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;
499 match->flags = assigned_dev->flags;
501 match->irq_source_id = -1;
504 list_add(&match->list, &kvm->arch.assigned_dev_head);
506 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
507 if (!kvm->arch.iommu_domain) {
508 r = kvm_iommu_map_guest(kvm);
512 r = kvm_assign_device(kvm, match);
518 mutex_unlock(&kvm->lock);
521 list_del(&match->list);
522 pci_release_regions(dev);
524 pci_disable_device(dev);
529 mutex_unlock(&kvm->lock);
534 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
535 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
536 struct kvm_assigned_pci_dev *assigned_dev)
539 struct kvm_assigned_dev_kernel *match;
541 mutex_lock(&kvm->lock);
543 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
544 assigned_dev->assigned_dev_id);
546 printk(KERN_INFO "%s: device hasn't been assigned before, "
547 "so cannot be deassigned\n", __func__);
552 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
553 kvm_deassign_device(kvm, match);
555 kvm_free_assigned_device(kvm, match);
558 mutex_unlock(&kvm->lock);
563 static inline int valid_vcpu(int n)
565 return likely(n >= 0 && n < KVM_MAX_VCPUS);
568 inline int kvm_is_mmio_pfn(pfn_t pfn)
571 return PageReserved(pfn_to_page(pfn));
577 * Switches to specified vcpu, until a matching vcpu_put()
579 void vcpu_load(struct kvm_vcpu *vcpu)
583 mutex_lock(&vcpu->mutex);
585 preempt_notifier_register(&vcpu->preempt_notifier);
586 kvm_arch_vcpu_load(vcpu, cpu);
590 void vcpu_put(struct kvm_vcpu *vcpu)
593 kvm_arch_vcpu_put(vcpu);
594 preempt_notifier_unregister(&vcpu->preempt_notifier);
596 mutex_unlock(&vcpu->mutex);
599 static void ack_flush(void *_completed)
603 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
608 struct kvm_vcpu *vcpu;
610 if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
614 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
615 vcpu = kvm->vcpus[i];
618 if (test_and_set_bit(req, &vcpu->requests))
621 if (cpus != NULL && cpu != -1 && cpu != me)
622 cpumask_set_cpu(cpu, cpus);
624 if (unlikely(cpus == NULL))
625 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
626 else if (!cpumask_empty(cpus))
627 smp_call_function_many(cpus, ack_flush, NULL, 1);
631 free_cpumask_var(cpus);
635 void kvm_flush_remote_tlbs(struct kvm *kvm)
637 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
638 ++kvm->stat.remote_tlb_flush;
641 void kvm_reload_remote_mmus(struct kvm *kvm)
643 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
646 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
651 mutex_init(&vcpu->mutex);
655 init_waitqueue_head(&vcpu->wq);
657 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
662 vcpu->run = page_address(page);
664 r = kvm_arch_vcpu_init(vcpu);
670 free_page((unsigned long)vcpu->run);
674 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
676 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
678 kvm_arch_vcpu_uninit(vcpu);
679 free_page((unsigned long)vcpu->run);
681 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
683 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
684 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
686 return container_of(mn, struct kvm, mmu_notifier);
689 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
690 struct mm_struct *mm,
691 unsigned long address)
693 struct kvm *kvm = mmu_notifier_to_kvm(mn);
697 * When ->invalidate_page runs, the linux pte has been zapped
698 * already but the page is still allocated until
699 * ->invalidate_page returns. So if we increase the sequence
700 * here the kvm page fault will notice if the spte can't be
701 * established because the page is going to be freed. If
702 * instead the kvm page fault establishes the spte before
703 * ->invalidate_page runs, kvm_unmap_hva will release it
706 * The sequence increase only need to be seen at spin_unlock
707 * time, and not at spin_lock time.
709 * Increasing the sequence after the spin_unlock would be
710 * unsafe because the kvm page fault could then establish the
711 * pte after kvm_unmap_hva returned, without noticing the page
712 * is going to be freed.
