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_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);
174 mutex_unlock(&assigned_dev->kvm->lock);
175 kvm_put_kvm(assigned_dev->kvm);
178 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
180 struct kvm_assigned_dev_kernel *assigned_dev =
181 (struct kvm_assigned_dev_kernel *) dev_id;
183 kvm_get_kvm(assigned_dev->kvm);
184 schedule_work(&assigned_dev->interrupt_work);
185 disable_irq_nosync(irq);
189 /* Ack the irq line for an assigned device */
190 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
192 struct kvm_assigned_dev_kernel *dev;
197 dev = container_of(kian, struct kvm_assigned_dev_kernel,
199 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
200 enable_irq(dev->host_irq);
203 static void kvm_free_assigned_irq(struct kvm *kvm,
204 struct kvm_assigned_dev_kernel *assigned_dev)
206 if (!irqchip_in_kernel(kvm))
209 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
211 if (assigned_dev->irq_source_id != -1)
212 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
213 assigned_dev->irq_source_id = -1;
215 if (!assigned_dev->irq_requested_type)
218 if (cancel_work_sync(&assigned_dev->interrupt_work))
219 /* We had pending work. That means we will have to take
220 * care of kvm_put_kvm.
224 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
226 if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
227 pci_disable_msi(assigned_dev->dev);
229 assigned_dev->irq_requested_type = 0;
233 static void kvm_free_assigned_device(struct kvm *kvm,
234 struct kvm_assigned_dev_kernel
237 kvm_free_assigned_irq(kvm, assigned_dev);
239 pci_reset_function(assigned_dev->dev);
241 pci_release_regions(assigned_dev->dev);
242 pci_disable_device(assigned_dev->dev);
243 pci_dev_put(assigned_dev->dev);
245 list_del(&assigned_dev->list);
249 void kvm_free_all_assigned_devices(struct kvm *kvm)
251 struct list_head *ptr, *ptr2;
252 struct kvm_assigned_dev_kernel *assigned_dev;
254 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
255 assigned_dev = list_entry(ptr,
256 struct kvm_assigned_dev_kernel,
259 kvm_free_assigned_device(kvm, assigned_dev);
263 static int assigned_device_update_intx(struct kvm *kvm,
264 struct kvm_assigned_dev_kernel *adev,
265 struct kvm_assigned_irq *airq)
267 adev->guest_irq = airq->guest_irq;
268 adev->ack_notifier.gsi = airq->guest_irq;
270 if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
273 if (irqchip_in_kernel(kvm)) {
275 adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) {
276 free_irq(adev->host_irq, (void *)kvm);
277 pci_disable_msi(adev->dev);
280 if (!capable(CAP_SYS_RAWIO))
284 adev->host_irq = airq->host_irq;
286 adev->host_irq = adev->dev->irq;
288 /* Even though this is PCI, we don't want to use shared
289 * interrupts. Sharing host devices with guest-assigned devices
290 * on the same interrupt line is not a happy situation: there
291 * are going to be long delays in accepting, acking, etc.
293 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
294 0, "kvm_assigned_intx_device", (void *)adev))
298 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
299 KVM_ASSIGNED_DEV_HOST_INTX;
304 static int assigned_device_update_msi(struct kvm *kvm,
305 struct kvm_assigned_dev_kernel *adev,
306 struct kvm_assigned_irq *airq)
310 if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
311 /* x86 don't care upper address of guest msi message addr */
312 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
313 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
314 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
315 adev->guest_msi.data = airq->guest_msi.data;
316 adev->ack_notifier.gsi = -1;
317 } else if (msi2intx) {
318 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
319 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
320 adev->guest_irq = airq->guest_irq;
321 adev->ack_notifier.gsi = airq->guest_irq;
324 if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
327 if (irqchip_in_kernel(kvm)) {
329 if (adev->irq_requested_type &
330 KVM_ASSIGNED_DEV_HOST_INTX)
331 free_irq(adev->host_irq, (void *)adev);
333 r = pci_enable_msi(adev->dev);
338 adev->host_irq = adev->dev->irq;
339 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
340 "kvm_assigned_msi_device", (void *)adev))
345 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
347 adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
352 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
353 struct kvm_assigned_irq
357 struct kvm_assigned_dev_kernel *match;
359 mutex_lock(&kvm->lock);
361 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
362 assigned_irq->assigned_dev_id);
364 mutex_unlock(&kvm->lock);
368 if (!match->irq_requested_type) {
369 INIT_WORK(&match->interrupt_work,
370 kvm_assigned_dev_interrupt_work_handler);
371 if (irqchip_in_kernel(kvm)) {
372 /* Register ack nofitier */
373 match->ack_notifier.gsi = -1;
374 match->ack_notifier.irq_acked =
375 kvm_assigned_dev_ack_irq;
376 kvm_register_irq_ack_notifier(kvm,
377 &match->ack_notifier);
379 /* Request IRQ source ID */
