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 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
51 #include "coalesced_mmio.h"
54 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
55 #include <linux/pci.h>
56 #include <linux/interrupt.h>
60 MODULE_AUTHOR("Qumranet");
61 MODULE_LICENSE("GPL");
63 DEFINE_SPINLOCK(kvm_lock);
66 static cpumask_t cpus_hardware_enabled;
68 struct kmem_cache *kvm_vcpu_cache;
69 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
71 static __read_mostly struct preempt_ops kvm_preempt_ops;
73 struct dentry *kvm_debugfs_dir;
75 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
80 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
81 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
84 struct list_head *ptr;
85 struct kvm_assigned_dev_kernel *match;
87 list_for_each(ptr, head) {
88 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
89 if (match->assigned_dev_id == assigned_dev_id)
95 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
97 struct kvm_assigned_dev_kernel *assigned_dev;
99 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
102 /* This is taken to safely inject irq inside the guest. When
103 * the interrupt injection (or the ioapic code) uses a
104 * finer-grained lock, update this
106 mutex_lock(&assigned_dev->kvm->lock);
107 kvm_set_irq(assigned_dev->kvm,
108 assigned_dev->irq_source_id,
109 assigned_dev->guest_irq, 1);
110 mutex_unlock(&assigned_dev->kvm->lock);
111 kvm_put_kvm(assigned_dev->kvm);
114 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
116 struct kvm_assigned_dev_kernel *assigned_dev =
117 (struct kvm_assigned_dev_kernel *) dev_id;
119 kvm_get_kvm(assigned_dev->kvm);
120 schedule_work(&assigned_dev->interrupt_work);
121 disable_irq_nosync(irq);
125 /* Ack the irq line for an assigned device */
126 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
128 struct kvm_assigned_dev_kernel *dev;
133 dev = container_of(kian, struct kvm_assigned_dev_kernel,
135 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
136 enable_irq(dev->host_irq);
139 static void kvm_free_assigned_device(struct kvm *kvm,
140 struct kvm_assigned_dev_kernel
143 if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested)
144 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
146 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
147 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
149 if (cancel_work_sync(&assigned_dev->interrupt_work))
150 /* We had pending work. That means we will have to take
151 * care of kvm_put_kvm.
155 pci_reset_function(assigned_dev->dev);
157 pci_release_regions(assigned_dev->dev);
158 pci_disable_device(assigned_dev->dev);
159 pci_dev_put(assigned_dev->dev);
161 list_del(&assigned_dev->list);
165 void kvm_free_all_assigned_devices(struct kvm *kvm)
167 struct list_head *ptr, *ptr2;
168 struct kvm_assigned_dev_kernel *assigned_dev;
170 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
171 assigned_dev = list_entry(ptr,
172 struct kvm_assigned_dev_kernel,
175 kvm_free_assigned_device(kvm, assigned_dev);
179 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
180 struct kvm_assigned_irq
184 struct kvm_assigned_dev_kernel *match;
186 mutex_lock(&kvm->lock);
188 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
189 assigned_irq->assigned_dev_id);
191 mutex_unlock(&kvm->lock);
195 if (!match->irq_requested) {
196 INIT_WORK(&match->interrupt_work,
197 kvm_assigned_dev_interrupt_work_handler);
198 if (irqchip_in_kernel(kvm)) {
199 /* Register ack nofitier */
200 match->ack_notifier.gsi = -1;
201 match->ack_notifier.irq_acked =
202 kvm_assigned_dev_ack_irq;
203 kvm_register_irq_ack_notifier(kvm,
204 &match->ack_notifier);
206 /* Request IRQ source ID */
207 r = kvm_request_irq_source_id(kvm);
211 match->irq_source_id = r;
214 match->guest_irq = assigned_irq->guest_irq;
215 match->ack_notifier.gsi = assigned_irq->guest_irq;
216 mutex_unlock(&kvm->lock);
220 if (irqchip_in_kernel(kvm)) {
221 if (!capable(CAP_SYS_RAWIO)) {
226 if (assigned_irq->host_irq)
227 match->host_irq = assigned_irq->host_irq;
229 match->host_irq = match->dev->irq;
230 match->guest_irq = assigned_irq->guest_irq;
231 match->ack_notifier.gsi = assigned_irq->guest_irq;
233 /* Even though this is PCI, we don't want to use shared
234 * interrupts. Sharing host devices with guest-assigned devices
235 * on the same interrupt line is not a happy situation: there
236 * are going to be long delays in accepting, acking, etc.
