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 MODULE_AUTHOR("Qumranet");
51 MODULE_LICENSE("GPL");
53 DEFINE_SPINLOCK(kvm_lock);
56 static cpumask_t cpus_hardware_enabled;
58 struct kmem_cache *kvm_vcpu_cache;
59 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
61 static __read_mostly struct preempt_ops kvm_preempt_ops;
63 struct dentry *debugfs_dir;
65 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
68 static inline int valid_vcpu(int n)
70 return likely(n >= 0 && n < KVM_MAX_VCPUS);
74 * Switches to specified vcpu, until a matching vcpu_put()
76 void vcpu_load(struct kvm_vcpu *vcpu)
80 mutex_lock(&vcpu->mutex);
82 preempt_notifier_register(&vcpu->preempt_notifier);
83 kvm_arch_vcpu_load(vcpu, cpu);
87 void vcpu_put(struct kvm_vcpu *vcpu)
90 kvm_arch_vcpu_put(vcpu);
91 preempt_notifier_unregister(&vcpu->preempt_notifier);
93 mutex_unlock(&vcpu->mutex);
96 static void ack_flush(void *_completed)
100 void kvm_flush_remote_tlbs(struct kvm *kvm)
104 struct kvm_vcpu *vcpu;
107 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
108 vcpu = kvm->vcpus[i];
111 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
114 if (cpu != -1 && cpu != raw_smp_processor_id())
117 if (cpus_empty(cpus))
119 ++kvm->stat.remote_tlb_flush;
120 smp_call_function_mask(cpus, ack_flush, NULL, 1);
123 void kvm_reload_remote_mmus(struct kvm *kvm)
127 struct kvm_vcpu *vcpu;
130 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
131 vcpu = kvm->vcpus[i];
134 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
137 if (cpu != -1 && cpu != raw_smp_processor_id())
140 if (cpus_empty(cpus))
142 smp_call_function_mask(cpus, ack_flush, NULL, 1);
146 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
151 mutex_init(&vcpu->mutex);
155 init_waitqueue_head(&vcpu->wq);
157 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
162 vcpu->run = page_address(page);
164 r = kvm_arch_vcpu_init(vcpu);
170 free_page((unsigned long)vcpu->run);
174 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
176 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
178 kvm_arch_vcpu_uninit(vcpu);
179 free_page((unsigned long)vcpu->run);
181 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
183 static struct kvm *kvm_create_vm(void)
185 struct kvm *kvm = kvm_arch_create_vm();
190 kvm->mm = current->mm;
191 atomic_inc(&kvm->mm->mm_count);
192 spin_lock_init(&kvm->mmu_lock);
193 kvm_io_bus_init(&kvm->pio_bus);
194 mutex_init(&kvm->lock);
195 kvm_io_bus_init(&kvm->mmio_bus);
196 init_rwsem(&kvm->slots_lock);
197 atomic_set(&kvm->users_count, 1);
198 spin_lock(&kvm_lock);
199 list_add(&kvm->vm_list, &vm_list);
200 spin_unlock(&kvm_lock);
206 * Free any memory in @free but not in @dont.
208 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
209 struct kvm_memory_slot *dont)
211 if (!dont || free->rmap != dont->rmap)
214 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
215 vfree(free->dirty_bitmap);
217 if (!dont || free->lpage_info != dont->lpage_info)
218 vfree(free->lpage_info);
221 free->dirty_bitmap = NULL;
223 free->lpage_info = NULL;
226 void kvm_free_physmem(struct kvm *kvm)
230 for (i = 0; i < kvm->nmemslots; ++i)
231 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
234 static void kvm_destroy_vm(struct kvm *kvm)
236 struct mm_struct *mm = kvm->mm;
238 spin_lock(&kvm_lock);
239 list_del(&kvm->vm_list);
240 spin_unlock(&kvm_lock);
241 kvm_io_bus_destroy(&kvm->pio_bus);
242 kvm_io_bus_destroy(&kvm->mmio_bus);
243 kvm_arch_destroy_vm(kvm);
247 void kvm_get_kvm(struct kvm *kvm)
249 atomic_inc(&kvm->users_count);
251 EXPORT_SYMBOL_GPL(kvm_get_kvm);
253 void kvm_put_kvm(struct kvm *kvm)
255 if (atomic_dec_and_test(&kvm->users_count))
258 EXPORT_SYMBOL_GPL(kvm_put_kvm);
261 static int kvm_vm_release(struct inode *inode, struct file *filp)
263 struct kvm *kvm = filp->private_data;
270 * Allocate some memory and give it an address in the guest physical address
273 * Discontiguous memory is allowed, mostly for framebuffers.
