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 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
534 static int kvm_vm_ioctl_deassign_device(struct kvm
*kvm
,
535 struct kvm_assigned_pci_dev
*assigned_dev
)
538 struct kvm_assigned_dev_kernel
*match
;
540 mutex_lock(&kvm
->lock
);
542 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
543 assigned_dev
->assigned_dev_id
);
545 printk(KERN_INFO
"%s: device hasn't been assigned before, "
546 "so cannot be deassigned\n", __func__
);
551 if (assigned_dev
->flags
& KVM_DEV_ASSIGN_ENABLE_IOMMU
)
552 kvm_deassign_device(kvm
, match
);
554 kvm_free_assigned_device(kvm
, match
);
557 mutex_unlock(&kvm
->lock
);
562 static inline int valid_vcpu(int n
)
564 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
567 inline int kvm_is_mmio_pfn(pfn_t pfn
)
570 return PageReserved(pfn_to_page(pfn
));
576 * Switches to specified vcpu, until a matching vcpu_put()
578 void vcpu_load(struct kvm_vcpu
*vcpu
)
582 mutex_lock(&vcpu
->mutex
);
584 preempt_notifier_register(&vcpu
->preempt_notifier
);
585 kvm_arch_vcpu_load(vcpu
, cpu
);
589 void vcpu_put(struct kvm_vcpu
*vcpu
)
592 kvm_arch_vcpu_put(vcpu
);
593 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
595 mutex_unlock(&vcpu
->mutex
);
598 static void ack_flush(void *_completed
)
602 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
607 struct kvm_vcpu
*vcpu
;
609 if (alloc_cpumask_var(&cpus
, GFP_ATOMIC
))
613 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
614 vcpu
= kvm
->vcpus
[i
];
617 if (test_and_set_bit(req
, &vcpu
->requests
))
620 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
)
621 cpumask_set_cpu(cpu
, cpus
);
623 if (unlikely(cpus
== NULL
))
624 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
625 else if (!cpumask_empty(cpus
))
626 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
630 free_cpumask_var(cpus
);
634 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
636 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
637 ++kvm
->stat
.remote_tlb_flush
;
640 void kvm_reload_remote_mmus(struct kvm
*kvm
)
642 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
645 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
650 mutex_init(&vcpu
->mutex
);
654 init_waitqueue_head(&vcpu
->wq
);
656 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
661 vcpu
->run
= page_address(page
);
663 r
= kvm_arch_vcpu_init(vcpu
);
669 free_page((unsigned long)vcpu
->run
);
673 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
675 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
677 kvm_arch_vcpu_uninit(vcpu
);
678 free_page((unsigned long)vcpu
->run
);
680 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
682 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
683 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
685 return container_of(mn
, struct kvm
, mmu_notifier
);
688 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
689 struct mm_struct
*mm
,
690 unsigned long address
)
692 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
696 * When ->invalidate_page runs, the linux pte has been zapped
697 * already but the page is still allocated until
698 * ->invalidate_page returns. So if we increase the sequence
699 * here the kvm page fault will notice if the spte can't be
700 * established because the page is going to be freed. If
701 * instead the kvm page fault establishes the spte before
702 * ->invalidate_page runs, kvm_unmap_hva will release it
705 * The sequence increase only need to be seen at spin_unlock
706 * time, and not at spin_lock time.
708 * Increasing the sequence after the spin_unlock would be
709 * unsafe because the kvm page fault could then establish the
710 * pte after kvm_unmap_hva returned, without noticing the page
711 * is going to be freed.
713 spin_lock(&kvm
->mmu_lock
);
714 kvm
->mmu_notifier_seq
++;
715 need_tlb_flush
= kvm_unmap_hva(kvm
, address
);
716 spin_unlock(&kvm
->mmu_lock
);
718 /* we've to flush the tlb before the pages can be freed */
720 kvm_flush_remote_tlbs(kvm
);
724 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
725 struct mm_struct
*mm
,
729 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
730 int need_tlb_flush
= 0;
732 spin_lock(&kvm
->mmu_lock
);
734 * The count increase must become visible at unlock time as no
735 * spte can be established without taking the mmu_lock and
736 * count is also read inside the mmu_lock critical section.
