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.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.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>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
51 #include <asm/processor.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
57 #include "coalesced_mmio.h"
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/kvm.h>
63 MODULE_AUTHOR("Qumranet");
64 MODULE_LICENSE("GPL");
69 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
72 DEFINE_SPINLOCK(kvm_lock
);
75 static cpumask_var_t cpus_hardware_enabled
;
76 static int kvm_usage_count
= 0;
77 static atomic_t hardware_enable_failed
;
79 struct kmem_cache
*kvm_vcpu_cache
;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
82 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
84 struct dentry
*kvm_debugfs_dir
;
86 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
88 static int hardware_enable_all(void);
89 static void hardware_disable_all(void);
91 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
93 static bool kvm_rebooting
;
95 static bool largepages_enabled
= true;
97 static struct page
*hwpoison_page
;
98 static pfn_t hwpoison_pfn
;
100 static struct page
*fault_page
;
101 static pfn_t fault_pfn
;
103 inline int kvm_is_mmio_pfn(pfn_t pfn
)
105 if (pfn_valid(pfn
)) {
106 struct page
*page
= compound_head(pfn_to_page(pfn
));
107 return PageReserved(page
);
114 * Switches to specified vcpu, until a matching vcpu_put()
116 void vcpu_load(struct kvm_vcpu
*vcpu
)
120 mutex_lock(&vcpu
->mutex
);
122 preempt_notifier_register(&vcpu
->preempt_notifier
);
123 kvm_arch_vcpu_load(vcpu
, cpu
);
127 void vcpu_put(struct kvm_vcpu
*vcpu
)
130 kvm_arch_vcpu_put(vcpu
);
131 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
133 mutex_unlock(&vcpu
->mutex
);
136 static void ack_flush(void *_completed
)
140 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
145 struct kvm_vcpu
*vcpu
;
147 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
149 raw_spin_lock(&kvm
->requests_lock
);
150 me
= smp_processor_id();
151 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
152 if (kvm_make_check_request(req
, vcpu
))
155 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
)
156 cpumask_set_cpu(cpu
, cpus
);
158 if (unlikely(cpus
== NULL
))
159 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
160 else if (!cpumask_empty(cpus
))
161 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
164 raw_spin_unlock(&kvm
->requests_lock
);
165 free_cpumask_var(cpus
);
169 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
171 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
172 ++kvm
->stat
.remote_tlb_flush
;
175 void kvm_reload_remote_mmus(struct kvm
*kvm
)
177 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
180 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
185 mutex_init(&vcpu
->mutex
);
189 init_waitqueue_head(&vcpu
->wq
);
190 kvm_async_pf_vcpu_init(vcpu
);
192 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
197 vcpu
->run
= page_address(page
);
199 r
= kvm_arch_vcpu_init(vcpu
);
205 free_page((unsigned long)vcpu
->run
);
209 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
211 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
213 kvm_arch_vcpu_uninit(vcpu
);
214 free_page((unsigned long)vcpu
->run
);
216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
219 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
221 return container_of(mn
, struct kvm
, mmu_notifier
);
224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
225 struct mm_struct
*mm
,
226 unsigned long address
)
228 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
229 int need_tlb_flush
, idx
;
232 * When ->invalidate_page runs, the linux pte has been zapped
233 * already but the page is still allocated until
234 * ->invalidate_page returns. So if we increase the sequence
235 * here the kvm page fault will notice if the spte can't be
236 * established because the page is going to be freed. If
237 * instead the kvm page fault establishes the spte before
238 * ->invalidate_page runs, kvm_unmap_hva will release it
241 * The sequence increase only need to be seen at spin_unlock
242 * time, and not at spin_lock time.
244 * Increasing the sequence after the spin_unlock would be
245 * unsafe because the kvm page fault could then establish the
246 * pte after kvm_unmap_hva returned, without noticing the page
247 * is going to be freed.
249 idx
= srcu_read_lock(&kvm
->srcu
);
250 spin_lock(&kvm
->mmu_lock
);
251 kvm
->mmu_notifier_seq
++;
252 need_tlb_flush
= kvm_unmap_hva(kvm
, address
);
253 spin_unlock(&kvm
->mmu_lock
);
254 srcu_read_unlock(&kvm
->srcu
, idx
);
256 /* we've to flush the tlb before the pages can be freed */
258 kvm_flush_remote_tlbs(kvm
);
262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
263 struct mm_struct
*mm
,
264 unsigned long address
,
267 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
270 idx
= srcu_read_lock(&kvm
->srcu
);
271 spin_lock(&kvm
->mmu_lock
);
272 kvm
->mmu_notifier_seq
++;
273 kvm_set_spte_hva(kvm
, address
, pte
);
274 spin_unlock(&kvm
->mmu_lock
);
275 srcu_read_unlock(&kvm
->srcu
, idx
);
278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
279 struct mm_struct
*mm
,
283 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
284 int need_tlb_flush
= 0, idx
;
286 idx
= srcu_read_lock(&kvm
->srcu
);
287 spin_lock(&kvm
->mmu_lock
);
289 * The count increase must become visible at unlock time as no
290 * spte can be established without taking the mmu_lock and
291 * count is also read inside the mmu_lock critical section.
293 kvm
->mmu_notifier_count
++;
294 for (; start
< end
; start
+= PAGE_SIZE
)
295 need_tlb_flush
|= kvm_unmap_hva(kvm
, start
);
296 spin_unlock(&kvm
->mmu_lock
);
297 srcu_read_unlock(&kvm
->srcu
, idx
);
299 /* we've to flush the tlb before the pages can be freed */
301 kvm_flush_remote_tlbs(kvm
);
304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
305 struct mm_struct
*mm
,
309 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
311 spin_lock(&kvm
->mmu_lock
);
313 * This sequence increase will notify the kvm page fault that
314 * the page that is going to be mapped in the spte could have
317 kvm
->mmu_notifier_seq
++;
319 * The above sequence increase must be visible before the
320 * below count decrease but both values are read by the kvm
321 * page fault under mmu_lock spinlock so we don't need to add
322 * a smb_wmb() here in between the two.