275 * Must be called holding mmap_sem for write.
277 int __kvm_set_memory_region(struct kvm *kvm,
278 struct kvm_userspace_memory_region *mem,
283 unsigned long npages;
285 struct kvm_memory_slot *memslot;
286 struct kvm_memory_slot old, new;
289 /* General sanity checks */
290 if (mem->memory_size & (PAGE_SIZE - 1))
292 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
294 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
296 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
299 memslot = &kvm->memslots[mem->slot];
300 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
301 npages = mem->memory_size >> PAGE_SHIFT;
304 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
306 new = old = *memslot;
308 new.base_gfn = base_gfn;
310 new.flags = mem->flags;
312 /* Disallow changing a memory slot's size. */
314 if (npages && old.npages && npages != old.npages)
317 /* Check for overlaps */
319 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
320 struct kvm_memory_slot *s = &kvm->memslots[i];
324 if (!((base_gfn + npages <= s->base_gfn) ||
325 (base_gfn >= s->base_gfn + s->npages)))
329 /* Free page dirty bitmap if unneeded */
330 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
331 new.dirty_bitmap = NULL;
335 /* Allocate if a slot is being created */
336 if (npages && !new.rmap) {
337 new.rmap = vmalloc(npages * sizeof(struct page *));
342 memset(new.rmap, 0, npages * sizeof(*new.rmap));
344 new.user_alloc = user_alloc;
345 new.userspace_addr = mem->userspace_addr;
347 if (npages && !new.lpage_info) {
348 int largepages = npages / KVM_PAGES_PER_HPAGE;
349 if (npages % KVM_PAGES_PER_HPAGE)
351 if (base_gfn % KVM_PAGES_PER_HPAGE)
354 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
359 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
361 if (base_gfn % KVM_PAGES_PER_HPAGE)
362 new.lpage_info[0].write_count = 1;
363 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
364 new.lpage_info[largepages-1].write_count = 1;
367 /* Allocate page dirty bitmap if needed */
368 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
369 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
371 new.dirty_bitmap = vmalloc(dirty_bytes);
372 if (!new.dirty_bitmap)
374 memset(new.dirty_bitmap, 0, dirty_bytes);
377 if (mem->slot >= kvm->nmemslots)
378 kvm->nmemslots = mem->slot + 1;
382 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
388 kvm_free_physmem_slot(&old, &new);
392 kvm_free_physmem_slot(&new, &old);
397 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
399 int kvm_set_memory_region(struct kvm *kvm,
400 struct kvm_userspace_memory_region *mem,
405 down_write(&kvm->slots_lock);
406 r = __kvm_set_memory_region(kvm, mem, user_alloc);
407 up_write(&kvm->slots_lock);
410 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
412 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
414 kvm_userspace_memory_region *mem,
417 if (mem->slot >= KVM_MEMORY_SLOTS)
419 return kvm_set_memory_region(kvm, mem, user_alloc);
422 int kvm_get_dirty_log(struct kvm *kvm,
423 struct kvm_dirty_log *log, int *is_dirty)
425 struct kvm_memory_slot *memslot;
428 unsigned long any = 0;
431 if (log->slot >= KVM_MEMORY_SLOTS)
434 memslot = &kvm->memslots[log->slot];
436 if (!memslot->dirty_bitmap)
439 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
441 for (i = 0; !any && i < n/sizeof(long); ++i)
442 any = memslot->dirty_bitmap[i];
445 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
456 int is_error_page(struct page *page)
