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>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
47 #include <asm/processor.h>
49 #include <asm/uaccess.h>
50 #include <asm/pgtable.h>
52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53 #include "coalesced_mmio.h"
56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
57 #include <linux/pci.h>
58 #include <linux/interrupt.h>
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/kvm.h>
65 MODULE_AUTHOR("Qumranet");
66 MODULE_LICENSE("GPL");
71 * kvm->slots_lock --> kvm->lock --> kvm->irq_lock
74 DEFINE_SPINLOCK(kvm_lock);
77 static cpumask_var_t cpus_hardware_enabled;
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
84 struct dentry *kvm_debugfs_dir;
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
89 static bool kvm_rebooting;
91 static bool largepages_enabled = true;
93 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
94 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
97 struct list_head *ptr;
98 struct kvm_assigned_dev_kernel *match;
100 list_for_each(ptr, head) {
101 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
102 if (match->assigned_dev_id == assigned_dev_id)
108 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
109 *assigned_dev, int irq)
112 struct msix_entry *host_msix_entries;
114 host_msix_entries = assigned_dev->host_msix_entries;
117 for (i = 0; i < assigned_dev->entries_nr; i++)
118 if (irq == host_msix_entries[i].vector) {
123 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
130 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
132 struct kvm_assigned_dev_kernel *assigned_dev;
136 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
138 kvm = assigned_dev->kvm;
140 mutex_lock(&kvm->irq_lock);
141 spin_lock_irq(&assigned_dev->assigned_dev_lock);
142 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
143 struct kvm_guest_msix_entry *guest_entries =
144 assigned_dev->guest_msix_entries;
145 for (i = 0; i < assigned_dev->entries_nr; i++) {
146 if (!(guest_entries[i].flags &
147 KVM_ASSIGNED_MSIX_PENDING))
149 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
150 kvm_set_irq(assigned_dev->kvm,
151 assigned_dev->irq_source_id,
152 guest_entries[i].vector, 1);
155 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
156 assigned_dev->guest_irq, 1);
158 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
159 mutex_unlock(&assigned_dev->kvm->irq_lock);
162 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
165 struct kvm_assigned_dev_kernel *assigned_dev =
166 (struct kvm_assigned_dev_kernel *) dev_id;
168 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
169 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
170 int index = find_index_from_host_irq(assigned_dev, irq);
173 assigned_dev->guest_msix_entries[index].flags |=
174 KVM_ASSIGNED_MSIX_PENDING;
177 schedule_work(&assigned_dev->interrupt_work);
179 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
180 disable_irq_nosync(irq);
181 assigned_dev->host_irq_disabled = true;
185 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
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;
198 dev = container_of(kian, struct kvm_assigned_dev_kernel,
201 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
203 /* The guest irq may be shared so this ack may be
204 * from another device.
206 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
207 if (dev->host_irq_disabled) {
208 enable_irq(dev->host_irq);
209 dev->host_irq_disabled = false;
211 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
214 static void deassign_guest_irq(struct kvm *kvm,
215 struct kvm_assigned_dev_kernel *assigned_dev)
217 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
218 assigned_dev->ack_notifier.gsi = -1;
220 if (assigned_dev->irq_source_id != -1)
221 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
222 assigned_dev->irq_source_id = -1;
223 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
226 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
227 static void deassign_host_irq(struct kvm *kvm,
228 struct kvm_assigned_dev_kernel *assigned_dev)
231 * In kvm_free_device_irq, cancel_work_sync return true if:
232 * 1. work is scheduled, and then cancelled.
233 * 2. work callback is executed.
235 * The first one ensured that the irq is disabled and no more events
236 * would happen. But for the second one, the irq may be enabled (e.g.
237 * for MSI). So we disable irq here to prevent further events.
239 * Notice this maybe result in nested disable if the interrupt type is
240 * INTx, but it's OK for we are going to free it.
242 * If this function is a part of VM destroy, please ensure that till
243 * now, the kvm state is still legal for probably we also have to wait
244 * interrupt_work done.
246 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
248 for (i = 0; i < assigned_dev->entries_nr; i++)
249 disable_irq_nosync(assigned_dev->
250 host_msix_entries[i].vector);
252 cancel_work_sync(&assigned_dev->interrupt_work);
254 for (i = 0; i < assigned_dev->entries_nr; i++)
255 free_irq(assigned_dev->host_msix_entries[i].vector,
256 (void *)assigned_dev);
258 assigned_dev->entries_nr = 0;
259 kfree(assigned_dev->host_msix_entries);
260 kfree(assigned_dev->guest_msix_entries);
261 pci_disable_msix(assigned_dev->dev);
263 /* Deal with MSI and INTx */
264 disable_irq_nosync(assigned_dev->host_irq);
265 cancel_work_sync(&assigned_dev->interrupt_work);
267 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
269 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
270 pci_disable_msi(assigned_dev->dev);
273 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
276 static int kvm_deassign_irq(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *assigned_dev,
278 unsigned long irq_requested_type)
280 unsigned long guest_irq_type, host_irq_type;
282 if (!irqchip_in_kernel(kvm))
284 /* no irq assignment to deassign */
285 if (!assigned_dev->irq_requested_type)
288 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
289 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
292 deassign_host_irq(kvm, assigned_dev);
294 deassign_guest_irq(kvm, assigned_dev);
299 static void kvm_free_assigned_irq(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *assigned_dev)
302 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
305 static void kvm_free_assigned_device(struct kvm *kvm,
306 struct kvm_assigned_dev_kernel
309 kvm_free_assigned_irq(kvm, assigned_dev);
311 pci_reset_function(assigned_dev->dev);
313 pci_release_regions(assigned_dev->dev);
314 pci_disable_device(assigned_dev->dev);
315 pci_dev_put(assigned_dev->dev);
317 list_del(&assigned_dev->list);
321 void kvm_free_all_assigned_devices(struct kvm *kvm)
323 struct list_head *ptr, *ptr2;
324 struct kvm_assigned_dev_kernel *assigned_dev;
326 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
327 assigned_dev = list_entry(ptr,
328 struct kvm_assigned_dev_kernel,
331 kvm_free_assigned_device(kvm, assigned_dev);
335 static int assigned_device_enable_host_intx(struct kvm *kvm,
336 struct kvm_assigned_dev_kernel *dev)
338 dev->host_irq = dev->dev->irq;
339 /* Even though this is PCI, we don't want to use shared
340 * interrupts. Sharing host devices with guest-assigned devices
341 * on the same interrupt line is not a happy situation: there
342 * are going to be long delays in accepting, acking, etc.