324 kvm
->mmu_notifier_count
--;
325 spin_unlock(&kvm
->mmu_lock
);
327 BUG_ON(kvm
->mmu_notifier_count
< 0);
330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
331 struct mm_struct
*mm
,
332 unsigned long address
)
334 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
337 idx
= srcu_read_lock(&kvm
->srcu
);
338 spin_lock(&kvm
->mmu_lock
);
339 young
= kvm_age_hva(kvm
, address
);
340 spin_unlock(&kvm
->mmu_lock
);
341 srcu_read_unlock(&kvm
->srcu
, idx
);
344 kvm_flush_remote_tlbs(kvm
);
349 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
350 struct mm_struct
*mm
)
352 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
355 idx
= srcu_read_lock(&kvm
->srcu
);
356 kvm_arch_flush_shadow(kvm
);
357 srcu_read_unlock(&kvm
->srcu
, idx
);
360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
361 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
362 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
363 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
364 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
365 .change_pte
= kvm_mmu_notifier_change_pte
,
366 .release
= kvm_mmu_notifier_release
,
369 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
371 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
372 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
375 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
377 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
384 static struct kvm
*kvm_create_vm(void)
387 struct kvm
*kvm
= kvm_arch_create_vm();
392 r
= hardware_enable_all();
394 goto out_err_nodisable
;
396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
397 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
398 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
402 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
405 if (init_srcu_struct(&kvm
->srcu
))
407 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
408 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
410 if (!kvm
->buses
[i
]) {
411 cleanup_srcu_struct(&kvm
->srcu
);
416 r
= kvm_init_mmu_notifier(kvm
);
418 cleanup_srcu_struct(&kvm
->srcu
);
422 kvm
->mm
= current
->mm
;
423 atomic_inc(&kvm
->mm
->mm_count
);
424 spin_lock_init(&kvm
->mmu_lock
);
425 raw_spin_lock_init(&kvm
->requests_lock
);
426 kvm_eventfd_init(kvm
);
427 mutex_init(&kvm
->lock
);
428 mutex_init(&kvm
->irq_lock
);
429 mutex_init(&kvm
->slots_lock
);
430 atomic_set(&kvm
->users_count
, 1);
431 spin_lock(&kvm_lock
);
432 list_add(&kvm
->vm_list
, &vm_list
);
433 spin_unlock(&kvm_lock
);
438 hardware_disable_all();
440 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
441 kfree(kvm
->buses
[i
]);
442 kfree(kvm
->memslots
);
448 * Free any memory in @free but not in @dont.
450 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
451 struct kvm_memory_slot
*dont
)
455 if (!dont
|| free
->rmap
!= dont
->rmap
)
458 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
459 vfree(free
->dirty_bitmap
);
462 for (i
= 0; i
< KVM_NR_PAGE_SIZES
- 1; ++i
) {
463 if (!dont
|| free
->lpage_info
[i
] != dont
->lpage_info
[i
]) {
464 vfree(free
->lpage_info
[i
]);
465 free
->lpage_info
[i
] = NULL
;
470 free
->dirty_bitmap
= NULL
;
474 void kvm_free_physmem(struct kvm
*kvm
)
477 struct kvm_memslots
*slots
= kvm
->memslots
;
479 for (i
= 0; i
< slots
->nmemslots
; ++i
)
480 kvm_free_physmem_slot(&slots
->memslots
[i
], NULL
);
482 kfree(kvm
->memslots
);
485 static void kvm_destroy_vm(struct kvm
*kvm
)
488 struct mm_struct
*mm
= kvm
->mm
;
490 kvm_arch_sync_events(kvm
);
491 spin_lock(&kvm_lock
);
492 list_del(&kvm
->vm_list
);
493 spin_unlock(&kvm_lock
);
494 kvm_free_irq_routing(kvm
);
495 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
496 kvm_io_bus_destroy(kvm
->buses
[i
]);
497 kvm_coalesced_mmio_free(kvm
);
498 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
499 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
501 kvm_arch_flush_shadow(kvm
);
503 kvm_arch_destroy_vm(kvm
);
504 hardware_disable_all();
508 void kvm_get_kvm(struct kvm
*kvm
)
510 atomic_inc(&kvm
->users_count
);
512 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
514 void kvm_put_kvm(struct kvm
*kvm
)
516 if (atomic_dec_and_test(&kvm
->users_count
))
519 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
522 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
524 struct kvm
*kvm
= filp
->private_data
;
526 kvm_irqfd_release(kvm
);
533 * Allocate some memory and give it an address in the guest physical address
536 * Discontiguous memory is allowed, mostly for framebuffers.
538 * Must be called holding mmap_sem for write.