714 spin_lock(&kvm->mmu_lock);
715 kvm->mmu_notifier_seq++;
716 need_tlb_flush = kvm_unmap_hva(kvm, address);
717 spin_unlock(&kvm->mmu_lock);
719 /* we've to flush the tlb before the pages can be freed */
721 kvm_flush_remote_tlbs(kvm);
725 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
726 struct mm_struct *mm,
730 struct kvm *kvm = mmu_notifier_to_kvm(mn);
731 int need_tlb_flush = 0;
733 spin_lock(&kvm->mmu_lock);
735 * The count increase must become visible at unlock time as no
736 * spte can be established without taking the mmu_lock and
737 * count is also read inside the mmu_lock critical section.
739 kvm->mmu_notifier_count++;
740 for (; start < end; start += PAGE_SIZE)
741 need_tlb_flush |= kvm_unmap_hva(kvm, start);
742 spin_unlock(&kvm->mmu_lock);
744 /* we've to flush the tlb before the pages can be freed */
746 kvm_flush_remote_tlbs(kvm);
749 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
750 struct mm_struct *mm,
754 struct kvm *kvm = mmu_notifier_to_kvm(mn);
756 spin_lock(&kvm->mmu_lock);
758 * This sequence increase will notify the kvm page fault that
759 * the page that is going to be mapped in the spte could have
762 kvm->mmu_notifier_seq++;
764 * The above sequence increase must be visible before the
765 * below count decrease but both values are read by the kvm
766 * page fault under mmu_lock spinlock so we don't need to add
767 * a smb_wmb() here in between the two.
769 kvm->mmu_notifier_count--;
770 spin_unlock(&kvm->mmu_lock);
772 BUG_ON(kvm->mmu_notifier_count < 0);
775 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
776 struct mm_struct *mm,
777 unsigned long address)
779 struct kvm *kvm = mmu_notifier_to_kvm(mn);
782 spin_lock(&kvm->mmu_lock);
783 young = kvm_age_hva(kvm, address);
784 spin_unlock(&kvm->mmu_lock);
787 kvm_flush_remote_tlbs(kvm);
792 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
793 .invalidate_page = kvm_mmu_notifier_invalidate_page,
794 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
795 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
796 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
798 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
800 static struct kvm *kvm_create_vm(void)
802 struct kvm *kvm = kvm_arch_create_vm();
803 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
810 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
811 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
814 return ERR_PTR(-ENOMEM);
816 kvm->coalesced_mmio_ring =
817 (struct kvm_coalesced_mmio_ring *)page_address(page);
820 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
823 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
824 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
826 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
835 kvm->mm = current->mm;
836 atomic_inc(&kvm->mm->mm_count);
837 spin_lock_init(&kvm->mmu_lock);
838 kvm_io_bus_init(&kvm->pio_bus);
839 mutex_init(&kvm->lock);
840 kvm_io_bus_init(&kvm->mmio_bus);
841 init_rwsem(&kvm->slots_lock);
842 atomic_set(&kvm->users_count, 1);
843 spin_lock(&kvm_lock);
844 list_add(&kvm->vm_list, &vm_list);
845 spin_unlock(&kvm_lock);
846 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
847 kvm_coalesced_mmio_init(kvm);
854 * Free any memory in @free but not in @dont.
856 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
857 struct kvm_memory_slot *dont)
859 if (!dont || free->rmap != dont->rmap)
862 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
863 vfree(free->dirty_bitmap);
865 if (!dont || free->lpage_info != dont->lpage_info)
866 vfree(free->lpage_info);
869 free->dirty_bitmap = NULL;
871 free->lpage_info = NULL;
874 void kvm_free_physmem(struct kvm *kvm)
878 for (i = 0; i < kvm->nmemslots; ++i)
879 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
882 static void kvm_destroy_vm(struct kvm *kvm)
884 struct mm_struct *mm = kvm->mm;
886 spin_lock(&kvm_lock);
887 list_del(&kvm->vm_list);
888 spin_unlock(&kvm_lock);
889 kvm_io_bus_destroy(&kvm->pio_bus);
890 kvm_io_bus_destroy(&kvm->mmio_bus);
891 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
892 if (kvm->coalesced_mmio_ring != NULL)
893 free_page((unsigned long)kvm->coalesced_mmio_ring);
895 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
896 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
898 kvm_arch_destroy_vm(kvm);
902 void kvm_get_kvm(struct kvm *kvm)
904 atomic_inc(&kvm->users_count);
906 EXPORT_SYMBOL_GPL(kvm_get_kvm);
908 void kvm_put_kvm(struct kvm *kvm)
910 if (atomic_dec_and_test(&kvm->users_count))
913 EXPORT_SYMBOL_GPL(kvm_put_kvm);
916 static int kvm_vm_release(struct inode *inode, struct file *filp)
918 struct kvm *kvm = filp->private_data;
925 * Allocate some memory and give it an address in the guest physical address
928 * Discontiguous memory is allowed, mostly for framebuffers.
