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_SPINLOCK(kvm_lock
);
74 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
77 static cpumask_var_t cpus_hardware_enabled
;
78 static int kvm_usage_count
= 0;
79 static atomic_t hardware_enable_failed
;
81 struct kmem_cache
*kvm_vcpu_cache
;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
84 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
86 struct dentry
*kvm_debugfs_dir
;
88 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
91 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
97 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
100 EXPORT_SYMBOL_GPL(kvm_rebooting
);
102 static bool largepages_enabled
= true;
104 bool kvm_is_mmio_pfn(pfn_t pfn
)
107 return PageReserved(pfn_to_page(pfn
));
113 * Switches to specified vcpu, until a matching vcpu_put()
115 int vcpu_load(struct kvm_vcpu
*vcpu
)
119 if (mutex_lock_killable(&vcpu
->mutex
))
121 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
122 /* The thread running this VCPU changed. */
123 struct pid
*oldpid
= vcpu
->pid
;
124 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
125 rcu_assign_pointer(vcpu
->pid
, newpid
);
130 preempt_notifier_register(&vcpu
->preempt_notifier
);
131 kvm_arch_vcpu_load(vcpu
, cpu
);
136 void vcpu_put(struct kvm_vcpu
*vcpu
)
139 kvm_arch_vcpu_put(vcpu
);
140 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
142 mutex_unlock(&vcpu
->mutex
);
145 static void ack_flush(void *_completed
)
149 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
154 struct kvm_vcpu
*vcpu
;
156 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
159 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
160 kvm_make_request(req
, vcpu
);
163 /* Set ->requests bit before we read ->mode */
166 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
167 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
168 cpumask_set_cpu(cpu
, cpus
);
170 if (unlikely(cpus
== NULL
))
171 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
172 else if (!cpumask_empty(cpus
))
173 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
177 free_cpumask_var(cpus
);
181 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
183 long dirty_count
= kvm
->tlbs_dirty
;
186 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
187 ++kvm
->stat
.remote_tlb_flush
;
188 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
190 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
192 void kvm_reload_remote_mmus(struct kvm
*kvm
)
194 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
197 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
199 make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
202 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
204 make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
207 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
212 mutex_init(&vcpu
->mutex
);
217 init_waitqueue_head(&vcpu
->wq
);
218 kvm_async_pf_vcpu_init(vcpu
);
220 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
225 vcpu
->run
= page_address(page
);
227 kvm_vcpu_set_in_spin_loop(vcpu
, false);
228 kvm_vcpu_set_dy_eligible(vcpu
, false);
229 vcpu
->preempted
= false;
231 r
= kvm_arch_vcpu_init(vcpu
);
237 free_page((unsigned long)vcpu
->run
);
241 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
243 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
246 kvm_arch_vcpu_uninit(vcpu
);
247 free_page((unsigned long)vcpu
->run
);
249 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
251 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
252 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
254 return container_of(mn
, struct kvm
, mmu_notifier
);
257 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
258 struct mm_struct
*mm
,
259 unsigned long address
)
261 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
262 int need_tlb_flush
, idx
;
265 * When ->invalidate_page runs, the linux pte has been zapped
266 * already but the page is still allocated until
267 * ->invalidate_page returns. So if we increase the sequence
268 * here the kvm page fault will notice if the spte can't be
269 * established because the page is going to be freed. If
270 * instead the kvm page fault establishes the spte before
271 * ->invalidate_page runs, kvm_unmap_hva will release it
274 * The sequence increase only need to be seen at spin_unlock
275 * time, and not at spin_lock time.
277 * Increasing the sequence after the spin_unlock would be
278 * unsafe because the kvm page fault could then establish the
279 * pte after kvm_unmap_hva returned, without noticing the page
280 * is going to be freed.
