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/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_RAW_SPINLOCK(kvm_lock
);
76 static cpumask_var_t cpus_hardware_enabled
;
77 static int kvm_usage_count
= 0;
78 static atomic_t hardware_enable_failed
;
80 struct kmem_cache
*kvm_vcpu_cache
;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
83 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
85 struct dentry
*kvm_debugfs_dir
;
87 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
90 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
99 EXPORT_SYMBOL_GPL(kvm_rebooting
);
101 static bool largepages_enabled
= true;
103 bool kvm_is_mmio_pfn(pfn_t pfn
)
105 if (pfn_valid(pfn
)) {
107 struct page
*tail
= pfn_to_page(pfn
);
108 struct page
*head
= compound_trans_head(tail
);
109 reserved
= PageReserved(head
);
112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
125 return PageReserved(tail
);
132 * Switches to specified vcpu, until a matching vcpu_put()
134 int vcpu_load(struct kvm_vcpu
*vcpu
)
138 if (mutex_lock_killable(&vcpu
->mutex
))
140 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
141 /* The thread running this VCPU changed. */
142 struct pid
*oldpid
= vcpu
->pid
;
143 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
144 rcu_assign_pointer(vcpu
->pid
, newpid
);
149 preempt_notifier_register(&vcpu
->preempt_notifier
);
150 kvm_arch_vcpu_load(vcpu
, cpu
);
155 void vcpu_put(struct kvm_vcpu
*vcpu
)
158 kvm_arch_vcpu_put(vcpu
);
159 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
161 mutex_unlock(&vcpu
->mutex
);
164 static void ack_flush(void *_completed
)
168 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
173 struct kvm_vcpu
*vcpu
;
175 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
178 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
179 kvm_make_request(req
, vcpu
);
182 /* Set ->requests bit before we read ->mode */
185 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
186 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
187 cpumask_set_cpu(cpu
, cpus
);
189 if (unlikely(cpus
== NULL
))
190 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
191 else if (!cpumask_empty(cpus
))
192 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
196 free_cpumask_var(cpus
);
200 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
202 long dirty_count
= kvm
->tlbs_dirty
;
205 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
206 ++kvm
->stat
.remote_tlb_flush
;
207 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
210 void kvm_reload_remote_mmus(struct kvm
*kvm
)
212 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
215 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
217 make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
220 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
222 make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
225 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
230 mutex_init(&vcpu
->mutex
);
235 init_waitqueue_head(&vcpu
->wq
);
236 kvm_async_pf_vcpu_init(vcpu
);
238 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
243 vcpu
->run
= page_address(page
);
245 kvm_vcpu_set_in_spin_loop(vcpu
, false);
246 kvm_vcpu_set_dy_eligible(vcpu
, false);
247 vcpu
->preempted
= false;
249 r
= kvm_arch_vcpu_init(vcpu
);
255 free_page((unsigned long)vcpu
->run
);
259 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
261 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
264 kvm_arch_vcpu_uninit(vcpu
);
265 free_page((unsigned long)vcpu
->run
);
267 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
269 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
270 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
272 return container_of(mn
, struct kvm
, mmu_notifier
);
275 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
276 struct mm_struct
*mm
,
277 unsigned long address
)
279 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
280 int need_tlb_flush
, idx
;
283 * When ->invalidate_page runs, the linux pte has been zapped
284 * already but the page is still allocated until
285 * ->invalidate_page returns. So if we increase the sequence
286 * here the kvm page fault will notice if the spte can't be
287 * established because the page is going to be freed. If
288 * instead the kvm page fault establishes the spte before
289 * ->invalidate_page runs, kvm_unmap_hva will release it
292 * The sequence increase only need to be seen at spin_unlock
293 * time, and not at spin_lock time.
295 * Increasing the sequence after the spin_unlock would be
296 * unsafe because the kvm page fault could then establish the
297 * pte after kvm_unmap_hva returned, without noticing the page
298 * is going to be freed.
300 idx
= srcu_read_lock(&kvm
->srcu
);
301 spin_lock(&kvm
->mmu_lock
);
303 kvm
->mmu_notifier_seq
++;
304 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
305 /* we've to flush the tlb before the pages can be freed */
307 kvm_flush_remote_tlbs(kvm
);
309 spin_unlock(&kvm
->mmu_lock
);
310 srcu_read_unlock(&kvm
->srcu
, idx
);
313 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
314 struct mm_struct
*mm
,
315 unsigned long address
,
318 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
321 idx
= srcu_read_lock(&kvm
->srcu
);
322 spin_lock(&kvm
->mmu_lock
);
323 kvm
->mmu_notifier_seq
++;
324 kvm_set_spte_hva(kvm
, address
, pte
);
325 spin_unlock(&kvm
->mmu_lock
);
326 srcu_read_unlock(&kvm
->srcu
, idx
);
329 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
330 struct mm_struct
*mm
,
334 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
335 int need_tlb_flush
= 0, idx
;
337 idx
= srcu_read_lock(&kvm
->srcu
);
338 spin_lock(&kvm
->mmu_lock
);
340 * The count increase must become visible at unlock time as no
341 * spte can be established without taking the mmu_lock and
342 * count is also read inside the mmu_lock critical section.
344 kvm
->mmu_notifier_count
++;
345 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
346 need_tlb_flush
|= kvm
->tlbs_dirty
;
347 /* we've to flush the tlb before the pages can be freed */
349 kvm_flush_remote_tlbs(kvm
);
351 spin_unlock(&kvm
->mmu_lock
);
352 srcu_read_unlock(&kvm
->srcu
, idx
);
355 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
356 struct mm_struct
*mm
,
360 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
362 spin_lock(&kvm
->mmu_lock
);
364 * This sequence increase will notify the kvm page fault that
365 * the page that is going to be mapped in the spte could have
368 kvm
->mmu_notifier_seq
++;
371 * The above sequence increase must be visible before the
372 * below count decrease, which is ensured by the smp_wmb above
373 * in conjunction with the smp_rmb in mmu_notifier_retry().
