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
)
672 * Insert memslot and re-sort memslots based on their size,
673 * so the larger slots will get better fit. Sorting algorithm
674 * takes advantage of having initially sorted array and
675 * known changed memslot position.
677 static void update_memslots(struct kvm_memslots
*slots
,
678 struct kvm_memory_slot
*new)
681 int i
= slots
->id_to_index
[id
];
682 struct kvm_memory_slot
*mslots
= slots
->memslots
;
684 WARN_ON(mslots
[i
].id
!= id
);
685 if (new->npages
!= mslots
[i
].npages
) {
686 if (new->npages
< mslots
[i
].npages
) {
687 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
688 new->npages
< mslots
[i
+ 1].npages
) {
689 mslots
[i
] = mslots
[i
+ 1];
690 slots
->id_to_index
[mslots
[i
].id
] = i
;
695 new->npages
> mslots
[i
- 1].npages
) {
696 mslots
[i
] = mslots
[i
- 1];
697 slots
->id_to_index
[mslots
[i
].id
] = i
;
704 slots
->id_to_index
[mslots
[i
].id
] = i
;
707 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
709 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
711 #ifdef __KVM_HAVE_READONLY_MEM
712 valid_flags
|= KVM_MEM_READONLY
;
715 if (mem
->flags
& ~valid_flags
)
721 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
722 struct kvm_memslots
*slots
)
724 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
727 * Set the low bit in the generation, which disables SPTE caching
728 * until the end of synchronize_srcu_expedited.
730 WARN_ON(old_memslots
->generation
& 1);
731 slots
->generation
= old_memslots
->generation
+ 1;
733 rcu_assign_pointer(kvm
->memslots
, slots
);
734 synchronize_srcu_expedited(&kvm
->srcu
);
737 * Increment the new memslot generation a second time. This prevents
738 * vm exits that race with memslot updates from caching a memslot
739 * generation that will (potentially) be valid forever.
743 kvm_arch_memslots_updated(kvm
);
749 * Allocate some memory and give it an address in the guest physical address
752 * Discontiguous memory is allowed, mostly for framebuffers.
754 * Must be called holding kvm->slots_lock for write.
756 int __kvm_set_memory_region(struct kvm
*kvm
,
757 struct kvm_userspace_memory_region
*mem
)
761 unsigned long npages
;
762 struct kvm_memory_slot
*slot
;
763 struct kvm_memory_slot old
, new;
764 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
765 enum kvm_mr_change change
;
767 r
= check_memory_region_flags(mem
);
772 /* General sanity checks */
773 if (mem
->memory_size
& (PAGE_SIZE
- 1))
775 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
777 /* We can read the guest memory with __xxx_user() later on. */
778 if ((mem
->slot
< KVM_USER_MEM_SLOTS
) &&
779 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
780 !access_ok(VERIFY_WRITE
,
781 (void __user
*)(unsigned long)mem
->userspace_addr
,
784 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
786 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
789 slot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
790 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
791 npages
= mem
->memory_size
>> PAGE_SHIFT
;
793 if (npages
> KVM_MEM_MAX_NR_PAGES
)
797 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
802 new.base_gfn
= base_gfn
;
804 new.flags
= mem
->flags
;
808 change
= KVM_MR_CREATE
;
809 else { /* Modify an existing slot. */
810 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
811 (npages
!= old
.npages
) ||
812 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
815 if (base_gfn
!= old
.base_gfn
)
816 change
= KVM_MR_MOVE
;
817 else if (new.flags
!= old
.flags
)
818 change
= KVM_MR_FLAGS_ONLY
;
819 else { /* Nothing to change. */
824 } else if (old
.npages
) {
825 change
= KVM_MR_DELETE
;
826 } else /* Modify a non-existent slot: disallowed. */
829 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
830 /* Check for overlaps */
832 kvm_for_each_memslot(slot
, kvm
->memslots
) {
833 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
834 (slot
->id
== mem
->slot
))
836 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
837 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
842 /* Free page dirty bitmap if unneeded */
843 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
844 new.dirty_bitmap
= NULL
;
847 if (change
== KVM_MR_CREATE
) {
848 new.userspace_addr
= mem
->userspace_addr
;
850 if (kvm_arch_create_memslot(kvm
, &new, npages
))
854 /* Allocate page dirty bitmap if needed */
855 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
856 if (kvm_create_dirty_bitmap(&new) < 0)
860 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
865 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
866 slot
= id_to_memslot(slots
, mem
->slot
);
867 slot
->flags
|= KVM_MEMSLOT_INVALID
;
869 old_memslots
= install_new_memslots(kvm
, slots
);
871 /* slot was deleted or moved, clear iommu mapping */
872 kvm_iommu_unmap_pages(kvm
, &old
);
873 /* From this point no new shadow pages pointing to a deleted,
874 * or moved, memslot will be created.
876 * validation of sp->gfn happens in:
877 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
878 * - kvm_is_visible_gfn (mmu_check_roots)
880 kvm_arch_flush_shadow_memslot(kvm
, slot
);
883 * We can re-use the old_memslots from above, the only difference
884 * from the currently installed memslots is the invalid flag. This
885 * will get overwritten by update_memslots anyway.
887 slots
= old_memslots
;
890 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
894 /* actual memory is freed via old in kvm_free_physmem_slot below */
895 if (change
== KVM_MR_DELETE
) {
896 new.dirty_bitmap
= NULL
;
897 memset(&new.arch
, 0, sizeof(new.arch
));
900 update_memslots(slots
, &new);
901 old_memslots
= install_new_memslots(kvm
, slots
);
903 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
905 kvm_free_physmem_slot(kvm
, &old
, &new);
909 * IOMMU mapping: New slots need to be mapped. Old slots need to be
910 * un-mapped and re-mapped if their base changes. Since base change
911 * unmapping is handled above with slot deletion, mapping alone is
912 * needed here. Anything else the iommu might care about for existing
913 * slots (size changes, userspace addr changes and read-only flag
914 * changes) is disallowed above, so any other attribute changes getting
915 * here can be skipped.
