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/syscore_ops.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>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
72 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
75 DEFINE_SPINLOCK(kvm_lock
);
76 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
79 static cpumask_var_t cpus_hardware_enabled
;
80 static int kvm_usage_count
= 0;
81 static atomic_t hardware_enable_failed
;
83 struct kmem_cache
*kvm_vcpu_cache
;
84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
86 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
88 struct dentry
*kvm_debugfs_dir
;
90 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
93 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
96 static int hardware_enable_all(void);
97 static void hardware_disable_all(void);
99 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
101 static void kvm_release_pfn_dirty(pfn_t pfn
);
102 static void mark_page_dirty_in_slot(struct kvm
*kvm
,
103 struct kvm_memory_slot
*memslot
, gfn_t gfn
);
105 __visible
bool kvm_rebooting
;
106 EXPORT_SYMBOL_GPL(kvm_rebooting
);
108 static bool largepages_enabled
= true;
110 bool kvm_is_mmio_pfn(pfn_t pfn
)
113 return !is_zero_pfn(pfn
) && PageReserved(pfn_to_page(pfn
));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu
*vcpu
)
125 if (mutex_lock_killable(&vcpu
->mutex
))
127 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
128 /* The thread running this VCPU changed. */
129 struct pid
*oldpid
= vcpu
->pid
;
130 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
131 rcu_assign_pointer(vcpu
->pid
, newpid
);
137 preempt_notifier_register(&vcpu
->preempt_notifier
);
138 kvm_arch_vcpu_load(vcpu
, cpu
);
143 void vcpu_put(struct kvm_vcpu
*vcpu
)
146 kvm_arch_vcpu_put(vcpu
);
147 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
149 mutex_unlock(&vcpu
->mutex
);
152 static void ack_flush(void *_completed
)
156 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
161 struct kvm_vcpu
*vcpu
;
163 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
166 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
167 kvm_make_request(req
, vcpu
);
170 /* Set ->requests bit before we read ->mode */
173 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
174 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
175 cpumask_set_cpu(cpu
, cpus
);
177 if (unlikely(cpus
== NULL
))
178 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
179 else if (!cpumask_empty(cpus
))
180 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
184 free_cpumask_var(cpus
);
188 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
190 long dirty_count
= kvm
->tlbs_dirty
;
193 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
194 ++kvm
->stat
.remote_tlb_flush
;
195 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
199 void kvm_reload_remote_mmus(struct kvm
*kvm
)
201 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
204 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
206 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
209 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
211 kvm_make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
214 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
219 mutex_init(&vcpu
->mutex
);
224 init_waitqueue_head(&vcpu
->wq
);
225 kvm_async_pf_vcpu_init(vcpu
);
227 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
232 vcpu
->run
= page_address(page
);
234 kvm_vcpu_set_in_spin_loop(vcpu
, false);
235 kvm_vcpu_set_dy_eligible(vcpu
, false);
236 vcpu
->preempted
= false;
238 r
= kvm_arch_vcpu_init(vcpu
);
244 free_page((unsigned long)vcpu
->run
);
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
250 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
253 kvm_arch_vcpu_uninit(vcpu
);
254 free_page((unsigned long)vcpu
->run
);
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
261 return container_of(mn
, struct kvm
, mmu_notifier
);
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
265 struct mm_struct
*mm
,
266 unsigned long address
)
268 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
269 int need_tlb_flush
, idx
;
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
289 idx
= srcu_read_lock(&kvm
->srcu
);
290 spin_lock(&kvm
->mmu_lock
);
292 kvm
->mmu_notifier_seq
++;
293 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
294 /* we've to flush the tlb before the pages can be freed */
296 kvm_flush_remote_tlbs(kvm
);
298 spin_unlock(&kvm
->mmu_lock
);
300 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
302 srcu_read_unlock(&kvm
->srcu
, idx
);
305 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
306 struct mm_struct
*mm
,
307 unsigned long address
,
310 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
313 idx
= srcu_read_lock(&kvm
->srcu
);
314 spin_lock(&kvm
->mmu_lock
);
315 kvm
->mmu_notifier_seq
++;
316 kvm_set_spte_hva(kvm
, address
, pte
);
317 spin_unlock(&kvm
->mmu_lock
);
318 srcu_read_unlock(&kvm
->srcu
, idx
);
321 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
322 struct mm_struct
*mm
,
326 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
327 int need_tlb_flush
= 0, idx
;
329 idx
= srcu_read_lock(&kvm
->srcu
);
330 spin_lock(&kvm
->mmu_lock
);
332 * The count increase must become visible at unlock time as no
333 * spte can be established without taking the mmu_lock and
334 * count is also read inside the mmu_lock critical section.
336 kvm
->mmu_notifier_count
++;
337 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
338 need_tlb_flush
|= kvm
->tlbs_dirty
;
339 /* we've to flush the tlb before the pages can be freed */
341 kvm_flush_remote_tlbs(kvm
);
343 spin_unlock(&kvm
->mmu_lock
);
344 srcu_read_unlock(&kvm
->srcu
, idx
);
347 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
348 struct mm_struct
*mm
,
352 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
354 spin_lock(&kvm
->mmu_lock
);
356 * This sequence increase will notify the kvm page fault that
357 * the page that is going to be mapped in the spte could have
360 kvm
->mmu_notifier_seq
++;
363 * The above sequence increase must be visible before the
364 * below count decrease, which is ensured by the smp_wmb above
365 * in conjunction with the smp_rmb in mmu_notifier_retry().
367 kvm
->mmu_notifier_count
--;
368 spin_unlock(&kvm
->mmu_lock
);
370 BUG_ON(kvm
->mmu_notifier_count
< 0);
373 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
374 struct mm_struct
*mm
,
378 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
381 idx
= srcu_read_lock(&kvm
->srcu
);
382 spin_lock(&kvm
->mmu_lock
);
384 young
= kvm_age_hva(kvm
, start
, end
);
386 kvm_flush_remote_tlbs(kvm
);
388 spin_unlock(&kvm
->mmu_lock
);
389 srcu_read_unlock(&kvm
->srcu
, idx
);
394 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
395 struct mm_struct
*mm
,
396 unsigned long address
)
398 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
401 idx
= srcu_read_lock(&kvm
->srcu
);
402 spin_lock(&kvm
->mmu_lock
);
403 young
= kvm_test_age_hva(kvm
, address
);
404 spin_unlock(&kvm
->mmu_lock
);
405 srcu_read_unlock(&kvm
->srcu
, idx
);
410 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
411 struct mm_struct
*mm
)
413 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
416 idx
= srcu_read_lock(&kvm
->srcu
);
417 kvm_arch_flush_shadow_all(kvm
);
418 srcu_read_unlock(&kvm
->srcu
, idx
);
421 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
422 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
423 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
424 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
425 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
426 .test_young
= kvm_mmu_notifier_test_young
,
427 .change_pte
= kvm_mmu_notifier_change_pte
,
428 .release
= kvm_mmu_notifier_release
,
431 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
433 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
434 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
437 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
439 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
444 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
446 static void kvm_init_memslots_id(struct kvm
*kvm
)
449 struct kvm_memslots
*slots
= kvm
->memslots
;
451 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
452 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
455 static struct kvm
*kvm_create_vm(unsigned long type
)
458 struct kvm
*kvm
= kvm_arch_alloc_vm();
461 return ERR_PTR(-ENOMEM
);
463 r
= kvm_arch_init_vm(kvm
, type
);
465 goto out_err_no_disable
;
467 r
= hardware_enable_all();
469 goto out_err_no_disable
;
471 #ifdef CONFIG_HAVE_KVM_IRQCHIP
472 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
474 #ifdef CONFIG_HAVE_KVM_IRQFD
475 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
478 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
481 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
483 goto out_err_no_srcu
;
486 * Init kvm generation close to the maximum to easily test the
487 * code of handling generation number wrap-around.