375 kvm
->mmu_notifier_count
--;
376 spin_unlock(&kvm
->mmu_lock
);
378 BUG_ON(kvm
->mmu_notifier_count
< 0);
381 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
382 struct mm_struct
*mm
,
383 unsigned long address
)
385 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
388 idx
= srcu_read_lock(&kvm
->srcu
);
389 spin_lock(&kvm
->mmu_lock
);
391 young
= kvm_age_hva(kvm
, address
);
393 kvm_flush_remote_tlbs(kvm
);
395 spin_unlock(&kvm
->mmu_lock
);
396 srcu_read_unlock(&kvm
->srcu
, idx
);
401 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
402 struct mm_struct
*mm
,
403 unsigned long address
)
405 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
408 idx
= srcu_read_lock(&kvm
->srcu
);
409 spin_lock(&kvm
->mmu_lock
);
410 young
= kvm_test_age_hva(kvm
, address
);
411 spin_unlock(&kvm
->mmu_lock
);
412 srcu_read_unlock(&kvm
->srcu
, idx
);
417 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
418 struct mm_struct
*mm
)
420 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
423 idx
= srcu_read_lock(&kvm
->srcu
);
424 kvm_arch_flush_shadow_all(kvm
);
425 srcu_read_unlock(&kvm
->srcu
, idx
);
428 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
429 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
430 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
431 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
432 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
433 .test_young
= kvm_mmu_notifier_test_young
,
434 .change_pte
= kvm_mmu_notifier_change_pte
,
435 .release
= kvm_mmu_notifier_release
,
438 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
440 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
441 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
444 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
446 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
451 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
453 static void kvm_init_memslots_id(struct kvm
*kvm
)
456 struct kvm_memslots
*slots
= kvm
->memslots
;
458 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
459 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
462 static struct kvm
*kvm_create_vm(unsigned long type
)
465 struct kvm
*kvm
= kvm_arch_alloc_vm();
468 return ERR_PTR(-ENOMEM
);
470 r
= kvm_arch_init_vm(kvm
, type
);
472 goto out_err_nodisable
;
474 r
= hardware_enable_all();
476 goto out_err_nodisable
;
478 #ifdef CONFIG_HAVE_KVM_IRQCHIP
479 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
480 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
483 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
486 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
489 kvm_init_memslots_id(kvm
);
490 if (init_srcu_struct(&kvm
->srcu
))
492 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
493 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
499 spin_lock_init(&kvm
->mmu_lock
);
500 kvm
->mm
= current
->mm
;
501 atomic_inc(&kvm
->mm
->mm_count
);
502 kvm_eventfd_init(kvm
);
503 mutex_init(&kvm
->lock
);
504 mutex_init(&kvm
->irq_lock
);
505 mutex_init(&kvm
->slots_lock
);
506 atomic_set(&kvm
->users_count
, 1);
508 r
= kvm_init_mmu_notifier(kvm
);
512 raw_spin_lock(&kvm_lock
);
513 list_add(&kvm
->vm_list
, &vm_list
);
514 raw_spin_unlock(&kvm_lock
);
519 cleanup_srcu_struct(&kvm
->srcu
);
521 hardware_disable_all();
523 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
524 kfree(kvm
->buses
[i
]);
525 kfree(kvm
->memslots
);
526 kvm_arch_free_vm(kvm
);
531 * Avoid using vmalloc for a small buffer.
532 * Should not be used when the size is statically known.
534 void *kvm_kvzalloc(unsigned long size
)
536 if (size
> PAGE_SIZE
)
537 return vzalloc(size
);
539 return kzalloc(size
, GFP_KERNEL
);
542 void kvm_kvfree(const void *addr
)
544 if (is_vmalloc_addr(addr
))
550 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
552 if (!memslot
->dirty_bitmap
)
555 kvm_kvfree(memslot
->dirty_bitmap
);
556 memslot
->dirty_bitmap
= NULL
;
560 * Free any memory in @free but not in @dont.
562 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
563 struct kvm_memory_slot
*dont
)
565 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
566 kvm_destroy_dirty_bitmap(free
);
568 kvm_arch_free_memslot(free
, dont
);
573 void kvm_free_physmem(struct kvm
*kvm
)
575 struct kvm_memslots
*slots
= kvm
->memslots
;
576 struct kvm_memory_slot
*memslot
;
578 kvm_for_each_memslot(memslot
, slots
)
579 kvm_free_physmem_slot(memslot
, NULL
);
581 kfree(kvm
->memslots
);
584 static void kvm_destroy_vm(struct kvm
*kvm
)
587 struct mm_struct
*mm
= kvm
->mm
;
589 kvm_arch_sync_events(kvm
);
590 raw_spin_lock(&kvm_lock
);
591 list_del(&kvm
->vm_list
);
592 raw_spin_unlock(&kvm_lock
);
593 kvm_free_irq_routing(kvm
);
594 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
595 kvm_io_bus_destroy(kvm
->buses
[i
]);
596 kvm_coalesced_mmio_free(kvm
);
597 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
598 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
600 kvm_arch_flush_shadow_all(kvm
);
602 kvm_arch_destroy_vm(kvm
);
603 kvm_free_physmem(kvm
);
604 cleanup_srcu_struct(&kvm
->srcu
);
605 kvm_arch_free_vm(kvm
);
606 hardware_disable_all();
610 void kvm_get_kvm(struct kvm
*kvm
)
612 atomic_inc(&kvm
->users_count
);
614 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
616 void kvm_put_kvm(struct kvm
*kvm
)
618 if (atomic_dec_and_test(&kvm
->users_count
))
621 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
624 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
626 struct kvm
*kvm
= filp
->private_data
;
628 kvm_irqfd_release(kvm
);
635 * Allocation size is twice as large as the actual dirty bitmap size.
636 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
638 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
641 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
643 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
644 if (!memslot
->dirty_bitmap
)
647 #endif /* !CONFIG_S390 */
651 static int cmp_memslot(const void *slot1
, const void *slot2
)
653 struct kvm_memory_slot
*s1
, *s2
;
655 s1
= (struct kvm_memory_slot
*)slot1
;
656 s2
= (struct kvm_memory_slot
*)slot2
;
658 if (s1
->npages
< s2
->npages
)
660 if (s1
->npages
> s2
->npages
)
667 * Sort the memslots base on its size, so the larger slots
668 * will get better fit.
670 static void sort_memslots(struct kvm_memslots
*slots
)
674 sort(slots
->memslots
, KVM_MEM_SLOTS_NUM
,
675 sizeof(struct kvm_memory_slot
), cmp_memslot
, NULL
);
677 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
678 slots
->id_to_index
[slots
->memslots
[i
].id
] = i
;
681 void update_memslots(struct kvm_memslots
*slots
, struct kvm_memory_slot
*new,
686 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
687 unsigned long npages
= old
->npages
;
690 if (new->npages
!= npages
)
691 sort_memslots(slots
);
694 slots
->generation
= last_generation
+ 1;
697 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
699 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
701 #ifdef KVM_CAP_READONLY_MEM
702 valid_flags
|= KVM_MEM_READONLY
;
705 if (mem
->flags
& ~valid_flags
)
711 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
712 struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
714 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
716 update_memslots(slots
, new, kvm
->memslots
->generation
);
717 rcu_assign_pointer(kvm
->memslots
, slots
);
718 synchronize_srcu_expedited(&kvm
->srcu
);
723 * Allocate some memory and give it an address in the guest physical address
726 * Discontiguous memory is allowed, mostly for framebuffers.
