2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
45 #include <asm/processor.h>
47 #include <asm/uaccess.h>
48 #include <asm/pgtable.h>
51 #include <asm/msidef.h>
54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
55 #include "coalesced_mmio.h"
58 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
59 #include <linux/pci.h>
60 #include <linux/interrupt.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
67 static int msi2intx
= 1;
68 module_param(msi2intx
, bool, 0);
70 DEFINE_SPINLOCK(kvm_lock
);
73 static cpumask_var_t cpus_hardware_enabled
;
75 struct kmem_cache
*kvm_vcpu_cache
;
76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
78 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
80 struct dentry
*kvm_debugfs_dir
;
82 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
85 static bool kvm_rebooting
;
87 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
90 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel
*dev
)
93 struct kvm_vcpu
*vcpu
;
94 struct kvm_ioapic
*ioapic
= ioapic_irqchip(dev
->kvm
);
95 int dest_id
= (dev
->guest_msi
.address_lo
& MSI_ADDR_DEST_ID_MASK
)
96 >> MSI_ADDR_DEST_ID_SHIFT
;
97 int vector
= (dev
->guest_msi
.data
& MSI_DATA_VECTOR_MASK
)
98 >> MSI_DATA_VECTOR_SHIFT
;
99 int dest_mode
= test_bit(MSI_ADDR_DEST_MODE_SHIFT
,
100 (unsigned long *)&dev
->guest_msi
.address_lo
);
101 int trig_mode
= test_bit(MSI_DATA_TRIGGER_SHIFT
,
102 (unsigned long *)&dev
->guest_msi
.data
);
103 int delivery_mode
= test_bit(MSI_DATA_DELIVERY_MODE_SHIFT
,
104 (unsigned long *)&dev
->guest_msi
.data
);
109 deliver_bitmask
= kvm_ioapic_get_delivery_bitmask(ioapic
,
111 /* IOAPIC delivery mode value is the same as MSI here */
112 switch (delivery_mode
) {
113 case IOAPIC_LOWEST_PRIORITY
:
114 vcpu
= kvm_get_lowest_prio_vcpu(ioapic
->kvm
, vector
,
117 kvm_apic_set_irq(vcpu
, vector
, trig_mode
);
119 printk(KERN_INFO
"kvm: null lowest priority vcpu!\n");
122 for (vcpu_id
= 0; deliver_bitmask
!= 0; vcpu_id
++) {
123 if (!(deliver_bitmask
& (1 << vcpu_id
)))
125 deliver_bitmask
&= ~(1 << vcpu_id
);
126 vcpu
= ioapic
->kvm
->vcpus
[vcpu_id
];
128 kvm_apic_set_irq(vcpu
, vector
, trig_mode
);
132 printk(KERN_INFO
"kvm: unsupported MSI delivery mode\n");
136 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel
*dev
) {}
139 static struct kvm_assigned_dev_kernel
*kvm_find_assigned_dev(struct list_head
*head
,
142 struct list_head
*ptr
;
143 struct kvm_assigned_dev_kernel
*match
;
145 list_for_each(ptr
, head
) {
146 match
= list_entry(ptr
, struct kvm_assigned_dev_kernel
, list
);
147 if (match
->assigned_dev_id
== assigned_dev_id
)
153 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct
*work
)
155 struct kvm_assigned_dev_kernel
*assigned_dev
;
157 assigned_dev
= container_of(work
, struct kvm_assigned_dev_kernel
,
160 /* This is taken to safely inject irq inside the guest. When
161 * the interrupt injection (or the ioapic code) uses a
162 * finer-grained lock, update this
164 mutex_lock(&assigned_dev
->kvm
->lock
);
165 if (assigned_dev
->irq_requested_type
& KVM_ASSIGNED_DEV_GUEST_INTX
)
166 kvm_set_irq(assigned_dev
->kvm
,
167 assigned_dev
->irq_source_id
,
168 assigned_dev
->guest_irq
, 1);
169 else if (assigned_dev
->irq_requested_type
&
170 KVM_ASSIGNED_DEV_GUEST_MSI
) {
171 assigned_device_msi_dispatch(assigned_dev
);
172 enable_irq(assigned_dev
->host_irq
);
173 assigned_dev
->host_irq_disabled
= false;
175 mutex_unlock(&assigned_dev
->kvm
->lock
);
176 kvm_put_kvm(assigned_dev
->kvm
);
179 static irqreturn_t
kvm_assigned_dev_intr(int irq
, void *dev_id
)
181 struct kvm_assigned_dev_kernel
*assigned_dev
=
182 (struct kvm_assigned_dev_kernel
*) dev_id
;
184 kvm_get_kvm(assigned_dev
->kvm
);
186 schedule_work(&assigned_dev
->interrupt_work
);
188 disable_irq_nosync(irq
);
189 assigned_dev
->host_irq_disabled
= true;
194 /* Ack the irq line for an assigned device */
195 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier
*kian
)
197 struct kvm_assigned_dev_kernel
*dev
;
202 dev
= container_of(kian
, struct kvm_assigned_dev_kernel
,
205 kvm_set_irq(dev
->kvm
, dev
->irq_source_id
, dev
->guest_irq
, 0);
207 /* The guest irq may be shared so this ack may be
208 * from another device.
210 if (dev
->host_irq_disabled
) {
211 enable_irq(dev
->host_irq
);
212 dev
->host_irq_disabled
= false;
216 static void kvm_free_assigned_irq(struct kvm
*kvm
,
217 struct kvm_assigned_dev_kernel
*assigned_dev
)
219 if (!irqchip_in_kernel(kvm
))
222 kvm_unregister_irq_ack_notifier(&assigned_dev
->ack_notifier
);
224 if (assigned_dev
->irq_source_id
!= -1)
225 kvm_free_irq_source_id(kvm
, assigned_dev
->irq_source_id
);
226 assigned_dev
->irq_source_id
= -1;
228 if (!assigned_dev
->irq_requested_type
)
231 if (cancel_work_sync(&assigned_dev
->interrupt_work
))
232 /* We had pending work. That means we will have to take
233 * care of kvm_put_kvm.
