val += *(u32 *)((void *)vcpu + offset);
1250 spin_unlock(&kvm_lock);
1254 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1256 static struct file_operations *stat_fops[] = {
1257 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1258 [KVM_STAT_VM] = &vm_stat_fops,
1261 static void kvm_init_debug(void)
1263 struct kvm_stats_debugfs_item *p;
1265 debugfs_dir = debugfs_create_dir("kvm", NULL);
1266 for (p = debugfs_entries; p->name; ++p)
1267 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1268 (void *)(long)p->offset,
1269 stat_fops[p->kind]);
1272 static void kvm_exit_debug(void)
1274 struct kvm_stats_debugfs_item *p;
1276 for (p = debugfs_entries; p->name; ++p)
1277 debugfs_remove(p->dentry);
1278 debugfs_remove(debugfs_dir);
1281 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1283 hardware_disable(NULL);
1287 static int kvm_resume(struct sys_device *dev)
1289 hardware_enable(NULL);
1293 static struct sysdev_class kvm_sysdev_class = {
1295 .suspend = kvm_suspend,
1296 .resume = kvm_resume,
1299 static struct sys_device kvm_sysdev = {
1301 .cls = &kvm_sysdev_class,
1304 struct page *bad_page;
1307 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1309 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1312 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1314 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1316 kvm_arch_vcpu_load(vcpu, cpu);
1319 static void kvm_sched_out(struct preempt_notifier *pn,
1320 struct task_struct *next)
1322 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1324 kvm_arch_vcpu_put(vcpu);
1327 int kvm_init(void *opaque, unsigned int vcpu_size,
1328 struct module *module)
1335 r = kvm_arch_init(opaque);
1339 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1341 if (bad_page == NULL) {
1346 r = kvm_arch_hardware_setup();
1350 for_each_online_cpu(cpu) {
1351 smp_call_function_single(cpu,
1352 kvm_arch_check_processor_compat,
1358 on_each_cpu(hardware_enable, NULL, 0, 1);
1359 r = register_cpu_notifier(&kvm_cpu_notifier);
1362 register_reboot_notifier(&kvm_reboot_notifier);
1364 r = sysdev_class_register(&kvm_sysdev_class);
1368 r = sysdev_register(&kvm_sysdev);
1372 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1373 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1374 __alignof__(struct kvm_vcpu),
1376 if (!kvm_vcpu_cache) {
1381 kvm_chardev_ops.owner = module;
1383 r = misc_register(&kvm_dev);
1385 printk(KERN_ERR "kvm: misc device register failed\n");
1389 kvm_preempt_ops.sched_in = kvm_sched_in;
1390 kvm_preempt_ops.sched_out = kvm_sched_out;
1395 kmem_cache_destroy(kvm_vcpu_cache);
1397 sysdev_unregister(&kvm_sysdev);
1399 sysdev_class_unregister(&kvm_sysdev_class);
1401 unregister_reboot_notifier(&kvm_reboot_notifier);
1402 unregister_cpu_notifier(&kvm_cpu_notifier);
1404 on_each_cpu(hardware_disable, NULL, 0, 1);
1406 kvm_arch_hardware_unsetup();
1408 __free_page(bad_page);
1415 EXPORT_SYMBOL_GPL(kvm_init);
1419 misc_deregister(&kvm_dev);
1420 kmem_cache_destroy(kvm_vcpu_cache);
1421 sysdev_unregister(&kvm_sysdev);
1422 sysdev_class_unregister(&kvm_sysdev_class);
1423 unregister_reboot_notifier(&kvm_reboot_notifier);
1424 unregister_cpu_notifier(&kvm_cpu_notifier);
1425 on_each_cpu(hardware_disable, NULL, 0, 1);
1426 kvm_arch_hardware_unsetup();
1429 __free_page(bad_page);
1431 EXPORT_SYMBOL_GPL(kvm_exit);