728 * Must be called holding mmap_sem for write.
730 int __kvm_set_memory_region(struct kvm
*kvm
,
731 struct kvm_userspace_memory_region
*mem
)
735 unsigned long npages
;
736 struct kvm_memory_slot
*slot
;
737 struct kvm_memory_slot old
, new;
738 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
739 enum kvm_mr_change change
;
741 r
= check_memory_region_flags(mem
);
746 /* General sanity checks */
747 if (mem
->memory_size
& (PAGE_SIZE
- 1))
749 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
751 /* We can read the guest memory with __xxx_user() later on. */
752 if ((mem
->slot
< KVM_USER_MEM_SLOTS
) &&
753 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
754 !access_ok(VERIFY_WRITE
,
755 (void __user
*)(unsigned long)mem
->userspace_addr
,
758 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
760 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
763 slot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
764 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
765 npages
= mem
->memory_size
>> PAGE_SHIFT
;
768 if (npages
> KVM_MEM_MAX_NR_PAGES
)
772 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
777 new.base_gfn
= base_gfn
;
779 new.flags
= mem
->flags
;
784 change
= KVM_MR_CREATE
;
785 else { /* Modify an existing slot. */
786 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
787 (npages
!= old
.npages
) ||
788 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
791 if (base_gfn
!= old
.base_gfn
)
792 change
= KVM_MR_MOVE
;
793 else if (new.flags
!= old
.flags
)
794 change
= KVM_MR_FLAGS_ONLY
;
795 else { /* Nothing to change. */
800 } else if (old
.npages
) {
801 change
= KVM_MR_DELETE
;
802 } else /* Modify a non-existent slot: disallowed. */
805 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
806 /* Check for overlaps */
808 kvm_for_each_memslot(slot
, kvm
->memslots
) {
809 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
810 (slot
->id
== mem
->slot
))
812 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
813 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
818 /* Free page dirty bitmap if unneeded */
819 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
820 new.dirty_bitmap
= NULL
;
823 if (change
== KVM_MR_CREATE
) {
824 new.userspace_addr
= mem
->userspace_addr
;
826 if (kvm_arch_create_memslot(&new, npages
))
830 /* Allocate page dirty bitmap if needed */
831 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
832 if (kvm_create_dirty_bitmap(&new) < 0)
836 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
838 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
842 slot
= id_to_memslot(slots
, mem
->slot
);
843 slot
->flags
|= KVM_MEMSLOT_INVALID
;
845 old_memslots
= install_new_memslots(kvm
, slots
, NULL
);
847 /* slot was deleted or moved, clear iommu mapping */
848 kvm_iommu_unmap_pages(kvm
, &old
);
849 /* From this point no new shadow pages pointing to a deleted,
850 * or moved, memslot will be created.
852 * validation of sp->gfn happens in:
853 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
854 * - kvm_is_visible_gfn (mmu_check_roots)
856 kvm_arch_flush_shadow_memslot(kvm
, slot
);
857 slots
= old_memslots
;
860 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
866 * We can re-use the old_memslots from above, the only difference
867 * from the currently installed memslots is the invalid flag. This
868 * will get overwritten by update_memslots anyway.
871 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
878 * IOMMU mapping: New slots need to be mapped. Old slots need to be
879 * un-mapped and re-mapped if their base changes. Since base change
880 * unmapping is handled above with slot deletion, mapping alone is
881 * needed here. Anything else the iommu might care about for existing
882 * slots (size changes, userspace addr changes and read-only flag
883 * changes) is disallowed above, so any other attribute changes getting
884 * here can be skipped.
886 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
887 r
= kvm_iommu_map_pages(kvm
, &new);
892 /* actual memory is freed via old in kvm_free_physmem_slot below */
893 if (change
== KVM_MR_DELETE
) {
894 new.dirty_bitmap
= NULL
;
895 memset(&new.arch
, 0, sizeof(new.arch
));
898 old_memslots
= install_new_memslots(kvm
, slots
, &new);
900 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
902 kvm_free_physmem_slot(&old
, &new);
910 kvm_free_physmem_slot(&new, &old
);
914 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
916 int kvm_set_memory_region(struct kvm
*kvm
,
917 struct kvm_userspace_memory_region
*mem
)
921 mutex_lock(&kvm
->slots_lock
);
922 r
= __kvm_set_memory_region(kvm
, mem
);
923 mutex_unlock(&kvm
->slots_lock
);
926 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
928 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
929 struct kvm_userspace_memory_region
*mem
)
931 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
933 return kvm_set_memory_region(kvm
, mem
);
936 int kvm_get_dirty_log(struct kvm
*kvm
,
937 struct kvm_dirty_log
*log
, int *is_dirty
)
939 struct kvm_memory_slot
*memslot
;
942 unsigned long any
= 0;
945 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
948 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
950 if (!memslot
->dirty_bitmap
)
953 n
= kvm_dirty_bitmap_bytes(memslot
);
955 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
956 any
= memslot
->dirty_bitmap
[i
];
959 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
970 bool kvm_largepages_enabled(void)
972 return largepages_enabled
;
975 void kvm_disable_largepages(void)
977 largepages_enabled
= false;
979 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
981 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
983 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
985 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
987 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
989 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
991 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
992 memslot
->flags
& KVM_MEMSLOT_INVALID
)
997 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
999 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1001 struct vm_area_struct
*vma
;
1002 unsigned long addr
, size
;
1006 addr
= gfn_to_hva(kvm
, gfn
);
1007 if (kvm_is_error_hva(addr
))
1010 down_read(¤t
->mm
->mmap_sem
);
1011 vma
= find_vma(current
->mm
, addr
);
1015 size
= vma_kernel_pagesize(vma
);
1018 up_read(¤t
->mm
->mmap_sem
);
1023 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1025 return slot
->flags
& KVM_MEM_READONLY
;
1028 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1029 gfn_t
*nr_pages
, bool write
)
1031 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1032 return KVM_HVA_ERR_BAD
;
1034 if (memslot_is_readonly(slot
) && write
)
1035 return KVM_HVA_ERR_RO_BAD
;
1038 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1040 return __gfn_to_hva_memslot(slot
, gfn
);
1043 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1046 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1049 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1052 return gfn_to_hva_many(slot
, gfn
, NULL
);
1054 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1056 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1058 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1060 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1063 * The hva returned by this function is only allowed to be read.