238 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
239 "kvm_assigned_device", (void *)match)) {
245 match->irq_requested = true;
246 mutex_unlock(&kvm->lock);
249 mutex_unlock(&kvm->lock);
250 kvm_free_assigned_device(kvm, match);
254 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
255 struct kvm_assigned_pci_dev *assigned_dev)
258 struct kvm_assigned_dev_kernel *match;
261 mutex_lock(&kvm->lock);
263 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
264 assigned_dev->assigned_dev_id);
266 /* device already assigned */
271 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
273 printk(KERN_INFO "%s: Couldn't allocate memory\n",
278 dev = pci_get_bus_and_slot(assigned_dev->busnr,
279 assigned_dev->devfn);
281 printk(KERN_INFO "%s: host device not found\n", __func__);
285 if (pci_enable_device(dev)) {
286 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
290 r = pci_request_regions(dev, "kvm_assigned_device");
292 printk(KERN_INFO "%s: Could not get access to device regions\n",
297 pci_reset_function(dev);
299 match->assigned_dev_id = assigned_dev->assigned_dev_id;
300 match->host_busnr = assigned_dev->busnr;
301 match->host_devfn = assigned_dev->devfn;
306 list_add(&match->list, &kvm->arch.assigned_dev_head);
308 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
309 r = kvm_iommu_map_guest(kvm, match);
315 mutex_unlock(&kvm->lock);
318 list_del(&match->list);
319 pci_release_regions(dev);
321 pci_disable_device(dev);
326 mutex_unlock(&kvm->lock);
331 static inline int valid_vcpu(int n)
333 return likely(n >= 0 && n < KVM_MAX_VCPUS);
336 inline int kvm_is_mmio_pfn(pfn_t pfn)
339 return PageReserved(pfn_to_page(pfn));
345 * Switches to specified vcpu, until a matching vcpu_put()
347 void vcpu_load(struct kvm_vcpu *vcpu)
351 mutex_lock(&vcpu->mutex);
353 preempt_notifier_register(&vcpu->preempt_notifier);
354 kvm_arch_vcpu_load(vcpu, cpu);
358 void vcpu_put(struct kvm_vcpu *vcpu)
361 kvm_arch_vcpu_put(vcpu);
362 preempt_notifier_unregister(&vcpu->preempt_notifier);
364 mutex_unlock(&vcpu->mutex);
367 static void ack_flush(void *_completed)
371 void kvm_flush_remote_tlbs(struct kvm *kvm)
375 struct kvm_vcpu *vcpu;
379 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
380 vcpu = kvm->vcpus[i];
383 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
386 if (cpu != -1 && cpu != me)
389 if (cpus_empty(cpus))
391 ++kvm->stat.remote_tlb_flush;
392 smp_call_function_mask(cpus, ack_flush, NULL, 1);
397 void kvm_reload_remote_mmus(struct kvm *kvm)
401 struct kvm_vcpu *vcpu;
405 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
406 vcpu = kvm->vcpus[i];
409 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
412 if (cpu != -1 && cpu != me)
415 if (cpus_empty(cpus))
417 smp_call_function_mask(cpus, ack_flush, NULL, 1);
423 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
428 mutex_init(&vcpu->mutex);
432 init_waitqueue_head(&vcpu->wq);
434 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
439 vcpu->run = page_address(page);
441 r = kvm_arch_vcpu_init(vcpu);
447 free_page((unsigned long)vcpu->run);
451 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
453 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
455 kvm_arch_vcpu_uninit(vcpu);
456 free_page((unsigned long)vcpu->run);
458 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
460 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
461 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
463 return container_of(mn, struct kvm, mmu_notifier);
466 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
467 struct mm_struct *mm,
468 unsigned long address)
470 struct kvm *kvm = mmu_notifier_to_kvm(mn);
474 * When ->invalidate_page runs, the linux pte has been zapped
475 * already but the page is still allocated until
476 * ->invalidate_page returns. So if we increase the sequence
477 * here the kvm page fault will notice if the spte can't be
478 * established because the page is going to be freed. If
479 * instead the kvm page fault establishes the spte before
480 * ->invalidate_page runs, kvm_unmap_hva will release it
483 * The sequence increase only need to be seen at spin_unlock
484 * time, and not at spin_lock time.
486 * Increasing the sequence after the spin_unlock would be
487 * unsafe because the kvm page fault could then establish the
488 * pte after kvm_unmap_hva returned, without noticing the page
489 * is going to be freed.
491 spin_lock(&kvm->mmu_lock);
492 kvm->mmu_notifier_seq++;
493 need_tlb_flush = kvm_unmap_hva(kvm, address);
494 spin_unlock(&kvm->mmu_lock);
496 /* we've to flush the tlb before the pages can be freed */
498 kvm_flush_remote_tlbs(kvm);
502 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
503 struct mm_struct *mm,
507 struct kvm *kvm = mmu_notifier_to_kvm(mn);
508 int need_tlb_flush = 0;
510 spin_lock(&kvm->mmu_lock);
512 * The count increase must become visible at unlock time as no
513 * spte can be established without taking the mmu_lock and
514 * count is also read inside the mmu_lock critical section.
