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>
50 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
51 #include "coalesced_mmio.h"
54 MODULE_AUTHOR("Qumranet");
55 MODULE_LICENSE("GPL");
57 DEFINE_SPINLOCK(kvm_lock
);
60 static cpumask_t cpus_hardware_enabled
;
62 struct kmem_cache
*kvm_vcpu_cache
;
63 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
65 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
67 struct dentry
*kvm_debugfs_dir
;
69 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
74 static inline int valid_vcpu(int n
)
76 return likely(n
>= 0 && n
< KVM_MAX_VCPUS
);
80 * Switches to specified vcpu, until a matching vcpu_put()
82 void vcpu_load(struct kvm_vcpu
*vcpu
)
86 mutex_lock(&vcpu
->mutex
);
88 preempt_notifier_register(&vcpu
->preempt_notifier
);
89 kvm_arch_vcpu_load(vcpu
, cpu
);
93 void vcpu_put(struct kvm_vcpu
*vcpu
)
96 kvm_arch_vcpu_put(vcpu
);
97 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
99 mutex_unlock(&vcpu
->mutex
);
102 static void ack_flush(void *_completed
)
106 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
110 struct kvm_vcpu
*vcpu
;
113 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
114 vcpu
= kvm
->vcpus
[i
];
117 if (test_and_set_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
120 if (cpu
!= -1 && cpu
!= raw_smp_processor_id())
123 if (cpus_empty(cpus
))
125 ++kvm
->stat
.remote_tlb_flush
;
126 smp_call_function_mask(cpus
, ack_flush
, NULL
, 1);
129 void kvm_reload_remote_mmus(struct kvm
*kvm
)
133 struct kvm_vcpu
*vcpu
;
136 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
137 vcpu
= kvm
->vcpus
[i
];
140 if (test_and_set_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
143 if (cpu
!= -1 && cpu
!= raw_smp_processor_id())
146 if (cpus_empty(cpus
))
148 smp_call_function_mask(cpus
, ack_flush
, NULL
, 1);
152 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
157 mutex_init(&vcpu
->mutex
);
161 init_waitqueue_head(&vcpu
->wq
);
163 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
168 vcpu
->run
= page_address(page
);
170 r
= kvm_arch_vcpu_init(vcpu
);
176 free_page((unsigned long)vcpu
->run
);
180 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
182 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
184 kvm_arch_vcpu_uninit(vcpu
);
185 free_page((unsigned long)vcpu
->run
);
187 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
189 static struct kvm
*kvm_create_vm(void)
191 struct kvm
*kvm
= kvm_arch_create_vm();
192 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
199 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
200 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
203 return ERR_PTR(-ENOMEM
);
205 kvm
->coalesced_mmio_ring
=
206 (struct kvm_coalesced_mmio_ring
*)page_address(page
);
209 kvm
->mm
= current
->mm
;
210 atomic_inc(&kvm
->mm
->mm_count
);
211 spin_lock_init(&kvm
->mmu_lock
);
212 kvm_io_bus_init(&kvm
->pio_bus
);
213 mutex_init(&kvm
->lock
);
214 kvm_io_bus_init(&kvm
->mmio_bus
);
215 init_rwsem(&kvm
->slots_lock
);
216 atomic_set(&kvm
->users_count
, 1);
217 spin_lock(&kvm_lock
);
218 list_add(&kvm
->vm_list
, &vm_list
);
219 spin_unlock(&kvm_lock
);
220 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
221 kvm_coalesced_mmio_init(kvm
);
228 * Free any memory in @free but not in @dont.