282 idx
= srcu_read_lock(&kvm
->srcu
);
283 spin_lock(&kvm
->mmu_lock
);
285 kvm
->mmu_notifier_seq
++;
286 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
287 /* we've to flush the tlb before the pages can be freed */
289 kvm_flush_remote_tlbs(kvm
);
291 spin_unlock(&kvm
->mmu_lock
);
292 srcu_read_unlock(&kvm
->srcu
, idx
);
295 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
296 struct mm_struct
*mm
,
297 unsigned long address
,
300 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
303 idx
= srcu_read_lock(&kvm
->srcu
);
304 spin_lock(&kvm
->mmu_lock
);
305 kvm
->mmu_notifier_seq
++;
306 kvm_set_spte_hva(kvm
, address
, pte
);
307 spin_unlock(&kvm
->mmu_lock
);
308 srcu_read_unlock(&kvm
->srcu
, idx
);
311 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
312 struct mm_struct
*mm
,
316 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
317 int need_tlb_flush
= 0, idx
;
319 idx
= srcu_read_lock(&kvm
->srcu
);
320 spin_lock(&kvm
->mmu_lock
);
322 * The count increase must become visible at unlock time as no
323 * spte can be established without taking the mmu_lock and
324 * count is also read inside the mmu_lock critical section.
326 kvm
->mmu_notifier_count
++;
327 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
328 need_tlb_flush
|= kvm
->tlbs_dirty
;
329 /* we've to flush the tlb before the pages can be freed */
331 kvm_flush_remote_tlbs(kvm
);
333 spin_unlock(&kvm
->mmu_lock
);
334 srcu_read_unlock(&kvm
->srcu
, idx
);
337 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
338 struct mm_struct
*mm
,
342 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
344 spin_lock(&kvm
->mmu_lock
);
346 * This sequence increase will notify the kvm page fault that
347 * the page that is going to be mapped in the spte could have
350 kvm
->mmu_notifier_seq
++;
353 * The above sequence increase must be visible before the
354 * below count decrease, which is ensured by the smp_wmb above
355 * in conjunction with the smp_rmb in mmu_notifier_retry().
357 kvm
->mmu_notifier_count
--;
358 spin_unlock(&kvm
->mmu_lock
);
360 BUG_ON(kvm
->mmu_notifier_count
< 0);
363 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
364 struct mm_struct
*mm
,
365 unsigned long address
)
367 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
370 idx
= srcu_read_lock(&kvm
->srcu
);
371 spin_lock(&kvm
->mmu_lock
);
373 young
= kvm_age_hva(kvm
, address
);
375 kvm_flush_remote_tlbs(kvm
);
377 spin_unlock(&kvm
->mmu_lock
);
378 srcu_read_unlock(&kvm
->srcu
, idx
);
383 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
384 struct mm_struct
*mm
,
385 unsigned long address
)
387 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
390 idx
= srcu_read_lock(&kvm
->srcu
);
391 spin_lock(&kvm
->mmu_lock
);
392 young
= kvm_test_age_hva(kvm
, address
);
393 spin_unlock(&kvm
->mmu_lock
);
394 srcu_read_unlock(&kvm
->srcu
, idx
);
399 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
400 struct mm_struct
*mm
)
402 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
405 idx
= srcu_read_lock(&kvm
->srcu
);
406 kvm_arch_flush_shadow_all(kvm
);
407 srcu_read_unlock(&kvm
->srcu
, idx
);
410 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
411 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
412 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
413 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
414 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
415 .test_young
= kvm_mmu_notifier_test_young
,
416 .change_pte
= kvm_mmu_notifier_change_pte
,
417 .release
= kvm_mmu_notifier_release
,
420 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
422 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
423 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
426 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
428 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
433 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
435 static void kvm_init_memslots_id(struct kvm
*kvm
)
438 struct kvm_memslots
*slots
= kvm
