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_change_pte(struct mmu_notifier *mn,
854 struct mm_struct *mm,
855 unsigned long address,
858 struct kvm *kvm = mmu_notifier_to_kvm(mn);
860 spin_lock(&kvm->mmu_lock);
861 kvm->mmu_notifier_seq++;
862 kvm_set_spte_hva(kvm, address, pte);
863 spin_unlock(&kvm->mmu_lock);
866 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
867 struct mm_struct *mm,
871 struct kvm *kvm = mmu_notifier_to_kvm(mn);
872 int need_tlb_flush = 0;
874 spin_lock(&kvm->mmu_lock);
876 * The count increase must become visible at unlock time as no
877 * spte can be established without taking the mmu_lock and
878 * count is also read inside the mmu_lock critical section.
880 kvm->mmu_notifier_count++;
881 for (; start < end; start += PAGE_SIZE)
882 need_tlb_flush |= kvm_unmap_hva(kvm, start);
883 spin_unlock(&kvm->mmu_lock);
885 /* we've to flush the tlb before the pages can be freed */
887 kvm_flush_remote_tlbs(kvm);
890 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
891 struct mm_struct *mm,
895 struct kvm *kvm = mmu_notifier_to_kvm(mn);
897 spin_lock(&kvm->mmu_lock);
899 * This sequence increase will notify the kvm page fault that
900 * the page that is going to be mapped in the spte could have
903 kvm->mmu_notifier_seq++;
905 * The above sequence increase must be visible before the
906 * below count decrease but both values are read by the kvm
907 * page fault under mmu_lock spinlock so we don't need to add
908 * a smb_wmb() here in between the two.
910 kvm->mmu_notifier_count--;
911 spin_unlock(&kvm->mmu_lock);
913 BUG_ON(kvm->mmu_notifier_count < 0);
916 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
917 struct mm_struct *mm,
918 unsigned long address)
920 struct kvm *kvm = mmu_notifier_to_kvm(mn);
923 spin_lock(&kvm->mmu_lock);
924 young = kvm_age_hva(kvm, address);
925 spin_unlock(&kvm->mmu_lock);
928 kvm_flush_remote_tlbs(kvm);
933 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
934 struct mm_struct *mm)
936 struct kvm *kvm = mmu_notifier_to_kvm(mn);
937 kvm_arch_flush_shadow(kvm);
940 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
941 .invalidate_page = kvm_mmu_notifier_invalidate_page,
942 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
943 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
944 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
945 .change_pte = kvm_mmu_notifier_change_pte,
946 .release = kvm_mmu_notifier_release,
948 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
950 static struct kvm *kvm_create_vm(void)
952 struct kvm *kvm = kvm_arch_create_vm();
953 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
959 #ifdef CONFIG_HAVE_KVM_IRQCHIP
960 INIT_LIST_HEAD(&kvm->irq_routing);
961 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
964 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
965 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
968 return ERR_PTR(-ENOMEM);
970 kvm->coalesced_mmio_ring =
971 (struct kvm_coalesced_mmio_ring *)page_address(page);
974 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
977 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
978 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
980 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
989 kvm->mm = current->mm;
990 atomic_inc(&kvm->mm->mm_count);
991 spin_lock_init(&kvm->mmu_lock);
992 spin_lock_init(&kvm->requests_lock);
993 kvm_io_bus_init(&kvm->pio_bus);
994 kvm_eventfd_init(kvm);
995 mutex_init(&kvm->lock);
996 mutex_init(&kvm->irq_lock);
997 kvm_io_bus_init(&kvm->mmio_bus);
998 init_rwsem(&kvm->slots_lock);
999 atomic_set(&kvm->users_count, 1);
1000 spin_lock(&kvm_lock);
1001 list_add(&kvm->vm_list, &vm_list);
1002 spin_unlock(&kvm_lock);
1003 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1004 kvm_coalesced_mmio_init(kvm);
1011 * Free any memory in @free but not in @dont.