344 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
345 0, "kvm_assigned_intx_device", (void *)dev))
350 #ifdef __KVM_HAVE_MSI
351 static int assigned_device_enable_host_msi(struct kvm *kvm,
352 struct kvm_assigned_dev_kernel *dev)
356 if (!dev->dev->msi_enabled) {
357 r = pci_enable_msi(dev->dev);
362 dev->host_irq = dev->dev->irq;
363 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
364 "kvm_assigned_msi_device", (void *)dev)) {
365 pci_disable_msi(dev->dev);
373 #ifdef __KVM_HAVE_MSIX
374 static int assigned_device_enable_host_msix(struct kvm *kvm,
375 struct kvm_assigned_dev_kernel *dev)
379 /* host_msix_entries and guest_msix_entries should have been
381 if (dev->entries_nr == 0)
384 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
388 for (i = 0; i < dev->entries_nr; i++) {
389 r = request_irq(dev->host_msix_entries[i].vector,
390 kvm_assigned_dev_intr, 0,
391 "kvm_assigned_msix_device",
393 /* FIXME: free requested_irq's on failure */
403 static int assigned_device_enable_guest_intx(struct kvm *kvm,
404 struct kvm_assigned_dev_kernel *dev,
405 struct kvm_assigned_irq *irq)
407 dev->guest_irq = irq->guest_irq;
408 dev->ack_notifier.gsi = irq->guest_irq;
412 #ifdef __KVM_HAVE_MSI
413 static int assigned_device_enable_guest_msi(struct kvm *kvm,
414 struct kvm_assigned_dev_kernel *dev,
415 struct kvm_assigned_irq *irq)
417 dev->guest_irq = irq->guest_irq;
418 dev->ack_notifier.gsi = -1;
419 dev->host_irq_disabled = false;
423 #ifdef __KVM_HAVE_MSIX
424 static int assigned_device_enable_guest_msix(struct kvm *kvm,
425 struct kvm_assigned_dev_kernel *dev,
426 struct kvm_assigned_irq *irq)
428 dev->guest_irq = irq->guest_irq;
429 dev->ack_notifier.gsi = -1;
430 dev->host_irq_disabled = false;
435 static int assign_host_irq(struct kvm *kvm,
436 struct kvm_assigned_dev_kernel *dev,
441 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
444 switch (host_irq_type) {
445 case KVM_DEV_IRQ_HOST_INTX:
446 r = assigned_device_enable_host_intx(kvm, dev);
448 #ifdef __KVM_HAVE_MSI
449 case KVM_DEV_IRQ_HOST_MSI:
450 r = assigned_device_enable_host_msi(kvm, dev);
453 #ifdef __KVM_HAVE_MSIX
454 case KVM_DEV_IRQ_HOST_MSIX:
455 r = assigned_device_enable_host_msix(kvm, dev);
463 dev->irq_requested_type |= host_irq_type;
468 static int assign_guest_irq(struct kvm *kvm,
469 struct kvm_assigned_dev_kernel *dev,
470 struct kvm_assigned_irq *irq,
471 unsigned long guest_irq_type)
476 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
479 id = kvm_request_irq_source_id(kvm);
483 dev->irq_source_id = id;
485 switch (guest_irq_type) {
486 case KVM_DEV_IRQ_GUEST_INTX:
487 r = assigned_device_enable_guest_intx(kvm, dev, irq);
489 #ifdef __KVM_HAVE_MSI
490 case KVM_DEV_IRQ_GUEST_MSI:
491 r = assigned_device_enable_guest_msi(kvm, dev, irq);
494 #ifdef __KVM_HAVE_MSIX
495 case KVM_DEV_IRQ_GUEST_MSIX:
496 r = assigned_device_enable_guest_msix(kvm, dev, irq);
504 dev->irq_requested_type |= guest_irq_type;
505 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
507 kvm_free_irq_source_id(kvm, dev->irq_source_id);
512 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
513 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
514 struct kvm_assigned_irq *assigned_irq)
517 struct kvm_assigned_dev_kernel *match;
518 unsigned long host_irq_type, guest_irq_type;
520 if (!capable(CAP_SYS_RAWIO))
523 if (!irqchip_in_kernel(kvm))
526 mutex_lock(&kvm->lock);
528 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
529 assigned_irq->assigned_dev_id);
533 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
534 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
537 /* can only assign one type at a time */
538 if (hweight_long(host_irq_type) > 1)
540 if (hweight_long(guest_irq_type) > 1)
542 if (host_irq_type == 0 && guest_irq_type == 0)
547 r = assign_host_irq(kvm, match, host_irq_type);
552 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
554 mutex_unlock(&kvm->lock);
558 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
559 struct kvm_assigned_irq
563 struct kvm_assigned_dev_kernel *match;
565 mutex_lock(&kvm->lock);
567 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
568 assigned_irq->assigned_dev_id);
572 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
574 mutex_unlock(&kvm->lock);
578 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
579 struct kvm_assigned_pci_dev *assigned_dev)
582 struct kvm_assigned_dev_kernel *match;
585 down_read(&kvm->slots_lock);
586 mutex_lock(&kvm->lock);
588 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
589 assigned_dev->assigned_dev_id);
591 /* device already assigned */
596 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
598 printk(KERN_INFO "%s: Couldn't allocate memory\n",
603 dev = pci_get_bus_and_slot(assigned_dev->busnr,
604 assigned_dev->devfn);
606 printk(KERN_INFO "%s: host device not found\n", __func__);
610 if (pci_enable_device(dev)) {
611 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
615 r = pci_request_regions(dev, "kvm_assigned_device");
617 printk(KERN_INFO "%s: Could not get access to device regions\n",
622 pci_reset_function(dev);
624 match->assigned_dev_id = assigned_dev->assigned_dev_id;
625 match->host_busnr = assigned_dev->busnr;
626 match->host_devfn = assigned_dev->devfn;
627 match->flags = assigned_dev->flags;
629 spin_lock_init(&match->assigned_dev_lock);
630 match->irq_source_id = -1;
632 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
633 INIT_WORK(&match->interrupt_work,
634 kvm_assigned_dev_interrupt_work_handler);
636 list_add(&match->list, &kvm->arch.assigned_dev_head);
638 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
639 if (!kvm->arch.iommu_domain) {
640 r = kvm_iommu_map_guest(kvm);
644 r = kvm_assign_device(kvm, match);
650 mutex_unlock(&kvm->lock);
651 up_read(&kvm->slots_lock);
654 list_del(&match->list);
655 pci_release_regions(dev);
657 pci_disable_device(dev);
662 mutex_unlock(&kvm->lock);
663 up_read(&kvm->slots_lock);
668 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
669 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
670 struct kvm_assigned_pci_dev *assigned_dev)
673 struct kvm_assigned_dev_kernel *match;
675 mutex_lock(&kvm->lock);
677 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
678 assigned_dev->assigned_dev_id);
680 printk(KERN_INFO "%s: device hasn't been assigned before, "
681 "so cannot be deassigned\n", __func__);
686 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
687 kvm_deassign_device(kvm, match);
689 kvm_free_assigned_device(kvm, match);
692 mutex_unlock(&kvm->lock);
697 inline int kvm_is_mmio_pfn(pfn_t pfn)
699 if (pfn_valid(pfn)) {