738 kvm
->mmu_notifier_count
++;
739 for (; start
< end
; start
+= PAGE_SIZE
)
740 need_tlb_flush
|= kvm_unmap_hva(kvm
, start
);
741 spin_unlock(&kvm
->mmu_lock
);
743 /* we've to flush the tlb before the pages can be freed */
745 kvm_flush_remote_tlbs(kvm
);
748 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
749 struct mm_struct
*mm
,
753 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
755 spin_lock(&kvm
->mmu_lock
);
757 * This sequence increase will notify the kvm page fault that
758 * the page that is going to be mapped in the spte could have
761 kvm
->mmu_notifier_seq
++;
763 * The above sequence increase must be visible before the
764 * below count decrease but both values are read by the kvm
765 * page fault under mmu_lock spinlock so we don't need to add
766 * a smb_wmb() here in between the two.
768 kvm
->mmu_notifier_count
--;
769 spin_unlock(&kvm
->mmu_lock
);
771 BUG_ON(kvm
->mmu_notifier_count
< 0);
774 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
775 struct mm_struct
*mm
,
776 unsigned long address
)
778 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
781 spin_lock(&kvm
->mmu_lock
);
782 young
= kvm_age_hva(kvm
, address
);
783 spin_unlock(&kvm
->mmu_lock
);
786 kvm_flush_remote_tlbs(kvm
);
791 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
792 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
793 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
794 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
795 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
797 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
799 static struct kvm
*kvm_create_vm(void)
801 struct kvm
*kvm
= kvm_arch_create_vm();
802 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
809 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
810 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
813 return ERR_PTR(-ENOMEM
);
815 kvm
->coalesced_mmio_ring
=
816 (struct kvm_coalesced_mmio_ring
*)page_address(page
);
819 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
822 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
823 err
= mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
825 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
834 kvm
->mm
= current
->mm
;
835 atomic_inc(&kvm
->mm
->mm_count
);
836 spin_lock_init(&kvm
->mmu_lock
);
837 kvm_io_bus_init(&kvm
->pio_bus
);
838 mutex_init(&kvm
->lock
);
839 kvm_io_bus_init(&kvm
->mmio_bus
);
840 init_rwsem(&kvm
->slots_lock
);
841 atomic_set(&kvm
->users_count
, 1);
842 spin_lock(&kvm_lock
);
843 list_add(&kvm
->vm_list
, &vm_list
);
844 spin_unlock(&kvm_lock
);
845 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
846 kvm_coalesced_mmio_init(kvm
);
853 * Free any memory in @free but not in @dont.
855 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
856 struct kvm_memory_slot
*dont
)
858 if (!dont
|| free
->rmap
!= dont
->rmap
)
861 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
862 vfree(free
->dirty_bitmap
);
864 if (!dont
|| free
->lpage_info
!= dont
->lpage_info
)
865 vfree(free
->lpage_info
);
868 free
->dirty_bitmap
= NULL
;
870 free
->lpage_info
= NULL
;
873 void kvm_free_physmem(struct kvm
*kvm
)
877 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
878 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
881 static void kvm_destroy_vm(struct kvm
*kvm
)
883 struct mm_struct
*mm
= kvm
->mm
;
885 spin_lock(&kvm_lock
);
886 list_del(&kvm
->vm_list
);
887 spin_unlock(&kvm_lock
);
888 kvm_io_bus_destroy(&kvm
->pio_bus
);
889 kvm_io_bus_destroy(&kvm
->mmio_bus
);
890 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
891 if (kvm
->coalesced_mmio_ring
!= NULL
)
892 free_page((unsigned long)kvm
->coalesced_mmio_ring
);
894 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
895 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
897 kvm_arch_destroy_vm(kvm
);
901 void kvm_get_kvm(struct kvm
*kvm
)
903 atomic_inc(&kvm
->users_count
);
905 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
907 void kvm_put_kvm(struct kvm
*kvm
)
909 if (atomic_dec_and_test(&kvm
->users_count
))
912 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
915 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
917 struct kvm
*kvm
= filp
->private_data
;
924 * Allocate some memory and give it an address in the guest physical address
927 * Discontiguous memory is allowed, mostly for framebuffers.
929 * Must be called holding mmap_sem for write.
