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>
51 #include <asm/msidef.h>
54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
55 #include "coalesced_mmio.h"
58 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
59 #include <linux/pci.h>
60 #include <linux/interrupt.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
67 static int msi2intx = 1;
68 module_param(msi2intx, bool, 0);
70 DEFINE_SPINLOCK(kvm_lock);
73 static cpumask_var_t cpus_hardware_enabled;
75 struct kmem_cache *kvm_vcpu_cache;
76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
78 static __read_mostly struct preempt_ops kvm_preempt_ops;
80 struct dentry *kvm_debugfs_dir;
82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
85 static bool kvm_rebooting;
87 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
90 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev)
93 struct kvm_vcpu *vcpu;
94 struct kvm_ioapic *ioapic = ioapic_irqchip(dev->kvm);
95 int dest_id = (dev->guest_msi.address_lo & MSI_ADDR_DEST_ID_MASK)
96 >> MSI_ADDR_DEST_ID_SHIFT;
97 int vector = (dev->guest_msi.data & MSI_DATA_VECTOR_MASK)
98 >> MSI_DATA_VECTOR_SHIFT;
99 int dest_mode = test_bit(MSI_ADDR_DEST_MODE_SHIFT,
100 (unsigned long *)&dev->guest_msi.address_lo);
101 int trig_mode = test_bit(MSI_DATA_TRIGGER_SHIFT,
102 (unsigned long *)&dev->guest_msi.data);
103 int delivery_mode = test_bit(MSI_DATA_DELIVERY_MODE_SHIFT,
104 (unsigned long *)&dev->guest_msi.data);
109 deliver_bitmask = kvm_ioapic_get_delivery_bitmask(ioapic,
111 /* IOAPIC delivery mode value is the same as MSI here */
112 switch (delivery_mode) {
113 case IOAPIC_LOWEST_PRIORITY:
114 vcpu = kvm_get_lowest_prio_vcpu(ioapic->kvm, vector,
117 kvm_apic_set_irq(vcpu, vector, trig_mode);
119 printk(KERN_INFO "kvm: null lowest priority vcpu!\n");
122 for (vcpu_id = 0; deliver_bitmask != 0; vcpu_id++) {
123 if (!(deliver_bitmask & (1 << vcpu_id)))
125 deliver_bitmask &= ~(1 << vcpu_id);
126 vcpu = ioapic->kvm->vcpus[vcpu_id];
128 kvm_apic_set_irq(vcpu, vector, trig_mode);
132 printk(KERN_INFO "kvm: unsupported MSI delivery mode\n");
136 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev) {}
139 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
142 struct list_head *ptr;
143 struct kvm_assigned_dev_kernel *match;
145 list_for_each(ptr, head) {
146 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
147 if (match->assigned_dev_id == assigned_dev_id)
153 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
155 struct kvm_assigned_dev_kernel *assigned_dev;
157 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
160 /* This is taken to safely inject irq inside the guest. When
161 * the interrupt injection (or the ioapic code) uses a
162 * finer-grained lock, update this
164 mutex_lock(&assigned_dev->kvm->lock);
165 if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_INTX)
166 kvm_set_irq(assigned_dev->kvm,
167 assigned_dev->irq_source_id,
168 assigned_dev->guest_irq, 1);
169 else if (assigned_dev->irq_requested_type &
170 KVM_ASSIGNED_DEV_GUEST_MSI) {
171 assigned_device_msi_dispatch(assigned_dev);
172 enable_irq(assigned_dev->host_irq);
173 assigned_dev->host_irq_disabled = false;
175 mutex_unlock(&assigned_dev->kvm->lock);
176 kvm_put_kvm(assigned_dev->kvm);
179 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
181 struct kvm_assigned_dev_kernel *assigned_dev =
182 (struct kvm_assigned_dev_kernel *) dev_id;
184 kvm_get_kvm(assigned_dev->kvm);
186 schedule_work(&assigned_dev->interrupt_work);
188 disable_irq_nosync(irq);
189 assigned_dev->host_irq_disabled = true;
194 /* Ack the irq line for an assigned device */
195 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
197 struct kvm_assigned_dev_kernel *dev;
202 dev = container_of(kian, struct kvm_assigned_dev_kernel,
205 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
207 /* The guest irq may be shared so this ack may be
208 * from another device.
210 if (dev->host_irq_disabled) {
211 enable_irq(dev->host_irq);
212 dev->host_irq_disabled = false;
216 static void kvm_free_assigned_irq(struct kvm *kvm,
217 struct kvm_assigned_dev_kernel *assigned_dev)
219 if (!irqchip_in_kernel(kvm))
222 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
224 if (assigned_dev->irq_source_id != -1)
225 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
226 assigned_dev->irq_source_id = -1;
228 if (!assigned_dev->irq_requested_type)
231 if (cancel_work_sync(&assigned_dev->interrupt_work))
232 /* We had pending work. That means we will have to take
233 * care of kvm_put_kvm.