489 kvm
->memslots
->generation
= -150;
491 kvm_init_memslots_id(kvm
);
492 if (init_srcu_struct(&kvm
->srcu
))
493 goto out_err_no_srcu
;
494 if (init_srcu_struct(&kvm
->irq_srcu
))
495 goto out_err_no_irq_srcu
;
496 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
497 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
503 spin_lock_init(&kvm
->mmu_lock
);
504 kvm
->mm
= current
->mm
;
505 atomic_inc(&kvm
->mm
->mm_count
);
506 kvm_eventfd_init(kvm
);
507 mutex_init(&kvm
->lock
);
508 mutex_init(&kvm
->irq_lock
);
509 mutex_init(&kvm
->slots_lock
);
510 atomic_set(&kvm
->users_count
, 1);
511 INIT_LIST_HEAD(&kvm
->devices
);
513 r
= kvm_init_mmu_notifier(kvm
);
517 spin_lock(&kvm_lock
);
518 list_add(&kvm
->vm_list
, &vm_list
);
519 spin_unlock(&kvm_lock
);
524 cleanup_srcu_struct(&kvm
->irq_srcu
);
526 cleanup_srcu_struct(&kvm
->srcu
);
528 hardware_disable_all();
530 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
531 kfree(kvm
->buses
[i
]);
532 kfree(kvm
->memslots
);
533 kvm_arch_free_vm(kvm
);
538 * Avoid using vmalloc for a small buffer.
539 * Should not be used when the size is statically known.
541 void *kvm_kvzalloc(unsigned long size
)
543 if (size
> PAGE_SIZE
)
544 return vzalloc(size
);
546 return kzalloc(size
, GFP_KERNEL
);
549 void kvm_kvfree(const void *addr
)
551 if (is_vmalloc_addr(addr
))
557 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
559 if (!memslot
->dirty_bitmap
)
562 kvm_kvfree(memslot
->dirty_bitmap
);
563 memslot
->dirty_bitmap
= NULL
;
567 * Free any memory in @free but not in @dont.
569 static void kvm_free_physmem_slot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
570 struct kvm_memory_slot
*dont
)
572 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
573 kvm_destroy_dirty_bitmap(free
);
575 kvm_arch_free_memslot(kvm
, free
, dont
);
580 static void kvm_free_physmem(struct kvm
*kvm
)
582 struct kvm_memslots
*slots
= kvm
->memslots
;
583 struct kvm_memory_slot
*memslot
;
585 kvm_for_each_memslot(memslot
, slots
)
586 kvm_free_physmem_slot(kvm
, memslot
, NULL
);
588 kfree(kvm
->memslots
);
591 static void kvm_destroy_devices(struct kvm
*kvm
)
593 struct list_head
*node
, *tmp
;
595 list_for_each_safe(node
, tmp
, &kvm
->devices
) {
596 struct kvm_device
*dev
=
597 list_entry(node
, struct kvm_device
, vm_node
);
600 dev
->ops
->destroy(dev
);
604 static void kvm_destroy_vm(struct kvm
*kvm
)
607 struct mm_struct
*mm
= kvm
->mm
;
609 kvm_arch_sync_events(kvm
);
610 spin_lock(&kvm_lock
);
611 list_del(&kvm
->vm_list
);
612 spin_unlock(&kvm_lock
);
613 kvm_free_irq_routing(kvm
);
614 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
615 kvm_io_bus_destroy(kvm
->buses
[i
]);
616 kvm_coalesced_mmio_free(kvm
);
617 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
618 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
620 kvm_arch_flush_shadow_all(kvm
);
622 kvm_arch_destroy_vm(kvm
);
623 kvm_destroy_devices(kvm
);
624 kvm_free_physmem(kvm
);
625 cleanup_srcu_struct(&kvm
->irq_srcu
);
626 cleanup_srcu_struct(&kvm
->srcu
);
627 kvm_arch_free_vm(kvm
);
628 hardware_disable_all();
632 void kvm_get_kvm(struct kvm
*kvm
)
634 atomic_inc(&kvm
->users_count
);
636 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
638 void kvm_put_kvm(struct kvm
*kvm
)
640 if (atomic_dec_and_test(&kvm
->users_count
))
643 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
646 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
648 struct kvm
*kvm
= filp
->private_data
;
650 kvm_irqfd_release(kvm
);
657 * Allocation size is twice as large as the actual dirty bitmap size.
658 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
660 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
662 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
664 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
665 if (!memslot
->dirty_bitmap
)
671 static int cmp_memslot(const void *slot1
, const void *slot2
)
673 struct kvm_memory_slot
*s1
, *s2
;
675 s1
= (struct kvm_memory_slot
*)slot1
;
676 s2
= (struct kvm_memory_slot
*)slot2
;
678 if (s1
->npages
< s2
->npages
)
680 if (s1
->npages
> s2
->npages
)
687 * Sort the memslots base on its size, so the larger slots
688 * will get better fit.
690 static void sort_memslots(struct kvm_memslots
*slots
)
694 sort(slots
->memslots
, KVM_MEM_SLOTS_NUM
,
695 sizeof(struct kvm_memory_slot
), cmp_memslot
, NULL
);
697 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
698 slots
->id_to_index
[slots
->memslots
[i
].id
] = i
;
701 static void update_memslots(struct kvm_memslots
*slots
,
702 struct kvm_memory_slot
*new)
706 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
707 unsigned long npages
= old
->npages
;
710 if (new->npages
!= npages
)
711 sort_memslots(slots
);
715 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
717 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
719 #ifdef __KVM_HAVE_READONLY_MEM
720 valid_flags
|= KVM_MEM_READONLY
;
723 if (mem
->flags
& ~valid_flags
)
729 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
730 struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
732 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
735 * Set the low bit in the generation, which disables SPTE caching
736 * until the end of synchronize_srcu_expedited.
738 WARN_ON(old_memslots
->generation
& 1);
739 slots
->generation
= old_memslots
->generation
+ 1;
741 update_memslots(slots
, new);
742 rcu_assign_pointer(kvm
->memslots
, slots
);
743 synchronize_srcu_expedited(&kvm
->srcu
);
746 * Increment the new memslot generation a second time. This prevents
747 * vm exits that race with memslot updates from caching a memslot
748 * generation that will (potentially) be valid forever.
752 kvm_arch_memslots_updated(kvm
);
758 * Allocate some memory and give it an address in the guest physical address
761 * Discontiguous memory is allowed, mostly for framebuffers.
763 * Must be called holding mmap_sem for write.