1013 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1014 struct kvm_memory_slot *dont)
1018 if (!dont || free->rmap != dont->rmap)
1021 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1022 vfree(free->dirty_bitmap);
1025 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1026 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1027 vfree(free->lpage_info[i]);
1028 free->lpage_info[i] = NULL;
1033 free->dirty_bitmap = NULL;
1037 void kvm_free_physmem(struct kvm *kvm)
1041 for (i = 0; i < kvm->nmemslots; ++i)
1042 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1045 static void kvm_destroy_vm(struct kvm *kvm)
1047 struct mm_struct *mm = kvm->mm;
1049 kvm_arch_sync_events(kvm);
1050 spin_lock(&kvm_lock);
1051 list_del(&kvm->vm_list);
1052 spin_unlock(&kvm_lock);
1053 kvm_free_irq_routing(kvm);
1054 kvm_io_bus_destroy(&kvm->pio_bus);
1055 kvm_io_bus_destroy(&kvm->mmio_bus);
1056 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1057 if (kvm->coalesced_mmio_ring != NULL)
1058 free_page((unsigned long)kvm->coalesced_mmio_ring);
1060 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1061 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1063 kvm_arch_flush_shadow(kvm);
1065 kvm_arch_destroy_vm(kvm);
1069 void kvm_get_kvm(struct kvm *kvm)
1071 atomic_inc(&kvm->users_count);
1073 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1075 void kvm_put_kvm(struct kvm *kvm)
1077 if (atomic_dec_and_test(&kvm->users_count))
1078 kvm_destroy_vm(kvm);
1080 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1083 static int kvm_vm_release(struct inode *inode, struct file *filp)
1085 struct kvm *kvm = filp->private_data;
1087 kvm_irqfd_release(kvm);
1094 * Allocate some memory and give it an address in the guest physical address
1097 * Discontiguous memory is allowed, mostly for framebuffers.
1099 * Must be called holding mmap_sem for write.
1101 int __kvm_set_memory_region(struct kvm *kvm,
1102 struct kvm_userspace_memory_region *mem,
1107 unsigned long npages;
1109 struct kvm_memory_slot *memslot;
1110 struct kvm_memory_slot old, new;
1113 /* General sanity checks */
1114 if (mem->memory_size & (PAGE_SIZE - 1))
1116 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1118 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1120 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1122 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1125 memslot = &kvm->memslots[mem->slot];
1126 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1127 npages = mem->memory_size >> PAGE_SHIFT;
1130 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1132 new = old = *memslot;
1134 new.base_gfn = base_gfn;
1135 new.npages = npages;
1136 new.flags = mem->flags;
1138 /* Disallow changing a memory slot's size. */
1140 if (npages && old.npages && npages != old.npages)
1143 /* Check for overlaps */
1145 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1146 struct kvm_memory_slot *s = &kvm->memslots[i];
1148 if (s == memslot || !s->npages)
1150 if (!((base_gfn + npages <= s->base_gfn) ||
1151 (base_gfn >= s->base_gfn + s->npages)))
1155 /* Free page dirty bitmap if unneeded */
1156 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1157 new.dirty_bitmap = NULL;
1161 /* Allocate if a slot is being created */
1163 if (npages && !new.rmap) {
1164 new.rmap = vmalloc(npages * sizeof(struct page *));
1169 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1171 new.user_alloc = user_alloc;
1173 * hva_to_rmmap() serialzies with the mmu_lock and to be
1174 * safe it has to ignore memslots with !user_alloc &&
1178 new.userspace_addr = mem->userspace_addr;
1180 new.userspace_addr = 0;
1185 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1191 /* Avoid unused variable warning if no large pages */
1194 if (new.lpage_info[i])
1197 lpages = 1 + (base_gfn + npages - 1) /
1198 KVM_PAGES_PER_HPAGE(level);
1199 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1201 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1203 if (!new.lpage_info[i])
1206 memset(new.lpage_info[i], 0,
1207 lpages * sizeof(*new.lpage_info[i]));
1209 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1210 new.lpage_info[i][0].write_count = 1;
1211 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1212 new.lpage_info[i][lpages - 1].write_count = 1;
1213 ugfn = new.userspace_addr >> PAGE_SHIFT;
1215 * If the gfn and userspace address are not aligned wrt each
1216 * other, or if explicitly asked to, disable large page
1217 * support for this slot
1219 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1220 !largepages_enabled)
1221 for (j = 0; j < lpages; ++j)
1222 new.lpage_info[i][j].write_count = 1;
1227 /* Allocate page dirty bitmap if needed */
1228 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1229 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1231 new.dirty_bitmap = vmalloc(dirty_bytes);
1232 if (!new.dirty_bitmap)
1234 memset(new.dirty_bitmap, 0, dirty_bytes);