1064 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1066 static unsigned long gfn_to_hva_read(struct kvm
*kvm
, gfn_t gfn
)
1068 return __gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
, false);
1071 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1073 return __copy_from_user(data
, hva
, len
);
1076 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1078 return __copy_from_user_inatomic(data
, hva
, len
);
1081 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1082 unsigned long start
, int write
, struct page
**page
)
1084 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1087 flags
|= FOLL_WRITE
;
1089 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1092 static inline int check_user_page_hwpoison(unsigned long addr
)
1094 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1096 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1097 flags
, NULL
, NULL
, NULL
);
1098 return rc
== -EHWPOISON
;
1102 * The atomic path to get the writable pfn which will be stored in @pfn,
1103 * true indicates success, otherwise false is returned.
1105 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1106 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1108 struct page
*page
[1];
1111 if (!(async
|| atomic
))
1115 * Fast pin a writable pfn only if it is a write fault request
1116 * or the caller allows to map a writable pfn for a read fault
1119 if (!(write_fault
|| writable
))
1122 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1124 *pfn
= page_to_pfn(page
[0]);
1135 * The slow path to get the pfn of the specified host virtual address,
1136 * 1 indicates success, -errno is returned if error is detected.
1138 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1139 bool *writable
, pfn_t
*pfn
)
1141 struct page
*page
[1];
1147 *writable
= write_fault
;
1150 down_read(¤t
->mm
->mmap_sem
);
1151 npages
= get_user_page_nowait(current
, current
->mm
,
1152 addr
, write_fault
, page
);
1153 up_read(¤t
->mm
->mmap_sem
);
1155 npages
= get_user_pages_fast(addr
, 1, write_fault
,
1160 /* map read fault as writable if possible */
1161 if (unlikely(!write_fault
) && writable
) {
1162 struct page
*wpage
[1];
1164 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1173 *pfn
= page_to_pfn(page
[0]);
1177 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1179 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1182 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1189 * Pin guest page in memory and return its pfn.
1190 * @addr: host virtual address which maps memory to the guest
1191 * @atomic: whether this function can sleep
1192 * @async: whether this function need to wait IO complete if the
1193 * host page is not in the memory
1194 * @write_fault: whether we should get a writable host page
1195 * @writable: whether it allows to map a writable host page for !@write_fault
1197 * The function will map a writable host page for these two cases:
1198 * 1): @write_fault = true
1199 * 2): @write_fault = false && @writable, @writable will tell the caller
1200 * whether the mapping is writable.
1202 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1203 bool write_fault
, bool *writable
)
1205 struct vm_area_struct
*vma
;
1209 /* we can do it either atomically or asynchronously, not both */
1210 BUG_ON(atomic
&& async
);
1212 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1216 return KVM_PFN_ERR_FAULT
;
1218 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1222 down_read(¤t
->mm
->mmap_sem
);
1223 if (npages
== -EHWPOISON
||
1224 (!async
&& check_user_page_hwpoison(addr
))) {
1225 pfn
= KVM_PFN_ERR_HWPOISON
;
1229 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1232 pfn
= KVM_PFN_ERR_FAULT
;
1233 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1234 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1236 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1238 if (async
&& vma_is_valid(vma
, write_fault
))
1240 pfn
= KVM_PFN_ERR_FAULT
;
1243 up_read(¤t
->mm
->mmap_sem
);
1248 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1249 bool *async
, bool write_fault
, bool *writable
)
1251 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1253 if (addr
== KVM_HVA_ERR_RO_BAD
)
1254 return KVM_PFN_ERR_RO_FAULT
;
1256 if (kvm_is_error_hva(addr
))
1257 return KVM_PFN_NOSLOT
;
1259 /* Do not map writable pfn in the readonly memslot. */
1260 if (writable
&& memslot_is_readonly(slot
)) {
1265 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1269 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1270 bool write_fault
, bool *writable
)
1272 struct kvm_memory_slot
*slot
;
1277 slot
= gfn_to_memslot(kvm
, gfn
);
1279 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1283 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1285 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1287 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1289 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1290 bool write_fault
, bool *writable
)
1292 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1294 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1296 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1298 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1300 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1302 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1305 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1307 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1309 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1311 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1314 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1316 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1318 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1320 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1326 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1327 if (kvm_is_error_hva(addr
))
1330 if (entry
< nr_pages
)
1333 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1335 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1337 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1339 if (is_error_noslot_pfn(pfn
))
1340 return KVM_ERR_PTR_BAD_PAGE
;
1342 if (kvm_is_mmio_pfn(pfn
)) {
1344 return KVM_ERR_PTR_BAD_PAGE
;
1347 return pfn_to_page(pfn
);
1350 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1354 pfn
= gfn_to_pfn(kvm
, gfn
);
1356 return kvm_pfn_to_page(pfn
);
1359 EXPORT_SYMBOL_GPL(gfn_to_page
);
1361 void kvm_release_page_clean(struct page
*page
)
1363 WARN_ON(is_error_page(page
));
1365 kvm_release_pfn_clean(page_to_pfn(page
));
1367 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1369 void kvm_release_pfn_clean(pfn_t pfn
)
1371 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1372 put_page(pfn_to_page(pfn
));
1374 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1376 void kvm_release_page_dirty(struct page
*page
)
1378 WARN_ON(is_error_page(page
));
1380 kvm_release_pfn_dirty(page_to_pfn(page
));
1382 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1384 void kvm_release_pfn_dirty(pfn_t pfn
)
1386 kvm_set_pfn_dirty(pfn
);
1387 kvm_release_pfn_clean(pfn
);
1389 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1391 void kvm_set_page_dirty(struct page
*page
)
1393 kvm_set_pfn_dirty(page_to_pfn(page
));
1395 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1397 void kvm_set_pfn_dirty(pfn_t pfn
)
1399 if (!kvm_is_mmio_pfn(pfn
)) {
1400 struct page
*page
= pfn_to_page(pfn
);
1401 if (!PageReserved(page
))
1405 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1407 void kvm_set_pfn_accessed(pfn_t pfn
)
1409 if (!kvm_is_mmio_pfn(pfn
))
1410 mark_page_accessed(pfn_to_page(pfn
));
1412 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1414 void kvm_get_pfn(pfn_t pfn
)
1416 if (!kvm_is_mmio_pfn(pfn
))
1417 get_page(pfn_to_page(pfn
));
1419 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1421 static int next_segment(unsigned long len
, int offset
)
1423 if (len
> PAGE_SIZE
- offset
)
1424 return PAGE_SIZE
- offset
;
1429 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1435 addr
= gfn_to_hva_read(kvm
, gfn
);
1436 if (kvm_is_error_hva(addr
))
1438 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1443 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1445 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1447 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1449 int offset
= offset_in_page(gpa
);
1452 while ((seg
= next_segment(len
, offset
)) != 0) {
1453 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1463 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1465 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1470 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1471 int offset