917 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
918 r
= kvm_iommu_map_pages(kvm
, &new);
927 kvm_free_physmem_slot(kvm
, &new, &old
);
931 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
933 int kvm_set_memory_region(struct kvm
*kvm
,
934 struct kvm_userspace_memory_region
*mem
)
938 mutex_lock(&kvm
->slots_lock
);
939 r
= __kvm_set_memory_region(kvm
, mem
);
940 mutex_unlock(&kvm
->slots_lock
);
943 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
945 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
946 struct kvm_userspace_memory_region
*mem
)
948 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
950 return kvm_set_memory_region(kvm
, mem
);
953 int kvm_get_dirty_log(struct kvm
*kvm
,
954 struct kvm_dirty_log
*log
, int *is_dirty
)
956 struct kvm_memory_slot
*memslot
;
959 unsigned long any
= 0;
962 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
965 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
967 if (!memslot
->dirty_bitmap
)
970 n
= kvm_dirty_bitmap_bytes(memslot
);
972 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
973 any
= memslot
->dirty_bitmap
[i
];
976 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
986 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
988 bool kvm_largepages_enabled(void)
990 return largepages_enabled
;
993 void kvm_disable_largepages(void)
995 largepages_enabled
= false;
997 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
999 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1001 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1003 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1005 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1007 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1009 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1010 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1015 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1017 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1019 struct vm_area_struct
*vma
;
1020 unsigned long addr
, size
;
1024 addr
= gfn_to_hva(kvm
, gfn
);
1025 if (kvm_is_error_hva(addr
))
1028 down_read(¤t
->mm
->mmap_sem
);
1029 vma
= find_vma(current
->mm
, addr
);
1033 size
= vma_kernel_pagesize(vma
);
1036 up_read(¤t
->mm
->mmap_sem
);
1041 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1043 return slot
->flags
& KVM_MEM_READONLY
;
1046 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1047 gfn_t
*nr_pages
, bool write
)
1049 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1050 return KVM_HVA_ERR_BAD
;
1052 if (memslot_is_readonly(slot
) && write
)
1053 return KVM_HVA_ERR_RO_BAD
;
1056 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1058 return __gfn_to_hva_memslot(slot
, gfn
);
1061 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1064 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1067 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1070 return gfn_to_hva_many(slot
, gfn
, NULL
);
1072 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1074 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1076 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1078 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1081 * If writable is set to false, the hva returned by this function is only
1082 * allowed to be read.
1084 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1085 gfn_t gfn
, bool *writable
)
1087 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1089 if (!kvm_is_error_hva(hva
) && writable
)
1090 *writable
= !memslot_is_readonly(slot
);
1095 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1097 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1099 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1102 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1104 return __copy_from_user(data
, hva
, len
);
1107 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1109 return __copy_from_user_inatomic(data
, hva
, len
);
1112 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1113 unsigned long start
, int write
, struct page
**page
)
1115 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1118 flags
|= FOLL_WRITE
;
1120 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1123 int kvm_get_user_page_io(struct task_struct
*tsk
, struct mm_struct
*mm
,
1124 unsigned long addr
, bool write_fault
,
1125 struct page
**pagep
)
1129 int flags
= FOLL_TOUCH
| FOLL_HWPOISON
|
1130 (pagep
? FOLL_GET
: 0) |
1131 (write_fault
? FOLL_WRITE
: 0);
1134 * If retrying the fault, we get here *not* having allowed the filemap
1135 * to wait on the page lock. We should now allow waiting on the IO with
1136 * the mmap semaphore released.
1138 down_read(&mm
->mmap_sem
);
1139 npages
= __get_user_pages(tsk
, mm
, addr
, 1, flags
, pagep
, NULL
,
1148 * The previous call has now waited on the IO. Now we can
1149 * retry and complete. Pass TRIED to ensure we do not re
1150 * schedule async IO (see e.g. filemap_fault).
1152 down_read(&mm
->mmap_sem
);
1153 npages
= __get_user_pages(tsk
, mm
, addr
, 1, flags
| FOLL_TRIED
,
1156 up_read(&mm
->mmap_sem
);
1160 static inline int check_user_page_hwpoison(unsigned long addr
)
1162 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1164 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1165 flags
, NULL
, NULL
, NULL
);
1166 return rc
== -EHWPOISON
;
1170 * The atomic path to get the writable pfn which will be stored in @pfn,
1171 * true indicates success, otherwise false is returned.
1173 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1174 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1176 struct page
*page
[1];
1179 if (!(async
|| atomic
))
1183 * Fast pin a writable pfn only if it is a write fault request
1184 * or the caller allows to map a writable pfn for a read fault
1187 if (!(write_fault
|| writable
))
1190 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1192 *pfn
= page_to_pfn(page
[0]);
1203 * The slow path to get the pfn of the specified host virtual address,
1204 * 1 indicates success, -errno is returned if error is detected.
1206 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1207 bool *writable
, pfn_t
*pfn
)
1209 struct page
*page
[1];
1215 *writable
= write_fault
;
1218 down_read(¤t
->mm
->mmap_sem
);
1219 npages
= get_user_page_nowait(current
, current
->mm
,
1220 addr
, write_fault
, page
);
1221 up_read(¤t
->mm
->mmap_sem
);
1224 * By now we have tried gup_fast, and possibly async_pf, and we
1225 * are certainly not atomic. Time to retry the gup, allowing
1226 * mmap semaphore to be relinquished in the case of IO.
1228 npages
= kvm_get_user_page_io(current
, current
->mm
, addr
,
1234 /* map read fault as writable if possible */
1235 if (unlikely(!write_fault
) && writable
) {
1236 struct page
*wpage
[1];
1238 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1247 *pfn
= page_to_pfn(page
[0]);
1251 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1253 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1256 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1263 * Pin guest page in memory and return its pfn.