700 struct page *page = compound_head(pfn_to_page(pfn));
701 return PageReserved(page);
708 * Switches to specified vcpu, until a matching vcpu_put()
710 void vcpu_load(struct kvm_vcpu *vcpu)
714 mutex_lock(&vcpu->mutex);
716 preempt_notifier_register(&vcpu->preempt_notifier);
717 kvm_arch_vcpu_load(vcpu, cpu);
721 void vcpu_put(struct kvm_vcpu *vcpu)
724 kvm_arch_vcpu_put(vcpu);
725 preempt_notifier_unregister(&vcpu->preempt_notifier);
727 mutex_unlock(&vcpu->mutex);
730 static void ack_flush(void *_completed)
734 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
739 struct kvm_vcpu *vcpu;
741 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
743 spin_lock(&kvm->requests_lock);
744 me = smp_processor_id();
745 kvm_for_each_vcpu(i, vcpu, kvm) {
746 if (test_and_set_bit(req, &vcpu->requests))
749 if (cpus != NULL && cpu != -1 && cpu != me)
750 cpumask_set_cpu(cpu, cpus);
752 if (unlikely(cpus == NULL))
753 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
754 else if (!cpumask_empty(cpus))
755 smp_call_function_many(cpus, ack_flush, NULL, 1);
758 spin_unlock(&kvm->requests_lock);
759 free_cpumask_var(cpus);
763 void kvm_flush_remote_tlbs(struct kvm *kvm)
765 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
766 ++kvm->stat.remote_tlb_flush;
769 void kvm_reload_remote_mmus(struct kvm *kvm)
771 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
774 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
779 mutex_init(&vcpu->mutex);
783 init_waitqueue_head(&vcpu->wq);
785 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
790 vcpu->run = page_address(page);
792 r = kvm_arch_vcpu_init(vcpu);
798 free_page((unsigned long)vcpu->run);
802 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
804 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
806 kvm_arch_vcpu_uninit(vcpu);
807 free_page((unsigned long)vcpu->run);
809 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
812 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
814 return container_of(mn, struct kvm, mmu_notifier);
817 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
818 struct mm_struct *mm,
819 unsigned long address)
821 struct kvm *kvm = mmu_notifier_to_kvm(mn);
825 * When ->invalidate_page runs, the linux pte has been zapped
826 * already but the page is still allocated until
827 * ->invalidate_page returns. So if we increase the sequence
828 * here the kvm page fault will notice if the spte can't be
829 * established because the page is going to be freed. If
830 * instead the kvm page fault establishes the spte before
831 * ->invalidate_page runs, kvm_unmap_hva will release it
834 * The sequence increase only need to be seen at spin_unlock
835 * time, and not at spin_lock time.
837 * Increasing the sequence after the spin_unlock would be
838 * unsafe because the kvm page fault could then establish the
839 * pte after kvm_unmap_hva returned, without noticing the page
840 * is going to be freed.
842 spin_lock(&kvm->mmu_lock);
843 kvm->mmu_notifier_seq++;
844 need_tlb_flush = kvm_unmap_hva(kvm, address);
845 spin_unlock(&kvm->mmu_lock);
847 /* we've to flush the tlb before the pages can be freed */
849 kvm_flush_remote_tlbs(kvm);
853 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
854 struct mm_struct *mm,
858 struct kvm *kvm = mmu_notifier_to_kvm(mn);
859 int need_tlb_flush = 0;
861 spin_lock(&kvm->mmu_lock);
863 * The count increase must become visible at unlock time as no
864 * spte can be established without taking the mmu_lock and
865 * count is also read inside the mmu_lock critical section.
867 kvm->mmu_notifier_count++;
868 for (; start < end; start += PAGE_SIZE)
869 need_tlb_flush |= kvm_unmap_hva(kvm, start);
870 spin_unlock(&kvm->mmu_lock);
872 /* we've to flush the tlb before the pages can be freed */
874 kvm_flush_remote_tlbs(kvm);
877 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
878 struct mm_struct *mm,
882 struct kvm *kvm = mmu_notifier_to_kvm(mn);
884 spin_lock(&kvm->mmu_lock);
886 * This sequence increase will notify the kvm page fault that
887 * the page that is going to be mapped in the spte could have
890 kvm->mmu_notifier_seq++;
892 * The above sequence increase must be visible before the
893 * below count decrease but both values are read by the kvm
894 * page fault under mmu_lock spinlock so we don't need to add
895 * a smb_wmb() here in between the two.
897 kvm->mmu_notifier_count--;
898 spin_unlock(&kvm->mmu_lock);
900 BUG_ON(kvm->mmu_notifier_count < 0);
903 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
904 struct mm_struct *mm,
905 unsigned long address)
907 struct kvm *kvm = mmu_notifier_to_kvm(mn);
910 spin_lock(&kvm->mmu_lock);
911 young = kvm_age_hva(kvm, address);
912 spin_unlock(&kvm->mmu_lock);
915 kvm_flush_remote_tlbs(kvm);
920 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
921 struct mm_struct *mm)
923 struct kvm *kvm = mmu_notifier_to_kvm(mn);
924 kvm_arch_flush_shadow(kvm);
927 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
928 .invalidate_page = kvm_mmu_notifier_invalidate_page,
929 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
930 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
931 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
932 .release = kvm_mmu_notifier_release,
934 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
936 static struct kvm *kvm_create_vm(void)
938 struct kvm *kvm = kvm_arch_create_vm();
939 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
945 #ifdef CONFIG_HAVE_KVM_IRQCHIP
946 INIT_LIST_HEAD(&kvm->irq_routing);
947 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
950 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
951 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
954 return ERR_PTR(-ENOMEM);
956 kvm->coalesced_mmio_ring =
957 (struct kvm_coalesced_mmio_ring *)page_address(page);
960 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
963 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
964 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
966 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
975 kvm->mm = current->mm;
976 atomic_inc(&kvm->mm->mm_count);
977 spin_lock_init(&kvm->mmu_lock);
978 spin_lock_init(&kvm->requests_lock);
979 kvm_io_bus_init(&kvm->pio_bus);
980 kvm_eventfd_init(kvm);
981 mutex_init(&kvm->lock);
982 mutex_init(&kvm->irq_lock);
983 kvm_io_bus_init(&kvm->mmio_bus);
984 init_rwsem(&kvm->slots_lock);
985 atomic_set(&kvm->users_count, 1);
986 spin_lock(&kvm_lock);
987 list_add(&kvm->vm_list, &vm_list);
988 spin_unlock(&kvm_lock);
989 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
990 kvm_coalesced_mmio_init(kvm);
997 * Free any memory in @free but not in @dont.
