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 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
65 DEFINE_SPINLOCK(kvm_lock);
68 static cpumask_var_t cpus_hardware_enabled;
70 struct kmem_cache *kvm_vcpu_cache;
71 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
73 static __read_mostly struct preempt_ops kvm_preempt_ops;
75 struct dentry *kvm_debugfs_dir;
77 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
80 static bool kvm_rebooting;
82 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
83 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
86 struct list_head *ptr;
87 struct kvm_assigned_dev_kernel *match;
89 list_for_each(ptr, head) {
90 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
91 if (match->assigned_dev_id == assigned_dev_id)
97 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
98 *assigned_dev, int irq)
101 struct msix_entry *host_msix_entries;
103 host_msix_entries = assigned_dev->host_msix_entries;
106 for (i = 0; i < assigned_dev->entries_nr; i++)
107 if (irq == host_msix_entries[i].vector) {
112 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
119 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
121 struct kvm_assigned_dev_kernel *assigned_dev;
125 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
127 kvm = assigned_dev->kvm;
129 /* This is taken to safely inject irq inside the guest. When
130 * the interrupt injection (or the ioapic code) uses a
131 * finer-grained lock, update this
133 mutex_lock(&kvm->lock);
134 spin_lock_irq(&assigned_dev->assigned_dev_lock);
135 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
136 struct kvm_guest_msix_entry *guest_entries =
137 assigned_dev->guest_msix_entries;
138 for (i = 0; i < assigned_dev->entries_nr; i++) {
139 if (!(guest_entries[i].flags &
140 KVM_ASSIGNED_MSIX_PENDING))
142 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
143 kvm_set_irq(assigned_dev->kvm,
144 assigned_dev->irq_source_id,
145 guest_entries[i].vector, 1);
146 irq = assigned_dev->host_msix_entries[i].vector;
149 assigned_dev->host_irq_disabled = false;
152 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
153 assigned_dev->guest_irq, 1);
154 if (assigned_dev->irq_requested_type &
155 KVM_DEV_IRQ_GUEST_MSI) {
156 enable_irq(assigned_dev->host_irq);
157 assigned_dev->host_irq_disabled = false;
161 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
162 mutex_unlock(&assigned_dev->kvm->lock);
165 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
168 struct kvm_assigned_dev_kernel *assigned_dev =
169 (struct kvm_assigned_dev_kernel *) dev_id;
171 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
172 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
173 int index = find_index_from_host_irq(assigned_dev, irq);
176 assigned_dev->guest_msix_entries[index].flags |=
177 KVM_ASSIGNED_MSIX_PENDING;
180 schedule_work(&assigned_dev->interrupt_work);
182 disable_irq_nosync(irq);
183 assigned_dev->host_irq_disabled = true;
186 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
190 /* Ack the irq line for an assigned device */
191 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
193 struct kvm_assigned_dev_kernel *dev;
199 dev = container_of(kian, struct kvm_assigned_dev_kernel,
202 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
204 /* The guest irq may be shared so this ack may be
205 * from another device.
207 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
208 if (dev->host_irq_disabled) {
209 enable_irq(dev->host_irq);
210 dev->host_irq_disabled = false;
212 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
215 static void deassign_guest_irq(struct kvm *kvm,
216 struct kvm_assigned_dev_kernel *assigned_dev)
218 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
219 assigned_dev->ack_notifier.gsi = -1;
221 if (assigned_dev->irq_source_id != -1)
222 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
223 assigned_dev->irq_source_id = -1;
224 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
227 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
228 static void deassign_host_irq(struct kvm *kvm,
229 struct kvm_assigned_dev_kernel *assigned_dev)
232 * In kvm_free_device_irq, cancel_work_sync return true if:
233 * 1. work is scheduled, and then cancelled.
234 * 2. work callback is executed.
236 * The first one ensured that the irq is disabled and no more events
237 * would happen. But for the second one, the irq may be enabled (e.g.
238 * for MSI). So we disable irq here to prevent further events.
240 * Notice this maybe result in nested disable if the interrupt type is
241 * INTx, but it's OK for we are going to free it.
243 * If this function is a part of VM destroy, please ensure that till
244 * now, the kvm state is still legal for probably we also have to wait
245 * interrupt_work done.
247 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
249 for (i = 0; i < assigned_dev->entries_nr; i++)
250 disable_irq_nosync(assigned_dev->
251 host_msix_entries[i].vector);
253 cancel_work_sync(&assigned_dev->interrupt_work);
255 for (i = 0; i < assigned_dev->entries_nr; i++)
256 free_irq(assigned_dev->host_msix_entries[i].vector,
257 (void *)assigned_dev);
259 assigned_dev->entries_nr = 0;
260 kfree(assigned_dev->host_msix_entries);
261 kfree(assigned_dev->guest_msix_entries);
262 pci_disable_msix(assigned_dev->dev);
264 /* Deal with MSI and INTx */
265 disable_irq_nosync(assigned_dev->host_irq);
266 cancel_work_sync(&assigned_dev->interrupt_work);
268 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
270 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
271 pci_disable_msi(assigned_dev->dev);
274 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
277 static int kvm_deassign_irq(struct kvm *kvm,
278 struct kvm_assigned_dev_kernel *assigned_dev,
279 unsigned long irq_requested_type)
281 unsigned long guest_irq_type, host_irq_type;
283 if (!irqchip_in_kernel(kvm))
285 /* no irq assignment to deassign */
286 if (!assigned_dev->irq_requested_type)
289 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
290 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
293 deassign_host_irq(kvm, assigned_dev);
295 deassign_guest_irq(kvm, assigned_dev);
300 static void kvm_free_assigned_irq(struct kvm *kvm,
301 struct kvm_assigned_dev_kernel *assigned_dev)
303 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
306 static void kvm_free_assigned_device(struct kvm *kvm,
307 struct kvm_assigned_dev_kernel
310 kvm_free_assigned_irq(kvm, assigned_dev);
312 pci_reset_function(assigned_dev->dev);
314 pci_release_regions(assigned_dev->dev);
315 pci_disable_device(assigned_dev->dev);
316 pci_dev_put(assigned_dev->dev);
318 list_del(&assigned_dev->list);
322 void kvm_free_all_assigned_devices(struct kvm *kvm)
324 struct list_head *ptr, *ptr2;
325 struct kvm_assigned_dev_kernel *assigned_dev;
327 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
328 assigned_dev = list_entry(ptr,
329 struct kvm_assigned_dev_kernel,
332 kvm_free_assigned_device(kvm, assigned_dev);
336 static int assigned_device_enable_host_intx(struct kvm *kvm,
337 struct kvm_assigned_dev_kernel *dev)
339 dev->host_irq = dev->dev->irq;
340 /* Even though this is PCI, we don't want to use shared
341 * interrupts. Sharing host devices with guest-assigned devices
342 * on the same interrupt line is not a happy situation: there
343 * are going to be long delays in accepting, acking, etc.