237 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
239 if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
240 pci_disable_msi(assigned_dev->dev);
242 assigned_dev->irq_requested_type = 0;
246 static void kvm_free_assigned_device(struct kvm *kvm,
247 struct kvm_assigned_dev_kernel
250 kvm_free_assigned_irq(kvm, assigned_dev);
252 pci_reset_function(assigned_dev->dev);
254 pci_release_regions(assigned_dev->dev);
255 pci_disable_device(assigned_dev->dev);
256 pci_dev_put(assigned_dev->dev);
258 list_del(&assigned_dev->list);
262 void kvm_free_all_assigned_devices(struct kvm *kvm)
264 struct list_head *ptr, *ptr2;
265 struct kvm_assigned_dev_kernel *assigned_dev;
267 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
268 assigned_dev = list_entry(ptr,
269 struct kvm_assigned_dev_kernel,
272 kvm_free_assigned_device(kvm, assigned_dev);
276 static int assigned_device_update_intx(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *adev,
278 struct kvm_assigned_irq *airq)
280 adev->guest_irq = airq->guest_irq;
281 adev->ack_notifier.gsi = airq->guest_irq;
283 if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
286 if (irqchip_in_kernel(kvm)) {
288 adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) {
289 free_irq(adev->host_irq, (void *)kvm);
290 pci_disable_msi(adev->dev);
293 if (!capable(CAP_SYS_RAWIO))
297 adev->host_irq = airq->host_irq;
299 adev->host_irq = adev->dev->irq;
301 /* Even though this is PCI, we don't want to use shared
302 * interrupts. Sharing host devices with guest-assigned devices
303 * on the same interrupt line is not a happy situation: there
304 * are going to be long delays in accepting, acking, etc.
306 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
307 0, "kvm_assigned_intx_device", (void *)adev))
311 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
312 KVM_ASSIGNED_DEV_HOST_INTX;
317 static int assigned_device_update_msi(struct kvm *kvm,
318 struct kvm_assigned_dev_kernel *adev,
319 struct kvm_assigned_irq *airq)
323 if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
324 /* x86 don't care upper address of guest msi message addr */
325 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
326 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
327 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
328 adev->guest_msi.data = airq->guest_msi.data;
329 adev->ack_notifier.gsi = -1;
330 } else if (msi2intx) {
331 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
332 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
333 adev->guest_irq = airq->guest_irq;
334 adev->ack_notifier.gsi = airq->guest_irq;
337 if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
340 if (irqchip_in_kernel(kvm)) {
342 if (adev->irq_requested_type &
343 KVM_ASSIGNED_DEV_HOST_INTX)
344 free_irq(adev->host_irq, (void *)adev);
346 r = pci_enable_msi(adev->dev);
351 adev->host_irq = adev->dev->irq;
352 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
353 "kvm_assigned_msi_device", (void *)adev))
358 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
360 adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
365 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
366 struct kvm_assigned_irq
370 struct kvm_assigned_dev_kernel *match;
372 mutex_lock(&kvm->lock);
374 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
375 assigned_irq->assigned_dev_id);
377 mutex_unlock(&kvm->lock);
381 if (!match->irq_requested_type) {
382 INIT_WORK(&match->interrupt_work,
383 kvm_assigned_dev_interrupt_work_handler);
384 if (irqchip_in_kernel(kvm)) {
385 /* Register ack nofitier */
386 match->ack_notifier.gsi = -1;
387 match->ack_notifier.irq_acked =
388 kvm_assigned_dev_ack_irq;
389 kvm_register_irq_ack_notifier(kvm,
390 &match->ack_notifier);
392 /* Request IRQ source ID */
393 r = kvm_request_irq_source_id(kvm);
397 match->irq_source_id = r;
400 /* Determine host device irq type, we can know the
401 * result from dev->msi_enabled */
403 pci_enable_msi(match->dev);
409 (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
410 (msi2intx && match->dev->msi_enabled)) {
412 r = assigned_device_update_msi(kvm, match, assigned_irq);
414 printk(KERN_WARNING "kvm: failed to enable "
421 } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
422 /* Host device IRQ 0 means don't support INTx */
425 "kvm: wait device to enable MSI!\n");
429 "kvm: failed to enable MSI device!\n");
434 /* Non-sharing INTx mode */
435 r = assigned_device_update_intx(kvm, match, assigned_irq);
437 printk(KERN_WARNING "kvm: failed to enable "
443 mutex_unlock(&kvm->lock);
446 mutex_unlock(&kvm->lock);
447 kvm_free_assigned_device(kvm, match);
451 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
452 struct kvm_assigned_pci_dev *assigned_dev)
455 struct kvm_assigned_dev_kernel *match;
458 mutex_lock(&kvm->lock);
460 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
461 assigned_dev->assigned_dev_id);
463 /* device already assigned */
468 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
470 printk(KERN_INFO "%s: Couldn't allocate memory\n",
475 dev = pci_get_bus_and_slot(assigned_dev->busnr,
476 assigned_dev->devfn);
478 printk(KERN_INFO "%s: host device not found\n", __func__);
482 if (pci_enable_device(dev)) {
483 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
487 r = pci_request_regions(dev, "kvm_assigned_device");
489 printk(KERN_INFO "%s: Could not get access to device regions\n",
494 pci_reset_function(dev);
496 match->assigned_dev_id = assigned_dev->assigned_dev_id;
497 match->host_busnr = assigned_dev->busnr;
498 match->host_devfn = assigned_dev->devfn;
500 match->irq_source_id = -1;
503 list_add(&match->list, &kvm->arch.assigned_dev_head);
505 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
506 r = kvm_iommu_map_guest(kvm, match);
512 mutex_unlock(&kvm->lock);
515 list_del(&match->list);
516 pci_release_regions(dev);
518 pci_disable_device(dev);
523 mutex_unlock(&kvm->lock);
528 static inline int valid_vcpu(int n)
530 return likely(n >= 0 && n < KVM_MAX_VCPUS);
533 inline int kvm_is_mmio_pfn(pfn_t pfn)
536 return PageReserved(pfn_to_page(pfn));