931 int __kvm_set_memory_region(struct kvm
*kvm
,
932 struct kvm_userspace_memory_region
*mem
,
937 unsigned long npages
;
939 struct kvm_memory_slot
*memslot
;
940 struct kvm_memory_slot old
, new;
943 /* General sanity checks */
944 if (mem
->memory_size
& (PAGE_SIZE
- 1))
946 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
948 if (user_alloc
&& (mem
->userspace_addr
& (PAGE_SIZE
- 1)))
950 if (mem
->slot
>= KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
)
952 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
955 memslot
= &kvm
->memslots
[mem
->slot
];
956 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
957 npages
= mem
->memory_size
>> PAGE_SHIFT
;
960 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
962 new = old
= *memslot
;
964 new.base_gfn
= base_gfn
;
966 new.flags
= mem
->flags
;
968 /* Disallow changing a memory slot's size. */
970 if (npages
&& old
.npages
&& npages
!= old
.npages
)
973 /* Check for overlaps */
975 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
976 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
980 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
981 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
985 /* Free page dirty bitmap if unneeded */
986 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
987 new.dirty_bitmap
= NULL
;
991 /* Allocate if a slot is being created */
993 if (npages
&& !new.rmap
) {
994 new.rmap
= vmalloc(npages
* sizeof(struct page
*));
999 memset(new.rmap
, 0, npages
* sizeof(*new.rmap
));
1001 new.user_alloc
= user_alloc
;
1003 * hva_to_rmmap() serialzies with the mmu_lock and to be
1004 * safe it has to ignore memslots with !user_alloc &&
1008 new.userspace_addr
= mem
->userspace_addr
;
1010 new.userspace_addr
= 0;
1012 if (npages
&& !new.lpage_info
) {
1013 int largepages
= npages
/ KVM_PAGES_PER_HPAGE
;
1014 if (npages
% KVM_PAGES_PER_HPAGE
)
1016 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
1019 new.lpage_info
= vmalloc(largepages
* sizeof(*new.lpage_info
));
1021 if (!new.lpage_info
)
1024 memset(new.lpage_info
, 0, largepages
* sizeof(*new.lpage_info
));
1026 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
1027 new.lpage_info
[0].write_count
= 1;
1028 if ((base_gfn
+npages
) % KVM_PAGES_PER_HPAGE
)
1029 new.lpage_info
[largepages
-1].write_count
= 1;
1032 /* Allocate page dirty bitmap if needed */
1033 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1034 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
1036 new.dirty_bitmap
= vmalloc(dirty_bytes
);
1037 if (!new.dirty_bitmap
)
1039 memset(new.dirty_bitmap
, 0, dirty_bytes
);
1041 #endif /* not defined CONFIG_S390 */
1044 kvm_arch_flush_shadow(kvm
);
1046 spin_lock(&kvm
->mmu_lock
);
1047 if (mem
->slot
>= kvm
->nmemslots
)
1048 kvm
->nmemslots
= mem
->slot
+ 1;
1051 spin_unlock(&kvm
->mmu_lock
);
1053 r
= kvm_arch_set_memory_region(kvm
, mem
, old
, user_alloc
);
1055 spin_lock(&kvm
->mmu_lock
);
1057 spin_unlock(&kvm
->mmu_lock
);
1061 kvm_free_physmem_slot(&old
, npages
? &new : NULL
);
1062 /* Slot deletion case: we have to update the current slot */
1066 /* map the pages in iommu page table */
1067 r
= kvm_iommu_map_pages(kvm
, base_gfn
, npages
);
1074 kvm_free_physmem_slot(&new, &old
);
1079 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1081 int kvm_set_memory_region(struct kvm
*kvm
,
1082 struct kvm_userspace_memory_region
*mem
,
1087 down_write(&kvm
->slots_lock
);
1088 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
1089 up_write(&kvm
->slots_lock
);
1092 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1094 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1096 kvm_userspace_memory_region
*mem
,
1099 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
1101 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
1104 int kvm_get_dirty_log(struct kvm
*kvm
,
1105 struct kvm_dirty_log
*log
, int *is_dirty
)
1107 struct kvm_memory_slot
*memslot
;
1110 unsigned long any
= 0;
1113 if (log
->slot
>= KVM_MEMORY_SLOTS
)
1116 memslot
= &kvm
->memslots
[log
->slot
];
1118 if (!memslot
->dirty_bitmap
)
1121 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1123 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1124 any
= memslot
->dirty_bitmap
[i
];
1127 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1138 int is_error_page(struct page
*page
)
1140 return page
== bad_page
;
1142 EXPORT_SYMBOL_GPL(is_error_page
);
1144 int is_error_pfn(pfn_t pfn
)
1146 return pfn
== bad_pfn
;
1148 EXPORT_SYMBOL_GPL(is_error_pfn
);
1150 static inline unsigned long bad_hva(void)