237 free_irq(assigned_dev
->host_irq
, (void *)assigned_dev
);
239 if (assigned_dev
->irq_requested_type
& KVM_ASSIGNED_DEV_HOST_MSI
)
240 pci_disable_msi(assigned_dev
->dev
);
242 assigned_dev
->irq_requested_type
= 0;
246 static void kvm_free_assigned_device(struct kvm
*kvm
,
247 struct kvm_assigned_dev_kernel
250 kvm_free_assigned_irq(kvm
, assigned_dev
);
252 pci_reset_function(assigned_dev
->dev
);
254 pci_release_regions(assigned_dev
->dev
);
255 pci_disable_device(assigned_dev
->dev
);
256 pci_dev_put(assigned_dev
->dev
);
258 list_del(&assigned_dev
->list
);
262 void kvm_free_all_assigned_devices(struct kvm
*kvm
)
264 struct list_head
*ptr
, *ptr2
;
265 struct kvm_assigned_dev_kernel
*assigned_dev
;
267 list_for_each_safe(ptr
, ptr2
, &kvm
->arch
.assigned_dev_head
) {
268 assigned_dev
= list_entry(ptr
,
269 struct kvm_assigned_dev_kernel
,
272 kvm_free_assigned_device(kvm
, assigned_dev
);
276 static int assigned_device_update_intx(struct kvm
*kvm
,
277 struct kvm_assigned_dev_kernel
*adev
,
278 struct kvm_assigned_irq
*airq
)
280 adev
->guest_irq
= airq
->guest_irq
;
281 adev
->ack_notifier
.gsi
= airq
->guest_irq
;
283 if (adev
->irq_requested_type
& KVM_ASSIGNED_DEV_HOST_INTX
)
286 if (irqchip_in_kernel(kvm
)) {
288 adev
->irq_requested_type
& KVM_ASSIGNED_DEV_HOST_MSI
) {
289 free_irq(adev
->host_irq
, (void *)kvm
);
290 pci_disable_msi(adev
->dev
);
293 if (!capable(CAP_SYS_RAWIO
))
297 adev
->host_irq
= airq
->host_irq
;
299 adev
->host_irq
= adev
->dev
->irq
;
301 /* Even though this is PCI, we don't want to use shared
302 * interrupts. Sharing host devices with guest-assigned devices
303 * on the same interrupt line is not a happy situation: there
304 * are going to be long delays in accepting, acking, etc.
306 if (request_irq(adev
->host_irq
, kvm_assigned_dev_intr
,
307 0, "kvm_assigned_intx_device", (void *)adev
))
311 adev
->irq_requested_type
= KVM_ASSIGNED_DEV_GUEST_INTX
|
312 KVM_ASSIGNED_DEV_HOST_INTX
;
317 static int assigned_device_update_msi(struct kvm
*kvm
,
318 struct kvm_assigned_dev_kernel
*adev
,
319 struct kvm_assigned_irq
*airq
)
323 if (airq
->flags
& KVM_DEV_IRQ_ASSIGN_ENABLE_MSI
) {
324 /* x86 don't care upper address of guest msi message addr */
325 adev
->irq_requested_type
|= KVM_ASSIGNED_DEV_GUEST_MSI
;
326 adev
->irq_requested_type
&= ~KVM_ASSIGNED_DEV_GUEST_INTX
;
327 adev
->guest_msi
.address_lo
= airq
->guest_msi
.addr_lo
;
328 adev
->guest_msi
.data
= airq
->guest_msi
.data
;
329 adev
->ack_notifier
.gsi
= -1;
330 } else if (msi2intx
) {
331 adev
->irq_requested_type
|= KVM_ASSIGNED_DEV_GUEST_INTX
;
332 adev
->irq_requested_type
&= ~KVM_ASSIGNED_DEV_GUEST_MSI
;
333 adev
->guest_irq
= airq
->guest_irq
;
334 adev
->ack_notifier
.gsi
= airq
->guest_irq
;
337 if (adev
->irq_requested_type
& KVM_ASSIGNED_DEV_HOST_MSI
)
340 if (irqchip_in_kernel(kvm
)) {
342 if (adev
->irq_requested_type
&
343 KVM_ASSIGNED_DEV_HOST_INTX
)
344 free_irq(adev
->host_irq
, (void *)adev
);
346 r
= pci_enable_msi(adev
->dev
);
351 adev
->host_irq
= adev
->dev
->irq
;
352 if (request_irq(adev
->host_irq
, kvm_assigned_dev_intr
, 0,
353 "kvm_assigned_msi_device", (void *)adev
))
358 adev
->irq_requested_type
= KVM_ASSIGNED_DEV_GUEST_MSI
;
360 adev
->irq_requested_type
|= KVM_ASSIGNED_DEV_HOST_MSI
;
365 static int kvm_vm_ioctl_assign_irq(struct kvm
*kvm
,
366 struct kvm_assigned_irq
370 struct kvm_assigned_dev_kernel
*match
;
372 mutex_lock(&kvm
->lock
);
374 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
375 assigned_irq
->assigned_dev_id
);
377 mutex_unlock(&kvm
->lock
);
381 if (!match
->irq_requested_type
) {
382 INIT_WORK(&match
->interrupt_work
,
383 kvm_assigned_dev_interrupt_work_handler
);
384 if (irqchip_in_kernel(kvm
)) {
385 /* Register ack nofitier */
386 match
->ack_notifier
.gsi
= -1;
387 match
->ack_notifier
.irq_acked
=
388 kvm_assigned_dev_ack_irq
;
389 kvm_register_irq_ack_notifier(kvm
,
390 &match
->ack_notifier
);
392 /* Request IRQ source ID */
393 r
= kvm_request_irq_source_id(kvm
);
397 match
->irq_source_id
= r
;
400 /* Determine host device irq type, we can know the
401 * result from dev->msi_enabled */
403 pci_enable_msi(match
->dev
);
409 (assigned_irq
->flags
& KVM_DEV_IRQ_ASSIGN_ENABLE_MSI
)) ||
410 (msi2intx
&& match
->dev
->msi_enabled
)) {
412 r
= assigned_device_update_msi(kvm
, match
, assigned_irq
);
414 printk(KERN_WARNING
"kvm: failed to enable "
421 } else if (assigned_irq
->host_irq
== 0 && match
->dev
->irq
== 0) {
422 /* Host device IRQ 0 means don't support INTx */
425 "kvm: wait device to enable MSI!\n");
429 "kvm: failed to enable MSI device!\n");
434 /* Non-sharing INTx mode */
435 r
= assigned_device_update_intx(kvm
, match
, assigned_irq
);
437 printk(KERN_WARNING
"kvm: failed to enable "
443 mutex_unlock(&kvm
->lock
);
446 mutex_unlock(&kvm
->lock
);
447 kvm_free_assigned_device(kvm
, match
);
451 static int kvm_vm_ioctl_assign_device(struct kvm
*kvm
,
452 struct kvm_assigned_pci_dev
*assigned_dev
)
455 struct kvm_assigned_dev_kernel
*match
;
458 mutex_lock(&kvm