1236 kvm_arch_flush_shadow(kvm);
1238 #else /* not defined CONFIG_S390 */
1239 new.user_alloc = user_alloc;
1241 new.userspace_addr = mem->userspace_addr;
1242 #endif /* not defined CONFIG_S390 */
1245 kvm_arch_flush_shadow(kvm);
1247 spin_lock(&kvm->mmu_lock);
1248 if (mem->slot >= kvm->nmemslots)
1249 kvm->nmemslots = mem->slot + 1;
1252 spin_unlock(&kvm->mmu_lock);
1254 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1256 spin_lock(&kvm->mmu_lock);
1258 spin_unlock(&kvm->mmu_lock);
1262 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1263 /* Slot deletion case: we have to update the current slot */
1264 spin_lock(&kvm->mmu_lock);
1267 spin_unlock(&kvm->mmu_lock);
1269 /* map the pages in iommu page table */
1270 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1277 kvm_free_physmem_slot(&new, &old);
1282 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1284 int kvm_set_memory_region(struct kvm *kvm,
1285 struct kvm_userspace_memory_region *mem,
1290 down_write(&kvm->slots_lock);
1291 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1292 up_write(&kvm->slots_lock);
1295 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1297 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1299 kvm_userspace_memory_region *mem,
1302 if (mem->slot >= KVM_MEMORY_SLOTS)
1304 return kvm_set_memory_region(kvm, mem, user_alloc);
1307 int kvm_get_dirty_log(struct kvm *kvm,
1308 struct kvm_dirty_log *log, int *is_dirty)
1310 struct kvm_memory_slot *memslot;
1313 unsigned long any = 0;
1316 if (log->slot >= KVM_MEMORY_SLOTS)
1319 memslot = &kvm->memslots[log->slot];
1321 if (!memslot->dirty_bitmap)
1324 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1326 for (i = 0; !any && i < n/sizeof(long); ++i)
1327 any = memslot->dirty_bitmap[i];
1330 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1341 void kvm_disable_largepages(void)
1343 largepages_enabled = false;
1345 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1347 int is_error_page(struct page *page)
1349 return page == bad_page;
1351 EXPORT_SYMBOL_GPL(is_error_page);
1353 int is_error_pfn(pfn_t pfn)
1355 return pfn == bad_pfn;
1357 EXPORT_SYMBOL_GPL(is_error_pfn);
1359 static inline unsigned long bad_hva(void)
1364 int kvm_is_error_hva(unsigned long addr)
1366 return addr == bad_hva();
1368 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1370 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1374 for (i = 0; i < kvm->nmemslots; ++i) {
1375 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1377 if (gfn >= memslot->base_gfn
1378 && gfn < memslot->base_gfn + memslot->npages)
1383 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1385 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1387 gfn = unalias_gfn(kvm, gfn);
1388 return gfn_to_memslot_unaliased(kvm, gfn);
1391 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1395 gfn = unalias_gfn(kvm, gfn);
1396 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1397 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1399 if (gfn >= memslot->base_gfn
1400 && gfn < memslot->base_gfn + memslot->npages)
1405 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1407 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1409 struct kvm_memory_slot *slot;
1411 gfn = unalias_gfn(kvm, gfn);
1412 slot = gfn_to_memslot_unaliased(kvm, gfn);
1415 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1417 EXPORT_SYMBOL_GPL(gfn_to_hva);
1419 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1421 struct page *page[1];
1428 addr = gfn_to_hva(kvm, gfn);
1429 if (kvm_is_error_hva(addr)) {
1431 return page_to_pfn(bad_page);
1434 npages = get_user_pages_fast(addr, 1, 1, page);
1436 if (unlikely(npages != 1)) {
1437 struct vm_area_struct *vma;
1439 down_read(¤t->mm->mmap_sem);
1440 vma = find_vma(current->mm, addr);
1442 if (vma == NULL || addr < vma->vm_start ||
1443 !(vma->vm_flags & VM_PFNMAP)) {
1444 up_read(¤t->mm->mmap_sem);
1446 return page_to_pfn(bad_page);
1449 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1450 up_read(¤t->mm->mmap_sem);
1451 BUG_ON(!kvm_is_mmio_pfn(pfn));
1453 pfn = page_to_pfn(page[0]);
1458 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1460 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1464 pfn = gfn_to_pfn(kvm, gfn);
1465 if (!kvm_is_mmio_pfn(pfn))
1466 return pfn_to_page(pfn);
1468 WARN_ON(kvm_is_mmio_pfn(pfn));
1474 EXPORT_SYMBOL_GPL(gfn_to_page);
1476 void kvm_release_page_clean(struct page *page)
1478 kvm_release_pfn_clean(page_to_pfn(page));
1480 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1482 void kvm_release_pfn_clean(pfn_t pfn)
1484 if (!kvm_is_mmio_pfn(pfn))
1485 put_page(pfn_to_page(pfn));
1487 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1489 void kvm_release_page_dirty(struct page *page)
1491 kvm_release_pfn_dirty(page_to_pfn(page));
1493 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1495 void kvm_release_pfn_dirty(pfn_t pfn)