230 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
231 struct kvm_memory_slot
*dont
)
233 if (!dont
|| free
->rmap
!= dont
->rmap
)
236 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
237 vfree(free
->dirty_bitmap
);
239 if (!dont
|| free
->lpage_info
!= dont
->lpage_info
)
240 vfree(free
->lpage_info
);
243 free
->dirty_bitmap
= NULL
;
245 free
->lpage_info
= NULL
;
248 void kvm_free_physmem(struct kvm
*kvm
)
252 for (i
= 0; i
< kvm
->nmemslots
; ++i
)
253 kvm_free_physmem_slot(&kvm
->memslots
[i
], NULL
);
256 static void kvm_destroy_vm(struct kvm
*kvm
)
258 struct mm_struct
*mm
= kvm
->mm
;
260 spin_lock(&kvm_lock
);
261 list_del(&kvm
->vm_list
);
262 spin_unlock(&kvm_lock
);
263 kvm_io_bus_destroy(&kvm
->pio_bus
);
264 kvm_io_bus_destroy(&kvm
->mmio_bus
);
265 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
266 if (kvm
->coalesced_mmio_ring
!= NULL
)
267 free_page((unsigned long)kvm
->coalesced_mmio_ring
);
269 kvm_arch_destroy_vm(kvm
);
273 void kvm_get_kvm(struct kvm
*kvm
)
275 atomic_inc(&kvm
->users_count
);
277 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
279 void kvm_put_kvm(struct kvm
*kvm
)
281 if (atomic_dec_and_test(&kvm
->users_count
))
284 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
287 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
289 struct kvm
*kvm
= filp
->private_data
;
296 * Allocate some memory and give it an address in the guest physical address
299 * Discontiguous memory is allowed, mostly for framebuffers.
301 * Must be called holding mmap_sem for write.
303 int __kvm_set_memory_region(struct kvm
*kvm
,
304 struct kvm_userspace_memory_region
*mem
,
309 unsigned long npages
;
311 struct kvm_memory_slot
*memslot
;
312 struct kvm_memory_slot old
, new;
315 /* General sanity checks */
316 if (mem
->memory_size
& (PAGE_SIZE
- 1))
318 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
320 if (mem
->slot
>= KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
)
322 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
325 memslot
= &kvm
->memslots
[mem
->slot
];
326 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
327 npages
= mem
->memory_size
>> PAGE_SHIFT
;
330 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
332 new = old
= *memslot
;
334 new.base_gfn
= base_gfn
;
336 new.flags
= mem
->flags
;
338 /* Disallow changing a memory slot's size. */
340 if (npages
&& old
.npages
&& npages
!= old
.npages
)
343 /* Check for overlaps */
345 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
346 struct kvm_memory_slot
*s
= &kvm
->memslots
[i
];
350 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
351 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
355 /* Free page dirty bitmap if unneeded */
356 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
357 new.dirty_bitmap
= NULL
;
361 /* Allocate if a slot is being created */
362 if (npages
&& !new.rmap
) {
363 new.rmap
= vmalloc(npages
* sizeof(struct page
*));
368 memset(new.rmap
, 0, npages
* sizeof(*new.rmap
));
370 new.user_alloc
= user_alloc
;
371 new.userspace_addr
= mem
->userspace_addr
;
373 if (npages
&& !new.lpage_info
) {
374 int largepages
= npages
/ KVM_PAGES_PER_HPAGE
;
375 if (npages
% KVM_PAGES_PER_HPAGE
)
377 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
380 new.lpage_info
= vmalloc(largepages
* sizeof(*new.lpage_info
));
385 memset(new.lpage_info
, 0, largepages
* sizeof(*new.lpage_info
));
387 if (base_gfn
% KVM_PAGES_PER_HPAGE
)
388 new.lpage_info
[0].write_count
= 1;
389 if ((base_gfn
+npages
) % KVM_PAGES_PER_HPAGE
)
390 new.lpage_info
[largepages
-1].write_count
= 1;
393 /* Allocate page dirty bitmap if needed */
394 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
395 unsigned dirty_bytes
= ALIGN(npages
, BITS_PER_LONG
) / 8;
397 new.dirty_bitmap
= vmalloc(dirty_bytes
);
398 if (!new.dirty_bitmap
)
400 memset(new.dirty_bitmap
, 0, dirty_bytes
);
403 if (mem
->slot
>= kvm
->nmemslots
)
404 kvm
->nmemslots
= mem
->slot
+ 1;
408 r
= kvm_arch_set_memory_region(kvm
, mem
, old
, user_alloc
);
414 kvm_free_physmem_slot(&old
, &new);
418 kvm_free_physmem_slot(&new, &old
);
423 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
425 int kvm_set_memory_region(struct kvm
*kvm
,