542 * Switches to specified vcpu, until a matching vcpu_put()
544 void vcpu_load(struct kvm_vcpu *vcpu)
548 mutex_lock(&vcpu->mutex);
550 preempt_notifier_register(&vcpu->preempt_notifier);
551 kvm_arch_vcpu_load(vcpu, cpu);
555 void vcpu_put(struct kvm_vcpu *vcpu)
558 kvm_arch_vcpu_put(vcpu);
559 preempt_notifier_unregister(&vcpu->preempt_notifier);
561 mutex_unlock(&vcpu->mutex);
564 static void ack_flush(void *_completed)
568 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
573 struct kvm_vcpu *vcpu;
575 if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
579 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
580 vcpu = kvm->vcpus[i];
583 if (test_and_set_bit(req, &vcpu->requests))
586 if (cpus != NULL && cpu != -1 && cpu != me)
587 cpumask_set_cpu(cpu, cpus);
589 if (unlikely(cpus == NULL))
590 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
591 else if (!cpumask_empty(cpus))
592 smp_call_function_many(cpus, ack_flush, NULL, 1);
596 free_cpumask_var(cpus);
600 void kvm_flush_remote_tlbs(struct kvm *kvm)
602 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
603 ++kvm->stat.remote_tlb_flush;
606 void kvm_reload_remote_mmus(struct kvm *kvm)
608 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
611 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
616 mutex_init(&vcpu->mutex);
620 init_waitqueue_head(&vcpu->wq);
622 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
627 vcpu->run = page_address(page);
629 r = kvm_arch_vcpu_init(vcpu);
635 free_page((unsigned long)vcpu->run);
639 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
641 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
643 kvm_arch_vcpu_uninit(vcpu);
644 free_page((unsigned long)vcpu->run);
646 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
648 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
649 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
651 return container_of(mn, struct kvm, mmu_notifier);
654 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
655 struct mm_struct *mm,
656 unsigned long address)
658 struct kvm *kvm = mmu_notifier_to_kvm(mn);
662 * When ->invalidate_page runs, the linux pte has been zapped
663 * already but the page is still allocated until
664 * ->invalidate_page returns. So if we increase the sequence
665 * here the kvm page fault will notice if the spte can't be
666 * established because the page is going to be freed. If
667 * instead the kvm page fault establishes the spte before
668 * ->invalidate_page runs, kvm_unmap_hva will release it
671 * The sequence increase only need to be seen at spin_unlock
672 * time, and not at spin_lock time.
674 * Increasing the sequence after the spin_unlock would be
675 * unsafe because the kvm page fault could then establish the
676 * pte after kvm_unmap_hva returned, without noticing the page
677 * is going to be freed.
679 spin_lock(&kvm->mmu_lock);
680 kvm->mmu_notifier_seq++;
681 need_tlb_flush = kvm_unmap_hva(kvm, address);
682 spin_unlock(&kvm->mmu_lock);
684 /* we've to flush the tlb before the pages can be freed */
686 kvm_flush_remote_tlbs(kvm);
690 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
691 struct mm_struct *mm,
695 struct kvm *kvm = mmu_notifier_to_kvm(mn);
696 int need_tlb_flush = 0;
698 spin_lock(&kvm->mmu_lock);
700 * The count increase must become visible at unlock time as no
701 * spte can be established without taking the mmu_lock and
702 * count is also read inside the mmu_lock critical section.
704 kvm->mmu_notifier_count++;
705 for (; start < end; start += PAGE_SIZE)
706 need_tlb_flush |= kvm_unmap_hva(kvm, start);
707 spin_unlock(&kvm->mmu_lock);
709 /* we've to flush the tlb before the pages can be freed */
711 kvm_flush_remote_tlbs(kvm);
714 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
715 struct mm_struct *mm,
719 struct kvm *kvm = mmu_notifier_to_kvm(mn);
721 spin_lock(&kvm->mmu_lock);
723 * This sequence increase will notify the kvm page fault that
724 * the page that is going to be mapped in the spte could have
727 kvm->mmu_notifier_seq++;
729 * The above sequence increase must be visible before the
730 * below count decrease but both values are read by the kvm
731 * page fault under mmu_lock spinlock so we don't need to add
732 * a smb_wmb() here in between the two.
734 kvm->mmu_notifier_count--;
735 spin_unlock(&kvm->mmu_lock);
737 BUG_ON(kvm->mmu_notifier_count < 0);
740 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
741 struct mm_struct *mm,
742 unsigned long address)
744 struct kvm *kvm = mmu_notifier_to_kvm(mn);
747 spin_lock(&kvm->mmu_lock);
748 young = kvm_age_hva(kvm, address);
749 spin_unlock(&kvm->mmu_lock);
752 kvm_flush_remote_tlbs(kvm);
757 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
758 .invalidate_page = kvm_mmu_notifier_invalidate_page,
759 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
760 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
761 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
763 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
765 static struct kvm *kvm_create_vm(void)
767 struct kvm *kvm = kvm_arch_create_vm();
768 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
775 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
776 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
779 return ERR_PTR(-ENOMEM);
781 kvm->coalesced_mmio_ring =
782 (struct kvm_coalesced_mmio_ring *)page_address(page);
785 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
788 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
789 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
791 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
800 kvm->mm = current->mm;
801 atomic_inc(&kvm->mm->mm_count);
802 spin_lock_init(&kvm->mmu_lock);
803 kvm_io_bus_init(&kvm->pio_bus);
804 mutex_init(&kvm->lock);
805 kvm_io_bus_init(&kvm->mmio_bus);
806 init_rwsem(&kvm->slots_lock);
807 atomic_set(&kvm->users_count, 1);
808 spin_lock(&kvm_lock);
809 list_add(&kvm->vm_list, &vm_list);
810 spin_unlock(&kvm_lock);
811 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
812 kvm_coalesced_mmio_init(kvm);
819 * Free any memory in @free but not in @dont.