1497 kvm_set_pfn_dirty(pfn);
1498 kvm_release_pfn_clean(pfn);
1500 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1502 void kvm_set_page_dirty(struct page *page)
1504 kvm_set_pfn_dirty(page_to_pfn(page));
1506 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1508 void kvm_set_pfn_dirty(pfn_t pfn)
1510 if (!kvm_is_mmio_pfn(pfn)) {
1511 struct page *page = pfn_to_page(pfn);
1512 if (!PageReserved(page))
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1518 void kvm_set_pfn_accessed(pfn_t pfn)
1520 if (!kvm_is_mmio_pfn(pfn))
1521 mark_page_accessed(pfn_to_page(pfn));
1523 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1525 void kvm_get_pfn(pfn_t pfn)
1527 if (!kvm_is_mmio_pfn(pfn))
1528 get_page(pfn_to_page(pfn));
1530 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1532 static int next_segment(unsigned long len, int offset)
1534 if (len > PAGE_SIZE - offset)
1535 return PAGE_SIZE - offset;
1540 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1546 addr = gfn_to_hva(kvm, gfn);
1547 if (kvm_is_error_hva(addr))
1549 r = copy_from_user(data, (void __user *)addr + offset, len);
1554 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1556 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1558 gfn_t gfn = gpa >> PAGE_SHIFT;
1560 int offset = offset_in_page(gpa);
1563 while ((seg = next_segment(len, offset)) != 0) {
1564 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1574 EXPORT_SYMBOL_GPL(kvm_read_guest);
1576 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1581 gfn_t gfn = gpa >> PAGE_SHIFT;
1582 int offset = offset_in_page(gpa);
1584 addr = gfn_to_hva(kvm, gfn);
1585 if (kvm_is_error_hva(addr))
1587 pagefault_disable();
1588 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1594 EXPORT_SYMBOL(kvm_read_guest_atomic);
1596 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1597 int offset, int len)
1602 addr = gfn_to_hva(kvm, gfn);
1603 if (kvm_is_error_hva(addr))
1605 r = copy_to_user((void __user *)addr + offset, data, len);
1608 mark_page_dirty(kvm, gfn);
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1613 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1616 gfn_t gfn = gpa >> PAGE_SHIFT;
1618 int offset = offset_in_page(gpa);
1621 while ((seg = next_segment(len, offset)) != 0) {
1622 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1633 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1635 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1637 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1639 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1641 gfn_t gfn = gpa >> PAGE_SHIFT;
1643 int offset = offset_in_page(gpa);
1646 while ((seg = next_segment(len, offset)) != 0) {
1647 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1656 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1658 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1660 struct kvm_memory_slot *memslot;
1662 gfn = unalias_gfn(kvm, gfn);
1663 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1664 if (memslot && memslot->dirty_bitmap) {
1665 unsigned long rel_gfn = gfn - memslot->base_gfn;
1668 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1669 set_bit(rel_gfn, memslot->dirty_bitmap);
1674 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1676 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1681 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1683 if (kvm_arch_vcpu_runnable(vcpu)) {
1684 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1687 if (kvm_cpu_has_pending_timer(vcpu))
1689 if (signal_pending(current))
1697 finish_wait(&vcpu->wq, &wait);
1700 void kvm_resched(struct kvm_vcpu *vcpu)
1702 if (!need_resched())
1706 EXPORT_SYMBOL_GPL(kvm_resched);
1708 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1710 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1713 if (vmf->pgoff == 0)
1714 page = virt_to_page(vcpu->run);
1716 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1717 page = virt_to_page(vcpu->arch.pio_data);
1719 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1720 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1721 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1724 return VM_FAULT_SIGBUS;
1730 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1731 .fault = kvm_vcpu_fault,
1734 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1736 vma->vm_ops = &kvm_vcpu_vm_ops;
1740 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1742 struct kvm_vcpu *vcpu = filp->private_data;
1744 kvm_put_kvm(vcpu->kvm);
1748 static struct file_operations kvm_vcpu_fops = {
1749 .release = kvm_vcpu_release,
1750 .unlocked_ioctl = kvm_vcpu_ioctl,
1751 .compat_ioctl = kvm_vcpu_ioctl,
1752 .mmap = kvm_vcpu_mmap,
1756 * Allocates an inode for the vcpu.
1758 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1760 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1764 * Creates some virtual cpus. Good luck creating more than one.