1155 int kvm_is_error_hva(unsigned long addr
)
1157 return addr
== bad_hva();
1159 EXPORT_SYMBOL_GPL(kvm_is_error_hva
);
1161 struct kvm_memory_slot
*gfn_to_memslot_unaliased(struct kvm
*kvm
, gfn_t gfn
)
1165 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
1166 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
1168 if (gfn
>= memslot
->base_gfn
1169 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
1174 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased
);
1176 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1178 gfn
= unalias_gfn(kvm
, gfn
);
1179 return gfn_to_memslot_unaliased(kvm
, gfn
);
1182 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1186 gfn
= unalias_gfn(kvm
, gfn
);
1187 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
1188 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
1190 if (gfn
>= memslot
->base_gfn
1191 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
1196 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1198 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1200 struct kvm_memory_slot
*slot
;
1202 gfn
= unalias_gfn(kvm
, gfn
);
1203 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
1206 return (slot
->userspace_addr
+ (gfn
- slot
->base_gfn
) * PAGE_SIZE
);
1208 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1210 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1212 struct page
*page
[1];
1219 addr
= gfn_to_hva(kvm
, gfn
);
1220 if (kvm_is_error_hva(addr
)) {
1222 return page_to_pfn(bad_page
);
1225 npages
= get_user_pages_fast(addr
, 1, 1, page
);
1227 if (unlikely(npages
!= 1)) {
1228 struct vm_area_struct
*vma
;
1230 down_read(¤t
->mm
->mmap_sem
);
1231 vma
= find_vma(current
->mm
, addr
);
1233 if (vma
== NULL
|| addr
< vma
->vm_start
||
1234 !(vma
->vm_flags
& VM_PFNMAP
)) {
1235 up_read(¤t
->mm
->mmap_sem
);
1237 return page_to_pfn(bad_page
);
1240 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1241 up_read(¤t
->mm
->mmap_sem
);
1242 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1244 pfn
= page_to_pfn(page
[0]);
1249 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1251 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1255 pfn
= gfn_to_pfn(kvm
, gfn
);
1256 if (!kvm_is_mmio_pfn(pfn
))
1257 return pfn_to_page(pfn
);
1259 WARN_ON(kvm_is_mmio_pfn(pfn
));
1265 EXPORT_SYMBOL_GPL(gfn_to_page
);
1267 void kvm_release_page_clean(struct page
*page
)
1269 kvm_release_pfn_clean(page_to_pfn(page
));
1271 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1273 void kvm_release_pfn_clean(pfn_t pfn
)
1275 if (!kvm_is_mmio_pfn(pfn
))
1276 put_page(pfn_to_page(pfn
));
1278 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1280 void kvm_release_page_dirty(struct page
*page
)
1282 kvm_release_pfn_dirty(page_to_pfn(page
));
1284 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1286 void kvm_release_pfn_dirty(pfn_t pfn
)
1288 kvm_set_pfn_dirty(pfn
);
1289 kvm_release_pfn_clean(pfn
);
1291 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1293 void kvm_set_page_dirty(struct page
*page
)
1295 kvm_set_pfn_dirty(page_to_pfn(page
));
1297 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1299 void kvm_set_pfn_dirty(pfn_t pfn
)
1301 if (!kvm_is_mmio_pfn(pfn
)) {
1302 struct page
*page
= pfn_to_page(pfn
);
1303 if (!PageReserved(page
))
1307 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1309 void kvm_set_pfn_accessed(pfn_t pfn
)
1311 if (!kvm_is_mmio_pfn(pfn
))
1312 mark_page_accessed(pfn_to_page(pfn
));
1314 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1316 void kvm_get_pfn(pfn_t pfn
)
1318 if (!kvm_is_mmio_pfn(pfn
))
1319 get_page(pfn_to_page(pfn
));
1321 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1323 static int next_segment(unsigned long len
, int offset
)
1325 if (len
> PAGE_SIZE
- offset
)
1326 return PAGE_SIZE
- offset
;
1331 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1337 addr
= gfn_to_hva(kvm
, gfn
);
1338 if (kvm_is_error_hva(addr
))
1340 r
= copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1345 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1347 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1349 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1351 int offset
= offset_in_page(gpa
);
1354 while ((seg
= next_segment(len
, offset
)) != 0) {
1355 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1365 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1367 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1372 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1373 int offset