930 * Must be called holding mmap_sem for write.
932 int __kvm_set_memory_region(struct kvm *kvm,
933 struct kvm_userspace_memory_region *mem,
938 unsigned long npages;
940 struct kvm_memory_slot *memslot;
941 struct kvm_memory_slot old, new;
944 /* General sanity checks */
945 if (mem->memory_size & (PAGE_SIZE - 1))
947 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
949 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
951 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
953 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
956 memslot = &kvm->memslots[mem->slot];
957 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
958 npages = mem->memory_size >> PAGE_SHIFT;
961 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
963 new = old = *memslot;
965 new.base_gfn = base_gfn;
967 new.flags = mem->flags;
969 /* Disallow changing a memory slot's size. */
971 if (npages && old.npages && npages != old.npages)
974 /* Check for overlaps */
976 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
977 struct kvm_memory_slot *s = &kvm->memslots[i];
981 if (!((base_gfn + npages <= s->base_gfn) ||
982 (base_gfn >= s->base_gfn + s->npages)))
986 /* Free page dirty bitmap if unneeded */
987 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
988 new.dirty_bitmap = NULL;
992 /* Allocate if a slot is being created */
994 if (npages && !new.rmap) {
995 new.rmap = vmalloc(npages * sizeof(struct page *));
1000 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1002 new.user_alloc = user_alloc;
1004 * hva_to_rmmap() serialzies with the mmu_lock and to be
1005 * safe it has to ignore memslots with !user_alloc &&
1009 new.userspace_addr = mem->userspace_addr;
1011 new.userspace_addr = 0;
1013 if (npages && !new.lpage_info) {
1014 int largepages = npages / KVM_PAGES_PER_HPAGE;
1015 if (npages % KVM_PAGES_PER_HPAGE)
1017 if (base_gfn % KVM_PAGES_PER_HPAGE)
1020 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1022 if (!new.lpage_info)
1025 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1027 if (base_gfn % KVM_PAGES_PER_HPAGE)
1028 new.lpage_info[0].write_count = 1;
1029 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1030 new.lpage_info[largepages-1].write_count = 1;
1033 /* Allocate page dirty bitmap if needed */
1034 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1035 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1037 new.dirty_bitmap = vmalloc(dirty_bytes);
1038 if (!new.dirty_bitmap)
1040 memset(new.dirty_bitmap, 0, dirty_bytes);
1042 #endif /* not defined CONFIG_S390 */
1045 kvm_arch_flush_shadow(kvm);
1047 spin_lock(&kvm->mmu_lock);
1048 if (mem->slot >= kvm->nmemslots)
1049 kvm->nmemslots = mem->slot + 1;
1052 spin_unlock(&kvm->mmu_lock);
1054 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1056 spin_lock(&kvm->mmu_lock);
1058 spin_unlock(&kvm->mmu_lock);
1062 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1063 /* Slot deletion case: we have to update the current slot */
1067 /* map the pages in iommu page table */
1068 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1075 kvm_free_physmem_slot(&new, &old);
1080 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1082 int kvm_set_memory_region(struct kvm *kvm,
1083 struct kvm_userspace_memory_region *mem,
1088 down_write(&kvm->slots_lock);
1089 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1090 up_write(&kvm->slots_lock);
1093 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1095 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1097 kvm_userspace_memory_region *mem,
1100 if (mem->slot >= KVM_MEMORY_SLOTS)
1102 return kvm_set_memory_region(kvm, mem, user_alloc);
1105 int kvm_get_dirty_log(struct kvm *kvm,
1106 struct kvm_dirty_log *log, int *is_dirty)
1108 struct kvm_memory_slot *memslot;
1111 unsigned long any = 0;
1114 if (log->slot >= KVM_MEMORY_SLOTS)
1117 memslot = &kvm->memslots[log->slot];
1119 if (!memslot->dirty_bitmap)
1122 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1124 for (i = 0; !any && i < n/sizeof(long); ++i)