1766 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1769 struct kvm_vcpu *vcpu, *v;
1771 vcpu = kvm_arch_vcpu_create(kvm, id);
1773 return PTR_ERR(vcpu);
1775 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1777 r = kvm_arch_vcpu_setup(vcpu);
1781 mutex_lock(&kvm->lock);
1782 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1787 kvm_for_each_vcpu(r, v, kvm)
1788 if (v->vcpu_id == id) {
1793 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1795 /* Now it's all set up, let userspace reach it */
1797 r = create_vcpu_fd(vcpu);
1803 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1805 atomic_inc(&kvm->online_vcpus);
1807 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1808 if (kvm->bsp_vcpu_id == id)
1809 kvm->bsp_vcpu = vcpu;
1811 mutex_unlock(&kvm->lock);
1815 mutex_unlock(&kvm->lock);
1816 kvm_arch_vcpu_destroy(vcpu);
1820 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1823 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1824 vcpu->sigset_active = 1;
1825 vcpu->sigset = *sigset;
1827 vcpu->sigset_active = 0;
1831 #ifdef __KVM_HAVE_MSIX
1832 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1833 struct kvm_assigned_msix_nr *entry_nr)
1836 struct kvm_assigned_dev_kernel *adev;
1838 mutex_lock(&kvm->lock);
1840 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1841 entry_nr->assigned_dev_id);
1847 if (adev->entries_nr == 0) {
1848 adev->entries_nr = entry_nr->entry_nr;
1849 if (adev->entries_nr == 0 ||
1850 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1855 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1858 if (!adev->host_msix_entries) {
1862 adev->guest_msix_entries = kzalloc(
1863 sizeof(struct kvm_guest_msix_entry) *
1864 entry_nr->entry_nr, GFP_KERNEL);
1865 if (!adev->guest_msix_entries) {
1866 kfree(adev->host_msix_entries);
1870 } else /* Not allowed set MSI-X number twice */
1873 mutex_unlock(&kvm->lock);
1877 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1878 struct kvm_assigned_msix_entry *entry)
1881 struct kvm_assigned_dev_kernel *adev;
1883 mutex_lock(&kvm->lock);
1885 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1886 entry->assigned_dev_id);
1890 goto msix_entry_out;
1893 for (i = 0; i < adev->entries_nr; i++)
1894 if (adev->guest_msix_entries[i].vector == 0 ||
1895 adev->guest_msix_entries[i].entry == entry->entry) {
1896 adev->guest_msix_entries[i].entry = entry->entry;
1897 adev->guest_msix_entries[i].vector = entry->gsi;
1898 adev->host_msix_entries[i].entry = entry->entry;
1901 if (i == adev->entries_nr) {
1903 goto msix_entry_out;
1907 mutex_unlock(&kvm->lock);
1913 static long kvm_vcpu_ioctl(struct file *filp,
1914 unsigned int ioctl, unsigned long arg)
1916 struct kvm_vcpu *vcpu = filp->private_data;
1917 void __user *argp = (void __user *)arg;
1919 struct kvm_fpu *fpu = NULL;
1920 struct kvm_sregs *kvm_sregs = NULL;
1922 if (vcpu->kvm->mm != current->mm)
1929 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1931 case KVM_GET_REGS: {
1932 struct kvm_regs *kvm_regs;
1935 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1938 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1942 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1949 case KVM_SET_REGS: {
1950 struct kvm_regs *kvm_regs;
1953 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1957 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1959 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1967 case KVM_GET_SREGS: {
1968 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1972 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1976 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1981 case KVM_SET_SREGS: {
1982 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1987 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1989 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1995 case KVM_GET_MP_STATE: {
1996 struct kvm_mp_state mp_state;
1998 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2002 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2007 case KVM_SET_MP_STATE: {
2008 struct kvm_mp_state mp_state;
2011 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2013 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2019 case KVM_TRANSLATE: {
2020 struct kvm_translation tr;
2023 if (copy_from_user(&tr, argp, sizeof tr))
2025 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2029 if (copy_to_user(argp, &tr, sizeof tr))
2034 case KVM_SET_GUEST_DEBUG: {
2035 struct kvm_guest_debug dbg;
2038 if (copy_from_user(&dbg, argp, sizeof dbg))
2040 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2046 case KVM_SET_SIGNAL_MASK: {
2047 struct kvm_signal_mask __user *sigmask_arg = argp;
2048 struct kvm_signal_mask kvm_sigmask;
2049 sigset_t sigset, *p;