345 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
346 0, "kvm_assigned_intx_device", (void *)dev))
351 #ifdef __KVM_HAVE_MSI
352 static int assigned_device_enable_host_msi(struct kvm *kvm,
353 struct kvm_assigned_dev_kernel *dev)
357 if (!dev->dev->msi_enabled) {
358 r = pci_enable_msi(dev->dev);
363 dev->host_irq = dev->dev->irq;
364 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
365 "kvm_assigned_msi_device", (void *)dev)) {
366 pci_disable_msi(dev->dev);
374 #ifdef __KVM_HAVE_MSIX
375 static int assigned_device_enable_host_msix(struct kvm *kvm,
376 struct kvm_assigned_dev_kernel *dev)
380 /* host_msix_entries and guest_msix_entries should have been
382 if (dev->entries_nr == 0)
385 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
389 for (i = 0; i < dev->entries_nr; i++) {
390 r = request_irq(dev->host_msix_entries[i].vector,
391 kvm_assigned_dev_intr, 0,
392 "kvm_assigned_msix_device",
394 /* FIXME: free requested_irq's on failure */
404 static int assigned_device_enable_guest_intx(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 = irq->guest_irq;
413 #ifdef __KVM_HAVE_MSI
414 static int assigned_device_enable_guest_msi(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;
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;
434 static int assign_host_irq(struct kvm *kvm,
435 struct kvm_assigned_dev_kernel *dev,
440 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
443 switch (host_irq_type) {
444 case KVM_DEV_IRQ_HOST_INTX:
445 r = assigned_device_enable_host_intx(kvm, dev);
447 #ifdef __KVM_HAVE_MSI
448 case KVM_DEV_IRQ_HOST_MSI:
449 r = assigned_device_enable_host_msi(kvm, dev);
452 #ifdef __KVM_HAVE_MSIX
453 case KVM_DEV_IRQ_HOST_MSIX:
454 r = assigned_device_enable_host_msix(kvm, dev);
462 dev->irq_requested_type |= host_irq_type;
467 static int assign_guest_irq(struct kvm *kvm,
468 struct kvm_assigned_dev_kernel *dev,
469 struct kvm_assigned_irq *irq,
470 unsigned long guest_irq_type)
475 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
478 id = kvm_request_irq_source_id(kvm);
482 dev->irq_source_id = id;
484 switch (guest_irq_type) {
485 case KVM_DEV_IRQ_GUEST_INTX:
486 r = assigned_device_enable_guest_intx(kvm, dev, irq);
488 #ifdef __KVM_HAVE_MSI
489 case KVM_DEV_IRQ_GUEST_MSI:
490 r = assigned_device_enable_guest_msi(kvm, dev, irq);
493 #ifdef __KVM_HAVE_MSIX
494 case KVM_DEV_IRQ_GUEST_MSIX:
495 r = assigned_device_enable_guest_msix(kvm, dev, irq);
503 dev->irq_requested_type |= guest_irq_type;
504 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
506 kvm_free_irq_source_id(kvm, dev->irq_source_id);
511 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
512 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
513 struct kvm_assigned_irq *assigned_irq)
516 struct kvm_assigned_dev_kernel *match;
517 unsigned long host_irq_type, guest_irq_type;
519 if (!capable(CAP_SYS_RAWIO))
522 if (!irqchip_in_kernel(kvm))
525 mutex_lock(&kvm->lock);
527 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
528 assigned_irq->assigned_dev_id);
532 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
533 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
536 /* can only assign one type at a time */
537 if (hweight_long(host_irq_type) > 1)
539 if (hweight_long(guest_irq_type) > 1)
541 if (host_irq_type == 0 && guest_irq_type == 0)
546 r = assign_host_irq(kvm, match, host_irq_type);
551 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
553 mutex_unlock(&kvm->lock);
557 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
558 struct kvm_assigned_irq
562 struct kvm_assigned_dev_kernel *match;
564 mutex_lock(&kvm->lock);
566 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
567 assigned_irq->assigned_dev_id);
571 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
573 mutex_unlock(&kvm->lock);
577 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
578 struct kvm_assigned_pci_dev *assigned_dev)
581 struct kvm_assigned_dev_kernel *match;
584 down_read(&kvm->slots_lock);
585 mutex_lock(&kvm->lock);
587 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
588 assigned_dev->assigned_dev_id);
590 /* device already assigned */
595 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
597 printk(KERN_INFO "%s: Couldn't allocate memory\n",
602 dev = pci_get_bus_and_slot(assigned_dev->busnr,
603 assigned_dev->devfn);
605 printk(KERN_INFO "%s: host device not found\n", __func__);
609 if (pci_enable_device(dev)) {
610 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
614 r = pci_request_regions(dev, "kvm_assigned_device");
616 printk(KERN_INFO "%s: Could not get access to device regions\n",
621 pci_reset_function(dev);
623 match->assigned_dev_id = assigned_dev->assigned_dev_id;
624 match->host_busnr = assigned_dev->busnr;
625 match->host_devfn = assigned_dev->devfn;
626 match->flags = assigned_dev->flags;
628 spin_lock_init(&match->assigned_dev_lock);
629 match->irq_source_id = -1;
631 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
632 INIT_WORK(&match->interrupt_work,
633 kvm_assigned_dev_interrupt_work_handler);
635 list_add(&match->list, &kvm->arch.assigned_dev_head);
637 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
638 if (!kvm->arch.iommu_domain) {
639 r = kvm_iommu_map_guest(kvm);
643 r = kvm_assign_device(kvm, match);
649 mutex_unlock(&kvm->lock);
650 up_read(&kvm->slots_lock);
653 list_del(&match->list);
654 pci_release_regions(dev);
656 pci_disable_device(dev);
661 mutex_unlock(&kvm->lock);
662 up_read(&kvm->slots_lock);
667 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
668 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
669 struct kvm_assigned_pci_dev *assigned_dev)
672 struct kvm_assigned_dev_kernel *match;
674 mutex_lock(&kvm->lock);
676 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
677 assigned_dev->assigned_dev_id);
679 printk(KERN_INFO "%s: device hasn't been assigned before, "
680 "so cannot be deassigned\n", __func__);
685 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
686 kvm_deassign_device(kvm, match);
688 kvm_free_assigned_device(kvm, match);
691 mutex_unlock(&kvm->lock);
696 static inline int valid_vcpu(int n)
698 return likely(n >= 0 && n < KVM_MAX_VCPUS);
701 inline int kvm_is_mmio_pfn(pfn_t pfn)
703 if (pfn_valid(pfn)) {
704 struct page *page = compound_head(pfn_to_page(pfn));
705 return PageReserved(page);
712 * Switches to specified vcpu, until a matching vcpu_put()
