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_IRQFD
472 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
475 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
478 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
480 goto out_err_no_srcu
;
483 * Init kvm generation close to the maximum to easily test the
484 * code of handling generation number wrap-around.
486 kvm
->memslots
->generation
= -150;
488 kvm_init_memslots_id(kvm
);
489 if (init_srcu_struct(&kvm
->srcu
))
490 goto out_err_no_srcu
;
491 if (init_srcu_struct(&kvm
->irq_srcu
))
492 goto out_err_no_irq_srcu
;
493 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
494 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
500 spin_lock_init(&kvm
->mmu_lock
);
501 kvm
->mm
= current
->mm
;
502 atomic_inc(&kvm
->mm
->mm_count
);
503 kvm_eventfd_init(kvm
);
504 mutex_init(&kvm
->lock
);
505 mutex_init(&kvm
->irq_lock
);
506 mutex_init(&kvm
->slots_lock
);
507 atomic_set(&kvm
->users_count
, 1);
508 INIT_LIST_HEAD(&kvm
->devices
);
510 r
= kvm_init_mmu_notifier(kvm
);
514 spin_lock(&kvm_lock
);
515 list_add(&kvm
->vm_list
, &vm_list
);
516 spin_unlock(&kvm_lock
);
521 cleanup_srcu_struct(&kvm
->irq_srcu
);
523 cleanup_srcu_struct(&kvm
->srcu
);
525 hardware_disable_all();
527 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
528 kfree(kvm
->buses
[i
]);
529 kfree(kvm
->memslots
);
530 kvm_arch_free_vm(kvm
);
535 * Avoid using vmalloc for a small buffer.
536 * Should not be used when the size is statically known.
538 void *kvm_kvzalloc(unsigned long size
)
540 if (size
> PAGE_SIZE
)
541 return vzalloc(size
);
543 return kzalloc(size
, GFP_KERNEL
);
546 void kvm_kvfree(const void *addr
)
548 if (is_vmalloc_addr(addr
))
554 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
556 if (!memslot
->dirty_bitmap
)
559 kvm_kvfree(memslot
->dirty_bitmap
);
560 memslot
->dirty_bitmap
= NULL
;
564 * Free any memory in @free but not in @dont.
566 static void kvm_free_physmem_slot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
567 struct kvm_memory_slot
*dont
)
569 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
570 kvm_destroy_dirty_bitmap(free
);
572 kvm_arch_free_memslot(kvm
, free
, dont
);
577 static void kvm_free_physmem(struct kvm
*kvm
)
579 struct kvm_memslots
*slots
= kvm
->memslots
;
580 struct kvm_memory_slot
*memslot
;
582 kvm_for_each_memslot(memslot
, slots
)
583 kvm_free_physmem_slot(kvm
, memslot
, NULL
);
585 kfree(kvm
->memslots
);
588 static void kvm_destroy_devices(struct kvm
*kvm
)
590 struct list_head
*node
, *tmp
;
592 list_for_each_safe(node
, tmp
, &kvm
->devices
) {
593 struct kvm_device
*dev
=
594 list_entry(node
, struct kvm_device
, vm_node
);
597 dev
->ops
->destroy(dev
);
601 static void kvm_destroy_vm(struct kvm
*kvm
)
604 struct mm_struct
*mm
= kvm
->mm
;
606 kvm_arch_sync_events(kvm
);
607 spin_lock(&kvm_lock
);
608 list_del(&kvm
->vm_list
);
609 spin_unlock(&kvm_lock
);
610 kvm_free_irq_routing(kvm
);
611 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
612 kvm_io_bus_destroy(kvm
->buses
[i
]);
613 kvm_coalesced_mmio_free(kvm
);
614 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
615 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
617 kvm_arch_flush_shadow_all(kvm
);
619 kvm_arch_destroy_vm(kvm
);
620 kvm_destroy_devices(kvm
);
621 kvm_free_physmem(kvm
);
622 cleanup_srcu_struct(&kvm
->irq_srcu
);
623 cleanup_srcu_struct(&kvm
->srcu
);
624 kvm_arch_free_vm(kvm
);
625 hardware_disable_all();
629 void kvm_get_kvm(struct kvm
*kvm
)
631 atomic_inc(&kvm
->users_count
);
633 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
635 void kvm_put_kvm(struct kvm
*kvm
)
637 if (atomic_dec_and_test(&kvm
->users_count
))
640 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
643 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
645 struct kvm
*kvm
= filp
->private_data
;
647 kvm_irqfd_release(kvm
);
654 * Allocation size is twice as large as the actual dirty bitmap size.
655 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
657 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
659 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
661 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
662 if (!memslot
->dirty_bitmap
)
669 * Insert memslot and re-sort memslots based on their GFN,
670 * so binary search could be used to lookup GFN.
671 * Sorting algorithm takes advantage of having initially
672 * sorted array and known changed memslot position.
674 static void update_memslots(struct kvm_memslots
*slots
,
675 struct kvm_memory_slot
*new)
678 int i
= slots
->id_to_index
[id
];
679 struct kvm_memory_slot
*mslots
= slots
->memslots
;
681 WARN_ON(mslots
[i
].id
!= id
);
684 if (mslots
[i
].npages
)
687 if (!mslots
[i
].npages
)
691 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
692 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
693 if (!mslots
[i
+ 1].npages
)
695 mslots
[i
] = mslots
[i
+ 1];
696 slots
->id_to_index
[mslots
[i
].id
] = i
;
700 new->base_gfn
> mslots
[i
- 1].base_gfn
) {
701 mslots
[i
] = mslots
[i
- 1];
702 slots
->id_to_index
[mslots
[i
].id
] = i
;
707 slots
->id_to_index
[mslots
[i
].id
] = i
;
710 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
712 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
714 #ifdef __KVM_HAVE_READONLY_MEM
715 valid_flags
|= KVM_MEM_READONLY
;
718 if (mem
->flags
& ~valid_flags
)
724 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
725 struct kvm_memslots
*slots
)
727 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
730 * Set the low bit in the generation, which disables SPTE caching
731 * until the end of synchronize_srcu_expedited.
733 WARN_ON(old_memslots
->generation
& 1);
734 slots
->generation
= old_memslots
->generation
+ 1;
736 rcu_assign_pointer(kvm
->memslots
, slots
);
737 synchronize_srcu_expedited(&kvm
->srcu
);
740 * Increment the new memslot generation a second time. This prevents
741 * vm exits that race with memslot updates from caching a memslot
742 * generation that will (potentially) be valid forever.
