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 atomic_set(&kvm->users_count, 1);
197 spin_lock(&kvm_lock);
198 list_add(&kvm->vm_list, &vm_list);
199 spin_unlock(&kvm_lock);
205 * Free any memory in @free but not in @dont.
207 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
208 struct kvm_memory_slot *dont)
210 if (!dont || free->rmap != dont->rmap)
213 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
214 vfree(free->dirty_bitmap);
216 if (!dont || free->lpage_info != dont->lpage_info)
217 vfree(free->lpage_info);
220 free->dirty_bitmap = NULL;
222 free->lpage_info = NULL;
225 void kvm_free_physmem(struct kvm *kvm)
229 for (i = 0; i < kvm->nmemslots; ++i)
230 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
233 static void kvm_destroy_vm(struct kvm *kvm)
235 struct mm_struct *mm = kvm->mm;
237 spin_lock(&kvm_lock);
238 list_del(&kvm->vm_list);
239 spin_unlock(&kvm_lock);
240 kvm_io_bus_destroy(&kvm->pio_bus);
241 kvm_io_bus_destroy(&kvm->mmio_bus);
242 kvm_arch_destroy_vm(kvm);
246 void kvm_get_kvm(struct kvm *kvm)
248 atomic_inc(&kvm->users_count);
250 EXPORT_SYMBOL_GPL(kvm_get_kvm);
252 void kvm_put_kvm(struct kvm *kvm)
254 if (atomic_dec_and_test(&kvm->users_count))
257 EXPORT_SYMBOL_GPL(kvm_put_kvm);
260 static int kvm_vm_release(struct inode *inode, struct file *filp)
262 struct kvm *kvm = filp->private_data;
269 * Allocate some memory and give it an address in the guest physical address
272 * Discontiguous memory is allowed, mostly for framebuffers.
274 * Must be called holding mmap_sem for write.
276 int __kvm_set_memory_region(struct kvm *kvm,
277 struct kvm_userspace_memory_region *mem,
282 unsigned long npages;
284 struct kvm_memory_slot *memslot;
285 struct kvm_memory_slot old, new;
288 /* General sanity checks */
289 if (mem->memory_size & (PAGE_SIZE - 1))
291 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
293 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
295 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
298 memslot = &kvm->memslots[mem->slot];
299 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
300 npages = mem->memory_size >> PAGE_SHIFT;
303 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
305 new = old = *memslot;
307 new.base_gfn = base_gfn;
309 new.flags = mem->flags;
311 /* Disallow changing a memory slot's size. */
313 if (npages && old.npages && npages != old.npages)
316 /* Check for overlaps */
318 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
319 struct kvm_memory_slot *s = &kvm->memslots[i];
323 if (!((base_gfn + npages <= s->base_gfn) ||
324 (base_gfn >= s->base_gfn + s->npages)))
328 /* Free page dirty bitmap if unneeded */
329 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
330 new.dirty_bitmap = NULL;
334 /* Allocate if a slot is being created */
335 if (npages && !new.rmap) {
336 new.rmap = vmalloc(npages * sizeof(struct page *));
341 memset(new.rmap, 0, npages * sizeof(*new.rmap));
343 new.user_alloc = user_alloc;
344 new.userspace_addr = mem->userspace_addr;
346 if (npages && !new.lpage_info) {
347 int largepages = npages / KVM_PAGES_PER_HPAGE;
348 if (npages % KVM_PAGES_PER_HPAGE)
350 if (base_gfn % KVM_PAGES_PER_HPAGE)
353 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
358 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
360 if (base_gfn % KVM_PAGES_PER_HPAGE)
361 new.lpage_info[0].write_count = 1;
362 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
363 new.lpage_info[largepages-1].write_count = 1;
366 /* Allocate page dirty bitmap if needed */
367 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
368 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
370 new.dirty_bitmap = vmalloc(dirty_bytes);
371 if (!new.dirty_bitmap)
373 memset(new.dirty_bitmap, 0, dirty_bytes);
376 if (mem->slot >= kvm->nmemslots)
377 kvm->nmemslots = mem->slot + 1;
381 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
387 kvm_free_physmem_slot(&old, &new);
391 kvm_free_physmem_slot(&new, &old);
396 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
398 int kvm_set_memory_region(struct kvm *kvm,
399 struct kvm_userspace_memory_region *mem,
404 down_write(&kvm->slots_lock);
405 r = __kvm_set_memory_region(kvm, mem, user_alloc);
406 up_write(&kvm->slots_lock);
409 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
411 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