426 struct kvm_userspace_memory_region
*mem
,
431 down_write(&kvm
->slots_lock
);
432 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
433 up_write(&kvm
->slots_lock
);
436 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
438 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
440 kvm_userspace_memory_region
*mem
,
443 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
445 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
448 int kvm_get_dirty_log(struct kvm
*kvm
,
449 struct kvm_dirty_log
*log
, int *is_dirty
)
451 struct kvm_memory_slot
*memslot
;
454 unsigned long any
= 0;
457 if (log
->slot
>= KVM_MEMORY_SLOTS
)
460 memslot
= &kvm
->memslots
[log
->slot
];
462 if (!memslot
->dirty_bitmap
)
465 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
467 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
468 any
= memslot
->dirty_bitmap
[i
];
471 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
482 int is_error_page(struct page
*page
)
484 return page
== bad_page
;
486 EXPORT_SYMBOL_GPL(is_error_page
);
488 int is_error_pfn(pfn_t pfn
)
490 return pfn
== bad_pfn
;
492 EXPORT_SYMBOL_GPL(is_error_pfn
);
494 static inline unsigned long bad_hva(void)
499 int kvm_is_error_hva(unsigned long addr
)
501 return addr
== bad_hva();
503 EXPORT_SYMBOL_GPL(kvm_is_error_hva
);
505 static struct kvm_memory_slot
*__gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
509 for (i
= 0; i
< kvm
->nmemslots
; ++i
) {
510 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
512 if (gfn
>= memslot
->base_gfn
513 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
519 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
521 gfn
= unalias_gfn(kvm
, gfn
);
522 return __gfn_to_memslot(kvm
, gfn
);
525 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
529 gfn
= unalias_gfn(kvm
, gfn
);
530 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
531 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
533 if (gfn
>= memslot
->base_gfn
534 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
539 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
541 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
543 struct kvm_memory_slot
*slot
;
545 gfn
= unalias_gfn(kvm
, gfn
);
546 slot
= __gfn_to_memslot(kvm
, gfn
);
549 return (slot
->userspace_addr
+ (gfn
- slot
->base_gfn
) * PAGE_SIZE
);
551 EXPORT_SYMBOL_GPL(gfn_to_hva
);
554 * Requires current->mm->mmap_sem to be held
556 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
558 struct page
*page
[1];
565 addr
= gfn_to_hva(kvm
, gfn
);
566 if (kvm_is_error_hva(addr
)) {
568 return page_to_pfn(bad_page
);
571 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 1, page
,
574 if (unlikely(npages
!= 1)) {
575 struct vm_area_struct
*vma
;
577 vma
= find_vma(current
->mm
, addr
);
578 if (vma
== NULL
|| addr
< vma
->vm_start
||
579 !(vma
->vm_flags
& VM_PFNMAP
)) {
581 return page_to_pfn(bad_page
);
584 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
585 BUG_ON(pfn_valid(pfn
));
587 pfn
= page_to_pfn(page
[0]);
592 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
594 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
598 pfn
= gfn_to_pfn(kvm
, gfn
);
600 return pfn_to_page(pfn
);
602 WARN_ON(!pfn_valid(pfn
));
608 EXPORT_SYMBOL_GPL(gfn_to_page
);
610 void kvm_release_page_clean(struct page
*page
)
612 kvm_release_pfn_clean(page_to_pfn(page
));
614 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
616 void kvm_release_pfn_clean(pfn_t pfn
)
619 put_page(pfn_to_page(pfn
));
621 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
623 void kvm_release_page_dirty(struct page
*page
)
625 kvm_release_pfn_dirty(page_to_pfn(page
));
627 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
629 void kvm_release_pfn_dirty(pfn_t pfn
)
631 kvm_set_pfn_dirty(pfn
);
632 kvm_release_pfn_clean(pfn
);
634 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
636 void kvm_set_page_dirty(struct page
*page
)
638 kvm_set_pfn_dirty(page_to_pfn(page
));
640 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
642 void kvm_set_pfn_dirty(pfn_t pfn
)
644 if (pfn_valid(pfn
)) {
645 struct page
*page
= pfn_to_page(pfn
);
646 if (!PageReserved(page
))
650 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