->memslots
;
440 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
441 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
444 static struct kvm
*kvm_create_vm(unsigned long type
)
447 struct kvm
*kvm
= kvm_arch_alloc_vm();
450 return ERR_PTR(-ENOMEM
);
452 r
= kvm_arch_init_vm(kvm
, type
);
454 goto out_err_nodisable
;
456 r
= hardware_enable_all();
458 goto out_err_nodisable
;
460 #ifdef CONFIG_HAVE_KVM_IRQCHIP
461 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
462 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
465 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
468 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
471 kvm_init_memslots_id(kvm
);
472 if (init_srcu_struct(&kvm
->srcu
))
474 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
475 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
481 spin_lock_init(&kvm
->mmu_lock
);
482 kvm
->mm
= current
->mm
;
483 atomic_inc(&kvm
->mm
->mm_count
);
484 kvm_eventfd_init(kvm
);
485 mutex_init(&kvm
->lock
);
486 mutex_init(&kvm
->irq_lock
);
487 mutex_init(&kvm
->slots_lock
);
488 atomic_set(&kvm
->users_count
, 1);
489 INIT_LIST_HEAD(&kvm
->devices
);
491 r
= kvm_init_mmu_notifier(kvm
);
495 spin_lock(&kvm_lock
);
496 list_add(&kvm
->vm_list
, &vm_list
);
497 spin_unlock(&kvm_lock
);
502 cleanup_srcu_struct(&kvm
->srcu
);
504 hardware_disable_all();
506 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
507 kfree(kvm
->buses
[i
]);
508 kfree(kvm
->memslots
);
509 kvm_arch_free_vm(kvm
);
514 * Avoid using vmalloc for a small buffer.
515 * Should not be used when the size is statically known.
517 void *kvm_kvzalloc(unsigned long size
)
519 if (size
> PAGE_SIZE
)
520 return vzalloc(size
);
522 return kzalloc(size
, GFP_KERNEL
);
525 void kvm_kvfree(const void *addr
)
527 if (is_vmalloc_addr(addr
))
533 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
535 if (!memslot
->dirty_bitmap
)
538 kvm_kvfree(memslot
->dirty_bitmap
);
539 memslot
->dirty_bitmap
= NULL
;
543 * Free any memory in @free but not in @dont.
545 static void kvm_free_physmem_slot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
546 struct kvm_memory_slot
*dont
)
548 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
549 kvm_destroy_dirty_bitmap(free
);
551 kvm_arch_free_memslot(kvm
, free
, dont
);
556 void kvm_free_physmem(struct kvm
*kvm
)
558 struct kvm_memslots
*slots
= kvm
->memslots
;
559 struct kvm_memory_slot
*memslot
;
561 kvm_for_each_memslot(memslot
, slots
)
562 kvm_free_physmem_slot(kvm
, memslot
, NULL
);
564 kfree(kvm
->memslots
);
567 static void kvm_destroy_devices(struct kvm
*kvm
)
569 struct list_head
*node
, *tmp
;
571 list_for_each_safe(node
, tmp
, &kvm
->devices
) {
572 struct kvm_device
*dev
=
573 list_entry(node
, struct kvm_device
, vm_node
);
576 dev
->ops
->destroy(dev
);
580 static void kvm_destroy_vm(struct kvm
*kvm
)
583 struct mm_struct
*mm
= kvm
->mm
;
585 kvm_arch_sync_events(kvm
);
586 spin_lock(&kvm_lock
);
587 list_del(&kvm
->vm_list
);
588 spin_unlock(&kvm_lock
);
589 kvm_free_irq_routing(kvm
);
590 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
591 kvm_io_bus_destroy(kvm
->buses
[i
]);
592 kvm_coalesced_mmio_free(kvm
);
593 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
594 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
596 kvm_arch_flush_shadow_all(kvm
);
598 kvm_arch_destroy_vm(kvm
);
599 kvm_destroy_devices(kvm
);
600 kvm_free_physmem(kvm
);
601 cleanup_srcu_struct(&kvm
->srcu
);
602 kvm_arch_free_vm(kvm
);
603 hardware_disable_all();
607 void kvm_get_kvm(struct kvm
*kvm
)
609 atomic_inc(&kvm
->users_count
);
611 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
613 void kvm_put_kvm(struct kvm
*kvm
)
615 if (atomic_dec_and_test(&kvm
->users_count
))
618 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
621 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
623 struct kvm
*kvm
= filp
->private_data
;
625 kvm_irqfd_release(kvm
);
632 * Allocation size is twice as large as the actual dirty bitmap size.
