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>
46 #include <asm/processor.h>
48 #include <asm/uaccess.h>
49 #include <asm/pgtable.h>
51 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
52 #include "coalesced_mmio.h"
55 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
56 #include <linux/pci.h>
57 #include <linux/interrupt.h>
61 MODULE_AUTHOR("Qumranet");
62 MODULE_LICENSE("GPL");
64 DEFINE_SPINLOCK(kvm_lock);
67 static cpumask_var_t cpus_hardware_enabled;
69 struct kmem_cache *kvm_vcpu_cache;
70 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
72 static __read_mostly struct preempt_ops kvm_preempt_ops;
74 struct dentry *kvm_debugfs_dir;
76 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
79 static bool kvm_rebooting;
81 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
82 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
85 struct list_head *ptr;
86 struct kvm_assigned_dev_kernel *match;
88 list_for_each(ptr, head) {
89 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
90 if (match->assigned_dev_id == assigned_dev_id)
96 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
97 *assigned_dev, int irq)
100 struct msix_entry *host_msix_entries;
102 host_msix_entries = assigned_dev->host_msix_entries;
105 for (i = 0; i < assigned_dev->entries_nr; i++)
106 if (irq == host_msix_entries[i].vector) {
111 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
118 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
120 struct kvm_assigned_dev_kernel *assigned_dev;
124 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
126 kvm = assigned_dev->kvm;
128 /* This is taken to safely inject irq inside the guest. When
129 * the interrupt injection (or the ioapic code) uses a
130 * finer-grained lock, update this
132 mutex_lock(&kvm->lock);
133 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
134 struct kvm_guest_msix_entry *guest_entries =
135 assigned_dev->guest_msix_entries;
136 for (i = 0; i < assigned_dev->entries_nr; i++) {
137 if (!(guest_entries[i].flags &
138 KVM_ASSIGNED_MSIX_PENDING))
140 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
141 kvm_set_irq(assigned_dev->kvm,
142 assigned_dev->irq_source_id,
143 guest_entries[i].vector, 1);
144 irq = assigned_dev->host_msix_entries[i].vector;
147 assigned_dev->host_irq_disabled = false;
150 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
151 assigned_dev->guest_irq, 1);
152 if (assigned_dev->irq_requested_type &
153 KVM_DEV_IRQ_GUEST_MSI) {
154 enable_irq(assigned_dev->host_irq);
155 assigned_dev->host_irq_disabled = false;
159 mutex_unlock(&assigned_dev->kvm->lock);
162 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
164 struct kvm_assigned_dev_kernel *assigned_dev =
165 (struct kvm_assigned_dev_kernel *) dev_id;
167 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
168 int index = find_index_from_host_irq(assigned_dev, irq);
171 assigned_dev->guest_msix_entries[index].flags |=
172 KVM_ASSIGNED_MSIX_PENDING;
175 schedule_work(&assigned_dev->interrupt_work);
177 disable_irq_nosync(irq);
178 assigned_dev->host_irq_disabled = true;
183 /* Ack the irq line for an assigned device */
184 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
186 struct kvm_assigned_dev_kernel *dev;
191 dev = container_of(kian, struct kvm_assigned_dev_kernel,
194 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
196 /* The guest irq may be shared so this ack may be
197 * from another device.
199 if (dev->host_irq_disabled) {
200 enable_irq(dev->host_irq);
201 dev->host_irq_disabled = false;
205 static void deassign_guest_irq(struct kvm *kvm,
206 struct kvm_assigned_dev_kernel *assigned_dev)
208 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
209 assigned_dev->ack_notifier.gsi = -1;
211 if (assigned_dev->irq_source_id != -1)
212 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
213 assigned_dev->irq_source_id = -1;
214 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
217 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
218 static void deassign_host_irq(struct kvm *kvm,
219 struct kvm_assigned_dev_kernel *assigned_dev)
222 * In kvm_free_device_irq, cancel_work_sync return true if:
223 * 1. work is scheduled, and then cancelled.
224 * 2. work callback is executed.
226 * The first one ensured that the irq is disabled and no more events
227 * would happen. But for the second one, the irq may be enabled (e.g.
228 * for MSI). So we disable irq here to prevent further events.
230 * Notice this maybe result in nested disable if the interrupt type is
231 * INTx, but it's OK for we are going to free it.
233 * If this function is a part of VM destroy, please ensure that till
234 * now, the kvm state is still legal for probably we also have to wait
235 * interrupt_work done.
237 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
239 for (i = 0; i < assigned_dev->entries_nr; i++)
240 disable_irq_nosync(assigned_dev->
241 host_msix_entries[i].vector);
243 cancel_work_sync(&assigned_dev->interrupt_work);
245 for (i = 0; i < assigned_dev->entries_nr; i++)
246 free_irq(assigned_dev->host_msix_entries[i].vector,
247 (void *)assigned_dev);
249 assigned_dev->entries_nr = 0;
250 kfree(assigned_dev->host_msix_entries);
251 kfree(assigned_dev->guest_msix_entries);
252 pci_disable_msix(assigned_dev->dev);
254 /* Deal with MSI and INTx */
255 disable_irq_nosync(assigned_dev->host_irq);
256 cancel_work_sync(&assigned_dev->interrupt_work);
258 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
260 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
261 pci_disable_msi(assigned_dev->dev);
264 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
267 static int kvm_deassign_irq(struct kvm *kvm,
268 struct kvm_assigned_dev_kernel *assigned_dev,
269 unsigned long irq_requested_type)
271 unsigned long guest_irq_type, host_irq_type;
273 if (!irqchip_in_kernel(kvm))
275 /* no irq assignment to deassign */
276 if (!assigned_dev->irq_requested_type)
279 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
280 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
283 deassign_host_irq(kvm, assigned_dev);
285 deassign_guest_irq(kvm, assigned_dev);
290 static void kvm_free_assigned_irq(struct kvm *kvm,
291 struct kvm_assigned_dev_kernel *assigned_dev)
293 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
296 static void kvm_free_assigned_device(struct kvm *kvm,
297 struct kvm_assigned_dev_kernel
300 kvm_free_assigned_irq(kvm, assigned_dev);
302 pci_reset_function(assigned_dev->dev);
304 pci_release_regions(assigned_dev->dev);
305 pci_disable_device(assigned_dev->dev);
306 pci_dev_put(assigned_dev->dev);
308 list_del(&assigned_dev->list);
312 void kvm_free_all_assigned_devices(struct kvm *kvm)
314 struct list_head *ptr, *ptr2;
315 struct kvm_assigned_dev_kernel *assigned_dev;
317 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
318 assigned_dev = list_entry(ptr,
319 struct kvm_assigned_dev_kernel,
322 kvm_free_assigned_device(kvm, assigned_dev);
326 static int assigned_device_enable_host_intx(struct kvm *kvm,
327 struct kvm_assigned_dev_kernel *dev)
329 dev->host_irq = dev->dev->irq;
330 /* Even though this is PCI, we don't want to use shared
331 * interrupts. Sharing host devices with guest-assigned devices
332 * on the same interrupt line is not a happy situation: there
333 * are going to be long delays in accepting, acking, etc.