821 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
822 struct kvm_memory_slot *dont)
824 if (!dont || free->rmap != dont->rmap)
827 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
828 vfree(free->dirty_bitmap);
830 if (!dont || free->lpage_info != dont->lpage_info)
831 vfree(free->lpage_info);
834 free->dirty_bitmap = NULL;
836 free->lpage_info = NULL;
839 void kvm_free_physmem(struct kvm *kvm)
843 for (i = 0; i < kvm->nmemslots; ++i)
844 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
847 static void kvm_destroy_vm(struct kvm *kvm)
849 struct mm_struct *mm = kvm->mm;
851 spin_lock(&kvm_lock);
852 list_del(&kvm->vm_list);
853 spin_unlock(&kvm_lock);
854 kvm_io_bus_destroy(&kvm->pio_bus);
855 kvm_io_bus_destroy(&kvm->mmio_bus);
856 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
857 if (kvm->coalesced_mmio_ring != NULL)
858 free_page((unsigned long)kvm->coalesced_mmio_ring);
860 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
861 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
863 kvm_arch_destroy_vm(kvm);
867 void kvm_get_kvm(struct kvm *kvm)
869 atomic_inc(&kvm->users_count);
871 EXPORT_SYMBOL_GPL(kvm_get_kvm);
873 void kvm_put_kvm(struct kvm *kvm)
875 if (atomic_dec_and_test(&kvm->users_count))
878 EXPORT_SYMBOL_GPL(kvm_put_kvm);
881 static int kvm_vm_release(struct inode *inode, struct file *filp)
883 struct kvm *kvm = filp->private_data;
890 * Allocate some memory and give it an address in the guest physical address
893 * Discontiguous memory is allowed, mostly for framebuffers.
895 * Must be called holding mmap_sem for write.
897 int __kvm_set_memory_region(struct kvm *kvm,
898 struct kvm_userspace_memory_region *mem,
903 unsigned long npages;
905 struct kvm_memory_slot *memslot;
906 struct kvm_memory_slot old, new;
909 /* General sanity checks */
910 if (mem->memory_size & (PAGE_SIZE - 1))
912 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
914 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
916 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
918 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
921 memslot = &kvm->memslots[mem->slot];
922 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
923 npages = mem->memory_size >> PAGE_SHIFT;
926 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
928 new = old = *memslot;
930 new.base_gfn = base_gfn;
932 new.flags = mem->flags;
934 /* Disallow changing a memory slot's size. */
936 if (npages && old.npages && npages != old.npages)
939 /* Check for overlaps */
941 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
942 struct kvm_memory_slot *s = &kvm->memslots[i];
946 if (!((base_gfn + npages <= s->base_gfn) ||
947 (base_gfn >= s->base_gfn + s->npages)))
951 /* Free page dirty bitmap if unneeded */
952 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
953 new.dirty_bitmap = NULL;
957 /* Allocate if a slot is being created */
959 if (npages && !new.rmap) {
960 new.rmap = vmalloc(npages * sizeof(struct page *));
965 memset(new.rmap, 0, npages * sizeof(*new.rmap));
967 new.user_alloc = user_alloc;
969 * hva_to_rmmap() serialzies with the mmu_lock and to be
970 * safe it has to ignore memslots with !user_alloc &&
974 new.userspace_addr = mem->userspace_addr;
976 new.userspace_addr = 0;
978 if (npages && !new.lpage_info) {
979 int largepages = npages / KVM_PAGES_PER_HPAGE;
980 if (npages % KVM_PAGES_PER_HPAGE)
982 if (base_gfn % KVM_PAGES_PER_HPAGE)
985 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
990 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
992 if (base_gfn % KVM_PAGES_PER_HPAGE)
993 new.lpage_info[0].write_count = 1;
994 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
995 new.lpage_info[largepages-1].write_count = 1;
998 /* Allocate page dirty bitmap if needed */
999 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1000 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1002 new.dirty_bitmap = vmalloc(dirty_bytes);
1003 if (!new.dirty_bitmap)
1005 memset(new.dirty_bitmap, 0, dirty_bytes);
1007 #endif /* not defined CONFIG_S390 */
1010 kvm_arch_flush_shadow(kvm);
1012 spin_lock(&kvm->mmu_lock);
1013 if (mem->slot >= kvm->nmemslots)
1014 kvm->nmemslots = mem->slot + 1;
1017 spin_unlock(&kvm->mmu_lock);
1019 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1021 spin_lock(&kvm->mmu_lock);
1023 spin_unlock(&kvm->mmu_lock);
1027 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1028 /* Slot deletion case: we have to update the current slot */
1032 /* map the pages in iommu page table */
1033 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1040 kvm_free_physmem_slot(&new, &old);
1045 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1047 int kvm_set_memory_region(struct kvm *kvm,
1048 struct kvm_userspace_memory_region *mem,
1053 down_write(&kvm->slots_lock);
1054 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1055 up_write(&kvm->slots_lock);
1058 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1060 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1062 kvm_userspace_memory_region *mem,
1065 if (mem->slot >= KVM_MEMORY_SLOTS)
1067 return kvm_set_memory_region(kvm, mem, user_alloc);
1070 int kvm_get_dirty_log(struct kvm *kvm,
1071 struct kvm_dirty_log *log, int *is_dirty)
1073 struct kvm_memory_slot *memslot;
1076 unsigned long any = 0;
1079 if (log->slot >= KVM_MEMORY_SLOTS)
1082 memslot = &kvm->memslots[log->slot];
1084 if (!memslot->dirty_bitmap)