458 return page == bad_page;
460 EXPORT_SYMBOL_GPL(is_error_page);
462 int is_error_pfn(pfn_t pfn)
464 return pfn == bad_pfn;
466 EXPORT_SYMBOL_GPL(is_error_pfn);
468 static inline unsigned long bad_hva(void)
473 int kvm_is_error_hva(unsigned long addr)
475 return addr == bad_hva();
477 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
479 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
483 for (i = 0; i < kvm->nmemslots; ++i) {
484 struct kvm_memory_slot *memslot = &kvm->memslots[i];
486 if (gfn >= memslot->base_gfn
487 && gfn < memslot->base_gfn + memslot->npages)
493 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
495 gfn = unalias_gfn(kvm, gfn);
496 return __gfn_to_memslot(kvm, gfn);
499 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
503 gfn = unalias_gfn(kvm, gfn);
504 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
505 struct kvm_memory_slot *memslot = &kvm->memslots[i];
507 if (gfn >= memslot->base_gfn
508 && gfn < memslot->base_gfn + memslot->npages)
513 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
515 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
517 struct kvm_memory_slot *slot;
519 gfn = unalias_gfn(kvm, gfn);
520 slot = __gfn_to_memslot(kvm, gfn);
523 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
527 * Requires current->mm->mmap_sem to be held
529 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
531 struct page *page[1];
537 addr = gfn_to_hva(kvm, gfn);
538 if (kvm_is_error_hva(addr)) {
540 return page_to_pfn(bad_page);
543 npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
548 return page_to_pfn(bad_page);
551 return page_to_pfn(page[0]);
554 EXPORT_SYMBOL_GPL(gfn_to_pfn);
556 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
558 return pfn_to_page(gfn_to_pfn(kvm, gfn));
561 EXPORT_SYMBOL_GPL(gfn_to_page);
563 void kvm_release_page_clean(struct page *page)
565 kvm_release_pfn_clean(page_to_pfn(page));
567 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
569 void kvm_release_pfn_clean(pfn_t pfn)
571 put_page(pfn_to_page(pfn));
573 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
575 void kvm_release_page_dirty(struct page *page)
577 kvm_release_pfn_dirty(page_to_pfn(page));
579 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
581 void kvm_release_pfn_dirty(pfn_t pfn)
583 kvm_set_pfn_dirty(pfn);
584 kvm_release_pfn_clean(pfn);
586 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
588 void kvm_set_page_dirty(struct page *page)
590 kvm_set_pfn_dirty(page_to_pfn(page));
592 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
594 void kvm_set_pfn_dirty(pfn_t pfn)
596 struct page *page = pfn_to_page(pfn);
597 if (!PageReserved(page))
600 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
602 void kvm_set_pfn_accessed(pfn_t pfn)
604 mark_page_accessed(pfn_to_page(pfn));
606 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
608 void kvm_get_pfn(pfn_t pfn)
610 get_page(pfn_to_page(pfn));
612 EXPORT_SYMBOL_GPL(kvm_get_pfn);
614 static int next_segment(unsigned long len, int offset)
616 if (len > PAGE_SIZE - offset)
617 return PAGE_SIZE - offset;
622 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
628 addr = gfn_to_hva(kvm, gfn);
629 if (kvm_is_error_hva(addr))
631 r = copy_from_user(data, (void __user *)addr + offset, len);
636 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
638 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
640 gfn_t gfn = gpa >> PAGE_SHIFT;
642 int offset = offset_in_page(gpa);