335 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
336 0, "kvm_assigned_intx_device", (void *)dev))
341 #ifdef __KVM_HAVE_MSI
342 static int assigned_device_enable_host_msi(struct kvm *kvm,
343 struct kvm_assigned_dev_kernel *dev)
347 if (!dev->dev->msi_enabled) {
348 r = pci_enable_msi(dev->dev);
353 dev->host_irq = dev->dev->irq;
354 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
355 "kvm_assigned_msi_device", (void *)dev)) {
356 pci_disable_msi(dev->dev);
364 #ifdef __KVM_HAVE_MSIX
365 static int assigned_device_enable_host_msix(struct kvm *kvm,
366 struct kvm_assigned_dev_kernel *dev)
370 /* host_msix_entries and guest_msix_entries should have been
372 if (dev->entries_nr == 0)
375 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
379 for (i = 0; i < dev->entries_nr; i++) {
380 r = request_irq(dev->host_msix_entries[i].vector,
381 kvm_assigned_dev_intr, 0,
382 "kvm_assigned_msix_device",
384 /* FIXME: free requested_irq's on failure */
394 static int assigned_device_enable_guest_intx(struct kvm *kvm,
395 struct kvm_assigned_dev_kernel *dev,
396 struct kvm_assigned_irq *irq)
398 dev->guest_irq = irq->guest_irq;
399 dev->ack_notifier.gsi = irq->guest_irq;
403 #ifdef __KVM_HAVE_MSI
404 static int assigned_device_enable_guest_msi(struct kvm *kvm,
405 struct kvm_assigned_dev_kernel *dev,
406 struct kvm_assigned_irq *irq)
408 dev->guest_irq = irq->guest_irq;
409 dev->ack_notifier.gsi = -1;
413 #ifdef __KVM_HAVE_MSIX
414 static int assigned_device_enable_guest_msix(struct kvm *kvm,
415 struct kvm_assigned_dev_kernel *dev,
416 struct kvm_assigned_irq *irq)
418 dev->guest_irq = irq->guest_irq;
419 dev->ack_notifier.gsi = -1;
424 static int assign_host_irq(struct kvm *kvm,
425 struct kvm_assigned_dev_kernel *dev,
430 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
433 switch (host_irq_type) {
434 case KVM_DEV_IRQ_HOST_INTX:
435 r = assigned_device_enable_host_intx(kvm, dev);
437 #ifdef __KVM_HAVE_MSI
438 case KVM_DEV_IRQ_HOST_MSI:
439 r = assigned_device_enable_host_msi(kvm, dev);
442 #ifdef __KVM_HAVE_MSIX
443 case KVM_DEV_IRQ_HOST_MSIX:
444 r = assigned_device_enable_host_msix(kvm, dev);
452 dev->irq_requested_type |= host_irq_type;
457 static int assign_guest_irq(struct kvm *kvm,
458 struct kvm_assigned_dev_kernel *dev,
459 struct kvm_assigned_irq *irq,
460 unsigned long guest_irq_type)
465 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
468 id = kvm_request_irq_source_id(kvm);
472 dev->irq_source_id = id;
474 switch (guest_irq_type) {
475 case KVM_DEV_IRQ_GUEST_INTX:
476 r = assigned_device_enable_guest_intx(kvm, dev, irq);
478 #ifdef __KVM_HAVE_MSI
479 case KVM_DEV_IRQ_GUEST_MSI:
480 r = assigned_device_enable_guest_msi(kvm, dev, irq);
483 #ifdef __KVM_HAVE_MSIX
484 case KVM_DEV_IRQ_GUEST_MSIX:
485 r = assigned_device_enable_guest_msix(kvm, dev, irq);
493 dev->irq_requested_type |= guest_irq_type;
494 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
496 kvm_free_irq_source_id(kvm, dev->irq_source_id);
501 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
502 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
503 struct kvm_assigned_irq *assigned_irq)
506 struct kvm_assigned_dev_kernel *match;
507 unsigned long host_irq_type, guest_irq_type;
509 if (!capable(CAP_SYS_RAWIO))
512 if (!irqchip_in_kernel(kvm))
515 mutex_lock(&kvm->lock);
517 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
518 assigned_irq->assigned_dev_id);
522 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
523 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
526 /* can only assign one type at a time */
527 if (hweight_long(host_irq_type) > 1)
529 if (hweight_long(guest_irq_type) > 1)
531 if (host_irq_type == 0 && guest_irq_type == 0)
536 r = assign_host_irq(kvm, match, host_irq_type);
541 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
543 mutex_unlock(&kvm->lock);
547 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
548 struct kvm_assigned_irq
552 struct kvm_assigned_dev_kernel *match;
554 mutex_lock(&kvm->lock);
556 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
557 assigned_irq->assigned_dev_id);
561 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
563 mutex_unlock(&kvm->lock);
567 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
568 struct kvm_assigned_pci_dev *assigned_dev)
571 struct kvm_assigned_dev_kernel *match;
574 down_read(&kvm->slots_lock);
575 mutex_lock(&kvm->lock);
577 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
578 assigned_dev->assigned_dev_id);
580 /* device already assigned */
585 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
587 printk(KERN_INFO "%s: Couldn't allocate memory\n",
592 dev = pci_get_bus_and_slot(assigned_dev->busnr,
593 assigned_dev->devfn);
595 printk(KERN_INFO "%s: host device not found\n", __func__);
599 if (pci_enable_device(dev)) {
600 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
604 r = pci_request_regions(dev, "kvm_assigned_device");
606 printk(KERN_INFO "%s: Could not get access to device regions\n",
611 pci_reset_function(dev);
613 match->assigned_dev_id = assigned_dev->assigned_dev_id;
614 match->host_busnr = assigned_dev->busnr;
615 match->host_devfn = assigned_dev->devfn;
616 match->flags = assigned_dev->flags;
618 match->irq_source_id = -1;
620 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
621 INIT_WORK(&match->interrupt_work,
622 kvm_assigned_dev_interrupt_work_handler);
624 list_add(&match->list, &kvm->arch.assigned_dev_head);