1264 * @addr: host virtual address which maps memory to the guest
1265 * @atomic: whether this function can sleep
1266 * @async: whether this function need to wait IO complete if the
1267 * host page is not in the memory
1268 * @write_fault: whether we should get a writable host page
1269 * @writable: whether it allows to map a writable host page for !@write_fault
1271 * The function will map a writable host page for these two cases:
1272 * 1): @write_fault = true
1273 * 2): @write_fault = false && @writable, @writable will tell the caller
1274 * whether the mapping is writable.
1276 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1277 bool write_fault
, bool *writable
)
1279 struct vm_area_struct
*vma
;
1283 /* we can do it either atomically or asynchronously, not both */
1284 BUG_ON(atomic
&& async
);
1286 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1290 return KVM_PFN_ERR_FAULT
;
1292 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1296 down_read(¤t
->mm
->mmap_sem
);
1297 if (npages
== -EHWPOISON
||
1298 (!async
&& check_user_page_hwpoison(addr
))) {
1299 pfn
= KVM_PFN_ERR_HWPOISON
;
1303 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1306 pfn
= KVM_PFN_ERR_FAULT
;
1307 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1308 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1310 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1312 if (async
&& vma_is_valid(vma
, write_fault
))
1314 pfn
= KVM_PFN_ERR_FAULT
;
1317 up_read(¤t
->mm
->mmap_sem
);
1322 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1323 bool *async
, bool write_fault
, bool *writable
)
1325 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1327 if (addr
== KVM_HVA_ERR_RO_BAD
)
1328 return KVM_PFN_ERR_RO_FAULT
;
1330 if (kvm_is_error_hva(addr
))
1331 return KVM_PFN_NOSLOT
;
1333 /* Do not map writable pfn in the readonly memslot. */
1334 if (writable
&& memslot_is_readonly(slot
)) {
1339 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1343 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1344 bool write_fault
, bool *writable
)
1346 struct kvm_memory_slot
*slot
;
1351 slot
= gfn_to_memslot(kvm
, gfn
);
1353 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1357 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1359 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1361 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1363 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1364 bool write_fault
, bool *writable
)
1366 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1368 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1370 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1372 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1374 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1376 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1379 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1381 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1383 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1385 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1388 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1390 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1392 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1394 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1400 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1401 if (kvm_is_error_hva(addr
))
1404 if (entry
< nr_pages
)
1407 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1409 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1411 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1413 if (is_error_noslot_pfn(pfn
))
1414 return KVM_ERR_PTR_BAD_PAGE
;
1416 if (kvm_is_mmio_pfn(pfn
)) {
1418 return KVM_ERR_PTR_BAD_PAGE
;
1421 return pfn_to_page(pfn
);
1424 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1428 pfn
= gfn_to_pfn(kvm
, gfn
);
1430 return kvm_pfn_to_page(pfn
);
1433 EXPORT_SYMBOL_GPL(gfn_to_page
);
1435 void kvm_release_page_clean(struct page
*page
)
1437 WARN_ON(is_error_page(page
));
1439 kvm_release_pfn_clean(page_to_pfn(page
));
1441 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1443 void kvm_release_pfn_clean(pfn_t pfn
)
1445 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1446 put_page(pfn_to_page(pfn
));
1448 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1450 void kvm_release_page_dirty(struct page
*page
)
1452 WARN_ON(is_error_page(page
));
1454 kvm_release_pfn_dirty(page_to_pfn(page
));
1456 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1458 static void kvm_release_pfn_dirty(pfn_t pfn
)
1460 kvm_set_pfn_dirty(pfn
);
1461 kvm_release_pfn_clean(pfn
);
1464 void kvm_set_pfn_dirty(pfn_t pfn
)
1466 if (!kvm_is_mmio_pfn(pfn
)) {
1467 struct page
*page
= pfn_to_page(pfn
);
1468 if (!PageReserved(page
))
1472 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1474 void kvm_set_pfn_accessed(pfn_t pfn
)
1476 if (!kvm_is_mmio_pfn(pfn
))
1477 mark_page_accessed(pfn_to_page(pfn
));
1479 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1481 void kvm_get_pfn(pfn_t pfn
)
1483 if (!kvm_is_mmio_pfn(pfn
))
1484 get_page(pfn_to_page(pfn
));
1486 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1488 static int next_segment(unsigned long len
, int offset
)
1490 if (len
> PAGE_SIZE
- offset
)
1491 return PAGE_SIZE
- offset
;
1496 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1502 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1503 if (kvm_is_error_hva(addr
))
1505 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1510 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1512 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1514 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1516 int offset
= offset_in_page(gpa
);
1519 while ((seg
= next_segment(len
, offset
)) != 0) {
1520 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1530 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1532 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1537 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1538 int offset
= offset_in_page(gpa
);
1540 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1541 if (kvm_is_error_hva(addr
))
1543 pagefault_disable();
1544 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1550 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1552 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1553 int offset
, int len
)
1558 addr
= gfn_to_hva(kvm
, gfn
);
1559 if (kvm_is_error_hva(addr
))
1561 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1564 mark_page_dirty(kvm
, gfn
);
1567 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1569 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1572 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1574 int offset
= offset_in_page(gpa
);
1577 while ((seg
= next_segment(len
, offset
)) != 0) {
1578 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1589 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1590 gpa_t gpa
, unsigned long len
)
1592 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1593 int offset
= offset_in_page(gpa
);
1594 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1595 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1596 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1597 gfn_t nr_pages_avail
;
1600 ghc
->generation
= slots
->generation
;
1602 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1603 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, &nr_pages_avail
);
1604 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_avail
>= nr_pages_needed
) {
1608 * If the requested region crosses two memslots, we still
1609 * verify that the entire region is valid here.
