2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
47 #include <asm/processor.h>
49 #include <asm/uaccess.h>
50 #include <asm/pgtable.h>
52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53 #include "coalesced_mmio.h"
56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
57 #include <linux/pci.h>
58 #include <linux/interrupt.h>
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
65 DEFINE_SPINLOCK(kvm_lock
);
68 static cpumask_var_t cpus_hardware_enabled
;
70 struct kmem_cache
*kvm_vcpu_cache
;
71 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
73 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
75 struct dentry
*kvm_debugfs_dir
;
77 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
80 static bool kvm_rebooting
;
82 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
83 static struct kvm_assigned_dev_kernel
*kvm_find_assigned_dev(struct list_head
*head
,
86 struct list_head
*ptr
;
87 struct kvm_assigned_dev_kernel
*match
;
89 list_for_each(ptr
, head
) {
90 match
= list_entry(ptr
, struct kvm_assigned_dev_kernel
, list
);
91 if (match
->assigned_dev_id
== assigned_dev_id
)
97 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
98 *assigned_dev
, int irq
)
101 struct msix_entry
*host_msix_entries
;
103 host_msix_entries
= assigned_dev
->host_msix_entries
;
106 for (i
= 0; i
< assigned_dev
->entries_nr
; i
++)
107 if (irq
== host_msix_entries
[i
].vector
) {
112 printk(KERN_WARNING
"Fail to find correlated MSI-X entry!\n");
119 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct
*work
)
121 struct kvm_assigned_dev_kernel
*assigned_dev
;
125 assigned_dev
= container_of(work
, struct kvm_assigned_dev_kernel
,
127 kvm
= assigned_dev
->kvm
;
129 /* This is taken to safely inject irq inside the guest. When
130 * the interrupt injection (or the ioapic code) uses a
131 * finer-grained lock, update this
133 mutex_lock(&kvm
->lock
);
134 spin_lock_irq(&assigned_dev
->assigned_dev_lock
);
135 if (assigned_dev
->irq_requested_type
& KVM_DEV_IRQ_HOST_MSIX
) {
136 struct kvm_guest_msix_entry
*guest_entries
=
137 assigned_dev
->guest_msix_entries
;
138 for (i
= 0; i
< assigned_dev
->entries_nr
; i
++) {
139 if (!(guest_entries
[i
].flags
&
140 KVM_ASSIGNED_MSIX_PENDING
))
142 guest_entries
[i
].flags
&= ~KVM_ASSIGNED_MSIX_PENDING
;
143 kvm_set_irq(assigned_dev
->kvm
,
144 assigned_dev
->irq_source_id
,
145 guest_entries
[i
].vector
, 1);
148 kvm_set_irq(assigned_dev
->kvm
, assigned_dev
->irq_source_id
,
149 assigned_dev
->guest_irq
, 1);
151 spin_unlock_irq(&assigned_dev
->assigned_dev_lock
);
152 mutex_unlock(&assigned_dev
->kvm
->lock
);
155 static irqreturn_t
kvm_assigned_dev_intr(int irq
, void *dev_id
)
158 struct kvm_assigned_dev_kernel
*assigned_dev
=
159 (struct kvm_assigned_dev_kernel
*) dev_id
;
161 spin_lock_irqsave(&assigned_dev
->assigned_dev_lock
, flags
);
162 if (assigned_dev
->irq_requested_type
& KVM_DEV_IRQ_HOST_MSIX
) {
163 int index
= find_index_from_host_irq(assigned_dev
, irq
);
166 assigned_dev
->guest_msix_entries
[index
].flags
|=
167 KVM_ASSIGNED_MSIX_PENDING
;
170 schedule_work(&assigned_dev
->interrupt_work
);
172 if (assigned_dev
->irq_requested_type
& KVM_DEV_IRQ_GUEST_INTX
) {
173 disable_irq_nosync(irq
);
174 assigned_dev
->host_irq_disabled
= true;
178 spin_unlock_irqrestore(&assigned_dev
->assigned_dev_lock
, flags
);
182 /* Ack the irq line for an assigned device */
183 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier
*kian
)
185 struct kvm_assigned_dev_kernel
*dev
;
191 dev
= container_of(kian
, struct kvm_assigned_dev_kernel
,
194 kvm_set_irq(dev
->kvm
, dev
->irq_source_id
, dev
->guest_irq
, 0);
196 /* The guest irq may be shared so this ack may be
197 * from another device.
199 spin_lock_irqsave(&dev
->assigned_dev_lock
, flags
);
200 if (dev
->host_irq_disabled
) {
201 enable_irq(dev
->host_irq
);
202 dev
->host_irq_disabled
= false;
204 spin_unlock_irqrestore(&dev
->assigned_dev_lock
, flags
);
207 static void deassign_guest_irq(struct kvm
*kvm
,
208 struct kvm_assigned_dev_kernel
*assigned_dev
)
210 kvm_unregister_irq_ack_notifier(&assigned_dev
->ack_notifier
);
211 assigned_dev
->ack_notifier
.gsi
= -1;
213 if (assigned_dev
->irq_source_id
!= -1)
214 kvm_free_irq_source_id(kvm
, assigned_dev
->irq_source_id
);
215 assigned_dev
->irq_source_id
= -1;
216 assigned_dev
->irq_requested_type
&= ~(KVM_DEV_IRQ_GUEST_MASK
);
219 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
220 static void deassign_host_irq(struct kvm
*kvm
,
221 struct kvm_assigned_dev_kernel
*assigned_dev
)
224 * In kvm_free_device_irq, cancel_work_sync return true if:
225 * 1. work is scheduled, and then cancelled.
226 * 2. work callback is executed.
228 * The first one ensured that the irq is disabled and no more events
229 * would happen. But for the second one, the irq may be enabled (e.g.
230 * for MSI). So we disable irq here to prevent further events.
232 * Notice this maybe result in nested disable if the interrupt type is
233 * INTx, but it's OK for we are going to free it.
235 * If this function is a part of VM destroy, please ensure that till
236 * now, the kvm state is still legal for probably we also have to wait
237 * interrupt_work done.
