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 "x86_emulate.h"
23 #include <linux/kvm.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/percpu.h>
27 #include <linux/gfp.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <linux/reboot.h>
32 #include <linux/debugfs.h>
33 #include <linux/highmem.h>
34 #include <linux/file.h>
35 #include <linux/sysdev.h>
36 #include <linux/cpu.h>
37 #include <linux/sched.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
46 #include <asm/processor.h>
49 #include <asm/uaccess.h>
52 MODULE_AUTHOR("Qumranet");
53 MODULE_LICENSE("GPL");
55 static DEFINE_SPINLOCK(kvm_lock);
56 static LIST_HEAD(vm_list);
58 static cpumask_t cpus_hardware_enabled;
60 struct kvm_x86_ops *kvm_x86_ops;
61 struct kmem_cache *kvm_vcpu_cache;
62 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
64 static __read_mostly struct preempt_ops kvm_preempt_ops;
66 static struct dentry *debugfs_dir;
68 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
71 static inline int valid_vcpu(int n)
73 return likely(n >= 0 && n < KVM_MAX_VCPUS);
77 * Switches to specified vcpu, until a matching vcpu_put()
79 void vcpu_load(struct kvm_vcpu *vcpu)
83 mutex_lock(&vcpu->mutex);
85 preempt_notifier_register(&vcpu->preempt_notifier);
86 kvm_arch_vcpu_load(vcpu, cpu);
90 void vcpu_put(struct kvm_vcpu *vcpu)
93 kvm_arch_vcpu_put(vcpu);
94 preempt_notifier_unregister(&vcpu->preempt_notifier);
96 mutex_unlock(&vcpu->mutex);
99 static void ack_flush(void *_completed)
103 void kvm_flush_remote_tlbs(struct kvm *kvm)
107 struct kvm_vcpu *vcpu;
110 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
111 vcpu = kvm->vcpus[i];
114 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
117 if (cpu != -1 && cpu != raw_smp_processor_id())
120 smp_call_function_mask(cpus, ack_flush, NULL, 1);
123 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
128 mutex_init(&vcpu->mutex);
130 vcpu->mmu.root_hpa = INVALID_PAGE;
133 if (!irqchip_in_kernel(kvm) || id == 0)
134 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
136 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
137 init_waitqueue_head(&vcpu->wq);
139 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
144 vcpu->run = page_address(page);
146 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
151 vcpu->pio_data = page_address(page);
153 r = kvm_mmu_create(vcpu);
155 goto fail_free_pio_data;
157 if (irqchip_in_kernel(kvm)) {
158 r = kvm_create_lapic(vcpu);
160 goto fail_mmu_destroy;
166 kvm_mmu_destroy(vcpu);
168 free_page((unsigned long)vcpu->pio_data);
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_free_lapic(vcpu);
179 kvm_mmu_destroy(vcpu);
180 free_page((unsigned long)vcpu->pio_data);
181 free_page((unsigned long)vcpu->run);
183 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
185 static struct kvm *kvm_create_vm(void)
187 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
190 return ERR_PTR(-ENOMEM);
192 kvm_io_bus_init(&kvm->pio_bus);
193 mutex_init(&kvm->lock);
194 INIT_LIST_HEAD(&kvm->active_mmu_pages);
195 kvm_io_bus_init(&kvm->mmio_bus);
196 spin_lock(&kvm_lock);
197 list_add(&kvm->vm_list, &vm_list);
198 spin_unlock(&kvm_lock);
203 * Free any memory in @free but not in @dont.
205 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
206 struct kvm_memory_slot *dont)
208 if (!dont || free->rmap != dont->rmap)
211 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
212 vfree(free->dirty_bitmap);
215 free->dirty_bitmap = NULL;
219 static void kvm_free_physmem(struct kvm *kvm)
223 for (i = 0; i < kvm->nmemslots; ++i)
224 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
227 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
230 kvm_mmu_unload(vcpu);
234 static void kvm_free_vcpus(struct kvm *kvm)
239 * Unpin any mmu pages first.