1611 while (start_gfn
<= end_gfn
) {
1612 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1613 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1615 if (kvm_is_error_hva(ghc
->hva
))
1617 start_gfn
+= nr_pages_avail
;
1619 /* Use the slow path for cross page reads and writes. */
1620 ghc
->memslot
= NULL
;
1624 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1626 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1627 void *data
, unsigned long len
)
1629 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1632 BUG_ON(len
> ghc
->len
);
1634 if (slots
->generation
!= ghc
->generation
)
1635 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1637 if (unlikely(!ghc
->memslot
))
1638 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1640 if (kvm_is_error_hva(ghc
->hva
))
1643 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1646 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1650 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1652 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1653 void *data
, unsigned long len
)
1655 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1658 BUG_ON(len
> ghc
->len
);
1660 if (slots
->generation
!= ghc
->generation
)
1661 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1663 if (unlikely(!ghc
->memslot
))
1664 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1666 if (kvm_is_error_hva(ghc
->hva
))
1669 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1675 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1677 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1679 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1681 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1683 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1685 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1687 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1689 int offset
= offset_in_page(gpa
);
1692 while ((seg
= next_segment(len
, offset
)) != 0) {
1693 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1702 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1704 static void mark_page_dirty_in_slot(struct kvm
*kvm
,
1705 struct kvm_memory_slot
*memslot
,
1708 if (memslot
&& memslot
->dirty_bitmap
) {
1709 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1711 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1715 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1717 struct kvm_memory_slot
*memslot
;
1719 memslot
= gfn_to_memslot(kvm
, gfn
);
1720 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1722 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1725 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1727 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1732 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1734 if (kvm_arch_vcpu_runnable(vcpu
)) {
1735 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1738 if (kvm_cpu_has_pending_timer(vcpu
))
1740 if (signal_pending(current
))
1746 finish_wait(&vcpu
->wq
, &wait
);
1748 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
1752 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1754 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1757 int cpu
= vcpu
->cpu
;
1758 wait_queue_head_t
*wqp
;
1760 wqp
= kvm_arch_vcpu_wq(vcpu
);
1761 if (waitqueue_active(wqp
)) {
1762 wake_up_interruptible(wqp
);
1763 ++vcpu
->stat
.halt_wakeup
;
1767 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1768 if (kvm_arch_vcpu_should_kick(vcpu
))
1769 smp_send_reschedule(cpu
);
1772 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1773 #endif /* !CONFIG_S390 */
1775 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1778 struct task_struct
*task
= NULL
;
1782 pid
= rcu_dereference(target
->pid
);
1784 task
= get_pid_task(pid
, PIDTYPE_PID
);
1788 if (task
->flags
& PF_VCPU
) {
1789 put_task_struct(task
);
1792 ret
= yield_to(task
, 1);
1793 put_task_struct(task
);
1797 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1800 * Helper that checks whether a VCPU is eligible for directed yield.
1801 * Most eligible candidate to yield is decided by following heuristics:
1803 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1804 * (preempted lock holder), indicated by @in_spin_loop.
1805 * Set at the beiginning and cleared at the end of interception/PLE handler.
1807 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1808 * chance last time (mostly it has become eligible now since we have probably
1809 * yielded to lockholder in last iteration. This is done by toggling
1810 * @dy_eligible each time a VCPU checked for eligibility.)
1812 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1813 * to preempted lock-holder could result in wrong VCPU selection and CPU
1814 * burning. Giving priority for a potential lock-holder increases lock
1817 * Since algorithm is based on heuristics, accessing another VCPU data without
1818 * locking does not harm. It may result in trying to yield to same VCPU, fail
1819 * and continue with next VCPU and so on.
1821 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1823 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1826 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1827 vcpu
->spin_loop
.dy_eligible
;
1829 if (vcpu
->spin_loop
.in_spin_loop
)
1830 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1838 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1840 struct kvm
*kvm
= me
->kvm
;
1841 struct kvm_vcpu
*vcpu
;
1842 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1848 kvm_vcpu_set_in_spin_loop(me
, true);
1850 * We boost the priority of a VCPU that is runnable but not
1851 * currently running, because it got preempted by something
1852 * else and called schedule in __vcpu_run. Hopefully that
1853 * VCPU is holding the lock that we need and will release it.
1854 * We approximate round-robin by starting at the last boosted VCPU.
1856 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1857 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1858 if (!pass
&& i
<= last_boosted_vcpu
) {
1859 i
= last_boosted_vcpu
;
1861 } else if (pass
&& i
> last_boosted_vcpu
)
1863 if (!ACCESS_ONCE(vcpu
->preempted
))
1867 if (waitqueue_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
1869 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1872 yielded
= kvm_vcpu_yield_to(vcpu
);
1874 kvm
->last_boosted_vcpu
= i
;
1876 } else if (yielded
< 0) {
1883 kvm_vcpu_set_in_spin_loop(me
, false);
1885 /* Ensure vcpu is not eligible during next spinloop */
1886 kvm_vcpu_set_dy_eligible(me
, false);
1888 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1890 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1892 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1895 if (vmf
->pgoff
== 0)
1896 page
= virt_to_page(vcpu
->run
);
1898 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1899 page
= virt_to_page(vcpu
->arch
.pio_data
);
1901 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1902 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1903 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1906 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1912 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1913 .fault
= kvm_vcpu_fault
,
1916 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1918 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1922 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1924 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1926 kvm_put_kvm(vcpu
->kvm
);
1930 static struct file_operations kvm_vcpu_fops
= {
1931 .release
= kvm_vcpu_release
,
1932 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1933 #ifdef CONFIG_COMPAT
1934 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1936 .mmap
= kvm_vcpu_mmap
,
1937 .llseek
= noop_llseek
,
1941 * Allocates an inode for the vcpu.
1943 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1945 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
1949 * Creates some virtual cpus. Good luck creating more than one.