652 void kvm_set_pfn_accessed(pfn_t pfn
)
655 mark_page_accessed(pfn_to_page(pfn
));
657 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
659 void kvm_get_pfn(pfn_t pfn
)
662 get_page(pfn_to_page(pfn
));
664 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
666 static int next_segment(unsigned long len
, int offset
)
668 if (len
> PAGE_SIZE
- offset
)
669 return PAGE_SIZE
- offset
;
674 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
680 addr
= gfn_to_hva(kvm
, gfn
);
681 if (kvm_is_error_hva(addr
))
683 r
= copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
688 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
690 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
692 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
694 int offset
= offset_in_page(gpa
);
697 while ((seg
= next_segment(len
, offset
)) != 0) {
698 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
708 EXPORT_SYMBOL_GPL(kvm_read_guest
);
710 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
715 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
716 int offset
= offset_in_page(gpa
);
718 addr
= gfn_to_hva(kvm
, gfn
);
719 if (kvm_is_error_hva(addr
))
722 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
728 EXPORT_SYMBOL(kvm_read_guest_atomic
);
730 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
736 addr
= gfn_to_hva(kvm
, gfn
);
737 if (kvm_is_error_hva(addr
))
739 r
= copy_to_user((void __user
*)addr
+ offset
, data
, len
);
742 mark_page_dirty(kvm
, gfn
);
745 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
747 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
750 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
752 int offset
= offset_in_page(gpa
);
755 while ((seg
= next_segment(len
, offset
)) != 0) {
756 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
767 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
769 return kvm_write_guest_page(kvm
, gfn
, empty_zero_page
, offset
, len
);
771 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
773 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
775 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
777 int offset
= offset_in_page(gpa
);
780 while ((seg
= next_segment(len
, offset
)) != 0) {
781 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
790 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
792 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
794 struct kvm_memory_slot
*memslot
;
796 gfn
= unalias_gfn(kvm
, gfn
);
797 memslot
= __gfn_to_memslot(kvm
, gfn
);
798 if (memslot
&& memslot
->dirty_bitmap
) {
799 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
802 if (!test_bit(rel_gfn
, memslot
->dirty_bitmap
))
803 set_bit(rel_gfn
, memslot
->dirty_bitmap
);
808 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
810 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
815 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
817 if (kvm_cpu_has_interrupt(vcpu
))
819 if (kvm_cpu_has_pending_timer(vcpu
))
821 if (kvm_arch_vcpu_runnable(vcpu
))
823 if (signal_pending(current
))
831 finish_wait(&vcpu
->wq
, &wait
);
834 void kvm_resched(struct kvm_vcpu
*vcpu
)
840 EXPORT_SYMBOL_GPL(kvm_resched
);
842 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
844 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
848 page
= virt_to_page(vcpu
->run
);
850 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
851 page
= virt_to_page(vcpu
->arch
.pio_data
);
853 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
854 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
855 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
858 return VM_FAULT_SIGBUS
;
864 static struct vm_operations_struct kvm_vcpu_vm_ops
= {
865 .fault
= kvm_vcpu_fault
,
868 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
870 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
874 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
876 struct kvm_vcpu
*vcpu
= filp
->private_data
;
878 kvm_put_kvm(vcpu
->kvm
);
882 static const struct file_operations kvm_vcpu_fops
= {
883 .release
= kvm_vcpu_release
,
884 .unlocked_ioctl
= kvm_vcpu_ioctl
,
885 .compat_ioctl
= kvm_vcpu_ioctl
,
886 .mmap
= kvm_vcpu_mmap
,
890 * Allocates an inode for the vcpu.