633 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
635 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
638 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
640 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
641 if (!memslot
->dirty_bitmap
)
644 #endif /* !CONFIG_S390 */
648 static int cmp_memslot(const void *slot1
, const void *slot2
)
650 struct kvm_memory_slot
*s1
, *s2
;
652 s1
= (struct kvm_memory_slot
*)slot1
;
653 s2
= (struct kvm_memory_slot
*)slot2
;
655 if (s1
->npages
< s2
->npages
)
657 if (s1
->npages
> s2
->npages
)
664 * Sort the memslots base on its size, so the larger slots
665 * will get better fit.
667 static void sort_memslots(struct kvm_memslots
*slots
)
671 sort(slots
->memslots
, KVM_MEM_SLOTS_NUM
,
672 sizeof(struct kvm_memory_slot
), cmp_memslot
, NULL
);
674 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
675 slots
->id_to_index
[slots
->memslots
[i
].id
] = i
;
678 void update_memslots(struct kvm_memslots
*slots
, struct kvm_memory_slot
*new,
683 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
684 unsigned long npages
= old
->npages
;
687 if (new->npages
!= npages
)
688 sort_memslots(slots
);
691 slots
->generation
= last_generation
+ 1;
694 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
696 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
698 #ifdef KVM_CAP_READONLY_MEM
699 valid_flags
|= KVM_MEM_READONLY
;
702 if (mem
->flags
& ~valid_flags
)
708 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
709 struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
711 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
713 update_memslots(slots
, new, kvm
->memslots
->generation
);
714 rcu_assign_pointer(kvm
->memslots
, slots
);
715 synchronize_srcu_expedited(&kvm
->srcu
);
717 kvm_arch_memslots_updated(kvm
);
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(kvm
, &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
),
877 /* actual memory is freed via old in kvm_free_physmem_slot below */
878 if (change
== KVM_MR_DELETE
) {
879 new.dirty_bitmap
= NULL
;
880 memset(&new.arch
, 0, sizeof(new.arch
));
883 old_memslots
= install_new_memslots(kvm
, slots
, &new);
885 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
887 kvm_free_physmem_slot(kvm
, &old
, &new);
891 * IOMMU mapping: New slots need to be mapped. Old slots need to be
892 * un-mapped and re-mapped if their base changes. Since base change
893 * unmapping is handled above with slot deletion, mapping alone is
894 * needed here. Anything else the iommu might care about for existing
895 * slots (size changes, userspace addr changes and read-only flag
896 * changes) is disallowed above, so any other attribute changes getting
897 * here can be skipped.
899 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
900 r
= kvm_iommu_map_pages(kvm
, &new);
909 kvm_free_physmem_slot(kvm
, &new, &old
);
913 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
915 int kvm_set_memory_region(struct kvm
*kvm
,
916 struct kvm_userspace_memory_region
*mem
)
920 mutex_lock(&kvm
->slots_lock
);
921 r
= __kvm_set_memory_region(kvm
, mem
);
922 mutex_unlock(&kvm
->slots_lock
);
925 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
927 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
928 struct kvm_userspace_memory_region
*mem
)
930 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
932 return kvm_set_memory_region(kvm
, mem
);
935 int kvm_get_dirty_log(struct kvm
*kvm
,
936 struct kvm_dirty_log
*log
, int *is_dirty
)
938 struct kvm_memory_slot
*memslot
;
941 unsigned long any
= 0;
944 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
947 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
949 if (!memslot
->dirty_bitmap
)
952 n
= kvm_dirty_bitmap_bytes(memslot
);
954 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
955 any
= memslot
->dirty_bitmap
[i
];
958 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
968 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
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 * If writable is set to false, the hva returned by this function is only
1064 * allowed to be read.