1951 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1954 struct kvm_vcpu
*vcpu
, *v
;
1956 if (id
>= KVM_MAX_VCPUS
)
1959 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1961 return PTR_ERR(vcpu
);
1963 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1965 r
= kvm_arch_vcpu_setup(vcpu
);
1969 mutex_lock(&kvm
->lock
);
1970 if (!kvm_vcpu_compatible(vcpu
)) {
1972 goto unlock_vcpu_destroy
;
1974 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1976 goto unlock_vcpu_destroy
;
1979 kvm_for_each_vcpu(r
, v
, kvm
)
1980 if (v
->vcpu_id
== id
) {
1982 goto unlock_vcpu_destroy
;
1985 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1987 /* Now it's all set up, let userspace reach it */
1989 r
= create_vcpu_fd(vcpu
);
1992 goto unlock_vcpu_destroy
;
1995 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1997 atomic_inc(&kvm
->online_vcpus
);
1999 mutex_unlock(&kvm
->lock
);
2000 kvm_arch_vcpu_postcreate(vcpu
);
2003 unlock_vcpu_destroy
:
2004 mutex_unlock(&kvm
->lock
);
2006 kvm_arch_vcpu_destroy(vcpu
);
2010 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2013 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2014 vcpu
->sigset_active
= 1;
2015 vcpu
->sigset
= *sigset
;
2017 vcpu
->sigset_active
= 0;
2021 static long kvm_vcpu_ioctl(struct file
*filp
,
2022 unsigned int ioctl
, unsigned long arg
)
2024 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2025 void __user
*argp
= (void __user
*)arg
;
2027 struct kvm_fpu
*fpu
= NULL
;
2028 struct kvm_sregs
*kvm_sregs
= NULL
;
2030 if (vcpu
->kvm
->mm
!= current
->mm
)
2033 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2036 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2038 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2039 * so vcpu_load() would break it.
2041 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
2042 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2046 r
= vcpu_load(vcpu
);
2054 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2055 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2057 case KVM_GET_REGS
: {
2058 struct kvm_regs
*kvm_regs
;
2061 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2064 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2068 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2075 case KVM_SET_REGS
: {
2076 struct kvm_regs
*kvm_regs
;
2079 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2080 if (IS_ERR(kvm_regs
)) {
2081 r
= PTR_ERR(kvm_regs
);
2084 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2088 case KVM_GET_SREGS
: {
2089 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2093 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2097 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2102 case KVM_SET_SREGS
: {
2103 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2104 if (IS_ERR(kvm_sregs
)) {
2105 r
= PTR_ERR(kvm_sregs
);
2109 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2112 case KVM_GET_MP_STATE
: {
2113 struct kvm_mp_state mp_state
;
2115 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2119 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2124 case KVM_SET_MP_STATE
: {
2125 struct kvm_mp_state mp_state
;
2128 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2130 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2133 case KVM_TRANSLATE
: {
2134 struct kvm_translation tr
;
2137 if (copy_from_user(&tr
, argp
, sizeof tr
))
2139 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2143 if (copy_to_user(argp
, &tr
, sizeof tr
))
2148 case KVM_SET_GUEST_DEBUG
: {
2149 struct kvm_guest_debug dbg
;
2152 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2154 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2157 case KVM_SET_SIGNAL_MASK
: {
2158 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2159 struct kvm_signal_mask kvm_sigmask
;
2160 sigset_t sigset
, *p
;
2165 if (copy_from_user(&kvm_sigmask
, argp
,
2166 sizeof kvm_sigmask
))
2169 if (kvm_sigmask
.len
!= sizeof sigset
)
2172 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2177 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2181 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2185 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2189 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2195 fpu
= memdup_user(argp
, sizeof(*fpu
));
2201 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2205 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2214 #ifdef CONFIG_COMPAT
2215 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2216 unsigned int ioctl
, unsigned long arg
)
2218 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2219 void __user
*argp
= compat_ptr(arg
);
2222 if (vcpu
->kvm
->mm
!= current
->mm
)
2226 case KVM_SET_SIGNAL_MASK
: {
2227 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2228 struct kvm_signal_mask kvm_sigmask
;
2229 compat_sigset_t csigset
;
2234 if (copy_from_user(&kvm_sigmask
, argp
,
2235 sizeof kvm_sigmask
))
2238 if (kvm_sigmask
.len
!= sizeof csigset
)
2241 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2244 sigset_from_compat(&sigset
, &csigset
);
2245 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2247 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2251 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2259 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2260 int (*accessor
)(struct kvm_device
*dev
,
2261 struct kvm_device_attr
*attr
),
2264 struct kvm_device_attr attr
;
2269 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2272 return accessor(dev
, &attr
);
2275 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2278 struct kvm_device
*dev
= filp
->private_data
;
2281 case KVM_SET_DEVICE_ATTR
:
2282 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2283 case KVM_GET_DEVICE_ATTR
:
2284 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2285 case KVM_HAS_DEVICE_ATTR
:
2286 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2288 if (dev
->ops
->ioctl
)
2289 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2295 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2297 struct kvm_device
*dev
= filp
->private_data
;
2298 struct kvm
*kvm
= dev
->kvm
;
2304 static const struct file_operations kvm_device_fops
= {
2305 .unlocked_ioctl
= kvm_device_ioctl