645 while ((seg = next_segment(len, offset)) != 0) {
646 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
656 EXPORT_SYMBOL_GPL(kvm_read_guest);
658 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
663 gfn_t gfn = gpa >> PAGE_SHIFT;
664 int offset = offset_in_page(gpa);
666 addr = gfn_to_hva(kvm, gfn);
667 if (kvm_is_error_hva(addr))
670 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
676 EXPORT_SYMBOL(kvm_read_guest_atomic);
678 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
684 addr = gfn_to_hva(kvm, gfn);
685 if (kvm_is_error_hva(addr))
687 r = copy_to_user((void __user *)addr + offset, data, len);
690 mark_page_dirty(kvm, gfn);
693 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
695 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
698 gfn_t gfn = gpa >> PAGE_SHIFT;
700 int offset = offset_in_page(gpa);
703 while ((seg = next_segment(len, offset)) != 0) {
704 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
715 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
717 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
719 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
721 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
723 gfn_t gfn = gpa >> PAGE_SHIFT;
725 int offset = offset_in_page(gpa);
728 while ((seg = next_segment(len, offset)) != 0) {
729 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
738 EXPORT_SYMBOL_GPL(kvm_clear_guest);
740 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
742 struct kvm_memory_slot *memslot;
744 gfn = unalias_gfn(kvm, gfn);
745 memslot = __gfn_to_memslot(kvm, gfn);
746 if (memslot && memslot->dirty_bitmap) {
747 unsigned long rel_gfn = gfn - memslot->base_gfn;
750 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
751 set_bit(rel_gfn, memslot->dirty_bitmap);
756 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
758 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
760 DECLARE_WAITQUEUE(wait, current);
762 add_wait_queue(&vcpu->wq, &wait);
765 * We will block until either an interrupt or a signal wakes us up
767 while (!kvm_cpu_has_interrupt(vcpu)
768 && !signal_pending(current)
769 && !kvm_arch_vcpu_runnable(vcpu)) {
770 set_current_state(TASK_INTERRUPTIBLE);
776 __set_current_state(TASK_RUNNING);
777 remove_wait_queue(&vcpu->wq, &wait);
780 void kvm_resched(struct kvm_vcpu *vcpu)
786 EXPORT_SYMBOL_GPL(kvm_resched);
788 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
790 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
794 page = virt_to_page(vcpu->run);
796 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
797 page = virt_to_page(vcpu->arch.pio_data);
800 return VM_FAULT_SIGBUS;
806 static struct vm_operations_struct kvm_vcpu_vm_ops = {
807 .fault = kvm_vcpu_fault,
810 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
812 vma->vm_ops = &kvm_vcpu_vm_ops;
816 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
818 struct kvm_vcpu *vcpu = filp->private_data;
820 fput(vcpu->kvm->filp);
824 static const struct file_operations kvm_vcpu_fops = {
825 .release = kvm_vcpu_release,
826 .unlocked_ioctl = kvm_vcpu_ioctl,
827 .compat_ioctl = kvm_vcpu_ioctl,
828 .mmap = kvm_vcpu_mmap,
832 * Allocates an inode for the vcpu.
834 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
840 r = anon_inode_getfd(&fd, &inode, &file,
841 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
844 atomic_inc(&vcpu->kvm->filp->f_count);
849 * Creates some virtual cpus. Good luck creating more than one.