626 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
627 if (!kvm->arch.iommu_domain) {
628 r = kvm_iommu_map_guest(kvm);
632 r = kvm_assign_device(kvm, match);
638 mutex_unlock(&kvm->lock);
639 up_read(&kvm->slots_lock);
642 list_del(&match->list);
643 pci_release_regions(dev);
645 pci_disable_device(dev);
650 mutex_unlock(&kvm->lock);
651 up_read(&kvm->slots_lock);
656 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
657 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
658 struct kvm_assigned_pci_dev *assigned_dev)
661 struct kvm_assigned_dev_kernel *match;
663 mutex_lock(&kvm->lock);
665 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
666 assigned_dev->assigned_dev_id);
668 printk(KERN_INFO "%s: device hasn't been assigned before, "
669 "so cannot be deassigned\n", __func__);
674 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
675 kvm_deassign_device(kvm, match);
677 kvm_free_assigned_device(kvm, match);
680 mutex_unlock(&kvm->lock);
685 static inline int valid_vcpu(int n)
687 return likely(n >= 0 && n < KVM_MAX_VCPUS);
690 inline int kvm_is_mmio_pfn(pfn_t pfn)
692 if (pfn_valid(pfn)) {
693 struct page *page = compound_head(pfn_to_page(pfn));
694 return PageReserved(page);
701 * Switches to specified vcpu, until a matching vcpu_put()
703 void vcpu_load(struct kvm_vcpu *vcpu)
707 mutex_lock(&vcpu->mutex);
709 preempt_notifier_register(&vcpu->preempt_notifier);
710 kvm_arch_vcpu_load(vcpu, cpu);
714 void vcpu_put(struct kvm_vcpu *vcpu)
717 kvm_arch_vcpu_put(vcpu);
718 preempt_notifier_unregister(&vcpu->preempt_notifier);
720 mutex_unlock(&vcpu->mutex);
723 static void ack_flush(void *_completed)
727 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
732 struct kvm_vcpu *vcpu;
734 if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
738 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
739 vcpu = kvm->vcpus[i];
742 if (test_and_set_bit(req, &vcpu->requests))
745 if (cpus != NULL && cpu != -1 && cpu != me)
746 cpumask_set_cpu(cpu, cpus);
748 if (unlikely(cpus == NULL))
749 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
750 else if (!cpumask_empty(cpus))
751 smp_call_function_many(cpus, ack_flush, NULL, 1);
755 free_cpumask_var(cpus);
759 void kvm_flush_remote_tlbs(struct kvm *kvm)
761 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
762 ++kvm->stat.remote_tlb_flush;
765 void kvm_reload_remote_mmus(struct kvm *kvm)
767 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
770 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
775 mutex_init(&vcpu->mutex);
779 init_waitqueue_head(&vcpu->wq);
781 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
786 vcpu->run = page_address(page);
788 r = kvm_arch_vcpu_init(vcpu);
794 free_page((unsigned long)vcpu->run);
798 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
800 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
802 kvm_arch_vcpu_uninit(vcpu);
803 free_page((unsigned long)vcpu->run);
805 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
807 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
808 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
810 return container_of(mn, struct kvm, mmu_notifier);
813 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
814 struct mm_struct *mm,
815 unsigned long address)
817 struct kvm *kvm = mmu_notifier_to_kvm(mn);
821 * When ->invalidate_page runs, the linux pte has been zapped
822 * already but the page is still allocated until
823 * ->invalidate_page returns. So if we increase the sequence
824 * here the kvm page fault will notice if the spte can't be
825 * established because the page is going to be freed. If
826 * instead the kvm page fault establishes the spte before
827 * ->invalidate_page runs, kvm_unmap_hva will release it
830 * The sequence increase only need to be seen at spin_unlock
831 * time, and not at spin_lock time.
833 * Increasing the sequence after the spin_unlock would be
834 * unsafe because the kvm page fault could then establish the
835 * pte after kvm_unmap_hva returned, without noticing the page
836 * is going to be freed.
838 spin_lock(&kvm->mmu_lock);
839 kvm->mmu_notifier_seq++;
840 need_tlb_flush = kvm_unmap_hva(kvm, address);
841 spin_unlock(&kvm->mmu_lock);
843 /* we've to flush the tlb before the pages can be freed */
845 kvm_flush_remote_tlbs(kvm);
849 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
850 struct mm_struct *mm,
854 struct kvm *kvm = mmu_notifier_to_kvm(mn);
855 int need_tlb_flush = 0;
857 spin_lock(&kvm->mmu_lock);
859 * The count increase must become visible at unlock time as no
860 * spte can be established without taking the mmu_lock and
861 * count is also read inside the mmu_lock critical section.
863 kvm->mmu_notifier_count++;
864 for (; start < end; start += PAGE_SIZE)
865 need_tlb_flush |= kvm_unmap_hva(kvm, start);
866 spin_unlock(&kvm->mmu_lock);
868 /* we've to flush the tlb before the pages can be freed */
870 kvm_flush_remote_tlbs(kvm);
873 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
874 struct mm_struct *mm,
878 struct kvm *kvm = mmu_notifier_to_kvm(mn);
880 spin_lock(&kvm->mmu_lock);
882 * This sequence increase will notify the kvm page fault that
883 * the page that is going to be mapped in the spte could have
886 kvm->mmu_notifier_seq++;
888 * The above sequence increase must be visible before the
889 * below count decrease but both values are read by the kvm
890 * page fault under mmu_lock spinlock so we don't need to add
891 * a smb_wmb() here in between the two.