,
2306 #ifdef CONFIG_COMPAT
2307 .compat_ioctl
= kvm_device_ioctl
,
2309 .release
= kvm_device_release
,
2312 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2314 if (filp
->f_op
!= &kvm_device_fops
)
2317 return filp
->private_data
;
2320 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2321 #ifdef CONFIG_KVM_MPIC
2322 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2323 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2326 #ifdef CONFIG_KVM_XICS
2327 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2331 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2333 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2336 if (kvm_device_ops_table
[type
] != NULL
)
2339 kvm_device_ops_table
[type
] = ops
;
2343 void kvm_unregister_device_ops(u32 type
)
2345 if (kvm_device_ops_table
[type
] != NULL
)
2346 kvm_device_ops_table
[type
] = NULL
;
2349 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2350 struct kvm_create_device
*cd
)
2352 struct kvm_device_ops
*ops
= NULL
;
2353 struct kvm_device
*dev
;
2354 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2357 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2360 ops
= kvm_device_ops_table
[cd
->type
];
2367 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2374 ret
= ops
->create(dev
, cd
->type
);
2380 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2386 list_add(&dev
->vm_node
, &kvm
->devices
);
2392 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2395 case KVM_CAP_USER_MEMORY
:
2396 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2397 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2398 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2399 case KVM_CAP_SET_BOOT_CPU_ID
:
2401 case KVM_CAP_INTERNAL_ERROR_DATA
:
2402 #ifdef CONFIG_HAVE_KVM_MSI
2403 case KVM_CAP_SIGNAL_MSI
:
2405 #ifdef CONFIG_HAVE_KVM_IRQFD
2406 case KVM_CAP_IRQFD_RESAMPLE
:
2408 case KVM_CAP_CHECK_EXTENSION_VM
:
2410 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2411 case KVM_CAP_IRQ_ROUTING
:
2412 return KVM_MAX_IRQ_ROUTES
;
2417 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2420 static long kvm_vm_ioctl(struct file
*filp
,
2421 unsigned int ioctl
, unsigned long arg
)
2423 struct kvm
*kvm
= filp
->private_data
;
2424 void __user
*argp
= (void __user
*)arg
;
2427 if (kvm
->mm
!= current
->mm
)
2430 case KVM_CREATE_VCPU
:
2431 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2433 case KVM_SET_USER_MEMORY_REGION
: {
2434 struct kvm_userspace_memory_region kvm_userspace_mem
;
2437 if (copy_from_user(&kvm_userspace_mem
, argp
,
2438 sizeof kvm_userspace_mem
))
2441 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2444 case KVM_GET_DIRTY_LOG
: {
2445 struct kvm_dirty_log log
;
2448 if (copy_from_user(&log
, argp
, sizeof log
))
2450 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2453 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2454 case KVM_REGISTER_COALESCED_MMIO
: {
2455 struct kvm_coalesced_mmio_zone zone
;
2457 if (copy_from_user(&zone
, argp
, sizeof zone
))
2459 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2462 case KVM_UNREGISTER_COALESCED_MMIO
: {
2463 struct kvm_coalesced_mmio_zone zone
;
2465 if (copy_from_user(&zone
, argp
, sizeof zone
))
2467 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2472 struct kvm_irqfd data
;
2475 if (copy_from_user(&data
, argp
, sizeof data
))
2477 r
= kvm_irqfd(kvm
, &data
);
2480 case KVM_IOEVENTFD
: {
2481 struct kvm_ioeventfd data
;
2484 if (copy_from_user(&data
, argp
, sizeof data
))
2486 r
= kvm_ioeventfd(kvm
, &data
);
2489 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2490 case KVM_SET_BOOT_CPU_ID
:
2492 mutex_lock(&kvm
->lock
);
2493 if (atomic_read(&kvm
->online_vcpus
) != 0)
2496 kvm
->bsp_vcpu_id
= arg
;
2497 mutex_unlock(&kvm
->lock
);
2500 #ifdef CONFIG_HAVE_KVM_MSI
2501 case KVM_SIGNAL_MSI
: {
2505 if (copy_from_user(&msi
, argp
, sizeof msi
))
2507 r
= kvm_send_userspace_msi(kvm
, &msi
);
2511 #ifdef __KVM_HAVE_IRQ_LINE
2512 case KVM_IRQ_LINE_STATUS
:
2513 case KVM_IRQ_LINE
: {
2514 struct kvm_irq_level irq_event
;
2517 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2520 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2521 ioctl
== KVM_IRQ_LINE_STATUS
);
2526 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2527 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2535 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2536 case KVM_SET_GSI_ROUTING
: {
2537 struct kvm_irq_routing routing
;
2538 struct kvm_irq_routing __user
*urouting
;
2539 struct kvm_irq_routing_entry
*entries
;
2542 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2545 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2550 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2555 if (copy_from_user(entries
, urouting
->entries
,
2556 routing
.nr
* sizeof(*entries
)))
2557 goto out_free_irq_routing
;
2558 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2560 out_free_irq_routing
:
2564 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2565 case KVM_CREATE_DEVICE
: {
2566 struct kvm_create_device cd
;
2569 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2572 r
= kvm_ioctl_create_device(kvm
, &cd
);
2577 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2583 case KVM_CHECK_EXTENSION
:
2584 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
2587 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2589 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2595 #ifdef CONFIG_COMPAT
2596 struct compat_kvm_dirty_log
{
2600 compat_uptr_t dirty_bitmap
; /* one bit per page */
2605 static long kvm_vm_compat_ioctl(struct file
*filp
,
2606 unsigned int ioctl
, unsigned long arg
)
2608 struct kvm
*kvm
= filp
->private_data
;
2611 if (kvm
->mm
!= current
->mm
)
2614 case KVM_GET_DIRTY_LOG
: {
2615 struct compat_kvm_dirty_log compat_log
;
2616 struct kvm_dirty_log log
;
2619 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2620 sizeof(compat_log
)))
2622 log
.slot
= compat_log
.slot
;
2623 log
.padding1
= compat_log
.padding1
;
2624 log
.padding2
= compat_log
.padding2
;
2625 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2627 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2631 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2639 static struct file_operations kvm_vm_fops