851 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
854 struct kvm_vcpu *vcpu;
859 vcpu = kvm_arch_vcpu_create(kvm, n);
861 return PTR_ERR(vcpu);
863 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
865 r = kvm_arch_vcpu_setup(vcpu);
869 mutex_lock(&kvm->lock);
872 mutex_unlock(&kvm->lock);
875 kvm->vcpus[n] = vcpu;
876 mutex_unlock(&kvm->lock);
878 /* Now it's all set up, let userspace reach it */
879 r = create_vcpu_fd(vcpu);
885 mutex_lock(&kvm->lock);
886 kvm->vcpus[n] = NULL;
887 mutex_unlock(&kvm->lock);
889 kvm_arch_vcpu_destroy(vcpu);
893 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
896 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
897 vcpu->sigset_active = 1;
898 vcpu->sigset = *sigset;
900 vcpu->sigset_active = 0;
904 static long kvm_vcpu_ioctl(struct file *filp,
905 unsigned int ioctl, unsigned long arg)
907 struct kvm_vcpu *vcpu = filp->private_data;
908 void __user *argp = (void __user *)arg;
911 if (vcpu->kvm->mm != current->mm)
918 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
921 struct kvm_regs *kvm_regs;
924 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
927 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
931 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
939 struct kvm_regs *kvm_regs;
942 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
946 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
948 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
956 case KVM_GET_SREGS: {
957 struct kvm_sregs kvm_sregs;
959 memset(&kvm_sregs, 0, sizeof kvm_sregs);
960 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
964 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
969 case KVM_SET_SREGS: {
970 struct kvm_sregs kvm_sregs;
973 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
975 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
981 case KVM_TRANSLATE: {
982 struct kvm_translation tr;
985 if (copy_from_user(&tr, argp, sizeof tr))
987 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
991 if (copy_to_user(argp, &tr, sizeof tr))
996 case KVM_DEBUG_GUEST: {
997 struct kvm_debug_guest dbg;
1000 if (copy_from_user(&dbg, argp, sizeof dbg))
1002 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1008 case KVM_SET_SIGNAL_MASK: {
1009 struct kvm_signal_mask __user *sigmask_arg = argp;
1010 struct kvm_signal_mask kvm_sigmask;
1011 sigset_t sigset, *p;
1016 if (copy_from_user(&kvm_sigmask, argp,
1017 sizeof kvm_sigmask))
1020 if (kvm_sigmask.len != sizeof sigset)
1023 if (copy_from_user(&sigset, sigmask_arg->sigset,
1028 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1034 memset(&fpu, 0, sizeof fpu);
1035 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1039 if (copy_to_user(argp, &fpu, sizeof fpu))
1048 if (copy_from_user(&fpu, argp, sizeof fpu))
1050 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1057 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1063 static long kvm_vm_ioctl(struct file *filp,
1064 unsigned int ioctl, unsigned long arg)
1066 struct kvm *kvm = filp->private_data;
1067 void __user *argp = (void __user *)arg;
1070 if (kvm->mm != current->mm)
1073 case KVM_CREATE_VCPU:
1074 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1078 case KVM_SET_USER_MEMORY_REGION: {
1079 struct kvm_userspace_memory_region kvm_userspace_mem;
1082 if (copy_from_user(&kvm_userspace_mem, argp,
1083 sizeof kvm_userspace_mem))
1086 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1091 case KVM_GET_DIRTY_LOG: {
1092 struct kvm_dirty_log log;
1095 if (copy_from_user(&log, argp, sizeof log))
1097 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1103 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1109 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1111 struct kvm *kvm = vma->vm_file->private_data;