892 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
894 int fd
= anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
);
896 kvm_put_kvm(vcpu
->kvm
);
901 * Creates some virtual cpus. Good luck creating more than one.
903 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, int n
)
906 struct kvm_vcpu
*vcpu
;
911 vcpu
= kvm_arch_vcpu_create(kvm
, n
);
913 return PTR_ERR(vcpu
);
915 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
917 r
= kvm_arch_vcpu_setup(vcpu
);
921 mutex_lock(&kvm
->lock
);
924 mutex_unlock(&kvm
->lock
);
927 kvm
->vcpus
[n
] = vcpu
;
928 mutex_unlock(&kvm
->lock
);
930 /* Now it's all set up, let userspace reach it */
932 r
= create_vcpu_fd(vcpu
);
938 mutex_lock(&kvm
->lock
);
939 kvm
->vcpus
[n
] = NULL
;
940 mutex_unlock(&kvm
->lock
);
942 kvm_arch_vcpu_destroy(vcpu
);
946 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
949 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
950 vcpu
->sigset_active
= 1;
951 vcpu
->sigset
= *sigset
;
953 vcpu
->sigset_active
= 0;
957 static long kvm_vcpu_ioctl(struct file
*filp
,
958 unsigned int ioctl
, unsigned long arg
)
960 struct kvm_vcpu
*vcpu
= filp
->private_data
;
961 void __user
*argp
= (void __user
*)arg
;
964 if (vcpu
->kvm
->mm
!= current
->mm
)
971 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
974 struct kvm_regs
*kvm_regs
;
977 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
980 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
984 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
992 struct kvm_regs
*kvm_regs
;
995 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
999 if (copy_from_user(kvm_regs
, argp
, sizeof(struct kvm_regs
)))
1001 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1009 case KVM_GET_SREGS
: {
1010 struct kvm_sregs kvm_sregs
;
1012 memset(&kvm_sregs
, 0, sizeof kvm_sregs
);
1013 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, &kvm_sregs
);
1017 if (copy_to_user(argp
, &kvm_sregs
, sizeof kvm_sregs
))
1022 case KVM_SET_SREGS
: {
1023 struct kvm_sregs kvm_sregs
;
1026 if (copy_from_user(&kvm_sregs
, argp
, sizeof kvm_sregs
))
1028 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, &kvm_sregs
);
1034 case KVM_GET_MP_STATE
: {
1035 struct kvm_mp_state mp_state
;
1037 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1041 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
1046 case KVM_SET_MP_STATE
: {
1047 struct kvm_mp_state mp_state
;
1050 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
1052 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
1058 case KVM_TRANSLATE
: {
1059 struct kvm_translation tr
;
1062 if (copy_from_user(&tr
, argp
, sizeof tr
))
1064 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
1068 if (copy_to_user(argp
, &tr
, sizeof tr
))
1073 case KVM_DEBUG_GUEST
: {
1074 struct kvm_debug_guest dbg
;
1077 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
1079 r
= kvm_arch_vcpu_ioctl_debug_guest(vcpu
, &dbg
);
1085 case KVM_SET_SIGNAL_MASK
: {
1086 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
1087 struct kvm_signal_mask kvm_sigmask
;
1088 sigset_t sigset
, *p
;
1093 if (copy_from_user(&kvm_sigmask
, argp
,
1094 sizeof kvm_sigmask
))
1097 if (kvm_sigmask
.len
!= sizeof sigset
)
1100 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