241 for (i = 0; i < KVM_MAX_VCPUS; ++i)
243 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
244 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
246 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
247 kvm->vcpus[i] = NULL;
253 static void kvm_destroy_vm(struct kvm *kvm)
255 spin_lock(&kvm_lock);
256 list_del(&kvm->vm_list);
257 spin_unlock(&kvm_lock);
258 kvm_io_bus_destroy(&kvm->pio_bus);
259 kvm_io_bus_destroy(&kvm->mmio_bus);
263 kvm_free_physmem(kvm);
267 static int kvm_vm_release(struct inode *inode, struct file *filp)
269 struct kvm *kvm = filp->private_data;
276 * Allocate some memory and give it an address in the guest physical address
279 * Discontiguous memory is allowed, mostly for framebuffers.
281 * Must be called holding kvm->lock.
283 int __kvm_set_memory_region(struct kvm *kvm,
284 struct kvm_userspace_memory_region *mem,
289 unsigned long npages;
291 struct kvm_memory_slot *memslot;
292 struct kvm_memory_slot old, new;
295 /* General sanity checks */
296 if (mem->memory_size & (PAGE_SIZE - 1))
298 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
300 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
302 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
305 memslot = &kvm->memslots[mem->slot];
306 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
307 npages = mem->memory_size >> PAGE_SHIFT;
310 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
312 new = old = *memslot;
314 new.base_gfn = base_gfn;
316 new.flags = mem->flags;
318 /* Disallow changing a memory slot's size. */
320 if (npages && old.npages && npages != old.npages)
323 /* Check for overlaps */
325 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
326 struct kvm_memory_slot *s = &kvm->memslots[i];
330 if (!((base_gfn + npages <= s->base_gfn) ||
331 (base_gfn >= s->base_gfn + s->npages)))
335 /* Free page dirty bitmap if unneeded */
336 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
337 new.dirty_bitmap = NULL;
341 /* Allocate if a slot is being created */
342 if (npages && !new.rmap) {
343 new.rmap = vmalloc(npages * sizeof(struct page *));
348 memset(new.rmap, 0, npages * sizeof(*new.rmap));
350 new.user_alloc = user_alloc;
352 new.userspace_addr = mem->userspace_addr;
354 down_write(¤t->mm->mmap_sem);
355 new.userspace_addr = do_mmap(NULL, 0,
357 PROT_READ | PROT_WRITE,
358 MAP_SHARED | MAP_ANONYMOUS,
360 up_write(¤t->mm->mmap_sem);
362 if (IS_ERR((void *)new.userspace_addr))
366 if (!old.user_alloc && old.rmap) {
369 down_write(¤t->mm->mmap_sem);
370 ret = do_munmap(current->mm, old.userspace_addr,
371 old.npages * PAGE_SIZE);
372 up_write(¤t->mm->mmap_sem);
375 "kvm_vm_ioctl_set_memory_region: "
376 "failed to munmap memory\n");
380 /* Allocate page dirty bitmap if needed */
381 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
382 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
384 new.dirty_bitmap = vmalloc(dirty_bytes);
385 if (!new.dirty_bitmap)
387 memset(new.dirty_bitmap, 0, dirty_bytes);
390 if (mem->slot >= kvm->nmemslots)
391 kvm->nmemslots = mem->slot + 1;
393 if (!kvm->n_requested_mmu_pages) {
394 unsigned int n_pages;
397 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
398 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
401 unsigned int nr_mmu_pages;
403 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
404 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
405 nr_mmu_pages = max(nr_mmu_pages,
406 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
407 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
413 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
414 kvm_flush_remote_tlbs(kvm);
416 kvm_free_physmem_slot(&old, &new);
420 kvm_free_physmem_slot(&new, &old);
425 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
427 int kvm_set_memory_region(struct kvm *kvm,
428 struct kvm_userspace_memory_region *mem,
433 mutex_lock(&kvm->lock);
434 r = __kvm_set_memory_region(kvm, mem, user_alloc);
435 mutex_unlock(&kvm->lock);
438 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
440 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
442 kvm_userspace_memory_region *mem,
445 if (mem->slot >= KVM_MEMORY_SLOTS)
447 return kvm_set_memory_region(kvm, mem, user_alloc);
451 * Get (and clear) the dirty memory log for a memory slot.