239 if (assigned_dev
->irq_requested_type
& KVM_DEV_IRQ_HOST_MSIX
) {
241 for (i
= 0; i
< assigned_dev
->entries_nr
; i
++)
242 disable_irq_nosync(assigned_dev
->
243 host_msix_entries
[i
].vector
);
245 cancel_work_sync(&assigned_dev
->interrupt_work
);
247 for (i
= 0; i
< assigned_dev
->entries_nr
; i
++)
248 free_irq(assigned_dev
->host_msix_entries
[i
].vector
,
249 (void *)assigned_dev
);
251 assigned_dev
->entries_nr
= 0;
252 kfree(assigned_dev
->host_msix_entries
);
253 kfree(assigned_dev
->guest_msix_entries
);
254 pci_disable_msix(assigned_dev
->dev
);
256 /* Deal with MSI and INTx */
257 disable_irq_nosync(assigned_dev
->host_irq
);
258 cancel_work_sync(&assigned_dev
->interrupt_work
);
260 free_irq(assigned_dev
->host_irq
, (void *)assigned_dev
);
262 if (assigned_dev
->irq_requested_type
& KVM_DEV_IRQ_HOST_MSI
)
263 pci_disable_msi(assigned_dev
->dev
);
266 assigned_dev
->irq_requested_type
&= ~(KVM_DEV_IRQ_HOST_MASK
);
269 static int kvm_deassign_irq(struct kvm
*kvm
,
270 struct kvm_assigned_dev_kernel
*assigned_dev
,
271 unsigned long irq_requested_type
)
273 unsigned long guest_irq_type
, host_irq_type
;
275 if (!irqchip_in_kernel(kvm
))
277 /* no irq assignment to deassign */
278 if (!assigned_dev
->irq_requested_type
)
281 host_irq_type
= irq_requested_type
& KVM_DEV_IRQ_HOST_MASK
;
282 guest_irq_type
= irq_requested_type
& KVM_DEV_IRQ_GUEST_MASK
;
285 deassign_host_irq(kvm
, assigned_dev
);
287 deassign_guest_irq(kvm
, assigned_dev
);
292 static void kvm_free_assigned_irq(struct kvm
*kvm
,
293 struct kvm_assigned_dev_kernel
*assigned_dev
)
295 kvm_deassign_irq(kvm
, assigned_dev
, assigned_dev
->irq_requested_type
);
298 static void kvm_free_assigned_device(struct kvm
*kvm
,
299 struct kvm_assigned_dev_kernel
302 kvm_free_assigned_irq(kvm
, assigned_dev
);
304 pci_reset_function(assigned_dev
->dev
);
306 pci_release_regions(assigned_dev
->dev
);
307 pci_disable_device(assigned_dev
->dev
);
308 pci_dev_put(assigned_dev
->dev
);
310 list_del(&assigned_dev
->list
);
314 void kvm_free_all_assigned_devices(struct kvm
*kvm
)
316 struct list_head
*ptr
, *ptr2
;
317 struct kvm_assigned_dev_kernel
*assigned_dev
;
319 list_for_each_safe(ptr
, ptr2
, &kvm
->arch
.assigned_dev_head
) {
320 assigned_dev
= list_entry(ptr
,
321 struct kvm_assigned_dev_kernel
,
324 kvm_free_assigned_device(kvm
, assigned_dev
);
328 static int assigned_device_enable_host_intx(struct kvm
*kvm
,
329 struct kvm_assigned_dev_kernel
*dev
)
331 dev
->host_irq
= dev
->dev
->irq
;
332 /* Even though this is PCI, we don't want to use shared
333 * interrupts. Sharing host devices with guest-assigned devices
334 * on the same interrupt line is not a happy situation: there
335 * are going to be long delays in accepting, acking, etc.
337 if (request_irq(dev
->host_irq
, kvm_assigned_dev_intr
,
338 0, "kvm_assigned_intx_device", (void *)dev
))
343 #ifdef __KVM_HAVE_MSI
344 static int assigned_device_enable_host_msi(struct kvm
*kvm
,
345 struct kvm_assigned_dev_kernel
*dev
)
349 if (!dev
->dev
->msi_enabled
) {
350 r
= pci_enable_msi(dev
->dev
);
355 dev
->host_irq
= dev
->dev
->irq
;
356 if (request_irq(dev
->host_irq
, kvm_assigned_dev_intr
, 0,
357 "kvm_assigned_msi_device", (void *)dev
)) {
358 pci_disable_msi(dev
->dev
);
366 #ifdef __KVM_HAVE_MSIX
367 static int assigned_device_enable_host_msix(struct kvm
*kvm
,
368 struct kvm_assigned_dev_kernel
*dev
)
372 /* host_msix_entries and guest_msix_entries should have been
374 if (dev
->entries_nr
== 0)
377 r
= pci_enable_msix(dev
->dev
, dev
->host_msix_entries
, dev
->entries_nr
);
381 for (i
= 0; i
< dev
->entries_nr
; i
++) {
382 r
= request_irq(dev
->host_msix_entries
[i
].vector
,
383 kvm_assigned_dev_intr
, 0,
384 "kvm_assigned_msix_device",
386 /* FIXME: free requested_irq's on failure */
396 static int assigned_device_enable_guest_intx(struct kvm
*kvm
,
397 struct kvm_assigned_dev_kernel
*dev
,
398 struct kvm_assigned_irq
*irq
)
400 dev
->guest_irq
= irq
->guest_irq
;
401 dev
->ack_notifier
.gsi
= irq
->guest_irq
;
405 #ifdef __KVM_HAVE_MSI
406 static int assigned_device_enable_guest_msi(struct kvm
*kvm
,
407 struct kvm_assigned_dev_kernel
*dev
,
408 struct kvm_assigned_irq
*irq
)
410 dev
->guest_irq
= irq
->guest_irq
;
411 dev
->ack_notifier
.gsi
= -1;
412 dev
->host_irq_disabled
= false;
416 #ifdef __KVM_HAVE_MSIX
417 static int assigned_device_enable_guest_msix(struct kvm
*kvm
,
418 struct kvm_assigned_dev_kernel
*dev
,
419 struct kvm_assigned_irq
*irq
)
421 dev
->guest_irq
= irq
->guest_irq
;
422 dev
->ack_notifier
.gsi
= -1;
423 dev
->host_irq_disabled
= false;
428 static int assign_host_irq(struct kvm
*kvm
,
429 struct kvm_assigned_dev_kernel
*dev
,
434 if (dev
->irq_requested_type
& KVM_DEV_IRQ_HOST_MASK
)
437 switch (host_irq_type
) {
438 case KVM_DEV_IRQ_HOST_INTX
:
439 r
= assigned_device_enable_host_intx(kvm
, dev
);
441 #ifdef __KVM_HAVE_MSI
442 case KVM_DEV_IRQ_HOST_MSI
:
443 r
= assigned_device_enable_host_msi(kvm
, dev
);
446 #ifdef __KVM_HAVE_MSIX
447 case KVM_DEV_IRQ_HOST_MSIX
:
448 r
= assigned_device_enable_host_msix(kvm
, dev
);
456 dev
->irq_requested_type
|= host_irq_type
;
461 static int assign_guest_irq(struct kvm
*kvm
,
462 struct kvm_assigned_dev_kernel
*dev
,
463 struct kvm_assigned_irq
*irq
,
464 unsigned long guest_irq_type
)
469 if (dev
->irq_requested_type
& KVM_DEV_IRQ_GUEST_MASK
)
472 id
= kvm_request_irq_source_id(kvm
);
476 dev
->irq_source_id
= id
;
478 switch (guest_irq_type
) {
479 case KVM_DEV_IRQ_GUEST_INTX
:
480 r
= assigned_device_enable_guest_intx(kvm
, dev
, irq
);
482 #ifdef __KVM_HAVE_MSI
483 case KVM_DEV_IRQ_GUEST_MSI
:
484 r
= assigned_device_enable_guest_msi(kvm
, dev
, irq
);
487 #ifdef __KVM_HAVE_MSIX
488 case KVM_DEV_IRQ_GUEST_MSIX
:
489 r
= assigned_device_enable_guest_msix(kvm
, dev
, irq
);
497 dev
->irq_requested_type
|= guest_irq_type
;
498 kvm_register_irq_ack_notifier(kvm
, &dev
->ack_notifier
);
500 kvm_free_irq_source_id(kvm
, dev
->irq_source_id
);
505 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
506 static int kvm_vm_ioctl_assign_irq(struct kvm
*kvm
,
507 struct kvm_assigned_irq
*assigned_irq
)
510 struct kvm_assigned_dev_kernel
*match
;
511 unsigned long host_irq_type
, guest_irq_type
;
513 if (!capable(CAP_SYS_RAWIO
))
516 if (!irqchip_in_kernel(kvm
))
519 mutex_lock(&kvm
->lock
);
521 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
522 assigned_irq
->assigned_dev_id
);
526 host_irq_type
= (assigned_irq
->flags