380 r = kvm_request_irq_source_id(kvm);
384 match->irq_source_id = r;
387 /* Determine host device irq type, we can know the
388 * result from dev->msi_enabled */
390 pci_enable_msi(match->dev);
396 (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
397 (msi2intx && match->dev->msi_enabled)) {
399 r = assigned_device_update_msi(kvm, match, assigned_irq);
401 printk(KERN_WARNING "kvm: failed to enable "
408 } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
409 /* Host device IRQ 0 means don't support INTx */
412 "kvm: wait device to enable MSI!\n");
416 "kvm: failed to enable MSI device!\n");
421 /* Non-sharing INTx mode */
422 r = assigned_device_update_intx(kvm, match, assigned_irq);
424 printk(KERN_WARNING "kvm: failed to enable "
430 mutex_unlock(&kvm->lock);
433 mutex_unlock(&kvm->lock);
434 kvm_free_assigned_device(kvm, match);
438 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
439 struct kvm_assigned_pci_dev *assigned_dev)
442 struct kvm_assigned_dev_kernel *match;
445 mutex_lock(&kvm->lock);
447 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
448 assigned_dev->assigned_dev_id);
450 /* device already assigned */
455 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
457 printk(KERN_INFO "%s: Couldn't allocate memory\n",
462 dev = pci_get_bus_and_slot(assigned_dev->busnr,
463 assigned_dev->devfn);
465 printk(KERN_INFO "%s: host device not found\n", __func__);
469 if (pci_enable_device(dev)) {
470 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
474 r = pci_request_regions(dev, "kvm_assigned_device");
476 printk(KERN_INFO "%s: Could not get access to device regions\n",
481 pci_reset_function(dev);
483 match->assigned_dev_id = assigned_dev->assigned_dev_id;
484 match->host_busnr = assigned_dev->busnr;
485 match->host_devfn = assigned_dev->devfn;
487 match->irq_source_id = -1;
490 list_add(&match->list, &kvm->arch.assigned_dev_head);
492 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
493 r = kvm_iommu_map_guest(kvm, match);
499 mutex_unlock(&kvm->lock);
502 list_del(&match->list);
503 pci_release_regions(dev);
505 pci_disable_device(dev);
510 mutex_unlock(&kvm->lock);
515 static inline int valid_vcpu(int n)
517 return likely(n >= 0 && n < KVM_MAX_VCPUS);
520 inline int kvm_is_mmio_pfn(pfn_t pfn)
523 return PageReserved(pfn_to_page(pfn));
529 * Switches to specified vcpu, until a matching vcpu_put()
531 void vcpu_load(struct kvm_vcpu *vcpu)
535 mutex_lock(&vcpu->mutex);
537 preempt_notifier_register(&vcpu->preempt_notifier);
538 kvm_arch_vcpu_load(vcpu, cpu);
542 void vcpu_put(struct kvm_vcpu *vcpu)
545 kvm_arch_vcpu_put(vcpu);
546 preempt_notifier_unregister(&vcpu->preempt_notifier);
548 mutex_unlock(&vcpu->mutex);
551 static void ack_flush(void *_completed)
555 void kvm_flush_remote_tlbs(struct kvm *kvm)
559 struct kvm_vcpu *vcpu;
563 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
564 vcpu = kvm->vcpus[i];
567 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
570 if (cpu != -1 && cpu != me)
573 if (cpus_empty(cpus))
575 ++kvm->stat.remote_tlb_flush;
576 smp_call_function_mask(cpus, ack_flush, NULL, 1);
581 void kvm_reload_remote_mmus(struct kvm *kvm)
585 struct kvm_vcpu *vcpu;
589 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
590 vcpu = kvm->vcpus[i];
593 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
596 if (cpu != -1 && cpu != me)
599 if (cpus_empty(cpus))
601 smp_call_function_mask(cpus, ack_flush, NULL, 1);
607 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
612 mutex_init(&vcpu->mutex);
616 init_waitqueue_head(&vcpu->wq);
618 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
623 vcpu->run = page_address(page);
625 r = kvm_arch_vcpu_init(vcpu);
631 free_page((unsigned long)vcpu->run);
635 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
637 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
639 kvm_arch_vcpu_uninit(vcpu);
640 free_page((unsigned long)vcpu->run);
642 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
644 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
645 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
647 return container_of(mn, struct kvm, mmu_notifier);
650 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
651 struct mm_struct *mm,
652 unsigned long address)
654 struct kvm *kvm = mmu_notifier_to_kvm(mn);
658 * When ->invalidate_page runs, the linux pte has been zapped
659 * already but the page is still allocated until
660 * ->invalidate_page returns. So if we increase the sequence
661 * here the kvm page fault will notice if the spte can't be
662 * established because the page is going to be freed. If
663 * instead the kvm page fault establishes the spte before
664 * ->invalidate_page runs, kvm_unmap_hva will release it
667 * The sequence increase only need to be seen at spin_unlock
668 * time, and not at spin_lock time.
670 * Increasing the sequence after the spin_unlock would be
671 * unsafe because the kvm page fault could then establish the
672 * pte after kvm_unmap_hva returned, without noticing the page
673 * is going to be freed.