540 int __kvm_set_memory_region(struct kvm
*kvm
,
541 struct kvm_userspace_memory_region
*mem
,
544 int r
, flush_shadow
= 0;
546 unsigned long npages
;
548 struct kvm_memory_slot
*memslot
;
549 struct kvm_memory_slot old
, new;
550 struct kvm_memslots
*slots
, *old_memslots
;
553 /* General sanity checks */
554 if (mem
->memory_size
& (PAGE_SIZE
- 1))
556 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
558 if (user_alloc
&& (mem
->userspace_addr
& (PAGE_SIZE
- 1)))
560 if (mem
->slot
>= KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
)
562 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
565 memslot
= &kvm
->memslots
->memslots
[mem
->slot
];
566 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
567 npages
= mem
->memory_size
>> PAGE_SHIFT
;
570 if (npages
> KVM_MEM_MAX_NR_PAGES
)
574 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
576 new = old
= *memslot
;
579 new.base_gfn
= base_gfn
;
581 new.flags
= mem
->flags
;
583 /* Disallow changing a memory slot's size. */
585 if (npages
&& old
.npages
&& npages
!= old
.npages
)
588 /* Check for overlaps */
590 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
591 struct kvm_memory_slot
*s
= &kvm
->memslots
->memslots
[i
];
593 if (s
== memslot
|| !s
->npages
)
595 if (!((base_gfn
+ npages
<= s
->base_gfn
) ||
596 (base_gfn
>= s
->base_gfn
+ s
->npages
)))
600 /* Free page dirty bitmap if unneeded */
601 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
602 new.dirty_bitmap
= NULL
;
606 /* Allocate if a slot is being created */
608 if (npages
&& !new.rmap
) {
609 new.rmap
= vmalloc(npages
* sizeof(*new.rmap
));
614 memset(new.rmap
, 0, npages
* sizeof(*new.rmap
));
616 new.user_alloc
= user_alloc
;
617 new.userspace_addr
= mem
->userspace_addr
;
622 for (i
= 0; i
< KVM_NR_PAGE_SIZES
- 1; ++i
) {
628 /* Avoid unused variable warning if no large pages */
631 if (new.lpage_info
[i
])
634 lpages
= 1 + ((base_gfn
+ npages
- 1)
635 >> KVM_HPAGE_GFN_SHIFT(level
));
636 lpages
-= base_gfn
>> KVM_HPAGE_GFN_SHIFT(level
);
638 new.lpage_info
[i
] = vmalloc(lpages
* sizeof(*new.lpage_info
[i
]));
640 if (!new.lpage_info
[i
])
643 memset(new.lpage_info
[i
], 0,
644 lpages
* sizeof(*new.lpage_info
[i
]));
646 if (base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
647 new.lpage_info
[i
][0].write_count
= 1;
648 if ((base_gfn
+npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
649 new.lpage_info
[i
][lpages
- 1].write_count
= 1;
650 ugfn
= new.userspace_addr
>> PAGE_SHIFT
;
652 * If the gfn and userspace address are not aligned wrt each
653 * other, or if explicitly asked to, disable large page
654 * support for this slot
656 if ((base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
658 for (j
= 0; j
< lpages
; ++j
)
659 new.lpage_info
[i
][j
].write_count
= 1;
664 /* Allocate page dirty bitmap if needed */
665 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
666 unsigned long dirty_bytes
= kvm_dirty_bitmap_bytes(&new);
668 new.dirty_bitmap
= vmalloc(dirty_bytes
);
669 if (!new.dirty_bitmap
)
671 memset(new.dirty_bitmap
, 0, dirty_bytes
);
672 /* destroy any largepage mappings for dirty tracking */
676 #else /* not defined CONFIG_S390 */
677 new.user_alloc
= user_alloc
;
679 new.userspace_addr
= mem
->userspace_addr
;
680 #endif /* not defined CONFIG_S390 */
684 slots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
687 memcpy(slots
, kvm
->memslots
, sizeof(struct kvm_memslots
));
688 if (mem
->slot
>= slots
->nmemslots
)
689 slots
->nmemslots
= mem
->slot
+ 1;
691 slots
->memslots
[mem
->slot
].flags
|= KVM_MEMSLOT_INVALID
;
693 old_memslots
= kvm
->memslots
;
694 rcu_assign_pointer(kvm
->memslots
, slots
);
695 synchronize_srcu_expedited(&kvm
->srcu
);
696 /* From this point no new shadow pages pointing to a deleted
697 * memslot will be created.
699 * validation of sp->gfn happens in:
700 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
701 * - kvm_is_visible_gfn (mmu_check_roots)
703 kvm_arch_flush_shadow(kvm
);
707 r
= kvm_arch_prepare_memory_region(kvm
, &new, old
, mem
, user_alloc
);
711 /* map the pages in iommu page table */
713 r
= kvm_iommu_map_pages(kvm
, &new);
719 slots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
722 memcpy(slots
, kvm
->memslots
, sizeof(struct kvm_memslots
));
723 if (mem
->slot
>= slots
->nmemslots
)
724 slots
->nmemslots
= mem
->slot
+ 1;
727 /* actual memory is freed via old in kvm_free_physmem_slot below */
730 new.dirty_bitmap
= NULL
;
731 for (i
= 0; i
< KVM_NR_PAGE_SIZES
- 1; ++i
)
732 new.lpage_info
[i
] = NULL
;
735 slots
->memslots
[mem
->slot
] = new;
736 old_memslots
= kvm
->memslots
;
737 rcu_assign_pointer(kvm
->memslots
, slots
);
738 synchronize_srcu_expedited(&kvm
->srcu
);
740 kvm_arch_commit_memory_region(kvm
, mem
, old
, user_alloc
);
742 kvm_free_physmem_slot(&old
, &new);
746 kvm_arch_flush_shadow(kvm
);
751 kvm_free_physmem_slot(&new, &old
);
756 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
758 int kvm_set_memory_region(struct kvm
*kvm
,
759 struct kvm_userspace_memory_region
*mem
,
764 mutex_lock(&kvm
->slots_lock
);
765 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
766 mutex_unlock(&kvm
->slots_lock
);
769 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
771 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
773 kvm_userspace_memory_region
*mem
,
776 if (mem
->slot
>= KVM_MEMORY_SLOTS
)
778 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
781 int kvm_get_dirty_log(struct kvm
*kvm
,
782 struct kvm_dirty_log
*log
, int *is_dirty
)
784 struct kvm_memory_slot
*memslot
;
787 unsigned long any
= 0;
790 if (log
->slot
>= KVM_MEMORY_SLOTS
)
793 memslot
= &kvm
->memslots
->memslots
[log
->slot
];
795 if (!memslot
->dirty_bitmap
)
798 n
= kvm_dirty_bitmap_bytes(memslot
);
800 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
801 any
= memslot
->dirty_bitmap
[i
];
804 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
815 void kvm_disable_largepages(void)
817 largepages_enabled
= false;
819 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
821 int is_error_page(struct page
*page
)
823 return page
== bad_page
|| page
== hwpoison_page
|| page
== fault_page
;
825 EXPORT_SYMBOL_GPL(is_error_page
);
827 int is_error_pfn(pfn_t pfn
)
829 return pfn
== bad_pfn
|| pfn
== hwpoison_pfn
|| pfn
== fault_pfn
;
831 EXPORT_SYMBOL_GPL(is_error_pfn
);
833 int is_hwpoison_pfn(pfn_t pfn
)
835 return pfn
== hwpoison_pfn
;
837 EXPORT_SYMBOL_GPL(is_hwpoison_pfn
);
839 int is_fault_pfn(pfn_t pfn
)
841 return pfn
== fault_pfn
;
843 EXPORT_SYMBOL_GPL(is_fault_pfn
);
845 static inline unsigned long bad_hva(void)
850 int kvm_is_error_hva(unsigned long addr
)
852 return addr
== bad_hva();
854 EXPORT_SYMBOL_GPL(kvm_is_error_hva
);
856 static struct kvm_memory_slot
*__gfn_to_memslot(struct kvm_memslots
*slots
,
861 for (i
= 0; i
< slots
->nmemslots
; ++i
) {
862 struct kvm_memory_slot
*memslot
= &slots
->memslots
[i
];
864 if (gfn
>= memslot