= offset_in_page(gpa
);
1375 addr
= gfn_to_hva(kvm
, gfn
);
1376 if (kvm_is_error_hva(addr
))
1378 pagefault_disable();
1379 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1385 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1387 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1388 int offset
, int len
)
1393 addr
= gfn_to_hva(kvm
, gfn
);
1394 if (kvm_is_error_hva(addr
))
1396 r
= copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1399 mark_page_dirty(kvm
, gfn
);
1402 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1404 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1407 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1409 int offset
= offset_in_page(gpa
);
1412 while ((seg
= next_segment(len
, offset
)) != 0) {
1413 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1424 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1426 return kvm_write_guest_page(kvm
, gfn
, empty_zero_page
, offset
, len
);
1428 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1430 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1432 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1434 int offset
= offset_in_page(gpa
);
1437 while ((seg
= next_segment(len
, offset
)) != 0) {
1438 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1447 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1449 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1451 struct kvm_memory_slot
*memslot
;
1453 gfn
= unalias_gfn(kvm
, gfn
);
1454 memslot
= gfn_to_memslot_unaliased(kvm
, gfn
);
1455 if (memslot
&& memslot
->dirty_bitmap
) {
1456 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1459 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
1460 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
1465 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1467 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1472 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1474 if (kvm_cpu_has_interrupt(vcpu
) ||
1475 kvm_cpu_has_pending_timer(vcpu
) ||
1476 kvm_arch_vcpu_runnable(vcpu
)) {
1477 set_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
1480 if (signal_pending(current
))
1488 finish_wait(&vcpu
->wq
, &wait
);
1491 void kvm_resched(struct kvm_vcpu
*vcpu
)
1493 if (!need_resched())
1497 EXPORT_SYMBOL_GPL(kvm_resched
);
1499 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1501 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1504 if (vmf
->pgoff
== 0)
1505 page
= virt_to_page(vcpu
->run
);
1507 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1508 page
= virt_to_page(vcpu
->arch
.pio_data
);
1510 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1511 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1512 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1515 return VM_FAULT_SIGBUS
;
1521 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
1522 .fault
= kvm_vcpu_fault
,
1525 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1527 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1531 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1533 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1535 kvm_put_kvm(vcpu
->kvm
);
1539 static struct file_operations kvm_vcpu_fops
= {
1540 .release
= kvm_vcpu_release
,
1541 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1542 .compat_ioctl
= kvm_vcpu_ioctl
,
1543 .mmap
= kvm_vcpu_mmap
,
1547 * Allocates an inode for the vcpu.
1549 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1551 int fd
= anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, 0);
1553 kvm_put_kvm(vcpu
->kvm
);
1558 * Creates some virtual cpus. Good luck creating more than one.
1560 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
1563 struct kvm_vcpu
*vcpu
;
1568 vcpu
= kvm_arch_vcpu_create(kvm
, n
);
1570 return PTR_ERR(vcpu
);
1572 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1574 r
= kvm_arch_vcpu_setup(vcpu
);
1578 mutex_lock(&kvm
->lock
);
1579 if (kvm
->vcpus
[n
]) {
1583 kvm
->vcpus
[n
] = vcpu
;
1584 mutex_unlock(&kvm
->lock
);
1586 /* Now it's all set up, let userspace reach it */
1588 r
= create_vcpu_fd(vcpu
);
1594 mutex_lock(&kvm
->lock
);
1595 kvm
->vcpus
[n
] = NULL
;
1597 mutex_unlock(&kvm
->lock
);
1598 kvm_arch_vcpu_destroy(vcpu
);
1602 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1605 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1606 vcpu
->sigset_active
= 1;
1607 vcpu
->sigset
= *sigset
;
1609 vcpu
->sigset_active
= 0;
1613 static long kvm_vcpu_ioctl(struct file
*filp
,
1614 unsigned int ioctl
, unsigned long arg