516 kvm->mmu_notifier_count++;
517 for (; start < end; start += PAGE_SIZE)
518 need_tlb_flush |= kvm_unmap_hva(kvm, start);
519 spin_unlock(&kvm->mmu_lock);
521 /* we've to flush the tlb before the pages can be freed */
523 kvm_flush_remote_tlbs(kvm);
526 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
527 struct mm_struct *mm,
531 struct kvm *kvm = mmu_notifier_to_kvm(mn);
533 spin_lock(&kvm->mmu_lock);
535 * This sequence increase will notify the kvm page fault that
536 * the page that is going to be mapped in the spte could have
539 kvm->mmu_notifier_seq++;
541 * The above sequence increase must be visible before the
542 * below count decrease but both values are read by the kvm
543 * page fault under mmu_lock spinlock so we don't need to add
544 * a smb_wmb() here in between the two.
546 kvm->mmu_notifier_count--;
547 spin_unlock(&kvm->mmu_lock);
549 BUG_ON(kvm->mmu_notifier_count < 0);
552 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
553 struct mm_struct *mm,
554 unsigned long address)
556 struct kvm *kvm = mmu_notifier_to_kvm(mn);
559 spin_lock(&kvm->mmu_lock);
560 young = kvm_age_hva(kvm, address);
561 spin_unlock(&kvm->mmu_lock);
564 kvm_flush_remote_tlbs(kvm);
569 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
570 .invalidate_page = kvm_mmu_notifier_invalidate_page,
571 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
572 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
573 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
575 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
577 static struct kvm *kvm_create_vm(void)
579 struct kvm *kvm = kvm_arch_create_vm();
580 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
587 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
588 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
591 return ERR_PTR(-ENOMEM);
593 kvm->coalesced_mmio_ring =
594 (struct kvm_coalesced_mmio_ring *)page_address(page);
597 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
600 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
601 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
603 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
612 kvm->mm = current->mm;
613 atomic_inc(&kvm->mm->mm_count);
614 spin_lock_init(&kvm->mmu_lock);
615 kvm_io_bus_init(&kvm->pio_bus);
616 mutex_init(&kvm->lock);
617 kvm_io_bus_init(&kvm->mmio_bus);
618 init_rwsem(&kvm->slots_lock);
619 atomic_set(&kvm->users_count, 1);
620 spin_lock(&kvm_lock);
621 list_add(&kvm->vm_list, &vm_list);
622 spin_unlock(&kvm_lock);
623 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
624 kvm_coalesced_mmio_init(kvm);
631 * Free any memory in @free but not in @dont.
633 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
634 struct kvm_memory_slot *dont)
636 if (!dont || free->rmap != dont->rmap)
639 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
640 vfree(free->dirty_bitmap);
642 if (!dont || free->lpage_info != dont->lpage_info)
643 vfree(free->lpage_info);
646 free->dirty_bitmap = NULL;
648 free->lpage_info = NULL;
651 void kvm_free_physmem(struct kvm *kvm)
655 for (i = 0; i < kvm->nmemslots; ++i)
656 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
659 static void kvm_destroy_vm(struct kvm *kvm)
661 struct mm_struct *mm = kvm->mm;
663 spin_lock(&kvm_lock);
664 list_del(&kvm->vm_list);
665 spin_unlock(&kvm_lock);
666 kvm_io_bus_destroy(&kvm->pio_bus);
667 kvm_io_bus_destroy(&kvm->mmio_bus);
668 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
669 if (kvm->coalesced_mmio_ring != NULL)
670 free_page((unsigned long)kvm->coalesced_mmio_ring);
672 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
673 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
675 kvm_arch_destroy_vm(kvm);
679 void kvm_get_kvm(struct kvm *kvm)
681 atomic_inc(&kvm->users_count);
683 EXPORT_SYMBOL_GPL(kvm_get_kvm);
685 void kvm_put_kvm(struct kvm *kvm)
687 if (atomic_dec_and_test(&kvm->users_count))
690 EXPORT_SYMBOL_GPL(kvm_put_kvm);
693 static int kvm_vm_release(struct inode *inode, struct file *filp)
695 struct kvm *kvm = filp->private_data;
702 * Allocate some memory and give it an address in the guest physical address
705 * Discontiguous memory is allowed, mostly for framebuffers.
707 * Must be called holding mmap_sem for write.