714 void vcpu_load(struct kvm_vcpu *vcpu)
718 mutex_lock(&vcpu->mutex);
720 preempt_notifier_register(&vcpu->preempt_notifier);
721 kvm_arch_vcpu_load(vcpu, cpu);
725 void vcpu_put(struct kvm_vcpu *vcpu)
728 kvm_arch_vcpu_put(vcpu);
729 preempt_notifier_unregister(&vcpu->preempt_notifier);
731 mutex_unlock(&vcpu->mutex);
734 static void ack_flush(void *_completed)
738 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
743 struct kvm_vcpu *vcpu;
745 if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
749 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
750 vcpu = kvm->vcpus[i];
753 if (test_and_set_bit(req, &vcpu->requests))
756 if (cpus != NULL && cpu != -1 && cpu != me)
757 cpumask_set_cpu(cpu, cpus);
759 if (unlikely(cpus == NULL))
760 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
761 else if (!cpumask_empty(cpus))
762 smp_call_function_many(cpus, ack_flush, NULL, 1);
766 free_cpumask_var(cpus);
770 void kvm_flush_remote_tlbs(struct kvm *kvm)
772 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
773 ++kvm->stat.remote_tlb_flush;
776 void kvm_reload_remote_mmus(struct kvm *kvm)
778 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
781 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
786 mutex_init(&vcpu->mutex);
790 init_waitqueue_head(&vcpu->wq);
792 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
797 vcpu->run = page_address(page);
799 r = kvm_arch_vcpu_init(vcpu);
805 free_page((unsigned long)vcpu->run);
809 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
811 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
813 kvm_arch_vcpu_uninit(vcpu);
814 free_page((unsigned long)vcpu->run);
816 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
818 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
819 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
821 return container_of(mn, struct kvm, mmu_notifier);
824 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
825 struct mm_struct *mm,
826 unsigned long address)
828 struct kvm *kvm = mmu_notifier_to_kvm(mn);
832 * When ->invalidate_page runs, the linux pte has been zapped
833 * already but the page is still allocated until
834 * ->invalidate_page returns. So if we increase the sequence
835 * here the kvm page fault will notice if the spte can't be
836 * established because the page is going to be freed. If
837 * instead the kvm page fault establishes the spte before
838 * ->invalidate_page runs, kvm_unmap_hva will release it
841 * The sequence increase only need to be seen at spin_unlock
842 * time, and not at spin_lock time.
844 * Increasing the sequence after the spin_unlock would be
845 * unsafe because the kvm page fault could then establish the
846 * pte after kvm_unmap_hva returned, without noticing the page
847 * is going to be freed.
849 spin_lock(&kvm->mmu_lock);
850 kvm->mmu_notifier_seq++;
851 need_tlb_flush = kvm_unmap_hva(kvm, address);
852 spin_unlock(&kvm->mmu_lock);
854 /* we've to flush the tlb before the pages can be freed */
856 kvm_flush_remote_tlbs(kvm);
860 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
861 struct mm_struct *mm,
865 struct kvm *kvm = mmu_notifier_to_kvm(mn);
866 int need_tlb_flush = 0;
868 spin_lock(&kvm->mmu_lock);
870 * The count increase must become visible at unlock time as no
871 * spte can be established without taking the mmu_lock and
872 * count is also read inside the mmu_lock critical section.
874 kvm->mmu_notifier_count++;
875 for (; start < end; start += PAGE_SIZE)
876 need_tlb_flush |= kvm_unmap_hva(kvm, start);
877 spin_unlock(&kvm->mmu_lock);
879 /* we've to flush the tlb before the pages can be freed */
881 kvm_flush_remote_tlbs(kvm);
884 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
885 struct mm_struct *mm,
889 struct kvm *kvm = mmu_notifier_to_kvm(mn);
891 spin_lock(&kvm->mmu_lock);
893 * This sequence increase will notify the kvm page fault that
894 * the page that is going to be mapped in the spte could have
897 kvm->mmu_notifier_seq++;
899 * The above sequence increase must be visible before the
900 * below count decrease but both values are read by the kvm
901 * page fault under mmu_lock spinlock so we don't need to add
902 * a smb_wmb() here in between the two.
904 kvm->mmu_notifier_count--;
905 spin_unlock(&kvm->mmu_lock);
907 BUG_ON(kvm->mmu_notifier_count < 0);
910 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
911 struct mm_struct *mm,
912 unsigned long address)
914 struct kvm *kvm = mmu_notifier_to_kvm(mn);
917 spin_lock(&kvm->mmu_lock);
918 young = kvm_age_hva(kvm, address);
919 spin_unlock(&kvm->mmu_lock);
922 kvm_flush_remote_tlbs(kvm);
927 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
928 struct mm_struct *mm)
930 struct kvm *kvm = mmu_notifier_to_kvm(mn);
931 kvm_arch_flush_shadow(kvm);
934 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
935 .invalidate_page = kvm_mmu_notifier_invalidate_page,
936 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
937 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
938 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
939 .release = kvm_mmu_notifier_release,
941 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
943 static struct kvm *kvm_create_vm(void)
945 struct kvm *kvm = kvm_arch_create_vm();
946 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
952 #ifdef CONFIG_HAVE_KVM_IRQCHIP
953 INIT_LIST_HEAD(&kvm->irq_routing);
954 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
957 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
958 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
961 return ERR_PTR(-ENOMEM);
963 kvm->coalesced_mmio_ring =
964 (struct kvm_coalesced_mmio_ring *)page_address(page);
967 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
970 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
971 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
973 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
982 kvm->mm = current->mm;
983 atomic_inc(&kvm->mm->mm_count);
984 spin_lock_init(&kvm->mmu_lock);
985 kvm_io_bus_init(&kvm->pio_bus);
986 mutex_init(&kvm->lock);
987 kvm_io_bus_init(&kvm->mmio_bus);
988 init_rwsem(&kvm->slots_lock);
989 atomic_set(&kvm->users_count, 1);
990 spin_lock(&kvm_lock);
991 list_add(&kvm->vm_list, &vm_list);
992 spin_unlock(&kvm_lock);
993 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
994 kvm_coalesced_mmio_init(kvm);
1001 * Free any memory in @free but not in @dont.