765 int __kvm_set_memory_region(struct kvm
*kvm
,
766 struct kvm_userspace_memory_region
*mem
)
770 unsigned long npages
;
771 struct kvm_memory_slot
*slot
;
772 struct kvm_memory_slot old
, new;
773 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
774 enum kvm_mr_change change
;
776 r
= check_memory_region_flags(mem
);
781 /* General sanity checks */
782 if (mem
->memory_size
& (PAGE_SIZE
- 1))
784 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
786 /* We can read the guest memory with __xxx_user() later on. */
787 if ((mem
->slot
< KVM_USER_MEM_SLOTS
) &&
788 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
789 !access_ok(VERIFY_WRITE
,
790 (void __user
*)(unsigned long)mem
->userspace_addr
,
793 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
795 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
798 slot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
799 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
800 npages
= mem
->memory_size
>> PAGE_SHIFT
;
802 if (npages
> KVM_MEM_MAX_NR_PAGES
)
806 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
811 new.base_gfn
= base_gfn
;
813 new.flags
= mem
->flags
;
817 change
= KVM_MR_CREATE
;
818 else { /* Modify an existing slot. */
819 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
820 (npages
!= old
.npages
) ||
821 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
824 if (base_gfn
!= old
.base_gfn
)
825 change
= KVM_MR_MOVE
;
826 else if (new.flags
!= old
.flags
)
827 change
= KVM_MR_FLAGS_ONLY
;
828 else { /* Nothing to change. */
833 } else if (old
.npages
) {
834 change
= KVM_MR_DELETE
;
835 } else /* Modify a non-existent slot: disallowed. */
838 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
839 /* Check for overlaps */
841 kvm_for_each_memslot(slot
, kvm
->memslots
) {
842 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
843 (slot
->id
== mem
->slot
))
845 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
846 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
851 /* Free page dirty bitmap if unneeded */
852 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
853 new.dirty_bitmap
= NULL
;
856 if (change
== KVM_MR_CREATE
) {
857 new.userspace_addr
= mem
->userspace_addr
;
859 if (kvm_arch_create_memslot(kvm
, &new, npages
))
863 /* Allocate page dirty bitmap if needed */
864 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
865 if (kvm_create_dirty_bitmap(&new) < 0)
869 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
870 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
874 slot
= id_to_memslot(slots
, mem
->slot
);
875 slot
->flags
|= KVM_MEMSLOT_INVALID
;
877 old_memslots
= install_new_memslots(kvm
, slots
, NULL
);
879 /* slot was deleted or moved, clear iommu mapping */
880 kvm_iommu_unmap_pages(kvm
, &old
);
881 /* From this point no new shadow pages pointing to a deleted,
882 * or moved, memslot will be created.
884 * validation of sp->gfn happens in:
885 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
886 * - kvm_is_visible_gfn (mmu_check_roots)
888 kvm_arch_flush_shadow_memslot(kvm
, slot
);
889 slots
= old_memslots
;
892 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
898 * We can re-use the old_memslots from above, the only difference
899 * from the currently installed memslots is the invalid flag. This
900 * will get overwritten by update_memslots anyway.
903 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
909 /* actual memory is freed via old in kvm_free_physmem_slot below */
910 if (change
== KVM_MR_DELETE
) {
911 new.dirty_bitmap
= NULL
;
912 memset(&new.arch
, 0, sizeof(new.arch
));
915 old_memslots
= install_new_memslots(kvm
, slots
, &new);
917 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
919 kvm_free_physmem_slot(kvm
, &old
, &new);
923 * IOMMU mapping: New slots need to be mapped. Old slots need to be
924 * un-mapped and re-mapped if their base changes. Since base change
925 * unmapping is handled above with slot deletion, mapping alone is
926 * needed here. Anything else the iommu might care about for existing
927 * slots (size changes, userspace addr changes and read-only flag
928 * changes) is disallowed above, so any other attribute changes getting
929 * here can be skipped.
931 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
932 r
= kvm_iommu_map_pages(kvm
, &new);
941 kvm_free_physmem_slot(kvm
, &new, &old
);
945 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
947 int kvm_set_memory_region(struct kvm
*kvm
,
948 struct kvm_userspace_memory_region
*mem
)
952 mutex_lock(&kvm
->slots_lock
);
953 r
= __kvm_set_memory_region(kvm
, mem
);
954 mutex_unlock(&kvm
->slots_lock
);
957 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
959 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
960 struct kvm_userspace_memory_region
*mem
)
962 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
964 return kvm_set_memory_region(kvm
, mem
);
967 int kvm_get_dirty_log(struct kvm
*kvm
,
968 struct kvm_dirty_log
*log
, int *is_dirty
)
970 struct kvm_memory_slot
*memslot
;
973 unsigned long any
= 0;
976 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
979 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
981 if (!memslot
->dirty_bitmap
)
984 n
= kvm_dirty_bitmap_bytes(memslot
);
986 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
987 any
= memslot
->dirty_bitmap
[i
];
990 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1000 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1002 bool kvm_largepages_enabled(void)
1004 return largepages_enabled
;
1007 void kvm_disable_largepages(void)
1009 largepages_enabled
= false;
1011 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1013 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1015 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1017 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1019 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1021 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1023 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1024 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1029 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1031 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1033 struct vm_area_struct
*vma
;
1034 unsigned long addr
, size
;
1038 addr
= gfn_to_hva(kvm
, gfn
);
1039 if (kvm_is_error_hva(addr
))
1042 down_read(¤t
->mm
->mmap_sem
);
1043 vma
= find_vma(current
->mm
, addr
);
1047 size
= vma_kernel_pagesize(vma
);
1050 up_read(¤t
->mm
->mmap_sem
);
1055 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1057 return slot
->flags
& KVM_MEM_READONLY
;
1060 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1061 gfn_t
*nr_pages
, bool write
)
1063 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1064 return KVM_HVA_ERR_BAD
;
1066 if (memslot_is_readonly(slot
) && write
)
1067 return KVM_HVA_ERR_RO_BAD
;
1070 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1072 return __gfn_to_hva_memslot(slot
, gfn
);
1075 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1078 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1081 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1084 return gfn_to_hva_many(slot
, gfn
, NULL
);
1086 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1088 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1090 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1092 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1095 * If writable is set to false, the hva returned by this function is only
1096 * allowed to be read.
1098 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1099 gfn_t gfn
, bool *writable
)
1101 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1103 if (!kvm_is_error_hva(hva
) && writable
)
1104 *writable
= !memslot_is_readonly(slot
);
1109 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1111 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1113 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1116 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1118 return __copy_from_user(data
, hva
, len
);
1121 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1123 return __copy_from_user_inatomic(data
, hva
, len
);
1126 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1127 unsigned long start
, int write
, struct page
**page
)
1129 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1132 flags
|= FOLL_WRITE
;
1134 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1137 int kvm_get_user_page_io(struct task_struct
*tsk
, struct mm_struct
*mm
,
1138 unsigned long addr
, bool write_fault
,
1139 struct page
**pagep
)
1143 int flags
= FOLL_TOUCH
| FOLL_HWPOISON
|
1144 (pagep
? FOLL_GET
: 0) |
1145 (write_fault
? FOLL_WRITE
: 0);
1148 * If retrying the fault, we get here *not* having allowed the filemap
1149 * to wait on the page lock. We should now allow waiting on the IO with
1150 * the mmap semaphore released.
1152 down_read(&mm
->mmap_sem
);
1153 npages
= __get_user_pages(tsk
, mm
, addr
, 1, flags
, pagep
, NULL
,
1162 * The previous call has now waited on the IO. Now we can
1163 * retry and complete. Pass TRIED to ensure we do not re
1164 * schedule async IO (see e.g. filemap_fault).