& KVM_DEV_IRQ_HOST_MASK
);
527 guest_irq_type
= (assigned_irq
->flags
& KVM_DEV_IRQ_GUEST_MASK
);
530 /* can only assign one type at a time */
531 if (hweight_long(host_irq_type
) > 1)
533 if (hweight_long(guest_irq_type
) > 1)
535 if (host_irq_type
== 0 && guest_irq_type
== 0)
540 r
= assign_host_irq(kvm
, match
, host_irq_type
);
545 r
= assign_guest_irq(kvm
, match
, assigned_irq
, guest_irq_type
);
547 mutex_unlock(&kvm
->lock
);
551 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm
*kvm
,
552 struct kvm_assigned_irq
556 struct kvm_assigned_dev_kernel
*match
;
558 mutex_lock(&kvm
->lock
);
560 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
561 assigned_irq
->assigned_dev_id
);
565 r
= kvm_deassign_irq(kvm
, match
, assigned_irq
->flags
);
567 mutex_unlock(&kvm
->lock
);
571 static int kvm_vm_ioctl_assign_device(struct kvm
*kvm
,
572 struct kvm_assigned_pci_dev
*assigned_dev
)
575 struct kvm_assigned_dev_kernel
*match
;
578 down_read(&kvm
->slots_lock
);
579 mutex_lock(&kvm
->lock
);
581 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
582 assigned_dev
->assigned_dev_id
);
584 /* device already assigned */
589 match
= kzalloc(sizeof(struct kvm_assigned_dev_kernel
), GFP_KERNEL
);
591 printk(KERN_INFO
"%s: Couldn't allocate memory\n",
596 dev
= pci_get_bus_and_slot(assigned_dev
->busnr
,
597 assigned_dev
->devfn
);
599 printk(KERN_INFO
"%s: host device not found\n", __func__
);
603 if (pci_enable_device(dev
)) {
604 printk(KERN_INFO
"%s: Could not enable PCI device\n", __func__
);
608 r
= pci_request_regions(dev
, "kvm_assigned_device");
610 printk(KERN_INFO
"%s: Could not get access to device regions\n",
615 pci_reset_function(dev
);
617 match
->assigned_dev_id
= assigned_dev
->assigned_dev_id
;
618 match
->host_busnr
= assigned_dev
->busnr
;
619 match
->host_devfn
= assigned_dev
->devfn
;
620 match
->flags
= assigned_dev
->flags
;
622 spin_lock_init(&match
->assigned_dev_lock
);
623 match
->irq_source_id
= -1;
625 match
->ack_notifier
.irq_acked
= kvm_assigned_dev_ack_irq
;
626 INIT_WORK(&match
->interrupt_work
,
627 kvm_assigned_dev_interrupt_work_handler
);
629 list_add(&match
->list
, &kvm
->arch
.assigned_dev_head
);
631 if (assigned_dev
->flags
& KVM_DEV_ASSIGN_ENABLE_IOMMU
) {
632 if (!kvm
->arch
.iommu_domain
) {
633 r
= kvm_iommu_map_guest(kvm
);
637 r
= kvm_assign_device(kvm
, match
);
643 mutex_unlock(&kvm
->lock
);
644 up_read(&kvm
->slots_lock
);
647 list_del(&match
->list
);
648 pci_release_regions(dev
);
650 pci_disable_device(dev
);
655 mutex_unlock(&kvm
->lock
);
656 up_read(&kvm
->slots_lock
);
661 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
662 static int kvm_vm_ioctl_deassign_device(struct kvm
*kvm
,
663 struct kvm_assigned_pci_dev
*assigned_dev
)
666 struct kvm_assigned_dev_kernel
*match
;
668 mutex_lock(&kvm
->lock
);
670 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
671 assigned_dev
->assigned_dev_id
);
673 printk(KERN_INFO
"%s: device hasn't been assigned before, "
674 "so cannot be deassigned\n", __func__
);
679 if (match
->flags
& KVM_DEV_ASSIGN_ENABLE_IOMMU
)
680 kvm_deassign_device(kvm
, match
);
682 kvm_free_assigned_device(kvm
, match
);
685 mutex_unlock(&kvm
->lock
);
690 static inline int valid_vcpu(int n
)
692 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
695 inline int kvm_is_mmio_pfn(pfn_t pfn
)
697 if (pfn_valid(pfn
)) {
698 struct page
*page
= compound_head(pfn_to_page(pfn
));
699 return PageReserved(page
);
706 * Switches to specified vcpu, until a matching vcpu_put()
708 void vcpu_load(struct kvm_vcpu
*vcpu
)
712 mutex_lock(&vcpu
->mutex
);
714 preempt_notifier_register(&vcpu
->preempt_notifier
);
715 kvm_arch_vcpu_load(vcpu
, cpu
);
719 void vcpu_put(struct kvm_vcpu
*vcpu
)
722 kvm_arch_vcpu_put(vcpu
);
723 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
725 mutex_unlock(&vcpu
->mutex
);
728 static void ack_flush(void *_completed
)
732 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
737 struct kvm_vcpu
*vcpu
;
739 if (alloc_cpumask_var(&cpus
, GFP_ATOMIC
))
743 spin_lock(&kvm
->requests_lock
);
744 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
745 vcpu
= kvm
->vcpus
[i
];
748 if (test_and_set_bit(req
, &vcpu
->requests
))
751 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
)
752 cpumask_set_cpu(cpu
, cpus
);
754 if (unlikely(cpus
== NULL
))
755 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
756 else if (!cpumask_empty(cpus
))
757 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
760 spin_unlock(&kvm
->requests_lock
);
762 free_cpumask_var(cpus
);
766 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
768 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
769 ++kvm
->stat
.remote_tlb_flush
;
772 void kvm_reload_remote_mmus(struct kvm
*kvm
)
774 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
777 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
782 mutex_init(&vcpu
->mutex
);
786 init_waitqueue_head(&vcpu
->wq
);
788 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
793 vcpu
->run
= page_address(page
);
795 r
= kvm_arch_vcpu_init(vcpu
);
801 free_page((unsigned long)vcpu
->run
);
805 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
807 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
809 kvm_arch_vcpu_uninit(vcpu
);
810 free_page((unsigned long)vcpu
->run
);
812 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
814 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
815 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
817 return container_of(mn
, struct kvm
, mmu_notifier
);
820 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
821 struct mm_struct
*mm
,
822 unsigned long address
)
824 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
828 * When ->invalidate_page runs, the linux pte has been zapped
829 * already but the page is still allocated until
830 * ->invalidate_page returns. So if we increase the sequence
831 * here the kvm page fault will notice if the spte can't be
832 * established because the page is going to be freed. If
833 * instead the kvm page fault establishes the spte before
834 * ->invalidate_page runs, kvm_unmap_hva will release it
837 * The sequence increase only need to be seen at spin_unlock
838 * time, and not at spin_lock time.
840 * Increasing the sequence after the spin_unlock would be
841 * unsafe because the kvm page fault could then establish the
842 * pte after kvm_unmap_hva returned, without noticing the page
843 * is going to be freed.