675 spin_lock(&kvm->mmu_lock);
676 kvm->mmu_notifier_seq++;
677 need_tlb_flush = kvm_unmap_hva(kvm, address);
678 spin_unlock(&kvm->mmu_lock);
680 /* we've to flush the tlb before the pages can be freed */
682 kvm_flush_remote_tlbs(kvm);
686 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
687 struct mm_struct *mm,
691 struct kvm *kvm = mmu_notifier_to_kvm(mn);
692 int need_tlb_flush = 0;
694 spin_lock(&kvm->mmu_lock);
696 * The count increase must become visible at unlock time as no
697 * spte can be established without taking the mmu_lock and
698 * count is also read inside the mmu_lock critical section.
700 kvm->mmu_notifier_count++;
701 for (; start < end; start += PAGE_SIZE)
702 need_tlb_flush |= kvm_unmap_hva(kvm, start);
703 spin_unlock(&kvm->mmu_lock);
705 /* we've to flush the tlb before the pages can be freed */
707 kvm_flush_remote_tlbs(kvm);
710 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
711 struct mm_struct *mm,
715 struct kvm *kvm = mmu_notifier_to_kvm(mn);
717 spin_lock(&kvm->mmu_lock);
719 * This sequence increase will notify the kvm page fault that
720 * the page that is going to be mapped in the spte could have
723 kvm->mmu_notifier_seq++;
725 * The above sequence increase must be visible before the
726 * below count decrease but both values are read by the kvm
727 * page fault under mmu_lock spinlock so we don't need to add
728 * a smb_wmb() here in between the two.
730 kvm->mmu_notifier_count--;
731 spin_unlock(&kvm->mmu_lock);
733 BUG_ON(kvm->mmu_notifier_count < 0);
736 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
737 struct mm_struct *mm,
738 unsigned long address)
740 struct kvm *kvm = mmu_notifier_to_kvm(mn);
743 spin_lock(&kvm->mmu_lock);
744 young = kvm_age_hva(kvm, address);
745 spin_unlock(&kvm->mmu_lock);
748 kvm_flush_remote_tlbs(kvm);
753 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
754 .invalidate_page = kvm_mmu_notifier_invalidate_page,
755 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
756 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
757 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
759 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
761 static struct kvm *kvm_create_vm(void)
763 struct kvm *kvm = kvm_arch_create_vm();
764 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
771 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
772 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
775 return ERR_PTR(-ENOMEM);
777 kvm->coalesced_mmio_ring =
778 (struct kvm_coalesced_mmio_ring *)page_address(page);
781 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
784 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
785 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
787 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
796 kvm->mm = current->mm;
797 atomic_inc(&kvm->mm->mm_count);
798 spin_lock_init(&kvm->mmu_lock);
799 kvm_io_bus_init(&kvm->pio_bus);
800 mutex_init(&kvm->lock);
801 kvm_io_bus_init(&kvm->mmio_bus);
802 init_rwsem(&kvm->slots_lock);
803 atomic_set(&kvm->users_count, 1);
804 spin_lock(&kvm_lock);
805 list_add(&kvm->vm_list, &vm_list);
806 spin_unlock(&kvm_lock);
807 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
808 kvm_coalesced_mmio_init(kvm);
815 * Free any memory in @free but not in @dont.
817 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
818 struct kvm_memory_slot *dont)
820 if (!dont || free->rmap != dont->rmap)
823 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
824 vfree(free->dirty_bitmap);
826 if (!dont || free->lpage_info != dont->lpage_info)
827 vfree(free->lpage_info);
830 free->dirty_bitmap = NULL;
832 free->lpage_info = NULL;
835 void kvm_free_physmem(struct kvm *kvm)
839 for (i = 0; i < kvm->nmemslots; ++i)
840 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
843 static void kvm_destroy_vm(struct kvm *kvm)
845 struct mm_struct *mm = kvm->mm;
847 spin_lock(&kvm_lock);
848 list_del(&kvm->vm_list);
849 spin_unlock(&kvm_lock);
850 kvm_io_bus_destroy(&kvm->pio_bus);
851 kvm_io_bus_destroy(&kvm->mmio_bus);
852 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
853 if (kvm->coalesced_mmio_ring != NULL)
854 free_page((unsigned long)kvm->coalesced_mmio_ring);
856 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
857 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
859 kvm_arch_destroy_vm(kvm);
863 void kvm_get_kvm(struct kvm *kvm)
865 atomic_inc(&kvm->users_count);
867 EXPORT_SYMBOL_GPL(kvm_get_kvm);
869 void kvm_put_kvm(struct kvm *kvm)
871 if (atomic_dec_and_test(&kvm->users_count))
874 EXPORT_SYMBOL_GPL(kvm_put_kvm);
877 static int kvm_vm_release(struct inode *inode, struct file *filp)
879 struct kvm *kvm = filp->private_data;
886 * Allocate some memory and give it an address in the guest physical address
889 * Discontiguous memory is allowed, mostly for framebuffers.