1166 down_read(&mm
->mmap_sem
);
1167 npages
= __get_user_pages(tsk
, mm
, addr
, 1, flags
| FOLL_TRIED
,
1170 up_read(&mm
->mmap_sem
);
1174 static inline int check_user_page_hwpoison(unsigned long addr
)
1176 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1178 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1179 flags
, NULL
, NULL
, NULL
);
1180 return rc
== -EHWPOISON
;
1184 * The atomic path to get the writable pfn which will be stored in @pfn,
1185 * true indicates success, otherwise false is returned.
1187 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1188 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1190 struct page
*page
[1];
1193 if (!(async
|| atomic
))
1197 * Fast pin a writable pfn only if it is a write fault request
1198 * or the caller allows to map a writable pfn for a read fault
1201 if (!(write_fault
|| writable
))
1204 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1206 *pfn
= page_to_pfn(page
[0]);
1217 * The slow path to get the pfn of the specified host virtual address,
1218 * 1 indicates success, -errno is returned if error is detected.
1220 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1221 bool *writable
, pfn_t
*pfn
)
1223 struct page
*page
[1];
1229 *writable
= write_fault
;
1232 down_read(¤t
->mm
->mmap_sem
);
1233 npages
= get_user_page_nowait(current
, current
->mm
,
1234 addr
, write_fault
, page
);
1235 up_read(¤t
->mm
->mmap_sem
);
1238 * By now we have tried gup_fast, and possibly async_pf, and we
1239 * are certainly not atomic. Time to retry the gup, allowing
1240 * mmap semaphore to be relinquished in the case of IO.
1242 npages
= kvm_get_user_page_io(current
, current
->mm
, addr
,
1248 /* map read fault as writable if possible */
1249 if (unlikely(!write_fault
) && writable
) {
1250 struct page
*wpage
[1];
1252 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1261 *pfn
= page_to_pfn(page
[0]);
1265 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1267 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1270 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1277 * Pin guest page in memory and return its pfn.
1278 * @addr: host virtual address which maps memory to the guest
1279 * @atomic: whether this function can sleep
1280 * @async: whether this function need to wait IO complete if the
1281 * host page is not in the memory
1282 * @write_fault: whether we should get a writable host page
1283 * @writable: whether it allows to map a writable host page for !@write_fault
1285 * The function will map a writable host page for these two cases:
1286 * 1): @write_fault = true
1287 * 2): @write_fault = false && @writable, @writable will tell the caller
1288 * whether the mapping is writable.
1290 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1291 bool write_fault
, bool *writable
)
1293 struct vm_area_struct
*vma
;
1297 /* we can do it either atomically or asynchronously, not both */
1298 BUG_ON(atomic
&& async
);
1300 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1304 return KVM_PFN_ERR_FAULT
;
1306 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1310 down_read(¤t
->mm
->mmap_sem
);
1311 if (npages
== -EHWPOISON
||
1312 (!async
&& check_user_page_hwpoison(addr
))) {
1313 pfn
= KVM_PFN_ERR_HWPOISON
;
1317 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1320 pfn
= KVM_PFN_ERR_FAULT
;
1321 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1322 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1324 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1326 if (async
&& vma_is_valid(vma
, write_fault
))
1328 pfn
= KVM_PFN_ERR_FAULT
;
1331 up_read(¤t
->mm
->mmap_sem
);
1336 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1337 bool *async
, bool write_fault
, bool *writable
)
1339 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1341 if (addr
== KVM_HVA_ERR_RO_BAD
)
1342 return KVM_PFN_ERR_RO_FAULT
;
1344 if (kvm_is_error_hva(addr
))
1345 return KVM_PFN_NOSLOT
;
1347 /* Do not map writable pfn in the readonly memslot. */
1348 if (writable
&& memslot_is_readonly(slot
)) {
1353 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1357 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1358 bool write_fault
, bool *writable
)
1360 struct kvm_memory_slot
*slot
;
1365 slot
= gfn_to_memslot(kvm
, gfn
);
1367 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1371 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1373 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1375 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1377 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1378 bool write_fault
, bool *writable
)
1380 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1382 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1384 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1386 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1388 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1390 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1393 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1395 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1397 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1399 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1402 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1404 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1406 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1408 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1414 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1415 if (kvm_is_error_hva(addr
))
1418 if (entry
< nr_pages
)
1421 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1423 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1425 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1427 if (is_error_noslot_pfn(pfn
))
1428 return KVM_ERR_PTR_BAD_PAGE
;
1430 if (kvm_is_mmio_pfn(pfn
)) {
1432 return KVM_ERR_PTR_BAD_PAGE
;
1435 return pfn_to_page(pfn
);
1438 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1442 pfn
= gfn_to_pfn(kvm
, gfn
);
1444 return kvm_pfn_to_page(pfn
);
1447 EXPORT_SYMBOL_GPL(gfn_to_page
);
1449 void kvm_release_page_clean(struct page
*page
)
1451 WARN_ON(is_error_page(page
));
1453 kvm_release_pfn_clean(page_to_pfn(page
));
1455 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1457 void kvm_release_pfn_clean(pfn_t pfn
)
1459 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1460 put_page(pfn_to_page(pfn
));
1462 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1464 void kvm_release_page_dirty(struct page
*page
)
1466 WARN_ON(is_error_page(page
));
1468 kvm_release_pfn_dirty(page_to_pfn(page
));
1470 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1472 static void kvm_release_pfn_dirty(pfn_t pfn
)
1474 kvm_set_pfn_dirty(pfn
);
1475 kvm_release_pfn_clean(pfn
);
1478 void kvm_set_pfn_dirty(pfn_t pfn
)
1480 if (!kvm_is_mmio_pfn(pfn
)) {
1481 struct page
*page
= pfn_to_page(pfn
);
1482 if (!PageReserved(page
))
1486 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1488 void kvm_set_pfn_accessed(pfn_t pfn
)
1490 if (!kvm_is_mmio_pfn(pfn
))
1491 mark_page_accessed(pfn_to_page(pfn
));
1493 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1495 void kvm_get_pfn(pfn_t pfn
)
1497 if (!kvm_is_mmio_pfn(pfn
))
1498 get_page(pfn_to_page(pfn
));
1500 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1502 static int next_segment(unsigned long len
, int offset
)
1504 if (len
> PAGE_SIZE
- offset
)
1505 return PAGE_SIZE
- offset
;
1510 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1516 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1517 if (kvm_is_error_hva(addr
))
1519 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1524 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1526 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1528 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1530 int offset
= offset_in_page(gpa
);
1533 while ((seg
= next_segment(len
, offset
)) != 0) {
1534 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1544 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1546 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1551 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1552 int offset
= offset_in_page(gpa
);
1554 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1555 if (kvm_is_error_hva(addr
))
1557 pagefault_disable();
1558 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1564 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1566 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1567 int offset
, int len
)
1572 addr
= gfn_to_hva(kvm
, gfn
);
1573 if (kvm_is_error_hva(addr
))
1575 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1578 mark_page_dirty(kvm
, gfn
);
1581 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1583 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1586 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1588 int offset
= offset_in_page(gpa
);
1591 while ((seg
= next_segment(len
, offset
)) != 0) {
1592 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1603 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1604 gpa_t gpa
, unsigned long len
)
1606 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1607 int offset
= offset_in_page(gpa
);
1608 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1609 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1610 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1611 gfn_t nr_pages_avail
;
1614 ghc
->generation
= slots
->generation
;
1616 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1617 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, &nr_pages_avail
);
1618 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_avail
>= nr_pages_needed
) {
1622 * If the requested region crosses two memslots, we still
1623 * verify that the entire region is valid here.