746 kvm_arch_memslots_updated(kvm
);
752 * Allocate some memory and give it an address in the guest physical address
755 * Discontiguous memory is allowed, mostly for framebuffers.
757 * Must be called holding kvm->slots_lock for write.
759 int __kvm_set_memory_region(struct kvm
*kvm
,
760 struct kvm_userspace_memory_region
*mem
)
764 unsigned long npages
;
765 struct kvm_memory_slot
*slot
;
766 struct kvm_memory_slot old
, new;
767 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
768 enum kvm_mr_change change
;
770 r
= check_memory_region_flags(mem
);
775 /* General sanity checks */
776 if (mem
->memory_size
& (PAGE_SIZE
- 1))
778 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
780 /* We can read the guest memory with __xxx_user() later on. */
781 if ((mem
->slot
< KVM_USER_MEM_SLOTS
) &&
782 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
783 !access_ok(VERIFY_WRITE
,
784 (void __user
*)(unsigned long)mem
->userspace_addr
,
787 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
789 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
792 slot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
793 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
794 npages
= mem
->memory_size
>> PAGE_SHIFT
;
796 if (npages
> KVM_MEM_MAX_NR_PAGES
)
800 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
805 new.base_gfn
= base_gfn
;
807 new.flags
= mem
->flags
;
811 change
= KVM_MR_CREATE
;
812 else { /* Modify an existing slot. */
813 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
814 (npages
!= old
.npages
) ||
815 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
818 if (base_gfn
!= old
.base_gfn
)
819 change
= KVM_MR_MOVE
;
820 else if (new.flags
!= old
.flags
)
821 change
= KVM_MR_FLAGS_ONLY
;
822 else { /* Nothing to change. */
827 } else if (old
.npages
) {
828 change
= KVM_MR_DELETE
;
829 } else /* Modify a non-existent slot: disallowed. */
832 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
833 /* Check for overlaps */
835 kvm_for_each_memslot(slot
, kvm
->memslots
) {
836 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
837 (slot
->id
== mem
->slot
))
839 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
840 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
845 /* Free page dirty bitmap if unneeded */
846 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
847 new.dirty_bitmap
= NULL
;
850 if (change
== KVM_MR_CREATE
) {
851 new.userspace_addr
= mem
->userspace_addr
;
853 if (kvm_arch_create_memslot(kvm
, &new, npages
))
857 /* Allocate page dirty bitmap if needed */
858 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
859 if (kvm_create_dirty_bitmap(&new) < 0)
863 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
868 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
869 slot
= id_to_memslot(slots
, mem
->slot
);
870 slot
->flags
|= KVM_MEMSLOT_INVALID
;
872 old_memslots
= install_new_memslots(kvm
, slots
);
874 /* slot was deleted or moved, clear iommu mapping */
875 kvm_iommu_unmap_pages(kvm
, &old
);
876 /* From this point no new shadow pages pointing to a deleted,
877 * or moved, memslot will be created.
879 * validation of sp->gfn happens in:
880 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
881 * - kvm_is_visible_gfn (mmu_check_roots)
883 kvm_arch_flush_shadow_memslot(kvm
, slot
);
886 * We can re-use the old_memslots from above, the only difference
887 * from the currently installed memslots is the invalid flag. This
888 * will get overwritten by update_memslots anyway.
890 slots
= old_memslots
;
893 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
897 /* actual memory is freed via old in kvm_free_physmem_slot below */
898 if (change
== KVM_MR_DELETE
) {
899 new.dirty_bitmap
= NULL
;
900 memset(&new.arch
, 0, sizeof(new.arch
));
903 update_memslots(slots
, &new);
904 old_memslots
= install_new_memslots(kvm
, slots
);
906 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
908 kvm_free_physmem_slot(kvm
, &old
, &new);
912 * IOMMU mapping: New slots need to be mapped. Old slots need to be
913 * un-mapped and re-mapped if their base changes. Since base change
914 * unmapping is handled above with slot deletion, mapping alone is
915 * needed here. Anything else the iommu might care about for existing
916 * slots (size changes, userspace addr changes and read-only flag
917 * changes) is disallowed above, so any other attribute changes getting
918 * here can be skipped.
920 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
921 r
= kvm_iommu_map_pages(kvm
, &new);
930 kvm_free_physmem_slot(kvm
, &new, &old
);
934 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
936 int kvm_set_memory_region(struct kvm
*kvm
,
937 struct kvm_userspace_memory_region
*mem
)
941 mutex_lock(&kvm
->slots_lock
);
942 r
= __kvm_set_memory_region(kvm
, mem
);
943 mutex_unlock(&kvm
->slots_lock
);
946 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
948 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
949 struct kvm_userspace_memory_region
*mem
)
951 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
953 return kvm_set_memory_region(kvm
, mem
);
956 int kvm_get_dirty_log(struct kvm
*kvm
,
957 struct kvm_dirty_log
*log
, int *is_dirty
)
959 struct kvm_memory_slot
*memslot
;
962 unsigned long any
= 0;
965 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
968 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
970 if (!memslot
->dirty_bitmap
)
973 n
= kvm_dirty_bitmap_bytes(memslot
);
975 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
976 any
= memslot
->dirty_bitmap
[i
];
979 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
989 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
991 bool kvm_largepages_enabled(void)
993 return largepages_enabled
;
996 void kvm_disable_largepages(void)
998 largepages_enabled
= false;
1000 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1002 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1004 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1006 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1008 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1010 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1012 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1013 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1018 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1020 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1022 struct vm_area_struct
*vma
;
1023 unsigned long addr
, size
;
1027 addr
= gfn_to_hva(kvm
, gfn
);
1028 if (kvm_is_error_hva(addr
))
1031 down_read(¤t
->mm
->mmap_sem
);
1032 vma
= find_vma(current
->mm
, addr
);
1036 size
= vma_kernel_pagesize(vma
);
1039 up_read(¤t
->mm
->mmap_sem
);
1044 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1046 return slot
->flags
& KVM_MEM_READONLY
;
1049 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1050 gfn_t
*nr_pages
, bool write
)
1052 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1053 return KVM_HVA_ERR_BAD
;
1055 if (memslot_is_readonly(slot
) && write
)
1056 return KVM_HVA_ERR_RO_BAD
;
1059 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1061 return __gfn_to_hva_memslot(slot
, gfn
);
1064 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1067 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1070 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1073 return gfn_to_hva_many(slot
, gfn
, NULL
);
1075 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1077 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1079 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1081 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1084 * If writable is set to false, the hva returned by this function is only
1085 * allowed to be read.