891 * Must be called holding mmap_sem for write.
893 int __kvm_set_memory_region(struct kvm *kvm,
894 struct kvm_userspace_memory_region *mem,
899 unsigned long npages;
901 struct kvm_memory_slot *memslot;
902 struct kvm_memory_slot old, new;
905 /* General sanity checks */
906 if (mem->memory_size & (PAGE_SIZE - 1))
908 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
910 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
912 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
914 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
917 memslot = &kvm->memslots[mem->slot];
918 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
919 npages = mem->memory_size >> PAGE_SHIFT;
922 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
924 new = old = *memslot;
926 new.base_gfn = base_gfn;
928 new.flags = mem->flags;
930 /* Disallow changing a memory slot's size. */
932 if (npages && old.npages && npages != old.npages)
935 /* Check for overlaps */
937 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
938 struct kvm_memory_slot *s = &kvm->memslots[i];
942 if (!((base_gfn + npages <= s->base_gfn) ||
943 (base_gfn >= s->base_gfn + s->npages)))
947 /* Free page dirty bitmap if unneeded */
948 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
949 new.dirty_bitmap = NULL;
953 /* Allocate if a slot is being created */
955 if (npages && !new.rmap) {
956 new.rmap = vmalloc(npages * sizeof(struct page *));
961 memset(new.rmap, 0, npages * sizeof(*new.rmap));
963 new.user_alloc = user_alloc;
965 * hva_to_rmmap() serialzies with the mmu_lock and to be
966 * safe it has to ignore memslots with !user_alloc &&
970 new.userspace_addr = mem->userspace_addr;
972 new.userspace_addr = 0;
974 if (npages && !new.lpage_info) {
975 int largepages = npages / KVM_PAGES_PER_HPAGE;
976 if (npages % KVM_PAGES_PER_HPAGE)
978 if (base_gfn % KVM_PAGES_PER_HPAGE)
981 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
986 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
988 if (base_gfn % KVM_PAGES_PER_HPAGE)
989 new.lpage_info[0].write_count = 1;
990 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
991 new.lpage_info[largepages-1].write_count = 1;
994 /* Allocate page dirty bitmap if needed */
995 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
996 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
998 new.dirty_bitmap = vmalloc(dirty_bytes);
999 if (!new.dirty_bitmap)
1001 memset(new.dirty_bitmap, 0, dirty_bytes);
1003 #endif /* not defined CONFIG_S390 */
1006 kvm_arch_flush_shadow(kvm);
1008 spin_lock(&kvm->mmu_lock);
1009 if (mem->slot >= kvm->nmemslots)
1010 kvm->nmemslots = mem->slot + 1;
1013 spin_unlock(&kvm->mmu_lock);
1015 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1017 spin_lock(&kvm->mmu_lock);
1019 spin_unlock(&kvm->mmu_lock);
1023 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1024 /* Slot deletion case: we have to update the current slot */
1028 /* map the pages in iommu page table */
1029 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1036 kvm_free_physmem_slot(&new, &old);
1041 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1043 int kvm_set_memory_region(struct kvm *kvm,
1044 struct kvm_userspace_memory_region *mem,
1049 down_write(&kvm->slots_lock);
1050 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1051 up_write(&kvm->slots_lock);
1054 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1056 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1058 kvm_userspace_memory_region *mem,
1061 if (mem->slot >= KVM_MEMORY_SLOTS)
1063 return kvm_set_memory_region(kvm, mem, user_alloc);
1066 int kvm_get_dirty_log(struct kvm *kvm,
1067 struct kvm_dirty_log *log, int *is_dirty)
1069 struct kvm_memory_slot *memslot;
1072 unsigned long any = 0;
1075 if (log->slot >= KVM_MEMORY_SLOTS)
1078 memslot = &kvm->memslots[log->slot];
1080 if (!memslot->dirty_bitmap)
1083 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1085 for (i = 0; !any && i < n/sizeof(long); ++i)
1086 any = memslot->dirty_bitmap[i];
1089 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1100 int is_error_page(struct page *page)
1102 return page == bad_page;
1104 EXPORT_SYMBOL_GPL(is_error_page);
1106 int is_error_pfn(pfn_t pfn)
1108 return pfn == bad_pfn;
1110 EXPORT_SYMBOL_GPL(is_error_pfn);
1112 static inline unsigned long bad_hva(void)
1117 int kvm_is_error_hva(unsigned long addr)
1119 return addr == bad_hva();
1121 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1123 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1127 for (i = 0; i < kvm->nmemslots; ++i) {
1128 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1130 if (gfn >= memslot->base_gfn
1131 && gfn < memslot->base_gfn + memslot->npages)
1136 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1138 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1140 gfn = unalias_gfn(kvm, gfn);
1141 return gfn_to_memslot_unaliased(kvm, gfn);
1144 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1148 gfn = unalias_gfn(kvm, gfn);
1149 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1150 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1152 if (gfn >= memslot->base_gfn
1153 && gfn < memslot->base_gfn + memslot->npages)
1158 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1160 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1162 struct kvm_memory_slot *slot;
1164 gfn = unalias_gfn(kvm, gfn);
1165 slot = gfn_to_memslot_unaliased(kvm, gfn);
1168 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1170 EXPORT_SYMBOL_GPL(gfn_to_hva);
1172 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1174 struct page *page[1];
1181 addr = gfn_to_hva(kvm, gfn);
1182 if (kvm_is_error_hva(addr)) {
1184 return page_to_pfn(bad_page);
1187 npages = get_user_pages_fast(addr, 1, 1, page);
1189 if (unlikely(npages != 1)) {
1190 struct vm_area_struct *vma;
1192 down_read(¤t->mm->mmap_sem);
1193 vma = find_vma(current->mm, addr);