= offset_in_page(gpa
);
1473 addr
= gfn_to_hva_read(kvm
, gfn
);
1474 if (kvm_is_error_hva(addr
))
1476 pagefault_disable();
1477 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1483 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1485 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1486 int offset
, int len
)
1491 addr
= gfn_to_hva(kvm
, gfn
);
1492 if (kvm_is_error_hva(addr
))
1494 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1497 mark_page_dirty(kvm
, gfn
);
1500 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1502 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1505 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1507 int offset
= offset_in_page(gpa
);
1510 while ((seg
= next_segment(len
, offset
)) != 0) {
1511 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1522 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1525 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1526 int offset
= offset_in_page(gpa
);
1527 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1530 ghc
->generation
= slots
->generation
;
1531 ghc
->memslot
= gfn_to_memslot(kvm
, gfn
);
1532 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, gfn
, NULL
);
1533 if (!kvm_is_error_hva(ghc
->hva
))
1540 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1542 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1543 void *data
, unsigned long len
)
1545 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1548 if (slots
->generation
!= ghc
->generation
)
1549 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1551 if (kvm_is_error_hva(ghc
->hva
))
1554 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1557 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1561 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1563 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1564 void *data
, unsigned long len
)
1566 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1569 if (slots
->generation
!= ghc
->generation
)
1570 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1572 if (kvm_is_error_hva(ghc
->hva
))
1575 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1581 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1583 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1585 return kvm_write_guest_page(kvm
, gfn
, (const void *) empty_zero_page
,
1588 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1590 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1592 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1594 int offset
= offset_in_page(gpa
);
1597 while ((seg
= next_segment(len
, offset
)) != 0) {
1598 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1607 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1609 void mark_page_dirty_in_slot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
,
1612 if (memslot
&& memslot
->dirty_bitmap
) {
1613 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1615 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1619 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1621 struct kvm_memory_slot
*memslot
;
1623 memslot
= gfn_to_memslot(kvm
, gfn
);
1624 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1628 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1630 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1635 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1637 if (kvm_arch_vcpu_runnable(vcpu
)) {
1638 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1641 if (kvm_cpu_has_pending_timer(vcpu
))
1643 if (signal_pending(current
))
1649 finish_wait(&vcpu
->wq
, &wait
);
1654 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1656 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1659 int cpu
= vcpu
->cpu
;
1660 wait_queue_head_t
*wqp
;
1662 wqp
= kvm_arch_vcpu_wq(vcpu
);
1663 if (waitqueue_active(wqp
)) {
1664 wake_up_interruptible(wqp
);
1665 ++vcpu
->stat
.halt_wakeup
;
1669 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1670 if (kvm_arch_vcpu_should_kick(vcpu
))
1671 smp_send_reschedule(cpu
);
1674 #endif /* !CONFIG_S390 */
1676 void kvm_resched(struct kvm_vcpu
*vcpu
)
1678 if (!need_resched())
1682 EXPORT_SYMBOL_GPL(kvm_resched
);
1684 bool kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1687 struct task_struct
*task
= NULL
;
1691 pid
= rcu_dereference(target
->pid
);
1693 task
= get_pid_task(target
->pid
, PIDTYPE_PID
);
1697 if (task
->flags
& PF_VCPU
) {
1698 put_task_struct(task
);
1701 ret
= yield_to(task
, 1);
1702 put_task_struct(task
);
1706 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1708 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1710 * Helper that checks whether a VCPU is eligible for directed yield.
1711 * Most eligible candidate to yield is decided by following heuristics:
1713 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1714 * (preempted lock holder), indicated by @in_spin_loop.
1715 * Set at the beiginning and cleared at the end of interception/PLE handler.
1717 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1718 * chance last time (mostly it has become eligible now since we have probably
1719 * yielded to lockholder in last iteration. This is done by toggling
1720 * @dy_eligible each time a VCPU checked for eligibility.)
1722 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1723 * to preempted lock-holder could result in wrong VCPU selection and CPU
1724 * burning. Giving priority for a potential lock-holder increases lock
1727 * Since algorithm is based on heuristics, accessing another VCPU data without
1728 * locking does not harm. It may result in trying to yield to same VCPU, fail
1729 * and continue with next VCPU and so on.
1731 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1735 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1736 (vcpu
->spin_loop
.in_spin_loop
&&
1737 vcpu
->spin_loop
.dy_eligible
);
1739 if (vcpu
->spin_loop
.in_spin_loop
)
1740 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1746 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1748 struct kvm
*kvm
= me
->kvm
;
1749 struct kvm_vcpu
*vcpu
;
1750 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1756 kvm_vcpu_set_in_spin_loop(me
, true);
1758 * We boost the priority of a VCPU that is runnable but not
1759 * currently running, because it got preempted by something
1760 * else and called schedule in __vcpu_run. Hopefully that
1761 * VCPU is holding the lock that we need and will release it.
1762 * We approximate round-robin by starting at the last boosted VCPU.
1764 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1765 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1766 if (!pass
&& i
<= last_boosted_vcpu
) {
1767 i
= last_boosted_vcpu
;
1769 } else if (pass
&& i
> last_boosted_vcpu
)
1771 if (!ACCESS_ONCE(vcpu
->preempted
))
1775 if (waitqueue_active(&vcpu
->wq
))
1777 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1780 yielded
= kvm_vcpu_yield_to(vcpu
);
1782 kvm
->last_boosted_vcpu
= i
;
1784 } else if (yielded
< 0) {
1791 kvm_vcpu_set_in_spin_loop(me
, false);
1793 /* Ensure vcpu is not eligible during next spinloop */
1794 kvm_vcpu_set_dy_eligible(me
, false);
1796 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1798 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1800 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1803 if (vmf
->pgoff
== 0)
1804 page
= virt_to_page(vcpu
->run
);
1806 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1807 page
= virt_to_page(vcpu
->arch
.pio_data
);
1809 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1810 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1811 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1814 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1820 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1821 .fault
= kvm_vcpu_fault
,
1824 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1826 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1830 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1832 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1834 kvm_put_kvm(vcpu
->kvm
);
1838 static struct file_operations kvm_vcpu_fops
= {
1839 .release
= kvm_vcpu_release
,
1840 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1841 #ifdef CONFIG_COMPAT
1842 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1844 .mmap
= kvm_vcpu_mmap
,
1845 .llseek
= noop_llseek
,
1849 * Allocates an inode for the vcpu.
1851 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1853 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
);
1857 * Creates some virtual cpus. Good luck creating more than one.