845 spin_lock(&kvm
->mmu_lock
);
846 kvm
->mmu_notifier_seq
++;
847 need_tlb_flush
= kvm_unmap_hva(kvm
, address
);
848 spin_unlock(&kvm
->mmu_lock
);
850 /* we've to flush the tlb before the pages can be freed */
852 kvm_flush_remote_tlbs(kvm
);
856 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
857 struct mm_struct
*mm
,
861 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
862 int need_tlb_flush
= 0;
864 spin_lock(&kvm
->mmu_lock
);
866 * The count increase must become visible at unlock time as no
867 * spte can be established without taking the mmu_lock and
868 * count is also read inside the mmu_lock critical section.
870 kvm
->mmu_notifier_count
++;
871 for (; start
< end
; start
+= PAGE_SIZE
)
872 need_tlb_flush
|= kvm_unmap_hva(kvm
, start
);
873 spin_unlock(&kvm
->mmu_lock
);
875 /* we've to flush the tlb before the pages can be freed */
877 kvm_flush_remote_tlbs(kvm
);
880 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
881 struct mm_struct
*mm
,
885 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
887 spin_lock(&kvm
->mmu_lock
);
889 * This sequence increase will notify the kvm page fault that
890 * the page that is going to be mapped in the spte could have
893 kvm
->mmu_notifier_seq
++;
895 * The above sequence increase must be visible before the
896 * below count decrease but both values are read by the kvm
897 * page fault under mmu_lock spinlock so we don't need to add
898 * a smb_wmb() here in between the two.
900 kvm
->mmu_notifier_count
--;
901 spin_unlock(&kvm
->mmu_lock
);
903 BUG_ON(kvm
->mmu_notifier_count
< 0);
906 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
907 struct mm_struct
*mm
,
908 unsigned long address
)
910 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
913 spin_lock(&kvm
->mmu_lock
);
914 young
= kvm_age_hva(kvm
, address
);
915 spin_unlock(&kvm
->mmu_lock
);
918 kvm_flush_remote_tlbs(kvm
);
923 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
924 struct mm_struct
*mm
)
926 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
927 kvm_arch_flush_shadow(kvm
);
930 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
931 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
932 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
933 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
934 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
935 .release
= kvm_mmu_notifier_release
,
937 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
939 static struct kvm
*kvm_create_vm(void)
941 struct kvm
*kvm
= kvm_arch_create_vm();
942 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
948 #ifdef CONFIG_HAVE_KVM_IRQCHIP
949 INIT_LIST_HEAD(&kvm
->irq_routing
);
950 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
953 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
954 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
957 return ERR_PTR(-ENOMEM
);
959 kvm
->coalesced_mmio_ring
=
960 (struct kvm_coalesced_mmio_ring
*)page_address(page
);
963 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
966 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
967 err
= mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
969 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
978 kvm
->mm
= current
->mm
;
979 atomic_inc(&kvm
->mm
->mm_count
);
980 spin_lock_init(&kvm
->mmu_lock
);
981 spin_lock_init(&kvm
->requests_lock
);
982 kvm_io_bus_init(&kvm
->pio_bus
);
984 mutex_init(&kvm
->lock
);
985 kvm_io_bus_init(&kvm
->mmio_bus
);
986 init_rwsem(&kvm
->slots_lock
);
987 atomic_set(&kvm
->users_count
, 1);
988 spin_lock(&kvm_lock
);
989 list_add(&kvm
->vm_list
, &vm_list
);
990 spin_unlock(&kvm_lock
);
991 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
992 kvm_coalesced_mmio_init(kvm
);
999 * Free any memory in @free but not in @dont.
1001 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
1002 struct kvm_memory_slot
*dont
)
1004 if (!dont
|| free
->rmap
!= dont
->rmap
)
1007 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
1008 vfree(free
->dirty_bitmap
);
1010 if (!dont
|| free
->lpage_info
!= dont
->lpage_info
)
1011 vfree(free
->lpage_info
);
1014 free
->dirty_bitmap
= NULL
;
1016 free
->lpage_info
= NULL
;
1019 void kvm_free_physmem(struct kvm
*kvm
)
1023 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
1024 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
1027 static void kvm_destroy_vm(struct kvm
*kvm
)
1029 struct mm_struct
*mm
= kvm
->mm
;
1031 kvm_arch_sync_events(kvm
);
1032 spin_lock(&kvm_lock
);
1033 list_del(&kvm
->vm_list
);
1034 spin_unlock(&kvm_lock
);
1035 kvm_free_irq_routing(kvm
);
1036 kvm_io_bus_destroy(&kvm
->pio_bus
);
1037 kvm_io_bus_destroy(&kvm
->mmio_bus
);
1038 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1039 if (kvm
->coalesced_mmio_ring
!= NULL
)
1040 free_page((unsigned long)kvm
->coalesced_mmio_ring
);
1042 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1043 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1045 kvm_arch_flush_shadow(kvm
);
1047 kvm_arch_destroy_vm(kvm
);
1051 void kvm_get_kvm(struct kvm
*kvm
)
1053 atomic_inc(&kvm
->users_count
);
1055 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1057 void kvm_put_kvm(struct kvm
*kvm
)
1059 if (atomic_dec_and_test(&kvm
->users_count
))
1060 kvm_destroy_vm(kvm
);
1062 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1065 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1067 struct kvm
*kvm
= filp
->private_data
;
1069 kvm_irqfd_release(kvm
);
1076 * Allocate some memory and give it an address in the guest physical address
1079 * Discontiguous memory is allowed, mostly for framebuffers.
1081 * Must be called holding mmap_sem for write.