999 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1000 struct kvm_memory_slot *dont)
1004 if (!dont || free->rmap != dont->rmap)
1007 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1008 vfree(free->dirty_bitmap);
1011 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1012 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1013 vfree(free->lpage_info[i]);
1014 free->lpage_info[i] = NULL;
1019 free->dirty_bitmap = NULL;
1023 void kvm_free_physmem(struct kvm *kvm)
1027 for (i = 0; i < kvm->nmemslots; ++i)
1028 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1031 static void kvm_destroy_vm(struct kvm *kvm)
1033 struct mm_struct *mm = kvm->mm;
1035 kvm_arch_sync_events(kvm);
1036 spin_lock(&kvm_lock);
1037 list_del(&kvm->vm_list);
1038 spin_unlock(&kvm_lock);
1039 kvm_free_irq_routing(kvm);
1040 kvm_io_bus_destroy(&kvm->pio_bus);
1041 kvm_io_bus_destroy(&kvm->mmio_bus);
1042 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1043 if (kvm->coalesced_mmio_ring != NULL)
1044 free_page((unsigned long)kvm->coalesced_mmio_ring);
1046 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1047 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1049 kvm_arch_flush_shadow(kvm);
1051 kvm_arch_destroy_vm(kvm);
1055 void kvm_get_kvm(struct kvm *kvm)
1057 atomic_inc(&kvm->users_count);
1059 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1061 void kvm_put_kvm(struct kvm *kvm)
1063 if (atomic_dec_and_test(&kvm->users_count))
1064 kvm_destroy_vm(kvm);
1066 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1069 static int kvm_vm_release(struct inode *inode, struct file *filp)
1071 struct kvm *kvm = filp->private_data;
1073 kvm_irqfd_release(kvm);
1080 * Allocate some memory and give it an address in the guest physical address
1083 * Discontiguous memory is allowed, mostly for framebuffers.
1085 * Must be called holding mmap_sem for write.
1087 int __kvm_set_memory_region(struct kvm *kvm,
1088 struct kvm_userspace_memory_region *mem,
1093 unsigned long npages;
1095 struct kvm_memory_slot *memslot;
1096 struct kvm_memory_slot old, new;
1099 /* General sanity checks */
1100 if (mem->memory_size & (PAGE_SIZE - 1))
1102 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1104 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1106 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1108 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1111 memslot = &kvm->memslots[mem->slot];
1112 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1113 npages = mem->memory_size >> PAGE_SHIFT;
1116 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1118 new = old = *memslot;
1120 new.base_gfn = base_gfn;
1121 new.npages = npages;
1122 new.flags = mem->flags;
1124 /* Disallow changing a memory slot's size. */
1126 if (npages && old.npages && npages != old.npages)
1129 /* Check for overlaps */
1131 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1132 struct kvm_memory_slot *s = &kvm->memslots[i];
1134 if (s == memslot || !s->npages)
1136 if (!((base_gfn + npages <= s->base_gfn) ||
1137 (base_gfn >= s->base_gfn + s->npages)))
1141 /* Free page dirty bitmap if unneeded */
1142 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1143 new.dirty_bitmap = NULL;
1147 /* Allocate if a slot is being created */
1149 if (npages && !new.rmap) {
1150 new.rmap = vmalloc(npages * sizeof(struct page *));
1155 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1157 new.user_alloc = user_alloc;
1159 * hva_to_rmmap() serialzies with the mmu_lock and to be
1160 * safe it has to ignore memslots with !user_alloc &&
1164 new.userspace_addr = mem->userspace_addr;
1166 new.userspace_addr = 0;
1171 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1177 /* Avoid unused variable warning if no large pages */
1180 if (new.lpage_info[i])
1183 lpages = 1 + (base_gfn + npages - 1) /
1184 KVM_PAGES_PER_HPAGE(level);
1185 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1187 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1189 if (!new.lpage_info[i])
1192 memset(new.lpage_info[i], 0,
1193 lpages * sizeof(*new.lpage_info[i]));
1195 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1196 new.lpage_info[i][0].write_count = 1;
1197 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1198 new.lpage_info[i][lpages - 1].write_count = 1;
1199 ugfn = new.userspace_addr >> PAGE_SHIFT;
1201 * If the gfn and userspace address are not aligned wrt each
1202 * other, or if explicitly asked to, disable large page
1203 * support for this slot
1205 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1206 !largepages_enabled)
1207 for (j = 0; j < lpages; ++j)
1208 new.lpage_info[i][j].write_count = 1;
1213 /* Allocate page dirty bitmap if needed */
1214 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1215 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1217 new.dirty_bitmap = vmalloc(dirty_bytes);
1218 if (!new.dirty_bitmap)
1220 memset(new.dirty_bitmap, 0, dirty_bytes);
1222 kvm_arch_flush_shadow(kvm);
1224 #else /* not defined CONFIG_S390 */
1225 new.user_alloc = user_alloc;
1227 new.userspace_addr = mem->userspace_addr;
1228 #endif /* not defined CONFIG_S390 */
1231 kvm_arch_flush_shadow(kvm);
1233 spin_lock(&kvm->mmu_lock);
1234 if (mem->slot >= kvm->nmemslots)
1235 kvm->nmemslots = mem->slot + 1;
1238 spin_unlock(&kvm->mmu_lock);
1240 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1242 spin_lock(&kvm->mmu_lock);
1244 spin_unlock(&kvm->mmu_lock);
1248 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1249 /* Slot deletion case: we have to update the current slot */
1250 spin_lock(&kvm->mmu_lock);
1253 spin_unlock(&kvm->mmu_lock);
1255 /* map the pages in iommu page table */
1256 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1263 kvm_free_physmem_slot(&new, &old);
1268 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1270 int kvm_set_memory_region(struct kvm *kvm,
1271 struct kvm_userspace_memory_region *mem,
1276 down_write(&kvm->slots_lock);
1277 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1278 up_write(&kvm->slots_lock);
1281 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1283 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1285 kvm_userspace_memory_region *mem,
1288 if (mem->slot >= KVM_MEMORY_SLOTS)
1290 return kvm_set_memory_region(kvm, mem, user_alloc);
1293 int kvm_get_dirty_log(struct kvm *kvm,
1294 struct kvm_dirty_log *log, int *is_dirty)
1296 struct kvm_memory_slot *memslot;
1299 unsigned long any = 0;
1302 if (log->slot >= KVM_MEMORY_SLOTS)
1305 memslot = &kvm->memslots[log->slot];
1307 if (!memslot->dirty_bitmap)
1310 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1312 for (i = 0; !any && i < n/sizeof(long); ++i)
1313 any = memslot->dirty_bitmap[i];
1316 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1327 void kvm_disable_largepages(void)
1329 largepages_enabled = false;
1331 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1333 int is_error_page(struct page *page)
1335 return page == bad_page;
1337 EXPORT_SYMBOL_GPL(is_error_page);
1339 int is_error_pfn(pfn_t pfn)
1341 return pfn == bad_pfn;
1343 EXPORT_SYMBOL_GPL(is_error_pfn);
1345 static inline unsigned long bad_hva(void)
1350 int kvm_is_error_hva(unsigned long addr)
1352 return addr == bad_hva();