1195 if (vma == NULL || addr < vma->vm_start ||
1196 !(vma->vm_flags & VM_PFNMAP)) {
1197 up_read(¤t->mm->mmap_sem);
1199 return page_to_pfn(bad_page);
1202 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1203 up_read(¤t->mm->mmap_sem);
1204 BUG_ON(!kvm_is_mmio_pfn(pfn));
1206 pfn = page_to_pfn(page[0]);
1211 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1213 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1217 pfn = gfn_to_pfn(kvm, gfn);
1218 if (!kvm_is_mmio_pfn(pfn))
1219 return pfn_to_page(pfn);
1221 WARN_ON(kvm_is_mmio_pfn(pfn));
1227 EXPORT_SYMBOL_GPL(gfn_to_page);
1229 void kvm_release_page_clean(struct page *page)
1231 kvm_release_pfn_clean(page_to_pfn(page));
1233 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1235 void kvm_release_pfn_clean(pfn_t pfn)
1237 if (!kvm_is_mmio_pfn(pfn))
1238 put_page(pfn_to_page(pfn));
1240 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1242 void kvm_release_page_dirty(struct page *page)
1244 kvm_release_pfn_dirty(page_to_pfn(page));
1246 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1248 void kvm_release_pfn_dirty(pfn_t pfn)
1250 kvm_set_pfn_dirty(pfn);
1251 kvm_release_pfn_clean(pfn);
1253 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1255 void kvm_set_page_dirty(struct page *page)
1257 kvm_set_pfn_dirty(page_to_pfn(page));
1259 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1261 void kvm_set_pfn_dirty(pfn_t pfn)
1263 if (!kvm_is_mmio_pfn(pfn)) {
1264 struct page *page = pfn_to_page(pfn);
1265 if (!PageReserved(page))
1269 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1271 void kvm_set_pfn_accessed(pfn_t pfn)
1273 if (!kvm_is_mmio_pfn(pfn))
1274 mark_page_accessed(pfn_to_page(pfn));
1276 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1278 void kvm_get_pfn(pfn_t pfn)
1280 if (!kvm_is_mmio_pfn(pfn))
1281 get_page(pfn_to_page(pfn));
1283 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1285 static int next_segment(unsigned long len, int offset)
1287 if (len > PAGE_SIZE - offset)
1288 return PAGE_SIZE - offset;
1293 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1299 addr = gfn_to_hva(kvm, gfn);
1300 if (kvm_is_error_hva(addr))
1302 r = copy_from_user(data, (void __user *)addr + offset, len);
1307 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1309 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1311 gfn_t gfn = gpa >> PAGE_SHIFT;
1313 int offset = offset_in_page(gpa);
1316 while ((seg = next_segment(len, offset)) != 0) {
1317 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1327 EXPORT_SYMBOL_GPL(kvm_read_guest);
1329 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1334 gfn_t gfn = gpa >> PAGE_SHIFT;
1335 int offset = offset_in_page(gpa);
1337 addr = gfn_to_hva(kvm, gfn);
1338 if (kvm_is_error_hva(addr))
1340 pagefault_disable();
1341 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1347 EXPORT_SYMBOL(kvm_read_guest_atomic);
1349 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1350 int offset, int len)
1355 addr = gfn_to_hva(kvm, gfn);
1356 if (kvm_is_error_hva(addr))
1358 r = copy_to_user((void __user *)addr + offset, data, len);
1361 mark_page_dirty(kvm, gfn);
1364 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1366 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1369 gfn_t gfn = gpa >> PAGE_SHIFT;
1371 int offset = offset_in_page(gpa);
1374 while ((seg = next_segment(len, offset)) != 0) {
1375 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1386 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1388 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1390 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1392 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1394 gfn_t gfn = gpa >> PAGE_SHIFT;
1396 int offset = offset_in_page(gpa);
1399 while ((seg = next_segment(len, offset)) != 0) {
1400 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1409 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1411 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1413 struct kvm_memory_slot *memslot;
1415 gfn = unalias_gfn(kvm, gfn);
1416 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1417 if (memslot && memslot->dirty_bitmap) {
1418 unsigned long rel_gfn = gfn - memslot->base_gfn;
1421 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1422 set_bit(rel_gfn, memslot->dirty_bitmap);
1427 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1429 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1434 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1436 if (kvm_cpu_has_interrupt(vcpu) ||
1437 kvm_cpu_has_pending_timer(vcpu) ||
1438 kvm_arch_vcpu_runnable(vcpu)) {
1439 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1442 if (signal_pending(current))
1450 finish_wait(&vcpu->wq, &wait);
1453 void kvm_resched(struct kvm_vcpu *vcpu)
1455 if (!need_resched())
1459 EXPORT_SYMBOL_GPL(kvm_resched);
1461 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1463 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1466 if (vmf->pgoff == 0)
1467 page = virt_to_page(vcpu->run);
1469 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1470 page = virt_to_page(vcpu->arch.pio_data);
1472 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1473 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1474 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1477 return VM_FAULT_SIGBUS;
1483 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1484 .fault = kvm_vcpu_fault,
1487 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1489 vma->vm_ops = &kvm_vcpu_vm_ops;
1493 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1495 struct kvm_vcpu *vcpu = filp->private_data;
1497 kvm_put_kvm(vcpu->kvm);
1501 static const struct file_operations kvm_vcpu_fops = {
1502 .release = kvm_vcpu_release,
1503 .unlocked_ioctl = kvm_vcpu_ioctl,
1504 .compat_ioctl = kvm_vcpu_ioctl,
1505 .mmap = kvm_vcpu_mmap,
1509 * Allocates an inode for the vcpu.