413 kvm_userspace_memory_region *mem,
416 if (mem->slot >= KVM_MEMORY_SLOTS)
418 return kvm_set_memory_region(kvm, mem, user_alloc);
421 int kvm_get_dirty_log(struct kvm *kvm,
422 struct kvm_dirty_log *log, int *is_dirty)
424 struct kvm_memory_slot *memslot;
427 unsigned long any = 0;
430 if (log->slot >= KVM_MEMORY_SLOTS)
433 memslot = &kvm->memslots[log->slot];
435 if (!memslot->dirty_bitmap)
438 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
440 for (i = 0; !any && i < n/sizeof(long); ++i)
441 any = memslot->dirty_bitmap[i];
444 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
455 int is_error_page(struct page *page)
457 return page == bad_page;
459 EXPORT_SYMBOL_GPL(is_error_page);
461 static inline unsigned long bad_hva(void)
466 int kvm_is_error_hva(unsigned long addr)
468 return addr == bad_hva();
470 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
472 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
476 for (i = 0; i < kvm->nmemslots; ++i) {
477 struct kvm_memory_slot *memslot = &kvm->memslots[i];
479 if (gfn >= memslot->base_gfn
480 && gfn < memslot->base_gfn + memslot->npages)
486 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
488 gfn = unalias_gfn(kvm, gfn);
489 return __gfn_to_memslot(kvm, gfn);
492 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
496 gfn = unalias_gfn(kvm, gfn);
497 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
498 struct kvm_memory_slot *memslot = &kvm->memslots[i];
500 if (gfn >= memslot->base_gfn
501 && gfn < memslot->base_gfn + memslot->npages)
506 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
508 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
510 struct kvm_memory_slot *slot;
512 gfn = unalias_gfn(kvm, gfn);
513 slot = __gfn_to_memslot(kvm, gfn);
516 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
520 * Requires current->mm->mmap_sem to be held
522 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
524 struct page *page[1];
530 addr = gfn_to_hva(kvm, gfn);
531 if (kvm_is_error_hva(addr)) {
536 npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
547 EXPORT_SYMBOL_GPL(gfn_to_page);
549 void kvm_release_page_clean(struct page *page)
553 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
555 void kvm_release_page_dirty(struct page *page)
557 if (!PageReserved(page))
561 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
563 static int next_segment(unsigned long len, int offset)
565 if (len > PAGE_SIZE - offset)
566 return PAGE_SIZE - offset;
571 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
577 addr = gfn_to_hva(kvm, gfn);
578 if (kvm_is_error_hva(addr))
580 r = copy_from_user(data, (void __user *)addr + offset, len);
585 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
587 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
589 gfn_t gfn = gpa >> PAGE_SHIFT;
591 int offset = offset_in_page(gpa);
594 while ((seg = next_segment(len, offset)) != 0) {
595 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
605 EXPORT_SYMBOL_GPL(kvm_read_guest);
607 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
612 gfn_t gfn = gpa >> PAGE_SHIFT;
613 int offset = offset_in_page(gpa);
615 addr = gfn_to_hva(kvm, gfn);
616 if (kvm_is_error_hva(addr))
619 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
625 EXPORT_SYMBOL(kvm_read_guest_atomic);
627 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
633 addr = gfn_to_hva(kvm, gfn);
634 if (kvm_is_error_hva(addr))
636 r = copy_to_user((void __user *)addr + offset, data, len);
639 mark_page_dirty(kvm, gfn);
642 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
644 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
647 gfn_t gfn = gpa >> PAGE_SHIFT;
649 int offset = offset_in_page(gpa);
652 while ((seg = next_segment(len, offset)) != 0) {
653 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
664 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
666 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
668 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
670 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
672 gfn_t gfn = gpa >> PAGE_SHIFT;
674 int offset = offset_in_page(gpa);
677 while ((seg = next_segment(len, offset)) != 0) {
678 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
687 EXPORT_SYMBOL_GPL(kvm_clear_guest);
689 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