1354 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1356 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1360 for (i = 0; i < kvm->nmemslots; ++i) {
1361 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1363 if (gfn >= memslot->base_gfn
1364 && gfn < memslot->base_gfn + memslot->npages)
1369 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1371 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1373 gfn = unalias_gfn(kvm, gfn);
1374 return gfn_to_memslot_unaliased(kvm, gfn);
1377 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1381 gfn = unalias_gfn(kvm, gfn);
1382 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1383 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1385 if (gfn >= memslot->base_gfn
1386 && gfn < memslot->base_gfn + memslot->npages)
1391 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1393 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1395 struct kvm_memory_slot *slot;
1397 gfn = unalias_gfn(kvm, gfn);
1398 slot = gfn_to_memslot_unaliased(kvm, gfn);
1401 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1403 EXPORT_SYMBOL_GPL(gfn_to_hva);
1405 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1407 struct page *page[1];
1414 addr = gfn_to_hva(kvm, gfn);
1415 if (kvm_is_error_hva(addr)) {
1417 return page_to_pfn(bad_page);
1420 npages = get_user_pages_fast(addr, 1, 1, page);
1422 if (unlikely(npages != 1)) {
1423 struct vm_area_struct *vma;
1425 down_read(¤t->mm->mmap_sem);
1426 vma = find_vma(current->mm, addr);
1428 if (vma == NULL || addr < vma->vm_start ||
1429 !(vma->vm_flags & VM_PFNMAP)) {
1430 up_read(¤t->mm->mmap_sem);
1432 return page_to_pfn(bad_page);
1435 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1436 up_read(¤t->mm->mmap_sem);
1437 BUG_ON(!kvm_is_mmio_pfn(pfn));
1439 pfn = page_to_pfn(page[0]);
1444 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1446 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1450 pfn = gfn_to_pfn(kvm, gfn);
1451 if (!kvm_is_mmio_pfn(pfn))
1452 return pfn_to_page(pfn);
1454 WARN_ON(kvm_is_mmio_pfn(pfn));
1460 EXPORT_SYMBOL_GPL(gfn_to_page);
1462 void kvm_release_page_clean(struct page *page)
1464 kvm_release_pfn_clean(page_to_pfn(page));
1466 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1468 void kvm_release_pfn_clean(pfn_t pfn)
1470 if (!kvm_is_mmio_pfn(pfn))
1471 put_page(pfn_to_page(pfn));
1473 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1475 void kvm_release_page_dirty(struct page *page)
1477 kvm_release_pfn_dirty(page_to_pfn(page));
1479 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1481 void kvm_release_pfn_dirty(pfn_t pfn)
1483 kvm_set_pfn_dirty(pfn);
1484 kvm_release_pfn_clean(pfn);
1486 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1488 void kvm_set_page_dirty(struct page *page)
1490 kvm_set_pfn_dirty(page_to_pfn(page));
1492 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1494 void kvm_set_pfn_dirty(pfn_t pfn)
1496 if (!kvm_is_mmio_pfn(pfn)) {
1497 struct page *page = pfn_to_page(pfn);
1498 if (!PageReserved(page))
1502 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1504 void kvm_set_pfn_accessed(pfn_t pfn)
1506 if (!kvm_is_mmio_pfn(pfn))
1507 mark_page_accessed(pfn_to_page(pfn));
1509 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1511 void kvm_get_pfn(pfn_t pfn)
1513 if (!kvm_is_mmio_pfn(pfn))
1514 get_page(pfn_to_page(pfn));
1516 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1518 static int next_segment(unsigned long len, int offset)
1520 if (len > PAGE_SIZE - offset)
1521 return PAGE_SIZE - offset;
1526 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1532 addr = gfn_to_hva(kvm, gfn);
1533 if (kvm_is_error_hva(addr))
1535 r = copy_from_user(data, (void __user *)addr + offset, len);
1540 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1542 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1544 gfn_t gfn = gpa >> PAGE_SHIFT;
1546 int offset = offset_in_page(gpa);
1549 while ((seg = next_segment(len, offset)) != 0) {
1550 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1560 EXPORT_SYMBOL_GPL(kvm_read_guest);
1562 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1567 gfn_t gfn = gpa >> PAGE_SHIFT;
1568 int offset = offset_in_page(gpa);
1570 addr = gfn_to_hva(kvm, gfn);
1571 if (kvm_is_error_hva(addr))
1573 pagefault_disable();
1574 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1580 EXPORT_SYMBOL(kvm_read_guest_atomic);
1582 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1583 int offset, int len)
1588 addr = gfn_to_hva(kvm, gfn);
1589 if (kvm_is_error_hva(addr))
1591 r = copy_to_user((void __user *)addr + offset, data, len);
1594 mark_page_dirty(kvm, gfn);
1597 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1599 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1602 gfn_t gfn = gpa >> PAGE_SHIFT;
1604 int offset = offset_in_page(gpa);
1607 while ((seg = next_segment(len, offset)) != 0) {
1608 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1619 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1621 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1623 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1625 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1627 gfn_t gfn = gpa >> PAGE_SHIFT;
1629 int offset = offset_in_page(gpa);
1632 while ((seg = next_segment(len, offset)) != 0) {
1633 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1642 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1644 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1646 struct kvm_memory_slot *memslot;
1648 gfn = unalias_gfn(kvm, gfn);
1649 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1650 if (memslot && memslot->dirty_bitmap) {
1651 unsigned long rel_gfn = gfn - memslot->base_gfn;
1654 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1655 set_bit(rel_gfn, memslot->dirty_bitmap);
1660 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1667 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1669 if (kvm_arch_vcpu_runnable(vcpu)) {
1670 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1673 if (kvm_cpu_has_pending_timer(vcpu))
1675 if (signal_pending(current))
1683 finish_wait(&vcpu->wq, &wait);
1686 void kvm_resched(struct kvm_vcpu *vcpu)
1688 if (!need_resched())
1692 EXPORT_SYMBOL_GPL(kvm_resched);
1694 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1696 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1699 if (vmf->pgoff == 0)
1700 page = virt_to_page(vcpu->run);
1702 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1703 page = virt_to_page(vcpu->arch.pio_data);
1705 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1706 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1707 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1710 return VM_FAULT_SIGBUS;
1716 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1717 .fault = kvm_vcpu_fault,
1720 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1722 vma->vm_ops = &kvm_vcpu_vm_ops;
1726 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1728 struct kvm_vcpu *vcpu = filp->private_data;
1730 kvm_put_kvm(vcpu->kvm);
1734 static struct file_operations kvm_vcpu_fops = {
1735 .release = kvm_vcpu_release,
1736 .unlocked_ioctl = kvm_vcpu_ioctl,
1737 .compat_ioctl = kvm_vcpu_ioctl,
1738 .mmap = kvm_vcpu_mmap,
1742 * Allocates an inode for the vcpu.