= {
2640 .release
= kvm_vm_release
,
2641 .unlocked_ioctl
= kvm_vm_ioctl
,
2642 #ifdef CONFIG_COMPAT
2643 .compat_ioctl
= kvm_vm_compat_ioctl
,
2645 .llseek
= noop_llseek
,
2648 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2653 kvm
= kvm_create_vm(type
);
2655 return PTR_ERR(kvm
);
2656 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2657 r
= kvm_coalesced_mmio_init(kvm
);
2663 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2670 static long kvm_dev_ioctl(struct file
*filp
,
2671 unsigned int ioctl
, unsigned long arg
)
2676 case KVM_GET_API_VERSION
:
2679 r
= KVM_API_VERSION
;
2682 r
= kvm_dev_ioctl_create_vm(arg
);
2684 case KVM_CHECK_EXTENSION
:
2685 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
2687 case KVM_GET_VCPU_MMAP_SIZE
:
2690 r
= PAGE_SIZE
; /* struct kvm_run */
2692 r
+= PAGE_SIZE
; /* pio data page */
2694 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2695 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2698 case KVM_TRACE_ENABLE
:
2699 case KVM_TRACE_PAUSE
:
2700 case KVM_TRACE_DISABLE
:
2704 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2710 static struct file_operations kvm_chardev_ops
= {
2711 .unlocked_ioctl
= kvm_dev_ioctl
,
2712 .compat_ioctl
= kvm_dev_ioctl
,
2713 .llseek
= noop_llseek
,
2716 static struct miscdevice kvm_dev
= {
2722 static void hardware_enable_nolock(void *junk
)
2724 int cpu
= raw_smp_processor_id();
2727 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2730 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2732 r
= kvm_arch_hardware_enable();
2735 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2736 atomic_inc(&hardware_enable_failed
);
2737 printk(KERN_INFO
"kvm: enabling virtualization on "
2738 "CPU%d failed\n", cpu
);
2742 static void hardware_enable(void)
2744 raw_spin_lock(&kvm_count_lock
);
2745 if (kvm_usage_count
)
2746 hardware_enable_nolock(NULL
);
2747 raw_spin_unlock(&kvm_count_lock
);
2750 static void hardware_disable_nolock(void *junk
)
2752 int cpu
= raw_smp_processor_id();
2754 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2756 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2757 kvm_arch_hardware_disable();
2760 static void hardware_disable(void)
2762 raw_spin_lock(&kvm_count_lock
);
2763 if (kvm_usage_count
)
2764 hardware_disable_nolock(NULL
);
2765 raw_spin_unlock(&kvm_count_lock
);
2768 static void hardware_disable_all_nolock(void)
2770 BUG_ON(!kvm_usage_count
);
2773 if (!kvm_usage_count
)
2774 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2777 static void hardware_disable_all(void)
2779 raw_spin_lock(&kvm_count_lock
);
2780 hardware_disable_all_nolock();
2781 raw_spin_unlock(&kvm_count_lock
);
2784 static int hardware_enable_all(void)
2788 raw_spin_lock(&kvm_count_lock
);
2791 if (kvm_usage_count
== 1) {
2792 atomic_set(&hardware_enable_failed
, 0);
2793 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2795 if (atomic_read(&hardware_enable_failed
)) {
2796 hardware_disable_all_nolock();
2801 raw_spin_unlock(&kvm_count_lock
);
2806 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2811 val
&= ~CPU_TASKS_FROZEN
;
2814 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2819 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2827 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2831 * Some (well, at least mine) BIOSes hang on reboot if
2834 * And Intel TXT required VMX off for all cpu when system shutdown.
2836 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2837 kvm_rebooting
= true;
2838 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2842 static struct notifier_block kvm_reboot_notifier
= {
2843 .notifier_call
= kvm_reboot
,
2847 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2851 for (i
= 0; i
< bus
->dev_count
; i
++) {
2852 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2854 kvm_iodevice_destructor(pos
);
2859 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
2860 const struct kvm_io_range
*r2
)
2862 if (r1
->addr
< r2
->addr
)
2864 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2869 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2871 return kvm_io_bus_cmp(p1
, p2
);
2874 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2875 gpa_t addr
, int len
)
2877 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2883 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2884 kvm_io_bus_sort_cmp
, NULL
);
2889 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2890 gpa_t addr
, int len
)
2892 struct kvm_io_range
*range
, key
;
2895 key
= (struct kvm_io_range
) {
2900 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2901 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2905 off
= range
- bus
->range
;
2907 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
2913 static int __kvm_io_bus_write(struct kvm_io_bus
*bus
,
2914 struct kvm_io_range
*range
, const void *val
)
2918 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2922 while (idx
< bus
->dev_count
&&
2923 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2924 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, range
->addr
,
2933 /* kvm_io_bus_write - called under kvm->slots_lock */
2934 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2935 int len
, const void *val
)
2937 struct kvm_io_bus
*bus
;
2938 struct kvm_io_range range
;
2941 range
= (struct kvm_io_range
) {
2946 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2947 r
= __kvm_io_bus_write(bus
, &range
, val
);
2948 return r
< 0 ? r
: 0;
2951 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2952 int kvm_io_bus_write_cookie(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2953 int len
, const void *val
, long cookie
)
2955 struct kvm_io_bus
*bus
;
2956 struct kvm_io_range range
;
2958 range
= (struct kvm_io_range
) {
2963 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2965 /* First try the device referenced by cookie. */
2966 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
2967 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
2968 if (!kvm_iodevice_write(bus
->range
[cookie
].dev
, addr
, len
,
2973 * cookie contained garbage; fall back to search and return the
2974 * correct cookie value.