->base_gfn
865 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
871 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
873 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
875 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
877 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
880 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
882 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
883 struct kvm_memory_slot
*memslot
= &slots
->memslots
[i
];
885 if (memslot
->flags
& KVM_MEMSLOT_INVALID
)
888 if (gfn
>= memslot
->base_gfn
889 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
894 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
896 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
898 struct vm_area_struct
*vma
;
899 unsigned long addr
, size
;
903 addr
= gfn_to_hva(kvm
, gfn
);
904 if (kvm_is_error_hva(addr
))
907 down_read(¤t
->mm
->mmap_sem
);
908 vma
= find_vma(current
->mm
, addr
);
912 size
= vma_kernel_pagesize(vma
);
915 up_read(¤t
->mm
->mmap_sem
);
920 int memslot_id(struct kvm
*kvm
, gfn_t gfn
)
923 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
924 struct kvm_memory_slot
*memslot
= NULL
;
926 for (i
= 0; i
< slots
->nmemslots
; ++i
) {
927 memslot
= &slots
->memslots
[i
];
929 if (gfn
>= memslot
->base_gfn
930 && gfn
< memslot
->base_gfn
+ memslot
->npages
)
934 return memslot
- slots
->memslots
;
937 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
940 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
944 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
946 return gfn_to_hva_memslot(slot
, gfn
);
949 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
951 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
953 EXPORT_SYMBOL_GPL(gfn_to_hva
);
955 static pfn_t
get_fault_pfn(void)
957 get_page(fault_page
);
961 static pfn_t
hva_to_pfn(struct kvm
*kvm
, unsigned long addr
, bool atomic
,
962 bool *async
, bool write_fault
, bool *writable
)
964 struct page
*page
[1];
968 /* we can do it either atomically or asynchronously, not both */
969 BUG_ON(atomic
&& async
);
971 BUG_ON(!write_fault
&& !writable
);
977 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
979 if (unlikely(npages
!= 1) && !atomic
) {
983 *writable
= write_fault
;
985 npages
= get_user_pages_fast(addr
, 1, write_fault
, page
);
987 /* map read fault as writable if possible */
988 if (unlikely(!write_fault
) && npages
== 1) {
989 struct page
*wpage
[1];
991 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1001 if (unlikely(npages
!= 1)) {
1002 struct vm_area_struct
*vma
;
1005 return get_fault_pfn();
1007 down_read(¤t
->mm
->mmap_sem
);
1008 if (is_hwpoison_address(addr
)) {
1009 up_read(¤t
->mm
->mmap_sem
);
1010 get_page(hwpoison_page
);
1011 return page_to_pfn(hwpoison_page
);
1014 vma
= find_vma_intersection(current
->mm
, addr
, addr
+1);
1017 pfn
= get_fault_pfn();
1018 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1019 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1021 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1023 if (async
&& (vma
->vm_flags
& VM_WRITE
))
1025 pfn
= get_fault_pfn();
1027 up_read(¤t
->mm
->mmap_sem
);
1029 pfn
= page_to_pfn(page
[0]);
1034 pfn_t
hva_to_pfn_atomic(struct kvm
*kvm
, unsigned long addr
)
1036 return hva_to_pfn(kvm
, addr
, true, NULL
, true, NULL
);
1038 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic
);
1040 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1041 bool write_fault
, bool *writable
)
1048 addr
= gfn_to_hva(kvm
, gfn
);
1049 if (kvm_is_error_hva(addr
)) {
1051 return page_to_pfn(bad_page
);
1054 return hva_to_pfn(kvm
, addr
, atomic
, async
, write_fault
, writable
);
1057 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1059 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1061 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1063 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1064 bool write_fault
, bool *writable
)
1066 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1068 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1070 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1072 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1074 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1076 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1079 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1081 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1083 pfn_t
gfn_to_pfn_memslot(struct kvm
*kvm
,
1084 struct kvm_memory_slot
*slot
, gfn_t gfn
)
1086 unsigned long addr
= gfn_to_hva_memslot(slot
, gfn
);
1087 return hva_to_pfn(kvm
, addr
, false, NULL
, true, NULL
);
1090 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1096 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1097 if (kvm_is_error_hva(addr
))
1100 if (entry
< nr_pages
)
1103 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1105 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1107 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1111 pfn
= gfn_to_pfn(kvm
, gfn
);
1112 if (!kvm_is_mmio_pfn(pfn
))
1113 return pfn_to_page(pfn
);
1115 WARN_ON(kvm_is_mmio_pfn(pfn
));
1121 EXPORT_SYMBOL_GPL(gfn_to_page
);
1123 void kvm_release_page_clean(struct page
*page
)
1125 kvm_release_pfn_clean(page_to_pfn(page
));
1127 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1129 void kvm_release_pfn_clean(pfn_t pfn
)
1131 if (!kvm_is_mmio_pfn(pfn
))
1132 put_page(pfn_to_page(pfn
));
1134 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1136 void kvm_release_page_dirty(struct page
*page
)
1138 kvm_release_pfn_dirty(page_to_pfn(page
));
1140 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1142 void kvm_release_pfn_dirty(pfn_t pfn
)
1144 kvm_set_pfn_dirty(pfn
);
1145 kvm_release_pfn_clean(pfn
);
1147 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1149 void kvm_set_page_dirty(struct page
*page
)
1151 kvm_set_pfn_dirty(page_to_pfn(page
));
1153 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1155 void kvm_set_pfn_dirty(pfn_t pfn
)
1157 if (!kvm_is_mmio_pfn(pfn
)) {
1158 struct page
*page
= pfn_to_page(pfn
);
1159 if (!PageReserved(page
))
1163 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1165 void kvm_set_pfn_accessed(pfn_t pfn
)
1167 if (!kvm_is_mmio_pfn(pfn
))
1168 mark_page_accessed(pfn_to_page(pfn
));
1170 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1172 void kvm_get_pfn(pfn_t pfn
)
1174 if (!kvm_is_mmio_pfn(pfn
))
1175 get_page(pfn_to_page(pfn
));
1177 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1179 static int next_segment(unsigned long len
, int offset
)
1181 if (len
> PAGE_SIZE
- offset
)
1182 return PAGE_SIZE
- offset
;
1187 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1193 addr
= gfn_to_hva(kvm
, gfn
);
1194 if (kvm_is_error_hva(addr
))
1196 r
= copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1201 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1203 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1205 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1207 int offset