2054 if (copy_from_user(&kvm_sigmask, argp,
2055 sizeof kvm_sigmask))
2058 if (kvm_sigmask.len != sizeof sigset)
2061 if (copy_from_user(&sigset, sigmask_arg->sigset,
2066 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2070 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2074 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2078 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2084 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2089 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2091 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2098 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2106 static long kvm_vm_ioctl(struct file *filp,
2107 unsigned int ioctl, unsigned long arg)
2109 struct kvm *kvm = filp->private_data;
2110 void __user *argp = (void __user *)arg;
2113 if (kvm->mm != current->mm)
2116 case KVM_CREATE_VCPU:
2117 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2121 case KVM_SET_USER_MEMORY_REGION: {
2122 struct kvm_userspace_memory_region kvm_userspace_mem;
2125 if (copy_from_user(&kvm_userspace_mem, argp,
2126 sizeof kvm_userspace_mem))
2129 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2134 case KVM_GET_DIRTY_LOG: {
2135 struct kvm_dirty_log log;
2138 if (copy_from_user(&log, argp, sizeof log))
2140 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2145 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2146 case KVM_REGISTER_COALESCED_MMIO: {
2147 struct kvm_coalesced_mmio_zone zone;
2149 if (copy_from_user(&zone, argp, sizeof zone))
2152 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2158 case KVM_UNREGISTER_COALESCED_MMIO: {
2159 struct kvm_coalesced_mmio_zone zone;
2161 if (copy_from_user(&zone, argp, sizeof zone))
2164 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2171 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2172 case KVM_ASSIGN_PCI_DEVICE: {
2173 struct kvm_assigned_pci_dev assigned_dev;
2176 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2178 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2183 case KVM_ASSIGN_IRQ: {
2187 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2188 case KVM_ASSIGN_DEV_IRQ: {
2189 struct kvm_assigned_irq assigned_irq;
2192 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2194 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2199 case KVM_DEASSIGN_DEV_IRQ: {
2200 struct kvm_assigned_irq assigned_irq;
2203 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2205 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2212 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2213 case KVM_DEASSIGN_PCI_DEVICE: {
2214 struct kvm_assigned_pci_dev assigned_dev;
2217 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2219 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2225 #ifdef KVM_CAP_IRQ_ROUTING
2226 case KVM_SET_GSI_ROUTING: {
2227 struct kvm_irq_routing routing;
2228 struct kvm_irq_routing __user *urouting;
2229 struct kvm_irq_routing_entry *entries;
2232 if (copy_from_user(&routing, argp, sizeof(routing)))
2235 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2240 entries = vmalloc(routing.nr * sizeof(*entries));
2245 if (copy_from_user(entries, urouting->entries,
2246 routing.nr * sizeof(*entries)))
2247 goto out_free_irq_routing;
2248 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2250 out_free_irq_routing:
2254 #endif /* KVM_CAP_IRQ_ROUTING */
2255 #ifdef __KVM_HAVE_MSIX
2256 case KVM_ASSIGN_SET_MSIX_NR: {
2257 struct kvm_assigned_msix_nr entry_nr;
2259 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2261 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2266 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2267 struct kvm_assigned_msix_entry entry;
2269 if (copy_from_user(&entry, argp, sizeof entry))
2271 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2278 struct kvm_irqfd data;
2281 if (copy_from_user(&data, argp, sizeof data))
2283 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2286 case KVM_IOEVENTFD: {
2287 struct kvm_ioeventfd data;
2290 if (copy_from_user(&data, argp, sizeof data))
2292 r = kvm_ioeventfd(kvm, &data);
2295 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2296 case KVM_SET_BOOT_CPU_ID:
2298 mutex_lock(&kvm->lock);
2299 if (atomic_read(&kvm->online_vcpus) != 0)
2302 kvm->bsp_vcpu_id = arg;
2303 mutex_unlock(&kvm->lock);
2307 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2313 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2315 struct page *page[1];
2318 gfn_t gfn = vmf->pgoff;
2319 struct kvm *kvm = vma->vm_file->private_data;
2321 addr = gfn_to_hva(kvm, gfn);
2322 if (kvm_is_error_hva(addr))
2323 return VM_FAULT_SIGBUS;
2325 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2327 if (unlikely(npages != 1))