1087 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1089 for (i = 0; !any && i < n/sizeof(long); ++i)
1090 any = memslot->dirty_bitmap[i];
1093 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1104 int is_error_page(struct page *page)
1106 return page == bad_page;
1108 EXPORT_SYMBOL_GPL(is_error_page);
1110 int is_error_pfn(pfn_t pfn)
1112 return pfn == bad_pfn;
1114 EXPORT_SYMBOL_GPL(is_error_pfn);
1116 static inline unsigned long bad_hva(void)
1121 int kvm_is_error_hva(unsigned long addr)
1123 return addr == bad_hva();
1125 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1127 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1131 for (i = 0; i < kvm->nmemslots; ++i) {
1132 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1134 if (gfn >= memslot->base_gfn
1135 && gfn < memslot->base_gfn + memslot->npages)
1140 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1142 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1144 gfn = unalias_gfn(kvm, gfn);
1145 return gfn_to_memslot_unaliased(kvm, gfn);
1148 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1152 gfn = unalias_gfn(kvm, gfn);
1153 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1154 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1156 if (gfn >= memslot->base_gfn
1157 && gfn < memslot->base_gfn + memslot->npages)
1162 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1164 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1166 struct kvm_memory_slot *slot;
1168 gfn = unalias_gfn(kvm, gfn);
1169 slot = gfn_to_memslot_unaliased(kvm, gfn);
1172 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1174 EXPORT_SYMBOL_GPL(gfn_to_hva);
1176 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1178 struct page *page[1];
1185 addr = gfn_to_hva(kvm, gfn);
1186 if (kvm_is_error_hva(addr)) {
1188 return page_to_pfn(bad_page);
1191 npages = get_user_pages_fast(addr, 1, 1, page);
1193 if (unlikely(npages != 1)) {
1194 struct vm_area_struct *vma;
1196 down_read(¤t->mm->mmap_sem);
1197 vma = find_vma(current->mm, addr);
1199 if (vma == NULL || addr < vma->vm_start ||
1200 !(vma->vm_flags & VM_PFNMAP)) {
1201 up_read(¤t->mm->mmap_sem);
1203 return page_to_pfn(bad_page);
1206 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1207 up_read(¤t->mm->mmap_sem);
1208 BUG_ON(!kvm_is_mmio_pfn(pfn));
1210 pfn = page_to_pfn(page[0]);
1215 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1217 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1221 pfn = gfn_to_pfn(kvm, gfn);
1222 if (!kvm_is_mmio_pfn(pfn))
1223 return pfn_to_page(pfn);
1225 WARN_ON(kvm_is_mmio_pfn(pfn));
1231 EXPORT_SYMBOL_GPL(gfn_to_page);
1233 void kvm_release_page_clean(struct page *page)
1235 kvm_release_pfn_clean(page_to_pfn(page));
1237 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1239 void kvm_release_pfn_clean(pfn_t pfn)
1241 if (!kvm_is_mmio_pfn(pfn))
1242 put_page(pfn_to_page(pfn));
1244 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1246 void kvm_release_page_dirty(struct page *page)
1248 kvm_release_pfn_dirty(page_to_pfn(page));
1250 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1252 void kvm_release_pfn_dirty(pfn_t pfn)
1254 kvm_set_pfn_dirty(pfn);
1255 kvm_release_pfn_clean(pfn);
1257 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1259 void kvm_set_page_dirty(struct page *page)
1261 kvm_set_pfn_dirty(page_to_pfn(page));
1263 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1265 void kvm_set_pfn_dirty(pfn_t pfn)
1267 if (!kvm_is_mmio_pfn(pfn)) {
1268 struct page *page = pfn_to_page(pfn);
1269 if (!PageReserved(page))
1273 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1275 void kvm_set_pfn_accessed(pfn_t pfn)
1277 if (!kvm_is_mmio_pfn(pfn))
1278 mark_page_accessed(pfn_to_page(pfn));
1280 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1282 void kvm_get_pfn(pfn_t pfn)
1284 if (!kvm_is_mmio_pfn(pfn))
1285 get_page(pfn_to_page(pfn));
1287 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1289 static int next_segment(unsigned long len, int offset)
1291 if (len > PAGE_SIZE - offset)
1292 return PAGE_SIZE - offset;
1297 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1303 addr = gfn_to_hva(kvm, gfn);
1304 if (kvm_is_error_hva(addr))
1306 r = copy_from_user(data, (void __user *)addr + offset, len);
1311 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1313 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1315 gfn_t gfn = gpa >> PAGE_SHIFT;
1317 int offset = offset_in_page(gpa);
1320 while ((seg = next_segment(len, offset)) != 0) {
1321 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1331 EXPORT_SYMBOL_GPL(kvm_read_guest);
1333 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1338 gfn_t gfn = gpa >> PAGE_SHIFT;
1339 int offset = offset_in_page(gpa);
1341 addr = gfn_to_hva(kvm, gfn);
1342 if (kvm_is_error_hva(addr))
1344 pagefault_disable();
1345 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1351 EXPORT_SYMBOL(kvm_read_guest_atomic);
1353 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1354 int offset, int len)
1359 addr = gfn_to_hva(kvm, gfn);
1360 if (kvm_is_error_hva(addr))
1362 r = copy_to_user((void __user *)addr + offset, data, len);
1365 mark_page_dirty(kvm, gfn);
1368 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1370 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1373 gfn_t gfn = gpa >> PAGE_SHIFT;
1375 int offset = offset_in_page(gpa);
1378 while ((seg = next_segment(len, offset)) != 0) {