1511 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1513 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1515 kvm_put_kvm(vcpu->kvm);
1520 * Creates some virtual cpus. Good luck creating more than one.
1522 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1525 struct kvm_vcpu *vcpu;
1530 vcpu = kvm_arch_vcpu_create(kvm, n);
1532 return PTR_ERR(vcpu);
1534 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1536 r = kvm_arch_vcpu_setup(vcpu);
1540 mutex_lock(&kvm->lock);
1541 if (kvm->vcpus[n]) {
1545 kvm->vcpus[n] = vcpu;
1546 mutex_unlock(&kvm->lock);
1548 /* Now it's all set up, let userspace reach it */
1550 r = create_vcpu_fd(vcpu);
1556 mutex_lock(&kvm->lock);
1557 kvm->vcpus[n] = NULL;
1559 mutex_unlock(&kvm->lock);
1560 kvm_arch_vcpu_destroy(vcpu);
1564 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1567 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1568 vcpu->sigset_active = 1;
1569 vcpu->sigset = *sigset;
1571 vcpu->sigset_active = 0;
1575 static long kvm_vcpu_ioctl(struct file *filp,
1576 unsigned int ioctl, unsigned long arg)
1578 struct kvm_vcpu *vcpu = filp->private_data;
1579 void __user *argp = (void __user *)arg;
1581 struct kvm_fpu *fpu = NULL;
1582 struct kvm_sregs *kvm_sregs = NULL;
1584 if (vcpu->kvm->mm != current->mm)
1591 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1593 case KVM_GET_REGS: {
1594 struct kvm_regs *kvm_regs;
1597 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1600 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1604 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1611 case KVM_SET_REGS: {
1612 struct kvm_regs *kvm_regs;
1615 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1619 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1621 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1629 case KVM_GET_SREGS: {
1630 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1634 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1638 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1643 case KVM_SET_SREGS: {
1644 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1649 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1651 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1657 case KVM_GET_MP_STATE: {
1658 struct kvm_mp_state mp_state;
1660 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1664 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1669 case KVM_SET_MP_STATE: {
1670 struct kvm_mp_state mp_state;
1673 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1675 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1681 case KVM_TRANSLATE: {
1682 struct kvm_translation tr;
1685 if (copy_from_user(&tr, argp, sizeof tr))
1687 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1691 if (copy_to_user(argp, &tr, sizeof tr))
1696 case KVM_DEBUG_GUEST: {
1697 struct kvm_debug_guest dbg;
1700 if (copy_from_user(&dbg, argp, sizeof dbg))
1702 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1708 case KVM_SET_SIGNAL_MASK: {
1709 struct kvm_signal_mask __user *sigmask_arg = argp;
1710 struct kvm_signal_mask kvm_sigmask;
1711 sigset_t sigset, *p;
1716 if (copy_from_user(&kvm_sigmask, argp,
1717 sizeof kvm_sigmask))
1720 if (kvm_sigmask.len != sizeof sigset)
1723 if (copy_from_user(&sigset, sigmask_arg->sigset,
1728 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1732 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1736 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1740 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1746 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1751 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1753 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1760 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1768 static long kvm_vm_ioctl(struct file *filp,
1769 unsigned int ioctl, unsigned long arg)
1771 struct kvm *kvm = filp->private_data;
1772 void __user *argp = (void __user *)arg;
1775 if (kvm->mm != current->mm)
1778 case KVM_CREATE_VCPU:
1779 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1783 case KVM_SET_USER_MEMORY_REGION: {
1784 struct kvm_userspace_memory_region kvm_userspace_mem;
1787 if (copy_from_user(&kvm_userspace_mem, argp,
1788 sizeof kvm_userspace_mem))
1791 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1796 case KVM_GET_DIRTY_LOG: {
1797 struct kvm_dirty_log log;
1800 if (copy_from_user(&log, argp, sizeof log))
1802 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1807 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1808 case KVM_REGISTER_COALESCED_MMIO: {