893 kvm->mmu_notifier_count--;
894 spin_unlock(&kvm->mmu_lock);
896 BUG_ON(kvm->mmu_notifier_count < 0);
899 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
900 struct mm_struct *mm,
901 unsigned long address)
903 struct kvm *kvm = mmu_notifier_to_kvm(mn);
906 spin_lock(&kvm->mmu_lock);
907 young = kvm_age_hva(kvm, address);
908 spin_unlock(&kvm->mmu_lock);
911 kvm_flush_remote_tlbs(kvm);
916 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
917 struct mm_struct *mm)
919 struct kvm *kvm = mmu_notifier_to_kvm(mn);
920 kvm_arch_flush_shadow(kvm);
923 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
924 .invalidate_page = kvm_mmu_notifier_invalidate_page,
925 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
926 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
927 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
928 .release = kvm_mmu_notifier_release,
930 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
932 static struct kvm *kvm_create_vm(void)
934 struct kvm *kvm = kvm_arch_create_vm();
935 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
941 #ifdef CONFIG_HAVE_KVM_IRQCHIP
942 INIT_LIST_HEAD(&kvm->irq_routing);
943 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
946 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
947 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
950 return ERR_PTR(-ENOMEM);
952 kvm->coalesced_mmio_ring =
953 (struct kvm_coalesced_mmio_ring *)page_address(page);
956 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
959 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
960 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
962 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
971 kvm->mm = current->mm;
972 atomic_inc(&kvm->mm->mm_count);
973 spin_lock_init(&kvm->mmu_lock);
974 kvm_io_bus_init(&kvm->pio_bus);
975 mutex_init(&kvm->lock);
976 kvm_io_bus_init(&kvm->mmio_bus);
977 init_rwsem(&kvm->slots_lock);
978 atomic_set(&kvm->users_count, 1);
979 spin_lock(&kvm_lock);
980 list_add(&kvm->vm_list, &vm_list);
981 spin_unlock(&kvm_lock);
982 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
983 kvm_coalesced_mmio_init(kvm);
990 * Free any memory in @free but not in @dont.
992 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
993 struct kvm_memory_slot *dont)
995 if (!dont || free->rmap != dont->rmap)
998 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
999 vfree(free->dirty_bitmap);
1001 if (!dont || free->lpage_info != dont->lpage_info)
1002 vfree(free->lpage_info);
1005 free->dirty_bitmap = NULL;
1007 free->lpage_info = NULL;
1010 void kvm_free_physmem(struct kvm *kvm)
1014 for (i = 0; i < kvm->nmemslots; ++i)
1015 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1018 static void kvm_destroy_vm(struct kvm *kvm)
1020 struct mm_struct *mm = kvm->mm;
1022 kvm_arch_sync_events(kvm);
1023 spin_lock(&kvm_lock);
1024 list_del(&kvm->vm_list);
1025 spin_unlock(&kvm_lock);
1026 kvm_free_irq_routing(kvm);
1027 kvm_io_bus_destroy(&kvm->pio_bus);
1028 kvm_io_bus_destroy(&kvm->mmio_bus);
1029 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1030 if (kvm->coalesced_mmio_ring != NULL)
1031 free_page((unsigned long)kvm->coalesced_mmio_ring);
1033 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1034 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1036 kvm_arch_destroy_vm(kvm);
1040 void kvm_get_kvm(struct kvm *kvm)
1042 atomic_inc(&kvm->users_count);
1044 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1046 void kvm_put_kvm(struct kvm *kvm)
1048 if (atomic_dec_and_test(&kvm->users_count))
1049 kvm_destroy_vm(kvm);
1051 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1054 static int kvm_vm_release(struct inode *inode, struct file *filp)
1056 struct kvm *kvm = filp->private_data;
1063 * Allocate some memory and give it an address in the guest physical address
1066 * Discontiguous memory is allowed, mostly for framebuffers.
1068 * Must be called holding mmap_sem for write.
1070 int __kvm_set_memory_region(struct kvm *kvm,
1071 struct kvm_userspace_memory_region *mem,
1076 unsigned long npages;
1079 struct kvm_memory_slot *memslot;
1080 struct kvm_memory_slot old, new;
1083 /* General sanity checks */
1084 if (mem->memory_size & (PAGE_SIZE - 1))
1086 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1088 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1090 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1092 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1095 memslot = &kvm->memslots[mem->slot];
1096 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1097 npages = mem->memory_size >> PAGE_SHIFT;
1100 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1102 new = old = *memslot;
1104 new.base_gfn = base_gfn;
1105 new.npages = npages;
1106 new.flags = mem->flags;
1108 /* Disallow changing a memory slot's size. */
1110 if (npages && old.npages && npages != old.npages)
1113 /* Check for overlaps */
1115 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1116 struct kvm_memory_slot *s = &kvm->memslots[i];
1118 if (s == memslot || !s->npages)
1120 if (!((base_gfn + npages <= s->base_gfn) ||
1121 (base_gfn >= s->base_gfn + s->npages)))
1125 /* Free page dirty bitmap if unneeded */
1126 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1127 new.dirty_bitmap = NULL;
1131 /* Allocate if a slot is being created */
1133 if (npages && !new.rmap) {
1134 new.rmap = vmalloc(npages * sizeof(struct page *));
1139 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1141 new.user_alloc = user_alloc;
1143 * hva_to_rmmap() serialzies with the mmu_lock and to be
1144 * safe it has to ignore memslots with !user_alloc &&
1148 new.userspace_addr = mem->userspace_addr;
1150 new.userspace_addr = 0;
1152 if (npages && !new.lpage_info) {
1153 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1154 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1156 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1158 if (!new.lpage_info)
1161 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1163 if (base_gfn % KVM_PAGES_PER_HPAGE)
1164 new.lpage_info[0].write_count = 1;
1165 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1166 new.lpage_info[largepages-1].write_count = 1;
1169 /* Allocate page dirty bitmap if needed */
1170 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1171 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1173 new.dirty_bitmap = vmalloc(dirty_bytes);
1174 if (!new.dirty_bitmap)
1176 memset(new.dirty_bitmap, 0, dirty_bytes);
1178 #endif /* not defined CONFIG_S390 */
1181 kvm_arch_flush_shadow(kvm);
1183 spin_lock(&kvm->mmu_lock);