1379 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1390 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1392 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1394 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1396 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1398 gfn_t gfn = gpa >> PAGE_SHIFT;
1400 int offset = offset_in_page(gpa);
1403 while ((seg = next_segment(len, offset)) != 0) {
1404 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1413 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1415 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1417 struct kvm_memory_slot *memslot;
1419 gfn = unalias_gfn(kvm, gfn);
1420 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1421 if (memslot && memslot->dirty_bitmap) {
1422 unsigned long rel_gfn = gfn - memslot->base_gfn;
1425 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1426 set_bit(rel_gfn, memslot->dirty_bitmap);
1431 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1433 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1438 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1440 if (kvm_cpu_has_interrupt(vcpu) ||
1441 kvm_cpu_has_pending_timer(vcpu) ||
1442 kvm_arch_vcpu_runnable(vcpu)) {
1443 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1446 if (signal_pending(current))
1454 finish_wait(&vcpu->wq, &wait);
1457 void kvm_resched(struct kvm_vcpu *vcpu)
1459 if (!need_resched())
1463 EXPORT_SYMBOL_GPL(kvm_resched);
1465 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1467 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1470 if (vmf->pgoff == 0)
1471 page = virt_to_page(vcpu->run);
1473 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1474 page = virt_to_page(vcpu->arch.pio_data);
1476 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1477 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1478 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1481 return VM_FAULT_SIGBUS;
1487 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1488 .fault = kvm_vcpu_fault,
1491 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1493 vma->vm_ops = &kvm_vcpu_vm_ops;
1497 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1499 struct kvm_vcpu *vcpu = filp->private_data;
1501 kvm_put_kvm(vcpu->kvm);
1505 static struct file_operations kvm_vcpu_fops = {
1506 .release = kvm_vcpu_release,
1507 .unlocked_ioctl = kvm_vcpu_ioctl,
1508 .compat_ioctl = kvm_vcpu_ioctl,
1509 .mmap = kvm_vcpu_mmap,
1513 * Allocates an inode for the vcpu.
1515 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1517 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1519 kvm_put_kvm(vcpu->kvm);
1524 * Creates some virtual cpus. Good luck creating more than one.
1526 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1529 struct kvm_vcpu *vcpu;
1534 vcpu = kvm_arch_vcpu_create(kvm, n);
1536 return PTR_ERR(vcpu);
1538 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1540 r = kvm_arch_vcpu_setup(vcpu);
1544 mutex_lock(&kvm->lock);
1545 if (kvm->vcpus[n]) {
1549 kvm->vcpus[n] = vcpu;
1550 mutex_unlock(&kvm->lock);
1552 /* Now it's all set up, let userspace reach it */
1554 r = create_vcpu_fd(vcpu);
1560 mutex_lock(&kvm->lock);
1561 kvm->vcpus[n] = NULL;
1563 mutex_unlock(&kvm->lock);
1564 kvm_arch_vcpu_destroy(vcpu);
1568 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1571 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1572 vcpu->sigset_active = 1;
1573 vcpu->sigset = *sigset;
1575 vcpu->sigset_active = 0;
1579 static long kvm_vcpu_ioctl(struct file *filp,
1580 unsigned int ioctl, unsigned long arg)
1582 struct kvm_vcpu *vcpu = filp->private_data;
1583 void __user *argp = (void __user *)arg;
1585 struct kvm_fpu *fpu = NULL;
1586 struct kvm_sregs *kvm_sregs = NULL;
1588 if (vcpu->kvm->mm != current->mm)
1595 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1597 case KVM_GET_REGS: {
1598 struct kvm_regs *kvm_regs;
1601 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1604 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1608 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1615 case KVM_SET_REGS: {
1616 struct kvm_regs *kvm_regs;
1619 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1623 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1625 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1633 case KVM_GET_SREGS: {
1634 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1638 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1642 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1647 case KVM_SET_SREGS: {
1648 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1653 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1655 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1661 case KVM_GET_MP_STATE: {
1662 struct kvm_mp_state mp_state;
1664 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1668 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1673 case KVM_SET_MP_STATE: {
1674 struct kvm_mp_state mp_state;
1677 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1679 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1685 case KVM_TRANSLATE: {
1686 struct kvm_translation tr;
1689 if (copy_from_user(&tr, argp, sizeof tr))
1691 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1695 if (copy_to_user(argp, &tr, sizeof tr))
1700 case KVM_DEBUG_GUEST: {
1701 struct kvm_debug_guest dbg;
1704 if (copy_from_user(&dbg, argp, sizeof dbg))
1706 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1712 case KVM_SET_SIGNAL_MASK: {