1087 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1088 gfn_t gfn
, bool *writable
)
1090 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1092 if (!kvm_is_error_hva(hva
) && writable
)
1093 *writable
= !memslot_is_readonly(slot
);
1098 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1100 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1102 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1105 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1107 return __copy_from_user(data
, hva
, len
);
1110 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1112 return __copy_from_user_inatomic(data
, hva
, len
);
1115 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1116 unsigned long start
, int write
, struct page
**page
)
1118 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1121 flags
|= FOLL_WRITE
;
1123 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1126 int kvm_get_user_page_io(struct task_struct
*tsk
, struct mm_struct
*mm
,
1127 unsigned long addr
, bool write_fault
,
1128 struct page
**pagep
)
1132 int flags
= FOLL_TOUCH
| FOLL_HWPOISON
|
1133 (pagep
? FOLL_GET
: 0) |
1134 (write_fault
? FOLL_WRITE
: 0);
1137 * If retrying the fault, we get here *not* having allowed the filemap
1138 * to wait on the page lock. We should now allow waiting on the IO with
1139 * the mmap semaphore released.
1141 down_read(&mm
->mmap_sem
);
1142 npages
= __get_user_pages(tsk
, mm
, addr
, 1, flags
, pagep
, NULL
,
1151 * The previous call has now waited on the IO. Now we can
1152 * retry and complete. Pass TRIED to ensure we do not re
1153 * schedule async IO (see e.g. filemap_fault).
1155 down_read(&mm
->mmap_sem
);
1156 npages
= __get_user_pages(tsk
, mm
, addr
, 1, flags
| FOLL_TRIED
,
1159 up_read(&mm
->mmap_sem
);
1163 static inline int check_user_page_hwpoison(unsigned long addr
)
1165 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1167 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1168 flags
, NULL
, NULL
, NULL
);
1169 return rc
== -EHWPOISON
;
1173 * The atomic path to get the writable pfn which will be stored in @pfn,
1174 * true indicates success, otherwise false is returned.
1176 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1177 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1179 struct page
*page
[1];
1182 if (!(async
|| atomic
))
1186 * Fast pin a writable pfn only if it is a write fault request
1187 * or the caller allows to map a writable pfn for a read fault
1190 if (!(write_fault
|| writable
))
1193 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1195 *pfn
= page_to_pfn(page
[0]);
1206 * The slow path to get the pfn of the specified host virtual address,
1207 * 1 indicates success, -errno is returned if error is detected.
1209 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1210 bool *writable
, pfn_t
*pfn
)
1212 struct page
*page
[1];
1218 *writable
= write_fault
;
1221 down_read(¤t
->mm
->mmap_sem
);
1222 npages
= get_user_page_nowait(current
, current
->mm
,
1223 addr
, write_fault
, page
);
1224 up_read(¤t
->mm
->mmap_sem
);
1227 * By now we have tried gup_fast, and possibly async_pf, and we
1228 * are certainly not atomic. Time to retry the gup, allowing
1229 * mmap semaphore to be relinquished in the case of IO.
1231 npages
= kvm_get_user_page_io(current
, current
->mm
, addr
,
1237 /* map read fault as writable if possible */
1238 if (unlikely(!write_fault
) && writable
) {
1239 struct page
*wpage
[1];
1241 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1250 *pfn
= page_to_pfn(page
[0]);
1254 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1256 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1259 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1266 * Pin guest page in memory and return its pfn.
1267 * @addr: host virtual address which maps memory to the guest
1268 * @atomic: whether this function can sleep
1269 * @async: whether this function need to wait IO complete if the
1270 * host page is not in the memory
1271 * @write_fault: whether we should get a writable host page
1272 * @writable: whether it allows to map a writable host page for !@write_fault
1274 * The function will map a writable host page for these two cases:
1275 * 1): @write_fault = true
1276 * 2): @write_fault = false && @writable, @writable will tell the caller
1277 * whether the mapping is writable.
1279 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1280 bool write_fault
, bool *writable
)
1282 struct vm_area_struct
*vma
;
1286 /* we can do it either atomically or asynchronously, not both */
1287 BUG_ON(atomic
&& async
);
1289 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1293 return KVM_PFN_ERR_FAULT
;
1295 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1299 down_read(¤t
->mm
->mmap_sem
);
1300 if (npages
== -EHWPOISON
||
1301 (!async
&& check_user_page_hwpoison(addr
))) {
1302 pfn
= KVM_PFN_ERR_HWPOISON
;
1306 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1309 pfn
= KVM_PFN_ERR_FAULT
;
1310 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1311 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1313 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1315 if (async
&& vma_is_valid(vma
, write_fault
))
1317 pfn
= KVM_PFN_ERR_FAULT
;
1320 up_read(¤t
->mm
->mmap_sem
);
1325 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1326 bool *async
, bool write_fault
, bool *writable
)
1328 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1330 if (addr
== KVM_HVA_ERR_RO_BAD
)
1331 return KVM_PFN_ERR_RO_FAULT
;
1333 if (kvm_is_error_hva(addr
))
1334 return KVM_PFN_NOSLOT
;
1336 /* Do not map writable pfn in the readonly memslot. */
1337 if (writable
&& memslot_is_readonly(slot
)) {
1342 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1346 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1347 bool write_fault
, bool *writable
)
1349 struct kvm_memory_slot
*slot
;
1354 slot
= gfn_to_memslot(kvm
, gfn
);
1356 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1360 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1362 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1364 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1366 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1367 bool write_fault
, bool *writable
)
1369 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1371 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1373 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1375 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1377 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1379 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1382 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1384 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1386 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1388 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1391 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1393 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1395 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1397 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1403 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1404 if (kvm_is_error_hva(addr
))
1407 if (entry
< nr_pages
)
1410 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1412 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1414 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1416 if (is_error_noslot_pfn(pfn
))
1417 return KVM_ERR_PTR_BAD_PAGE
;
1419 if (kvm_is_mmio_pfn(pfn
)) {
1421 return KVM_ERR_PTR_BAD_PAGE
;
1424 return pfn_to_page(pfn
);
1427 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1431 pfn
= gfn_to_pfn(kvm
, gfn
);
1433 return kvm_pfn_to_page(pfn
);
1436 EXPORT_SYMBOL_GPL(gfn_to_page
);
1438 void kvm_release_page_clean(struct page
*page
)
1440 WARN_ON(is_error_page(page
));
1442 kvm_release_pfn_clean(page_to_pfn(page
));
1444 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1446 void kvm_release_pfn_clean(pfn_t pfn
)
1448 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1449 put_page(pfn_to_page(pfn
));
1451 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1453 void kvm_release_page_dirty(struct page
*page
)
1455 WARN_ON(is_error_page(page
));
1457 kvm_release_pfn_dirty(page_to_pfn(page