691 struct kvm_memory_slot *memslot;
693 gfn = unalias_gfn(kvm, gfn);
694 memslot = __gfn_to_memslot(kvm, gfn);
695 if (memslot && memslot->dirty_bitmap) {
696 unsigned long rel_gfn = gfn - memslot->base_gfn;
699 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
700 set_bit(rel_gfn, memslot->dirty_bitmap);
705 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
707 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
709 DECLARE_WAITQUEUE(wait, current);
711 add_wait_queue(&vcpu->wq, &wait);
714 * We will block until either an interrupt or a signal wakes us up
716 while (!kvm_cpu_has_interrupt(vcpu)
717 && !signal_pending(current)
718 && !kvm_arch_vcpu_runnable(vcpu)) {
719 set_current_state(TASK_INTERRUPTIBLE);
725 __set_current_state(TASK_RUNNING);
726 remove_wait_queue(&vcpu->wq, &wait);
729 void kvm_resched(struct kvm_vcpu *vcpu)
735 EXPORT_SYMBOL_GPL(kvm_resched);
737 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
739 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
743 page = virt_to_page(vcpu->run);
745 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
746 page = virt_to_page(vcpu->arch.pio_data);
749 return VM_FAULT_SIGBUS;
755 static struct vm_operations_struct kvm_vcpu_vm_ops = {
756 .fault = kvm_vcpu_fault,
759 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
761 vma->vm_ops = &kvm_vcpu_vm_ops;
765 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
767 struct kvm_vcpu *vcpu = filp->private_data;
769 fput(vcpu->kvm->filp);
773 static const struct file_operations kvm_vcpu_fops = {
774 .release = kvm_vcpu_release,
775 .unlocked_ioctl = kvm_vcpu_ioctl,
776 .compat_ioctl = kvm_vcpu_ioctl,
777 .mmap = kvm_vcpu_mmap,
781 * Allocates an inode for the vcpu.
783 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
789 r = anon_inode_getfd(&fd, &inode, &file,
790 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
793 atomic_inc(&vcpu->kvm->filp->f_count);
798 * Creates some virtual cpus. Good luck creating more than one.
800 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
803 struct kvm_vcpu *vcpu;
808 vcpu = kvm_arch_vcpu_create(kvm, n);
810 return PTR_ERR(vcpu);
812 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
814 r = kvm_arch_vcpu_setup(vcpu);
818 mutex_lock(&kvm->lock);
821 mutex_unlock(&kvm->lock);
824 kvm->vcpus[n] = vcpu;
825 mutex_unlock(&kvm->lock);
827 /* Now it's all set up, let userspace reach it */
828 r = create_vcpu_fd(vcpu);
834 mutex_lock(&kvm->lock);
835 kvm->vcpus[n] = NULL;
836 mutex_unlock(&kvm->lock);
838 kvm_arch_vcpu_destroy(vcpu);
842 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
845 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
846 vcpu->sigset_active = 1;
847 vcpu->sigset = *sigset;
849 vcpu->sigset_active = 0;
853 static long kvm_vcpu_ioctl(struct file *filp,
854 unsigned int ioctl, unsigned long arg)
856 struct kvm_vcpu *vcpu = filp->private_data;
857 void __user *argp = (void __user *)arg;
860 if (vcpu->kvm->mm != current->mm)
867 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
870 struct kvm_regs *kvm_regs;
873 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
876 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
880 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
888 struct kvm_regs *kvm_regs;
891 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
895 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
897 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
905 case KVM_GET_SREGS: {
906 struct kvm_sregs kvm_sregs;
908 memset(&kvm_sregs, 0, sizeof kvm_sregs);
909 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
913 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
918 case KVM_SET_SREGS: {
919 struct kvm_sregs kvm_sregs;
922 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
924 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
930 case KVM_TRANSLATE: {
931 struct kvm_translation tr;
934 if (copy_from_user(&tr, argp, sizeof tr))
936 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
940 if (copy_to_user(argp, &tr, sizeof tr))
945 case KVM_DEBUG_GUEST: {
946 struct kvm_debug_guest dbg;
949 if (copy_from_user(&dbg, argp, sizeof dbg))
951 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
957 case KVM_SET_SIGNAL_MASK: {
958 struct kvm_signal_mask __user *sigmask_arg = argp;
959 struct kvm_signal_mask kvm_sigmask;
965 if (copy_from_user(&kvm_sigmask, argp,