1713 struct kvm_signal_mask __user *sigmask_arg = argp;
1714 struct kvm_signal_mask kvm_sigmask;
1715 sigset_t sigset, *p;
1720 if (copy_from_user(&kvm_sigmask, argp,
1721 sizeof kvm_sigmask))
1724 if (kvm_sigmask.len != sizeof sigset)
1727 if (copy_from_user(&sigset, sigmask_arg->sigset,
1732 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1736 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1740 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1744 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1750 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1755 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1757 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1764 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1772 static long kvm_vm_ioctl(struct file *filp,
1773 unsigned int ioctl, unsigned long arg)
1775 struct kvm *kvm = filp->private_data;
1776 void __user *argp = (void __user *)arg;
1779 if (kvm->mm != current->mm)
1782 case KVM_CREATE_VCPU:
1783 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1787 case KVM_SET_USER_MEMORY_REGION: {
1788 struct kvm_userspace_memory_region kvm_userspace_mem;
1791 if (copy_from_user(&kvm_userspace_mem, argp,
1792 sizeof kvm_userspace_mem))
1795 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1800 case KVM_GET_DIRTY_LOG: {
1801 struct kvm_dirty_log log;
1804 if (copy_from_user(&log, argp, sizeof log))
1806 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1811 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1812 case KVM_REGISTER_COALESCED_MMIO: {
1813 struct kvm_coalesced_mmio_zone zone;
1815 if (copy_from_user(&zone, argp, sizeof zone))
1818 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1824 case KVM_UNREGISTER_COALESCED_MMIO: {
1825 struct kvm_coalesced_mmio_zone zone;
1827 if (copy_from_user(&zone, argp, sizeof zone))
1830 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1837 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1838 case KVM_ASSIGN_PCI_DEVICE: {
1839 struct kvm_assigned_pci_dev assigned_dev;
1842 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1844 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1849 case KVM_ASSIGN_IRQ: {
1850 struct kvm_assigned_irq assigned_irq;
1853 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1855 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1862 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1868 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1870 struct page *page[1];
1873 gfn_t gfn = vmf->pgoff;
1874 struct kvm *kvm = vma->vm_file->private_data;
1876 addr = gfn_to_hva(kvm, gfn);
1877 if (kvm_is_error_hva(addr))
1878 return VM_FAULT_SIGBUS;
1880 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1882 if (unlikely(npages != 1))
1883 return VM_FAULT_SIGBUS;
1885 vmf->page = page[0];
1889 static struct vm_operations_struct kvm_vm_vm_ops = {
1890 .fault = kvm_vm_fault,
1893 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1895 vma->vm_ops = &kvm_vm_vm_ops;
1899 static struct file_operations kvm_vm_fops = {
1900 .release = kvm_vm_release,
1901 .unlocked_ioctl = kvm_vm_ioctl,
1902 .compat_ioctl = kvm_vm_ioctl,
1903 .mmap = kvm_vm_mmap,
1906 static int kvm_dev_ioctl_create_vm(void)
1911 kvm = kvm_create_vm();
1913 return PTR_ERR(kvm);
1914 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1921 static long kvm_dev_ioctl_check_extension_generic(long arg)
1924 case KVM_CAP_USER_MEMORY:
1925 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1930 return kvm_dev_ioctl_check_extension(arg);
1933 static long kvm_dev_ioctl(struct file *filp,
1934 unsigned int ioctl, unsigned long arg)
1939 case KVM_GET_API_VERSION:
1943 r = KVM_API_VERSION;
1949 r = kvm_dev_ioctl_create_vm();
1951 case KVM_CHECK_EXTENSION:
1952 r = kvm_dev_ioctl_check_extension_generic(arg);
1954 case KVM_GET_VCPU_MMAP_SIZE:
1958 r = PAGE_SIZE; /* struct kvm_run */
1960 r += PAGE_SIZE; /* pio data page */
1962 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1963 r += PAGE_SIZE; /* coalesced mmio ring page */
1966 case KVM_TRACE_ENABLE:
1967 case KVM_TRACE_PAUSE:
1968 case KVM_TRACE_DISABLE:
1969 r = kvm_trace_ioctl(ioctl, arg);
1972 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1978 static struct file_operations kvm_chardev_ops = {
1979 .unlocked_ioctl = kvm_dev_ioctl,
1980 .compat_ioctl = kvm_dev_ioctl,
1983 static struct miscdevice kvm_dev = {
1989 static void hardware_enable(void *junk)
1991 int cpu = raw_smp_processor_id();
1993 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1995 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1996 kvm_arch_hardware_enable(NULL);
1999 static void hardware_disable(void *junk)
2001 int cpu = raw_smp_processor_id();
2003 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2005 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2006 kvm_arch_hardware_disable(NULL);
2009 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2014 val &= ~CPU_TASKS_FROZEN;
2017 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2019 hardware_disable(NULL);
2021 case CPU_UP_CANCELED:
2022 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2024 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2027 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2029 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2036 asmlinkage void kvm_handle_fault_on_reboot(void)
2039 /* spin while reset goes on */
2042 /* Fault while not rebooting. We want the trace. */