1184 if (mem->slot >= kvm->nmemslots)
1185 kvm->nmemslots = mem->slot + 1;
1188 spin_unlock(&kvm->mmu_lock);
1190 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1192 spin_lock(&kvm->mmu_lock);
1194 spin_unlock(&kvm->mmu_lock);
1198 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1199 /* Slot deletion case: we have to update the current slot */
1203 /* map the pages in iommu page table */
1204 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1211 kvm_free_physmem_slot(&new, &old);
1216 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1218 int kvm_set_memory_region(struct kvm *kvm,
1219 struct kvm_userspace_memory_region *mem,
1224 down_write(&kvm->slots_lock);
1225 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1226 up_write(&kvm->slots_lock);
1229 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1231 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1233 kvm_userspace_memory_region *mem,
1236 if (mem->slot >= KVM_MEMORY_SLOTS)
1238 return kvm_set_memory_region(kvm, mem, user_alloc);
1241 int kvm_get_dirty_log(struct kvm *kvm,
1242 struct kvm_dirty_log *log, int *is_dirty)
1244 struct kvm_memory_slot *memslot;
1247 unsigned long any = 0;
1250 if (log->slot >= KVM_MEMORY_SLOTS)
1253 memslot = &kvm->memslots[log->slot];
1255 if (!memslot->dirty_bitmap)
1258 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1260 for (i = 0; !any && i < n/sizeof(long); ++i)
1261 any = memslot->dirty_bitmap[i];
1264 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1275 int is_error_page(struct page *page)
1277 return page == bad_page;
1279 EXPORT_SYMBOL_GPL(is_error_page);
1281 int is_error_pfn(pfn_t pfn)
1283 return pfn == bad_pfn;
1285 EXPORT_SYMBOL_GPL(is_error_pfn);
1287 static inline unsigned long bad_hva(void)
1292 int kvm_is_error_hva(unsigned long addr)
1294 return addr == bad_hva();
1296 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1298 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1302 for (i = 0; i < kvm->nmemslots; ++i) {
1303 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1305 if (gfn >= memslot->base_gfn
1306 && gfn < memslot->base_gfn + memslot->npages)
1311 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1313 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1315 gfn = unalias_gfn(kvm, gfn);
1316 return gfn_to_memslot_unaliased(kvm, gfn);
1319 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1323 gfn = unalias_gfn(kvm, gfn);
1324 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1325 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1327 if (gfn >= memslot->base_gfn
1328 && gfn < memslot->base_gfn + memslot->npages)
1333 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1335 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1337 struct kvm_memory_slot *slot;
1339 gfn = unalias_gfn(kvm, gfn);
1340 slot = gfn_to_memslot_unaliased(kvm, gfn);
1343 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1345 EXPORT_SYMBOL_GPL(gfn_to_hva);
1347 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1349 struct page *page[1];
1356 addr = gfn_to_hva(kvm, gfn);
1357 if (kvm_is_error_hva(addr)) {
1359 return page_to_pfn(bad_page);
1362 npages = get_user_pages_fast(addr, 1, 1, page);
1364 if (unlikely(npages != 1)) {
1365 struct vm_area_struct *vma;
1367 down_read(¤t->mm->mmap_sem);
1368 vma = find_vma(current->mm, addr);
1370 if (vma == NULL || addr < vma->vm_start ||
1371 !(vma->vm_flags & VM_PFNMAP)) {
1372 up_read(¤t->mm->mmap_sem);
1374 return page_to_pfn(bad_page);
1377 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1378 up_read(¤t->mm->mmap_sem);
1379 BUG_ON(!kvm_is_mmio_pfn(pfn));
1381 pfn = page_to_pfn(page[0]);
1386 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1388 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1392 pfn = gfn_to_pfn(kvm, gfn);
1393 if (!kvm_is_mmio_pfn(pfn))
1394 return pfn_to_page(pfn);
1396 WARN_ON(kvm_is_mmio_pfn(pfn));
1402 EXPORT_SYMBOL_GPL(gfn_to_page);
1404 void kvm_release_page_clean(struct page *page)
1406 kvm_release_pfn_clean(page_to_pfn(page));
1408 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1410 void kvm_release_pfn_clean(pfn_t pfn)
1412 if (!kvm_is_mmio_pfn(pfn))
1413 put_page(pfn_to_page(pfn));
1415 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1417 void kvm_release_page_dirty(struct page *page)
1419 kvm_release_pfn_dirty(page_to_pfn(page));
1421 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1423 void kvm_release_pfn_dirty(pfn_t pfn)
1425 kvm_set_pfn_dirty(pfn);
1426 kvm_release_pfn_clean(pfn);
1428 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1430 void kvm_set_page_dirty(struct page *page)
1432 kvm_set_pfn_dirty(page_to_pfn(page));
1434 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1436 void kvm_set_pfn_dirty(pfn_t pfn)
1438 if (!kvm_is_mmio_pfn(pfn)) {
1439 struct page *page = pfn_to_page(pfn);
1440 if (!PageReserved(page))
1444 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1446 void kvm_set_pfn_accessed(pfn_t pfn)
1448 if (!kvm_is_mmio_pfn(pfn))
1449 mark_page_accessed(pfn_to_page(pfn));
1451 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1453 void kvm_get_pfn(pfn_t pfn)
1455 if (!kvm_is_mmio_pfn(pfn))
1456 get_page(pfn_to_page(pfn));
1458 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1460 static int next_segment(unsigned long len, int offset)
1462 if (len > PAGE_SIZE - offset)
1463 return PAGE_SIZE - offset;
1468 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1474 addr = gfn_to_hva(kvm, gfn);
1475 if (kvm_is_error_hva(addr))
1477 r = copy_from_user(data, (void __user *)addr + offset, len);
1482 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1484 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1486 gfn_t gfn = gpa >> PAGE_SHIFT;
1488 int offset = offset_in_page(gpa);
1491 while ((seg = next_segment(len, offset)) != 0) {
1492 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1502 EXPORT_SYMBOL_GPL(kvm_read_guest);
1504 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1509 gfn_t gfn = gpa >> PAGE_SHIFT;
1510 int offset = offset_in_page(gpa);
1512 addr = gfn_to_hva(kvm, gfn);
1513 if (kvm_is_error_hva(addr))
1515 pagefault_disable();
1516 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1522 EXPORT_SYMBOL(kvm_read_guest_atomic);
1524 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1525 int offset, int len)
1530 addr = gfn_to_hva(kvm, gfn);
1531 if (kvm_is_error_hva(addr))
1533 r = copy_to_user((void __user *)addr + offset, data, len);
1536 mark_page_dirty(kvm, gfn);
1539 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1541 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