1625 while (start_gfn
<= end_gfn
) {
1626 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1627 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1629 if (kvm_is_error_hva(ghc
->hva
))
1631 start_gfn
+= nr_pages_avail
;
1633 /* Use the slow path for cross page reads and writes. */
1634 ghc
->memslot
= NULL
;
1638 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1640 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1641 void *data
, unsigned long len
)
1643 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1646 BUG_ON(len
> ghc
->len
);
1648 if (slots
->generation
!= ghc
->generation
)
1649 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1651 if (unlikely(!ghc
->memslot
))
1652 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1654 if (kvm_is_error_hva(ghc
->hva
))
1657 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1660 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1664 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1666 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1667 void *data
, unsigned long len
)
1669 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1672 BUG_ON(len
> ghc
->len
);
1674 if (slots
->generation
!= ghc
->generation
)
1675 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1677 if (unlikely(!ghc
->memslot
))
1678 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1680 if (kvm_is_error_hva(ghc
->hva
))
1683 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1689 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1691 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1693 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1695 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1697 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1699 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1701 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1703 int offset
= offset_in_page(gpa
);
1706 while ((seg
= next_segment(len
, offset
)) != 0) {
1707 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1716 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1718 static void mark_page_dirty_in_slot(struct kvm
*kvm
,
1719 struct kvm_memory_slot
*memslot
,
1722 if (memslot
&& memslot
->dirty_bitmap
) {
1723 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1725 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1729 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1731 struct kvm_memory_slot
*memslot
;
1733 memslot
= gfn_to_memslot(kvm
, gfn
);
1734 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1736 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1739 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1741 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1746 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1748 if (kvm_arch_vcpu_runnable(vcpu
)) {
1749 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1752 if (kvm_cpu_has_pending_timer(vcpu
))
1754 if (signal_pending(current
))
1760 finish_wait(&vcpu
->wq
, &wait
);
1762 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
1766 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1768 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1771 int cpu
= vcpu
->cpu
;
1772 wait_queue_head_t
*wqp
;
1774 wqp
= kvm_arch_vcpu_wq(vcpu
);
1775 if (waitqueue_active(wqp
)) {
1776 wake_up_interruptible(wqp
);
1777 ++vcpu
->stat
.halt_wakeup
;
1781 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1782 if (kvm_arch_vcpu_should_kick(vcpu
))
1783 smp_send_reschedule(cpu
);
1786 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1787 #endif /* !CONFIG_S390 */
1789 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1792 struct task_struct
*task
= NULL
;
1796 pid
= rcu_dereference(target
->pid
);
1798 task
= get_pid_task(pid
, PIDTYPE_PID
);
1802 if (task
->flags
& PF_VCPU
) {
1803 put_task_struct(task
);
1806 ret
= yield_to(task
, 1);
1807 put_task_struct(task
);
1811 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1814 * Helper that checks whether a VCPU is eligible for directed yield.
1815 * Most eligible candidate to yield is decided by following heuristics:
1817 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1818 * (preempted lock holder), indicated by @in_spin_loop.
1819 * Set at the beiginning and cleared at the end of interception/PLE handler.
1821 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1822 * chance last time (mostly it has become eligible now since we have probably
1823 * yielded to lockholder in last iteration. This is done by toggling
1824 * @dy_eligible each time a VCPU checked for eligibility.)
1826 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1827 * to preempted lock-holder could result in wrong VCPU selection and CPU
1828 * burning. Giving priority for a potential lock-holder increases lock
1831 * Since algorithm is based on heuristics, accessing another VCPU data without
1832 * locking does not harm. It may result in trying to yield to same VCPU, fail
1833 * and continue with next VCPU and so on.
1835 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1837 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1840 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1841 vcpu
->spin_loop
.dy_eligible
;
1843 if (vcpu
->spin_loop
.in_spin_loop
)
1844 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1852 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1854 struct kvm
*kvm
= me
->kvm
;
1855 struct kvm_vcpu
*vcpu
;
1856 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1862 kvm_vcpu_set_in_spin_loop(me
, true);
1864 * We boost the priority of a VCPU that is runnable but not
1865 * currently running, because it got preempted by something
1866 * else and called schedule in __vcpu_run. Hopefully that
1867 * VCPU is holding the lock that we need and will release it.
1868 * We approximate round-robin by starting at the last boosted VCPU.
1870 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1871 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1872 if (!pass
&& i
<= last_boosted_vcpu
) {
1873 i
= last_boosted_vcpu
;
1875 } else if (pass
&& i
> last_boosted_vcpu
)
1877 if (!ACCESS_ONCE(vcpu
->preempted
))
1881 if (waitqueue_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
1883 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1886 yielded
= kvm_vcpu_yield_to(vcpu
);
1888 kvm
->last_boosted_vcpu
= i
;
1890 } else if (yielded
< 0) {
1897 kvm_vcpu_set_in_spin_loop(me
, false);
1899 /* Ensure vcpu is not eligible during next spinloop */
1900 kvm_vcpu_set_dy_eligible(me
, false);
1902 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1904 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1906 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1909 if (vmf
->pgoff
== 0)
1910 page
= virt_to_page(vcpu
->run
);
1912 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1913 page
= virt_to_page(vcpu
->arch
.pio_data
);
1915 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1916 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1917 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1920 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1926 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1927 .fault
= kvm_vcpu_fault
,
1930 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1932 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1936 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1938 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1940 kvm_put_kvm(vcpu
->kvm
);
1944 static struct file_operations kvm_vcpu_fops
= {
1945 .release
= kvm_vcpu_release
,
1946 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1947 #ifdef CONFIG_COMPAT
1948 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1950 .mmap
= kvm_vcpu_mmap
,
1951 .llseek
= noop_llseek
,
1955 * Allocates an inode for the vcpu.
1957 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1959 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
1963 * Creates some virtual cpus. Good luck creating more than one.
1965 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1968 struct kvm_vcpu
*vcpu
, *v
;
1970 if (id
>= KVM_MAX_VCPUS
)
1973 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1975 return PTR_ERR(vcpu
);
1977 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1979 r
= kvm_arch_vcpu_setup(vcpu
);
1983 mutex_lock(&kvm
->lock
);
1984 if (!kvm_vcpu_compatible(vcpu
)) {
1986 goto unlock_vcpu_destroy
;
1988 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1990 goto unlock_vcpu_destroy
;
1993 kvm_for_each_vcpu(r
, v
, kvm
)
1994 if (v
->vcpu_id
== id
) {
1996 goto unlock_vcpu_destroy
;
1999 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2001 /* Now it's all set up, let userspace reach it */
2003 r
= create_vcpu_fd(vcpu
);
2006 goto unlock_vcpu_destroy
;
2009 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2011 atomic_inc(&kvm
->online_vcpus
);
2013 mutex_unlock(&kvm
->lock
);
2014 kvm_arch_vcpu_postcreate(vcpu
);
2017 unlock_vcpu_destroy
:
2018 mutex_unlock(&kvm
->lock
);
2020 kvm_arch_vcpu_destroy(vcpu
);
2024 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2027 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2028 vcpu
->sigset_active
= 1;
2029 vcpu
->sigset
= *sigset
;
2031 vcpu
->sigset_active
= 0;
2035 static long kvm_vcpu_ioctl(struct file
*filp
,
2036 unsigned int ioctl
, unsigned long arg
)
2038 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2039 void __user
*argp
= (void __user
*)arg
;
2041 struct kvm_fpu
*fpu
= NULL
;
2042 struct kvm_sregs
*kvm_sregs
= NULL
;
2044 if (vcpu
->kvm
->mm
!= current
->mm
)
2047 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2050 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2052 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2053 * so vcpu_load() would break it.