));
1459 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1461 static void kvm_release_pfn_dirty(pfn_t pfn
)
1463 kvm_set_pfn_dirty(pfn
);
1464 kvm_release_pfn_clean(pfn
);
1467 void kvm_set_pfn_dirty(pfn_t pfn
)
1469 if (!kvm_is_mmio_pfn(pfn
)) {
1470 struct page
*page
= pfn_to_page(pfn
);
1471 if (!PageReserved(page
))
1475 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1477 void kvm_set_pfn_accessed(pfn_t pfn
)
1479 if (!kvm_is_mmio_pfn(pfn
))
1480 mark_page_accessed(pfn_to_page(pfn
));
1482 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1484 void kvm_get_pfn(pfn_t pfn
)
1486 if (!kvm_is_mmio_pfn(pfn
))
1487 get_page(pfn_to_page(pfn
));
1489 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1491 static int next_segment(unsigned long len
, int offset
)
1493 if (len
> PAGE_SIZE
- offset
)
1494 return PAGE_SIZE
- offset
;
1499 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1505 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1506 if (kvm_is_error_hva(addr
))
1508 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1513 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1515 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1517 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1519 int offset
= offset_in_page(gpa
);
1522 while ((seg
= next_segment(len
, offset
)) != 0) {
1523 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1533 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1535 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1540 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1541 int offset
= offset_in_page(gpa
);
1543 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1544 if (kvm_is_error_hva(addr
))
1546 pagefault_disable();
1547 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1553 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1555 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1556 int offset
, int len
)
1561 addr
= gfn_to_hva(kvm
, gfn
);
1562 if (kvm_is_error_hva(addr
))
1564 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1567 mark_page_dirty(kvm
, gfn
);
1570 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1572 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1575 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1577 int offset
= offset_in_page(gpa
);
1580 while ((seg
= next_segment(len
, offset
)) != 0) {
1581 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1592 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1593 gpa_t gpa
, unsigned long len
)
1595 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1596 int offset
= offset_in_page(gpa
);
1597 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1598 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1599 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1600 gfn_t nr_pages_avail
;
1603 ghc
->generation
= slots
->generation
;
1605 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1606 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, &nr_pages_avail
);
1607 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_avail
>= nr_pages_needed
) {
1611 * If the requested region crosses two memslots, we still
1612 * verify that the entire region is valid here.
1614 while (start_gfn
<= end_gfn
) {
1615 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1616 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1618 if (kvm_is_error_hva(ghc
->hva
))
1620 start_gfn
+= nr_pages_avail
;
1622 /* Use the slow path for cross page reads and writes. */
1623 ghc
->memslot
= NULL
;
1627 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1629 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1630 void *data
, unsigned long len
)
1632 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1635 BUG_ON(len
> ghc
->len
);
1637 if (slots
->generation
!= ghc
->generation
)
1638 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1640 if (unlikely(!ghc
->memslot
))
1641 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1643 if (kvm_is_error_hva(ghc
->hva
))
1646 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1649 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1653 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1655 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1656 void *data
, unsigned long len
)
1658 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1661 BUG_ON(len
> ghc
->len
);
1663 if (slots
->generation
!= ghc
->generation
)
1664 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1666 if (unlikely(!ghc
->memslot
))
1667 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1669 if (kvm_is_error_hva(ghc
->hva
))
1672 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1678 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1680 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1682 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1684 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1686 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1688 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1690 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1692 int offset
= offset_in_page(gpa
);
1695 while ((seg
= next_segment(len
, offset
)) != 0) {
1696 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1705 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1707 static void mark_page_dirty_in_slot(struct kvm
*kvm
,
1708 struct kvm_memory_slot
*memslot
,
1711 if (memslot
&& memslot
->dirty_bitmap
) {
1712 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1714 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1718 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1720 struct kvm_memory_slot
*memslot
;
1722 memslot
= gfn_to_memslot(kvm
, gfn
);
1723 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1725 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1728 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1730 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1735 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1737 if (kvm_arch_vcpu_runnable(vcpu
)) {
1738 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1741 if (kvm_cpu_has_pending_timer(vcpu
))
1743 if (signal_pending(current
))
1749 finish_wait(&vcpu
->wq
, &wait
);
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
1755 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1757 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1760 int cpu
= vcpu
->cpu
;
1761 wait_queue_head_t
*wqp
;
1763 wqp
= kvm_arch_vcpu_wq(vcpu
);
1764 if (waitqueue_active(wqp
)) {
1765 wake_up_interruptible(wqp
);
1766 ++vcpu
->stat
.halt_wakeup
;
1770 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1771 if (kvm_arch_vcpu_should_kick(vcpu
))
1772 smp_send_reschedule(cpu
);
1775 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1776 #endif /* !CONFIG_S390 */
1778 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1781 struct task_struct
*task
= NULL
;
1785 pid
= rcu_dereference(target
->pid
);
1787 task
= get_pid_task(pid
, PIDTYPE_PID
);
1791 if (task
->flags
& PF_VCPU
) {
1792 put_task_struct(task
);
1795 ret
= yield_to(task
, 1);
1796 put_task_struct(task
);
1800 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1803 * Helper that checks whether a VCPU is eligible for directed yield.
1804 * Most eligible candidate to yield is decided by following heuristics:
1806 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1807 * (preempted lock holder), indicated by @in_spin_loop.
1808 * Set at the beiginning and cleared at the end of interception/PLE handler.
1810 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1811 * chance last time (mostly it has become eligible now since we have probably
1812 * yielded to lockholder in last iteration. This is done by toggling
1813 * @dy_eligible each time a VCPU checked for eligibility.)
1815 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1816 * to preempted lock-holder could result in wrong VCPU selection and CPU
1817 * burning. Giving priority for a potential lock-holder increases lock
1820 * Since algorithm is based on heuristics, accessing another VCPU data without
1821 * locking does not harm. It may result in trying to yield to same VCPU, fail
1822 * and continue with next VCPU and so on.