453 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
454 struct kvm_dirty_log *log)
456 struct kvm_memory_slot *memslot;
459 unsigned long any = 0;
461 mutex_lock(&kvm->lock);
464 if (log->slot >= KVM_MEMORY_SLOTS)
467 memslot = &kvm->memslots[log->slot];
469 if (!memslot->dirty_bitmap)
472 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
474 for (i = 0; !any && i < n/sizeof(long); ++i)
475 any = memslot->dirty_bitmap[i];
478 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
481 /* If nothing is dirty, don't bother messing with page tables. */
483 kvm_mmu_slot_remove_write_access(kvm, log->slot);
484 kvm_flush_remote_tlbs(kvm);
485 memset(memslot->dirty_bitmap, 0, n);
491 mutex_unlock(&kvm->lock);
495 int is_error_page(struct page *page)
497 return page == bad_page;
499 EXPORT_SYMBOL_GPL(is_error_page);
501 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
504 struct kvm_mem_alias *alias;
506 for (i = 0; i < kvm->naliases; ++i) {
507 alias = &kvm->aliases[i];
508 if (gfn >= alias->base_gfn
509 && gfn < alias->base_gfn + alias->npages)
510 return alias->target_gfn + gfn - alias->base_gfn;
515 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
519 for (i = 0; i < kvm->nmemslots; ++i) {
520 struct kvm_memory_slot *memslot = &kvm->memslots[i];
522 if (gfn >= memslot->base_gfn
523 && gfn < memslot->base_gfn + memslot->npages)
529 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
531 gfn = unalias_gfn(kvm, gfn);
532 return __gfn_to_memslot(kvm, gfn);
535 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
539 gfn = unalias_gfn(kvm, gfn);
540 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
541 struct kvm_memory_slot *memslot = &kvm->memslots[i];
543 if (gfn >= memslot->base_gfn
544 && gfn < memslot->base_gfn + memslot->npages)
549 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
552 * Requires current->mm->mmap_sem to be held
554 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
556 struct kvm_memory_slot *slot;
557 struct page *page[1];
562 gfn = unalias_gfn(kvm, gfn);
563 slot = __gfn_to_memslot(kvm, gfn);
569 npages = get_user_pages(current, current->mm,
571 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
581 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
585 down_read(¤t->mm->mmap_sem);
586 page = __gfn_to_page(kvm, gfn);
587 up_read(¤t->mm->mmap_sem);
592 EXPORT_SYMBOL_GPL(gfn_to_page);
594 void kvm_release_page(struct page *page)
596 if (!PageReserved(page))
600 EXPORT_SYMBOL_GPL(kvm_release_page);
602 static int next_segment(unsigned long len, int offset)
604 if (len > PAGE_SIZE - offset)
605 return PAGE_SIZE - offset;
610 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
616 page = gfn_to_page(kvm, gfn);
617 if (is_error_page(page)) {
618 kvm_release_page(page);
621 page_virt = kmap_atomic(page, KM_USER0);
623 memcpy(data, page_virt + offset, len);
625 kunmap_atomic(page_virt, KM_USER0);
626 kvm_release_page(page);
629 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
631 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
633 gfn_t gfn = gpa >> PAGE_SHIFT;
635 int offset = offset_in_page(gpa);
638 while ((seg = next_segment(len, offset)) != 0) {
639 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
649 EXPORT_SYMBOL_GPL(kvm_read_guest);
651 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
657 page = gfn_to_page(kvm, gfn);
658 if (is_error_page(page)) {
659 kvm_release_page(page);
662 page_virt = kmap_atomic(page, KM_USER0);
664 memcpy(page_virt + offset, data, len);
666 kunmap_atomic(page_virt, KM_USER0);
667 mark_page_dirty(kvm, gfn);
668 kvm_release_page(page);
671 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
673 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