2055 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
2056 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2060 r
= vcpu_load(vcpu
);
2068 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2069 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2071 case KVM_GET_REGS
: {
2072 struct kvm_regs
*kvm_regs
;
2075 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2078 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2082 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2089 case KVM_SET_REGS
: {
2090 struct kvm_regs
*kvm_regs
;
2093 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2094 if (IS_ERR(kvm_regs
)) {
2095 r
= PTR_ERR(kvm_regs
);
2098 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2102 case KVM_GET_SREGS
: {
2103 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2107 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2111 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2116 case KVM_SET_SREGS
: {
2117 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2118 if (IS_ERR(kvm_sregs
)) {
2119 r
= PTR_ERR(kvm_sregs
);
2123 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2126 case KVM_GET_MP_STATE
: {
2127 struct kvm_mp_state mp_state
;
2129 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2133 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2138 case KVM_SET_MP_STATE
: {
2139 struct kvm_mp_state mp_state
;
2142 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2144 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2147 case KVM_TRANSLATE
: {
2148 struct kvm_translation tr
;
2151 if (copy_from_user(&tr
, argp
, sizeof tr
))
2153 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2157 if (copy_to_user(argp
, &tr
, sizeof tr
))
2162 case KVM_SET_GUEST_DEBUG
: {
2163 struct kvm_guest_debug dbg
;
2166 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2168 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2171 case KVM_SET_SIGNAL_MASK
: {
2172 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2173 struct kvm_signal_mask kvm_sigmask
;
2174 sigset_t sigset
, *p
;
2179 if (copy_from_user(&kvm_sigmask
, argp
,
2180 sizeof kvm_sigmask
))
2183 if (kvm_sigmask
.len
!= sizeof sigset
)
2186 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2191 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2195 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2199 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2203 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2209 fpu
= memdup_user(argp
, sizeof(*fpu
));
2215 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2219 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2228 #ifdef CONFIG_COMPAT
2229 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2230 unsigned int ioctl
, unsigned long arg
)
2232 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2233 void __user
*argp
= compat_ptr(arg
);
2236 if (vcpu
->kvm
->mm
!= current
->mm
)
2240 case KVM_SET_SIGNAL_MASK
: {
2241 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2242 struct kvm_signal_mask kvm_sigmask
;
2243 compat_sigset_t csigset
;
2248 if (copy_from_user(&kvm_sigmask
, argp
,
2249 sizeof kvm_sigmask
))
2252 if (kvm_sigmask
.len
!= sizeof csigset
)
2255 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2258 sigset_from_compat(&sigset
, &csigset
);
2259 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2261 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2265 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2273 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2274 int (*accessor
)(struct kvm_device
*dev
,
2275 struct kvm_device_attr
*attr
),
2278 struct kvm_device_attr attr
;
2283 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2286 return accessor(dev
, &attr
);
2289 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2292 struct kvm_device
*dev
= filp
->private_data
;
2295 case KVM_SET_DEVICE_ATTR
:
2296 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2297 case KVM_GET_DEVICE_ATTR
:
2298 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2299 case KVM_HAS_DEVICE_ATTR
:
2300 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2302 if (dev
->ops
->ioctl
)
2303 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2309 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2311 struct kvm_device
*dev
= filp
->private_data
;
2312 struct kvm
*kvm
= dev
->kvm
;
2318 static const struct file_operations kvm_device_fops
= {
2319 .unlocked_ioctl
= kvm_device_ioctl
,
2320 #ifdef CONFIG_COMPAT
2321 .compat_ioctl
= kvm_device_ioctl
,
2323 .release
= kvm_device_release
,
2326 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2328 if (filp
->f_op
!= &kvm_device_fops
)
2331 return filp
->private_data
;
2334 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2335 #ifdef CONFIG_KVM_MPIC
2336 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2337 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2340 #ifdef CONFIG_KVM_XICS
2341 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2345 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2347 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2350 if (kvm_device_ops_table
[type
] != NULL
)
2353 kvm_device_ops_table
[type
] = ops
;
2357 void kvm_unregister_device_ops(u32 type
)
2359 if (kvm_device_ops_table
[type
] != NULL
)
2360 kvm_device_ops_table
[type
] = NULL
;
2363 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2364 struct kvm_create_device
*cd
)
2366 struct kvm_device_ops
*ops
= NULL
;
2367 struct kvm_device
*dev
;
2368 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2371 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2374 ops
= kvm_device_ops_table
[cd
->type
];
2381 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2388 ret
= ops
->create(dev
, cd
->type
);
2394 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2400 list_add(&dev
->vm_node
, &kvm
->devices
);
2406 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2409 case KVM_CAP_USER_MEMORY
:
2410 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2411 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2412 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2413 case KVM_CAP_SET_BOOT_CPU_ID
:
2415 case KVM_CAP_INTERNAL_ERROR_DATA
:
2416 #ifdef CONFIG_HAVE_KVM_MSI
2417 case KVM_CAP_SIGNAL_MSI
:
2419 #ifdef CONFIG_HAVE_KVM_IRQFD
2420 case KVM_CAP_IRQFD_RESAMPLE
:
2422 case KVM_CAP_CHECK_EXTENSION_VM
:
2424 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2425 case KVM_CAP_IRQ_ROUTING
:
2426 return KVM_MAX_IRQ_ROUTES
;
2431 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2434 static long kvm_vm_ioctl(struct file
*filp
,
2435 unsigned int ioctl
, unsigned long arg
)
2437 struct kvm
*kvm
= filp
->private_data
;
2438 void __user
*argp
= (void __user
*)arg
;
2441 if (kvm
->mm
!= current
->mm
)
2444 case KVM_CREATE_VCPU
:
2445 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2447 case KVM_SET_USER_MEMORY_REGION
: {
2448 struct kvm_userspace_memory_region kvm_userspace_mem
;
2451 if (copy_from_user(&kvm_userspace_mem
, argp
,
2452 sizeof kvm_userspace_mem
))
2455 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2458 case KVM_GET_DIRTY_LOG
: {
2459 struct kvm_dirty_log log
;
2462 if (copy_from_user(&log
, argp
, sizeof log
))
2464 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2467 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2468 case KVM_REGISTER_COALESCED_MMIO
: {
2469 struct kvm_coalesced_mmio_zone zone
;
2471 if (copy_from_user(&zone
, argp
, sizeof zone
))
2473 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2476 case KVM_UNREGISTER_COALESCED_MMIO
: {
2477 struct kvm_coalesced_mmio_zone zone
;
2479 if (copy_from_user(&zone
, argp
, sizeof zone
))
2481 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2486 struct kvm_irqfd data
;
2489 if (copy_from_user(&data
, argp
, sizeof data
))
2491 r
= kvm_irqfd(kvm
, &data
);
2494 case KVM_IOEVENTFD
: {
2495 struct kvm_ioeventfd data
;
2498 if (copy_from_user(&data
, argp
, sizeof data
))
2500 r
= kvm_ioeventfd(kvm
, &data
);
2503 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2504 case KVM_SET_BOOT_CPU_ID
:
2506 mutex_lock(&kvm
->lock
);
2507 if (atomic_read(&kvm
->online_vcpus
) != 0)
2510 kvm
->bsp_vcpu_id
= arg
;
2511 mutex_unlock(&kvm
->lock
);
2514 #ifdef CONFIG_HAVE_KVM_MSI
2515 case KVM_SIGNAL_MSI
: {
2519 if (copy_from_user(&msi
, argp
, sizeof msi
))
2521 r
= kvm_send_userspace_msi(kvm
, &msi
);
2525 #ifdef __KVM_HAVE_IRQ_LINE
2526 case KVM_IRQ_LINE_STATUS