1066 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1068 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1069 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1071 if (!kvm_is_error_hva(hva
) && writable
)
1072 *writable
= !memslot_is_readonly(slot
);
1077 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1079 return __copy_from_user(data
, hva
, len
);
1082 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1084 return __copy_from_user_inatomic(data
, hva
, len
);
1087 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1088 unsigned long start
, int write
, struct page
**page
)
1090 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1093 flags
|= FOLL_WRITE
;
1095 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1098 static inline int check_user_page_hwpoison(unsigned long addr
)
1100 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1102 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1103 flags
, NULL
, NULL
, NULL
);
1104 return rc
== -EHWPOISON
;
1108 * The atomic path to get the writable pfn which will be stored in @pfn,
1109 * true indicates success, otherwise false is returned.
1111 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1112 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1114 struct page
*page
[1];
1117 if (!(async
|| atomic
))
1121 * Fast pin a writable pfn only if it is a write fault request
1122 * or the caller allows to map a writable pfn for a read fault
1125 if (!(write_fault
|| writable
))
1128 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1130 *pfn
= page_to_pfn(page
[0]);
1141 * The slow path to get the pfn of the specified host virtual address,
1142 * 1 indicates success, -errno is returned if error is detected.
1144 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1145 bool *writable
, pfn_t
*pfn
)
1147 struct page
*page
[1];
1153 *writable
= write_fault
;
1156 down_read(¤t
->mm
->mmap_sem
);
1157 npages
= get_user_page_nowait(current
, current
->mm
,
1158 addr
, write_fault
, page
);
1159 up_read(¤t
->mm
->mmap_sem
);
1161 npages
= get_user_pages_fast(addr
, 1, write_fault
,
1166 /* map read fault as writable if possible */
1167 if (unlikely(!write_fault
) && writable
) {
1168 struct page
*wpage
[1];
1170 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1179 *pfn
= page_to_pfn(page
[0]);
1183 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1185 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1188 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1195 * Pin guest page in memory and return its pfn.
1196 * @addr: host virtual address which maps memory to the guest
1197 * @atomic: whether this function can sleep
1198 * @async: whether this function need to wait IO complete if the
1199 * host page is not in the memory
1200 * @write_fault: whether we should get a writable host page
1201 * @writable: whether it allows to map a writable host page for !@write_fault
1203 * The function will map a writable host page for these two cases:
1204 * 1): @write_fault = true
1205 * 2): @write_fault = false && @writable, @writable will tell the caller
1206 * whether the mapping is writable.
1208 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1209 bool write_fault
, bool *writable
)
1211 struct vm_area_struct
*vma
;
1215 /* we can do it either atomically or asynchronously, not both */
1216 BUG_ON(atomic
&& async
);
1218 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1222 return KVM_PFN_ERR_FAULT
;
1224 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1228 down_read(¤t
->mm
->mmap_sem
);
1229 if (npages
== -EHWPOISON
||
1230 (!async
&& check_user_page_hwpoison(addr
))) {
1231 pfn
= KVM_PFN_ERR_HWPOISON
;
1235 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1238 pfn
= KVM_PFN_ERR_FAULT
;
1239 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1240 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1242 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1244 if (async
&& vma_is_valid(vma
, write_fault
))
1246 pfn
= KVM_PFN_ERR_FAULT
;
1249 up_read(¤t
->mm
->mmap_sem
);
1254 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1255 bool *async
, bool write_fault
, bool *writable
)
1257 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1259 if (addr
== KVM_HVA_ERR_RO_BAD
)
1260 return KVM_PFN_ERR_RO_FAULT
;
1262 if (kvm_is_error_hva(addr
))
1263 return KVM_PFN_NOSLOT
;
1265 /* Do not map writable pfn in the readonly memslot. */
1266 if (writable
&& memslot_is_readonly(slot
)) {
1271 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1275 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1276 bool write_fault
, bool *writable
)
1278 struct kvm_memory_slot
*slot
;
1283 slot
= gfn_to_memslot(kvm
, gfn
);
1285 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1289 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1291 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1293 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1295 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1296 bool write_fault
, bool *writable
)