->lock
);
460 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
461 assigned_dev
->assigned_dev_id
);
463 /* device already assigned */
468 match
= kzalloc(sizeof(struct kvm_assigned_dev_kernel
), GFP_KERNEL
);
470 printk(KERN_INFO
"%s: Couldn't allocate memory\n",
475 dev
= pci_get_bus_and_slot(assigned_dev
->busnr
,
476 assigned_dev
->devfn
);
478 printk(KERN_INFO
"%s: host device not found\n", __func__
);
482 if (pci_enable_device(dev
)) {
483 printk(KERN_INFO
"%s: Could not enable PCI device\n", __func__
);
487 r
= pci_request_regions(dev
, "kvm_assigned_device");
489 printk(KERN_INFO
"%s: Could not get access to device regions\n",
494 pci_reset_function(dev
);
496 match
->assigned_dev_id
= assigned_dev
->assigned_dev_id
;
497 match
->host_busnr
= assigned_dev
->busnr
;
498 match
->host_devfn
= assigned_dev
->devfn
;
500 match
->irq_source_id
= -1;
503 list_add(&match
->list
, &kvm
->arch
.assigned_dev_head
);
505 if (assigned_dev
->flags
& KVM_DEV_ASSIGN_ENABLE_IOMMU
) {
506 if (!kvm
->arch
.intel_iommu_domain
) {
507 r
= kvm_iommu_map_guest(kvm
);
511 r
= kvm_assign_device(kvm
, match
);
517 mutex_unlock(&kvm
->lock
);
520 list_del(&match
->list
);
521 pci_release_regions(dev
);
523 pci_disable_device(dev
);
528 mutex_unlock(&kvm
->lock
);
533 static inline int valid_vcpu(int n
)
535 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
538 inline int kvm_is_mmio_pfn(pfn_t pfn
)
541 return PageReserved(pfn_to_page(pfn
));
547 * Switches to specified vcpu, until a matching vcpu_put()
549 void vcpu_load(struct kvm_vcpu
*vcpu
)
553 mutex_lock(&vcpu
->mutex
);
555 preempt_notifier_register(&vcpu
->preempt_notifier
);
556 kvm_arch_vcpu_load(vcpu
, cpu
);
560 void vcpu_put(struct kvm_vcpu
*vcpu
)
563 kvm_arch_vcpu_put(vcpu
);
564 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
566 mutex_unlock(&vcpu
->mutex
);
569 static void ack_flush(void *_completed
)
573 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
578 struct kvm_vcpu
*vcpu
;
580 if (alloc_cpumask_var(&cpus
, GFP_ATOMIC
))
584 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
585 vcpu
= kvm
->vcpus
[i
];
588 if (test_and_set_bit(req
, &vcpu
->requests
))
591 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
)
592 cpumask_set_cpu(cpu
, cpus
);
594 if (unlikely(cpus
== NULL
))
595 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
596 else if (!cpumask_empty(cpus
))
597 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
601 free_cpumask_var(cpus
);
605 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
607 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
608 ++kvm
->stat
.remote_tlb_flush
;
611 void kvm_reload_remote_mmus(struct kvm
*kvm
)
613 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
616 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
621 mutex_init(&vcpu
->mutex
);
625 init_waitqueue_head(&vcpu
->wq
);
627 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
632 vcpu
->run
= page_address(page
);
634 r
= kvm_arch_vcpu_init(vcpu
);
640 free_page((unsigned long)vcpu
->run
);
644 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
646 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
648 kvm_arch_vcpu_uninit(vcpu
);
649 free_page((unsigned long)vcpu
->run
);
651 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
653 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
654 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
656 return container_of(mn
, struct kvm
, mmu_notifier
);
659 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
660 struct mm_struct
*mm
,
661 unsigned long address
)
663 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
667 * When ->invalidate_page runs, the linux pte has been zapped
668 * already but the page is still allocated until
669 * ->invalidate_page returns. So if we increase the sequence
670 * here the kvm page fault will notice if the spte can't be
671 * established because the page is going to be freed. If
672 * instead the kvm page fault establishes the spte before
673 * ->invalidate_page runs, kvm_unmap_hva will release it
676 * The sequence increase only need to be seen at spin_unlock
677 * time, and not at spin_lock time.
679 * Increasing the sequence after the spin_unlock would be
680 * unsafe because the kvm page fault could then establish the
681 * pte after kvm_unmap_hva returned, without noticing the page
682 * is going to be freed.
684 spin_lock(&kvm
->mmu_lock
);
685 kvm
->mmu_notifier_seq
++;
686 need_tlb_flush
= kvm_unmap_hva(kvm
, address
);
687 spin_unlock(&kvm
->mmu_lock
);
689 /* we've to flush the tlb before the pages can be freed */
691 kvm_flush_remote_tlbs(kvm
);
695 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
696 struct mm_struct
*mm
,
700 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
701 int need_tlb_flush
= 0;
703 spin_lock(&kvm
->mmu_lock
);
705 * The count increase must become visible at unlock time as no
706 * spte can be established without taking the mmu_lock and
707 * count is also read inside the mmu_lock critical section.