1125 any = memslot->dirty_bitmap[i];
1128 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1139 int is_error_page(struct page *page)
1141 return page == bad_page;
1143 EXPORT_SYMBOL_GPL(is_error_page);
1145 int is_error_pfn(pfn_t pfn)
1147 return pfn == bad_pfn;
1149 EXPORT_SYMBOL_GPL(is_error_pfn);
1151 static inline unsigned long bad_hva(void)
1156 int kvm_is_error_hva(unsigned long addr)
1158 return addr == bad_hva();
1160 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1162 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1166 for (i = 0; i < kvm->nmemslots; ++i) {
1167 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1169 if (gfn >= memslot->base_gfn
1170 && gfn < memslot->base_gfn + memslot->npages)
1175 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1177 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1179 gfn = unalias_gfn(kvm, gfn);
1180 return gfn_to_memslot_unaliased(kvm, gfn);
1183 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1187 gfn = unalias_gfn(kvm, gfn);
1188 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1189 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1191 if (gfn >= memslot->base_gfn
1192 && gfn < memslot->base_gfn + memslot->npages)
1197 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1199 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1201 struct kvm_memory_slot *slot;
1203 gfn = unalias_gfn(kvm, gfn);
1204 slot = gfn_to_memslot_unaliased(kvm, gfn);
1207 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1209 EXPORT_SYMBOL_GPL(gfn_to_hva);
1211 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1213 struct page *page[1];
1220 addr = gfn_to_hva(kvm, gfn);
1221 if (kvm_is_error_hva(addr)) {
1223 return page_to_pfn(bad_page);
1226 npages = get_user_pages_fast(addr, 1, 1, page);
1228 if (unlikely(npages != 1)) {
1229 struct vm_area_struct *vma;
1231 down_read(¤t->mm->mmap_sem);
1232 vma = find_vma(current->mm, addr);
1234 if (vma == NULL || addr < vma->vm_start ||
1235 !(vma->vm_flags & VM_PFNMAP)) {
1236 up_read(¤t->mm->mmap_sem);
1238 return page_to_pfn(bad_page);
1241 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1242 up_read(¤t->mm->mmap_sem);
1243 BUG_ON(!kvm_is_mmio_pfn(pfn));
1245 pfn = page_to_pfn(page[0]);
1250 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1252 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1256 pfn = gfn_to_pfn(kvm, gfn);
1257 if (!kvm_is_mmio_pfn(pfn))
1258 return pfn_to_page(pfn);
1260 WARN_ON(kvm_is_mmio_pfn(pfn));
1266 EXPORT_SYMBOL_GPL(gfn_to_page);
1268 void kvm_release_page_clean(struct page *page)
1270 kvm_release_pfn_clean(page_to_pfn(page));
1272 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1274 void kvm_release_pfn_clean(pfn_t pfn)
1276 if (!kvm_is_mmio_pfn(pfn))
1277 put_page(pfn_to_page(pfn));
1279 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1281 void kvm_release_page_dirty(struct page *page)
1283 kvm_release_pfn_dirty(page_to_pfn(page));
1285 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1287 void kvm_release_pfn_dirty(pfn_t pfn)
1289 kvm_set_pfn_dirty(pfn);
1290 kvm_release_pfn_clean(pfn);
1292 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1294 void kvm_set_page_dirty(struct page *page)
1296 kvm_set_pfn_dirty(page_to_pfn(page));
1298 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1300 void kvm_set_pfn_dirty(pfn_t pfn)
1302 if (!kvm_is_mmio_pfn(pfn)) {
1303 struct page *page = pfn_to_page(pfn);
1304 if (!PageReserved(page))
1308 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1310 void kvm_set_pfn_accessed(pfn_t pfn)
1312 if (!kvm_is_mmio_pfn(pfn))
1313 mark_page_accessed(pfn_to_page(pfn));
1315 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1317 void kvm_get_pfn(pfn_t pfn)
1319 if (!kvm_is_mmio_pfn(pfn))
1320 get_page(pfn_to_page(pfn));
1322 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1324 static int next_segment(unsigned long len, int offset)
1326 if (len > PAGE_SIZE - offset)
1327 return PAGE_SIZE - offset;