= offset_in_page(gpa
);
1210 while ((seg
= next_segment(len
, offset
)) != 0) {
1211 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1221 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1223 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1228 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1229 int offset
= offset_in_page(gpa
);
1231 addr
= gfn_to_hva(kvm
, gfn
);
1232 if (kvm_is_error_hva(addr
))
1234 pagefault_disable();
1235 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1241 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1243 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1244 int offset
, int len
)
1249 addr
= gfn_to_hva(kvm
, gfn
);
1250 if (kvm_is_error_hva(addr
))
1252 r
= copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1255 mark_page_dirty(kvm
, gfn
);
1258 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1260 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1263 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1265 int offset
= offset_in_page(gpa
);
1268 while ((seg
= next_segment(len
, offset
)) != 0) {
1269 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1280 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1283 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1284 int offset
= offset_in_page(gpa
);
1285 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1288 ghc
->generation
= slots
->generation
;
1289 ghc
->memslot
= __gfn_to_memslot(slots
, gfn
);
1290 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, gfn
, NULL
);
1291 if (!kvm_is_error_hva(ghc
->hva
))
1298 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1300 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1301 void *data
, unsigned long len
)
1303 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1306 if (slots
->generation
!= ghc
->generation
)
1307 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1309 if (kvm_is_error_hva(ghc
->hva
))
1312 r
= copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1315 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1319 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1321 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1323 return kvm_write_guest_page(kvm
, gfn
, empty_zero_page
, offset
, len
);
1325 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1327 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1329 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1331 int offset
= offset_in_page(gpa
);
1334 while ((seg
= next_segment(len
, offset
)) != 0) {
1335 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1344 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1346 void mark_page_dirty_in_slot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
,
1349 if (memslot
&& memslot
->dirty_bitmap
) {
1350 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1352 generic___set_le_bit(rel_gfn
, memslot
->dirty_bitmap
);
1356 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1358 struct kvm_memory_slot
*memslot
;
1360 memslot
= gfn_to_memslot(kvm
, gfn
);
1361 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1365 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1367 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1372 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1374 if (kvm_arch_vcpu_runnable(vcpu
)) {
1375 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1378 if (kvm_cpu_has_pending_timer(vcpu
))
1380 if (signal_pending(current
))
1386 finish_wait(&vcpu
->wq
, &wait
);
1389 void kvm_resched(struct kvm_vcpu
*vcpu
)
1391 if (!need_resched())
1395 EXPORT_SYMBOL_GPL(kvm_resched
);
1397 void kvm_vcpu_on_spin(struct kvm_vcpu
*vcpu
)
1402 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1404 /* Sleep for 100 us, and hope lock-holder got scheduled */
1405 expires
= ktime_add_ns(ktime_get(), 100000UL);
1406 schedule_hrtimeout(&expires
, HRTIMER_MODE_ABS
);
1408 finish_wait(&vcpu
->wq
, &wait
);
1410 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1412 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1414 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1417 if (vmf
->pgoff
== 0)
1418 page
= virt_to_page(vcpu
->run
);
1420 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1421 page
= virt_to_page(vcpu
->arch
.pio_data
);
1423 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1424 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1425 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1428 return VM_FAULT_SIGBUS
;
1434 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1435 .fault
= kvm_vcpu_fault
,
1438 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1440 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1444 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1446 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1448 kvm_put_kvm(vcpu
->kvm
);
1452 static struct file_operations kvm_vcpu_fops
= {
1453 .release
= kvm_vcpu_release
,
1454 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1455 .compat_ioctl
= kvm_vcpu_ioctl
,
1456 .mmap
= kvm_vcpu_mmap
,
1457 .llseek
= noop_llseek
,
1461 * Allocates an inode for the vcpu.
1463 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1465 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
);
1469 * Creates some virtual cpus. Good luck creating more than one.
1471 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1474 struct kvm_vcpu
*vcpu
, *v
;
1476 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1478 return PTR_ERR(vcpu
);
1480 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1482 r
= kvm_arch_vcpu_setup(vcpu
);
1486 mutex_lock(&kvm
->lock
);
1487 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1492 kvm_for_each_vcpu(r
, v
, kvm
)
1493 if (v
->vcpu_id
== id
) {
1498 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1500 /* Now it's all set up, let userspace reach it */
1502 r
= create_vcpu_fd(vcpu
);
1508 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1510 atomic_inc(&kvm
->online_vcpus
);
1512 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1513 if (kvm
->bsp_vcpu_id
== id
)
1514 kvm
->bsp_vcpu
= vcpu
;
1516 mutex_unlock(&kvm
->lock
);
1520 mutex_unlock(&kvm
->lock
);
1521 kvm_arch_vcpu_destroy(vcpu
);
1525 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1528 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1529 vcpu
->sigset_active
= 1;
1530 vcpu
->sigset
= *sigset
;
1532 vcpu
->sigset_active
= 0;
1536 static long kvm_vcpu_ioctl(struct file
*filp
,
1537 unsigned int ioctl
, unsigned long arg
)
1539 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1540 void __user
*argp
= (void __user
*)arg
;
1542 struct kvm_fpu
*fpu
= NULL
;
1543 struct kvm_sregs
*kvm_sregs
= NULL
;
1545 if (vcpu
->kvm
->mm
!= current
->mm
)
1548 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1550 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1551 * so vcpu_load() would break it.