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_write_guest_page(kvm, gfn, data, offset, seg);
1561 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1563 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1565 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1567 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1569 gfn_t gfn = gpa >> PAGE_SHIFT;
1571 int offset = offset_in_page(gpa);
1574 while ((seg = next_segment(len, offset)) != 0) {
1575 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1584 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1586 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1588 struct kvm_memory_slot *memslot;
1590 gfn = unalias_gfn(kvm, gfn);
1591 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1592 if (memslot && memslot->dirty_bitmap) {
1593 unsigned long rel_gfn = gfn - memslot->base_gfn;
1596 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1597 set_bit(rel_gfn, memslot->dirty_bitmap);
1602 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1604 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1609 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1611 if (kvm_cpu_has_interrupt(vcpu) ||
1612 kvm_cpu_has_pending_timer(vcpu) ||
1613 kvm_arch_vcpu_runnable(vcpu)) {
1614 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1617 if (signal_pending(current))
1625 finish_wait(&vcpu->wq, &wait);
1628 void kvm_resched(struct kvm_vcpu *vcpu)
1630 if (!need_resched())
1634 EXPORT_SYMBOL_GPL(kvm_resched);
1636 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1638 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1641 if (vmf->pgoff == 0)
1642 page = virt_to_page(vcpu->run);
1644 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1645 page = virt_to_page(vcpu->arch.pio_data);
1647 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1648 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1649 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1652 return VM_FAULT_SIGBUS;
1658 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1659 .fault = kvm_vcpu_fault,
1662 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1664 vma->vm_ops = &kvm_vcpu_vm_ops;
1668 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1670 struct kvm_vcpu *vcpu = filp->private_data;
1672 kvm_put_kvm(vcpu->kvm);
1676 static struct file_operations kvm_vcpu_fops = {
1677 .release = kvm_vcpu_release,
1678 .unlocked_ioctl = kvm_vcpu_ioctl,
1679 .compat_ioctl = kvm_vcpu_ioctl,
1680 .mmap = kvm_vcpu_mmap,
1684 * Allocates an inode for the vcpu.
1686 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1688 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1690 kvm_put_kvm(vcpu->kvm);
1695 * Creates some virtual cpus. Good luck creating more than one.
1697 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1700 struct kvm_vcpu *vcpu;
1705 vcpu = kvm_arch_vcpu_create(kvm, n);
1707 return PTR_ERR(vcpu);
1709 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1711 r = kvm_arch_vcpu_setup(vcpu);
1715 mutex_lock(&kvm->lock);
1716 if (kvm->vcpus[n]) {
1720 kvm->vcpus[n] = vcpu;
1721 mutex_unlock(&kvm->lock);
1723 /* Now it's all set up, let userspace reach it */
1725 r = create_vcpu_fd(vcpu);
1731 mutex_lock(&kvm->lock);
1732 kvm->vcpus[n] = NULL;
1734 mutex_unlock(&kvm->lock);
1735 kvm_arch_vcpu_destroy(vcpu);
1739 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1742 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1743 vcpu->sigset_active = 1;
1744 vcpu->sigset = *sigset;
1746 vcpu->sigset_active = 0;
1750 #ifdef __KVM_HAVE_MSIX
1751 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1752 struct kvm_assigned_msix_nr *entry_nr)
1755 struct kvm_assigned_dev_kernel *adev;
1757 mutex_lock(&kvm->lock);
1759 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1760 entry_nr->assigned_dev_id);
1766 if (adev->entries_nr == 0) {
1767 adev->entries_nr = entry_nr->entry_nr;
1768 if (adev->entries_nr == 0 ||
1769 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1774 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1777 if (!adev->host_msix_entries) {
1781 adev->guest_msix_entries = kzalloc(
1782 sizeof(struct kvm_guest_msix_entry) *
1783 entry_nr->entry_nr, GFP_KERNEL);
1784 if (!adev->guest_msix_entries) {
1785 kfree(adev->host_msix_entries);
1789 } else /* Not allowed set MSI-X number twice */
1792 mutex_unlock(&kvm->lock);
1796 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1797 struct kvm_assigned_msix_entry *entry)
1800 struct kvm_assigned_dev_kernel *adev;
1802 mutex_lock(&kvm->lock);
1804 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1805 entry->assigned_dev_id);
1809 goto msix_entry_out;
1812 for (i = 0; i < adev->entries_nr; i++)
1813 if (adev->guest_msix_entries[i].vector == 0 ||
1814 adev->guest_msix_entries[i].entry == entry->entry) {
1815 adev->guest_msix_entries[i].entry = entry->entry;
1816 adev->guest_msix_entries[i].vector = entry->gsi;
1817 adev->host_msix_entries[i].entry = entry->entry;
1820 if (i == adev->entries_nr) {
1822 goto msix_entry_out;
1826 mutex_unlock(&kvm->lock);
1832 static long kvm_vcpu_ioctl(struct file *filp,
1833 unsigned int ioctl, unsigned long arg)
1835 struct kvm_vcpu *vcpu = filp->private_data;
1836 void __user *argp = (void __user *)arg;
1838 struct kvm_fpu *fpu = NULL;
1839 struct kvm_sregs *kvm_sregs = NULL;
1841 if (vcpu->kvm->mm != current->mm)
1848 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1850 case KVM_GET_REGS: {
1851 struct kvm_regs *kvm_regs;
1854 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1857 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1861 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1868 case KVM_SET_REGS: {
1869 struct kvm_regs *kvm_regs;
1872 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1876 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1878 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1886 case KVM_GET_SREGS: {
1887 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1891 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1895 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1900 case KVM_SET_SREGS: {
1901 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1906 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1908 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1914 case KVM_GET_MP_STATE: {
1915 struct kvm_mp_state mp_state;
1917 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1921 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1926 case KVM_SET_MP_STATE: {
1927 struct kvm_mp_state mp_state;