2328 return VM_FAULT_SIGBUS;
2330 vmf->page = page[0];
2334 static const struct vm_operations_struct kvm_vm_vm_ops = {
2335 .fault = kvm_vm_fault,
2338 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2340 vma->vm_ops = &kvm_vm_vm_ops;
2344 static struct file_operations kvm_vm_fops = {
2345 .release = kvm_vm_release,
2346 .unlocked_ioctl = kvm_vm_ioctl,
2347 .compat_ioctl = kvm_vm_ioctl,
2348 .mmap = kvm_vm_mmap,
2351 static int kvm_dev_ioctl_create_vm(void)
2356 kvm = kvm_create_vm();
2358 return PTR_ERR(kvm);
2359 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2366 static long kvm_dev_ioctl_check_extension_generic(long arg)
2369 case KVM_CAP_USER_MEMORY:
2370 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2371 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2372 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID:
2376 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2377 case KVM_CAP_IRQ_ROUTING:
2378 return KVM_MAX_IRQ_ROUTES;
2383 return kvm_dev_ioctl_check_extension(arg);
2386 static long kvm_dev_ioctl(struct file *filp,
2387 unsigned int ioctl, unsigned long arg)
2392 case KVM_GET_API_VERSION:
2396 r = KVM_API_VERSION;
2402 r = kvm_dev_ioctl_create_vm();
2404 case KVM_CHECK_EXTENSION:
2405 r = kvm_dev_ioctl_check_extension_generic(arg);
2407 case KVM_GET_VCPU_MMAP_SIZE:
2411 r = PAGE_SIZE; /* struct kvm_run */
2413 r += PAGE_SIZE; /* pio data page */
2415 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2416 r += PAGE_SIZE; /* coalesced mmio ring page */
2419 case KVM_TRACE_ENABLE:
2420 case KVM_TRACE_PAUSE:
2421 case KVM_TRACE_DISABLE:
2425 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2431 static struct file_operations kvm_chardev_ops = {
2432 .unlocked_ioctl = kvm_dev_ioctl,
2433 .compat_ioctl = kvm_dev_ioctl,
2436 static struct miscdevice kvm_dev = {
2442 static void hardware_enable(void *junk)
2444 int cpu = raw_smp_processor_id();
2446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2448 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2449 kvm_arch_hardware_enable(NULL);
2452 static void hardware_disable(void *junk)
2454 int cpu = raw_smp_processor_id();
2456 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2458 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2459 kvm_arch_hardware_disable(NULL);
2462 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2467 val &= ~CPU_TASKS_FROZEN;
2470 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2472 hardware_disable(NULL);
2474 case CPU_UP_CANCELED:
2475 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2477 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2480 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2482 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2489 asmlinkage void kvm_handle_fault_on_reboot(void)
2492 /* spin while reset goes on */
2495 /* Fault while not rebooting. We want the trace. */
2498 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2500 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2504 * Some (well, at least mine) BIOSes hang on reboot if
2507 * And Intel TXT required VMX off for all cpu when system shutdown.
2509 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2510 kvm_rebooting = true;
2511 on_each_cpu(hardware_disable, NULL, 1);
2515 static struct notifier_block kvm_reboot_notifier = {
2516 .notifier_call = kvm_reboot,
2520 void kvm_io_bus_init(struct kvm_io_bus *bus)
2522 memset(bus, 0, sizeof(*bus));
2525 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2529 for (i = 0; i < bus->dev_count; i++) {
2530 struct kvm_io_device *pos = bus->devs[i];
2532 kvm_iodevice_destructor(pos);
2536 /* kvm_io_bus_write - called under kvm->slots_lock */
2537 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2538 int len, const void *val)
2541 for (i = 0; i < bus->dev_count; i++)
2542 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2547 /* kvm_io_bus_read - called under kvm->slots_lock */
2548 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2551 for (i = 0; i < bus->dev_count; i++)
2552 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2557 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2558 struct kvm_io_device *dev)
2562 down_write(&kvm->slots_lock);
2563 ret = __kvm_io_bus_register_dev(bus, dev);
2564 up_write(&kvm->slots_lock);
2569 /* An unlocked version. Caller must have write lock on slots_lock. */
2570 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2571 struct kvm_io_device *dev)
2573 if (bus->dev_count > NR_IOBUS_DEVS-1)
2576 bus->devs[bus->dev_count++] = dev;
2581 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2582 struct kvm_io_bus *bus,
2583 struct kvm_io_device *dev)
2585 down_write(&kvm->slots_lock);