1083 int __kvm_set_memory_region(struct kvm
*kvm
,
1084 struct kvm_userspace_memory_region
*mem
,
1089 unsigned long npages
, ugfn
;
1090 unsigned long largepages
, i
;
1091 struct kvm_memory_slot
*memslot
;
1092 struct kvm_memory_slot old
, new;
1095 /* General sanity checks */
1096 if (mem
->memory_size
& (PAGE_SIZE
- 1))
1098 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
1100 if (user_alloc
&& (mem
->userspace_addr
& (PAGE_SIZE
- 1)))
1102 if (mem
->slot
>= KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
)
1104 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
1107 memslot
= &kvm
->memslots
[mem
->slot
];
1108 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
1109 npages
= mem
->memory_size
>> PAGE_SHIFT
;
1112 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
1114 new = old
= *memslot
;
1116 new.base_gfn
= base_gfn
;
1117 new.npages
= npages
;
1118 new.flags
= mem
->flags
;
1120 /* Disallow changing a memory slot's size. */
1122 if (npages
&& old
.npages
&& npages
!= old
.npages
)
1125 /* Check for overlaps */
1127 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
1128 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
1130 if (s
== memslot
|| !s
->npages
)
1132 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
1133 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
1137 /* Free page dirty bitmap if unneeded */
1138 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1139 new.dirty_bitmap
= NULL
;
1143 /* Allocate if a slot is being created */
1145 if (npages
&& !new.rmap
) {
1146 new.rmap
= vmalloc(npages
* sizeof(struct page
*));
1151 memset(new.rmap
, 0, npages
* sizeof(*new.rmap
));
1153 new.user_alloc
= user_alloc
;
1155 * hva_to_rmmap() serialzies with the mmu_lock and to be
1156 * safe it has to ignore memslots with !user_alloc &&
1160 new.userspace_addr
= mem
->userspace_addr
;
1162 new.userspace_addr
= 0;
1164 if (npages
&& !new.lpage_info
) {
1165 largepages
= 1 + (base_gfn
+ npages
- 1) / KVM_PAGES_PER_HPAGE
;
1166 largepages
-= base_gfn
/ KVM_PAGES_PER_HPAGE
;
1168 new.lpage_info
= vmalloc(largepages
* sizeof(*new.lpage_info
));
1170 if (!new.lpage_info
)
1173 memset(new.lpage_info
, 0, largepages
* sizeof(*new.lpage_info
));
1175 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
1176 new.lpage_info
[0].write_count
= 1;
1177 if ((base_gfn
+npages
) % KVM_PAGES_PER_HPAGE
)
1178 new.lpage_info
[largepages
-1].write_count
= 1;
1179 ugfn
= new.userspace_addr
>> PAGE_SHIFT
;
1181 * If the gfn and userspace address are not aligned wrt each
1182 * other, disable large page support for this slot
1184 if ((base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE
- 1))
1185 for (i
= 0; i
< largepages
; ++i
)
1186 new.lpage_info
[i
].write_count
= 1;
1189 /* Allocate page dirty bitmap if needed */
1190 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1191 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
1193 new.dirty_bitmap
= vmalloc(dirty_bytes
);
1194 if (!new.dirty_bitmap
)
1196 memset(new.dirty_bitmap
, 0, dirty_bytes
);
1198 kvm_arch_flush_shadow(kvm
);
1200 #endif /* not defined CONFIG_S390 */
1203 kvm_arch_flush_shadow(kvm
);
1205 spin_lock(&kvm
->mmu_lock
);
1206 if (mem
->slot
>= kvm
->nmemslots
)
1207 kvm
->nmemslots
= mem
->slot
+ 1;
1210 spin_unlock(&kvm
->mmu_lock
);
1212 r
= kvm_arch_set_memory_region(kvm
, mem
, old
, user_alloc
);
1214 spin_lock(&kvm
->mmu_lock
);
1216 spin_unlock(&kvm
->mmu_lock
);
1220 kvm_free_physmem_slot(&old
, npages
? &new : NULL
);
1221 /* Slot deletion case: we have to update the current slot */
1222 spin_lock(&kvm
->mmu_lock
);
1225 spin_unlock(&kvm
->mmu_lock
);
1227 /* map the pages in iommu page table */
1228 r
= kvm_iommu_map_pages(kvm
, base_gfn
, npages
);
1235 kvm_free_physmem_slot(&new, &old
);
1240 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1242 int kvm_set_memory_region(struct kvm
*kvm
,
1243 struct kvm_userspace_memory_region
*mem
,
1248 down_write(&kvm
->slots_lock
);
1249 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
1250 up_write(&kvm
->slots_lock
);
1253 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1255 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1257 kvm_userspace_memory_region
*mem
,
1260 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
1262 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
1265 int kvm_get_dirty_log(struct kvm
*kvm
,
1266 struct kvm_dirty_log
*log
, int *is_dirty
)
1268 struct kvm_memory_slot
*memslot
;
1271 unsigned long any
= 0;
1274 if (log
->slot
>= KVM_MEMORY_SLOTS
)
1277 memslot
= &kvm
->memslots
[log
->slot
];
1279 if (!memslot
->dirty_bitmap
)
1282 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1284 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1285 any
= memslot
->dirty_bitmap
[i
];
1288 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1299 int is_error_page(struct page
*page
)
1301 return page
== bad_page
;
1303 EXPORT_SYMBOL_GPL(is_error_page
);
1305 int is_error_pfn(pfn_t pfn
)
1307 return pfn
== bad_pfn
;
1309 EXPORT_SYMBOL_GPL(is_error_pfn
);
1311 static inline unsigned long bad_hva(void)
1316 int kvm_is_error_hva(unsigned long addr
)
1318 return addr
== bad_hva();
1320 EXPORT_SYMBOL_GPL(kvm_is_error_hva
);
1322 struct kvm_memory_slot
*gfn_to_memslot_unaliased(struct kvm
*kvm
, gfn_t gfn
)
1326 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
1327 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
1329 if (gfn
>= memslot
->base_gfn
1330 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
1335 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased
);
1337 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1339 gfn
= unalias_gfn(kvm
, gfn
);
1340 return gfn_to_memslot_unaliased(kvm
, gfn
);
1343 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1347 gfn
= unalias_gfn(kvm
, gfn
);
1348 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
1349 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
1351 if (gfn
>= memslot
->base_gfn
1352 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
1357 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1359 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1361 struct kvm_memory_slot
*slot
;
1363 gfn
= unalias_gfn(kvm
, gfn
);
1364 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
1367 return (slot
->userspace_addr
+ (gfn
- slot
->base_gfn
) * PAGE_SIZE
);
1369 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1371 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1373 struct page
*page
[1];
1380 addr
= gfn_to_hva(kvm
, gfn
);
1381 if (kvm_is_error_hva(addr
)) {
1383 return page_to_pfn(bad_page
);
1386 npages
= get_user_pages_fast(addr
, 1, 1, page
);
1388 if (unlikely(npages
!= 1)) {
1389 struct vm_area_struct
*vma
;
1391 down_read(¤t
->mm
->mmap_sem
);
1392 vma
= find_vma(current
->mm
, addr
);
1394 if (vma
== NULL
|| addr
< vma
->vm_start
||
1395 !(vma
->vm_flags
& VM_PFNMAP
)) {
1396 up_read(¤t
->mm
->mmap_sem
);
1398 return page_to_pfn(bad_page
);
1401 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1402 up_read(¤t
->mm
->mmap_sem
);
1403 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1405 pfn
= page_to_pfn(page
[0]);
1410 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1412 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1416 pfn
= gfn_to_pfn(kvm
, gfn
);
1417 if (!kvm_is_mmio_pfn(pfn
))
1418 return pfn_to_page(pfn
);
1420 WARN_ON(kvm_is_mmio_pfn(pfn
));
1426 EXPORT_SYMBOL_GPL(gfn_to_page
);
1428 void kvm_release_page_clean(struct page
*page
)
1430 kvm_release_pfn_clean(page_to_pfn(page
));
1432 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1434 void kvm_release_pfn_clean(pfn_t pfn
)
1436 if (!kvm_is_mmio_pfn(pfn
))
1437 put_page(pfn_to_page(pfn
));
1439 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1441 void kvm_release_page_dirty(struct page
*page
)
1443 kvm_release_pfn_dirty(page_to_pfn(page
));
1445 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1447 void kvm_release_pfn_dirty(pfn_t pfn
)
1449 kvm_set_pfn_dirty(pfn
);
1450 kvm_release_pfn_clean(pfn
);
1452 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1454 void kvm_set_page_dirty(struct page
*page
)
1456 kvm_set_pfn_dirty(page_to_pfn(page
));
1458 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1460 void kvm_set_pfn_dirty(pfn_t pfn
)
1462 if (!kvm_is_mmio_pfn(pfn
)) {
1463 struct page
*page
= pfn_to_page(pfn
);
1464 if (!PageReserved(page
))
1468 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1470 void kvm_set_pfn_accessed(pfn_t pfn
)
1472 if (!kvm_is_mmio_pfn(pfn
))