1930 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1932 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1938 case KVM_TRANSLATE: {
1939 struct kvm_translation tr;
1942 if (copy_from_user(&tr, argp, sizeof tr))
1944 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1948 if (copy_to_user(argp, &tr, sizeof tr))
1953 case KVM_SET_GUEST_DEBUG: {
1954 struct kvm_guest_debug dbg;
1957 if (copy_from_user(&dbg, argp, sizeof dbg))
1959 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1965 case KVM_SET_SIGNAL_MASK: {
1966 struct kvm_signal_mask __user *sigmask_arg = argp;
1967 struct kvm_signal_mask kvm_sigmask;
1968 sigset_t sigset, *p;
1973 if (copy_from_user(&kvm_sigmask, argp,
1974 sizeof kvm_sigmask))
1977 if (kvm_sigmask.len != sizeof sigset)
1980 if (copy_from_user(&sigset, sigmask_arg->sigset,
1985 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1989 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1993 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1997 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2003 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2008 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2010 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2017 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2025 static long kvm_vm_ioctl(struct file *filp,
2026 unsigned int ioctl, unsigned long arg)
2028 struct kvm *kvm = filp->private_data;
2029 void __user *argp = (void __user *)arg;
2032 if (kvm->mm != current->mm)
2035 case KVM_CREATE_VCPU:
2036 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2040 case KVM_SET_USER_MEMORY_REGION: {
2041 struct kvm_userspace_memory_region kvm_userspace_mem;
2044 if (copy_from_user(&kvm_userspace_mem, argp,
2045 sizeof kvm_userspace_mem))
2048 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2053 case KVM_GET_DIRTY_LOG: {
2054 struct kvm_dirty_log log;
2057 if (copy_from_user(&log, argp, sizeof log))
2059 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2064 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2065 case KVM_REGISTER_COALESCED_MMIO: {
2066 struct kvm_coalesced_mmio_zone zone;
2068 if (copy_from_user(&zone, argp, sizeof zone))
2071 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2077 case KVM_UNREGISTER_COALESCED_MMIO: {
2078 struct kvm_coalesced_mmio_zone zone;
2080 if (copy_from_user(&zone, argp, sizeof zone))
2083 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2090 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2091 case KVM_ASSIGN_PCI_DEVICE: {
2092 struct kvm_assigned_pci_dev assigned_dev;
2095 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2097 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2102 case KVM_ASSIGN_IRQ: {
2106 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2107 case KVM_ASSIGN_DEV_IRQ: {
2108 struct kvm_assigned_irq assigned_irq;
2111 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2113 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2118 case KVM_DEASSIGN_DEV_IRQ: {
2119 struct kvm_assigned_irq assigned_irq;
2122 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2124 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2131 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2132 case KVM_DEASSIGN_PCI_DEVICE: {
2133 struct kvm_assigned_pci_dev assigned_dev;
2136 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2138 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2144 #ifdef KVM_CAP_IRQ_ROUTING
2145 case KVM_SET_GSI_ROUTING: {
2146 struct kvm_irq_routing routing;
2147 struct kvm_irq_routing __user *urouting;
2148 struct kvm_irq_routing_entry *entries;
2151 if (copy_from_user(&routing, argp, sizeof(routing)))
2154 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2159 entries = vmalloc(routing.nr * sizeof(*entries));
2164 if (copy_from_user(entries, urouting->entries,
2165 routing.nr * sizeof(*entries)))
2166 goto out_free_irq_routing;
2167 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2169 out_free_irq_routing:
2173 #ifdef __KVM_HAVE_MSIX
2174 case KVM_ASSIGN_SET_MSIX_NR: {
2175 struct kvm_assigned_msix_nr entry_nr;
2177 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2179 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2184 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2185 struct kvm_assigned_msix_entry entry;
2187 if (copy_from_user(&entry, argp, sizeof entry))
2189 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2195 #endif /* KVM_CAP_IRQ_ROUTING */
2197 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2203 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2205 struct page *page[1];
2208 gfn_t gfn = vmf->pgoff;
2209 struct kvm *kvm = vma->vm_file->private_data;
2211 addr = gfn_to_hva(kvm, gfn);
2212 if (kvm_is_error_hva(addr))
2213 return VM_FAULT_SIGBUS;
2215 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2217 if (unlikely(npages != 1))
2218 return VM_FAULT_SIGBUS;
2220 vmf->page = page[0];
2224 static struct vm_operations_struct kvm_vm_vm_ops = {
2225 .fault = kvm_vm_fault,
2228 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2230 vma->vm_ops = &kvm_vm_vm_ops;
2234 static struct file_operations kvm_vm_fops = {
2235 .release = kvm_vm_release,
2236 .unlocked_ioctl = kvm_vm_ioctl,
2237 .compat_ioctl = kvm_vm_ioctl,
2238 .mmap = kvm_vm_mmap,
2241 static int kvm_dev_ioctl_create_vm(void)
2246 kvm = kvm_create_vm();
2248 return PTR_ERR(kvm);
2249 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2256 static long kvm_dev_ioctl_check_extension_generic(long arg)
2259 case KVM_CAP_USER_MEMORY:
2260 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2261 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2263 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2264 case KVM_CAP_IRQ_ROUTING:
2265 return KVM_MAX_IRQ_ROUTES;
2270 return kvm_dev_ioctl_check_extension(arg);
2273 static long kvm_dev_ioctl(struct file *filp,
2274 unsigned int ioctl, unsigned long arg)
2279 case KVM_GET_API_VERSION:
2283 r = KVM_API_VERSION;
2289 r = kvm_dev_ioctl_create_vm();
2291 case KVM_CHECK_EXTENSION:
2292 r = kvm_dev_ioctl_check_extension_generic(arg);
2294 case KVM_GET_VCPU_MMAP_SIZE:
2298 r = PAGE_SIZE; /* struct kvm_run */
2300 r += PAGE_SIZE; /* pio data page */
2302 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2303 r += PAGE_SIZE; /* coalesced mmio ring page */
2306 case KVM_TRACE_ENABLE:
2307 case KVM_TRACE_PAUSE:
2308 case KVM_TRACE_DISABLE:
2309 r = kvm_trace_ioctl(ioctl, arg);
2312 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2318 static struct file_operations kvm_chardev_ops = {
2319 .unlocked_ioctl = kvm_dev_ioctl,
2320 .compat_ioctl = kvm_dev_ioctl,