1003 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1004 struct kvm_memory_slot *dont)
1006 if (!dont || free->rmap != dont->rmap)
1009 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1010 vfree(free->dirty_bitmap);
1012 if (!dont || free->lpage_info != dont->lpage_info)
1013 vfree(free->lpage_info);
1016 free->dirty_bitmap = NULL;
1018 free->lpage_info = NULL;
1021 void kvm_free_physmem(struct kvm *kvm)
1025 for (i = 0; i < kvm->nmemslots; ++i)
1026 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1029 static void kvm_destroy_vm(struct kvm *kvm)
1031 struct mm_struct *mm = kvm->mm;
1033 kvm_arch_sync_events(kvm);
1034 spin_lock(&kvm_lock);
1035 list_del(&kvm->vm_list);
1036 spin_unlock(&kvm_lock);
1037 kvm_free_irq_routing(kvm);
1038 kvm_io_bus_destroy(&kvm->pio_bus);
1039 kvm_io_bus_destroy(&kvm->mmio_bus);
1040 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1041 if (kvm->coalesced_mmio_ring != NULL)
1042 free_page((unsigned long)kvm->coalesced_mmio_ring);
1044 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1045 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1047 kvm_arch_flush_shadow(kvm);
1049 kvm_arch_destroy_vm(kvm);
1053 void kvm_get_kvm(struct kvm *kvm)
1055 atomic_inc(&kvm->users_count);
1057 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1059 void kvm_put_kvm(struct kvm *kvm)
1061 if (atomic_dec_and_test(&kvm->users_count))
1062 kvm_destroy_vm(kvm);
1064 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1067 static int kvm_vm_release(struct inode *inode, struct file *filp)
1069 struct kvm *kvm = filp->private_data;
1076 * Allocate some memory and give it an address in the guest physical address
1079 * Discontiguous memory is allowed, mostly for framebuffers.
1081 * Must be called holding mmap_sem for write.
1083 int __kvm_set_memory_region(struct kvm *kvm,
1084 struct kvm_userspace_memory_region *mem,
1089 unsigned long npages, ugfn;
1090 unsigned long largepages, i;
1091 struct kvm_memory_slot *memslot;
1092 struct kvm_memory_slot old, new;
1095 /* General sanity checks */
1096 if (mem->memory_size & (PAGE_SIZE - 1))
1098 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1100 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1102 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1104 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1107 memslot = &kvm->memslots[mem->slot];
1108 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1109 npages = mem->memory_size >> PAGE_SHIFT;
1112 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1114 new = old = *memslot;
1116 new.base_gfn = base_gfn;
1117 new.npages = npages;
1118 new.flags = mem->flags;
1120 /* Disallow changing a memory slot's size. */
1122 if (npages && old.npages && npages != old.npages)
1125 /* Check for overlaps */
1127 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1128 struct kvm_memory_slot *s = &kvm->memslots[i];
1130 if (s == memslot || !s->npages)
1132 if (!((base_gfn + npages <= s->base_gfn) ||
1133 (base_gfn >= s->base_gfn + s->npages)))
1137 /* Free page dirty bitmap if unneeded */
1138 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1139 new.dirty_bitmap = NULL;
1143 /* Allocate if a slot is being created */
1145 if (npages && !new.rmap) {
1146 new.rmap = vmalloc(npages * sizeof(struct page *));
1151 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1153 new.user_alloc = user_alloc;
1155 * hva_to_rmmap() serialzies with the mmu_lock and to be
1156 * safe it has to ignore memslots with !user_alloc &&
1160 new.userspace_addr = mem->userspace_addr;
1162 new.userspace_addr = 0;
1164 if (npages && !new.lpage_info) {
1165 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1166 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1168 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1170 if (!new.lpage_info)
1173 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1175 if (base_gfn % KVM_PAGES_PER_HPAGE)
1176 new.lpage_info[0].write_count = 1;
1177 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1178 new.lpage_info[largepages-1].write_count = 1;
1179 ugfn = new.userspace_addr >> PAGE_SHIFT;
1181 * If the gfn and userspace address are not aligned wrt each
1182 * other, disable large page support for this slot
1184 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1185 for (i = 0; i < largepages; ++i)
1186 new.lpage_info[i].write_count = 1;
1189 /* Allocate page dirty bitmap if needed */
1190 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1191 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1193 new.dirty_bitmap = vmalloc(dirty_bytes);
1194 if (!new.dirty_bitmap)
1196 memset(new.dirty_bitmap, 0, dirty_bytes);
1198 kvm_arch_flush_shadow(kvm);
1200 #endif /* not defined CONFIG_S390 */
1203 kvm_arch_flush_shadow(kvm);
1205 spin_lock(&kvm->mmu_lock);
1206 if (mem->slot >= kvm->nmemslots)
1207 kvm->nmemslots = mem->slot + 1;
1210 spin_unlock(&kvm->mmu_lock);
1212 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1214 spin_lock(&kvm->mmu_lock);
1216 spin_unlock(&kvm->mmu_lock);
1220 kvm_free_physmem_slot(&old, npages ? &new : NULL);
1221 /* Slot deletion case: we have to update the current slot */
1222 spin_lock(&kvm->mmu_lock);
1225 spin_unlock(&kvm->mmu_lock);
1227 /* map the pages in iommu page table */
1228 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1235 kvm_free_physmem_slot(&new, &old);
1240 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1242 int kvm_set_memory_region(struct kvm *kvm,
1243 struct kvm_userspace_memory_region *mem,
1248 down_write(&kvm->slots_lock);
1249 r = __kvm_set_memory_region(kvm, mem, user_alloc);
1250 up_write(&kvm->slots_lock);
1253 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1255 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1257 kvm_userspace_memory_region *mem,
1260 if (mem->slot >= KVM_MEMORY_SLOTS)
1262 return kvm_set_memory_region(kvm, mem, user_alloc);
1265 int kvm_get_dirty_log(struct kvm *kvm,
1266 struct kvm_dirty_log *log, int *is_dirty)
1268 struct kvm_memory_slot *memslot;
1271 unsigned long any = 0;
1274 if (log->slot >= KVM_MEMORY_SLOTS)
1277 memslot = &kvm->memslots[log->slot];
1279 if (!memslot->dirty_bitmap)
1282 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1284 for (i = 0; !any && i < n/sizeof(long); ++i)
1285 any = memslot->dirty_bitmap[i];
1288 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1299 int is_error_page(struct page *page)
1301 return page == bad_page;
1303 EXPORT_SYMBOL_GPL(is_error_page);
1305 int is_error_pfn(pfn_t pfn)
1307 return pfn == bad_pfn;