709 int __kvm_set_memory_region(struct kvm *kvm,
710 struct kvm_userspace_memory_region *mem,
715 unsigned long npages;
717 struct kvm_memory_slot *memslot;
718 struct kvm_memory_slot old, new;
721 /* General sanity checks */
722 if (mem->memory_size & (PAGE_SIZE - 1))
724 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
726 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
728 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
730 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
733 memslot = &kvm->memslots[mem->slot];
734 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
735 npages = mem->memory_size >> PAGE_SHIFT;
738 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
740 new = old = *memslot;
742 new.base_gfn = base_gfn;
744 new.flags = mem->flags;
746 /* Disallow changing a memory slot's size. */
748 if (npages && old.npages && npages != old.npages)
751 /* Check for overlaps */
753 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
754 struct kvm_memory_slot *s = &kvm->memslots[i];
758 if (!((base_gfn + npages <= s->base_gfn) ||
759 (base_gfn >= s->base_gfn + s->npages)))
763 /* Free page dirty bitmap if unneeded */
764 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
765 new.dirty_bitmap = NULL;
769 /* Allocate if a slot is being created */
771 if (npages && !new.rmap) {
772 new.rmap = vmalloc(npages * sizeof(struct page *));
777 memset(new.rmap, 0, npages * sizeof(*new.rmap));
779 new.user_alloc = user_alloc;
781 * hva_to_rmmap() serialzies with the mmu_lock and to be
782 * safe it has to ignore memslots with !user_alloc &&
786 new.userspace_addr = mem->userspace_addr;
788 new.userspace_addr = 0;
790 if (npages && !new.lpage_info) {
791 int largepages = npages / KVM_PAGES_PER_HPAGE;
792 if (npages % KVM_PAGES_PER_HPAGE)
794 if (base_gfn % KVM_PAGES_PER_HPAGE)
797 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
802 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
804 if (base_gfn % KVM_PAGES_PER_HPAGE)
805 new.lpage_info[0].write_count = 1;
806 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
807 new.lpage_info[largepages-1].write_count = 1;
810 /* Allocate page dirty bitmap if needed */
811 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
812 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
814 new.dirty_bitmap = vmalloc(dirty_bytes);
815 if (!new.dirty_bitmap)
817 memset(new.dirty_bitmap, 0, dirty_bytes);
819 #endif /* not defined CONFIG_S390 */
822 kvm_arch_flush_shadow(kvm);
824 spin_lock(&kvm->mmu_lock);
825 if (mem->slot >= kvm->nmemslots)
826 kvm->nmemslots = mem->slot + 1;
829 spin_unlock(&kvm->mmu_lock);
831 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
833 spin_lock(&kvm->mmu_lock);
835 spin_unlock(&kvm->mmu_lock);
839 kvm_free_physmem_slot(&old, &new);
841 /* map the pages in iommu page table */
842 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
849 kvm_free_physmem_slot(&new, &old);
854 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
856 int kvm_set_memory_region(struct kvm *kvm,
857 struct kvm_userspace_memory_region *mem,
862 down_write(&kvm->slots_lock);
863 r = __kvm_set_memory_region(kvm, mem, user_alloc);
864 up_write(&kvm->slots_lock);
867 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
869 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
871 kvm_userspace_memory_region *mem,
874 if (mem->slot >= KVM_MEMORY_SLOTS)
876 return kvm_set_memory_region(kvm, mem, user_alloc);
879 int kvm_get_dirty_log(struct kvm *kvm,
880 struct kvm_dirty_log *log, int *is_dirty)
882 struct kvm_memory_slot *memslot;
885 unsigned long any = 0;
888 if (log->slot >= KVM_MEMORY_SLOTS)
891 memslot = &kvm->memslots[log->slot];
893 if (!memslot->dirty_bitmap)
896 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
898 for (i = 0; !any && i < n/sizeof(long); ++i)
899 any = memslot->dirty_bitmap[i];
902 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
913 int is_error_page(struct page *page)
915 return page == bad_page;
917 EXPORT_SYMBOL_GPL(is_error_page);
919 int is_error_pfn(pfn_t pfn)
921 return pfn == bad_pfn;
923 EXPORT_SYMBOL_GPL(is_error_pfn);
925 static inline unsigned long bad_hva(void)
930 int kvm_is_error_hva(unsigned long addr)
932 return addr == bad_hva();
934 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
936 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
940 for (i = 0; i < kvm->nmemslots; ++i) {
941 struct kvm_memory_slot *memslot = &kvm->memslots[i];
943 if (gfn >= memslot->base_gfn
944 && gfn < memslot->base_gfn + memslot->npages)
949 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
951 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
953 gfn = unalias_gfn(kvm, gfn);
954 return gfn_to_memslot_unaliased(kvm, gfn);
957 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
961 gfn = unalias_gfn(kvm, gfn);
962 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
963 struct kvm_memory_slot *memslot = &kvm->memslots[i];
965 if (gfn >= memslot->base_gfn
966 && gfn < memslot->base_gfn + memslot->npages)
971 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
973 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
975 struct kvm_memory_slot *slot;
977 gfn = unalias_gfn(kvm, gfn);
978 slot = gfn_to_memslot_unaliased(kvm, gfn);
981 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
983 EXPORT_SYMBOL_GPL(gfn_to_hva);
985 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
987 struct page *page[1];
994 addr = gfn_to_hva(kvm, gfn);
995 if (kvm_is_error_hva(addr)) {
997 return page_to_pfn(bad_page);
1000 npages = get_user_pages_fast(addr, 1, 1, page);
1002 if (unlikely(npages != 1)) {
1003 struct vm_area_struct *vma;
1005 down_read(¤t->mm->mmap_sem);
1006 vma = find_vma(current->mm, addr);
1008 if (vma == NULL || addr < vma->vm_start ||
1009 !(vma->vm_flags & VM_PFNMAP)) {