676 gfn_t gfn = gpa >> PAGE_SHIFT;
678 int offset = offset_in_page(gpa);
681 while ((seg = next_segment(len, offset)) != 0) {
682 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
693 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
698 page = gfn_to_page(kvm, gfn);
699 if (is_error_page(page)) {
700 kvm_release_page(page);
703 page_virt = kmap_atomic(page, KM_USER0);
705 memset(page_virt + offset, 0, len);
707 kunmap_atomic(page_virt, KM_USER0);
708 kvm_release_page(page);
709 mark_page_dirty(kvm, gfn);
712 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
714 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
716 gfn_t gfn = gpa >> PAGE_SHIFT;
718 int offset = offset_in_page(gpa);
721 while ((seg = next_segment(len, offset)) != 0) {
722 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
731 EXPORT_SYMBOL_GPL(kvm_clear_guest);
733 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
735 struct kvm_memory_slot *memslot;
737 gfn = unalias_gfn(kvm, gfn);
738 memslot = __gfn_to_memslot(kvm, gfn);
739 if (memslot && memslot->dirty_bitmap) {
740 unsigned long rel_gfn = gfn - memslot->base_gfn;
743 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
744 set_bit(rel_gfn, memslot->dirty_bitmap);
749 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
751 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
753 DECLARE_WAITQUEUE(wait, current);
755 add_wait_queue(&vcpu->wq, &wait);
758 * We will block until either an interrupt or a signal wakes us up
760 while (!kvm_cpu_has_interrupt(vcpu)
761 && !signal_pending(current)
762 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
763 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
764 set_current_state(TASK_INTERRUPTIBLE);
770 __set_current_state(TASK_RUNNING);
771 remove_wait_queue(&vcpu->wq, &wait);
774 void kvm_resched(struct kvm_vcpu *vcpu)
780 EXPORT_SYMBOL_GPL(kvm_resched);
783 * Translate a guest virtual address to a guest physical address.
785 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
786 struct kvm_translation *tr)
788 unsigned long vaddr = tr->linear_address;
792 mutex_lock(&vcpu->kvm->lock);
793 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
794 tr->physical_address = gpa;
795 tr->valid = gpa != UNMAPPED_GVA;
798 mutex_unlock(&vcpu->kvm->lock);
804 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
805 struct kvm_interrupt *irq)
807 if (irq->irq < 0 || irq->irq >= 256)
809 if (irqchip_in_kernel(vcpu->kvm))
813 set_bit(irq->irq, vcpu->irq_pending);
814 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
821 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
822 unsigned long address,
825 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
829 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
831 page = virt_to_page(vcpu->run);
832 else if (pgoff == KVM_PIO_PAGE_OFFSET)
833 page = virt_to_page(vcpu->pio_data);
835 return NOPAGE_SIGBUS;
838 *type = VM_FAULT_MINOR;
843 static struct vm_operations_struct kvm_vcpu_vm_ops = {
844 .nopage = kvm_vcpu_nopage,
847 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
849 vma->vm_ops = &kvm_vcpu_vm_ops;
853 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
855 struct kvm_vcpu *vcpu = filp->private_data;
857 fput(vcpu->kvm->filp);
861 static struct file_operations kvm_vcpu_fops = {
862 .release = kvm_vcpu_release,
863 .unlocked_ioctl = kvm_vcpu_ioctl,
864 .compat_ioctl = kvm_vcpu_ioctl,
865 .mmap = kvm_vcpu_mmap,
869 * Allocates an inode for the vcpu.
871 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
877 r = anon_inode_getfd(&fd, &inode, &file,
878 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
881 atomic_inc(&vcpu->kvm->filp->f_count);
886 * Creates some virtual cpus. Good luck creating more than one.