1553 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1554 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1564 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1565 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
1567 case KVM_GET_REGS
: {
1568 struct kvm_regs
*kvm_regs
;
1571 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1574 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1578 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1585 case KVM_SET_REGS
: {
1586 struct kvm_regs
*kvm_regs
;
1589 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1593 if (copy_from_user(kvm_regs
, argp
, sizeof(struct kvm_regs
)))
1595 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1603 case KVM_GET_SREGS
: {
1604 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1608 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1612 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
1617 case KVM_SET_SREGS
: {
1618 kvm_sregs
= kmalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1623 if (copy_from_user(kvm_sregs
, argp
, sizeof(struct kvm_sregs
)))
1625 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
1631 case KVM_GET_MP_STATE
: {
1632 struct kvm_mp_state mp_state
;
1634 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1638 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
1643 case KVM_SET_MP_STATE
: {
1644 struct kvm_mp_state mp_state
;
1647 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
1649 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
1655 case KVM_TRANSLATE
: {
1656 struct kvm_translation tr
;
1659 if (copy_from_user(&tr
, argp
, sizeof tr
))
1661 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
1665 if (copy_to_user(argp
, &tr
, sizeof tr
))
1670 case KVM_SET_GUEST_DEBUG
: {
1671 struct kvm_guest_debug dbg
;
1674 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
1676 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
1682 case KVM_SET_SIGNAL_MASK
: {
1683 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
1684 struct kvm_signal_mask kvm_sigmask
;
1685 sigset_t sigset
, *p
;
1690 if (copy_from_user(&kvm_sigmask
, argp
,
1691 sizeof kvm_sigmask
))
1694 if (kvm_sigmask
.len
!= sizeof sigset
)
1697 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
1702 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
1706 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
1710 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
1714 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
1720 fpu
= kmalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
1725 if (copy_from_user(fpu
, argp
, sizeof(struct kvm_fpu
)))
1727 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
1734 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1743 static long kvm_vm_ioctl(struct file
*filp
,
1744 unsigned int ioctl
, unsigned long arg
)
1746 struct kvm
*kvm
= filp
->private_data
;
1747 void __user
*argp
= (void __user
*)arg
;
1750 if (kvm
->mm
!= current
->mm
)
1753 case KVM_CREATE_VCPU
:
1754 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
1758 case KVM_SET_USER_MEMORY_REGION
: {
1759 struct kvm_userspace_memory_region kvm_userspace_mem
;
1762 if (copy_from_user(&kvm_userspace_mem
, argp
,
1763 sizeof kvm_userspace_mem
))
1766 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 1);
1771 case KVM_GET_DIRTY_LOG
: {
1772 struct kvm_dirty_log log
;
1775 if (copy_from_user(&log
, argp
, sizeof log
))
1777 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
1782 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1783 case KVM_REGISTER_COALESCED_MMIO
: {
1784 struct kvm_coalesced_mmio_zone zone
;
1786 if (copy_from_user(&zone
, argp
, sizeof zone
))
1788 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
1794 case KVM_UNREGISTER_COALESCED_MMIO
: {
1795 struct kvm_coalesced_mmio_zone zone
;
1797 if (copy_from_user(&zone
, argp
, sizeof zone
))
1799 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
1807 struct kvm_irqfd data
;
1810 if (copy_from_user(&data
, argp
, sizeof data
))
1812 r
= kvm_irqfd(kvm
, data
.fd
, data
.gsi
, data
.flags
);
1815 case KVM_IOEVENTFD
: {
1816 struct kvm_ioeventfd data
;
1819 if (copy_from_user(&data
, argp
, sizeof data
))
1821 r
= kvm_ioeventfd(kvm
, &data
);
1824 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1825 case KVM_SET_BOOT_CPU_ID
:
1827 mutex_lock(&kvm
->lock
);
1828 if (atomic_read(&kvm
->online_vcpus
) != 0)
1831 kvm
->bsp_vcpu_id
= arg
;
1832 mutex_unlock(&kvm
->lock
);
1836 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
1838 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
1844 #ifdef CONFIG_COMPAT
1845 struct compat_kvm_dirty_log
{
1849 compat_uptr_t dirty_bitmap
; /* one bit per page */
1854 static long kvm_vm_compat_ioctl(struct file
*filp
,
1855 unsigned int ioctl
, unsigned long arg
)
1857 struct kvm
*kvm
= filp
->private_data
;
1860 if (kvm
->mm
!= current
->mm
)
1863 case KVM_GET_DIRTY_LOG
: {
1864 struct compat_kvm_dirty_log compat_log
;