1473 mark_page_accessed(pfn_to_page(pfn
));
1475 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1477 void kvm_get_pfn(pfn_t pfn
)
1479 if (!kvm_is_mmio_pfn(pfn
))
1480 get_page(pfn_to_page(pfn
));
1482 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1484 static int next_segment(unsigned long len
, int offset
)
1486 if (len
> PAGE_SIZE
- offset
)
1487 return PAGE_SIZE
- offset
;
1492 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1498 addr
= gfn_to_hva(kvm
, gfn
);
1499 if (kvm_is_error_hva(addr
))
1501 r
= copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1506 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1508 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1510 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1512 int offset
= offset_in_page(gpa
);
1515 while ((seg
= next_segment(len
, offset
)) != 0) {
1516 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1526 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1528 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1533 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1534 int offset
= offset_in_page(gpa
);
1536 addr
= gfn_to_hva(kvm
, gfn
);
1537 if (kvm_is_error_hva(addr
))
1539 pagefault_disable();
1540 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1546 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1548 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1549 int offset
, int len
)
1554 addr
= gfn_to_hva(kvm
, gfn
);
1555 if (kvm_is_error_hva(addr
))
1557 r
= copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1560 mark_page_dirty(kvm
, gfn
);
1563 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1565 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1568 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1570 int offset
= offset_in_page(gpa
);
1573 while ((seg
= next_segment(len
, offset
)) != 0) {
1574 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1585 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1587 return kvm_write_guest_page(kvm
, gfn
, empty_zero_page
, offset
, len
);
1589 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1591 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1593 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1595 int offset
= offset_in_page(gpa
);
1598 while ((seg
= next_segment(len
, offset
)) != 0) {
1599 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1608 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1610 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1612 struct kvm_memory_slot
*memslot
;
1614 gfn
= unalias_gfn(kvm
, gfn
);
1615 memslot
= gfn_to_memslot_unaliased(kvm
, gfn
);
1616 if (memslot
&& memslot
->dirty_bitmap
) {
1617 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1620 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
1621 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
1626 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1628 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1633 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1635 if ((kvm_arch_interrupt_allowed(vcpu
) &&
1636 kvm_cpu_has_interrupt(vcpu
)) ||
1637 kvm_arch_vcpu_runnable(vcpu
)) {
1638 set_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
1641 if (kvm_cpu_has_pending_timer(vcpu
))
1643 if (signal_pending(current
))
1651 finish_wait(&vcpu
->wq
, &wait
);
1654 void kvm_resched(struct kvm_vcpu
*vcpu
)
1656 if (!need_resched())
1660 EXPORT_SYMBOL_GPL(kvm_resched
);
1662 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1664 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1667 if (vmf
->pgoff
== 0)
1668 page
= virt_to_page(vcpu
->run
);
1670 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1671 page
= virt_to_page(vcpu
->arch
.pio_data
);
1673 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1674 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1675 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1678 return VM_FAULT_SIGBUS
;
1684 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
1685 .fault
= kvm_vcpu_fault
,
1688 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1690 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1694 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1696 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1698 kvm_put_kvm(vcpu
->kvm
);
1702 static struct file_operations kvm_vcpu_fops
= {
1703 .release
= kvm_vcpu_release
,
1704 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1705 .compat_ioctl
= kvm_vcpu_ioctl
,
1706 .mmap
= kvm_vcpu_mmap
,
1710 * Allocates an inode for the vcpu.
1712 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1714 int fd
= anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, 0);
1716 kvm_put_kvm(vcpu
->kvm
);
1721 * Creates some virtual cpus. Good luck creating more than one.
1723 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
1726 struct kvm_vcpu
*vcpu
;
1731 vcpu
= kvm_arch_vcpu_create(kvm
, n
);
1733 return PTR_ERR(vcpu
);
1735 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1737 r
= kvm_arch_vcpu_setup(vcpu
);
1741 mutex_lock(&kvm
->lock
);
1742 if (kvm
->vcpus
[n
]) {
1746 kvm
->vcpus
[n
] = vcpu
;
1747 mutex_unlock(&kvm
->lock
);
1749 /* Now it's all set up, let userspace reach it */
1751 r
= create_vcpu_fd(vcpu
);
1757 mutex_lock(&kvm
->lock
);
1758 kvm
->vcpus
[n
] = NULL
;
1760 mutex_unlock(&kvm
->lock
);
1761 kvm_arch_vcpu_destroy(vcpu
);
1765 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1768 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1769 vcpu
->sigset_active
= 1;
1770 vcpu
->sigset
= *sigset
;
1772 vcpu
->sigset_active
= 0;
1776 #ifdef __KVM_HAVE_MSIX
1777 static int kvm_vm_ioctl_set_msix_nr(struct kvm
*kvm
,
1778 struct kvm_assigned_msix_nr
*entry_nr
)
1781 struct kvm_assigned_dev_kernel
*adev
;
1783 mutex_lock(&kvm
->lock
);
1785 adev
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
1786 entry_nr
->assigned_dev_id
);
1792 if (adev
->entries_nr
== 0) {
1793 adev
->entries_nr
= entry_nr
->entry_nr
;
1794 if (adev
->entries_nr
== 0 ||
1795 adev
->entries_nr
>= KVM_MAX_MSIX_PER_DEV
) {
1800 adev
->host_msix_entries
= kzalloc(sizeof(struct msix_entry
) *
1803 if (!adev
->host_msix_entries
) {
1807 adev
->guest_msix_entries
= kzalloc(
1808 sizeof(struct kvm_guest_msix_entry
) *
1809 entry_nr
->entry_nr
, GFP_KERNEL
);
1810 if (!adev
->guest_msix_entries
) {
1811 kfree(adev
->host_msix_entries
);
1815 } else /* Not allowed set MSI-X number twice */
1818 mutex_unlock(&kvm
->lock
);
1822 static int kvm_vm_ioctl_set_msix_entry(struct kvm
*kvm
,
1823 struct kvm_assigned_msix_entry
*entry
)
1826 struct kvm_assigned_dev_kernel
*adev
;
1828 mutex_lock(&kvm
->lock
);
1830 adev
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
1831 entry
->assigned_dev_id
);
1835 goto msix_entry_out
;
1838 for (i
= 0; i
< adev
->entries_nr
; i
++)
1839 if (adev
->guest_msix_entries
[i
].vector
== 0 ||
1840 adev
->guest_msix_entries
[i
].entry
== entry
->entry
) {
1841 adev
->guest_msix_entries
[i
].entry
= entry
->entry
;
1842 adev
->guest_msix_entries
[i
].vector
= entry
->gsi
;
1843 adev
->host_msix_entries
[i
].entry
= entry
->entry
;
1846 if (i
== adev
->entries_nr
) {
1848 goto msix_entry_out
;
1852 mutex_unlock(&kvm
->lock
);
1858 static long kvm_vcpu_ioctl(struct file
*filp
,
1859 unsigned int ioctl
, unsigned long arg
)
1861 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1862 void __user
*argp
= (void __user
*)arg
;
1864 struct kvm_fpu
*fpu
= NULL
;
1865 struct kvm_sregs
*kvm_sregs
= NULL
;
1867 if (vcpu
->kvm
->mm
!= current
->mm
)
1874 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1876 case KVM_GET_REGS
: {
1877 struct kvm_regs
*kvm_regs
;
1880 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1883 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1887 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1894 case KVM_SET_REGS
: {
1895 struct kvm_regs
*kvm_regs
;
1898 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1902 if (copy_from_user(kvm_regs
, argp
, sizeof(struct kvm_regs
)))
1904 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1912 case KVM_GET_SREGS
: {
1913 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1917 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1921 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
1926 case KVM_SET_SREGS
: {
1927 kvm_sregs
= kmalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1932 if (copy_from_user(kvm_sregs
, argp
, sizeof(struct kvm_sregs
)))
1934 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
1940 case KVM_GET_MP_STATE
: {
1941 struct kvm_mp_state mp_state
;