709 kvm
->mmu_notifier_count
++;
710 for (; start
< end
; start
+= PAGE_SIZE
)
711 need_tlb_flush
|= kvm_unmap_hva(kvm
, start
);
712 spin_unlock(&kvm
->mmu_lock
);
714 /* we've to flush the tlb before the pages can be freed */
716 kvm_flush_remote_tlbs(kvm
);
719 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
720 struct mm_struct
*mm
,
724 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
726 spin_lock(&kvm
->mmu_lock
);
728 * This sequence increase will notify the kvm page fault that
729 * the page that is going to be mapped in the spte could have
732 kvm
->mmu_notifier_seq
++;
734 * The above sequence increase must be visible before the
735 * below count decrease but both values are read by the kvm
736 * page fault under mmu_lock spinlock so we don't need to add
737 * a smb_wmb() here in between the two.
739 kvm
->mmu_notifier_count
--;
740 spin_unlock(&kvm
->mmu_lock
);
742 BUG_ON(kvm
->mmu_notifier_count
< 0);
745 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
746 struct mm_struct
*mm
,
747 unsigned long address
)
749 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
752 spin_lock(&kvm
->mmu_lock
);
753 young
= kvm_age_hva(kvm
, address
);
754 spin_unlock(&kvm
->mmu_lock
);
757 kvm_flush_remote_tlbs(kvm
);
762 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
763 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
764 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
765 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
766 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
768 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
770 static struct kvm
*kvm_create_vm(void)
772 struct kvm
*kvm
= kvm_arch_create_vm();
773 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
780 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
781 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
784 return ERR_PTR(-ENOMEM
);
786 kvm
->coalesced_mmio_ring
=
787 (struct kvm_coalesced_mmio_ring
*)page_address(page
);
790 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
793 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
794 err
= mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
796 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
805 kvm
->mm
= current
->mm
;
806 atomic_inc(&kvm
->mm
->mm_count
);
807 spin_lock_init(&kvm
->mmu_lock
);
808 kvm_io_bus_init(&kvm
->pio_bus
);
809 mutex_init(&kvm
->lock
);
810 kvm_io_bus_init(&kvm
->mmio_bus
);
811 init_rwsem(&kvm
->slots_lock
);
812 atomic_set(&kvm
->users_count
, 1);
813 spin_lock(&kvm_lock
);
814 list_add(&kvm
->vm_list
, &vm_list
);
815 spin_unlock(&kvm_lock
);
816 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
817 kvm_coalesced_mmio_init(kvm
);
824 * Free any memory in @free but not in @dont.
826 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
827 struct kvm_memory_slot
*dont
)
829 if (!dont
|| free
->rmap
!= dont
->rmap
)
832 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
833 vfree(free
->dirty_bitmap
);
835 if (!dont
|| free
->lpage_info
!= dont
->lpage_info
)
836 vfree(free
->lpage_info
);
839 free
->dirty_bitmap
= NULL
;
841 free
->lpage_info
= NULL
;
844 void kvm_free_physmem(struct kvm
*kvm
)
848 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
849 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
852 static void kvm_destroy_vm(struct kvm
*kvm
)
854 struct mm_struct
*mm
= kvm
->mm
;
856 spin_lock(&kvm_lock
);
857 list_del(&kvm
->vm_list
);
858 spin_unlock(&kvm_lock
);
859 kvm_io_bus_destroy(&kvm
->pio_bus
);
860 kvm_io_bus_destroy(&kvm
->mmio_bus
);
861 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
862 if (kvm
->coalesced_mmio_ring
!= NULL
)
863 free_page((unsigned long)kvm
->coalesced_mmio_ring
);
865 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
866 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
868 kvm_arch_destroy_vm(kvm
);
872 void kvm_get_kvm(struct kvm
*kvm
)
874 atomic_inc(&kvm
->users_count
);
876 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
878 void kvm_put_kvm(struct kvm
*kvm
)
880 if (atomic_dec_and_test(&kvm
->users_count
))
883 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
886 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
888 struct kvm
*kvm
= filp
->private_data
;
895 * Allocate some memory and give it an address in the guest physical address
898 * Discontiguous memory is allowed, mostly for framebuffers.
900 * Must be called holding mmap_sem for write.
902 int __kvm_set_memory_region(struct kvm
*kvm
,
903 struct kvm_userspace_memory_region
*mem
,
908 unsigned long npages
;
910 struct kvm_memory_slot
*memslot
;
911 struct kvm_memory_slot old
, new;
914 /* General sanity checks */
915 if (mem
->memory_size
& (PAGE_SIZE
- 1))
917 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
919 if (user_alloc
&& (mem
->userspace_addr
& (PAGE_SIZE
- 1)))
921 if (mem
->slot
>= KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
)
923 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
926 memslot
= &kvm
->memslots
[mem
->slot
];
927 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
928 npages
= mem
->memory_size
>> PAGE_SHIFT
;
931 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
933 new = old
= *memslot
;
935 new.base_gfn
= base_gfn
;
937 new.flags
= mem
->flags
;
939 /* Disallow changing a memory slot's size. */
941 if (npages
&& old
.npages
&& npages
!= old
.npages
)
944 /* Check for overlaps */
946 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
947 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
951 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
952 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
956 /* Free page dirty bitmap if unneeded */
957 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
958 new.dirty_bitmap
= NULL
;
962 /* Allocate if a slot is being created */
964 if (npages
&& !new.rmap
) {
965 new.rmap
= vmalloc(npages
* sizeof(struct page
*));
970 memset(new.rmap
, 0, npages
* sizeof(*new.rmap
));
972 new.user_alloc
= user_alloc
;
974 * hva_to_rmmap() serialzies with the mmu_lock and to be
975 * safe it has to ignore memslots with !user_alloc &&
979 new.userspace_addr
= mem
->userspace_addr
;
981 new.userspace_addr
= 0;
983 if (npages
&& !new.lpage_info
) {
984 int largepages
= npages
/ KVM_PAGES_PER_HPAGE
;