969 if (kvm_sigmask.len != sizeof sigset)
972 if (copy_from_user(&sigset, sigmask_arg->sigset,
977 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
983 memset(&fpu, 0, sizeof fpu);
984 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
988 if (copy_to_user(argp, &fpu, sizeof fpu))
997 if (copy_from_user(&fpu, argp, sizeof fpu))
999 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1006 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1012 static long kvm_vm_ioctl(struct file *filp,
1013 unsigned int ioctl, unsigned long arg)
1015 struct kvm *kvm = filp->private_data;
1016 void __user *argp = (void __user *)arg;
1019 if (kvm->mm != current->mm)
1022 case KVM_CREATE_VCPU:
1023 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1027 case KVM_SET_USER_MEMORY_REGION: {
1028 struct kvm_userspace_memory_region kvm_userspace_mem;
1031 if (copy_from_user(&kvm_userspace_mem, argp,
1032 sizeof kvm_userspace_mem))
1035 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1040 case KVM_GET_DIRTY_LOG: {
1041 struct kvm_dirty_log log;
1044 if (copy_from_user(&log, argp, sizeof log))
1046 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1052 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1058 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1060 struct kvm *kvm = vma->vm_file->private_data;
1063 if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1064 return VM_FAULT_SIGBUS;
1065 page = gfn_to_page(kvm, vmf->pgoff);
1066 if (is_error_page(page)) {
1067 kvm_release_page_clean(page);
1068 return VM_FAULT_SIGBUS;
1074 static struct vm_operations_struct kvm_vm_vm_ops = {
1075 .fault = kvm_vm_fault,
1078 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1080 vma->vm_ops = &kvm_vm_vm_ops;
1084 static const struct file_operations kvm_vm_fops = {
1085 .release = kvm_vm_release,
1086 .unlocked_ioctl = kvm_vm_ioctl,
1087 .compat_ioctl = kvm_vm_ioctl,
1088 .mmap = kvm_vm_mmap,
1091 static int kvm_dev_ioctl_create_vm(void)
1094 struct inode *inode;
1098 kvm = kvm_create_vm();
1100 return PTR_ERR(kvm);
1101 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1103 kvm_destroy_vm(kvm);
1112 static long kvm_dev_ioctl(struct file *filp,
1113 unsigned int ioctl, unsigned long arg)
1115 void __user *argp = (void __user *)arg;
1119 case KVM_GET_API_VERSION:
1123 r = KVM_API_VERSION;
1129 r = kvm_dev_ioctl_create_vm();
1131 case KVM_CHECK_EXTENSION:
1132 r = kvm_dev_ioctl_check_extension((long)argp);
1134 case KVM_GET_VCPU_MMAP_SIZE:
1138 r = PAGE_SIZE; /* struct kvm_run */
1140 r += PAGE_SIZE; /* pio data page */
1144 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1150 static struct file_operations kvm_chardev_ops = {
1151 .unlocked_ioctl = kvm_dev_ioctl,
1152 .compat_ioctl = kvm_dev_ioctl,
1155 static struct miscdevice kvm_dev = {
1161 static void hardware_enable(void *junk)
1163 int cpu = raw_smp_processor_id();
1165 if (cpu_isset(cpu, cpus_hardware_enabled))
1167 cpu_set(cpu, cpus_hardware_enabled);
1168 kvm_arch_hardware_enable(NULL);
1171 static void hardware_disable(void *junk)
1173 int cpu = raw_smp_processor_id();
1175 if (!cpu_isset(cpu, cpus_hardware_enabled))
1177 cpu_clear(cpu, cpus_hardware_enabled);
1178 decache_vcpus_on_cpu(cpu);
1179 kvm_arch_hardware_disable(NULL);
1182 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1187 val &= ~CPU_TASKS_FROZEN;
1190 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1192 hardware_disable(NULL);
1194 case CPU_UP_CANCELED:
1195 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1197 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1200 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1202 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1208 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1211 if (val == SYS_RESTART) {
1213 * Some (well, at least mine) BIOSes hang on reboot if
1216 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1217 on_each_cpu(hardware_disable, NULL, 0, 1);
1222 static struct notifier_block kvm_reboot_notifier = {
1223 .notifier_call = kvm_reboot,
1227 void kvm_io_bus_init(struct kvm_io_bus *bus)
1229 memset(bus, 0, sizeof(*bus));
1232 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1236 for (i = 0; i < bus->dev_count; i++) {
1237 struct kvm_io_device *pos = bus->devs[i];
1239 kvm_iodevice_destructor(pos);
1243 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1247 for (i = 0; i < bus->dev_count; i++) {