)
1616 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1617 void __user
*argp
= (void __user
*)arg
;
1619 struct kvm_fpu
*fpu
= NULL
;
1620 struct kvm_sregs
*kvm_sregs
= NULL
;
1622 if (vcpu
->kvm
->mm
!= current
->mm
)
1629 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1631 case KVM_GET_REGS
: {
1632 struct kvm_regs
*kvm_regs
;
1635 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1638 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1642 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1649 case KVM_SET_REGS
: {
1650 struct kvm_regs
*kvm_regs
;
1653 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1657 if (copy_from_user(kvm_regs
, argp
, sizeof(struct kvm_regs
)))
1659 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1667 case KVM_GET_SREGS
: {
1668 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1672 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1676 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
1681 case KVM_SET_SREGS
: {
1682 kvm_sregs
= kmalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1687 if (copy_from_user(kvm_sregs
, argp
, sizeof(struct kvm_sregs
)))
1689 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
1695 case KVM_GET_MP_STATE
: {
1696 struct kvm_mp_state mp_state
;
1698 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1702 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
1707 case KVM_SET_MP_STATE
: {
1708 struct kvm_mp_state mp_state
;
1711 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
1713 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
1719 case KVM_TRANSLATE
: {
1720 struct kvm_translation tr
;
1723 if (copy_from_user(&tr
, argp
, sizeof tr
))
1725 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
1729 if (copy_to_user(argp
, &tr
, sizeof tr
))
1734 case KVM_DEBUG_GUEST
: {
1735 struct kvm_debug_guest dbg
;
1738 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
1740 r
= kvm_arch_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
1746 case KVM_SET_SIGNAL_MASK
: {
1747 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
1748 struct kvm_signal_mask kvm_sigmask
;
1749 sigset_t sigset
, *p
;
1754 if (copy_from_user(&kvm_sigmask
, argp
,
1755 sizeof kvm_sigmask
))
1758 if (kvm_sigmask
.len
!= sizeof sigset
)
1761 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
1766 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
1770 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
1774 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
1778 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
1784 fpu
= kmalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
1789 if (copy_from_user(fpu
, argp
, sizeof(struct kvm_fpu
)))
1791 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
1798 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1806 static long kvm_vm_ioctl(struct file
*filp
,
1807 unsigned int ioctl
, unsigned long arg
)
1809 struct kvm
*kvm
= filp
->private_data
;
1810 void __user
*argp
= (void __user
*)arg
;
1813 if (kvm
->mm
!= current
->mm
)
1816 case KVM_CREATE_VCPU
:
1817 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
1821 case KVM_SET_USER_MEMORY_REGION
: {
1822 struct kvm_userspace_memory_region kvm_userspace_mem
;
1825 if (copy_from_user(&kvm_userspace_mem
, argp
,
1826 sizeof kvm_userspace_mem
))
1829 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 1);
1834 case KVM_GET_DIRTY_LOG
: {
1835 struct kvm_dirty_log log
;
1838 if (copy_from_user(&log
, argp
, sizeof log
))
1840 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
1845 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1846 case KVM_REGISTER_COALESCED_MMIO
: {
1847 struct kvm_coalesced_mmio_zone zone
;
1849 if (copy_from_user(&zone
, argp
, sizeof zone
))
1852 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
1858 case KVM_UNREGISTER_COALESCED_MMIO
: {
1859 struct kvm_coalesced_mmio_zone zone
;
1861 if (copy_from_user(&zone
, argp
, sizeof zone
))
1864 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
1871 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1872 case KVM_ASSIGN_PCI_DEVICE
: {
1873 struct kvm_assigned_pci_dev assigned_dev
;
1876 if (copy_from_user(&assigned_dev
, argp
, sizeof assigned_dev
))
1878 r
= kvm_vm_ioctl_assign_device(kvm
, &assigned_dev
);
1883 case KVM_ASSIGN_IRQ
: {
1884 struct kvm_assigned_irq assigned_irq
;
1887 if (copy_from_user(&assigned_irq
, argp
, sizeof assigned_irq
))
1889 r
= kvm_vm_ioctl_assign_irq(kvm
, &assigned_irq
);
1895 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1896 case KVM_DEASSIGN_PCI_DEVICE