1744 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1746 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1750 * Creates some virtual cpus. Good luck creating more than one.
1752 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1755 struct kvm_vcpu *vcpu, *v;
1757 vcpu = kvm_arch_vcpu_create(kvm, id);
1759 return PTR_ERR(vcpu);
1761 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1763 r = kvm_arch_vcpu_setup(vcpu);
1767 mutex_lock(&kvm->lock);
1768 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1773 kvm_for_each_vcpu(r, v, kvm)
1774 if (v->vcpu_id == id) {
1779 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1781 /* Now it's all set up, let userspace reach it */
1783 r = create_vcpu_fd(vcpu);
1789 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1791 atomic_inc(&kvm->online_vcpus);
1793 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1794 if (kvm->bsp_vcpu_id == id)
1795 kvm->bsp_vcpu = vcpu;
1797 mutex_unlock(&kvm->lock);
1801 mutex_unlock(&kvm->lock);
1802 kvm_arch_vcpu_destroy(vcpu);
1806 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1809 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1810 vcpu->sigset_active = 1;
1811 vcpu->sigset = *sigset;
1813 vcpu->sigset_active = 0;
1817 #ifdef __KVM_HAVE_MSIX
1818 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1819 struct kvm_assigned_msix_nr *entry_nr)
1822 struct kvm_assigned_dev_kernel *adev;
1824 mutex_lock(&kvm->lock);
1826 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1827 entry_nr->assigned_dev_id);
1833 if (adev->entries_nr == 0) {
1834 adev->entries_nr = entry_nr->entry_nr;
1835 if (adev->entries_nr == 0 ||
1836 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1841 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1844 if (!adev->host_msix_entries) {
1848 adev->guest_msix_entries = kzalloc(
1849 sizeof(struct kvm_guest_msix_entry) *
1850 entry_nr->entry_nr, GFP_KERNEL);
1851 if (!adev->guest_msix_entries) {
1852 kfree(adev->host_msix_entries);
1856 } else /* Not allowed set MSI-X number twice */
1859 mutex_unlock(&kvm->lock);
1863 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1864 struct kvm_assigned_msix_entry *entry)
1867 struct kvm_assigned_dev_kernel *adev;
1869 mutex_lock(&kvm->lock);
1871 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1872 entry->assigned_dev_id);
1876 goto msix_entry_out;
1879 for (i = 0; i < adev->entries_nr; i++)
1880 if (adev->guest_msix_entries[i].vector == 0 ||
1881 adev->guest_msix_entries[i].entry == entry->entry) {
1882 adev->guest_msix_entries[i].entry = entry->entry;
1883 adev->guest_msix_entries[i].vector = entry->gsi;
1884 adev->host_msix_entries[i].entry = entry->entry;
1887 if (i == adev->entries_nr) {
1889 goto msix_entry_out;
1893 mutex_unlock(&kvm->lock);
1899 static long kvm_vcpu_ioctl(struct file *filp,
1900 unsigned int ioctl, unsigned long arg)
1902 struct kvm_vcpu *vcpu = filp->private_data;
1903 void __user *argp = (void __user *)arg;
1905 struct kvm_fpu *fpu = NULL;
1906 struct kvm_sregs *kvm_sregs = NULL;
1908 if (vcpu->kvm->mm != current->mm)
1915 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1917 case KVM_GET_REGS: {
1918 struct kvm_regs *kvm_regs;
1921 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1924 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1928 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1935 case KVM_SET_REGS: {
1936 struct kvm_regs *kvm_regs;
1939 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1943 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1945 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1953 case KVM_GET_SREGS: {
1954 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1958 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1962 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1967 case KVM_SET_SREGS: {
1968 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1973 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1975 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1981 case KVM_GET_MP_STATE: {
1982 struct kvm_mp_state mp_state;
1984 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1988 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1993 case KVM_SET_MP_STATE: {
1994 struct kvm_mp_state mp_state;
1997 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1999 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2005 case KVM_TRANSLATE: {
2006 struct kvm_translation tr;
2009 if (copy_from_user(&tr, argp, sizeof tr))
2011 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2015 if (copy_to_user(argp, &tr, sizeof tr))
2020 case KVM_SET_GUEST_DEBUG: {
2021 struct kvm_guest_debug dbg;
2024 if (copy_from_user(&dbg, argp, sizeof dbg))
2026 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2032 case KVM_SET_SIGNAL_MASK: {
2033 struct kvm_signal_mask __user *sigmask_arg = argp;
2034 struct kvm_signal_mask kvm_sigmask;
2035 sigset_t sigset, *p;
2040 if (copy_from_user(&kvm_sigmask, argp,
2041 sizeof kvm_sigmask))
2044 if (kvm_sigmask.len != sizeof sigset)
2047 if (copy_from_user(&sigset, sigmask_arg->sigset,
2052 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2056 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2060 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2064 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2070 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2075 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2077 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2084 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2092 static long kvm_vm_ioctl(struct file *filp,
2093 unsigned int ioctl, unsigned long arg)
2095 struct kvm *kvm = filp->private_data;
2096 void __user *argp = (void __user *)arg;
2099 if (kvm->mm != current->mm)
2102 case KVM_CREATE_VCPU:
2103 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2107 case KVM_SET_USER_MEMORY_REGION: {
2108 struct kvm_userspace_memory_region kvm_userspace_mem;
2111 if (copy_from_user(&kvm_userspace_mem, argp,
2112 sizeof kvm_userspace_mem))
2115 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2120 case KVM_GET_DIRTY_LOG: {
2121 struct kvm_dirty_log log;
2124 if (copy_from_user(&log, argp, sizeof log))
2126 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2131 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2132 case KVM_REGISTER_COALESCED_MMIO: {
2133 struct kvm_coalesced_mmio_zone zone;