1824 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1826 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1829 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1830 vcpu
->spin_loop
.dy_eligible
;
1832 if (vcpu
->spin_loop
.in_spin_loop
)
1833 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1841 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1843 struct kvm
*kvm
= me
->kvm
;
1844 struct kvm_vcpu
*vcpu
;
1845 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1851 kvm_vcpu_set_in_spin_loop(me
, true);
1853 * We boost the priority of a VCPU that is runnable but not
1854 * currently running, because it got preempted by something
1855 * else and called schedule in __vcpu_run. Hopefully that
1856 * VCPU is holding the lock that we need and will release it.
1857 * We approximate round-robin by starting at the last boosted VCPU.
1859 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1860 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1861 if (!pass
&& i
<= last_boosted_vcpu
) {
1862 i
= last_boosted_vcpu
;
1864 } else if (pass
&& i
> last_boosted_vcpu
)
1866 if (!ACCESS_ONCE(vcpu
->preempted
))
1870 if (waitqueue_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
1872 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1875 yielded
= kvm_vcpu_yield_to(vcpu
);
1877 kvm
->last_boosted_vcpu
= i
;
1879 } else if (yielded
< 0) {
1886 kvm_vcpu_set_in_spin_loop(me
, false);
1888 /* Ensure vcpu is not eligible during next spinloop */
1889 kvm_vcpu_set_dy_eligible(me
, false);
1891 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1893 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1895 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1898 if (vmf
->pgoff
== 0)
1899 page
= virt_to_page(vcpu
->run
);
1901 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1902 page
= virt_to_page(vcpu
->arch
.pio_data
);
1904 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1905 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1906 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1909 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1915 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1916 .fault
= kvm_vcpu_fault
,
1919 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1921 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1925 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1927 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1929 kvm_put_kvm(vcpu
->kvm
);
1933 static struct file_operations kvm_vcpu_fops
= {
1934 .release
= kvm_vcpu_release
,
1935 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1936 #ifdef CONFIG_COMPAT
1937 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1939 .mmap
= kvm_vcpu_mmap
,
1940 .llseek
= noop_llseek
,
1944 * Allocates an inode for the vcpu.
1946 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1948 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
1952 * Creates some virtual cpus. Good luck creating more than one.
1954 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1957 struct kvm_vcpu
*vcpu
, *v
;
1959 if (id
>= KVM_MAX_VCPUS
)
1962 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1964 return PTR_ERR(vcpu
);
1966 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1968 r
= kvm_arch_vcpu_setup(vcpu
);
1972 mutex_lock(&kvm
->lock
);
1973 if (!kvm_vcpu_compatible(vcpu
)) {
1975 goto unlock_vcpu_destroy
;
1977 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1979 goto unlock_vcpu_destroy
;
1982 kvm_for_each_vcpu(r
, v
, kvm
)
1983 if (v
->vcpu_id
== id
) {
1985 goto unlock_vcpu_destroy
;
1988 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1990 /* Now it's all set up, let userspace reach it */
1992 r
= create_vcpu_fd(vcpu
);
1995 goto unlock_vcpu_destroy
;
1998 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2000 atomic_inc(&kvm
->online_vcpus
);
2002 mutex_unlock(&kvm
->lock
);
2003 kvm_arch_vcpu_postcreate(vcpu
);
2006 unlock_vcpu_destroy
:
2007 mutex_unlock(&kvm
->lock
);
2009 kvm_arch_vcpu_destroy(vcpu
);
2013 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2016 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2017 vcpu
->sigset_active
= 1;
2018 vcpu
->sigset
= *sigset
;
2020 vcpu
->sigset_active
= 0;
2024 static long kvm_vcpu_ioctl(struct file
*filp
,
2025 unsigned int ioctl
, unsigned long arg
)
2027 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2028 void __user
*argp
= (void __user
*)arg
;
2030 struct kvm_fpu
*fpu
= NULL
;
2031 struct kvm_sregs
*kvm_sregs
= NULL
;
2033 if (vcpu
->kvm
->mm
!= current
->mm
)
2036 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2039 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2041 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2042 * so vcpu_load() would break it.
2044 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
2045 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2049 r
= vcpu_load(vcpu
);
2057 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2058 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2060 case KVM_GET_REGS
: {
2061 struct kvm_regs
*kvm_regs
;
2064 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2067 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2071 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2078 case KVM_SET_REGS
: {
2079 struct kvm_regs
*kvm_regs
;
2082 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2083 if (IS_ERR(kvm_regs
)) {
2084 r
= PTR_ERR(kvm_regs
);
2087 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2091 case KVM_GET_SREGS
: {
2092 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2096 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2100 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2105 case KVM_SET_SREGS
: {
2106 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2107 if (IS_ERR(kvm_sregs
)) {
2108 r
= PTR_ERR(kvm_sregs
);
2112 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2115 case KVM_GET_MP_STATE
: {
2116 struct kvm_mp_state mp_state
;
2118 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2122 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2127 case KVM_SET_MP_STATE
: {
2128 struct kvm_mp_state mp_state
;
2131 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2133 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2136 case KVM_TRANSLATE
: {
2137 struct kvm_translation tr
;
2140 if (copy_from_user(&tr
, argp
, sizeof tr
))
2142 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2146 if (copy_to_user(argp
, &tr
, sizeof tr
))
2151 case KVM_SET_GUEST_DEBUG
: {
2152 struct kvm_guest_debug dbg
;
2155 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2157 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2160 case KVM_SET_SIGNAL_MASK
: {
2161 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2162 struct kvm_signal_mask kvm_sigmask
;
2163 sigset_t sigset
, *p
;
2168 if (copy_from_user(&kvm_sigmask
, argp
,
2169 sizeof kvm_sigmask
))
2172 if (kvm_sigmask
.len
!= sizeof sigset
)
2175 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2180 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2184 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2188 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2192 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2198 fpu