1859 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1862 struct kvm_vcpu
*vcpu
, *v
;
1864 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1866 return PTR_ERR(vcpu
);
1868 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1870 r
= kvm_arch_vcpu_setup(vcpu
);
1874 mutex_lock(&kvm
->lock
);
1875 if (!kvm_vcpu_compatible(vcpu
)) {
1877 goto unlock_vcpu_destroy
;
1879 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1881 goto unlock_vcpu_destroy
;
1884 kvm_for_each_vcpu(r
, v
, kvm
)
1885 if (v
->vcpu_id
== id
) {
1887 goto unlock_vcpu_destroy
;
1890 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1892 /* Now it's all set up, let userspace reach it */
1894 r
= create_vcpu_fd(vcpu
);
1897 goto unlock_vcpu_destroy
;
1900 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1902 atomic_inc(&kvm
->online_vcpus
);
1904 mutex_unlock(&kvm
->lock
);
1905 kvm_arch_vcpu_postcreate(vcpu
);
1908 unlock_vcpu_destroy
:
1909 mutex_unlock(&kvm
->lock
);
1911 kvm_arch_vcpu_destroy(vcpu
);
1915 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1918 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1919 vcpu
->sigset_active
= 1;
1920 vcpu
->sigset
= *sigset
;
1922 vcpu
->sigset_active
= 0;
1926 static long kvm_vcpu_ioctl(struct file
*filp
,
1927 unsigned int ioctl
, unsigned long arg
)
1929 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1930 void __user
*argp
= (void __user
*)arg
;
1932 struct kvm_fpu
*fpu
= NULL
;
1933 struct kvm_sregs
*kvm_sregs
= NULL
;
1935 if (vcpu
->kvm
->mm
!= current
->mm
)
1938 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1940 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1941 * so vcpu_load() would break it.
1943 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1944 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1948 r
= vcpu_load(vcpu
);
1956 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1957 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
1959 case KVM_GET_REGS
: {
1960 struct kvm_regs
*kvm_regs
;
1963 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1966 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1970 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1977 case KVM_SET_REGS
: {
1978 struct kvm_regs
*kvm_regs
;
1981 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
1982 if (IS_ERR(kvm_regs
)) {
1983 r
= PTR_ERR(kvm_regs
);
1986 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1990 case KVM_GET_SREGS
: {
1991 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1995 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1999 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2004 case KVM_SET_SREGS
: {
2005 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2006 if (IS_ERR(kvm_sregs
)) {
2007 r
= PTR_ERR(kvm_sregs
);
2011 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2014 case KVM_GET_MP_STATE
: {
2015 struct kvm_mp_state mp_state
;
2017 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2021 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2026 case KVM_SET_MP_STATE
: {
2027 struct kvm_mp_state mp_state
;
2030 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2032 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2035 case KVM_TRANSLATE
: {
2036 struct kvm_translation tr
;
2039 if (copy_from_user(&tr
, argp
, sizeof tr
))
2041 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2045 if (copy_to_user(argp
, &tr
, sizeof tr
))
2050 case KVM_SET_GUEST_DEBUG
: {
2051 struct kvm_guest_debug dbg
;
2054 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2056 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2059 case KVM_SET_SIGNAL_MASK
: {
2060 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2061 struct kvm_signal_mask kvm_sigmask
;
2062 sigset_t sigset
, *p
;
2067 if (copy_from_user(&kvm_sigmask
, argp
,
2068 sizeof kvm_sigmask
))
2071 if (kvm_sigmask
.len
!= sizeof sigset
)
2074 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2079 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2083 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2087 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2091 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2097 fpu
= memdup_user(argp
, sizeof(*fpu
));
2103 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2107 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2116 #ifdef CONFIG_COMPAT
2117 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2118 unsigned int ioctl
, unsigned long arg
)
2120 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2121 void __user
*argp
= compat_ptr(arg
);
2124 if (vcpu
->kvm
->mm
!= current
->mm
)
2128 case KVM_SET_SIGNAL_MASK
: {
2129 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2130 struct kvm_signal_mask kvm_sigmask
;
2131 compat_sigset_t csigset
;
2136 if (copy_from_user(&kvm_sigmask
, argp
,
2137 sizeof kvm_sigmask
))
2140 if (kvm_sigmask
.len
!= sizeof csigset
)
2143 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2146 sigset_from_compat(&sigset
, &csigset
);
2147 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2149 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2153 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2161 static long kvm_vm_ioctl(struct file
*filp
,
2162 unsigned int ioctl
, unsigned long arg
)
2164 struct kvm
*kvm
= filp
->private_data
;
2165 void __user
*argp
= (void __user
*)arg
;
2168 if (kvm
->mm
!= current
->mm
)
2171 case KVM_CREATE_VCPU
:
2172 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2174 case KVM_SET_USER_MEMORY_REGION
: {
2175 struct kvm_userspace_memory_region kvm_userspace_mem
;
2178 if (copy_from_user(&kvm_userspace_mem
, argp
,
2179 sizeof kvm_userspace_mem
))
2182 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2185 case KVM_GET_DIRTY_LOG
: {
2186 struct kvm_dirty_log log
;
2189 if (copy_from_user(&log
, argp
, sizeof log
))
2191 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2194 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2195 case KVM_REGISTER_COALESCED_MMIO
: {
2196 struct kvm_coalesced_mmio_zone zone
;
2198 if (copy_from_user(&zone
, argp
, sizeof zone
))
2200 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2203 case KVM_UNREGISTER_COALESCED_MMIO
: {
2204 struct kvm_coalesced_mmio_zone zone
;
2206 if (copy_from_user(&zone
, argp
, sizeof zone
))
2208 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2213 struct kvm_irqfd data