888 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
891 struct kvm_vcpu *vcpu;
896 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
898 return PTR_ERR(vcpu);
900 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
902 /* We do fxsave: this must be aligned. */
903 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
906 r = kvm_x86_ops->vcpu_reset(vcpu);
908 r = kvm_mmu_setup(vcpu);
913 mutex_lock(&kvm->lock);
916 mutex_unlock(&kvm->lock);
919 kvm->vcpus[n] = vcpu;
920 mutex_unlock(&kvm->lock);
922 /* Now it's all set up, let userspace reach it */
923 r = create_vcpu_fd(vcpu);
929 mutex_lock(&kvm->lock);
930 kvm->vcpus[n] = NULL;
931 mutex_unlock(&kvm->lock);
935 kvm_mmu_unload(vcpu);
939 kvm_x86_ops->vcpu_free(vcpu);
943 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
946 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
947 vcpu->sigset_active = 1;
948 vcpu->sigset = *sigset;
950 vcpu->sigset_active = 0;
954 static long kvm_vcpu_ioctl(struct file *filp,
955 unsigned int ioctl, unsigned long arg)
957 struct kvm_vcpu *vcpu = filp->private_data;
958 void __user *argp = (void __user *)arg;
966 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
969 struct kvm_regs kvm_regs;
971 memset(&kvm_regs, 0, sizeof kvm_regs);
972 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
976 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
982 struct kvm_regs kvm_regs;
985 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
987 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
993 case KVM_GET_SREGS: {
994 struct kvm_sregs kvm_sregs;
996 memset(&kvm_sregs, 0, sizeof kvm_sregs);
997 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1001 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1006 case KVM_SET_SREGS: {
1007 struct kvm_sregs kvm_sregs;
1010 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1012 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1018 case KVM_TRANSLATE: {
1019 struct kvm_translation tr;
1022 if (copy_from_user(&tr, argp, sizeof tr))
1024 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
1028 if (copy_to_user(argp, &tr, sizeof tr))
1033 case KVM_INTERRUPT: {
1034 struct kvm_interrupt irq;
1037 if (copy_from_user(&irq, argp, sizeof irq))
1039 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1045 case KVM_DEBUG_GUEST: {
1046 struct kvm_debug_guest dbg;
1049 if (copy_from_user(&dbg, argp, sizeof dbg))
1051 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1057 case KVM_SET_SIGNAL_MASK: {
1058 struct kvm_signal_mask __user *sigmask_arg = argp;
1059 struct kvm_signal_mask kvm_sigmask;
1060 sigset_t sigset, *p;
1065 if (copy_from_user(&kvm_sigmask, argp,
1066 sizeof kvm_sigmask))
1069 if (kvm_sigmask.len != sizeof sigset)
1072 if (copy_from_user(&sigset, sigmask_arg->sigset,
1077 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1083 memset(&fpu, 0, sizeof fpu);
1084 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1088 if (copy_to_user(argp, &fpu, sizeof fpu))
1097 if (copy_from_user(&fpu, argp, sizeof fpu))
1099 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1106 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1112 static long kvm_vm_ioctl(struct file *filp,
1113 unsigned int ioctl, unsigned long arg)
1115 struct kvm *kvm = filp->private_data;
1116 void __user *argp = (void __user *)arg;
1120 case KVM_CREATE_VCPU:
1121 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1125 case KVM_SET_USER_MEMORY_REGION: {
1126 struct kvm_userspace_memory_region kvm_userspace_mem;
1129 if (copy_from_user(&kvm_userspace_mem, argp,
1130 sizeof kvm_userspace_mem))
1133 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1138 case KVM_GET_DIRTY_LOG: {
1139 struct kvm_dirty_log log;
1142 if (copy_from_user(&log, argp, sizeof log))
1144 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1150 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1156 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1157 unsigned long address,
1160 struct kvm *kvm = vma->vm_file->private_data;
1161 unsigned long pgoff;
1164 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1165 if (!kvm_is_visible_gfn(kvm, pgoff))