: {
1897 struct kvm_assigned_pci_dev assigned_dev
;
1900 if (copy_from_user(&assigned_dev
, argp
, sizeof assigned_dev
))
1902 r
= kvm_vm_ioctl_deassign_device(kvm
, &assigned_dev
);
1909 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
1915 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1917 struct page
*page
[1];
1920 gfn_t gfn
= vmf
->pgoff
;
1921 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1923 addr
= gfn_to_hva(kvm
, gfn
);
1924 if (kvm_is_error_hva(addr
))
1925 return VM_FAULT_SIGBUS
;
1927 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
1929 if (unlikely(npages
!= 1))
1930 return VM_FAULT_SIGBUS
;
1932 vmf
->page
= page
[0];
1936 static struct vm_operations_struct kvm_vm_vm_ops
= {
1937 .fault
= kvm_vm_fault
,
1940 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1942 vma
->vm_ops
= &kvm_vm_vm_ops
;
1946 static struct file_operations kvm_vm_fops
= {
1947 .release
= kvm_vm_release
,
1948 .unlocked_ioctl
= kvm_vm_ioctl
,
1949 .compat_ioctl
= kvm_vm_ioctl
,
1950 .mmap
= kvm_vm_mmap
,
1953 static int kvm_dev_ioctl_create_vm(void)
1958 kvm
= kvm_create_vm();
1960 return PTR_ERR(kvm
);
1961 fd
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, 0);
1968 static long kvm_dev_ioctl_check_extension_generic(long arg
)
1971 case KVM_CAP_USER_MEMORY
:
1972 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
1977 return kvm_dev_ioctl_check_extension(arg
);
1980 static long kvm_dev_ioctl(struct file
*filp
,
1981 unsigned int ioctl
, unsigned long arg
)
1986 case KVM_GET_API_VERSION
:
1990 r
= KVM_API_VERSION
;
1996 r
= kvm_dev_ioctl_create_vm();
1998 case KVM_CHECK_EXTENSION
:
1999 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2001 case KVM_GET_VCPU_MMAP_SIZE
:
2005 r
= PAGE_SIZE
; /* struct kvm_run */
2007 r
+= PAGE_SIZE
; /* pio data page */
2009 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2010 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2013 case KVM_TRACE_ENABLE
:
2014 case KVM_TRACE_PAUSE
:
2015 case KVM_TRACE_DISABLE
:
2016 r
= kvm_trace_ioctl(ioctl
, arg
);
2019 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2025 static struct file_operations kvm_chardev_ops
= {
2026 .unlocked_ioctl
= kvm_dev_ioctl
,
2027 .compat_ioctl
= kvm_dev_ioctl
,
2030 static struct miscdevice kvm_dev
= {
2036 static void hardware_enable(void *junk
)
2038 int cpu
= raw_smp_processor_id();
2040 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2042 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2043 kvm_arch_hardware_enable(NULL
);
2046 static void hardware_disable(void *junk
)
2048 int cpu
= raw_smp_processor_id();
2050 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2052 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2053 kvm_arch_hardware_disable(NULL
);
2056 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2061 val
&= ~CPU_TASKS_FROZEN
;
2064 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2066 hardware_disable(NULL
);
2068 case CPU_UP_CANCELED
:
2069 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2071 smp_call_function_single(cpu
, hardware_disable
, NULL
, 1);
2074 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2076 smp_call_function_single(cpu
, hardware_enable
, NULL
, 1);
2083 asmlinkage
void kvm_handle_fault_on_reboot(void)
2086 /* spin while reset goes on */
2089 /* Fault while not rebooting. We want the trace. */
2092 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot
);
2094 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2097 if (val
== SYS_RESTART
) {
2099 * Some (well, at least mine) BIOSes hang on reboot if
2102 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2103 kvm_rebooting
= true;
2104 on_each_cpu(hardware_disable
, NULL
, 1);
2109 static struct notifier_block kvm_reboot_notifier
= {
2110 .notifier_call
= kvm_reboot
,
2114 void kvm_io_bus_init(struct kvm_io_bus
*bus
)
2116 memset(bus
, 0, sizeof(*bus
));
2119 void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2123 for (i
= 0; i
< bus
->dev_count
; i
++) {
2124 struct kvm_io_device
*pos
= bus
->devs
[i
];
2126 kvm_iodevice_destructor(pos
);
2130 struct kvm_io_device
*kvm_io_bus_find_dev(struct kvm_io_bus
*bus
,
2131 gpa_t addr
, int len
, int is_write
)
2135 for (i
= 0; i
< bus
->dev_count
; i
++) {
2136 struct kvm_io_device
*pos
= bus
->devs
[i
];
2138 if (pos
->in_range(pos
, addr
, len
, is_write
))
2145 void kvm_io_bus_register_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
)
2147 BUG_ON(bus
->dev_count
> (NR_IOBUS_DEVS
-1));
2149 bus
->devs
[bus
->dev_count