1248 struct kvm_io_device *pos = bus->devs[i];
1250 if (pos->in_range(pos, addr))
1257 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1259 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1261 bus->devs[bus->dev_count++] = dev;
1264 static struct notifier_block kvm_cpu_notifier = {
1265 .notifier_call = kvm_cpu_hotplug,
1266 .priority = 20, /* must be > scheduler priority */
1269 static int vm_stat_get(void *_offset, u64 *val)
1271 unsigned offset = (long)_offset;
1275 spin_lock(&kvm_lock);
1276 list_for_each_entry(kvm, &vm_list, vm_list)
1277 *val += *(u32 *)((void *)kvm + offset);
1278 spin_unlock(&kvm_lock);
1282 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1284 static int vcpu_stat_get(void *_offset, u64 *val)
1286 unsigned offset = (long)_offset;
1288 struct kvm_vcpu *vcpu;
1292 spin_lock(&kvm_lock);
1293 list_for_each_entry(kvm, &vm_list, vm_list)
1294 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1295 vcpu = kvm->vcpus[i];
1297 *val += *(u32 *)((void *)vcpu + offset);
1299 spin_unlock(&kvm_lock);
1303 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1305 static struct file_operations *stat_fops[] = {
1306 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1307 [KVM_STAT_VM] = &vm_stat_fops,
1310 static void kvm_init_debug(void)
1312 struct kvm_stats_debugfs_item *p;
1314 debugfs_dir = debugfs_create_dir("kvm", NULL);
1315 for (p = debugfs_entries; p->name; ++p)
1316 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1317 (void *)(long)p->offset,
1318 stat_fops[p->kind]);
1321 static void kvm_exit_debug(void)
1323 struct kvm_stats_debugfs_item *p;
1325 for (p = debugfs_entries; p->name; ++p)
1326 debugfs_remove(p->dentry);
1327 debugfs_remove(debugfs_dir);
1330 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1332 hardware_disable(NULL);
1336 static int kvm_resume(struct sys_device *dev)
1338 hardware_enable(NULL);
1342 static struct sysdev_class kvm_sysdev_class = {
1344 .suspend = kvm_suspend,
1345 .resume = kvm_resume,
1348 static struct sys_device kvm_sysdev = {
1350 .cls = &kvm_sysdev_class,
1353 struct page *bad_page;
1356 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1358 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1361 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1363 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1365 kvm_arch_vcpu_load(vcpu, cpu);
1368 static void kvm_sched_out(struct preempt_notifier *pn,
1369 struct task_struct *next)
1371 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1373 kvm_arch_vcpu_put(vcpu);
1376 int kvm_init(void *opaque, unsigned int vcpu_size,
1377 struct module *module)
1384 r = kvm_arch_init(opaque);
1388 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1390 if (bad_page == NULL) {
1395 r = kvm_arch_hardware_setup();
1399 for_each_online_cpu(cpu) {
1400 smp_call_function_single(cpu,
1401 kvm_arch_check_processor_compat,
1407 on_each_cpu(hardware_enable, NULL, 0, 1);
1408 r = register_cpu_notifier(&kvm_cpu_notifier);
1411 register_reboot_notifier(&kvm_reboot_notifier);
1413 r = sysdev_class_register(&kvm_sysdev_class);
1417 r = sysdev_register(&kvm_sysdev);
1421 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1422 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1423 __alignof__(struct kvm_vcpu),
1425 if (!kvm_vcpu_cache) {
1430 kvm_chardev_ops.owner = module;
1432 r = misc_register(&kvm_dev);
1434 printk(KERN_ERR "kvm: misc device register failed\n");
1438 kvm_preempt_ops.sched_in = kvm_sched_in;
1439 kvm_preempt_ops.sched_out = kvm_sched_out;
1444 kmem_cache_destroy(kvm_vcpu_cache);
1446 sysdev_unregister(&kvm_sysdev);
1448 sysdev_class_unregister(&kvm_sysdev_class);
1450 unregister_reboot_notifier(&kvm_reboot_notifier);
1451 unregister_cpu_notifier(&kvm_cpu_notifier);
1453 on_each_cpu(hardware_disable, NULL, 0, 1);
1455 kvm_arch_hardware_unsetup();
1457 __free_page(bad_page);
1464 EXPORT_SYMBOL_GPL(kvm_init);
1468 misc_deregister(&kvm_dev);
1469 kmem_cache_destroy(kvm_vcpu_cache);
1470 sysdev_unregister(&kvm_sysdev);
1471 sysdev_class_unregister(&kvm_sysdev_class);
1472 unregister_reboot_notifier(&kvm_reboot_notifier);
1473 unregister_cpu_notifier(&kvm_cpu_notifier);
1474 on_each_cpu(hardware_disable, NULL, 0, 1);
1475 kvm_arch_hardware_unsetup();
1478 __free_page(bad_page);
1480 EXPORT_SYMBOL_GPL(kvm_exit);