1865 struct kvm_dirty_log log
;
1868 if (copy_from_user(&compat_log
, (void __user
*)arg
,
1869 sizeof(compat_log
)))
1871 log
.slot
= compat_log
.slot
;
1872 log
.padding1
= compat_log
.padding1
;
1873 log
.padding2
= compat_log
.padding2
;
1874 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
1876 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
1882 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
1890 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1892 struct page
*page
[1];
1895 gfn_t gfn
= vmf
->pgoff
;
1896 struct kvm
*kvm
= vma
->vm_file
->private_data
;
1898 addr
= gfn_to_hva(kvm
, gfn
);
1899 if (kvm_is_error_hva(addr
))
1900 return VM_FAULT_SIGBUS
;
1902 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
1904 if (unlikely(npages
!= 1))
1905 return VM_FAULT_SIGBUS
;
1907 vmf
->page
= page
[0];
1911 static const struct vm_operations_struct kvm_vm_vm_ops
= {
1912 .fault
= kvm_vm_fault
,
1915 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1917 vma
->vm_ops
= &kvm_vm_vm_ops
;
1921 static struct file_operations kvm_vm_fops
= {
1922 .release
= kvm_vm_release
,
1923 .unlocked_ioctl
= kvm_vm_ioctl
,
1924 #ifdef CONFIG_COMPAT
1925 .compat_ioctl
= kvm_vm_compat_ioctl
,
1927 .mmap
= kvm_vm_mmap
,
1928 .llseek
= noop_llseek
,
1931 static int kvm_dev_ioctl_create_vm(void)
1936 kvm
= kvm_create_vm();
1938 return PTR_ERR(kvm
);
1939 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1940 r
= kvm_coalesced_mmio_init(kvm
);
1946 fd
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
1953 static long kvm_dev_ioctl_check_extension_generic(long arg
)
1956 case KVM_CAP_USER_MEMORY
:
1957 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
1958 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
1959 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1960 case KVM_CAP_SET_BOOT_CPU_ID
:
1962 case KVM_CAP_INTERNAL_ERROR_DATA
:
1964 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1965 case KVM_CAP_IRQ_ROUTING
:
1966 return KVM_MAX_IRQ_ROUTES
;
1971 return kvm_dev_ioctl_check_extension(arg
);
1974 static long kvm_dev_ioctl(struct file
*filp
,
1975 unsigned int ioctl
, unsigned long arg
)
1980 case KVM_GET_API_VERSION
:
1984 r
= KVM_API_VERSION
;
1990 r
= kvm_dev_ioctl_create_vm();
1992 case KVM_CHECK_EXTENSION
:
1993 r
= kvm_dev_ioctl_check_extension_generic(arg
);
1995 case KVM_GET_VCPU_MMAP_SIZE
:
1999 r
= PAGE_SIZE
; /* struct kvm_run */
2001 r
+= PAGE_SIZE
; /* pio data page */
2003 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2004 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2007 case KVM_TRACE_ENABLE
:
2008 case KVM_TRACE_PAUSE
:
2009 case KVM_TRACE_DISABLE
:
2013 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2019 static struct file_operations kvm_chardev_ops
= {
2020 .unlocked_ioctl
= kvm_dev_ioctl
,
2021 .compat_ioctl
= kvm_dev_ioctl
,
2022 .llseek
= noop_llseek
,
2025 static struct miscdevice kvm_dev
= {
2031 static void hardware_enable(void *junk
)
2033 int cpu
= raw_smp_processor_id();
2036 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2039 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2041 r
= kvm_arch_hardware_enable(NULL
);
2044 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2045 atomic_inc(&hardware_enable_failed
);
2046 printk(KERN_INFO
"kvm: enabling virtualization on "
2047 "CPU%d failed\n", cpu
);
2051 static void hardware_disable(void *junk
)
2053 int cpu
= raw_smp_processor_id();
2055 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2057 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2058 kvm_arch_hardware_disable(NULL
);
2061 static void hardware_disable_all_nolock(void)
2063 BUG_ON(!kvm_usage_count
);
2066 if (!kvm_usage_count
)
2067 on_each_cpu(hardware_disable
, NULL
, 1);
2070 static void hardware_disable_all(void)
2072 spin_lock(&kvm_lock
);
2073 hardware_disable_all_nolock();
2074 spin_unlock(&kvm_lock
);
2077 static int hardware_enable_all(void)
2081 spin_lock(&kvm_lock
);
2084 if (kvm_usage_count
== 1) {
2085 atomic_set(&hardware_enable_failed
, 0);
2086 on_each_cpu(hardware_enable
, NULL
, 1);
2088 if (atomic_read(&hardware_enable_failed
)) {
2089 hardware_disable_all_nolock();
2094 spin_unlock(&kvm_lock
);
2099 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2104 if (!kvm_usage_count
)
2107 val
&= ~CPU_TASKS_FROZEN
;
2110 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2112 hardware_disable(NULL
);
2115 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2117 spin_lock(&kvm_lock
);
2118 hardware_enable(NULL
);
2119 spin_unlock(&kvm_lock
);
2126 asmlinkage
void kvm_handle_fault_on_reboot(void)
2128 if (kvm_rebooting
) {
2129 /* spin while reset goes on */
2134 /* Fault while not rebooting. We want the trace. */
2137 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot
);
2139 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2143 * Some (well, at least mine) BIOSes hang on reboot if
2146 * And Intel TXT required VMX off for all cpu when system shutdown.