= memdup_user(argp
, sizeof(*fpu
));
2204 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2208 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2217 #ifdef CONFIG_COMPAT
2218 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2219 unsigned int ioctl
, unsigned long arg
)
2221 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2222 void __user
*argp
= compat_ptr(arg
);
2225 if (vcpu
->kvm
->mm
!= current
->mm
)
2229 case KVM_SET_SIGNAL_MASK
: {
2230 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2231 struct kvm_signal_mask kvm_sigmask
;
2232 compat_sigset_t csigset
;
2237 if (copy_from_user(&kvm_sigmask
, argp
,
2238 sizeof kvm_sigmask
))
2241 if (kvm_sigmask
.len
!= sizeof csigset
)
2244 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2247 sigset_from_compat(&sigset
, &csigset
);
2248 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2250 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2254 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2262 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2263 int (*accessor
)(struct kvm_device
*dev
,
2264 struct kvm_device_attr
*attr
),
2267 struct kvm_device_attr attr
;
2272 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2275 return accessor(dev
, &attr
);
2278 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2281 struct kvm_device
*dev
= filp
->private_data
;
2284 case KVM_SET_DEVICE_ATTR
:
2285 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2286 case KVM_GET_DEVICE_ATTR
:
2287 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2288 case KVM_HAS_DEVICE_ATTR
:
2289 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2291 if (dev
->ops
->ioctl
)
2292 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2298 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2300 struct kvm_device
*dev
= filp
->private_data
;
2301 struct kvm
*kvm
= dev
->kvm
;
2307 static const struct file_operations kvm_device_fops
= {
2308 .unlocked_ioctl
= kvm_device_ioctl
,
2309 #ifdef CONFIG_COMPAT
2310 .compat_ioctl
= kvm_device_ioctl
,
2312 .release
= kvm_device_release
,
2315 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2317 if (filp
->f_op
!= &kvm_device_fops
)
2320 return filp
->private_data
;
2323 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2324 #ifdef CONFIG_KVM_MPIC
2325 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2326 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2329 #ifdef CONFIG_KVM_XICS
2330 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2334 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2336 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2339 if (kvm_device_ops_table
[type
] != NULL
)
2342 kvm_device_ops_table
[type
] = ops
;
2346 void kvm_unregister_device_ops(u32 type
)
2348 if (kvm_device_ops_table
[type
] != NULL
)
2349 kvm_device_ops_table
[type
] = NULL
;
2352 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2353 struct kvm_create_device
*cd
)
2355 struct kvm_device_ops
*ops
= NULL
;
2356 struct kvm_device
*dev
;
2357 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2360 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2363 ops
= kvm_device_ops_table
[cd
->type
];
2370 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2377 ret
= ops
->create(dev
, cd
->type
);
2383 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2389 list_add(&dev
->vm_node
, &kvm
->devices
);
2395 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2398 case KVM_CAP_USER_MEMORY
:
2399 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2400 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2401 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2402 case KVM_CAP_SET_BOOT_CPU_ID
:
2404 case KVM_CAP_INTERNAL_ERROR_DATA
:
2405 #ifdef CONFIG_HAVE_KVM_MSI
2406 case KVM_CAP_SIGNAL_MSI
:
2408 #ifdef CONFIG_HAVE_KVM_IRQFD
2409 case KVM_CAP_IRQFD_RESAMPLE
:
2411 case KVM_CAP_CHECK_EXTENSION_VM
:
2413 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2414 case KVM_CAP_IRQ_ROUTING
:
2415 return KVM_MAX_IRQ_ROUTES
;
2420 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2423 static long kvm_vm_ioctl(struct file
*filp
,
2424 unsigned int ioctl
, unsigned long arg
)
2426 struct kvm
*kvm
= filp
->private_data
;
2427 void __user
*argp
= (void __user
*)arg
;
2430 if (kvm
->mm
!= current
->mm
)
2433 case KVM_CREATE_VCPU
:
2434 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2436 case KVM_SET_USER_MEMORY_REGION
: {
2437 struct kvm_userspace_memory_region kvm_userspace_mem
;
2440 if (copy_from_user(&kvm_userspace_mem
, argp
,
2441 sizeof kvm_userspace_mem
))
2444 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2447 case KVM_GET_DIRTY_LOG
: {
2448 struct kvm_dirty_log log
;
2451 if (copy_from_user(&log
, argp
, sizeof log
))
2453 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2456 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2457 case KVM_REGISTER_COALESCED_MMIO
: {
2458 struct kvm_coalesced_mmio_zone zone
;
2460 if (copy_from_user(&zone
, argp
, sizeof zone
))
2462 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2465 case KVM_UNREGISTER_COALESCED_MMIO
: {
2466 struct kvm_coalesced_mmio_zone zone
;
2468 if (copy_from_user(&zone
, argp
, sizeof zone
))
2470 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2475 struct kvm_irqfd data
;
2478 if (copy_from_user(&data
, argp
, sizeof data
))
2480 r
= kvm_irqfd(kvm
, &data
);
2483 case KVM_IOEVENTFD
: {
2484 struct kvm_ioeventfd data
;
2487 if (copy_from_user(&data
, argp
, sizeof data
))
2489 r
= kvm_ioeventfd(kvm
, &data
);
2492 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2493 case KVM_SET_BOOT_CPU_ID
:
2495 mutex_lock(&kvm
->lock
);
2496 if (atomic_read(&kvm
->online_vcpus
) != 0)
2499 kvm
->bsp_vcpu_id
= arg
;
2500 mutex_unlock(&kvm
->lock
);
2503 #ifdef CONFIG_HAVE_KVM_MSI
2504 case KVM_SIGNAL_MSI
: {
2508 if (copy_from_user(&msi
, argp
, sizeof msi
))
2510 r
= kvm_send_userspace_msi(kvm
, &msi
);
2514 #ifdef __KVM_HAVE_IRQ_LINE
2515 case KVM_IRQ_LINE_STATUS
:
2516 case KVM_IRQ_LINE
: {
2517 struct kvm_irq_level irq_event
;
2520 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2523 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2524 ioctl
== KVM_IRQ_LINE_STATUS
);
2529 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2530 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2538 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2539 case KVM_SET_GSI_ROUTING
: {
2540 struct kvm_irq_routing routing
;
2541 struct kvm_irq_routing __user
*urouting
;
2542 struct kvm_irq_routing_entry
*entries
;
2545 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2548 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2553 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2558 if (copy_from_user(entries
, urouting
->entries
,
2559 routing
.nr
* sizeof(*entries
)))
2560 goto out_free_irq_routing
;
2561 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2563 out_free_irq_routing
:
2567 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2568 case KVM_CREATE_DEVICE
: {
2569 struct kvm_create_device cd
;
2572 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2575 r
= kvm_ioctl_create_device(kvm
, &cd
);