1114 if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1115 return VM_FAULT_SIGBUS;
1116 page = gfn_to_page(kvm, vmf->pgoff);
1117 if (is_error_page(page)) {
1118 kvm_release_page_clean(page);
1119 return VM_FAULT_SIGBUS;
1125 static struct vm_operations_struct kvm_vm_vm_ops = {
1126 .fault = kvm_vm_fault,
1129 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1131 vma->vm_ops = &kvm_vm_vm_ops;
1135 static const struct file_operations kvm_vm_fops = {
1136 .release = kvm_vm_release,
1137 .unlocked_ioctl = kvm_vm_ioctl,
1138 .compat_ioctl = kvm_vm_ioctl,
1139 .mmap = kvm_vm_mmap,
1142 static int kvm_dev_ioctl_create_vm(void)
1145 struct inode *inode;
1149 kvm = kvm_create_vm();
1151 return PTR_ERR(kvm);
1152 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1154 kvm_destroy_vm(kvm);
1163 static long kvm_dev_ioctl(struct file *filp,
1164 unsigned int ioctl, unsigned long arg)
1166 void __user *argp = (void __user *)arg;
1170 case KVM_GET_API_VERSION:
1174 r = KVM_API_VERSION;
1180 r = kvm_dev_ioctl_create_vm();
1182 case KVM_CHECK_EXTENSION:
1183 r = kvm_dev_ioctl_check_extension((long)argp);
1185 case KVM_GET_VCPU_MMAP_SIZE:
1189 r = PAGE_SIZE; /* struct kvm_run */
1191 r += PAGE_SIZE; /* pio data page */
1194 case KVM_TRACE_ENABLE:
1195 case KVM_TRACE_PAUSE:
1196 case KVM_TRACE_DISABLE:
1197 r = kvm_trace_ioctl(ioctl, arg);
1200 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1206 static struct file_operations kvm_chardev_ops = {
1207 .unlocked_ioctl = kvm_dev_ioctl,
1208 .compat_ioctl = kvm_dev_ioctl,
1211 static struct miscdevice kvm_dev = {
1217 static void hardware_enable(void *junk)
1219 int cpu = raw_smp_processor_id();
1221 if (cpu_isset(cpu, cpus_hardware_enabled))
1223 cpu_set(cpu, cpus_hardware_enabled);
1224 kvm_arch_hardware_enable(NULL);
1227 static void hardware_disable(void *junk)
1229 int cpu = raw_smp_processor_id();
1231 if (!cpu_isset(cpu, cpus_hardware_enabled))
1233 cpu_clear(cpu, cpus_hardware_enabled);
1234 decache_vcpus_on_cpu(cpu);
1235 kvm_arch_hardware_disable(NULL);
1238 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1243 val &= ~CPU_TASKS_FROZEN;
1246 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1248 hardware_disable(NULL);
1250 case CPU_UP_CANCELED:
1251 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1253 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1256 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1258 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1264 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1267 if (val == SYS_RESTART) {
1269 * Some (well, at least mine) BIOSes hang on reboot if
1272 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1273 on_each_cpu(hardware_disable, NULL, 0, 1);
1278 static struct notifier_block kvm_reboot_notifier = {
1279 .notifier_call = kvm_reboot,
1283 void kvm_io_bus_init(struct kvm_io_bus *bus)
1285 memset(bus, 0, sizeof(*bus));
1288 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1292 for (i = 0; i < bus->dev_count; i++) {
1293 struct kvm_io_device *pos = bus->devs[i];
1295 kvm_iodevice_destructor(pos);
1299 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1303 for (i = 0; i < bus->dev_count; i++) {
1304 struct kvm_io_device *pos = bus->devs[i];
1306 if (pos->in_range(pos, addr))
1313 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1315 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1317 bus->devs[bus->dev_count++] = dev;
1320 static struct notifier_block kvm_cpu_notifier = {
1321 .notifier_call = kvm_cpu_hotplug,
1322 .priority = 20, /* must be > scheduler priority */
1325 static int vm_stat_get(void *_offset, u64 *val)
1327 unsigned offset = (long)_offset;
1331 spin_lock(&kvm_lock);
1332 list_for_each_entry(kvm, &vm_list, vm_list)
1333 *val += *(u32 *)((void *)kvm + offset);