1809 struct kvm_coalesced_mmio_zone zone;
1811 if (copy_from_user(&zone, argp, sizeof zone))
1814 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1820 case KVM_UNREGISTER_COALESCED_MMIO: {
1821 struct kvm_coalesced_mmio_zone zone;
1823 if (copy_from_user(&zone, argp, sizeof zone))
1826 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1833 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1834 case KVM_ASSIGN_PCI_DEVICE: {
1835 struct kvm_assigned_pci_dev assigned_dev;
1838 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1840 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1845 case KVM_ASSIGN_IRQ: {
1846 struct kvm_assigned_irq assigned_irq;
1849 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1851 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1858 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1864 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1866 struct page *page[1];
1869 gfn_t gfn = vmf->pgoff;
1870 struct kvm *kvm = vma->vm_file->private_data;
1872 addr = gfn_to_hva(kvm, gfn);
1873 if (kvm_is_error_hva(addr))
1874 return VM_FAULT_SIGBUS;
1876 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1878 if (unlikely(npages != 1))
1879 return VM_FAULT_SIGBUS;
1881 vmf->page = page[0];
1885 static struct vm_operations_struct kvm_vm_vm_ops = {
1886 .fault = kvm_vm_fault,
1889 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1891 vma->vm_ops = &kvm_vm_vm_ops;
1895 static const struct file_operations kvm_vm_fops = {
1896 .release = kvm_vm_release,
1897 .unlocked_ioctl = kvm_vm_ioctl,
1898 .compat_ioctl = kvm_vm_ioctl,
1899 .mmap = kvm_vm_mmap,
1902 static int kvm_dev_ioctl_create_vm(void)
1907 kvm = kvm_create_vm();
1909 return PTR_ERR(kvm);
1910 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1917 static long kvm_dev_ioctl(struct file *filp,
1918 unsigned int ioctl, unsigned long arg)
1923 case KVM_GET_API_VERSION:
1927 r = KVM_API_VERSION;
1933 r = kvm_dev_ioctl_create_vm();
1935 case KVM_CHECK_EXTENSION:
1936 r = kvm_dev_ioctl_check_extension(arg);
1938 case KVM_GET_VCPU_MMAP_SIZE:
1942 r = PAGE_SIZE; /* struct kvm_run */
1944 r += PAGE_SIZE; /* pio data page */
1946 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1947 r += PAGE_SIZE; /* coalesced mmio ring page */
1950 case KVM_TRACE_ENABLE:
1951 case KVM_TRACE_PAUSE:
1952 case KVM_TRACE_DISABLE:
1953 r = kvm_trace_ioctl(ioctl, arg);
1956 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1962 static struct file_operations kvm_chardev_ops = {
1963 .unlocked_ioctl = kvm_dev_ioctl,
1964 .compat_ioctl = kvm_dev_ioctl,
1967 static struct miscdevice kvm_dev = {
1973 static void hardware_enable(void *junk)
1975 int cpu = raw_smp_processor_id();
1977 if (cpu_isset(cpu, cpus_hardware_enabled))
1979 cpu_set(cpu, cpus_hardware_enabled);
1980 kvm_arch_hardware_enable(NULL);
1983 static void hardware_disable(void *junk)
1985 int cpu = raw_smp_processor_id();
1987 if (!cpu_isset(cpu, cpus_hardware_enabled))
1989 cpu_clear(cpu, cpus_hardware_enabled);
1990 kvm_arch_hardware_disable(NULL);
1993 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1998 val &= ~CPU_TASKS_FROZEN;
2001 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2003 hardware_disable(NULL);
2005 case CPU_UP_CANCELED:
2006 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2008 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2011 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2013 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2020 asmlinkage void kvm_handle_fault_on_reboot(void)
2023 /* spin while reset goes on */
2026 /* Fault while not rebooting. We want the trace. */
2029 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2031 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2034 if (val == SYS_RESTART) {
2036 * Some (well, at least mine) BIOSes hang on reboot if
2039 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2040 kvm_rebooting = true;
2041 on_each_cpu(hardware_disable, NULL, 1);
2046 static struct notifier_block kvm_reboot_notifier = {
2047 .notifier_call = kvm_reboot,
2051 void kvm_io_bus_init(struct kvm_io_bus *bus)
2053 memset(bus, 0, sizeof(*bus));
2056 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2060 for (i = 0; i < bus->dev_count; i++) {
2061 struct kvm_io_device *pos = bus->devs[i];
2063 kvm_iodevice_destructor(pos);
2067 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2068 gpa_t addr, int len, int is_write)