1309 EXPORT_SYMBOL_GPL(is_error_pfn);
1311 static inline unsigned long bad_hva(void)
1316 int kvm_is_error_hva(unsigned long addr)
1318 return addr == bad_hva();
1320 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1322 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1326 for (i = 0; i < kvm->nmemslots; ++i) {
1327 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1329 if (gfn >= memslot->base_gfn
1330 && gfn < memslot->base_gfn + memslot->npages)
1335 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1337 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1339 gfn = unalias_gfn(kvm, gfn);
1340 return gfn_to_memslot_unaliased(kvm, gfn);
1343 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1347 gfn = unalias_gfn(kvm, gfn);
1348 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1349 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1351 if (gfn >= memslot->base_gfn
1352 && gfn < memslot->base_gfn + memslot->npages)
1357 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1359 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1361 struct kvm_memory_slot *slot;
1363 gfn = unalias_gfn(kvm, gfn);
1364 slot = gfn_to_memslot_unaliased(kvm, gfn);
1367 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1369 EXPORT_SYMBOL_GPL(gfn_to_hva);
1371 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1373 struct page *page[1];
1380 addr = gfn_to_hva(kvm, gfn);
1381 if (kvm_is_error_hva(addr)) {
1383 return page_to_pfn(bad_page);
1386 npages = get_user_pages_fast(addr, 1, 1, page);
1388 if (unlikely(npages != 1)) {
1389 struct vm_area_struct *vma;
1391 down_read(¤t->mm->mmap_sem);
1392 vma = find_vma(current->mm, addr);
1394 if (vma == NULL || addr < vma->vm_start ||
1395 !(vma->vm_flags & VM_PFNMAP)) {
1396 up_read(¤t->mm->mmap_sem);
1398 return page_to_pfn(bad_page);
1401 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1402 up_read(¤t->mm->mmap_sem);
1403 BUG_ON(!kvm_is_mmio_pfn(pfn));
1405 pfn = page_to_pfn(page[0]);
1410 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1412 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1416 pfn = gfn_to_pfn(kvm, gfn);
1417 if (!kvm_is_mmio_pfn(pfn))
1418 return pfn_to_page(pfn);
1420 WARN_ON(kvm_is_mmio_pfn(pfn));
1426 EXPORT_SYMBOL_GPL(gfn_to_page);
1428 void kvm_release_page_clean(struct page *page)
1430 kvm_release_pfn_clean(page_to_pfn(page));
1432 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1434 void kvm_release_pfn_clean(pfn_t pfn)
1436 if (!kvm_is_mmio_pfn(pfn))
1437 put_page(pfn_to_page(pfn));
1439 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1441 void kvm_release_page_dirty(struct page *page)
1443 kvm_release_pfn_dirty(page_to_pfn(page));
1445 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1447 void kvm_release_pfn_dirty(pfn_t pfn)
1449 kvm_set_pfn_dirty(pfn);
1450 kvm_release_pfn_clean(pfn);
1452 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1454 void kvm_set_page_dirty(struct page *page)
1456 kvm_set_pfn_dirty(page_to_pfn(page));
1458 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1460 void kvm_set_pfn_dirty(pfn_t pfn)
1462 if (!kvm_is_mmio_pfn(pfn)) {
1463 struct page *page = pfn_to_page(pfn);
1464 if (!PageReserved(page))
1468 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1470 void kvm_set_pfn_accessed(pfn_t pfn)
1472 if (!kvm_is_mmio_pfn(pfn))
1473 mark_page_accessed(pfn_to_page(pfn));
1475 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1477 void kvm_get_pfn(pfn_t pfn)
1479 if (!kvm_is_mmio_pfn(pfn))
1480 get_page(pfn_to_page(pfn));
1482 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1484 static int next_segment(unsigned long len, int offset)
1486 if (len > PAGE_SIZE - offset)
1487 return PAGE_SIZE - offset;
1492 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1498 addr = gfn_to_hva(kvm, gfn);
1499 if (kvm_is_error_hva(addr))
1501 r = copy_from_user(data, (void __user *)addr + offset, len);
1506 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1508 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1510 gfn_t gfn = gpa >> PAGE_SHIFT;
1512 int offset = offset_in_page(gpa);
1515 while ((seg = next_segment(len, offset)) != 0) {
1516 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1526 EXPORT_SYMBOL_GPL(kvm_read_guest);
1528 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1533 gfn_t gfn = gpa >> PAGE_SHIFT;
1534 int offset = offset_in_page(gpa);
1536 addr = gfn_to_hva(kvm, gfn);
1537 if (kvm_is_error_hva(addr))
1539 pagefault_disable();
1540 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1546 EXPORT_SYMBOL(kvm_read_guest_atomic);
1548 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1549 int offset, int len)
1554 addr = gfn_to_hva(kvm, gfn);
1555 if (kvm_is_error_hva(addr))
1557 r = copy_to_user((void __user *)addr + offset, data, len);
1560 mark_page_dirty(kvm, gfn);
1563 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1565 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1568 gfn_t gfn = gpa >> PAGE_SHIFT;
1570 int offset = offset_in_page(gpa);
1573 while ((seg = next_segment(len, offset)) != 0) {
1574 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1585 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1587 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1589 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1591 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1593 gfn_t gfn = gpa >> PAGE_SHIFT;
1595 int offset = offset_in_page(gpa);
1598 while ((seg = next_segment(len, offset)) != 0) {
1599 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1608 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1610 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1612 struct kvm_memory_slot *memslot;
1614 gfn = unalias_gfn(kvm, gfn);
1615 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1616 if (memslot && memslot->dirty_bitmap) {
1617 unsigned long rel_gfn = gfn - memslot->base_gfn;
1620 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1621 set_bit(rel_gfn, memslot->dirty_bitmap);
1626 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1628 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1633 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1635 if ((kvm_arch_interrupt_allowed(vcpu) &&
1636 kvm_cpu_has_interrupt(vcpu)) ||
1637 kvm_arch_vcpu_runnable(vcpu)) {
1638 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1641 if (kvm_cpu_has_pending_timer(vcpu))
1643 if (signal_pending(current))
1651 finish_wait(&vcpu->wq, &wait);
1654 void kvm_resched(struct kvm_vcpu *vcpu)
1656 if (!need_resched())