1105 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
1111 memset(&fpu
, 0, sizeof fpu
);
1112 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, &fpu
);
1116 if (copy_to_user(argp
, &fpu
, sizeof fpu
))
1125 if (copy_from_user(&fpu
, argp
, sizeof fpu
))
1127 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, &fpu
);
1134 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1140 static long kvm_vm_ioctl(struct file
*filp
,
1141 unsigned int ioctl
, unsigned long arg
)
1143 struct kvm
*kvm
= filp
->private_data
;
1144 void __user
*argp
= (void __user
*)arg
;
1147 if (kvm
->mm
!= current
->mm
)
1150 case KVM_CREATE_VCPU
:
1151 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
1155 case KVM_SET_USER_MEMORY_REGION
: {
1156 struct kvm_userspace_memory_region kvm_userspace_mem
;
1159 if (copy_from_user(&kvm_userspace_mem
, argp
,
1160 sizeof kvm_userspace_mem
))
1163 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 1);
1168 case KVM_GET_DIRTY_LOG
: {
1169 struct kvm_dirty_log log
;
1172 if (copy_from_user(&log
, argp
, sizeof log
))
1174 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
1179 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1180 case KVM_REGISTER_COALESCED_MMIO
: {
1181 struct kvm_coalesced_mmio_zone zone
;
1183 if (copy_from_user(&zone
, argp
, sizeof zone
))
1186 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
1192 case KVM_UNREGISTER_COALESCED_MMIO
: {
1193 struct kvm_coalesced_mmio_zone zone
;
1195 if (copy_from_user(&zone
, argp
, sizeof zone
))
1198 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
1206 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
1212 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1214 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1217 if (!kvm_is_visible_gfn(kvm
, vmf
->pgoff
))
1218 return VM_FAULT_SIGBUS
;
1219 page
= gfn_to_page(kvm
, vmf
->pgoff
);
1220 if (is_error_page(page
)) {
1221 kvm_release_page_clean(page
);
1222 return VM_FAULT_SIGBUS
;
1228 static struct vm_operations_struct kvm_vm_vm_ops
= {
1229 .fault
= kvm_vm_fault
,
1232 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1234 vma
->vm_ops
= &kvm_vm_vm_ops
;
1238 static const struct file_operations kvm_vm_fops
= {
1239 .release
= kvm_vm_release
,
1240 .unlocked_ioctl
= kvm_vm_ioctl
,
1241 .compat_ioctl
= kvm_vm_ioctl
,
1242 .mmap
= kvm_vm_mmap
,
1245 static int kvm_dev_ioctl_create_vm(void)
1250 kvm
= kvm_create_vm();
1252 return PTR_ERR(kvm
);
1253 fd
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
);
1260 static long kvm_dev_ioctl(struct file
*filp
,
1261 unsigned int ioctl
, unsigned long arg
)
1266 case KVM_GET_API_VERSION
:
1270 r
= KVM_API_VERSION
;
1276 r
= kvm_dev_ioctl_create_vm();
1278 case KVM_CHECK_EXTENSION
:
1279 r
= kvm_dev_ioctl_check_extension(arg
);
1281 case KVM_GET_VCPU_MMAP_SIZE
:
1285 r
= PAGE_SIZE
; /* struct kvm_run */
1287 r
+= PAGE_SIZE
; /* pio data page */
1289 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1290 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
1293 case KVM_TRACE_ENABLE
:
1294 case KVM_TRACE_PAUSE
:
1295 case KVM_TRACE_DISABLE
:
1296 r
= kvm_trace_ioctl(ioctl
, arg
);
1299 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
1305 static struct file_operations kvm_chardev_ops
= {
1306 .unlocked_ioctl
= kvm_dev_ioctl