1943 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1947 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
1952 case KVM_SET_MP_STATE
: {
1953 struct kvm_mp_state mp_state
;
1956 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
1958 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
1964 case KVM_TRANSLATE
: {
1965 struct kvm_translation tr
;
1968 if (copy_from_user(&tr
, argp
, sizeof tr
))
1970 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
1974 if (copy_to_user(argp
, &tr
, sizeof tr
))
1979 case KVM_SET_GUEST_DEBUG
: {
1980 struct kvm_guest_debug dbg
;
1983 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
1985 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
1991 case KVM_SET_SIGNAL_MASK
: {
1992 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
1993 struct kvm_signal_mask kvm_sigmask
;
1994 sigset_t sigset
, *p
;
1999 if (copy_from_user(&kvm_sigmask
, argp
,
2000 sizeof kvm_sigmask
))
2003 if (kvm_sigmask
.len
!= sizeof sigset
)
2006 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2011 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2015 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2019 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2023 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2029 fpu
= kmalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2034 if (copy_from_user(fpu
, argp
, sizeof(struct kvm_fpu
)))
2036 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2043 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2051 static long kvm_vm_ioctl(struct file
*filp
,
2052 unsigned int ioctl
, unsigned long arg
)
2054 struct kvm
*kvm
= filp
->private_data
;
2055 void __user
*argp
= (void __user
*)arg
;
2058 if (kvm
->mm
!= current
->mm
)
2061 case KVM_CREATE_VCPU
:
2062 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2066 case KVM_SET_USER_MEMORY_REGION
: {
2067 struct kvm_userspace_memory_region kvm_userspace_mem
;
2070 if (copy_from_user(&kvm_userspace_mem
, argp
,
2071 sizeof kvm_userspace_mem
))
2074 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 1);
2079 case KVM_GET_DIRTY_LOG
: {
2080 struct kvm_dirty_log log
;
2083 if (copy_from_user(&log
, argp
, sizeof log
))
2085 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2090 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2091 case KVM_REGISTER_COALESCED_MMIO
: {
2092 struct kvm_coalesced_mmio_zone zone
;
2094 if (copy_from_user(&zone
, argp
, sizeof zone
))
2097 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2103 case KVM_UNREGISTER_COALESCED_MMIO
: {
2104 struct kvm_coalesced_mmio_zone zone
;
2106 if (copy_from_user(&zone
, argp
, sizeof zone
))
2109 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2116 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2117 case KVM_ASSIGN_PCI_DEVICE
: {
2118 struct kvm_assigned_pci_dev assigned_dev
;
2121 if (copy_from_user(&assigned_dev
, argp
, sizeof assigned_dev
))
2123 r
= kvm_vm_ioctl_assign_device(kvm
, &assigned_dev
);
2128 case KVM_ASSIGN_IRQ
: {
2132 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2133 case KVM_ASSIGN_DEV_IRQ
: {
2134 struct kvm_assigned_irq assigned_irq
;
2137 if (copy_from_user(&assigned_irq
, argp
, sizeof assigned_irq
))
2139 r
= kvm_vm_ioctl_assign_irq(kvm
, &assigned_irq
);
2144 case KVM_DEASSIGN_DEV_IRQ
: {
2145 struct kvm_assigned_irq assigned_irq
;
2148 if (copy_from_user(&assigned_irq
, argp
, sizeof assigned_irq
))
2150 r
= kvm_vm_ioctl_deassign_dev_irq(kvm
, &assigned_irq
);
2157 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2158 case KVM_DEASSIGN_PCI_DEVICE
: {
2159 struct kvm_assigned_pci_dev assigned_dev
;
2162 if (copy_from_user(&assigned_dev
, argp
, sizeof assigned_dev
))
2164 r
= kvm_vm_ioctl_deassign_device(kvm
, &assigned_dev
);
2170 #ifdef KVM_CAP_IRQ_ROUTING
2171 case KVM_SET_GSI_ROUTING
: {
2172 struct kvm_irq_routing routing
;
2173 struct kvm_irq_routing __user
*urouting
;
2174 struct kvm_irq_routing_entry
*entries
;
2177 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2180 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2185 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2190 if (copy_from_user(entries
, urouting
->entries
,
2191 routing
.nr
* sizeof(*entries
)))
2192 goto out_free_irq_routing
;
2193 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2195 out_free_irq_routing
:
2199 #ifdef __KVM_HAVE_MSIX
2200 case KVM_ASSIGN_SET_MSIX_NR
: {
2201 struct kvm_assigned_msix_nr entry_nr
;
2203 if (copy_from_user(&entry_nr
, argp
, sizeof entry_nr
))
2205 r
= kvm_vm_ioctl_set_msix_nr(kvm
, &entry_nr
);
2210 case KVM_ASSIGN_SET_MSIX_ENTRY
: {
2211 struct kvm_assigned_msix_entry entry
;
2213 if (copy_from_user(&entry
, argp
, sizeof entry
))
2215 r
= kvm_vm_ioctl_set_msix_entry(kvm
, &entry
);
2221 #endif /* KVM_CAP_IRQ_ROUTING */
2223 struct kvm_irqfd data
;
2226 if (copy_from_user(&data
, argp
, sizeof data
))
2228 r
= kvm_irqfd(kvm
, data
.fd
, data
.gsi
, data
.flags
);
2232 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2238 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2240 struct page
*page
[1];
2243 gfn_t gfn
= vmf
->pgoff
;
2244 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2246 addr
= gfn_to_hva(kvm
, gfn
);
2247 if (kvm_is_error_hva(addr
))
2248 return VM_FAULT_SIGBUS
;
2250 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
2252 if (unlikely(npages
!= 1))
2253 return VM_FAULT_SIGBUS
;
2255 vmf
->page
= page
[0];
2259 static struct vm_operations_struct kvm_vm_vm_ops
= {
2260 .fault
= kvm_vm_fault
,
2263 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2265 vma
->vm_ops
= &kvm_vm_vm_ops
;
2269 static struct file_operations kvm_vm_fops
= {
2270 .release
= kvm_vm_release
,
2271 .unlocked_ioctl
= kvm_vm_ioctl
,
2272 .compat_ioctl
= kvm_vm_ioctl
,
2273 .mmap
= kvm_vm_mmap
,
2276 static int kvm_dev_ioctl_create_vm(void)
2281 kvm
= kvm_create_vm();
2283 return PTR_ERR(kvm
);
2284 fd
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, 0);
2291 static long kvm_dev_ioctl_check_extension_generic(long arg
)
2294 case KVM_CAP_USER_MEMORY
:
2295 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2296 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2298 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2299 case KVM_CAP_IRQ_ROUTING
:
2300 return KVM_MAX_IRQ_ROUTES
;
2305 return kvm_dev_ioctl_check_extension(arg
);
2308 static long kvm_dev_ioctl(struct file
*filp
,
2309 unsigned int ioctl
, unsigned long arg
)
2314 case KVM_GET_API_VERSION
:
2318 r
= KVM_API_VERSION
;
2324 r
= kvm_dev_ioctl_create_vm();
2326 case KVM_CHECK_EXTENSION
:
2327 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2329 case KVM_GET_VCPU_MMAP_SIZE
:
2333 r
= PAGE_SIZE
; /* struct kvm_run */
2335 r
+= PAGE_SIZE
; /* pio data page */
2337 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2338 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2341 case KVM_TRACE_ENABLE
:
2342 case KVM_TRACE_PAUSE
:
2343 case KVM_TRACE_DISABLE
:
2344 r
= kvm_trace_ioctl(ioctl
, arg
);
2347 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2353 static struct file_operations kvm_chardev_ops
= {
2354 .unlocked_ioctl
= kvm_dev_ioctl
,
2355 .compat_ioctl
= kvm_dev_ioctl
,
2358 static struct miscdevice kvm_dev
= {
2364 static void hardware_enable(void *junk
)
2366 int cpu
= raw_smp_processor_id();
2368 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2370 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2371 kvm_arch_hardware_enable(NULL
);
2374 static void hardware_disable(void *junk
)
2376 int cpu
= raw_smp_processor_id();
2378 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2380 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2381 kvm_arch_hardware_disable(NULL
);
2384 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2389 val
&= ~CPU_TASKS_FROZEN
;
2392 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2394 hardware_disable(NULL
);
2396 case CPU_UP_CANCELED
:
2397 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2399 smp_call_function_single(cpu
, hardware_disable
, NULL
, 1);
2402 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2404 smp_call_function_single(cpu
, hardware_enable
, NULL
, 1);
2411 asmlinkage
void kvm_handle_fault_on_reboot(void)
2414 /* spin while reset goes on */
2417 /* Fault while not rebooting. We want the trace. */
2420 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot
);
2422 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2426 * Some (well, at least mine) BIOSes hang on reboot if
2429 * And Intel TXT required VMX off for all cpu when system shutdown.