1334 spin_unlock(&kvm_lock);
1338 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1340 static int vcpu_stat_get(void *_offset, u64 *val)
1342 unsigned offset = (long)_offset;
1344 struct kvm_vcpu *vcpu;
1348 spin_lock(&kvm_lock);
1349 list_for_each_entry(kvm, &vm_list, vm_list)
1350 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1351 vcpu = kvm->vcpus[i];
1353 *val += *(u32 *)((void *)vcpu + offset);
1355 spin_unlock(&kvm_lock);
1359 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1361 static struct file_operations *stat_fops[] = {
1362 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1363 [KVM_STAT_VM] = &vm_stat_fops,
1366 static void kvm_init_debug(void)
1368 struct kvm_stats_debugfs_item *p;
1370 debugfs_dir = debugfs_create_dir("kvm", NULL);
1371 for (p = debugfs_entries; p->name; ++p)
1372 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1373 (void *)(long)p->offset,
1374 stat_fops[p->kind]);
1377 static void kvm_exit_debug(void)
1379 struct kvm_stats_debugfs_item *p;
1381 for (p = debugfs_entries; p->name; ++p)
1382 debugfs_remove(p->dentry);
1383 debugfs_remove(debugfs_dir);
1386 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1388 hardware_disable(NULL);
1392 static int kvm_resume(struct sys_device *dev)
1394 hardware_enable(NULL);
1398 static struct sysdev_class kvm_sysdev_class = {
1400 .suspend = kvm_suspend,
1401 .resume = kvm_resume,
1404 static struct sys_device kvm_sysdev = {
1406 .cls = &kvm_sysdev_class,
1409 struct page *bad_page;
1413 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1415 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1418 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1420 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1422 kvm_arch_vcpu_load(vcpu, cpu);
1425 static void kvm_sched_out(struct preempt_notifier *pn,
1426 struct task_struct *next)
1428 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1430 kvm_arch_vcpu_put(vcpu);
1433 int kvm_init(void *opaque, unsigned int vcpu_size,
1434 struct module *module)
1441 r = kvm_arch_init(opaque);
1445 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1447 if (bad_page == NULL) {
1452 bad_pfn = page_to_pfn(bad_page);
1454 r = kvm_arch_hardware_setup();
1458 for_each_online_cpu(cpu) {
1459 smp_call_function_single(cpu,
1460 kvm_arch_check_processor_compat,
1466 on_each_cpu(hardware_enable, NULL, 0, 1);
1467 r = register_cpu_notifier(&kvm_cpu_notifier);
1470 register_reboot_notifier(&kvm_reboot_notifier);
1472 r = sysdev_class_register(&kvm_sysdev_class);
1476 r = sysdev_register(&kvm_sysdev);
1480 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1481 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1482 __alignof__(struct kvm_vcpu),
1484 if (!kvm_vcpu_cache) {
1489 kvm_chardev_ops.owner = module;
1491 r = misc_register(&kvm_dev);
1493 printk(KERN_ERR "kvm: misc device register failed\n");
1497 kvm_preempt_ops.sched_in = kvm_sched_in;
1498 kvm_preempt_ops.sched_out = kvm_sched_out;
1503 kmem_cache_destroy(kvm_vcpu_cache);
1505 sysdev_unregister(&kvm_sysdev);
1507 sysdev_class_unregister(&kvm_sysdev_class);
1509 unregister_reboot_notifier(&kvm_reboot_notifier);
1510 unregister_cpu_notifier(&kvm_cpu_notifier);
1512 on_each_cpu(hardware_disable, NULL, 0, 1);
1514 kvm_arch_hardware_unsetup();
1516 __free_page(bad_page);
1523 EXPORT_SYMBOL_GPL(kvm_init);
1527 kvm_trace_cleanup();
1528 misc_deregister(&kvm_dev);
1529 kmem_cache_destroy(kvm_vcpu_cache);
1530 sysdev_unregister(&kvm_sysdev);
1531 sysdev_class_unregister(&kvm_sysdev_class);
1532 unregister_reboot_notifier(&kvm_reboot_notifier);
1533 unregister_cpu_notifier(&kvm_cpu_notifier);
1534 on_each_cpu(hardware_disable, NULL, 0, 1);
1535 kvm_arch_hardware_unsetup();
1538 __free_page(bad_page);
1540 EXPORT_SYMBOL_GPL(kvm_exit);