++] = dev
;
2152 static struct notifier_block kvm_cpu_notifier
= {
2153 .notifier_call
= kvm_cpu_hotplug
,
2154 .priority
= 20, /* must be > scheduler priority */
2157 static int vm_stat_get(void *_offset
, u64
*val
)
2159 unsigned offset
= (long)_offset
;
2163 spin_lock(&kvm_lock
);
2164 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2165 *val
+= *(u32
*)((void *)kvm
+ offset
);
2166 spin_unlock(&kvm_lock
);
2170 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2172 static int vcpu_stat_get(void *_offset
, u64
*val
)
2174 unsigned offset
= (long)_offset
;
2176 struct kvm_vcpu
*vcpu
;
2180 spin_lock(&kvm_lock
);
2181 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2182 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2183 vcpu
= kvm
->vcpus
[i
];
2185 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2187 spin_unlock(&kvm_lock
);
2191 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2193 static struct file_operations
*stat_fops
[] = {
2194 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2195 [KVM_STAT_VM
] = &vm_stat_fops
,
2198 static void kvm_init_debug(void)
2200 struct kvm_stats_debugfs_item
*p
;
2202 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2203 for (p
= debugfs_entries
; p
->name
; ++p
)
2204 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2205 (void *)(long)p
->offset
,
2206 stat_fops
[p
->kind
]);
2209 static void kvm_exit_debug(void)
2211 struct kvm_stats_debugfs_item
*p
;
2213 for (p
= debugfs_entries
; p
->name
; ++p
)
2214 debugfs_remove(p
->dentry
);
2215 debugfs_remove(kvm_debugfs_dir
);
2218 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2220 hardware_disable(NULL
);
2224 static int kvm_resume(struct sys_device
*dev
)
2226 hardware_enable(NULL
);
2230 static struct sysdev_class kvm_sysdev_class
= {
2232 .suspend
= kvm_suspend
,
2233 .resume
= kvm_resume
,
2236 static struct sys_device kvm_sysdev
= {
2238 .cls
= &kvm_sysdev_class
,
2241 struct page
*bad_page
;
2245 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2247 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2250 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2252 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2254 kvm_arch_vcpu_load(vcpu
, cpu
);
2257 static void kvm_sched_out(struct preempt_notifier
*pn
,
2258 struct task_struct
*next
)
2260 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2262 kvm_arch_vcpu_put(vcpu
);
2265 int kvm_init(void *opaque
, unsigned int vcpu_size
,
2266 struct module
*module
)
2273 r
= kvm_arch_init(opaque
);
2277 bad_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2279 if (bad_page
== NULL
) {
2284 bad_pfn
= page_to_pfn(bad_page
);
2286 if (!alloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2291 r
= kvm_arch_hardware_setup();
2295 for_each_online_cpu(cpu
) {
2296 smp_call_function_single(cpu
,
2297 kvm_arch_check_processor_compat
,
2303 on_each_cpu(hardware_enable
, NULL
, 1);
2304 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2307 register_reboot_notifier(&kvm_reboot_notifier
);
2309 r
= sysdev_class_register(&kvm_sysdev_class
);
2313 r
= sysdev_register(&kvm_sysdev
);
2317 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2318 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
,
2319 __alignof__(struct kvm_vcpu
),
2321 if (!kvm_vcpu_cache
) {
2326 kvm_chardev_ops
.owner
= module
;
2327 kvm_vm_fops
.owner
= module
;
2328 kvm_vcpu_fops
.owner
= module
;
2330 r
= misc_register(&kvm_dev
);
2332 printk(KERN_ERR
"kvm: misc device register failed\n");
2336 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2337 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2345 kmem_cache_destroy(kvm_vcpu_cache
);
2347 sysdev_unregister(&kvm_sysdev
);
2349 sysdev_class_unregister(&kvm_sysdev_class
);
2351 unregister_reboot_notifier(&kvm_reboot_notifier
);
2352 unregister_cpu_notifier(&kvm_cpu_notifier
);
2354 on_each_cpu(hardware_disable
, NULL
, 1);
2356 kvm_arch_hardware_unsetup();
2358 free_cpumask_var(cpus_hardware_enabled
);
2360 __free_page(bad_page
);
2367 EXPORT_SYMBOL_GPL(kvm_init
);
2371 kvm_trace_cleanup();
2372 misc_deregister(&kvm_dev
);
2373 kmem_cache_destroy(kvm_vcpu_cache
);
2374 sysdev_unregister(&kvm_sysdev
);
2375 sysdev_class_unregister(&kvm_sysdev_class
);
2376 unregister_reboot_notifier(&kvm_reboot_notifier
);
2377 unregister_cpu_notifier(&kvm_cpu_notifier
);
2378 on_each_cpu(hardware_disable
, NULL
, 1);
2379 kvm_arch_hardware_unsetup();
2382 free_cpumask_var(cpus_hardware_enabled
);
2383 __free_page(bad_page
);
2385 EXPORT_SYMBOL_GPL(kvm_exit
);