1166 return NOPAGE_SIGBUS;
1167 /* current->mm->mmap_sem is already held so call lockless version */
1168 page = __gfn_to_page(kvm, pgoff);
1169 if (is_error_page(page)) {
1170 kvm_release_page(page);
1171 return NOPAGE_SIGBUS;
1174 *type = VM_FAULT_MINOR;
1179 static struct vm_operations_struct kvm_vm_vm_ops = {
1180 .nopage = kvm_vm_nopage,
1183 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1185 vma->vm_ops = &kvm_vm_vm_ops;
1189 static struct file_operations kvm_vm_fops = {
1190 .release = kvm_vm_release,
1191 .unlocked_ioctl = kvm_vm_ioctl,
1192 .compat_ioctl = kvm_vm_ioctl,
1193 .mmap = kvm_vm_mmap,
1196 static int kvm_dev_ioctl_create_vm(void)
1199 struct inode *inode;
1203 kvm = kvm_create_vm();
1205 return PTR_ERR(kvm);
1206 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1208 kvm_destroy_vm(kvm);
1217 static long kvm_dev_ioctl(struct file *filp,
1218 unsigned int ioctl, unsigned long arg)
1220 void __user *argp = (void __user *)arg;
1224 case KVM_GET_API_VERSION:
1228 r = KVM_API_VERSION;
1234 r = kvm_dev_ioctl_create_vm();
1236 case KVM_CHECK_EXTENSION: {
1237 int ext = (long)argp;
1240 case KVM_CAP_IRQCHIP:
1242 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1243 case KVM_CAP_USER_MEMORY:
1244 case KVM_CAP_SET_TSS_ADDR:
1253 case KVM_GET_VCPU_MMAP_SIZE:
1260 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1266 static struct file_operations kvm_chardev_ops = {
1267 .unlocked_ioctl = kvm_dev_ioctl,
1268 .compat_ioctl = kvm_dev_ioctl,
1271 static struct miscdevice kvm_dev = {
1278 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
1281 static void decache_vcpus_on_cpu(int cpu)
1284 struct kvm_vcpu *vcpu;
1287 spin_lock(&kvm_lock);
1288 list_for_each_entry(vm, &vm_list, vm_list)
1289 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1290 vcpu = vm->vcpus[i];
1294 * If the vcpu is locked, then it is running on some
1295 * other cpu and therefore it is not cached on the
1298 * If it's not locked, check the last cpu it executed
1301 if (mutex_trylock(&vcpu->mutex)) {
1302 if (vcpu->cpu == cpu) {
1303 kvm_x86_ops->vcpu_decache(vcpu);
1306 mutex_unlock(&vcpu->mutex);
1309 spin_unlock(&kvm_lock);
1312 static void hardware_enable(void *junk)
1314 int cpu = raw_smp_processor_id();
1316 if (cpu_isset(cpu, cpus_hardware_enabled))
1318 cpu_set(cpu, cpus_hardware_enabled);
1319 kvm_x86_ops->hardware_enable(NULL);
1322 static void hardware_disable(void *junk)
1324 int cpu = raw_smp_processor_id();
1326 if (!cpu_isset(cpu, cpus_hardware_enabled))
1328 cpu_clear(cpu, cpus_hardware_enabled);
1329 decache_vcpus_on_cpu(cpu);
1330 kvm_x86_ops->hardware_disable(NULL);
1333 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1340 case CPU_DYING_FROZEN:
1341 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1343 hardware_disable(NULL);
1345 case CPU_UP_CANCELED:
1346 case CPU_UP_CANCELED_FROZEN:
1347 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1349 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1352 case CPU_ONLINE_FROZEN:
1353 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1355 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1361 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1364 if (val == SYS_RESTART) {
1366 * Some (well, at least mine) BIOSes hang on reboot if
1369 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1370 on_each_cpu(hardware_disable, NULL, 0, 1);
1375 static struct notifier_block kvm_reboot_notifier = {
1376 .notifier_call = kvm_reboot,
1380 void kvm_io_bus_init(struct kvm_io_bus *bus)
1382 memset(bus, 0, sizeof(*bus));
1385 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1389 for (i = 0; i < bus->dev_count; i++) {
1390 struct kvm_io_device *pos = bus->devs[i];
1392 kvm_iodevice_destructor(pos);
1396 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1400 for (i = 0; i < bus->dev_count; i++) {
1401 struct kvm_io_device *pos = bus->devs[i];
1403 if (pos->in_range(pos, addr))
1410 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1412 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1414 bus->devs[bus->dev_count++] = dev;