:
2527 case KVM_IRQ_LINE
: {
2528 struct kvm_irq_level irq_event
;
2531 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2534 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2535 ioctl
== KVM_IRQ_LINE_STATUS
);
2540 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2541 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2549 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2550 case KVM_SET_GSI_ROUTING
: {
2551 struct kvm_irq_routing routing
;
2552 struct kvm_irq_routing __user
*urouting
;
2553 struct kvm_irq_routing_entry
*entries
;
2556 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2559 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2564 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2569 if (copy_from_user(entries
, urouting
->entries
,
2570 routing
.nr
* sizeof(*entries
)))
2571 goto out_free_irq_routing
;
2572 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2574 out_free_irq_routing
:
2578 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2579 case KVM_CREATE_DEVICE
: {
2580 struct kvm_create_device cd
;
2583 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2586 r
= kvm_ioctl_create_device(kvm
, &cd
);
2591 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2597 case KVM_CHECK_EXTENSION
:
2598 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
2601 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2603 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2609 #ifdef CONFIG_COMPAT
2610 struct compat_kvm_dirty_log
{
2614 compat_uptr_t dirty_bitmap
; /* one bit per page */
2619 static long kvm_vm_compat_ioctl(struct file
*filp
,
2620 unsigned int ioctl
, unsigned long arg
)
2622 struct kvm
*kvm
= filp
->private_data
;
2625 if (kvm
->mm
!= current
->mm
)
2628 case KVM_GET_DIRTY_LOG
: {
2629 struct compat_kvm_dirty_log compat_log
;
2630 struct kvm_dirty_log log
;
2633 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2634 sizeof(compat_log
)))
2636 log
.slot
= compat_log
.slot
;
2637 log
.padding1
= compat_log
.padding1
;
2638 log
.padding2
= compat_log
.padding2
;
2639 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2641 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2645 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2653 static struct file_operations kvm_vm_fops
= {
2654 .release
= kvm_vm_release
,
2655 .unlocked_ioctl
= kvm_vm_ioctl
,
2656 #ifdef CONFIG_COMPAT
2657 .compat_ioctl
= kvm_vm_compat_ioctl
,
2659 .llseek
= noop_llseek
,
2662 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2667 kvm
= kvm_create_vm(type
);
2669 return PTR_ERR(kvm
);
2670 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2671 r
= kvm_coalesced_mmio_init(kvm
);
2677 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2684 static long kvm_dev_ioctl(struct file
*filp
,
2685 unsigned int ioctl
, unsigned long arg
)
2690 case KVM_GET_API_VERSION
:
2693 r
= KVM_API_VERSION
;
2696 r
= kvm_dev_ioctl_create_vm(arg
);
2698 case KVM_CHECK_EXTENSION
:
2699 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
2701 case KVM_GET_VCPU_MMAP_SIZE
:
2704 r
= PAGE_SIZE
; /* struct kvm_run */
2706 r
+= PAGE_SIZE
; /* pio data page */
2708 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2709 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2712 case KVM_TRACE_ENABLE
:
2713 case KVM_TRACE_PAUSE
:
2714 case KVM_TRACE_DISABLE
:
2718 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2724 static struct file_operations kvm_chardev_ops
= {
2725 .unlocked_ioctl
= kvm_dev_ioctl
,
2726 .compat_ioctl
= kvm_dev_ioctl
,
2727 .llseek
= noop_llseek
,
2730 static struct miscdevice kvm_dev
= {
2736 static void hardware_enable_nolock(void *junk
)
2738 int cpu
= raw_smp_processor_id();
2741 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2744 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2746 r
= kvm_arch_hardware_enable();
2749 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2750 atomic_inc(&hardware_enable_failed
);
2751 printk(KERN_INFO
"kvm: enabling virtualization on "
2752 "CPU%d failed\n", cpu
);
2756 static void hardware_enable(void)
2758 raw_spin_lock(&kvm_count_lock
);
2759 if (kvm_usage_count
)
2760 hardware_enable_nolock(NULL
);
2761 raw_spin_unlock(&kvm_count_lock
);
2764 static void hardware_disable_nolock(void *junk
)
2766 int cpu
= raw_smp_processor_id();
2768 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2770 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2771 kvm_arch_hardware_disable();
2774 static void hardware_disable(void)
2776 raw_spin_lock(&kvm_count_lock
);
2777 if (kvm_usage_count
)
2778 hardware_disable_nolock(NULL
);
2779 raw_spin_unlock(&kvm_count_lock
);
2782 static void hardware_disable_all_nolock(void)
2784 BUG_ON(!kvm_usage_count
);
2787 if (!kvm_usage_count
)
2788 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2791 static void hardware_disable_all(void)
2793 raw_spin_lock(&kvm_count_lock
);
2794 hardware_disable_all_nolock();
2795 raw_spin_unlock(&kvm_count_lock
);
2798 static int hardware_enable_all(void)
2802 raw_spin_lock(&kvm_count_lock
);
2805 if (kvm_usage_count
== 1) {
2806 atomic_set(&hardware_enable_failed
, 0);
2807 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2809 if (atomic_read(&hardware_enable_failed
)) {
2810 hardware_disable_all_nolock();
2815 raw_spin_unlock(&kvm_count_lock
);
2820 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2825 val
&= ~CPU_TASKS_FROZEN
;
2828 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2833 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2841 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2845 * Some (well, at least mine) BIOSes hang on reboot if
2848 * And Intel TXT required VMX off for all cpu when system shutdown.
2850 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2851 kvm_rebooting
= true;
2852 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2856 static struct notifier_block kvm_reboot_notifier
= {
2857 .notifier_call
= kvm_reboot
,
2861 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2865 for (i
= 0; i
< bus
->dev_count
; i
++) {
2866 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2868 kvm_iodevice_destructor(pos
);
2873 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
2874 const struct kvm_io_range
*r2
)
2876 if (r1
->addr
< r2
->addr
)
2878 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2883 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2885 return kvm_io_bus_cmp(p1
, p2
);
2888 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2889 gpa_t addr
, int len
)
2891 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2897 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2898 kvm_io_bus_sort_cmp
, NULL
);
2903 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2904 gpa_t addr
, int len
)
2906 struct kvm_io_range
*range
, key
;
2909 key
= (struct kvm_io_range
) {
2914 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2915 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2919 off
= range
- bus
->range
;
2921 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
2927 static int __kvm_io_bus_write(struct kvm_io_bus
*bus
,
2928 struct kvm_io_range
*range
, const void *val
)
2932 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2936 while (idx
< bus
->dev_count
&&
2937 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2938 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, range
->addr
,
2947 /* kvm_io_bus_write - called under kvm->slots_lock */
2948 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2949 int len
, const void *val
)
2951 struct kvm_io_bus
*bus
;
2952 struct kvm_io_range range
;
2955 range
= (struct kvm_io_range
) {
2960 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2961 r
= __kvm_io_bus_write(bus
, &range
, val
);
2962 return r
< 0 ? r
: 0;
2965 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2966 int kvm_io_bus_write_cookie(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2967 int len
, const void *val
, long cookie
)
2969 struct kvm_io_bus
*bus
;
2970 struct kvm_io_range range
;
2972 range
= (struct kvm_io_range
) {
2977 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2979 /* First try the device referenced by cookie. */
2980 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
2981 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
2982 if (!kvm_iodevice_write(bus
->range
[cookie
].dev
, addr
, len
,
2987 * cookie contained garbage; fall back to search and return the
2988 * correct cookie value.