1332 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1338 addr = gfn_to_hva(kvm, gfn);
1339 if (kvm_is_error_hva(addr))
1341 r = copy_from_user(data, (void __user *)addr + offset, len);
1346 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1348 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1350 gfn_t gfn = gpa >> PAGE_SHIFT;
1352 int offset = offset_in_page(gpa);
1355 while ((seg = next_segment(len, offset)) != 0) {
1356 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1366 EXPORT_SYMBOL_GPL(kvm_read_guest);
1368 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1373 gfn_t gfn = gpa >> PAGE_SHIFT;
1374 int offset = offset_in_page(gpa);
1376 addr = gfn_to_hva(kvm, gfn);
1377 if (kvm_is_error_hva(addr))
1379 pagefault_disable();
1380 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1386 EXPORT_SYMBOL(kvm_read_guest_atomic);
1388 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1389 int offset, int len)
1394 addr = gfn_to_hva(kvm, gfn);
1395 if (kvm_is_error_hva(addr))
1397 r = copy_to_user((void __user *)addr + offset, data, len);
1400 mark_page_dirty(kvm, gfn);
1403 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1405 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1408 gfn_t gfn = gpa >> PAGE_SHIFT;
1410 int offset = offset_in_page(gpa);
1413 while ((seg = next_segment(len, offset)) != 0) {
1414 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1425 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1427 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1429 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1431 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1433 gfn_t gfn = gpa >> PAGE_SHIFT;
1435 int offset = offset_in_page(gpa);
1438 while ((seg = next_segment(len, offset)) != 0) {
1439 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1448 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1450 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1452 struct kvm_memory_slot *memslot;
1454 gfn = unalias_gfn(kvm, gfn);
1455 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1456 if (memslot && memslot->dirty_bitmap) {
1457 unsigned long rel_gfn = gfn - memslot->base_gfn;
1460 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1461 set_bit(rel_gfn, memslot->dirty_bitmap);
1466 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1468 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1473 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1475 if (kvm_cpu_has_interrupt(vcpu) ||
1476 kvm_cpu_has_pending_timer(vcpu) ||
1477 kvm_arch_vcpu_runnable(vcpu)) {
1478 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1481 if (signal_pending(current))
1489 finish_wait(&vcpu->wq, &wait);
1492 void kvm_resched(struct kvm_vcpu *vcpu)
1494 if (!need_resched())
1498 EXPORT_SYMBOL_GPL(kvm_resched);
1500 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1502 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1505 if (vmf->pgoff == 0)
1506 page = virt_to_page(vcpu->run);
1508 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1509 page = virt_to_page(vcpu->arch.pio_data);
1511 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1512 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1513 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1516 return VM_FAULT_SIGBUS;
1522 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1523 .fault = kvm_vcpu_fault,
1526 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1528 vma->vm_ops = &kvm_vcpu_vm_ops;
1532 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1534 struct kvm_vcpu *vcpu = filp->private_data;
1536 kvm_put_kvm(vcpu->kvm);
1540 static struct file_operations kvm_vcpu_fops = {
1541 .release = kvm_vcpu_release,
1542 .unlocked_ioctl = kvm_vcpu_ioctl,
1543 .compat_ioctl = kvm_vcpu_ioctl,
1544 .mmap = kvm_vcpu_mmap,
1548 * Allocates an inode for the vcpu.
1550 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1552 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1554 kvm_put_kvm(vcpu->kvm);
1559 * Creates some virtual cpus. Good luck creating more than one.