2580 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2586 case KVM_CHECK_EXTENSION
:
2587 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
2590 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2596 #ifdef CONFIG_COMPAT
2597 struct compat_kvm_dirty_log
{
2601 compat_uptr_t dirty_bitmap
; /* one bit per page */
2606 static long kvm_vm_compat_ioctl(struct file
*filp
,
2607 unsigned int ioctl
, unsigned long arg
)
2609 struct kvm
*kvm
= filp
->private_data
;
2612 if (kvm
->mm
!= current
->mm
)
2615 case KVM_GET_DIRTY_LOG
: {
2616 struct compat_kvm_dirty_log compat_log
;
2617 struct kvm_dirty_log log
;
2620 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2621 sizeof(compat_log
)))
2623 log
.slot
= compat_log
.slot
;
2624 log
.padding1
= compat_log
.padding1
;
2625 log
.padding2
= compat_log
.padding2
;
2626 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2628 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2632 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2640 static struct file_operations kvm_vm_fops
= {
2641 .release
= kvm_vm_release
,
2642 .unlocked_ioctl
= kvm_vm_ioctl
,
2643 #ifdef CONFIG_COMPAT
2644 .compat_ioctl
= kvm_vm_compat_ioctl
,
2646 .llseek
= noop_llseek
,
2649 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2654 kvm
= kvm_create_vm(type
);
2656 return PTR_ERR(kvm
);
2657 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2658 r
= kvm_coalesced_mmio_init(kvm
);
2664 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2671 static long kvm_dev_ioctl(struct file
*filp
,
2672 unsigned int ioctl
, unsigned long arg
)
2677 case KVM_GET_API_VERSION
:
2680 r
= KVM_API_VERSION
;
2683 r
= kvm_dev_ioctl_create_vm(arg
);
2685 case KVM_CHECK_EXTENSION
:
2686 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
2688 case KVM_GET_VCPU_MMAP_SIZE
:
2691 r
= PAGE_SIZE
; /* struct kvm_run */
2693 r
+= PAGE_SIZE
; /* pio data page */
2695 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2696 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2699 case KVM_TRACE_ENABLE
:
2700 case KVM_TRACE_PAUSE
:
2701 case KVM_TRACE_DISABLE
:
2705 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2711 static struct file_operations kvm_chardev_ops
= {
2712 .unlocked_ioctl
= kvm_dev_ioctl
,
2713 .compat_ioctl
= kvm_dev_ioctl
,
2714 .llseek
= noop_llseek
,
2717 static struct miscdevice kvm_dev
= {
2723 static void hardware_enable_nolock(void *junk
)
2725 int cpu
= raw_smp_processor_id();
2728 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2731 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2733 r
= kvm_arch_hardware_enable();
2736 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2737 atomic_inc(&hardware_enable_failed
);
2738 printk(KERN_INFO
"kvm: enabling virtualization on "
2739 "CPU%d failed\n", cpu
);
2743 static void hardware_enable(void)
2745 raw_spin_lock(&kvm_count_lock
);
2746 if (kvm_usage_count
)
2747 hardware_enable_nolock(NULL
);
2748 raw_spin_unlock(&kvm_count_lock
);
2751 static void hardware_disable_nolock(void *junk
)
2753 int cpu
= raw_smp_processor_id();
2755 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2757 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2758 kvm_arch_hardware_disable();
2761 static void hardware_disable(void)
2763 raw_spin_lock(&kvm_count_lock
);
2764 if (kvm_usage_count
)
2765 hardware_disable_nolock(NULL
);
2766 raw_spin_unlock(&kvm_count_lock
);
2769 static void hardware_disable_all_nolock(void)
2771 BUG_ON(!kvm_usage_count
);
2774 if (!kvm_usage_count
)
2775 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2778 static void hardware_disable_all(void)
2780 raw_spin_lock(&kvm_count_lock
);
2781 hardware_disable_all_nolock();
2782 raw_spin_unlock(&kvm_count_lock
);
2785 static int hardware_enable_all(void)
2789 raw_spin_lock(&kvm_count_lock
);
2792 if (kvm_usage_count
== 1) {
2793 atomic_set(&hardware_enable_failed
, 0);
2794 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2796 if (atomic_read(&hardware_enable_failed
)) {
2797 hardware_disable_all_nolock();
2802 raw_spin_unlock(&kvm_count_lock
);
2807 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2812 val
&= ~CPU_TASKS_FROZEN
;
2815 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2820 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2828 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2832 * Some (well, at least mine) BIOSes hang on reboot if
2835 * And Intel TXT required VMX off for all cpu when system shutdown.
2837 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2838 kvm_rebooting
= true;
2839 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2843 static struct notifier_block kvm_reboot_notifier
= {
2844 .notifier_call
= kvm_reboot
,
2848 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2852 for (i
= 0; i
< bus
->dev_count
; i
++) {
2853 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2855 kvm_iodevice_destructor(pos
);
2860 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
2861 const struct kvm_io_range
*r2
)
2863 if (r1
->addr
< r2
->addr
)
2865 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2870 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2872 return kvm_io_bus_cmp(p1
, p2
);
2875 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2876 gpa_t addr
, int len
)
2878 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2884 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2885 kvm_io_bus_sort_cmp
, NULL
);
2890 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2891 gpa_t addr
, int len
)
2893 struct kvm_io_range
*range
, key
;
2896 key
= (struct kvm_io_range
) {
2901 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2902 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2906 off
= range
- bus
->range
;
2908 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
2914 static int __kvm_io_bus_write(struct kvm_io_bus
*bus
,
2915 struct kvm_io_range
*range
, const void *val
)
2919 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2923 while (idx
< bus
->dev_count
&&
2924 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2925 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, range
->addr
,
2934 /* kvm_io_bus_write - called under kvm->slots_lock */
2935 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2936 int len
, const void *val
)
2938 struct kvm_io_bus
*bus
;
2939 struct kvm_io_range range
;
2942 range
= (struct kvm_io_range
) {
2947 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2948 r
= __kvm_io_bus_write(bus
, &range
, val
);
2949 return r
< 0 ? r
: 0;
2952 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2953 int kvm_io_bus_write_cookie(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2954 int len
, const void *val
, long cookie
)
2956 struct kvm_io_bus
*bus
;
2957 struct kvm_io_range range
;
2959 range
= (struct kvm_io_range
) {
2964 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2966 /* First try the device referenced by cookie. */
2967 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
2968 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
2969 if (!kvm_iodevice_write(bus
->range
[cookie
].dev
, addr
, len
,
2974 * cookie contained garbage; fall back to search and return the
2975 * correct cookie value.