2976 return __kvm_io_bus_write(bus
, &range
, val
);
2979 static int __kvm_io_bus_read(struct kvm_io_bus
*bus
, struct kvm_io_range
*range
,
2984 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2988 while (idx
< bus
->dev_count
&&
2989 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2990 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, range
->addr
,
2998 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3000 /* kvm_io_bus_read - called under kvm->slots_lock */
3001 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3004 struct kvm_io_bus
*bus
;
3005 struct kvm_io_range range
;
3008 range
= (struct kvm_io_range
) {
3013 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3014 r
= __kvm_io_bus_read(bus
, &range
, val
);
3015 return r
< 0 ? r
: 0;
3019 /* Caller must hold slots_lock. */
3020 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3021 int len
, struct kvm_io_device
*dev
)
3023 struct kvm_io_bus
*new_bus
, *bus
;
3025 bus
= kvm
->buses
[bus_idx
];
3026 /* exclude ioeventfd which is limited by maximum fd */
3027 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3030 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3031 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3034 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3035 sizeof(struct kvm_io_range
)));
3036 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3037 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3038 synchronize_srcu_expedited(&kvm
->srcu
);
3044 /* Caller must hold slots_lock. */
3045 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3046 struct kvm_io_device
*dev
)
3049 struct kvm_io_bus
*new_bus
, *bus
;
3051 bus
= kvm
->buses
[bus_idx
];
3053 for (i
= 0; i
< bus
->dev_count
; i
++)
3054 if (bus
->range
[i
].dev
== dev
) {
3062 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3063 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3067 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3068 new_bus
->dev_count
--;
3069 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3070 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3072 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3073 synchronize_srcu_expedited(&kvm
->srcu
);
3078 static struct notifier_block kvm_cpu_notifier
= {
3079 .notifier_call
= kvm_cpu_hotplug
,
3082 static int vm_stat_get(void *_offset
, u64
*val
)
3084 unsigned offset
= (long)_offset
;
3088 spin_lock(&kvm_lock
);
3089 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3090 *val
+= *(u32
*)((void *)kvm
+ offset
);
3091 spin_unlock(&kvm_lock
);
3095 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3097 static int vcpu_stat_get(void *_offset
, u64
*val
)
3099 unsigned offset
= (long)_offset
;
3101 struct kvm_vcpu
*vcpu
;
3105 spin_lock(&kvm_lock
);
3106 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3107 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3108 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3110 spin_unlock(&kvm_lock
);
3114 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3116 static const struct file_operations
*stat_fops
[] = {
3117 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3118 [KVM_STAT_VM
] = &vm_stat_fops
,
3121 static int kvm_init_debug(void)
3124 struct kvm_stats_debugfs_item
*p
;
3126 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3127 if (kvm_debugfs_dir
== NULL
)
3130 for (p
= debugfs_entries
; p
->name
; ++p
) {
3131 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3132 (void *)(long)p
->offset
,
3133 stat_fops
[p
->kind
]);
3134 if (p
->dentry
== NULL
)
3141 debugfs_remove_recursive(kvm_debugfs_dir
);
3146 static void kvm_exit_debug(void)
3148 struct kvm_stats_debugfs_item
*p
;
3150 for (p
= debugfs_entries
; p
->name
; ++p
)
3151 debugfs_remove(p
->dentry
);
3152 debugfs_remove(kvm_debugfs_dir
);
3155 static int kvm_suspend(void)
3157 if (kvm_usage_count
)
3158 hardware_disable_nolock(NULL
);
3162 static void kvm_resume(void)
3164 if (kvm_usage_count
) {
3165 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3166 hardware_enable_nolock(NULL
);
3170 static struct syscore_ops kvm_syscore_ops
= {
3171 .suspend
= kvm_suspend
,
3172 .resume
= kvm_resume
,
3176 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3178 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3181 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3183 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3184 if (vcpu
->preempted
)
3185 vcpu
->preempted
= false;
3187 kvm_arch_sched_in(vcpu
, cpu
);
3189 kvm_arch_vcpu_load(vcpu
, cpu
);
3192 static void kvm_sched_out(struct preempt_notifier
*pn
,
3193 struct task_struct
*next
)
3195 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3197 if (current
->state
== TASK_RUNNING
)
3198 vcpu
->preempted
= true;
3199 kvm_arch_vcpu_put(vcpu
);
3202 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3203 struct module
*module
)
3208 r
= kvm_arch_init(opaque
);
3213 * kvm_arch_init makes sure there's at most one caller
3214 * for architectures that support multiple implementations,
3215 * like intel and amd on x86.
3216 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3217 * conflicts in case kvm is already setup for another implementation.
3219 r
= kvm_irqfd_init();
3223 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3228 r
= kvm_arch_hardware_setup();
3232 for_each_online_cpu(cpu
) {
3233 smp_call_function_single(cpu
,
3234 kvm_arch_check_processor_compat
,
3240 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3243 register_reboot_notifier(&kvm_reboot_notifier
);
3245 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3247 vcpu_align
= __alignof__(struct kvm_vcpu
);
3248 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3250 if (!kvm_vcpu_cache
) {
3255 r
= kvm_async_pf_init();
3259 kvm_chardev_ops
.owner
= module
;
3260 kvm_vm_fops
.owner
= module
;
3261 kvm_vcpu_fops
.owner
= module
;
3263 r
= misc_register(&kvm_dev
);
3265 printk(KERN_ERR
"kvm: misc device register failed\n");
3269 register_syscore_ops(&kvm_syscore_ops
);
3271 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3272 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3274 r
= kvm_init_debug();
3276 printk(KERN_ERR
"kvm: create debugfs files failed\n");
3280 r
= kvm_vfio_ops_init();
3286 unregister_syscore_ops(&kvm_syscore_ops
);
3287 misc_deregister(&kvm_dev
);
3289 kvm_async_pf_deinit();
3291 kmem_cache_destroy(kvm_vcpu_cache
);
3293 unregister_reboot_notifier(&kvm_reboot_notifier
);
3294 unregister_cpu_notifier(&kvm_cpu_notifier
);
3297 kvm_arch_hardware_unsetup();
3299 free_cpumask_var(cpus_hardware_enabled
);
3307 EXPORT_SYMBOL_GPL(kvm_init
);
3312 misc_deregister(&kvm_dev
);
3313 kmem_cache_destroy(kvm_vcpu_cache
);
3314 kvm_async_pf_deinit();
3315 unregister_syscore_ops(&kvm_syscore_ops
);
3316 unregister_reboot_notifier(&kvm_reboot_notifier
);
3317 unregister_cpu_notifier(&kvm_cpu_notifier
);
3318 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3319 kvm_arch_hardware_unsetup();
3322 free_cpumask_var(cpus_hardware_enabled
);
3323 kvm_vfio_ops_exit();
3325 EXPORT_SYMBOL_GPL(kvm_exit
);