1561 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1564 struct kvm_vcpu *vcpu;
1569 vcpu = kvm_arch_vcpu_create(kvm, n);
1571 return PTR_ERR(vcpu);
1573 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1575 r = kvm_arch_vcpu_setup(vcpu);
1579 mutex_lock(&kvm->lock);
1580 if (kvm->vcpus[n]) {
1584 kvm->vcpus[n] = vcpu;
1585 mutex_unlock(&kvm->lock);
1587 /* Now it's all set up, let userspace reach it */
1589 r = create_vcpu_fd(vcpu);
1595 mutex_lock(&kvm->lock);
1596 kvm->vcpus[n] = NULL;
1598 mutex_unlock(&kvm->lock);
1599 kvm_arch_vcpu_destroy(vcpu);
1603 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1606 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1607 vcpu->sigset_active = 1;
1608 vcpu->sigset = *sigset;
1610 vcpu->sigset_active = 0;
1614 static long kvm_vcpu_ioctl(struct file *filp,
1615 unsigned int ioctl, unsigned long arg)
1617 struct kvm_vcpu *vcpu = filp->private_data;
1618 void __user *argp = (void __user *)arg;
1620 struct kvm_fpu *fpu = NULL;
1621 struct kvm_sregs *kvm_sregs = NULL;
1623 if (vcpu->kvm->mm != current->mm)
1630 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1632 case KVM_GET_REGS: {
1633 struct kvm_regs *kvm_regs;
1636 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1639 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1643 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1650 case KVM_SET_REGS: {
1651 struct kvm_regs *kvm_regs;
1654 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1658 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1660 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1668 case KVM_GET_SREGS: {
1669 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1673 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1677 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1682 case KVM_SET_SREGS: {
1683 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1688 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1690 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1696 case KVM_GET_MP_STATE: {
1697 struct kvm_mp_state mp_state;
1699 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1703 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1708 case KVM_SET_MP_STATE: {
1709 struct kvm_mp_state mp_state;
1712 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1714 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1720 case KVM_TRANSLATE: {
1721 struct kvm_translation tr;
1724 if (copy_from_user(&tr, argp, sizeof tr))
1726 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1730 if (copy_to_user(argp, &tr, sizeof tr))
1735 case KVM_DEBUG_GUEST: {
1736 struct kvm_debug_guest dbg;
1739 if (copy_from_user(&dbg, argp, sizeof dbg))
1741 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1747 case KVM_SET_SIGNAL_MASK: {
1748 struct kvm_signal_mask __user *sigmask_arg = argp;
1749 struct kvm_signal_mask kvm_sigmask;
1750 sigset_t sigset, *p;
1755 if (copy_from_user(&kvm_sigmask, argp,
1756 sizeof kvm_sigmask))
1759 if (kvm_sigmask.len != sizeof sigset)
1762 if (copy_from_user(&sigset, sigmask_arg->sigset,
1767 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1771 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1775 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1779 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1785 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1790 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1792 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1799 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1807 static long kvm_vm_ioctl(struct file *filp,
1808 unsigned int ioctl, unsigned long arg)
1810 struct kvm *kvm = filp->private_data;
1811 void __user *argp = (void __user *)arg;
1814 if (kvm->mm != current->mm)
1817 case KVM_CREATE_VCPU:
1818 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1822 case KVM_SET_USER_MEMORY_REGION: {
1823 struct kvm_userspace_memory_region kvm_userspace_mem;
1826 if (copy_from_user(&kvm_userspace_mem, argp,
1827 sizeof kvm_userspace_mem))
1830 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1835 case KVM_GET_DIRTY_LOG: {
1836 struct kvm_dirty_log log;
1839 if (copy_from_user(&log, argp, sizeof log))
1841 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1846 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1847 case KVM_REGISTER_COALESCED_MMIO: {
1848 struct kvm_coalesced_mmio_zone zone;
1850 if (copy_from_user(&zone, argp, sizeof zone))
1853 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1859 case KVM_UNREGISTER_COALESCED_MMIO: {
1860 struct kvm_coalesced_mmio_zone zone;
1862 if (copy_from_user(&zone, argp, sizeof zone))