2135 if (copy_from_user(&zone, argp, sizeof zone))
2138 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2144 case KVM_UNREGISTER_COALESCED_MMIO: {
2145 struct kvm_coalesced_mmio_zone zone;
2147 if (copy_from_user(&zone, argp, sizeof zone))
2150 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2157 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2158 case KVM_ASSIGN_PCI_DEVICE: {
2159 struct kvm_assigned_pci_dev assigned_dev;
2162 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2164 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2169 case KVM_ASSIGN_IRQ: {
2173 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2174 case KVM_ASSIGN_DEV_IRQ: {
2175 struct kvm_assigned_irq assigned_irq;
2178 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2180 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2185 case KVM_DEASSIGN_DEV_IRQ: {
2186 struct kvm_assigned_irq assigned_irq;
2189 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2191 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2198 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2199 case KVM_DEASSIGN_PCI_DEVICE: {
2200 struct kvm_assigned_pci_dev assigned_dev;
2203 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2205 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2211 #ifdef KVM_CAP_IRQ_ROUTING
2212 case KVM_SET_GSI_ROUTING: {
2213 struct kvm_irq_routing routing;
2214 struct kvm_irq_routing __user *urouting;
2215 struct kvm_irq_routing_entry *entries;
2218 if (copy_from_user(&routing, argp, sizeof(routing)))
2221 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2226 entries = vmalloc(routing.nr * sizeof(*entries));
2231 if (copy_from_user(entries, urouting->entries,
2232 routing.nr * sizeof(*entries)))
2233 goto out_free_irq_routing;
2234 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2236 out_free_irq_routing:
2240 #endif /* KVM_CAP_IRQ_ROUTING */
2241 #ifdef __KVM_HAVE_MSIX
2242 case KVM_ASSIGN_SET_MSIX_NR: {
2243 struct kvm_assigned_msix_nr entry_nr;
2245 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2247 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2252 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2253 struct kvm_assigned_msix_entry entry;
2255 if (copy_from_user(&entry, argp, sizeof entry))
2257 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2264 struct kvm_irqfd data;
2267 if (copy_from_user(&data, argp, sizeof data))
2269 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2272 case KVM_IOEVENTFD: {
2273 struct kvm_ioeventfd data;
2276 if (copy_from_user(&data, argp, sizeof data))
2278 r = kvm_ioeventfd(kvm, &data);
2281 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2282 case KVM_SET_BOOT_CPU_ID:
2284 mutex_lock(&kvm->lock);
2285 if (atomic_read(&kvm->online_vcpus) != 0)
2288 kvm->bsp_vcpu_id = arg;
2289 mutex_unlock(&kvm->lock);
2293 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2299 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2301 struct page *page[1];
2304 gfn_t gfn = vmf->pgoff;
2305 struct kvm *kvm = vma->vm_file->private_data;
2307 addr = gfn_to_hva(kvm, gfn);
2308 if (kvm_is_error_hva(addr))
2309 return VM_FAULT_SIGBUS;
2311 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2313 if (unlikely(npages != 1))
2314 return VM_FAULT_SIGBUS;
2316 vmf->page = page[0];
2320 static const struct vm_operations_struct kvm_vm_vm_ops = {
2321 .fault = kvm_vm_fault,
2324 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2326 vma->vm_ops = &kvm_vm_vm_ops;
2330 static struct file_operations kvm_vm_fops = {
2331 .release = kvm_vm_release,
2332 .unlocked_ioctl = kvm_vm_ioctl,
2333 .compat_ioctl = kvm_vm_ioctl,
2334 .mmap = kvm_vm_mmap,
2337 static int kvm_dev_ioctl_create_vm(void)
2342 kvm = kvm_create_vm();
2344 return PTR_ERR(kvm);
2345 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2352 static long kvm_dev_ioctl_check_extension_generic(long arg)
2355 case KVM_CAP_USER_MEMORY:
2356 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2357 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2358 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2359 case KVM_CAP_SET_BOOT_CPU_ID:
2362 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2363 case KVM_CAP_IRQ_ROUTING:
2364 return KVM_MAX_IRQ_ROUTES;
2369 return kvm_dev_ioctl_check_extension(arg);
2372 static long kvm_dev_ioctl(struct file *filp,
2373 unsigned int ioctl, unsigned long arg)
2378 case KVM_GET_API_VERSION:
2382 r = KVM_API_VERSION;
2388 r = kvm_dev_ioctl_create_vm();
2390 case KVM_CHECK_EXTENSION:
2391 r = kvm_dev_ioctl_check_extension_generic(arg);
2393 case KVM_GET_VCPU_MMAP_SIZE:
2397 r = PAGE_SIZE; /* struct kvm_run */
2399 r += PAGE_SIZE; /* pio data page */
2401 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2402 r += PAGE_SIZE; /* coalesced mmio ring page */
2405 case KVM_TRACE_ENABLE:
2406 case KVM_TRACE_PAUSE:
2407 case KVM_TRACE_DISABLE:
2411 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2417 static struct file_operations kvm_chardev_ops = {
2418 .unlocked_ioctl = kvm_dev_ioctl,
2419 .compat_ioctl = kvm_dev_ioctl,
2422 static struct miscdevice kvm_dev = {
2428 static void hardware_enable(void *junk)
2430 int cpu = raw_smp_processor_id();
2432 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2434 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2435 kvm_arch_hardware_enable(NULL);
2438 static void hardware_disable(void *junk)
2440 int cpu = raw_smp_processor_id();
2442 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2444 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2445 kvm_arch_hardware_disable(NULL);
2448 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2453 val &= ~CPU_TASKS_FROZEN;
2456 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2458 hardware_disable(NULL);
2460 case CPU_UP_CANCELED:
2461 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2463 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2466 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2468 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2475 asmlinkage void kvm_handle_fault_on_reboot(void)
2478 /* spin while reset goes on */
2481 /* Fault while not rebooting. We want the trace. */
2484 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2486 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2490 * Some (well, at least mine) BIOSes hang on reboot if
2493 * And Intel TXT required VMX off for all cpu when system shutdown.