;
2216 if (copy_from_user(&data
, argp
, sizeof data
))
2218 r
= kvm_irqfd(kvm
, &data
);
2221 case KVM_IOEVENTFD
: {
2222 struct kvm_ioeventfd data
;
2225 if (copy_from_user(&data
, argp
, sizeof data
))
2227 r
= kvm_ioeventfd(kvm
, &data
);
2230 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2231 case KVM_SET_BOOT_CPU_ID
:
2233 mutex_lock(&kvm
->lock
);
2234 if (atomic_read(&kvm
->online_vcpus
) != 0)
2237 kvm
->bsp_vcpu_id
= arg
;
2238 mutex_unlock(&kvm
->lock
);
2241 #ifdef CONFIG_HAVE_KVM_MSI
2242 case KVM_SIGNAL_MSI
: {
2246 if (copy_from_user(&msi
, argp
, sizeof msi
))
2248 r
= kvm_send_userspace_msi(kvm
, &msi
);
2252 #ifdef __KVM_HAVE_IRQ_LINE
2253 case KVM_IRQ_LINE_STATUS
:
2254 case KVM_IRQ_LINE
: {
2255 struct kvm_irq_level irq_event
;
2258 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2261 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2262 ioctl
== KVM_IRQ_LINE_STATUS
);
2267 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2268 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2277 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2279 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2285 #ifdef CONFIG_COMPAT
2286 struct compat_kvm_dirty_log
{
2290 compat_uptr_t dirty_bitmap
; /* one bit per page */
2295 static long kvm_vm_compat_ioctl(struct file
*filp
,
2296 unsigned int ioctl
, unsigned long arg
)
2298 struct kvm
*kvm
= filp
->private_data
;
2301 if (kvm
->mm
!= current
->mm
)
2304 case KVM_GET_DIRTY_LOG
: {
2305 struct compat_kvm_dirty_log compat_log
;
2306 struct kvm_dirty_log log
;
2309 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2310 sizeof(compat_log
)))
2312 log
.slot
= compat_log
.slot
;
2313 log
.padding1
= compat_log
.padding1
;
2314 log
.padding2
= compat_log
.padding2
;
2315 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2317 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2321 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2329 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2331 struct page
*page
[1];
2334 gfn_t gfn
= vmf
->pgoff
;
2335 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2337 addr
= gfn_to_hva(kvm
, gfn
);
2338 if (kvm_is_error_hva(addr
))
2339 return VM_FAULT_SIGBUS
;
2341 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
2343 if (unlikely(npages
!= 1))
2344 return VM_FAULT_SIGBUS
;
2346 vmf
->page
= page
[0];
2350 static const struct vm_operations_struct kvm_vm_vm_ops
= {
2351 .fault
= kvm_vm_fault
,
2354 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2356 vma
->vm_ops
= &kvm_vm_vm_ops
;
2360 static struct file_operations kvm_vm_fops
= {
2361 .release
= kvm_vm_release
,
2362 .unlocked_ioctl
= kvm_vm_ioctl
,
2363 #ifdef CONFIG_COMPAT
2364 .compat_ioctl
= kvm_vm_compat_ioctl
,
2366 .mmap
= kvm_vm_mmap
,
2367 .llseek
= noop_llseek
,
2370 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2375 kvm
= kvm_create_vm(type
);
2377 return PTR_ERR(kvm
);
2378 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2379 r
= kvm_coalesced_mmio_init(kvm
);
2385 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
2392 static long kvm_dev_ioctl_check_extension_generic(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
:
2406 #ifdef KVM_CAP_IRQ_ROUTING
2407 case KVM_CAP_IRQ_ROUTING
:
2408 return KVM_MAX_IRQ_ROUTES
;
2413 return kvm_dev_ioctl_check_extension(arg
);
2416 static long kvm_dev_ioctl(struct file
*filp
,
2417 unsigned int ioctl
, unsigned long arg
)
2422 case KVM_GET_API_VERSION
:
2426 r
= KVM_API_VERSION
;
2429 r
= kvm_dev_ioctl_create_vm(arg
);
2431 case KVM_CHECK_EXTENSION
:
2432 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2434 case KVM_GET_VCPU_MMAP_SIZE
:
2438 r
= PAGE_SIZE
; /* struct kvm_run */
2440 r
+= PAGE_SIZE
; /* pio data page */
2442 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2443 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2446 case KVM_TRACE_ENABLE
:
2447 case KVM_TRACE_PAUSE
:
2448 case KVM_TRACE_DISABLE
:
2452 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2458 static struct file_operations kvm_chardev_ops
= {
2459 .unlocked_ioctl
= kvm_dev_ioctl
,
2460 .compat_ioctl
= kvm_dev_ioctl
,
2461 .llseek
= noop_llseek
,
2464 static struct miscdevice kvm_dev
= {
2470 static void hardware_enable_nolock(void *junk
)
2472 int cpu
= raw_smp_processor_id();
2475 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2478 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2480 r
= kvm_arch_hardware_enable(NULL
);
2483 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2484 atomic_inc(&hardware_enable_failed
);
2485 printk(KERN_INFO
"kvm: enabling virtualization on "
2486 "CPU%d failed\n", cpu
);
2490 static void hardware_enable(void *junk
)
2492 raw_spin_lock(&kvm_lock
);
2493 hardware_enable_nolock(junk
);
2494 raw_spin_unlock(&kvm_lock
);
2497 static void hardware_disable_nolock(void *junk
)
2499 int cpu
= raw_smp_processor_id();
2501 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2503 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2504 kvm_arch_hardware_disable(NULL
);
2507 static void hardware_disable(void *junk
)
2509 raw_spin_lock(&kvm_lock
);
2510 hardware_disable_nolock(junk
);
2511 raw_spin_unlock(&kvm_lock
);
2514 static void hardware_disable_all_nolock(void)
2516 BUG_ON(!kvm_usage_count
);
2519 if (!kvm_usage_count
)
2520 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2523 static void hardware_disable_all(void)
2525 raw_spin_lock(&kvm_lock
);
2526 hardware_disable_all_nolock();
2527 raw_spin_unlock(&kvm_lock
);
2530 static int hardware_enable_all(void)
2534 raw_spin_lock(&kvm_lock
);
2537 if (kvm_usage_count
== 1) {
2538 atomic_set(&hardware_enable_failed
, 0);
2539 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2541 if (atomic_read(&hardware_enable_failed
)) {
2542 hardware_disable_all_nolock();
2547 raw_spin_unlock(&kvm_lock
);
2552 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2557 if (!kvm_usage_count
)
2560 val
&= ~CPU_TASKS_FROZEN
;
2563 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2565 hardware_disable(NULL
);
2568 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2570 hardware_enable(NULL
);
2576 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2580 * Some (well, at least mine) BIOSes hang on reboot if
2583 * And Intel TXT required VMX off for all cpu when system shutdown.