1660 EXPORT_SYMBOL_GPL(kvm_resched);
1662 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1664 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1667 if (vmf->pgoff == 0)
1668 page = virt_to_page(vcpu->run);
1670 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1671 page = virt_to_page(vcpu->arch.pio_data);
1673 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1674 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1675 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1678 return VM_FAULT_SIGBUS;
1684 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1685 .fault = kvm_vcpu_fault,
1688 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1690 vma->vm_ops = &kvm_vcpu_vm_ops;
1694 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1696 struct kvm_vcpu *vcpu = filp->private_data;
1698 kvm_put_kvm(vcpu->kvm);
1702 static struct file_operations kvm_vcpu_fops = {
1703 .release = kvm_vcpu_release,
1704 .unlocked_ioctl = kvm_vcpu_ioctl,
1705 .compat_ioctl = kvm_vcpu_ioctl,
1706 .mmap = kvm_vcpu_mmap,
1710 * Allocates an inode for the vcpu.
1712 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1714 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1716 kvm_put_kvm(vcpu->kvm);
1721 * Creates some virtual cpus. Good luck creating more than one.
1723 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1726 struct kvm_vcpu *vcpu;
1731 vcpu = kvm_arch_vcpu_create(kvm, n);
1733 return PTR_ERR(vcpu);
1735 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1737 r = kvm_arch_vcpu_setup(vcpu);
1741 mutex_lock(&kvm->lock);
1742 if (kvm->vcpus[n]) {
1746 kvm->vcpus[n] = vcpu;
1747 mutex_unlock(&kvm->lock);
1749 /* Now it's all set up, let userspace reach it */
1751 r = create_vcpu_fd(vcpu);
1757 mutex_lock(&kvm->lock);
1758 kvm->vcpus[n] = NULL;
1760 mutex_unlock(&kvm->lock);
1761 kvm_arch_vcpu_destroy(vcpu);
1765 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1768 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1769 vcpu->sigset_active = 1;
1770 vcpu->sigset = *sigset;
1772 vcpu->sigset_active = 0;
1776 #ifdef __KVM_HAVE_MSIX
1777 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1778 struct kvm_assigned_msix_nr *entry_nr)
1781 struct kvm_assigned_dev_kernel *adev;
1783 mutex_lock(&kvm->lock);
1785 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1786 entry_nr->assigned_dev_id);
1792 if (adev->entries_nr == 0) {
1793 adev->entries_nr = entry_nr->entry_nr;
1794 if (adev->entries_nr == 0 ||
1795 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1800 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1803 if (!adev->host_msix_entries) {
1807 adev->guest_msix_entries = kzalloc(
1808 sizeof(struct kvm_guest_msix_entry) *
1809 entry_nr->entry_nr, GFP_KERNEL);
1810 if (!adev->guest_msix_entries) {
1811 kfree(adev->host_msix_entries);
1815 } else /* Not allowed set MSI-X number twice */
1818 mutex_unlock(&kvm->lock);
1822 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1823 struct kvm_assigned_msix_entry *entry)
1826 struct kvm_assigned_dev_kernel *adev;
1828 mutex_lock(&kvm->lock);
1830 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1831 entry->assigned_dev_id);
1835 goto msix_entry_out;
1838 for (i = 0; i < adev->entries_nr; i++)
1839 if (adev->guest_msix_entries[i].vector == 0 ||
1840 adev->guest_msix_entries[i].entry == entry->entry) {
1841 adev->guest_msix_entries[i].entry = entry->entry;
1842 adev->guest_msix_entries[i].vector = entry->gsi;
1843 adev->host_msix_entries[i].entry = entry->entry;
1846 if (i == adev->entries_nr) {
1848 goto msix_entry_out;
1852 mutex_unlock(&kvm->lock);
1858 static long kvm_vcpu_ioctl(struct file *filp,
1859 unsigned int ioctl, unsigned long arg)
1861 struct kvm_vcpu *vcpu = filp->private_data;
1862 void __user *argp = (void __user *)arg;
1864 struct kvm_fpu *fpu = NULL;
1865 struct kvm_sregs *kvm_sregs = NULL;
1867 if (vcpu->kvm->mm != current->mm)
1874 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1876 case KVM_GET_REGS: {
1877 struct kvm_regs *kvm_regs;
1880 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1883 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1887 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1894 case KVM_SET_REGS: {
1895 struct kvm_regs *kvm_regs;
1898 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1902 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1904 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1912 case KVM_GET_SREGS: {
1913 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1917 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1921 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1926 case KVM_SET_SREGS: {
1927 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1932 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1934 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1940 case KVM_GET_MP_STATE: {
1941 struct kvm_mp_state mp_state;
1943 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1947 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1952 case KVM_SET_MP_STATE: {
1953 struct kvm_mp_state mp_state;
1956 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1958 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1964 case KVM_TRANSLATE: {
1965 struct kvm_translation tr;
1968 if (copy_from_user(&tr, argp, sizeof tr))
1970 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1974 if (copy_to_user(argp, &tr, sizeof tr))
1979 case KVM_SET_GUEST_DEBUG: {
1980 struct kvm_guest_debug dbg;
1983 if (copy_from_user(&dbg, argp, sizeof dbg))
1985 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1991 case KVM_SET_SIGNAL_MASK: {
1992 struct kvm_signal_mask __user *sigmask_arg = argp;
1993 struct kvm_signal_mask kvm_sigmask;
1994 sigset_t sigset, *p;
1999 if (copy_from_user(&kvm_sigmask, argp,
2000 sizeof kvm_sigmask))
2003 if (kvm_sigmask.len != sizeof sigset)
2006 if (copy_from_user(&sigset, sigmask_arg->sigset,
2011 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2015 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2019 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2023 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2029 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2034 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2036 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2043 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2051 static long kvm_vm_ioctl(struct file *filp,
2052 unsigned int ioctl, unsigned long arg)
2054 struct kvm *kvm = filp->private_data;
2055 void __user *argp = (void __user *)arg;