2586 __kvm_io_bus_unregister_dev(bus, dev);
2587 up_write(&kvm->slots_lock);
2590 /* An unlocked version. Caller must have write lock on slots_lock. */
2591 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2592 struct kvm_io_device *dev)
2596 for (i = 0; i < bus->dev_count; i++)
2597 if (bus->devs[i] == dev) {
2598 bus->devs[i] = bus->devs[--bus->dev_count];
2603 static struct notifier_block kvm_cpu_notifier = {
2604 .notifier_call = kvm_cpu_hotplug,
2605 .priority = 20, /* must be > scheduler priority */
2608 static int vm_stat_get(void *_offset, u64 *val)
2610 unsigned offset = (long)_offset;
2614 spin_lock(&kvm_lock);
2615 list_for_each_entry(kvm, &vm_list, vm_list)
2616 *val += *(u32 *)((void *)kvm + offset);
2617 spin_unlock(&kvm_lock);
2621 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2623 static int vcpu_stat_get(void *_offset, u64 *val)
2625 unsigned offset = (long)_offset;
2627 struct kvm_vcpu *vcpu;
2631 spin_lock(&kvm_lock);
2632 list_for_each_entry(kvm, &vm_list, vm_list)
2633 kvm_for_each_vcpu(i, vcpu, kvm)
2634 *val += *(u32 *)((void *)vcpu + offset);
2636 spin_unlock(&kvm_lock);
2640 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2642 static const struct file_operations *stat_fops[] = {
2643 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2644 [KVM_STAT_VM] = &vm_stat_fops,
2647 static void kvm_init_debug(void)
2649 struct kvm_stats_debugfs_item *p;
2651 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2652 for (p = debugfs_entries; p->name; ++p)
2653 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2654 (void *)(long)p->offset,
2655 stat_fops[p->kind]);
2658 static void kvm_exit_debug(void)
2660 struct kvm_stats_debugfs_item *p;
2662 for (p = debugfs_entries; p->name; ++p)
2663 debugfs_remove(p->dentry);
2664 debugfs_remove(kvm_debugfs_dir);
2667 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2669 hardware_disable(NULL);
2673 static int kvm_resume(struct sys_device *dev)
2675 hardware_enable(NULL);
2679 static struct sysdev_class kvm_sysdev_class = {
2681 .suspend = kvm_suspend,
2682 .resume = kvm_resume,
2685 static struct sys_device kvm_sysdev = {
2687 .cls = &kvm_sysdev_class,
2690 struct page *bad_page;
2694 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2696 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2699 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2701 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2703 kvm_arch_vcpu_load(vcpu, cpu);
2706 static void kvm_sched_out(struct preempt_notifier *pn,
2707 struct task_struct *next)
2709 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2711 kvm_arch_vcpu_put(vcpu);
2714 int kvm_init(void *opaque, unsigned int vcpu_size,
2715 struct module *module)
2720 r = kvm_arch_init(opaque);
2724 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2726 if (bad_page == NULL) {
2731 bad_pfn = page_to_pfn(bad_page);
2733 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2738 r = kvm_arch_hardware_setup();
2742 for_each_online_cpu(cpu) {
2743 smp_call_function_single(cpu,
2744 kvm_arch_check_processor_compat,
2750 on_each_cpu(hardware_enable, NULL, 1);
2751 r = register_cpu_notifier(&kvm_cpu_notifier);
2754 register_reboot_notifier(&kvm_reboot_notifier);
2756 r = sysdev_class_register(&kvm_sysdev_class);
2760 r = sysdev_register(&kvm_sysdev);
2764 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2765 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2766 __alignof__(struct kvm_vcpu),
2768 if (!kvm_vcpu_cache) {
2773 kvm_chardev_ops.owner = module;
2774 kvm_vm_fops.owner = module;
2775 kvm_vcpu_fops.owner = module;
2777 r = misc_register(&kvm_dev);
2779 printk(KERN_ERR "kvm: misc device register failed\n");
2783 kvm_preempt_ops.sched_in = kvm_sched_in;
2784 kvm_preempt_ops.sched_out = kvm_sched_out;
2791 kmem_cache_destroy(kvm_vcpu_cache);
2793 sysdev_unregister(&kvm_sysdev);
2795 sysdev_class_unregister(&kvm_sysdev_class);
2797 unregister_reboot_notifier(&kvm_reboot_notifier);
2798 unregister_cpu_notifier(&kvm_cpu_notifier);
2800 on_each_cpu(hardware_disable, NULL, 1);
2802 kvm_arch_hardware_unsetup();
2804 free_cpumask_var(cpus_hardware_enabled);
2806 __free_page(bad_page);
2812 EXPORT_SYMBOL_GPL(kvm_init);
2816 tracepoint_synchronize_unregister();
2818 misc_deregister(&kvm_dev);
2819 kmem_cache_destroy(kvm_vcpu_cache);
2820 sysdev_unregister(&kvm_sysdev);
2821 sysdev_class_unregister(&kvm_sysdev_class);
2822 unregister_reboot_notifier(&kvm_reboot_notifier);
2823 unregister_cpu_notifier(&kvm_cpu_notifier);
2824 on_each_cpu(hardware_disable, NULL, 1);
2825 kvm_arch_hardware_unsetup();
2827 free_cpumask_var(cpus_hardware_enabled);
2828 __free_page(bad_page);
2830 EXPORT_SYMBOL_GPL(kvm_exit);