1010 up_read(¤t->mm->mmap_sem);
1012 return page_to_pfn(bad_page);
1015 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1016 up_read(¤t->mm->mmap_sem);
1017 BUG_ON(!kvm_is_mmio_pfn(pfn));
1019 pfn = page_to_pfn(page[0]);
1024 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1026 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1030 pfn = gfn_to_pfn(kvm, gfn);
1031 if (!kvm_is_mmio_pfn(pfn))
1032 return pfn_to_page(pfn);
1034 WARN_ON(kvm_is_mmio_pfn(pfn));
1040 EXPORT_SYMBOL_GPL(gfn_to_page);
1042 void kvm_release_page_clean(struct page *page)
1044 kvm_release_pfn_clean(page_to_pfn(page));
1046 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1048 void kvm_release_pfn_clean(pfn_t pfn)
1050 if (!kvm_is_mmio_pfn(pfn))
1051 put_page(pfn_to_page(pfn));
1053 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1055 void kvm_release_page_dirty(struct page *page)
1057 kvm_release_pfn_dirty(page_to_pfn(page));
1059 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1061 void kvm_release_pfn_dirty(pfn_t pfn)
1063 kvm_set_pfn_dirty(pfn);
1064 kvm_release_pfn_clean(pfn);
1066 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1068 void kvm_set_page_dirty(struct page *page)
1070 kvm_set_pfn_dirty(page_to_pfn(page));
1072 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1074 void kvm_set_pfn_dirty(pfn_t pfn)
1076 if (!kvm_is_mmio_pfn(pfn)) {
1077 struct page *page = pfn_to_page(pfn);
1078 if (!PageReserved(page))
1082 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1084 void kvm_set_pfn_accessed(pfn_t pfn)
1086 if (!kvm_is_mmio_pfn(pfn))
1087 mark_page_accessed(pfn_to_page(pfn));
1089 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1091 void kvm_get_pfn(pfn_t pfn)
1093 if (!kvm_is_mmio_pfn(pfn))
1094 get_page(pfn_to_page(pfn));
1096 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1098 static int next_segment(unsigned long len, int offset)
1100 if (len > PAGE_SIZE - offset)
1101 return PAGE_SIZE - offset;
1106 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1112 addr = gfn_to_hva(kvm, gfn);
1113 if (kvm_is_error_hva(addr))
1115 r = copy_from_user(data, (void __user *)addr + offset, len);
1120 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1122 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1124 gfn_t gfn = gpa >> PAGE_SHIFT;
1126 int offset = offset_in_page(gpa);
1129 while ((seg = next_segment(len, offset)) != 0) {
1130 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1140 EXPORT_SYMBOL_GPL(kvm_read_guest);
1142 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1147 gfn_t gfn = gpa >> PAGE_SHIFT;
1148 int offset = offset_in_page(gpa);
1150 addr = gfn_to_hva(kvm, gfn);
1151 if (kvm_is_error_hva(addr))
1153 pagefault_disable();
1154 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1160 EXPORT_SYMBOL(kvm_read_guest_atomic);
1162 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1163 int offset, int len)
1168 addr = gfn_to_hva(kvm, gfn);
1169 if (kvm_is_error_hva(addr))
1171 r = copy_to_user((void __user *)addr + offset, data, len);
1174 mark_page_dirty(kvm, gfn);
1177 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1179 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1182 gfn_t gfn = gpa >> PAGE_SHIFT;
1184 int offset = offset_in_page(gpa);
1187 while ((seg = next_segment(len, offset)) != 0) {
1188 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1199 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1201 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1203 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1205 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1207 gfn_t gfn = gpa >> PAGE_SHIFT;
1209 int offset = offset_in_page(gpa);
1212 while ((seg = next_segment(len, offset)) != 0) {
1213 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1222 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1224 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1226 struct kvm_memory_slot *memslot;
1228 gfn = unalias_gfn(kvm, gfn);
1229 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1230 if (memslot && memslot->dirty_bitmap) {
1231 unsigned long rel_gfn = gfn - memslot->base_gfn;
1234 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1235 set_bit(rel_gfn, memslot->dirty_bitmap);
1240 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1242 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1247 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1249 if (kvm_cpu_has_interrupt(vcpu) ||
1250 kvm_cpu_has_pending_timer(vcpu) ||
1251 kvm_arch_vcpu_runnable(vcpu)) {
1252 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1255 if (signal_pending(current))
1263 finish_wait(&vcpu->wq, &wait);
1266 void kvm_resched(struct kvm_vcpu *vcpu)
1268 if (!need_resched())
1272 EXPORT_SYMBOL_GPL(kvm_resched);
1274 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1276 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1279 if (vmf->pgoff == 0)
1280 page = virt_to_page(vcpu->run);
1282 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1283 page = virt_to_page(vcpu->arch.pio_data);
1285 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1286 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1287 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1290 return VM_FAULT_SIGBUS;
1296 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1297 .fault = kvm_vcpu_fault,
1300 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1302 vma->vm_ops = &kvm_vcpu_vm_ops;
1306 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1308 struct kvm_vcpu *vcpu = filp->private_data;
1310 kvm_put_kvm(vcpu->kvm);
1314 static const struct file_operations kvm_vcpu_fops = {
1315 .release = kvm_vcpu_release,
1316 .unlocked_ioctl = kvm_vcpu_ioctl,
1317 .compat_ioctl = kvm_vcpu_ioctl,
1318 .mmap = kvm_vcpu_mmap,
1322 * Allocates an inode for the vcpu.