1298 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1300 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1302 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1304 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1306 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1308 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1311 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1315 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1317 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1320 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1322 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1324 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1326 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1332 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1333 if (kvm_is_error_hva(addr
))
1336 if (entry
< nr_pages
)
1339 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1341 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1343 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1345 if (is_error_noslot_pfn(pfn
))
1346 return KVM_ERR_PTR_BAD_PAGE
;
1348 if (kvm_is_mmio_pfn(pfn
)) {
1350 return KVM_ERR_PTR_BAD_PAGE
;
1353 return pfn_to_page(pfn
);
1356 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1360 pfn
= gfn_to_pfn(kvm
, gfn
);
1362 return kvm_pfn_to_page(pfn
);
1365 EXPORT_SYMBOL_GPL(gfn_to_page
);
1367 void kvm_release_page_clean(struct page
*page
)
1369 WARN_ON(is_error_page(page
));
1371 kvm_release_pfn_clean(page_to_pfn(page
));
1373 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1375 void kvm_release_pfn_clean(pfn_t pfn
)
1377 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1378 put_page(pfn_to_page(pfn
));
1380 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1382 void kvm_release_page_dirty(struct page
*page
)
1384 WARN_ON(is_error_page(page
));
1386 kvm_release_pfn_dirty(page_to_pfn(page
));
1388 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1390 void kvm_release_pfn_dirty(pfn_t pfn
)
1392 kvm_set_pfn_dirty(pfn
);
1393 kvm_release_pfn_clean(pfn
);
1395 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1397 void kvm_set_page_dirty(struct page
*page
)
1399 kvm_set_pfn_dirty(page_to_pfn(page
));
1401 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1403 void kvm_set_pfn_dirty(pfn_t pfn
)
1405 if (!kvm_is_mmio_pfn(pfn
)) {
1406 struct page
*page
= pfn_to_page(pfn
);
1407 if (!PageReserved(page
))
1411 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1413 void kvm_set_pfn_accessed(pfn_t pfn
)
1415 if (!kvm_is_mmio_pfn(pfn
))
1416 mark_page_accessed(pfn_to_page(pfn
));
1418 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1420 void kvm_get_pfn(pfn_t pfn
)
1422 if (!kvm_is_mmio_pfn(pfn
))
1423 get_page(pfn_to_page(pfn
));
1425 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1427 static int next_segment(unsigned long len
, int offset
)
1429 if (len
> PAGE_SIZE
- offset
)
1430 return PAGE_SIZE
- offset
;
1435 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1441 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1442 if (kvm_is_error_hva(addr
))
1444 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1449 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1451 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1453 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1455 int offset
= offset_in_page(gpa
);
1458 while ((seg
= next_segment(len
, offset
)) != 0) {
1459 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1469 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1471 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1476 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1477 int offset
= offset_in_page(gpa
);
1479 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1480 if (kvm_is_error_hva(addr
))
1482 pagefault_disable();
1483 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1489 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1491 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1492 int offset
, int len
)
1497 addr
= gfn_to_hva(kvm
, gfn
);
1498 if (kvm_is_error_hva(addr
))
1500 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1503 mark_page_dirty(kvm
, gfn
);
1506 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1508 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1511 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1513 int offset
= offset_in_page(gpa
);
1516 while ((seg
= next_segment(len
, offset
)) != 0) {
1517 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1528 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1529 gpa_t gpa
, unsigned long len
)
1531 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1532 int offset
= offset_in_page(gpa
);
1533 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1534 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1535 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1536 gfn_t nr_pages_avail
;
1539 ghc
->generation
= slots
->generation
;
1541 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1542 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, &nr_pages_avail
);
1543 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_avail
>= nr_pages_needed
) {
1547 * If the requested region crosses two memslots, we still
1548 * verify that the entire region is valid here.