2045 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2047 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2050 if (val == SYS_RESTART) {
2052 * Some (well, at least mine) BIOSes hang on reboot if
2055 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2056 kvm_rebooting = true;
2057 on_each_cpu(hardware_disable, NULL, 1);
2062 static struct notifier_block kvm_reboot_notifier = {
2063 .notifier_call = kvm_reboot,
2067 void kvm_io_bus_init(struct kvm_io_bus *bus)
2069 memset(bus, 0, sizeof(*bus));
2072 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2076 for (i = 0; i < bus->dev_count; i++) {
2077 struct kvm_io_device *pos = bus->devs[i];
2079 kvm_iodevice_destructor(pos);
2083 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2084 gpa_t addr, int len, int is_write)
2088 for (i = 0; i < bus->dev_count; i++) {
2089 struct kvm_io_device *pos = bus->devs[i];
2091 if (pos->in_range(pos, addr, len, is_write))
2098 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2100 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2102 bus->devs[bus->dev_count++] = dev;
2105 static struct notifier_block kvm_cpu_notifier = {
2106 .notifier_call = kvm_cpu_hotplug,
2107 .priority = 20, /* must be > scheduler priority */
2110 static int vm_stat_get(void *_offset, u64 *val)
2112 unsigned offset = (long)_offset;
2116 spin_lock(&kvm_lock);
2117 list_for_each_entry(kvm, &vm_list, vm_list)
2118 *val += *(u32 *)((void *)kvm + offset);
2119 spin_unlock(&kvm_lock);
2123 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2125 static int vcpu_stat_get(void *_offset, u64 *val)
2127 unsigned offset = (long)_offset;
2129 struct kvm_vcpu *vcpu;
2133 spin_lock(&kvm_lock);
2134 list_for_each_entry(kvm, &vm_list, vm_list)
2135 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2136 vcpu = kvm->vcpus[i];
2138 *val += *(u32 *)((void *)vcpu + offset);
2140 spin_unlock(&kvm_lock);
2144 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2146 static struct file_operations *stat_fops[] = {
2147 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2148 [KVM_STAT_VM] = &vm_stat_fops,
2151 static void kvm_init_debug(void)
2153 struct kvm_stats_debugfs_item *p;
2155 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2156 for (p = debugfs_entries; p->name; ++p)
2157 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2158 (void *)(long)p->offset,
2159 stat_fops[p->kind]);
2162 static void kvm_exit_debug(void)
2164 struct kvm_stats_debugfs_item *p;
2166 for (p = debugfs_entries; p->name; ++p)
2167 debugfs_remove(p->dentry);
2168 debugfs_remove(kvm_debugfs_dir);
2171 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2173 hardware_disable(NULL);
2177 static int kvm_resume(struct sys_device *dev)
2179 hardware_enable(NULL);
2183 static struct sysdev_class kvm_sysdev_class = {
2185 .suspend = kvm_suspend,
2186 .resume = kvm_resume,
2189 static struct sys_device kvm_sysdev = {
2191 .cls = &kvm_sysdev_class,
2194 struct page *bad_page;
2198 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2200 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2203 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2205 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2207 kvm_arch_vcpu_load(vcpu, cpu);
2210 static void kvm_sched_out(struct preempt_notifier *pn,
2211 struct task_struct *next)
2213 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2215 kvm_arch_vcpu_put(vcpu);
2218 int kvm_init(void *opaque, unsigned int vcpu_size,
2219 struct module *module)
2226 r = kvm_arch_init(opaque);
2230 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2232 if (bad_page == NULL) {
2237 bad_pfn = page_to_pfn(bad_page);
2239 if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2244 r = kvm_arch_hardware_setup();
2248 for_each_online_cpu(cpu) {
2249 smp_call_function_single(cpu,
2250 kvm_arch_check_processor_compat,
2256 on_each_cpu(hardware_enable, NULL, 1);
2257 r = register_cpu_notifier(&kvm_cpu_notifier);
2260 register_reboot_notifier(&kvm_reboot_notifier);
2262 r = sysdev_class_register(&kvm_sysdev_class);
2266 r = sysdev_register(&kvm_sysdev);
2270 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2271 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2272 __alignof__(struct kvm_vcpu),
2274 if (!kvm_vcpu_cache) {
2279 kvm_chardev_ops.owner = module;
2280 kvm_vm_fops.owner = module;
2281 kvm_vcpu_fops.owner = module;
2283 r = misc_register(&kvm_dev);
2285 printk(KERN_ERR "kvm: misc device register failed\n");
2289 kvm_preempt_ops.sched_in = kvm_sched_in;
2290 kvm_preempt_ops.sched_out = kvm_sched_out;
2298 kmem_cache_destroy(kvm_vcpu_cache);
2300 sysdev_unregister(&kvm_sysdev);
2302 sysdev_class_unregister(&kvm_sysdev_class);
2304 unregister_reboot_notifier(&kvm_reboot_notifier);
2305 unregister_cpu_notifier(&kvm_cpu_notifier);
2307 on_each_cpu(hardware_disable, NULL, 1);
2309 kvm_arch_hardware_unsetup();
2311 free_cpumask_var(cpus_hardware_enabled);
2313 __free_page(bad_page);
2320 EXPORT_SYMBOL_GPL(kvm_init);
2324 kvm_trace_cleanup();
2325 misc_deregister(&kvm_dev);
2326 kmem_cache_destroy(kvm_vcpu_cache);
2327 sysdev_unregister(&kvm_sysdev);
2328 sysdev_class_unregister(&kvm_sysdev_class);
2329 unregister_reboot_notifier(&kvm_reboot_notifier);
2330 unregister_cpu_notifier(&kvm_cpu_notifier);
2331 on_each_cpu(hardware_disable, NULL, 1);
2332 kvm_arch_hardware_unsetup();
2335 free_cpumask_var(cpus_hardware_enabled);
2336 __free_page(bad_page);
2338 EXPORT_SYMBOL_GPL(kvm_exit);