2072 for (i = 0; i < bus->dev_count; i++) {
2073 struct kvm_io_device *pos = bus->devs[i];
2075 if (pos->in_range(pos, addr, len, is_write))
2082 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2084 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2086 bus->devs[bus->dev_count++] = dev;
2089 static struct notifier_block kvm_cpu_notifier = {
2090 .notifier_call = kvm_cpu_hotplug,
2091 .priority = 20, /* must be > scheduler priority */
2094 static int vm_stat_get(void *_offset, u64 *val)
2096 unsigned offset = (long)_offset;
2100 spin_lock(&kvm_lock);
2101 list_for_each_entry(kvm, &vm_list, vm_list)
2102 *val += *(u32 *)((void *)kvm + offset);
2103 spin_unlock(&kvm_lock);
2107 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2109 static int vcpu_stat_get(void *_offset, u64 *val)
2111 unsigned offset = (long)_offset;
2113 struct kvm_vcpu *vcpu;
2117 spin_lock(&kvm_lock);
2118 list_for_each_entry(kvm, &vm_list, vm_list)
2119 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2120 vcpu = kvm->vcpus[i];
2122 *val += *(u32 *)((void *)vcpu + offset);
2124 spin_unlock(&kvm_lock);
2128 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2130 static struct file_operations *stat_fops[] = {
2131 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2132 [KVM_STAT_VM] = &vm_stat_fops,
2135 static void kvm_init_debug(void)
2137 struct kvm_stats_debugfs_item *p;
2139 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2140 for (p = debugfs_entries; p->name; ++p)
2141 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2142 (void *)(long)p->offset,
2143 stat_fops[p->kind]);
2146 static void kvm_exit_debug(void)
2148 struct kvm_stats_debugfs_item *p;
2150 for (p = debugfs_entries; p->name; ++p)
2151 debugfs_remove(p->dentry);
2152 debugfs_remove(kvm_debugfs_dir);
2155 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2157 hardware_disable(NULL);
2161 static int kvm_resume(struct sys_device *dev)
2163 hardware_enable(NULL);
2167 static struct sysdev_class kvm_sysdev_class = {
2169 .suspend = kvm_suspend,
2170 .resume = kvm_resume,
2173 static struct sys_device kvm_sysdev = {
2175 .cls = &kvm_sysdev_class,
2178 struct page *bad_page;
2182 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2184 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2187 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2189 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2191 kvm_arch_vcpu_load(vcpu, cpu);
2194 static void kvm_sched_out(struct preempt_notifier *pn,
2195 struct task_struct *next)
2197 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2199 kvm_arch_vcpu_put(vcpu);
2202 int kvm_init(void *opaque, unsigned int vcpu_size,
2203 struct module *module)
2210 r = kvm_arch_init(opaque);
2214 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2216 if (bad_page == NULL) {
2221 bad_pfn = page_to_pfn(bad_page);
2223 r = kvm_arch_hardware_setup();
2227 for_each_online_cpu(cpu) {
2228 smp_call_function_single(cpu,
2229 kvm_arch_check_processor_compat,
2235 on_each_cpu(hardware_enable, NULL, 1);
2236 r = register_cpu_notifier(&kvm_cpu_notifier);
2239 register_reboot_notifier(&kvm_reboot_notifier);
2241 r = sysdev_class_register(&kvm_sysdev_class);
2245 r = sysdev_register(&kvm_sysdev);
2249 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2250 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2251 __alignof__(struct kvm_vcpu),
2253 if (!kvm_vcpu_cache) {
2258 kvm_chardev_ops.owner = module;
2260 r = misc_register(&kvm_dev);
2262 printk(KERN_ERR "kvm: misc device register failed\n");
2266 kvm_preempt_ops.sched_in = kvm_sched_in;
2267 kvm_preempt_ops.sched_out = kvm_sched_out;
2275 kmem_cache_destroy(kvm_vcpu_cache);
2277 sysdev_unregister(&kvm_sysdev);
2279 sysdev_class_unregister(&kvm_sysdev_class);
2281 unregister_reboot_notifier(&kvm_reboot_notifier);
2282 unregister_cpu_notifier(&kvm_cpu_notifier);
2284 on_each_cpu(hardware_disable, NULL, 1);
2286 kvm_arch_hardware_unsetup();
2288 __free_page(bad_page);
2295 EXPORT_SYMBOL_GPL(kvm_init);
2299 kvm_trace_cleanup();
2300 misc_deregister(&kvm_dev);
2301 kmem_cache_destroy(kvm_vcpu_cache);
2302 sysdev_unregister(&kvm_sysdev);
2303 sysdev_class_unregister(&kvm_sysdev_class);
2304 unregister_reboot_notifier(&kvm_reboot_notifier);
2305 unregister_cpu_notifier(&kvm_cpu_notifier);
2306 on_each_cpu(hardware_disable, NULL, 1);
2307 kvm_arch_hardware_unsetup();
2310 __free_page(bad_page);
2312 EXPORT_SYMBOL_GPL(kvm_exit);