2977 return __kvm_io_bus_write(bus
, &range
, val
);
2980 static int __kvm_io_bus_read(struct kvm_io_bus
*bus
, struct kvm_io_range
*range
,
2985 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2989 while (idx
< bus
->dev_count
&&
2990 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2991 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, range
->addr
,
2999 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3001 /* kvm_io_bus_read - called under kvm->slots_lock */
3002 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3005 struct kvm_io_bus
*bus
;
3006 struct kvm_io_range range
;
3009 range
= (struct kvm_io_range
) {
3014 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3015 r
= __kvm_io_bus_read(bus
, &range
, val
);
3016 return r
< 0 ? r
: 0;
3020 /* Caller must hold slots_lock. */
3021 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3022 int len
, struct kvm_io_device
*dev
)
3024 struct kvm_io_bus
*new_bus
, *bus
;
3026 bus
= kvm
->buses
[bus_idx
];
3027 /* exclude ioeventfd which is limited by maximum fd */
3028 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3031 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3032 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3035 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3036 sizeof(struct kvm_io_range
)));
3037 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3038 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3039 synchronize_srcu_expedited(&kvm
->srcu
);
3045 /* Caller must hold slots_lock. */
3046 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3047 struct kvm_io_device
*dev
)
3050 struct kvm_io_bus
*new_bus
, *bus
;
3052 bus
= kvm
->buses
[bus_idx
];
3054 for (i
= 0; i
< bus
->dev_count
; i
++)
3055 if (bus
->range
[i
].dev
== dev
) {
3063 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3064 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3068 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3069 new_bus
->dev_count
--;
3070 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3071 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3073 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3074 synchronize_srcu_expedited(&kvm
->srcu
);
3079 static struct notifier_block kvm_cpu_notifier
= {
3080 .notifier_call
= kvm_cpu_hotplug
,
3083 static int vm_stat_get(void *_offset
, u64
*val
)
3085 unsigned offset
= (long)_offset
;
3089 spin_lock(&kvm_lock
);
3090 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3091 *val
+= *(u32
*)((void *)kvm
+ offset
);
3092 spin_unlock(&kvm_lock
);
3096 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3098 static int vcpu_stat_get(void *_offset
, u64
*val
)
3100 unsigned offset
= (long)_offset
;
3102 struct kvm_vcpu
*vcpu
;
3106 spin_lock(&kvm_lock
);
3107 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3108 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3109 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3111 spin_unlock(&kvm_lock
);
3115 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3117 static const struct file_operations
*stat_fops
[] = {
3118 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3119 [KVM_STAT_VM
] = &vm_stat_fops
,
3122 static int kvm_init_debug(void)
3125 struct kvm_stats_debugfs_item
*p
;
3127 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3128 if (kvm_debugfs_dir
== NULL
)
3131 for (p
= debugfs_entries
; p
->name
; ++p
) {
3132 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3133 (void *)(long)p
->offset
,
3134 stat_fops
[p
->kind
]);
3135 if (p
->dentry
== NULL
)
3142 debugfs_remove_recursive(kvm_debugfs_dir
);
3147 static void kvm_exit_debug(void)
3149 struct kvm_stats_debugfs_item
*p
;
3151 for (p
= debugfs_entries
; p
->name
; ++p
)
3152 debugfs_remove(p
->dentry
);
3153 debugfs_remove(kvm_debugfs_dir
);
3156 static int kvm_suspend(void)
3158 if (kvm_usage_count
)
3159 hardware_disable_nolock(NULL
);
3163 static void kvm_resume(void)
3165 if (kvm_usage_count
) {
3166 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3167 hardware_enable_nolock(NULL
);
3171 static struct syscore_ops kvm_syscore_ops
= {
3172 .suspend
= kvm_suspend
,
3173 .resume
= kvm_resume
,
3177 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3179 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3182 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3184 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3185 if (vcpu
->preempted
)
3186 vcpu
->preempted
= false;
3188 kvm_arch_sched_in(vcpu
, cpu
);
3190 kvm_arch_vcpu_load(vcpu
, cpu
);
3193 static void kvm_sched_out(struct preempt_notifier
*pn
,
3194 struct task_struct
*next
)
3196 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3198 if (current
->state
== TASK_RUNNING
)
3199 vcpu
->preempted
= true;
3200 kvm_arch_vcpu_put(vcpu
);
3203 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3204 struct module
*module
)
3209 r
= kvm_arch_init(opaque
);
3214 * kvm_arch_init makes sure there's at most one caller
3215 * for architectures that support multiple implementations,
3216 * like intel and amd on x86.
3217 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3218 * conflicts in case kvm is already setup for another implementation.
3220 r
= kvm_irqfd_init();
3224 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3229 r
= kvm_arch_hardware_setup();
3233 for_each_online_cpu(cpu
) {
3234 smp_call_function_single(cpu
,
3235 kvm_arch_check_processor_compat
,
3241 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3244 register_reboot_notifier(&kvm_reboot_notifier
);
3246 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3248 vcpu_align
= __alignof__(struct kvm_vcpu
);
3249 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3251 if (!kvm_vcpu_cache
) {
3256 r
= kvm_async_pf_init();
3260 kvm_chardev_ops
.owner
= module
;
3261 kvm_vm_fops
.owner
= module
;
3262 kvm_vcpu_fops
.owner
= module
;
3264 r
= misc_register(&kvm_dev
);
3266 printk(KERN_ERR
"kvm: misc device register failed\n");
3270 register_syscore_ops(&kvm_syscore_ops
);
3272 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3273 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3275 r
= kvm_init_debug();
3277 printk(KERN_ERR
"kvm: create debugfs files failed\n");
3281 r
= kvm_vfio_ops_init();
3287 unregister_syscore_ops(&kvm_syscore_ops
);
3288 misc_deregister(&kvm_dev
);
3290 kvm_async_pf_deinit();
3292 kmem_cache_destroy(kvm_vcpu_cache
);
3294 unregister_reboot_notifier(&kvm_reboot_notifier
);
3295 unregister_cpu_notifier(&kvm_cpu_notifier
);
3298 kvm_arch_hardware_unsetup();
3300 free_cpumask_var(cpus_hardware_enabled
);
3308 EXPORT_SYMBOL_GPL(kvm_init
);
3313 misc_deregister(&kvm_dev
);
3314 kmem_cache_destroy(kvm_vcpu_cache
);
3315 kvm_async_pf_deinit();
3316 unregister_syscore_ops(&kvm_syscore_ops
);
3317 unregister_reboot_notifier(&kvm_reboot_notifier
);
3318 unregister_cpu_notifier(&kvm_cpu_notifier
);
3319 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3320 kvm_arch_hardware_unsetup();
3323 free_cpumask_var(cpus_hardware_enabled
);
3324 kvm_vfio_ops_exit();
3326 EXPORT_SYMBOL_GPL(kvm_exit
);