1324 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1326 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1328 kvm_put_kvm(vcpu->kvm);
1333 * Creates some virtual cpus. Good luck creating more than one.
1335 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1338 struct kvm_vcpu *vcpu;
1343 vcpu = kvm_arch_vcpu_create(kvm, n);
1345 return PTR_ERR(vcpu);
1347 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1349 r = kvm_arch_vcpu_setup(vcpu);
1353 mutex_lock(&kvm->lock);
1354 if (kvm->vcpus[n]) {
1358 kvm->vcpus[n] = vcpu;
1359 mutex_unlock(&kvm->lock);
1361 /* Now it's all set up, let userspace reach it */
1363 r = create_vcpu_fd(vcpu);
1369 mutex_lock(&kvm->lock);
1370 kvm->vcpus[n] = NULL;
1372 mutex_unlock(&kvm->lock);
1373 kvm_arch_vcpu_destroy(vcpu);
1377 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1380 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1381 vcpu->sigset_active = 1;
1382 vcpu->sigset = *sigset;
1384 vcpu->sigset_active = 0;
1388 static long kvm_vcpu_ioctl(struct file *filp,
1389 unsigned int ioctl, unsigned long arg)
1391 struct kvm_vcpu *vcpu = filp->private_data;
1392 void __user *argp = (void __user *)arg;
1394 struct kvm_fpu *fpu = NULL;
1395 struct kvm_sregs *kvm_sregs = NULL;
1397 if (vcpu->kvm->mm != current->mm)
1404 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1406 case KVM_GET_REGS: {
1407 struct kvm_regs *kvm_regs;
1410 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1413 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1417 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1424 case KVM_SET_REGS: {
1425 struct kvm_regs *kvm_regs;
1428 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1432 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1434 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1442 case KVM_GET_SREGS: {
1443 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1447 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1451 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1456 case KVM_SET_SREGS: {
1457 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1462 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1464 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1470 case KVM_GET_MP_STATE: {
1471 struct kvm_mp_state mp_state;
1473 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1477 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1482 case KVM_SET_MP_STATE: {
1483 struct kvm_mp_state mp_state;
1486 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1488 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1494 case KVM_TRANSLATE: {
1495 struct kvm_translation tr;
1498 if (copy_from_user(&tr, argp, sizeof tr))
1500 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1504 if (copy_to_user(argp, &tr, sizeof tr))
1509 case KVM_DEBUG_GUEST: {
1510 struct kvm_debug_guest dbg;
1513 if (copy_from_user(&dbg, argp, sizeof dbg))
1515 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1521 case KVM_SET_SIGNAL_MASK: {
1522 struct kvm_signal_mask __user *sigmask_arg = argp;
1523 struct kvm_signal_mask kvm_sigmask;
1524 sigset_t sigset, *p;
1529 if (copy_from_user(&kvm_sigmask, argp,
1530 sizeof kvm_sigmask))
1533 if (kvm_sigmask.len != sizeof sigset)
1536 if (copy_from_user(&sigset, sigmask_arg->sigset,
1541 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1545 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1549 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1553 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1559 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1564 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1566 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1573 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1581 static long kvm_vm_ioctl(struct file *filp,
1582 unsigned int ioctl, unsigned long arg)
1584 struct kvm *kvm = filp->private_data;
1585 void __user *argp = (void __user *)arg;
1588 if (kvm->mm != current->mm)
1591 case KVM_CREATE_VCPU:
1592 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1596 case KVM_SET_USER_MEMORY_REGION: {
1597 struct kvm_userspace_memory_region kvm_userspace_mem;
1600 if (copy_from_user(&kvm_userspace_mem, argp,
1601 sizeof kvm_userspace_mem))
1604 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1609 case KVM_GET_DIRTY_LOG: {
1610 struct kvm_dirty_log log;
1613 if (copy_from_user(&log, argp, sizeof log))
1615 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1620 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1621 case KVM_REGISTER_COALESCED_MMIO: {