985 if (npages
% KVM_PAGES_PER_HPAGE
)
987 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
990 new.lpage_info
= vmalloc(largepages
* sizeof(*new.lpage_info
));
995 memset(new.lpage_info
, 0, largepages
* sizeof(*new.lpage_info
));
997 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
998 new.lpage_info
[0].write_count
= 1;
999 if ((base_gfn
+npages
) % KVM_PAGES_PER_HPAGE
)
1000 new.lpage_info
[largepages
-1].write_count
= 1;
1003 /* Allocate page dirty bitmap if needed */
1004 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1005 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
1007 new.dirty_bitmap
= vmalloc(dirty_bytes
);
1008 if (!new.dirty_bitmap
)
1010 memset(new.dirty_bitmap
, 0, dirty_bytes
);
1012 #endif /* not defined CONFIG_S390 */
1015 kvm_arch_flush_shadow(kvm
);
1017 spin_lock(&kvm
->mmu_lock
);
1018 if (mem
->slot
>= kvm
->nmemslots
)
1019 kvm
->nmemslots
= mem
->slot
+ 1;
1022 spin_unlock(&kvm
->mmu_lock
);
1024 r
= kvm_arch_set_memory_region(kvm
, mem
, old
, user_alloc
);
1026 spin_lock(&kvm
->mmu_lock
);
1028 spin_unlock(&kvm
->mmu_lock
);
1032 kvm_free_physmem_slot(&old
, npages
? &new : NULL
);
1033 /* Slot deletion case: we have to update the current slot */
1037 /* map the pages in iommu page table */
1038 r
= kvm_iommu_map_pages(kvm
, base_gfn
, npages
);
1045 kvm_free_physmem_slot(&new, &old
);
1050 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1052 int kvm_set_memory_region(struct kvm
*kvm
,
1053 struct kvm_userspace_memory_region
*mem
,
1058 down_write(&kvm
->slots_lock
);
1059 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
1060 up_write(&kvm
->slots_lock
);
1063 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1065 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1067 kvm_userspace_memory_region
*mem
,
1070 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
1072 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
1075 int kvm_get_dirty_log(struct kvm
*kvm
,
1076 struct kvm_dirty_log
*log
, int *is_dirty
)
1078 struct kvm_memory_slot
*memslot
;
1081 unsigned long any
= 0;
1084 if (log
->slot
>= KVM_MEMORY_SLOTS
)
1087 memslot
= &kvm
->memslots
[log
->slot
];
1089 if (!memslot
->dirty_bitmap
)
1092 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1094 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1095 any
= memslot
->dirty_bitmap
[i
];
1098 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1109 int is_error_page(struct page
*page
)
1111 return page
== bad_page
;
1113 EXPORT_SYMBOL_GPL(is_error_page
);
1115 int is_error_pfn(pfn_t pfn
)
1117 return pfn
== bad_pfn
;
1119 EXPORT_SYMBOL_GPL(is_error_pfn
);
1121 static inline unsigned long bad_hva(void)
1126 int kvm_is_error_hva(unsigned long addr
)
1128 return addr
== bad_hva();
1130 EXPORT_SYMBOL_GPL(kvm_is_error_hva
);
1132 struct kvm_memory_slot
*gfn_to_memslot_unaliased(struct kvm
*kvm
, gfn_t gfn
)
1136 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
1137 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
1139 if (gfn
>= memslot
->base_gfn
1140 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
1145 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased
);
1147 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1149 gfn
= unalias_gfn(kvm
, gfn
);
1150 return gfn_to_memslot_unaliased(kvm
, gfn
);
1153 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1157 gfn
= unalias_gfn(kvm
, gfn
);
1158 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
1159 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
1161 if (gfn
>= memslot
->base_gfn
1162 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
1167 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1169 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1171 struct kvm_memory_slot
*slot
;
1173 gfn
= unalias_gfn(kvm
, gfn
);
1174 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
1177 return (slot
->userspace_addr
+ (gfn
- slot
->base_gfn
) * PAGE_SIZE
);
1179 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1181 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1183 struct page
*page
[1];
1190 addr
= gfn_to_hva(kvm
, gfn
);
1191 if (kvm_is_error_hva(addr
)) {
1193 return page_to_pfn(bad_page
);
1196 npages
= get_user_pages_fast(addr
, 1, 1, page
);
1198 if (unlikely(npages
!= 1)) {
1199 struct vm_area_struct
*vma
;
1201 down_read(¤t
->mm
->mmap_sem
);
1202 vma
= find_vma(current
->mm
, addr
);
1204 if (vma
== NULL
|| addr
< vma
->vm_start
||
1205 !(vma
->vm_flags
& VM_PFNMAP
)) {
1206 up_read(¤t
->mm
->mmap_sem
);
1208 return page_to_pfn(bad_page
);
1211 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1212 up_read(¤t
->mm
->mmap_sem
);
1213 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1215 pfn
= page_to_pfn(page
[0]);
1220 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1222 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1226 pfn
= gfn_to_pfn(kvm
, gfn
);
1227 if (!kvm_is_mmio_pfn(pfn
))
1228 return pfn_to_page(pfn
);
1230 WARN_ON(kvm_is_mmio_pfn(pfn
));
1236 EXPORT_SYMBOL_GPL(gfn_to_page
);
1238 void kvm_release_page_clean(struct page
*page
)
1240 kvm_release_pfn_clean(page_to_pfn(page
));
1242 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1244 void kvm_release_pfn_clean(pfn_t pfn
)
1246 if (!kvm_is_mmio_pfn(pfn
))
1247 put_page(pfn_to_page(pfn
));
1249 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1251 void kvm_release_page_dirty(struct page
*page
)
1253 kvm_release_pfn_dirty(page_to_pfn(page
));
1255 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1257 void kvm_release_pfn_dirty(pfn_t pfn
)
1259 kvm_set_pfn_dirty(pfn
);
1260 kvm_release_pfn_clean(pfn
);
1262 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1264 void kvm_set_page_dirty(struct page
*page
)
1266 kvm_set_pfn_dirty(page_to_pfn(page
));
1268 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1270 void kvm_set_pfn_dirty(pfn_t pfn
)
1272 if (!kvm_is_mmio_pfn(pfn
)) {
1273 struct page
*page
= pfn_to_page(pfn
);
1274 if (!PageReserved(page
))
1278 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1280 void kvm_set_pfn_accessed(pfn_t pfn
)
1282 if (!kvm_is_mmio_pfn(pfn
))
1283 mark_page_accessed(pfn_to_page(pfn
));
1285 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1287 void kvm_get_pfn(pfn_t pfn
)
1289 if (!kvm_is_mmio_pfn(pfn