,
1307 .compat_ioctl
= kvm_dev_ioctl
,
1310 static struct miscdevice kvm_dev
= {
1316 static void hardware_enable(void *junk
)
1318 int cpu
= raw_smp_processor_id();
1320 if (cpu_isset(cpu
, cpus_hardware_enabled
))
1322 cpu_set(cpu
, cpus_hardware_enabled
);
1323 kvm_arch_hardware_enable(NULL
);
1326 static void hardware_disable(void *junk
)
1328 int cpu
= raw_smp_processor_id();
1330 if (!cpu_isset(cpu
, cpus_hardware_enabled
))
1332 cpu_clear(cpu
, cpus_hardware_enabled
);
1333 kvm_arch_hardware_disable(NULL
);
1336 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
1341 val
&= ~CPU_TASKS_FROZEN
;
1344 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
1346 hardware_disable(NULL
);
1348 case CPU_UP_CANCELED
:
1349 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
1351 smp_call_function_single(cpu
, hardware_disable
, NULL
, 1);
1354 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
1356 smp_call_function_single(cpu
, hardware_enable
, NULL
, 1);
1363 asmlinkage
void kvm_handle_fault_on_reboot(void)
1366 /* spin while reset goes on */
1369 /* Fault while not rebooting. We want the trace. */
1372 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot
);
1374 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
1377 if (val
== SYS_RESTART
) {
1379 * Some (well, at least mine) BIOSes hang on reboot if
1382 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
1383 kvm_rebooting
= true;
1384 on_each_cpu(hardware_disable
, NULL
, 1);
1389 static struct notifier_block kvm_reboot_notifier
= {
1390 .notifier_call
= kvm_reboot
,
1394 void kvm_io_bus_init(struct kvm_io_bus
*bus
)
1396 memset(bus
, 0, sizeof(*bus
));
1399 void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
1403 for (i
= 0; i
< bus
->dev_count
; i
++) {
1404 struct kvm_io_device
*pos
= bus
->devs
[i
];
1406 kvm_iodevice_destructor(pos
);
1410 struct kvm_io_device
*kvm_io_bus_find_dev(struct kvm_io_bus
*bus
,
1411 gpa_t addr
, int len
, int is_write
)
1415 for (i
= 0; i
< bus
->dev_count
; i
++) {
1416 struct kvm_io_device
*pos
= bus
->devs
[i
];
1418 if (pos
->in_range(pos
, addr
, len
, is_write
))
1425 void kvm_io_bus_register_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
)
1427 BUG_ON(bus
->dev_count
> (NR_IOBUS_DEVS
-1));
1429 bus
->devs
[bus
->dev_count
++] = dev
;
1432 static struct notifier_block kvm_cpu_notifier
= {
1433 .notifier_call
= kvm_cpu_hotplug
,
1434 .priority
= 20, /* must be > scheduler priority */
1437 static int vm_stat_get(void *_offset
, u64
*val
)
1439 unsigned offset
= (long)_offset
;
1443 spin_lock(&kvm_lock
);
1444 list_for_each_entry(kvm
, &vm_list
, vm_list
)
1445 *val
+= *(u32
*)((void *)kvm
+ offset
);
1446 spin_unlock(&kvm_lock
);
1450 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
1452 static int vcpu_stat_get(void *_offset
, u64
*val
)
1454 unsigned offset
= (long)_offset
;
1456 struct kvm_vcpu
*vcpu
;
1460 spin_lock(&kvm_lock
);
1461 list_for_each_entry(kvm
, &vm_list
, vm_list
)
1462 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
1463 vcpu
= kvm
->vcpus
[i
];
1465 *val
+= *(u32
*)((void *)vcpu
+ offset
);
1467 spin_unlock(&kvm_lock
);
1471 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
1473 static struct file_operations
*stat_fops
[] = {
1474 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