2058 if (kvm->mm != current->mm)
2061 case KVM_CREATE_VCPU:
2062 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2066 case KVM_SET_USER_MEMORY_REGION: {
2067 struct kvm_userspace_memory_region kvm_userspace_mem;
2070 if (copy_from_user(&kvm_userspace_mem, argp,
2071 sizeof kvm_userspace_mem))
2074 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2079 case KVM_GET_DIRTY_LOG: {
2080 struct kvm_dirty_log log;
2083 if (copy_from_user(&log, argp, sizeof log))
2085 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2090 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2091 case KVM_REGISTER_COALESCED_MMIO: {
2092 struct kvm_coalesced_mmio_zone zone;
2094 if (copy_from_user(&zone, argp, sizeof zone))
2097 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2103 case KVM_UNREGISTER_COALESCED_MMIO: {
2104 struct kvm_coalesced_mmio_zone zone;
2106 if (copy_from_user(&zone, argp, sizeof zone))
2109 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2116 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2117 case KVM_ASSIGN_PCI_DEVICE: {
2118 struct kvm_assigned_pci_dev assigned_dev;
2121 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2123 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2128 case KVM_ASSIGN_IRQ: {
2132 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2133 case KVM_ASSIGN_DEV_IRQ: {
2134 struct kvm_assigned_irq assigned_irq;
2137 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2139 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2144 case KVM_DEASSIGN_DEV_IRQ: {
2145 struct kvm_assigned_irq assigned_irq;
2148 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2150 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2157 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2158 case KVM_DEASSIGN_PCI_DEVICE: {
2159 struct kvm_assigned_pci_dev assigned_dev;
2162 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2164 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2170 #ifdef KVM_CAP_IRQ_ROUTING
2171 case KVM_SET_GSI_ROUTING: {
2172 struct kvm_irq_routing routing;
2173 struct kvm_irq_routing __user *urouting;
2174 struct kvm_irq_routing_entry *entries;
2177 if (copy_from_user(&routing, argp, sizeof(routing)))
2180 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2185 entries = vmalloc(routing.nr * sizeof(*entries));
2190 if (copy_from_user(entries, urouting->entries,
2191 routing.nr * sizeof(*entries)))
2192 goto out_free_irq_routing;
2193 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2195 out_free_irq_routing:
2199 #ifdef __KVM_HAVE_MSIX
2200 case KVM_ASSIGN_SET_MSIX_NR: {
2201 struct kvm_assigned_msix_nr entry_nr;
2203 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2205 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2210 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2211 struct kvm_assigned_msix_entry entry;
2213 if (copy_from_user(&entry, argp, sizeof entry))
2215 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2221 #endif /* KVM_CAP_IRQ_ROUTING */
2223 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2229 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2231 struct page *page[1];
2234 gfn_t gfn = vmf->pgoff;
2235 struct kvm *kvm = vma->vm_file->private_data;
2237 addr = gfn_to_hva(kvm, gfn);
2238 if (kvm_is_error_hva(addr))
2239 return VM_FAULT_SIGBUS;
2241 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2243 if (unlikely(npages != 1))
2244 return VM_FAULT_SIGBUS;
2246 vmf->page = page[0];
2250 static struct vm_operations_struct kvm_vm_vm_ops = {
2251 .fault = kvm_vm_fault,
2254 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2256 vma->vm_ops = &kvm_vm_vm_ops;
2260 static struct file_operations kvm_vm_fops = {
2261 .release = kvm_vm_release,
2262 .unlocked_ioctl = kvm_vm_ioctl,
2263 .compat_ioctl = kvm_vm_ioctl,
2264 .mmap = kvm_vm_mmap,
2267 static int kvm_dev_ioctl_create_vm(void)
2272 kvm = kvm_create_vm();
2274 return PTR_ERR(kvm);
2275 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2282 static long kvm_dev_ioctl_check_extension_generic(long arg)
2285 case KVM_CAP_USER_MEMORY:
2286 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2287 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2289 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2290 case KVM_CAP_IRQ_ROUTING:
2291 return KVM_MAX_IRQ_ROUTES;
2296 return kvm_dev_ioctl_check_extension(arg);
2299 static long kvm_dev_ioctl(struct file *filp,
2300 unsigned int ioctl, unsigned long arg)
2305 case KVM_GET_API_VERSION:
2309 r = KVM_API_VERSION;
2315 r = kvm_dev_ioctl_create_vm();
2317 case KVM_CHECK_EXTENSION:
2318 r = kvm_dev_ioctl_check_extension_generic(arg);
2320 case KVM_GET_VCPU_MMAP_SIZE:
2324 r = PAGE_SIZE; /* struct kvm_run */
2326 r += PAGE_SIZE; /* pio data page */
2328 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2329 r += PAGE_SIZE; /* coalesced mmio ring page */
2332 case KVM_TRACE_ENABLE:
2333 case KVM_TRACE_PAUSE:
2334 case KVM_TRACE_DISABLE:
2335 r = kvm_trace_ioctl(ioctl, arg);
2338 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2344 static struct file_operations kvm_chardev_ops = {
2345 .unlocked_ioctl = kvm_dev_ioctl,
2346 .compat_ioctl = kvm_dev_ioctl,
2349 static struct miscdevice kvm_dev = {
2355 static void hardware_enable(void *junk)
2357 int cpu = raw_smp_processor_id();
2359 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2361 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2362 kvm_arch_hardware_enable(NULL);
2365 static void hardware_disable(void *junk)
2367 int cpu = raw_smp_processor_id();
2369 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2371 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2372 kvm_arch_hardware_disable(NULL);
2375 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2380 val &= ~CPU_TASKS_FROZEN;
2383 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2385 hardware_disable(NULL);
2387 case CPU_UP_CANCELED:
2388 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2390 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2393 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2395 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2402 asmlinkage void kvm_handle_fault_on_reboot(void)
2405 /* spin while reset goes on */
2408 /* Fault while not rebooting. We want the trace. */
2411 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2413 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2417 * Some (well, at least mine) BIOSes hang on reboot if
2420 * And Intel TXT required VMX off for all cpu when system shutdown.