))
1290 get_page(pfn_to_page(pfn
));
1292 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1294 static int next_segment(unsigned long len
, int offset
)
1296 if (len
> PAGE_SIZE
- offset
)
1297 return PAGE_SIZE
- offset
;
1302 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1308 addr
= gfn_to_hva(kvm
, gfn
);
1309 if (kvm_is_error_hva(addr
))
1311 r
= copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1316 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1318 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1320 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1322 int offset
= offset_in_page(gpa
);
1325 while ((seg
= next_segment(len
, offset
)) != 0) {
1326 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1336 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1338 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1343 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1344 int offset
= offset_in_page(gpa
);
1346 addr
= gfn_to_hva(kvm
, gfn
);
1347 if (kvm_is_error_hva(addr
))
1349 pagefault_disable();
1350 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1356 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1358 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1359 int offset
, int len
)
1364 addr
= gfn_to_hva(kvm
, gfn
);
1365 if (kvm_is_error_hva(addr
))
1367 r
= copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1370 mark_page_dirty(kvm
, gfn
);
1373 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1375 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1378 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1380 int offset
= offset_in_page(gpa
);
1383 while ((seg
= next_segment(len
, offset
)) != 0) {
1384 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1395 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1397 return kvm_write_guest_page(kvm
, gfn
, empty_zero_page
, offset
, len
);
1399 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1401 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1403 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1405 int offset
= offset_in_page(gpa
);
1408 while ((seg
= next_segment(len
, offset
)) != 0) {
1409 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1418 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1420 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1422 struct kvm_memory_slot
*memslot
;
1424 gfn
= unalias_gfn(kvm
, gfn
);
1425 memslot
= gfn_to_memslot_unaliased(kvm
, gfn
);
1426 if (memslot
&& memslot
->dirty_bitmap
) {
1427 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1430 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
1431 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
1436 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1438 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1443 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1445 if (kvm_cpu_has_interrupt(vcpu
) ||
1446 kvm_cpu_has_pending_timer(vcpu
) ||
1447 kvm_arch_vcpu_runnable(vcpu
)) {
1448 set_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
1451 if (signal_pending(current
))
1459 finish_wait(&vcpu
->wq
, &wait
);
1462 void kvm_resched(struct kvm_vcpu
*vcpu
)
1464 if (!need_resched())
1468 EXPORT_SYMBOL_GPL(kvm_resched
);
1470 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1472 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1475 if (vmf
->pgoff
== 0)
1476 page
= virt_to_page(vcpu
->run
);
1478 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1479 page
= virt_to_page(vcpu
->arch
.pio_data
);
1481 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1482 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1483 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1486 return VM_FAULT_SIGBUS
;
1492 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
1493 .fault
= kvm_vcpu_fault
,
1496 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1498 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1502 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1504 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1506 kvm_put_kvm(vcpu
->kvm
);
1510 static struct file_operations kvm_vcpu_fops
= {
1511 .release
= kvm_vcpu_release
,
1512 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1513 .compat_ioctl
= kvm_vcpu_ioctl
,
1514 .mmap
= kvm_vcpu_mmap
,
1518 * Allocates an inode for the vcpu.
1520 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1522 int fd
= anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, 0);
1524 kvm_put_kvm(vcpu
->kvm
);
1529 * Creates some virtual cpus. Good luck creating more than one.
1531 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
1534 struct kvm_vcpu
*vcpu
;
1539 vcpu
= kvm_arch_vcpu_create(kvm
, n
);
1541 return PTR_ERR(vcpu
);
1543 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1545 r
= kvm_arch_vcpu_setup(vcpu
);
1549 mutex_lock(&kvm
->lock
);
1550 if (kvm
->vcpus
[n
]) {
1554 kvm
->vcpus
[n
] = vcpu
;
1555 mutex_unlock(&kvm
->lock
);
1557 /* Now it's all set up, let userspace reach it */
1559 r
= create_vcpu_fd(vcpu
);
1565 mutex_lock(&kvm
->lock
);
1566 kvm
->vcpus
[n
] = NULL
;
1568 mutex_unlock(&kvm
->lock
);
1569 kvm_arch_vcpu_destroy(vcpu
);
1573 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1576 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1577 vcpu
->sigset_active
= 1;
1578 vcpu
->sigset
= *sigset
;
1580 vcpu
->sigset_active
= 0;
1584 static long kvm_vcpu_ioctl(struct file
*filp
,
1585 unsigned int ioctl
, unsigned long arg
)
1587 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1588 void __user
*argp
= (void __user
*)arg
;
1590 struct kvm_fpu
*fpu
= NULL
;
1591 struct kvm_sregs
*kvm_sregs
= NULL
;
1593 if (vcpu
->kvm
->mm
!= current
->mm
)
1600 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1602 case KVM_GET_REGS
: {
1603 struct kvm_regs
*kvm_regs
;
1606 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1609 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1613 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1620 case KVM_SET_REGS
: {
1621 struct kvm_regs
*kvm_regs
;
1624 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1628 if (copy_from_user(kvm_regs
, argp
, sizeof(struct kvm_regs
)))
1630 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1638 case KVM_GET_SREGS
: {