2990 return __kvm_io_bus_write(bus
, &range
, val
);
2993 static int __kvm_io_bus_read(struct kvm_io_bus
*bus
, struct kvm_io_range
*range
,
2998 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3002 while (idx
< bus
->dev_count
&&
3003 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3004 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, range
->addr
,
3012 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3014 /* kvm_io_bus_read - called under kvm->slots_lock */
3015 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3018 struct kvm_io_bus
*bus
;
3019 struct kvm_io_range range
;
3022 range
= (struct kvm_io_range
) {
3027 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3028 r
= __kvm_io_bus_read(bus
, &range
, val
);
3029 return r
< 0 ? r
: 0;
3033 /* Caller must hold slots_lock. */
3034 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3035 int len
, struct kvm_io_device
*dev
)
3037 struct kvm_io_bus
*new_bus
, *bus
;
3039 bus
= kvm
->buses
[bus_idx
];
3040 /* exclude ioeventfd which is limited by maximum fd */
3041 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3044 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3045 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3048 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3049 sizeof(struct kvm_io_range
)));
3050 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3051 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3052 synchronize_srcu_expedited(&kvm
->srcu
);
3058 /* Caller must hold slots_lock. */
3059 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3060 struct kvm_io_device
*dev
)
3063 struct kvm_io_bus
*new_bus
, *bus
;
3065 bus
= kvm
->buses
[bus_idx
];
3067 for (i
= 0; i
< bus
->dev_count
; i
++)
3068 if (bus
->range
[i
].dev
== dev
) {
3076 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3077 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3081 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3082 new_bus
->dev_count
--;
3083 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3084 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3086 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3087 synchronize_srcu_expedited(&kvm
->srcu
);
3092 static struct notifier_block kvm_cpu_notifier
= {
3093 .notifier_call
= kvm_cpu_hotplug
,
3096 static int vm_stat_get(void *_offset
, u64
*val
)
3098 unsigned offset
= (long)_offset
;
3102 spin_lock(&kvm_lock
);
3103 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3104 *val
+= *(u32
*)((void *)kvm
+ offset
);
3105 spin_unlock(&kvm_lock
);
3109 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3111 static int vcpu_stat_get(void *_offset
, u64
*val
)
3113 unsigned offset
= (long)_offset
;
3115 struct kvm_vcpu
*vcpu
;
3119 spin_lock(&kvm_lock
);
3120 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3121 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3122 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3124 spin_unlock(&kvm_lock
);
3128 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3130 static const struct file_operations
*stat_fops
[] = {
3131 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3132 [KVM_STAT_VM
] = &vm_stat_fops
,
3135 static int kvm_init_debug(void)
3138 struct kvm_stats_debugfs_item
*p
;
3140 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3141 if (kvm_debugfs_dir
== NULL
)
3144 for (p
= debugfs_entries
; p
->name
; ++p
) {
3145 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3146 (void *)(long)p
->offset
,
3147 stat_fops
[p
->kind
]);
3148 if (p
->dentry
== NULL
)
3155 debugfs_remove_recursive(kvm_debugfs_dir
);
3160 static void kvm_exit_debug(void)
3162 struct kvm_stats_debugfs_item
*p
;
3164 for (p
= debugfs_entries
; p
->name
; ++p
)
3165 debugfs_remove(p
->dentry
);
3166 debugfs_remove(kvm_debugfs_dir
);
3169 static int kvm_suspend(void)
3171 if (kvm_usage_count
)
3172 hardware_disable_nolock(NULL
);
3176 static void kvm_resume(void)
3178 if (kvm_usage_count
) {
3179 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3180 hardware_enable_nolock(NULL
);
3184 static struct syscore_ops kvm_syscore_ops
= {
3185 .suspend
= kvm_suspend
,
3186 .resume
= kvm_resume
,
3190 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3192 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3195 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3197 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3198 if (vcpu
->preempted
)
3199 vcpu
->preempted
= false;
3201 kvm_arch_sched_in(vcpu
, cpu
);
3203 kvm_arch_vcpu_load(vcpu
, cpu
);
3206 static void kvm_sched_out(struct preempt_notifier
*pn
,
3207 struct task_struct
*next
)
3209 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3211 if (current
->state
== TASK_RUNNING
)
3212 vcpu
->preempted
= true;
3213 kvm_arch_vcpu_put(vcpu
);
3216 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3217 struct module
*module
)
3222 r
= kvm_arch_init(opaque
);
3227 * kvm_arch_init makes sure there's at most one caller
3228 * for architectures that support multiple implementations,
3229 * like intel and amd on x86.
3230 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3231 * conflicts in case kvm is already setup for another implementation.
3233 r
= kvm_irqfd_init();
3237 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3242 r
= kvm_arch_hardware_setup();
3246 for_each_online_cpu(cpu
) {
3247 smp_call_function_single(cpu
,
3248 kvm_arch_check_processor_compat
,
3254 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3257 register_reboot_notifier(&kvm_reboot_notifier
);
3259 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3261 vcpu_align
= __alignof__(struct kvm_vcpu
);
3262 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3264 if (!kvm_vcpu_cache
) {
3269 r
= kvm_async_pf_init();
3273 kvm_chardev_ops
.owner
= module
;
3274 kvm_vm_fops
.owner
= module
;
3275 kvm_vcpu_fops
.owner
= module
;
3277 r
= misc_register(&kvm_dev
);
3279 printk(KERN_ERR
"kvm: misc device register failed\n");
3283 register_syscore_ops(&kvm_syscore_ops
);
3285 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3286 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3288 r
= kvm_init_debug();
3290 printk(KERN_ERR
"kvm: create debugfs files failed\n");
3294 r
= kvm_vfio_ops_init();
3300 unregister_syscore_ops(&kvm_syscore_ops
);
3301 misc_deregister(&kvm_dev
);
3303 kvm_async_pf_deinit();
3305 kmem_cache_destroy(kvm_vcpu_cache
);
3307 unregister_reboot_notifier(&kvm_reboot_notifier
);
3308 unregister_cpu_notifier(&kvm_cpu_notifier
);
3311 kvm_arch_hardware_unsetup();
3313 free_cpumask_var(cpus_hardware_enabled
);
3321 EXPORT_SYMBOL_GPL(kvm_init
);
3326 misc_deregister(&kvm_dev
);
3327 kmem_cache_destroy(kvm_vcpu_cache
);
3328 kvm_async_pf_deinit();
3329 unregister_syscore_ops(&kvm_syscore_ops
);
3330 unregister_reboot_notifier(&kvm_reboot_notifier
);
3331 unregister_cpu_notifier(&kvm_cpu_notifier
);
3332 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3333 kvm_arch_hardware_unsetup();
3336 free_cpumask_var(cpus_hardware_enabled
);
3337 kvm_vfio_ops_exit();
3339 EXPORT_SYMBOL_GPL(kvm_exit
);