2422 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2423 kvm_rebooting = true;
2424 on_each_cpu(hardware_disable, NULL, 1);
2428 static struct notifier_block kvm_reboot_notifier = {
2429 .notifier_call = kvm_reboot,
2433 void kvm_io_bus_init(struct kvm_io_bus *bus)
2435 memset(bus, 0, sizeof(*bus));
2438 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2442 for (i = 0; i < bus->dev_count; i++) {
2443 struct kvm_io_device *pos = bus->devs[i];
2445 kvm_iodevice_destructor(pos);
2449 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2450 gpa_t addr, int len, int is_write)
2454 for (i = 0; i < bus->dev_count; i++) {
2455 struct kvm_io_device *pos = bus->devs[i];
2457 if (pos->in_range(pos, addr, len, is_write))
2464 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2466 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2468 bus->devs[bus->dev_count++] = dev;
2471 static struct notifier_block kvm_cpu_notifier = {
2472 .notifier_call = kvm_cpu_hotplug,
2473 .priority = 20, /* must be > scheduler priority */
2476 static int vm_stat_get(void *_offset, u64 *val)
2478 unsigned offset = (long)_offset;
2482 spin_lock(&kvm_lock);
2483 list_for_each_entry(kvm, &vm_list, vm_list)
2484 *val += *(u32 *)((void *)kvm + offset);
2485 spin_unlock(&kvm_lock);
2489 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2491 static int vcpu_stat_get(void *_offset, u64 *val)
2493 unsigned offset = (long)_offset;
2495 struct kvm_vcpu *vcpu;
2499 spin_lock(&kvm_lock);
2500 list_for_each_entry(kvm, &vm_list, vm_list)
2501 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2502 vcpu = kvm->vcpus[i];
2504 *val += *(u32 *)((void *)vcpu + offset);
2506 spin_unlock(&kvm_lock);
2510 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2512 static struct file_operations *stat_fops[] = {
2513 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2514 [KVM_STAT_VM] = &vm_stat_fops,
2517 static void kvm_init_debug(void)
2519 struct kvm_stats_debugfs_item *p;
2521 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2522 for (p = debugfs_entries; p->name; ++p)
2523 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2524 (void *)(long)p->offset,
2525 stat_fops[p->kind]);
2528 static void kvm_exit_debug(void)
2530 struct kvm_stats_debugfs_item *p;
2532 for (p = debugfs_entries; p->name; ++p)
2533 debugfs_remove(p->dentry);
2534 debugfs_remove(kvm_debugfs_dir);
2537 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2539 hardware_disable(NULL);
2543 static int kvm_resume(struct sys_device *dev)
2545 hardware_enable(NULL);
2549 static struct sysdev_class kvm_sysdev_class = {
2551 .suspend = kvm_suspend,
2552 .resume = kvm_resume,
2555 static struct sys_device kvm_sysdev = {
2557 .cls = &kvm_sysdev_class,
2560 struct page *bad_page;
2564 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2566 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2569 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2571 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2573 kvm_arch_vcpu_load(vcpu, cpu);
2576 static void kvm_sched_out(struct preempt_notifier *pn,
2577 struct task_struct *next)
2579 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2581 kvm_arch_vcpu_put(vcpu);
2584 int kvm_init(void *opaque, unsigned int vcpu_size,
2585 struct module *module)
2592 r = kvm_arch_init(opaque);
2596 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2598 if (bad_page == NULL) {
2603 bad_pfn = page_to_pfn(bad_page);
2605 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2610 r = kvm_arch_hardware_setup();
2614 for_each_online_cpu(cpu) {
2615 smp_call_function_single(cpu,
2616 kvm_arch_check_processor_compat,
2622 on_each_cpu(hardware_enable, NULL, 1);
2623 r = register_cpu_notifier(&kvm_cpu_notifier);
2626 register_reboot_notifier(&kvm_reboot_notifier);
2628 r = sysdev_class_register(&kvm_sysdev_class);
2632 r = sysdev_register(&kvm_sysdev);
2636 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2637 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2638 __alignof__(struct kvm_vcpu),
2640 if (!kvm_vcpu_cache) {
2645 kvm_chardev_ops.owner = module;
2646 kvm_vm_fops.owner = module;
2647 kvm_vcpu_fops.owner = module;
2649 r = misc_register(&kvm_dev);
2651 printk(KERN_ERR "kvm: misc device register failed\n");
2655 kvm_preempt_ops.sched_in = kvm_sched_in;
2656 kvm_preempt_ops.sched_out = kvm_sched_out;
2661 kmem_cache_destroy(kvm_vcpu_cache);
2663 sysdev_unregister(&kvm_sysdev);
2665 sysdev_class_unregister(&kvm_sysdev_class);
2667 unregister_reboot_notifier(&kvm_reboot_notifier);
2668 unregister_cpu_notifier(&kvm_cpu_notifier);
2670 on_each_cpu(hardware_disable, NULL, 1);
2672 kvm_arch_hardware_unsetup();
2674 free_cpumask_var(cpus_hardware_enabled);
2676 __free_page(bad_page);
2683 EXPORT_SYMBOL_GPL(kvm_init);
2687 kvm_trace_cleanup();
2688 misc_deregister(&kvm_dev);
2689 kmem_cache_destroy(kvm_vcpu_cache);
2690 sysdev_unregister(&kvm_sysdev);
2691 sysdev_class_unregister(&kvm_sysdev_class);
2692 unregister_reboot_notifier(&kvm_reboot_notifier);
2693 unregister_cpu_notifier(&kvm_cpu_notifier);
2694 on_each_cpu(hardware_disable, NULL, 1);
2695 kvm_arch_hardware_unsetup();
2698 free_cpumask_var(cpus_hardware_enabled);
2699 __free_page(bad_page);
2701 EXPORT_SYMBOL_GPL(kvm_exit);