1417 static struct notifier_block kvm_cpu_notifier = {
1418 .notifier_call = kvm_cpu_hotplug,
1419 .priority = 20, /* must be > scheduler priority */
1422 static u64 stat_get(void *_offset)
1424 unsigned offset = (long)_offset;
1427 struct kvm_vcpu *vcpu;
1430 spin_lock(&kvm_lock);
1431 list_for_each_entry(kvm, &vm_list, vm_list)
1432 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1433 vcpu = kvm->vcpus[i];
1435 total += *(u32 *)((void *)vcpu + offset);
1437 spin_unlock(&kvm_lock);
1441 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1443 static __init void kvm_init_debug(void)
1445 struct kvm_stats_debugfs_item *p;
1447 debugfs_dir = debugfs_create_dir("kvm", NULL);
1448 for (p = debugfs_entries; p->name; ++p)
1449 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1450 (void *)(long)p->offset,
1454 static void kvm_exit_debug(void)
1456 struct kvm_stats_debugfs_item *p;
1458 for (p = debugfs_entries; p->name; ++p)
1459 debugfs_remove(p->dentry);
1460 debugfs_remove(debugfs_dir);
1463 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1465 hardware_disable(NULL);
1469 static int kvm_resume(struct sys_device *dev)
1471 hardware_enable(NULL);
1475 static struct sysdev_class kvm_sysdev_class = {
1477 .suspend = kvm_suspend,
1478 .resume = kvm_resume,
1481 static struct sys_device kvm_sysdev = {
1483 .cls = &kvm_sysdev_class,
1486 struct page *bad_page;
1489 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1491 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1494 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1496 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1498 kvm_x86_ops->vcpu_load(vcpu, cpu);
1501 static void kvm_sched_out(struct preempt_notifier *pn,
1502 struct task_struct *next)
1504 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1506 kvm_x86_ops->vcpu_put(vcpu);
1509 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
1510 struct module *module)
1516 printk(KERN_ERR "kvm: already loaded the other module\n");
1520 if (!ops->cpu_has_kvm_support()) {
1521 printk(KERN_ERR "kvm: no hardware support\n");
1524 if (ops->disabled_by_bios()) {
1525 printk(KERN_ERR "kvm: disabled by bios\n");
1531 r = kvm_x86_ops->hardware_setup();
1535 for_each_online_cpu(cpu) {
1536 smp_call_function_single(cpu,
1537 kvm_x86_ops->check_processor_compatibility,
1543 on_each_cpu(hardware_enable, NULL, 0, 1);
1544 r = register_cpu_notifier(&kvm_cpu_notifier);
1547 register_reboot_notifier(&kvm_reboot_notifier);
1549 r = sysdev_class_register(&kvm_sysdev_class);
1553 r = sysdev_register(&kvm_sysdev);
1557 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1558 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1559 __alignof__(struct kvm_vcpu), 0, 0);
1560 if (!kvm_vcpu_cache) {
1565 kvm_chardev_ops.owner = module;
1567 r = misc_register(&kvm_dev);
1569 printk(KERN_ERR "kvm: misc device register failed\n");
1573 kvm_preempt_ops.sched_in = kvm_sched_in;
1574 kvm_preempt_ops.sched_out = kvm_sched_out;
1576 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
1581 kmem_cache_destroy(kvm_vcpu_cache);
1583 sysdev_unregister(&kvm_sysdev);
1585 sysdev_class_unregister(&kvm_sysdev_class);
1587 unregister_reboot_notifier(&kvm_reboot_notifier);
1588 unregister_cpu_notifier(&kvm_cpu_notifier);
1590 on_each_cpu(hardware_disable, NULL, 0, 1);
1592 kvm_x86_ops->hardware_unsetup();
1597 EXPORT_SYMBOL_GPL(kvm_init_x86);
1599 void kvm_exit_x86(void)
1601 misc_deregister(&kvm_dev);
1602 kmem_cache_destroy(kvm_vcpu_cache);
1603 sysdev_unregister(&kvm_sysdev);
1604 sysdev_class_unregister(&kvm_sysdev_class);
1605 unregister_reboot_notifier(&kvm_reboot_notifier);
1606 unregister_cpu_notifier(&kvm_cpu_notifier);
1607 on_each_cpu(hardware_disable, NULL, 0, 1);
1608 kvm_x86_ops->hardware_unsetup();
1611 EXPORT_SYMBOL_GPL(kvm_exit_x86);
1613 static __init int kvm_init(void)
1617 r = kvm_mmu_module_init();
1625 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1627 if (bad_page == NULL) {
1636 kvm_mmu_module_exit();
1641 static __exit void kvm_exit(void)
1644 __free_page(bad_page);
1645 kvm_mmu_module_exit();
1648 module_init(kvm_init)
1649 module_exit(kvm_exit)