2585 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2586 kvm_rebooting
= true;
2587 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2591 static struct notifier_block kvm_reboot_notifier
= {
2592 .notifier_call
= kvm_reboot
,
2596 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2600 for (i
= 0; i
< bus
->dev_count
; i
++) {
2601 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2603 kvm_iodevice_destructor(pos
);
2608 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2610 const struct kvm_io_range
*r1
= p1
;
2611 const struct kvm_io_range
*r2
= p2
;
2613 if (r1
->addr
< r2
->addr
)
2615 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2620 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2621 gpa_t addr
, int len
)
2623 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2629 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2630 kvm_io_bus_sort_cmp
, NULL
);
2635 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2636 gpa_t addr
, int len
)
2638 struct kvm_io_range
*range
, key
;
2641 key
= (struct kvm_io_range
) {
2646 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2647 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2651 off
= range
- bus
->range
;
2653 while (off
> 0 && kvm_io_bus_sort_cmp(&key
, &bus
->range
[off
-1]) == 0)
2659 /* kvm_io_bus_write - called under kvm->slots_lock */
2660 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2661 int len
, const void *val
)
2664 struct kvm_io_bus
*bus
;
2665 struct kvm_io_range range
;
2667 range
= (struct kvm_io_range
) {
2672 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2673 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2677 while (idx
< bus
->dev_count
&&
2678 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2679 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, addr
, len
, val
))
2687 /* kvm_io_bus_read - called under kvm->slots_lock */
2688 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2692 struct kvm_io_bus
*bus
;
2693 struct kvm_io_range range
;
2695 range
= (struct kvm_io_range
) {
2700 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2701 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2705 while (idx
< bus
->dev_count
&&
2706 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2707 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, addr
, len
, val
))
2715 /* Caller must hold slots_lock. */
2716 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2717 int len
, struct kvm_io_device
*dev
)
2719 struct kvm_io_bus
*new_bus
, *bus
;
2721 bus
= kvm
->buses
[bus_idx
];
2722 if (bus
->dev_count
> NR_IOBUS_DEVS
- 1)
2725 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
2726 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2729 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
2730 sizeof(struct kvm_io_range
)));
2731 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
2732 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2733 synchronize_srcu_expedited(&kvm
->srcu
);
2739 /* Caller must hold slots_lock. */
2740 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2741 struct kvm_io_device
*dev
)
2744 struct kvm_io_bus
*new_bus
, *bus
;
2746 bus
= kvm
->buses
[bus_idx
];
2748 for (i
= 0; i
< bus
->dev_count
; i
++)
2749 if (bus
->range
[i
].dev
== dev
) {
2757 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
2758 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2762 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
2763 new_bus
->dev_count
--;
2764 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
2765 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
2767 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2768 synchronize_srcu_expedited(&kvm
->srcu
);
2773 static struct notifier_block kvm_cpu_notifier
= {
2774 .notifier_call
= kvm_cpu_hotplug
,
2777 static int vm_stat_get(void *_offset
, u64
*val
)
2779 unsigned offset
= (long)_offset
;
2783 raw_spin_lock(&kvm_lock
);
2784 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2785 *val
+= *(u32
*)((void *)kvm
+ offset
);
2786 raw_spin_unlock(&kvm_lock
);
2790 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2792 static int vcpu_stat_get(void *_offset
, u64
*val
)
2794 unsigned offset
= (long)_offset
;
2796 struct kvm_vcpu
*vcpu
;
2800 raw_spin_lock(&kvm_lock
);
2801 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2802 kvm_for_each_vcpu(i
, vcpu
, kvm
)
2803 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2805 raw_spin_unlock(&kvm_lock
);
2809 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2811 static const struct file_operations
*stat_fops
[] = {
2812 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2813 [KVM_STAT_VM
] = &vm_stat_fops
,
2816 static int kvm_init_debug(void)
2819 struct kvm_stats_debugfs_item
*p
;
2821 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2822 if (kvm_debugfs_dir
== NULL
)
2825 for (p
= debugfs_entries
; p
->name
; ++p
) {
2826 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2827 (void *)(long)p
->offset
,
2828 stat_fops
[p
->kind
]);
2829 if (p
->dentry
== NULL
)
2836 debugfs_remove_recursive(kvm_debugfs_dir
);
2841 static void kvm_exit_debug(void)
2843 struct kvm_stats_debugfs_item
*p
;
2845 for (p
= debugfs_entries
; p
->name
; ++p
)
2846 debugfs_remove(p
->dentry
);
2847 debugfs_remove(kvm_debugfs_dir
);
2850 static int kvm_suspend(void)
2852 if (kvm_usage_count
)
2853 hardware_disable_nolock(NULL
);
2857 static void kvm_resume(void)
2859 if (kvm_usage_count
) {
2860 WARN_ON(raw_spin_is_locked(&kvm_lock
));
2861 hardware_enable_nolock(NULL
);
2865 static struct syscore_ops kvm_syscore_ops
= {
2866 .suspend
= kvm_suspend
,
2867 .resume
= kvm_resume
,
2871 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2873 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2876 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2878 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2879 if (vcpu
->preempted
)
2880 vcpu
->preempted
= false;
2882 kvm_arch_vcpu_load(vcpu
, cpu
);
2885 static void kvm_sched_out(struct preempt_notifier
*pn
,
2886 struct task_struct
*next
)
2888 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2890 if (current
->state
== TASK_RUNNING
)
2891 vcpu
->preempted
= true;
2892 kvm_arch_vcpu_put(vcpu
);
2895 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
2896 struct module
*module
)
2901 r
= kvm_irqfd_init();
2904 r
= kvm_arch_init(opaque
);
2908 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2913 r
= kvm_arch_hardware_setup();
2917 for_each_online_cpu(cpu
) {
2918 smp_call_function_single(cpu
,
2919 kvm_arch_check_processor_compat
,
2925 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2928 register_reboot_notifier(&kvm_reboot_notifier
);
2930 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2932 vcpu_align
= __alignof__(struct kvm_vcpu
);
2933 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
2935 if (!kvm_vcpu_cache
) {
2940 r
= kvm_async_pf_init();
2944 kvm_chardev_ops
.owner
= module
;
2945 kvm_vm_fops
.owner
= module
;
2946 kvm_vcpu_fops
.owner
= module
;
2948 r
= misc_register(&kvm_dev
);
2950 printk(KERN_ERR
"kvm: misc device register failed\n");
2954 register_syscore_ops(&kvm_syscore_ops
);
2956 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2957 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2959 r
= kvm_init_debug();
2961 printk(KERN_ERR
"kvm: create debugfs files failed\n");
2968 unregister_syscore_ops(&kvm_syscore_ops
);
2970 kvm_async_pf_deinit();
2972 kmem_cache_destroy(kvm_vcpu_cache
);
2974 unregister_reboot_notifier(&kvm_reboot_notifier
);
2975 unregister_cpu_notifier(&kvm_cpu_notifier
);
2978 kvm_arch_hardware_unsetup();
2980 free_cpumask_var(cpus_hardware_enabled
);
2988 EXPORT_SYMBOL_GPL(kvm_init
);
2993 misc_deregister(&kvm_dev
);
2994 kmem_cache_destroy(kvm_vcpu_cache
);
2995 kvm_async_pf_deinit();
2996 unregister_syscore_ops(&kvm_syscore_ops
);
2997 unregister_reboot_notifier(&kvm_reboot_notifier
);
2998 unregister_cpu_notifier(&kvm_cpu_notifier
);
2999 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3000 kvm_arch_hardware_unsetup();
3003 free_cpumask_var(cpus_hardware_enabled
);
3005 EXPORT_SYMBOL_GPL(kvm_exit
);