1639 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1643 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1647 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
1652 case KVM_SET_SREGS
: {
1653 kvm_sregs
= kmalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1658 if (copy_from_user(kvm_sregs
, argp
, sizeof(struct kvm_sregs
)))
1660 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
1666 case KVM_GET_MP_STATE
: {
1667 struct kvm_mp_state mp_state
;
1669 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1673 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
1678 case KVM_SET_MP_STATE
: {
1679 struct kvm_mp_state mp_state
;
1682 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
1684 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
1690 case KVM_TRANSLATE
: {
1691 struct kvm_translation tr
;
1694 if (copy_from_user(&tr
, argp
, sizeof tr
))
1696 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
1700 if (copy_to_user(argp
, &tr
, sizeof tr
))
1705 case KVM_DEBUG_GUEST
: {
1706 struct kvm_debug_guest dbg
;
1709 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
1711 r
= kvm_arch_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
1717 case KVM_SET_SIGNAL_MASK
: {
1718 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
1719 struct kvm_signal_mask kvm_sigmask
;
1720 sigset_t sigset
, *p
;
1725 if (copy_from_user(&kvm_sigmask
, argp
,
1726 sizeof kvm_sigmask
))
1729 if (kvm_sigmask
.len
!= sizeof sigset
)
1732 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
1737 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
1741 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
1745 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
1749 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
1755 fpu
= kmalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
1760 if (copy_from_user(fpu
, argp
, sizeof(struct kvm_fpu
)))
1762 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
1769 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1777 static long kvm_vm_ioctl(struct file
*filp
,
1778 unsigned int ioctl
, unsigned long arg
)
1780 struct kvm
*kvm
= filp
->private_data
;
1781 void __user
*argp
= (void __user
*)arg
;
1784 if (kvm
->mm
!= current
->mm
)
1787 case KVM_CREATE_VCPU
:
1788 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
1792 case KVM_SET_USER_MEMORY_REGION
: {
1793 struct kvm_userspace_memory_region kvm_userspace_mem
;
1796 if (copy_from_user(&kvm_userspace_mem
, argp
,
1797 sizeof kvm_userspace_mem
))
1800 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 1);
1805 case KVM_GET_DIRTY_LOG
: {
1806 struct kvm_dirty_log log
;
1809 if (copy_from_user(&log
, argp
, sizeof log
))
1811 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
1816 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1817 case KVM_REGISTER_COALESCED_MMIO
: {
1818 struct kvm_coalesced_mmio_zone zone
;
1820 if (copy_from_user(&zone
, argp
, sizeof zone
))
1823 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
1829 case KVM_UNREGISTER_COALESCED_MMIO
: {
1830 struct kvm_coalesced_mmio_zone zone
;
1832 if (copy_from_user(&zone
, argp
, sizeof zone
))
1835 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
1842 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1843 case KVM_ASSIGN_PCI_DEVICE
: {
1844 struct kvm_assigned_pci_dev assigned_dev
;
1847 if (copy_from_user(&assigned_dev
, argp
, sizeof assigned_dev
))
1849 r
= kvm_vm_ioctl_assign_device(kvm
, &assigned_dev
);
1854 case KVM_ASSIGN_IRQ
: {
1855 struct kvm_assigned_irq assigned_irq
;
1858 if (copy_from_user(&assigned_irq
, argp
, sizeof assigned_irq
))
1860 r
= kvm_vm_ioctl_assign_irq(kvm
, &assigned_irq
);
1867 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
1873 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1875 struct page
*page
[1];
1878 gfn_t gfn
= vmf
->pgoff
;
1879 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1881 addr
= gfn_to_hva(kvm
, gfn
);
1882 if (kvm_is_error_hva(addr
))
1883 return VM_FAULT_SIGBUS
;
1885 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
1887 if (unlikely(npages
!= 1))
1888 return VM_FAULT_SIGBUS
;
1890 vmf
->page
= page
[0];
1894 static struct vm_operations_struct kvm_vm_vm_ops
= {
1895 .fault
= kvm_vm_fault
,
1898 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1900 vma
->vm_ops
= &kvm_vm_vm_ops
;
1904 static struct file_operations kvm_vm_fops
= {
1905 .release
= kvm_vm_release
,
1906 .unlocked_ioctl
= kvm_vm_ioctl
,
1907 .compat_ioctl
= kvm_vm_ioctl
,
1908 .mmap
= kvm_vm_mmap
,
1911 static int kvm_dev_ioctl_create_vm(void)
1916 kvm
= kvm_create_vm();
1918 return PTR_ERR(kvm
);
1919 fd
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, 0);
1926 static long kvm_dev_ioctl_check_extension_generic(long arg
)
1929 case KVM_CAP_USER_MEMORY
:
1930 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
1935 return kvm_dev_ioctl_check_extension(arg
);
1938 static long kvm_dev_ioctl(struct file
*filp
,
1939 unsigned int ioctl
, unsigned long arg
)
1944 case KVM_GET_API_VERSION
:
1948 r
= KVM_API_VERSION
;
1954 r
= kvm_dev_ioctl_create_vm();
1956 case KVM_CHECK_EXTENSION
:
1957 r
= kvm_dev_ioctl_check_extension_generic(arg
);
1959 case KVM_GET_VCPU_MMAP_SIZE
:
1963 r
= PAGE_SIZE
; /* struct kvm_run */
1965 r
+= PAGE_SIZE
; /* pio data page */
1967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1968 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
1971 case KVM_TRACE_ENABLE
:
1972 case KVM_TRACE_PAUSE
:
1973 case KVM_TRACE_DISABLE
:
1974 r
= kvm_trace_ioctl(ioctl
, arg
);
1977 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
1983 static struct file_operations kvm_chardev_ops
= {
1984 .unlocked_ioctl
= kvm_dev_ioctl
,
1985 .compat_ioctl
= kvm_dev_ioctl
,
1988 static struct miscdevice kvm_dev
= {
1994 static void hardware_enable(void *junk
)
1996 int cpu
= raw_smp_processor_id();
1998 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2000 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2001 kvm_arch_hardware_enable(NULL
);
2004 static void hardware_disable(void *junk
)
2006 int cpu
= raw_smp_processor_id();
2008 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2010 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2011 kvm_arch_hardware_disable(NULL
);