1865 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1872 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1873 case KVM_ASSIGN_PCI_DEVICE: {
1874 struct kvm_assigned_pci_dev assigned_dev;
1877 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1879 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1884 case KVM_ASSIGN_IRQ: {
1885 struct kvm_assigned_irq assigned_irq;
1888 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1890 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1896 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1897 case KVM_DEASSIGN_PCI_DEVICE: {
1898 struct kvm_assigned_pci_dev assigned_dev;
1901 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1903 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1910 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1916 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1918 struct page *page[1];
1921 gfn_t gfn = vmf->pgoff;
1922 struct kvm *kvm = vma->vm_file->private_data;
1924 addr = gfn_to_hva(kvm, gfn);
1925 if (kvm_is_error_hva(addr))
1926 return VM_FAULT_SIGBUS;
1928 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1930 if (unlikely(npages != 1))
1931 return VM_FAULT_SIGBUS;
1933 vmf->page = page[0];
1937 static struct vm_operations_struct kvm_vm_vm_ops = {
1938 .fault = kvm_vm_fault,
1941 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1943 vma->vm_ops = &kvm_vm_vm_ops;
1947 static struct file_operations kvm_vm_fops = {
1948 .release = kvm_vm_release,
1949 .unlocked_ioctl = kvm_vm_ioctl,
1950 .compat_ioctl = kvm_vm_ioctl,
1951 .mmap = kvm_vm_mmap,
1954 static int kvm_dev_ioctl_create_vm(void)
1959 kvm = kvm_create_vm();
1961 return PTR_ERR(kvm);
1962 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1969 static long kvm_dev_ioctl_check_extension_generic(long arg)
1972 case KVM_CAP_USER_MEMORY:
1973 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1978 return kvm_dev_ioctl_check_extension(arg);
1981 static long kvm_dev_ioctl(struct file *filp,
1982 unsigned int ioctl, unsigned long arg)
1987 case KVM_GET_API_VERSION:
1991 r = KVM_API_VERSION;
1997 r = kvm_dev_ioctl_create_vm();
1999 case KVM_CHECK_EXTENSION:
2000 r = kvm_dev_ioctl_check_extension_generic(arg);
2002 case KVM_GET_VCPU_MMAP_SIZE:
2006 r = PAGE_SIZE; /* struct kvm_run */
2008 r += PAGE_SIZE; /* pio data page */
2010 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2011 r += PAGE_SIZE; /* coalesced mmio ring page */
2014 case KVM_TRACE_ENABLE:
2015 case KVM_TRACE_PAUSE:
2016 case KVM_TRACE_DISABLE:
2017 r = kvm_trace_ioctl(ioctl, arg);
2020 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2026 static struct file_operations kvm_chardev_ops = {
2027 .unlocked_ioctl = kvm_dev_ioctl,
2028 .compat_ioctl = kvm_dev_ioctl,
2031 static struct miscdevice kvm_dev = {
2037 static void hardware_enable(void *junk)
2039 int cpu = raw_smp_processor_id();
2041 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2043 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2044 kvm_arch_hardware_enable(NULL);
2047 static void hardware_disable(void *junk)
2049 int cpu = raw_smp_processor_id();
2051 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2053 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2054 kvm_arch_hardware_disable(NULL);
2057 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2062 val &= ~CPU_TASKS_FROZEN;
2065 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2067 hardware_disable(NULL);
2069 case CPU_UP_CANCELED:
2070 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2072 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2075 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2077 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2084 asmlinkage void kvm_handle_fault_on_reboot(void)
2087 /* spin while reset goes on */
2090 /* Fault while not rebooting. We want the trace. */
2093 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2095 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2098 if (val == SYS_RESTART) {
2100 * Some (well, at least mine) BIOSes hang on reboot if
2103 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2104 kvm_rebooting = true;
2105 on_each_cpu(hardware_disable, NULL, 1);
2110 static struct notifier_block kvm_reboot_notifier = {
2111 .notifier_call = kvm_reboot,
2115 void kvm_io_bus_init(struct kvm_io_bus *bus)
2117 memset(bus, 0, sizeof(*bus));
2120 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2124 for (i = 0; i < bus->dev_count; i++) {
2125 struct kvm_io_device *pos = bus->devs[i];
2127 kvm_iodevice_destructor(pos);
2131 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2132 gpa_t addr, int len, int is_write)