2495 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2496 kvm_rebooting = true;
2497 on_each_cpu(hardware_disable, NULL, 1);
2501 static struct notifier_block kvm_reboot_notifier = {
2502 .notifier_call = kvm_reboot,
2506 void kvm_io_bus_init(struct kvm_io_bus *bus)
2508 memset(bus, 0, sizeof(*bus));
2511 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2515 for (i = 0; i < bus->dev_count; i++) {
2516 struct kvm_io_device *pos = bus->devs[i];
2518 kvm_iodevice_destructor(pos);
2522 /* kvm_io_bus_write - called under kvm->slots_lock */
2523 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2524 int len, const void *val)
2527 for (i = 0; i < bus->dev_count; i++)
2528 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2533 /* kvm_io_bus_read - called under kvm->slots_lock */
2534 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2537 for (i = 0; i < bus->dev_count; i++)
2538 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2543 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2544 struct kvm_io_device *dev)
2548 down_write(&kvm->slots_lock);
2549 ret = __kvm_io_bus_register_dev(bus, dev);
2550 up_write(&kvm->slots_lock);
2555 /* An unlocked version. Caller must have write lock on slots_lock. */
2556 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2557 struct kvm_io_device *dev)
2559 if (bus->dev_count > NR_IOBUS_DEVS-1)
2562 bus->devs[bus->dev_count++] = dev;
2567 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2568 struct kvm_io_bus *bus,
2569 struct kvm_io_device *dev)
2571 down_write(&kvm->slots_lock);
2572 __kvm_io_bus_unregister_dev(bus, dev);
2573 up_write(&kvm->slots_lock);
2576 /* An unlocked version. Caller must have write lock on slots_lock. */
2577 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2578 struct kvm_io_device *dev)
2582 for (i = 0; i < bus->dev_count; i++)
2583 if (bus->devs[i] == dev) {
2584 bus->devs[i] = bus->devs[--bus->dev_count];
2589 static struct notifier_block kvm_cpu_notifier = {
2590 .notifier_call = kvm_cpu_hotplug,
2591 .priority = 20, /* must be > scheduler priority */
2594 static int vm_stat_get(void *_offset, u64 *val)
2596 unsigned offset = (long)_offset;
2600 spin_lock(&kvm_lock);
2601 list_for_each_entry(kvm, &vm_list, vm_list)
2602 *val += *(u32 *)((void *)kvm + offset);
2603 spin_unlock(&kvm_lock);
2607 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2609 static int vcpu_stat_get(void *_offset, u64 *val)
2611 unsigned offset = (long)_offset;
2613 struct kvm_vcpu *vcpu;
2617 spin_lock(&kvm_lock);
2618 list_for_each_entry(kvm, &vm_list, vm_list)
2619 kvm_for_each_vcpu(i, vcpu, kvm)
2620 *val += *(u32 *)((void *)vcpu + offset);
2622 spin_unlock(&kvm_lock);
2626 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2628 static struct file_operations *stat_fops[] = {
2629 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2630 [KVM_STAT_VM] = &vm_stat_fops,
2633 static void kvm_init_debug(void)
2635 struct kvm_stats_debugfs_item *p;
2637 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2638 for (p = debugfs_entries; p->name; ++p)
2639 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2640 (void *)(long)p->offset,
2641 stat_fops[p->kind]);
2644 static void kvm_exit_debug(void)
2646 struct kvm_stats_debugfs_item *p;
2648 for (p = debugfs_entries; p->name; ++p)
2649 debugfs_remove(p->dentry);
2650 debugfs_remove(kvm_debugfs_dir);
2653 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2655 hardware_disable(NULL);
2659 static int kvm_resume(struct sys_device *dev)
2661 hardware_enable(NULL);
2665 static struct sysdev_class kvm_sysdev_class = {
2667 .suspend = kvm_suspend,
2668 .resume = kvm_resume,
2671 static struct sys_device kvm_sysdev = {
2673 .cls = &kvm_sysdev_class,
2676 struct page *bad_page;
2680 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2682 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2685 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2687 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2689 kvm_arch_vcpu_load(vcpu, cpu);
2692 static void kvm_sched_out(struct preempt_notifier *pn,
2693 struct task_struct *next)
2695 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2697 kvm_arch_vcpu_put(vcpu);
2700 int kvm_init(void *opaque, unsigned int vcpu_size,
2701 struct module *module)
2708 r = kvm_arch_init(opaque);
2712 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2714 if (bad_page == NULL) {
2719 bad_pfn = page_to_pfn(bad_page);
2721 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2726 r = kvm_arch_hardware_setup();
2730 for_each_online_cpu(cpu) {
2731 smp_call_function_single(cpu,
2732 kvm_arch_check_processor_compat,
2738 on_each_cpu(hardware_enable, NULL, 1);
2739 r = register_cpu_notifier(&kvm_cpu_notifier);
2742 register_reboot_notifier(&kvm_reboot_notifier);
2744 r = sysdev_class_register(&kvm_sysdev_class);
2748 r = sysdev_register(&kvm_sysdev);
2752 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2753 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2754 __alignof__(struct kvm_vcpu),
2756 if (!kvm_vcpu_cache) {
2761 kvm_chardev_ops.owner = module;
2762 kvm_vm_fops.owner = module;
2763 kvm_vcpu_fops.owner = module;
2765 r = misc_register(&kvm_dev);
2767 printk(KERN_ERR "kvm: misc device register failed\n");
2771 kvm_preempt_ops.sched_in = kvm_sched_in;
2772 kvm_preempt_ops.sched_out = kvm_sched_out;
2777 kmem_cache_destroy(kvm_vcpu_cache);
2779 sysdev_unregister(&kvm_sysdev);
2781 sysdev_class_unregister(&kvm_sysdev_class);
2783 unregister_reboot_notifier(&kvm_reboot_notifier);
2784 unregister_cpu_notifier(&kvm_cpu_notifier);
2786 on_each_cpu(hardware_disable, NULL, 1);
2788 kvm_arch_hardware_unsetup();
2790 free_cpumask_var(cpus_hardware_enabled);
2792 __free_page(bad_page);
2799 EXPORT_SYMBOL_GPL(kvm_init);
2803 tracepoint_synchronize_unregister();
2804 misc_deregister(&kvm_dev);
2805 kmem_cache_destroy(kvm_vcpu_cache);
2806 sysdev_unregister(&kvm_sysdev);
2807 sysdev_class_unregister(&kvm_sysdev_class);
2808 unregister_reboot_notifier(&kvm_reboot_notifier);
2809 unregister_cpu_notifier(&kvm_cpu_notifier);
2810 on_each_cpu(hardware_disable, NULL, 1);
2811 kvm_arch_hardware_unsetup();
2814 free_cpumask_var(cpus_hardware_enabled);
2815 __free_page(bad_page);
2817 EXPORT_SYMBOL_GPL(kvm_exit);