2148 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2149 kvm_rebooting
= true;
2150 on_each_cpu(hardware_disable
, NULL
, 1);
2154 static struct notifier_block kvm_reboot_notifier
= {
2155 .notifier_call
= kvm_reboot
,
2159 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2163 for (i
= 0; i
< bus
->dev_count
; i
++) {
2164 struct kvm_io_device
*pos
= bus
->devs
[i
];
2166 kvm_iodevice_destructor(pos
);
2171 /* kvm_io_bus_write - called under kvm->slots_lock */
2172 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2173 int len
, const void *val
)
2176 struct kvm_io_bus
*bus
;
2178 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2179 for (i
= 0; i
< bus
->dev_count
; i
++)
2180 if (!kvm_iodevice_write(bus
->devs
[i
], addr
, len
, val
))
2185 /* kvm_io_bus_read - called under kvm->slots_lock */
2186 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2190 struct kvm_io_bus
*bus
;
2192 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2193 for (i
= 0; i
< bus
->dev_count
; i
++)
2194 if (!kvm_iodevice_read(bus
->devs
[i
], addr
, len
, val
))
2199 /* Caller must hold slots_lock. */
2200 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2201 struct kvm_io_device
*dev
)
2203 struct kvm_io_bus
*new_bus
, *bus
;
2205 bus
= kvm
->buses
[bus_idx
];
2206 if (bus
->dev_count
> NR_IOBUS_DEVS
-1)
2209 new_bus
= kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
);
2212 memcpy(new_bus
, bus
, sizeof(struct kvm_io_bus
));
2213 new_bus
->devs
[new_bus
->dev_count
++] = dev
;
2214 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2215 synchronize_srcu_expedited(&kvm
->srcu
);
2221 /* Caller must hold slots_lock. */
2222 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2223 struct kvm_io_device
*dev
)
2226 struct kvm_io_bus
*new_bus
, *bus
;
2228 new_bus
= kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
);
2232 bus
= kvm
->buses
[bus_idx
];
2233 memcpy(new_bus
, bus
, sizeof(struct kvm_io_bus
));
2236 for (i
= 0; i
< new_bus
->dev_count
; i
++)
2237 if (new_bus
->devs
[i
] == dev
) {
2239 new_bus
->devs
[i
] = new_bus
->devs
[--new_bus
->dev_count
];
2248 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2249 synchronize_srcu_expedited(&kvm
->srcu
);
2254 static struct notifier_block kvm_cpu_notifier
= {
2255 .notifier_call
= kvm_cpu_hotplug
,
2258 static int vm_stat_get(void *_offset
, u64
*val
)
2260 unsigned offset
= (long)_offset
;
2264 spin_lock(&kvm_lock
);
2265 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2266 *val
+= *(u32
*)((void *)kvm
+ offset
);
2267 spin_unlock(&kvm_lock
);
2271 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2273 static int vcpu_stat_get(void *_offset
, u64
*val
)
2275 unsigned offset
= (long)_offset
;
2277 struct kvm_vcpu
*vcpu
;
2281 spin_lock(&kvm_lock
);
2282 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2283 kvm_for_each_vcpu(i
, vcpu
, kvm
)
2284 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2286 spin_unlock(&kvm_lock
);
2290 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2292 static const struct file_operations
*stat_fops
[] = {
2293 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2294 [KVM_STAT_VM
] = &vm_stat_fops
,
2297 static void kvm_init_debug(void)
2299 struct kvm_stats_debugfs_item
*p
;
2301 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2302 for (p
= debugfs_entries
; p
->name
; ++p
)
2303 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2304 (void *)(long)p
->offset
,
2305 stat_fops
[p
->kind
]);
2308 static void kvm_exit_debug(void)
2310 struct kvm_stats_debugfs_item
*p
;
2312 for (p
= debugfs_entries
; p
->name
; ++p
)
2313 debugfs_remove(p
->dentry
);
2314 debugfs_remove(kvm_debugfs_dir
);
2317 static int kvm_suspend(struct sys_device
*dev
, pm_message_t state
)
2319 if (kvm_usage_count
)
2320 hardware_disable(NULL
);
2324 static int kvm_resume(struct sys_device
*dev
)
2326 if (kvm_usage_count
) {
2327 WARN_ON(spin_is_locked(&kvm_lock
));
2328 hardware_enable(NULL
);
2333 static struct sysdev_class kvm_sysdev_class
= {
2335 .suspend
= kvm_suspend
,
2336 .resume
= kvm_resume
,
2339 static struct sys_device kvm_sysdev
= {
2341 .cls
= &kvm_sysdev_class
,
2344 struct page
*bad_page
;
2348 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2350 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2353 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2355 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2357 kvm_arch_vcpu_load(vcpu
, cpu
);
2360 static void kvm_sched_out(struct preempt_notifier
*pn
,
2361 struct task_struct
*next
)
2363 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2365 kvm_arch_vcpu_put(vcpu
);
2368 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
2369 struct module
*module
)
2374 r
= kvm_arch_init(opaque
);
2378 bad_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2380 if (bad_page
== NULL
) {
2385 bad_pfn
= page_to_pfn(bad_page
);
2387 hwpoison_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2389 if (hwpoison_page
== NULL
) {
2394 hwpoison_pfn
= page_to_pfn(hwpoison_page
);
2396 fault_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2398 if (fault_page
== NULL
) {
2403 fault_pfn
= page_to_pfn(fault_page
);
2405 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2410 r
= kvm_arch_hardware_setup();
2414 for_each_online_cpu(cpu
) {
2415 smp_call_function_single(cpu
,
2416 kvm_arch_check_processor_compat
,
2422 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2425 register_reboot_notifier(&kvm_reboot_notifier
);
2427 r
= sysdev_class_register(&kvm_sysdev_class
);
2431 r
= sysdev_register(&kvm_sysdev
);
2435 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2437 vcpu_align
= __alignof__(struct kvm_vcpu
);
2438 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
2440 if (!kvm_vcpu_cache
) {
2445 r
= kvm_async_pf_init();
2449 kvm_chardev_ops
.owner
= module
;
2450 kvm_vm_fops
.owner
= module
;
2451 kvm_vcpu_fops
.owner
= module
;
2453 r
= misc_register(&kvm_dev
);
2455 printk(KERN_ERR
"kvm: misc device register failed\n");
2459 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2460 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2467 kvm_async_pf_deinit();
2469 kmem_cache_destroy(kvm_vcpu_cache
);
2471 sysdev_unregister(&kvm_sysdev
);
2473 sysdev_class_unregister(&kvm_sysdev_class
);
2475 unregister_reboot_notifier(&kvm_reboot_notifier
);
2476 unregister_cpu_notifier(&kvm_cpu_notifier
);
2479 kvm_arch_hardware_unsetup();
2481 free_cpumask_var(cpus_hardware_enabled
);
2484 __free_page(fault_page
);
2486 __free_page(hwpoison_page
);
2487 __free_page(bad_page
);
2493 EXPORT_SYMBOL_GPL(kvm_init
);
2498 misc_deregister(&kvm_dev
);
2499 kmem_cache_destroy(kvm_vcpu_cache
);
2500 kvm_async_pf_deinit();
2501 sysdev_unregister(&kvm_sysdev
);
2502 sysdev_class_unregister(&kvm_sysdev_class
);
2503 unregister_reboot_notifier(&kvm_reboot_notifier
);
2504 unregister_cpu_notifier(&kvm_cpu_notifier
);
2505 on_each_cpu(hardware_disable
, NULL
, 1);
2506 kvm_arch_hardware_unsetup();
2508 free_cpumask_var(cpus_hardware_enabled
);
2509 __free_page(hwpoison_page
);
2510 __free_page(bad_page
);
2512 EXPORT_SYMBOL_GPL(kvm_exit
);