1622 struct kvm_coalesced_mmio_zone zone;
1624 if (copy_from_user(&zone, argp, sizeof zone))
1627 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1633 case KVM_UNREGISTER_COALESCED_MMIO: {
1634 struct kvm_coalesced_mmio_zone zone;
1636 if (copy_from_user(&zone, argp, sizeof zone))
1639 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1646 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1647 case KVM_ASSIGN_PCI_DEVICE: {
1648 struct kvm_assigned_pci_dev assigned_dev;
1651 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1653 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1658 case KVM_ASSIGN_IRQ: {
1659 struct kvm_assigned_irq assigned_irq;
1662 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1664 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1671 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1677 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1679 struct page *page[1];
1682 gfn_t gfn = vmf->pgoff;
1683 struct kvm *kvm = vma->vm_file->private_data;
1685 addr = gfn_to_hva(kvm, gfn);
1686 if (kvm_is_error_hva(addr))
1687 return VM_FAULT_SIGBUS;
1689 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1691 if (unlikely(npages != 1))
1692 return VM_FAULT_SIGBUS;
1694 vmf->page = page[0];
1698 static struct vm_operations_struct kvm_vm_vm_ops = {
1699 .fault = kvm_vm_fault,
1702 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1704 vma->vm_ops = &kvm_vm_vm_ops;
1708 static const struct file_operations kvm_vm_fops = {
1709 .release = kvm_vm_release,
1710 .unlocked_ioctl = kvm_vm_ioctl,
1711 .compat_ioctl = kvm_vm_ioctl,
1712 .mmap = kvm_vm_mmap,
1715 static int kvm_dev_ioctl_create_vm(void)
1720 kvm = kvm_create_vm();
1722 return PTR_ERR(kvm);
1723 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1730 static long kvm_dev_ioctl(struct file *filp,
1731 unsigned int ioctl, unsigned long arg)
1736 case KVM_GET_API_VERSION:
1740 r = KVM_API_VERSION;
1746 r = kvm_dev_ioctl_create_vm();
1748 case KVM_CHECK_EXTENSION:
1749 r = kvm_dev_ioctl_check_extension(arg);
1751 case KVM_GET_VCPU_MMAP_SIZE:
1755 r = PAGE_SIZE; /* struct kvm_run */
1757 r += PAGE_SIZE; /* pio data page */
1759 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1760 r += PAGE_SIZE; /* coalesced mmio ring page */
1763 case KVM_TRACE_ENABLE:
1764 case KVM_TRACE_PAUSE:
1765 case KVM_TRACE_DISABLE:
1766 r = kvm_trace_ioctl(ioctl, arg);
1769 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1775 static struct file_operations kvm_chardev_ops = {
1776 .unlocked_ioctl = kvm_dev_ioctl,
1777 .compat_ioctl = kvm_dev_ioctl,
1780 static struct miscdevice kvm_dev = {
1786 static void hardware_enable(void *junk)
1788 int cpu = raw_smp_processor_id();
1790 if (cpu_isset(cpu, cpus_hardware_enabled))
1792 cpu_set(cpu, cpus_hardware_enabled);
1793 kvm_arch_hardware_enable(NULL);
1796 static void hardware_disable(void *junk)
1798 int cpu = raw_smp_processor_id();
1800 if (!cpu_isset(cpu, cpus_hardware_enabled))
1802 cpu_clear(cpu, cpus_hardware_enabled);
1803 kvm_arch_hardware_disable(NULL);
1806 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1811 val &= ~CPU_TASKS_FROZEN;
1814 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1816 hardware_disable(NULL);
1818 case CPU_UP_CANCELED:
1819 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1821 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1824 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1826 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1833 asmlinkage void kvm_handle_fault_on_reboot(void)
1836 /* spin while reset goes on */
1839 /* Fault while not rebooting. We want the trace. */
1842 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1844 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1847 if (val == SYS_RESTART) {
1849 * Some (well, at least mine) BIOSes hang on reboot if
1852 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1853 kvm_rebooting = true;
1854 on_each_cpu(hardware_disable, NULL, 1);
1859 static struct notifier_block kvm_reboot_notifier = {
1860 .notifier_call = kvm_reboot,
1864 void kvm_io_bus_init(struct kvm_io_bus *bus)
1866 memset(bus, 0, sizeof(*bus));
1869 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1873 for (i = 0; i < bus->dev_count; i++) {
1874 struct kvm_io_device *pos = bus->devs[i];
1876 kvm_iodevice_destructor(pos);
1880 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1881 gpa_t addr, int len, int is_write)