2323 static struct miscdevice kvm_dev = {
2329 static void hardware_enable(void *junk)
2331 int cpu = raw_smp_processor_id();
2333 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2335 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2336 kvm_arch_hardware_enable(NULL);
2339 static void hardware_disable(void *junk)
2341 int cpu = raw_smp_processor_id();
2343 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2345 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2346 kvm_arch_hardware_disable(NULL);
2349 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2354 val &= ~CPU_TASKS_FROZEN;
2357 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2359 hardware_disable(NULL);
2361 case CPU_UP_CANCELED:
2362 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2364 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2367 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2369 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2376 asmlinkage void kvm_handle_fault_on_reboot(void)
2379 /* spin while reset goes on */
2382 /* Fault while not rebooting. We want the trace. */
2385 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2387 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2390 if (val == SYS_RESTART) {
2392 * Some (well, at least mine) BIOSes hang on reboot if
2395 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2396 kvm_rebooting = true;
2397 on_each_cpu(hardware_disable, NULL, 1);
2402 static struct notifier_block kvm_reboot_notifier = {
2403 .notifier_call = kvm_reboot,
2407 void kvm_io_bus_init(struct kvm_io_bus *bus)
2409 memset(bus, 0, sizeof(*bus));
2412 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2416 for (i = 0; i < bus->dev_count; i++) {
2417 struct kvm_io_device *pos = bus->devs[i];
2419 kvm_iodevice_destructor(pos);
2423 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2424 gpa_t addr, int len, int is_write)
2428 for (i = 0; i < bus->dev_count; i++) {
2429 struct kvm_io_device *pos = bus->devs[i];
2431 if (pos->in_range(pos, addr, len, is_write))
2438 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2440 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2442 bus->devs[bus->dev_count++] = dev;
2445 static struct notifier_block kvm_cpu_notifier = {
2446 .notifier_call = kvm_cpu_hotplug,
2447 .priority = 20, /* must be > scheduler priority */
2450 static int vm_stat_get(void *_offset, u64 *val)
2452 unsigned offset = (long)_offset;
2456 spin_lock(&kvm_lock);
2457 list_for_each_entry(kvm, &vm_list, vm_list)
2458 *val += *(u32 *)((void *)kvm + offset);
2459 spin_unlock(&kvm_lock);
2463 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2465 static int vcpu_stat_get(void *_offset, u64 *val)
2467 unsigned offset = (long)_offset;
2469 struct kvm_vcpu *vcpu;
2473 spin_lock(&kvm_lock);
2474 list_for_each_entry(kvm, &vm_list, vm_list)
2475 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2476 vcpu = kvm->vcpus[i];
2478 *val += *(u32 *)((void *)vcpu + offset);
2480 spin_unlock(&kvm_lock);
2484 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2486 static struct file_operations *stat_fops[] = {
2487 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2488 [KVM_STAT_VM] = &vm_stat_fops,
2491 static void kvm_init_debug(void)
2493 struct kvm_stats_debugfs_item *p;
2495 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2496 for (p = debugfs_entries; p->name; ++p)
2497 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2498 (void *)(long)p->offset,
2499 stat_fops[p->kind]);
2502 static void kvm_exit_debug(void)
2504 struct kvm_stats_debugfs_item *p;
2506 for (p = debugfs_entries; p->name; ++p)
2507 debugfs_remove(p->dentry);
2508 debugfs_remove(kvm_debugfs_dir);
2511 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2513 hardware_disable(NULL);
2517 static int kvm_resume(struct sys_device *dev)
2519 hardware_enable(NULL);
2523 static struct sysdev_class kvm_sysdev_class = {
2525 .suspend = kvm_suspend,
2526 .resume = kvm_resume,
2529 static struct sys_device kvm_sysdev = {
2531 .cls = &kvm_sysdev_class,
2534 struct page *bad_page;
2538 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2540 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2543 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2545 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2547 kvm_arch_vcpu_load(vcpu, cpu);
2550 static void kvm_sched_out(struct preempt_notifier *pn,
2551 struct task_struct *next)
2553 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2555 kvm_arch_vcpu_put(vcpu);
2558 int kvm_init(void *opaque, unsigned int vcpu_size,
2559 struct module *module)
2566 r = kvm_arch_init(opaque);
2570 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2572 if (bad_page == NULL) {
2577 bad_pfn = page_to_pfn(bad_page);
2579 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2583 cpumask_clear(cpus_hardware_enabled);
2585 r = kvm_arch_hardware_setup();
2589 for_each_online_cpu(cpu) {
2590 smp_call_function_single(cpu,
2591 kvm_arch_check_processor_compat,
2597 on_each_cpu(hardware_enable, NULL, 1);
2598 r = register_cpu_notifier(&kvm_cpu_notifier);
2601 register_reboot_notifier(&kvm_reboot_notifier);
2603 r = sysdev_class_register(&kvm_sysdev_class);
2607 r = sysdev_register(&kvm_sysdev);
2611 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2612 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2613 __alignof__(struct kvm_vcpu),
2615 if (!kvm_vcpu_cache) {
2620 kvm_chardev_ops.owner = module;
2621 kvm_vm_fops.owner = module;
2622 kvm_vcpu_fops.owner = module;
2624 r = misc_register(&kvm_dev);
2626 printk(KERN_ERR "kvm: misc device register failed\n");
2630 kvm_preempt_ops.sched_in = kvm_sched_in;
2631 kvm_preempt_ops.sched_out = kvm_sched_out;
2636 kmem_cache_destroy(kvm_vcpu_cache);
2638 sysdev_unregister(&kvm_sysdev);
2640 sysdev_class_unregister(&kvm_sysdev_class);
2642 unregister_reboot_notifier(&kvm_reboot_notifier);
2643 unregister_cpu_notifier(&kvm_cpu_notifier);
2645 on_each_cpu(hardware_disable, NULL, 1);
2647 kvm_arch_hardware_unsetup();
2649 free_cpumask_var(cpus_hardware_enabled);
2651 __free_page(bad_page);
2658 EXPORT_SYMBOL_GPL(kvm_init);
2662 kvm_trace_cleanup();
2663 misc_deregister(&kvm_dev);
2664 kmem_cache_destroy(kvm_vcpu_cache);
2665 sysdev_unregister(&kvm_sysdev);
2666 sysdev_class_unregister(&kvm_sysdev_class);
2667 unregister_reboot_notifier(&kvm_reboot_notifier);
2668 unregister_cpu_notifier(&kvm_cpu_notifier);
2669 on_each_cpu(hardware_disable, NULL, 1);
2670 kvm_arch_hardware_unsetup();
2673 free_cpumask_var(cpus_hardware_enabled);
2674 __free_page(bad_page);
2676 EXPORT_SYMBOL_GPL(kvm_exit);