2014 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2019 val
&= ~CPU_TASKS_FROZEN
;
2022 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2024 hardware_disable(NULL
);
2026 case CPU_UP_CANCELED
:
2027 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2029 smp_call_function_single(cpu
, hardware_disable
, NULL
, 1);
2032 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2034 smp_call_function_single(cpu
, hardware_enable
, NULL
, 1);
2041 asmlinkage
void kvm_handle_fault_on_reboot(void)
2044 /* spin while reset goes on */
2047 /* Fault while not rebooting. We want the trace. */
2050 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot
);
2052 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2055 if (val
== SYS_RESTART
) {
2057 * Some (well, at least mine) BIOSes hang on reboot if
2060 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2061 kvm_rebooting
= true;
2062 on_each_cpu(hardware_disable
, NULL
, 1);
2067 static struct notifier_block kvm_reboot_notifier
= {
2068 .notifier_call
= kvm_reboot
,
2072 void kvm_io_bus_init(struct kvm_io_bus
*bus
)
2074 memset(bus
, 0, sizeof(*bus
));
2077 void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2081 for (i
= 0; i
< bus
->dev_count
; i
++) {
2082 struct kvm_io_device
*pos
= bus
->devs
[i
];
2084 kvm_iodevice_destructor(pos
);
2088 struct kvm_io_device
*kvm_io_bus_find_dev(struct kvm_io_bus
*bus
,
2089 gpa_t addr
, int len
, int is_write
)
2093 for (i
= 0; i
< bus
->dev_count
; i
++) {
2094 struct kvm_io_device
*pos
= bus
->devs
[i
];
2096 if (pos
->in_range(pos
, addr
, len
, is_write
))
2103 void kvm_io_bus_register_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
)
2105 BUG_ON(bus
->dev_count
> (NR_IOBUS_DEVS
-1));
2107 bus
->devs
[bus
->dev_count
++] = dev
;
2110 static struct notifier_block kvm_cpu_notifier
= {
2111 .notifier_call
= kvm_cpu_hotplug
,
2112 .priority
= 20, /* must be > scheduler priority */
2115 static int vm_stat_get(void *_offset
, u64
*val
)
2117 unsigned offset
= (long)_offset
;
2121 spin_lock(&kvm_lock
);
2122 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2123 *val
+= *(u32
*)((void *)kvm
+ offset
);
2124 spin_unlock(&kvm_lock
);
2128 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2130 static int vcpu_stat_get(void *_offset
, u64
*val
)
2132 unsigned offset
= (long)_offset
;
2134 struct kvm_vcpu
*vcpu
;
2138 spin_lock(&kvm_lock
);
2139 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2140 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2141 vcpu
= kvm
->vcpus
[i
];
2143 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2145 spin_unlock(&kvm_lock
);
2149 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2151 static struct file_operations
*stat_fops
[] = {
2152 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2153 [KVM_STAT_VM
] = &vm_stat_fops
,
2156 static void kvm_init_debug(void)
2158 struct kvm_stats_debugfs_item
*p
;
2160 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2161 for (p
= debugfs_entries
; p
->name
; ++p
)
2162 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2163 (void *)(long)p
->offset
,
2164 stat_fops
[p
->kind
]);
2167 static void kvm_exit_debug(void)
2169 struct kvm_stats_debugfs_item
*p
;
2171 for (p
= debugfs_entries
; p
->name
; ++p
)
2172 debugfs_remove(p
->dentry
);
2173 debugfs_remove(kvm_debugfs_dir
);
2176 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2178 hardware_disable(NULL
);
2182 static int kvm_resume(struct sys_device
*dev
)
2184 hardware_enable(NULL
);
2188 static struct sysdev_class kvm_sysdev_class
= {
2190 .suspend
= kvm_suspend
,
2191 .resume
= kvm_resume
,
2194 static struct sys_device kvm_sysdev
= {
2196 .cls
= &kvm_sysdev_class
,
2199 struct page
*bad_page
;
2203 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2205 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2208 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2210 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2212 kvm_arch_vcpu_load(vcpu
, cpu
);
2215 static void kvm_sched_out(struct preempt_notifier
*pn
,
2216 struct task_struct
*next
)
2218 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2220 kvm_arch_vcpu_put(vcpu
);
2223 int kvm_init(void *opaque
, unsigned int vcpu_size
,
2224 struct module
*module
)
2231 r
= kvm_arch_init(opaque
);
2235 bad_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2237 if (bad_page
== NULL
) {
2242 bad_pfn
= page_to_pfn(bad_page
);
2244 if (!alloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2249 r
= kvm_arch_hardware_setup();
2253 for_each_online_cpu(cpu
) {
2254 smp_call_function_single(cpu
,
2255 kvm_arch_check_processor_compat
,
2261 on_each_cpu(hardware_enable
, NULL
, 1);
2262 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2265 register_reboot_notifier(&kvm_reboot_notifier
);
2267 r
= sysdev_class_register(&kvm_sysdev_class
);
2271 r
= sysdev_register(&kvm_sysdev
);
2275 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2276 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
,
2277 __alignof__(struct kvm_vcpu
),
2279 if (!kvm_vcpu_cache
) {
2284 kvm_chardev_ops
.owner
= module
;
2285 kvm_vm_fops
.owner
= module
;
2286 kvm_vcpu_fops
.owner
= module
;
2288 r
= misc_register(&kvm_dev
);
2290 printk(KERN_ERR
"kvm: misc device register failed\n");
2294 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2295 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2303 kmem_cache_destroy(kvm_vcpu_cache
);
2305 sysdev_unregister(&kvm_sysdev
);
2307 sysdev_class_unregister(&kvm_sysdev_class
);
2309 unregister_reboot_notifier(&kvm_reboot_notifier
);
2310 unregister_cpu_notifier(&kvm_cpu_notifier
);
2312 on_each_cpu(hardware_disable
, NULL
, 1);
2314 kvm_arch_hardware_unsetup();
2316 free_cpumask_var(cpus_hardware_enabled
);
2318 __free_page(bad_page
);
2325 EXPORT_SYMBOL_GPL(kvm_init
);
2329 kvm_trace_cleanup();
2330 misc_deregister(&kvm_dev
);
2331 kmem_cache_destroy(kvm_vcpu_cache
);
2332 sysdev_unregister(&kvm_sysdev
);
2333 sysdev_class_unregister(&kvm_sysdev_class
);
2334 unregister_reboot_notifier(&kvm_reboot_notifier
);
2335 unregister_cpu_notifier(&kvm_cpu_notifier
);
2336 on_each_cpu(hardware_disable
, NULL
, 1);
2337 kvm_arch_hardware_unsetup();
2340 free_cpumask_var(cpus_hardware_enabled
);
2341 __free_page(bad_page
);
2343 EXPORT_SYMBOL_GPL(kvm_exit
);