2431 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2432 kvm_rebooting
= true;
2433 on_each_cpu(hardware_disable
, NULL
, 1);
2437 static struct notifier_block kvm_reboot_notifier
= {
2438 .notifier_call
= kvm_reboot
,
2442 void kvm_io_bus_init(struct kvm_io_bus
*bus
)
2444 memset(bus
, 0, sizeof(*bus
));
2447 void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2451 for (i
= 0; i
< bus
->dev_count
; i
++) {
2452 struct kvm_io_device
*pos
= bus
->devs
[i
];
2454 kvm_iodevice_destructor(pos
);
2458 struct kvm_io_device
*kvm_io_bus_find_dev(struct kvm_io_bus
*bus
,
2459 gpa_t addr
, int len
, int is_write
)
2463 for (i
= 0; i
< bus
->dev_count
; i
++) {
2464 struct kvm_io_device
*pos
= bus
->devs
[i
];
2466 if (kvm_iodevice_in_range(pos
, addr
, len
, is_write
))
2473 void kvm_io_bus_register_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
)
2475 BUG_ON(bus
->dev_count
> (NR_IOBUS_DEVS
-1));
2477 bus
->devs
[bus
->dev_count
++] = dev
;
2480 static struct notifier_block kvm_cpu_notifier
= {
2481 .notifier_call
= kvm_cpu_hotplug
,
2482 .priority
= 20, /* must be > scheduler priority */
2485 static int vm_stat_get(void *_offset
, u64
*val
)
2487 unsigned offset
= (long)_offset
;
2491 spin_lock(&kvm_lock
);
2492 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2493 *val
+= *(u32
*)((void *)kvm
+ offset
);
2494 spin_unlock(&kvm_lock
);
2498 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2500 static int vcpu_stat_get(void *_offset
, u64
*val
)
2502 unsigned offset
= (long)_offset
;
2504 struct kvm_vcpu
*vcpu
;
2508 spin_lock(&kvm_lock
);
2509 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2510 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2511 vcpu
= kvm
->vcpus
[i
];
2513 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2515 spin_unlock(&kvm_lock
);
2519 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2521 static struct file_operations
*stat_fops
[] = {
2522 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2523 [KVM_STAT_VM
] = &vm_stat_fops
,
2526 static void kvm_init_debug(void)
2528 struct kvm_stats_debugfs_item
*p
;
2530 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2531 for (p
= debugfs_entries
; p
->name
; ++p
)
2532 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2533 (void *)(long)p
->offset
,
2534 stat_fops
[p
->kind
]);
2537 static void kvm_exit_debug(void)
2539 struct kvm_stats_debugfs_item
*p
;
2541 for (p
= debugfs_entries
; p
->name
; ++p
)
2542 debugfs_remove(p
->dentry
);
2543 debugfs_remove(kvm_debugfs_dir
);
2546 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2548 hardware_disable(NULL
);
2552 static int kvm_resume(struct sys_device
*dev
)
2554 hardware_enable(NULL
);
2558 static struct sysdev_class kvm_sysdev_class
= {
2560 .suspend
= kvm_suspend
,
2561 .resume
= kvm_resume
,
2564 static struct sys_device kvm_sysdev
= {
2566 .cls
= &kvm_sysdev_class
,
2569 struct page
*bad_page
;
2573 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2575 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2578 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2580 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2582 kvm_arch_vcpu_load(vcpu
, cpu
);
2585 static void kvm_sched_out(struct preempt_notifier
*pn
,
2586 struct task_struct
*next
)
2588 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2590 kvm_arch_vcpu_put(vcpu
);
2593 int kvm_init(void *opaque
, unsigned int vcpu_size
,
2594 struct module
*module
)
2601 r
= kvm_arch_init(opaque
);
2605 bad_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2607 if (bad_page
== NULL
) {
2612 bad_pfn
= page_to_pfn(bad_page
);
2614 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2619 r
= kvm_arch_hardware_setup();
2623 for_each_online_cpu(cpu
) {
2624 smp_call_function_single(cpu
,
2625 kvm_arch_check_processor_compat
,
2631 on_each_cpu(hardware_enable
, NULL
, 1);
2632 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2635 register_reboot_notifier(&kvm_reboot_notifier
);
2637 r
= sysdev_class_register(&kvm_sysdev_class
);
2641 r
= sysdev_register(&kvm_sysdev
);
2645 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2646 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
,
2647 __alignof__(struct kvm_vcpu
),
2649 if (!kvm_vcpu_cache
) {
2654 kvm_chardev_ops
.owner
= module
;
2655 kvm_vm_fops
.owner
= module
;
2656 kvm_vcpu_fops
.owner
= module
;
2658 r
= misc_register(&kvm_dev
);
2660 printk(KERN_ERR
"kvm: misc device register failed\n");
2664 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2665 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2670 kmem_cache_destroy(kvm_vcpu_cache
);
2672 sysdev_unregister(&kvm_sysdev
);
2674 sysdev_class_unregister(&kvm_sysdev_class
);
2676 unregister_reboot_notifier(&kvm_reboot_notifier
);
2677 unregister_cpu_notifier(&kvm_cpu_notifier
);
2679 on_each_cpu(hardware_disable
, NULL
, 1);
2681 kvm_arch_hardware_unsetup();
2683 free_cpumask_var(cpus_hardware_enabled
);
2685 __free_page(bad_page
);
2692 EXPORT_SYMBOL_GPL(kvm_init
);
2696 kvm_trace_cleanup();
2697 misc_deregister(&kvm_dev
);
2698 kmem_cache_destroy(kvm_vcpu_cache
);
2699 sysdev_unregister(&kvm_sysdev
);
2700 sysdev_class_unregister(&kvm_sysdev_class
);
2701 unregister_reboot_notifier(&kvm_reboot_notifier
);
2702 unregister_cpu_notifier(&kvm_cpu_notifier
);
2703 on_each_cpu(hardware_disable
, NULL
, 1);
2704 kvm_arch_hardware_unsetup();
2707 free_cpumask_var(cpus_hardware_enabled
);
2708 __free_page(bad_page
);
2710 EXPORT_SYMBOL_GPL(kvm_exit
);