2136 for (i = 0; i < bus->dev_count; i++) {
2137 struct kvm_io_device *pos = bus->devs[i];
2139 if (pos->in_range(pos, addr, len, is_write))
2146 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2148 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2150 bus->devs[bus->dev_count++] = dev;
2153 static struct notifier_block kvm_cpu_notifier = {
2154 .notifier_call = kvm_cpu_hotplug,
2155 .priority = 20, /* must be > scheduler priority */
2158 static int vm_stat_get(void *_offset, u64 *val)
2160 unsigned offset = (long)_offset;
2164 spin_lock(&kvm_lock);
2165 list_for_each_entry(kvm, &vm_list, vm_list)
2166 *val += *(u32 *)((void *)kvm + offset);
2167 spin_unlock(&kvm_lock);
2171 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2173 static int vcpu_stat_get(void *_offset, u64 *val)
2175 unsigned offset = (long)_offset;
2177 struct kvm_vcpu *vcpu;
2181 spin_lock(&kvm_lock);
2182 list_for_each_entry(kvm, &vm_list, vm_list)
2183 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2184 vcpu = kvm->vcpus[i];
2186 *val += *(u32 *)((void *)vcpu + offset);
2188 spin_unlock(&kvm_lock);
2192 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2194 static struct file_operations *stat_fops[] = {
2195 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2196 [KVM_STAT_VM] = &vm_stat_fops,
2199 static void kvm_init_debug(void)
2201 struct kvm_stats_debugfs_item *p;
2203 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2204 for (p = debugfs_entries; p->name; ++p)
2205 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2206 (void *)(long)p->offset,
2207 stat_fops[p->kind]);
2210 static void kvm_exit_debug(void)
2212 struct kvm_stats_debugfs_item *p;
2214 for (p = debugfs_entries; p->name; ++p)
2215 debugfs_remove(p->dentry);
2216 debugfs_remove(kvm_debugfs_dir);
2219 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2221 hardware_disable(NULL);
2225 static int kvm_resume(struct sys_device *dev)
2227 hardware_enable(NULL);
2231 static struct sysdev_class kvm_sysdev_class = {
2233 .suspend = kvm_suspend,
2234 .resume = kvm_resume,
2237 static struct sys_device kvm_sysdev = {
2239 .cls = &kvm_sysdev_class,
2242 struct page *bad_page;
2246 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2248 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2251 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2253 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2255 kvm_arch_vcpu_load(vcpu, cpu);
2258 static void kvm_sched_out(struct preempt_notifier *pn,
2259 struct task_struct *next)
2261 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2263 kvm_arch_vcpu_put(vcpu);
2266 int kvm_init(void *opaque, unsigned int vcpu_size,
2267 struct module *module)
2274 r = kvm_arch_init(opaque);
2278 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2280 if (bad_page == NULL) {
2285 bad_pfn = page_to_pfn(bad_page);
2287 if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2292 r = kvm_arch_hardware_setup();
2296 for_each_online_cpu(cpu) {
2297 smp_call_function_single(cpu,
2298 kvm_arch_check_processor_compat,
2304 on_each_cpu(hardware_enable, NULL, 1);
2305 r = register_cpu_notifier(&kvm_cpu_notifier);
2308 register_reboot_notifier(&kvm_reboot_notifier);
2310 r = sysdev_class_register(&kvm_sysdev_class);
2314 r = sysdev_register(&kvm_sysdev);
2318 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2319 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2320 __alignof__(struct kvm_vcpu),
2322 if (!kvm_vcpu_cache) {
2327 kvm_chardev_ops.owner = module;
2328 kvm_vm_fops.owner = module;
2329 kvm_vcpu_fops.owner = module;
2331 r = misc_register(&kvm_dev);
2333 printk(KERN_ERR "kvm: misc device register failed\n");
2337 kvm_preempt_ops.sched_in = kvm_sched_in;
2338 kvm_preempt_ops.sched_out = kvm_sched_out;
2346 kmem_cache_destroy(kvm_vcpu_cache);
2348 sysdev_unregister(&kvm_sysdev);
2350 sysdev_class_unregister(&kvm_sysdev_class);
2352 unregister_reboot_notifier(&kvm_reboot_notifier);
2353 unregister_cpu_notifier(&kvm_cpu_notifier);
2355 on_each_cpu(hardware_disable, NULL, 1);
2357 kvm_arch_hardware_unsetup();
2359 free_cpumask_var(cpus_hardware_enabled);
2361 __free_page(bad_page);
2368 EXPORT_SYMBOL_GPL(kvm_init);
2372 kvm_trace_cleanup();
2373 misc_deregister(&kvm_dev);
2374 kmem_cache_destroy(kvm_vcpu_cache);
2375 sysdev_unregister(&kvm_sysdev);
2376 sysdev_class_unregister(&kvm_sysdev_class);
2377 unregister_reboot_notifier(&kvm_reboot_notifier);
2378 unregister_cpu_notifier(&kvm_cpu_notifier);
2379 on_each_cpu(hardware_disable, NULL, 1);
2380 kvm_arch_hardware_unsetup();
2383 free_cpumask_var(cpus_hardware_enabled);
2384 __free_page(bad_page);
2386 EXPORT_SYMBOL_GPL(kvm_exit);