1475 [KVM_STAT_VM
] = &vm_stat_fops
,
1478 static void kvm_init_debug(void)
1480 struct kvm_stats_debugfs_item
*p
;
1482 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
1483 for (p
= debugfs_entries
; p
->name
; ++p
)
1484 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
1485 (void *)(long)p
->offset
,
1486 stat_fops
[p
->kind
]);
1489 static void kvm_exit_debug(void)
1491 struct kvm_stats_debugfs_item
*p
;
1493 for (p
= debugfs_entries
; p
->name
; ++p
)
1494 debugfs_remove(p
->dentry
);
1495 debugfs_remove(kvm_debugfs_dir
);
1498 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
1500 hardware_disable(NULL
);
1504 static int kvm_resume(struct sys_device
*dev
)
1506 hardware_enable(NULL
);
1510 static struct sysdev_class kvm_sysdev_class
= {
1512 .suspend
= kvm_suspend
,
1513 .resume
= kvm_resume
,
1516 static struct sys_device kvm_sysdev
= {
1518 .cls
= &kvm_sysdev_class
,
1521 struct page
*bad_page
;
1525 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
1527 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
1530 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
1532 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
1534 kvm_arch_vcpu_load(vcpu
, cpu
);
1537 static void kvm_sched_out(struct preempt_notifier
*pn
,
1538 struct task_struct
*next
)
1540 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
1542 kvm_arch_vcpu_put(vcpu
);
1545 int kvm_init(void *opaque
, unsigned int vcpu_size
,
1546 struct module
*module
)
1553 r
= kvm_arch_init(opaque
);
1557 bad_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
1559 if (bad_page
== NULL
) {
1564 bad_pfn
= page_to_pfn(bad_page
);
1566 r
= kvm_arch_hardware_setup();
1570 for_each_online_cpu(cpu
) {
1571 smp_call_function_single(cpu
,
1572 kvm_arch_check_processor_compat
,
1578 on_each_cpu(hardware_enable
, NULL
, 1);
1579 r
= register_cpu_notifier(&kvm_cpu_notifier
);
1582 register_reboot_notifier(&kvm_reboot_notifier
);
1584 r
= sysdev_class_register(&kvm_sysdev_class
);
1588 r
= sysdev_register(&kvm_sysdev
);
1592 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1593 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
,
1594 __alignof__(struct kvm_vcpu
),
1596 if (!kvm_vcpu_cache
) {
1601 kvm_chardev_ops
.owner
= module
;
1603 r
= misc_register(&kvm_dev
);
1605 printk(KERN_ERR
"kvm: misc device register failed\n");
1609 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
1610 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
1615 kmem_cache_destroy(kvm_vcpu_cache
);
1617 sysdev_unregister(&kvm_sysdev
);
1619 sysdev_class_unregister(&kvm_sysdev_class
);
1621 unregister_reboot_notifier(&kvm_reboot_notifier
);
1622 unregister_cpu_notifier(&kvm_cpu_notifier
);
1624 on_each_cpu(hardware_disable
, NULL
, 1);
1626 kvm_arch_hardware_unsetup();
1628 __free_page(bad_page
);
1635 EXPORT_SYMBOL_GPL(kvm_init
);
1639 kvm_trace_cleanup();
1640 misc_deregister(&kvm_dev
);
1641 kmem_cache_destroy(kvm_vcpu_cache
);
1642 sysdev_unregister(&kvm_sysdev
);
1643 sysdev_class_unregister(&kvm_sysdev_class
);
1644 unregister_reboot_notifier(&kvm_reboot_notifier
);
1645 unregister_cpu_notifier(&kvm_cpu_notifier
);
1646 on_each_cpu(hardware_disable
, NULL
, 1);
1647 kvm_arch_hardware_unsetup();
1650 __free_page(bad_page
);
1652 EXPORT_SYMBOL_GPL(kvm_exit
);