1885 for (i = 0; i < bus->dev_count; i++) {
1886 struct kvm_io_device *pos = bus->devs[i];
1888 if (pos->in_range(pos, addr, len, is_write))
1895 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1897 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1899 bus->devs[bus->dev_count++] = dev;
1902 static struct notifier_block kvm_cpu_notifier = {
1903 .notifier_call = kvm_cpu_hotplug,
1904 .priority = 20, /* must be > scheduler priority */
1907 static int vm_stat_get(void *_offset, u64 *val)
1909 unsigned offset = (long)_offset;
1913 spin_lock(&kvm_lock);
1914 list_for_each_entry(kvm, &vm_list, vm_list)
1915 *val += *(u32 *)((void *)kvm + offset);
1916 spin_unlock(&kvm_lock);
1920 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1922 static int vcpu_stat_get(void *_offset, u64 *val)
1924 unsigned offset = (long)_offset;
1926 struct kvm_vcpu *vcpu;
1930 spin_lock(&kvm_lock);
1931 list_for_each_entry(kvm, &vm_list, vm_list)
1932 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1933 vcpu = kvm->vcpus[i];
1935 *val += *(u32 *)((void *)vcpu + offset);
1937 spin_unlock(&kvm_lock);
1941 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1943 static struct file_operations *stat_fops[] = {
1944 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1945 [KVM_STAT_VM] = &vm_stat_fops,
1948 static void kvm_init_debug(void)
1950 struct kvm_stats_debugfs_item *p;
1952 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1953 for (p = debugfs_entries; p->name; ++p)
1954 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1955 (void *)(long)p->offset,
1956 stat_fops[p->kind]);
1959 static void kvm_exit_debug(void)
1961 struct kvm_stats_debugfs_item *p;
1963 for (p = debugfs_entries; p->name; ++p)
1964 debugfs_remove(p->dentry);
1965 debugfs_remove(kvm_debugfs_dir);
1968 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1970 hardware_disable(NULL);
1974 static int kvm_resume(struct sys_device *dev)
1976 hardware_enable(NULL);
1980 static struct sysdev_class kvm_sysdev_class = {
1982 .suspend = kvm_suspend,
1983 .resume = kvm_resume,
1986 static struct sys_device kvm_sysdev = {
1988 .cls = &kvm_sysdev_class,
1991 struct page *bad_page;
1995 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1997 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2000 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2002 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2004 kvm_arch_vcpu_load(vcpu, cpu);
2007 static void kvm_sched_out(struct preempt_notifier *pn,
2008 struct task_struct *next)
2010 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2012 kvm_arch_vcpu_put(vcpu);
2015 int kvm_init(void *opaque, unsigned int vcpu_size,
2016 struct module *module)
2023 r = kvm_arch_init(opaque);
2027 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2029 if (bad_page == NULL) {
2034 bad_pfn = page_to_pfn(bad_page);
2036 r = kvm_arch_hardware_setup();
2040 for_each_online_cpu(cpu) {
2041 smp_call_function_single(cpu,
2042 kvm_arch_check_processor_compat,
2048 on_each_cpu(hardware_enable, NULL, 1);
2049 r = register_cpu_notifier(&kvm_cpu_notifier);
2052 register_reboot_notifier(&kvm_reboot_notifier);
2054 r = sysdev_class_register(&kvm_sysdev_class);
2058 r = sysdev_register(&kvm_sysdev);
2062 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2063 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2064 __alignof__(struct kvm_vcpu),
2066 if (!kvm_vcpu_cache) {
2071 kvm_chardev_ops.owner = module;
2073 r = misc_register(&kvm_dev);
2075 printk(KERN_ERR "kvm: misc device register failed\n");
2079 kvm_preempt_ops.sched_in = kvm_sched_in;
2080 kvm_preempt_ops.sched_out = kvm_sched_out;
2085 kmem_cache_destroy(kvm_vcpu_cache);
2087 sysdev_unregister(&kvm_sysdev);
2089 sysdev_class_unregister(&kvm_sysdev_class);
2091 unregister_reboot_notifier(&kvm_reboot_notifier);
2092 unregister_cpu_notifier(&kvm_cpu_notifier);
2094 on_each_cpu(hardware_disable, NULL, 1);
2096 kvm_arch_hardware_unsetup();
2098 __free_page(bad_page);
2105 EXPORT_SYMBOL_GPL(kvm_init);
2109 kvm_trace_cleanup();
2110 misc_deregister(&kvm_dev);
2111 kmem_cache_destroy(kvm_vcpu_cache);
2112 sysdev_unregister(&kvm_sysdev);
2113 sysdev_class_unregister(&kvm_sysdev_class);
2114 unregister_reboot_notifier(&kvm_reboot_notifier);
2115 unregister_cpu_notifier(&kvm_cpu_notifier);
2116 on_each_cpu(hardware_disable, NULL, 1);
2117 kvm_arch_hardware_unsetup();
2120 __free_page(bad_page);
2122 EXPORT_SYMBOL_GPL(kvm_exit);