1550 while (start_gfn
<= end_gfn
) {
1551 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1552 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1554 if (kvm_is_error_hva(ghc
->hva
))
1556 start_gfn
+= nr_pages_avail
;
1558 /* Use the slow path for cross page reads and writes. */
1559 ghc
->memslot
= NULL
;
1563 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1565 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1566 void *data
, unsigned long len
)
1568 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1571 BUG_ON(len
> ghc
->len
);
1573 if (slots
->generation
!= ghc
->generation
)
1574 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1576 if (unlikely(!ghc
->memslot
))
1577 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1579 if (kvm_is_error_hva(ghc
->hva
))
1582 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1585 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1589 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1591 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1592 void *data
, unsigned long len
)
1594 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1597 BUG_ON(len
> ghc
->len
);
1599 if (slots
->generation
!= ghc
->generation
)
1600 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1602 if (unlikely(!ghc
->memslot
))
1603 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1605 if (kvm_is_error_hva(ghc
->hva
))
1608 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1614 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1616 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1618 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1620 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1622 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1624 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1626 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1628 int offset
= offset_in_page(gpa
);
1631 while ((seg
= next_segment(len
, offset
)) != 0) {
1632 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1641 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1643 void mark_page_dirty_in_slot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
,
1646 if (memslot
&& memslot
->dirty_bitmap
) {
1647 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1649 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1653 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1655 struct kvm_memory_slot
*memslot
;
1657 memslot
= gfn_to_memslot(kvm
, gfn
);
1658 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1660 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1663 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1665 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1670 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1672 if (kvm_arch_vcpu_runnable(vcpu
)) {
1673 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1676 if (kvm_cpu_has_pending_timer(vcpu
))
1678 if (signal_pending(current
))
1684 finish_wait(&vcpu
->wq
, &wait
);
1686 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
1690 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1692 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1695 int cpu
= vcpu
->cpu
;
1696 wait_queue_head_t
*wqp
;
1698 wqp
= kvm_arch_vcpu_wq(vcpu
);
1699 if (waitqueue_active(wqp
)) {
1700 wake_up_interruptible(wqp
);
1701 ++vcpu
->stat
.halt_wakeup
;
1705 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1706 if (kvm_arch_vcpu_should_kick(vcpu
))
1707 smp_send_reschedule(cpu
);
1710 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1711 #endif /* !CONFIG_S390 */
1713 void kvm_resched(struct kvm_vcpu
*vcpu
)
1715 if (!need_resched())
1719 EXPORT_SYMBOL_GPL(kvm_resched
);
1721 bool kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1724 struct task_struct
*task
= NULL
;
1728 pid
= rcu_dereference(target
->pid
);
1730 task
= get_pid_task(target
->pid
, PIDTYPE_PID
);
1734 if (task
->flags
& PF_VCPU
) {
1735 put_task_struct(task
);
1738 ret
= yield_to(task
, 1);
1739 put_task_struct(task
);
1743 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1745 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1747 * Helper that checks whether a VCPU is eligible for directed yield.
1748 * Most eligible candidate to yield is decided by following heuristics:
1750 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1751 * (preempted lock holder), indicated by @in_spin_loop.
1752 * Set at the beiginning and cleared at the end of interception/PLE handler.
1754 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1755 * chance last time (mostly it has become eligible now since we have probably
1756 * yielded to lockholder in last iteration. This is done by toggling
1757 * @dy_eligible each time a VCPU checked for eligibility.)
1759 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1760 * to preempted lock-holder could result in wrong VCPU selection and CPU
1761 * burning. Giving priority for a potential lock-holder increases lock
1764 * Since algorithm is based on heuristics, accessing another VCPU data without
1765 * locking does not harm. It may result in trying to yield to same VCPU, fail
1766 * and continue with next VCPU and so on.
1768 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1772 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1773 (vcpu
->spin_loop
.in_spin_loop
&&
1774 vcpu
->spin_loop
.dy_eligible
);
1776 if (vcpu
->spin_loop
.in_spin_loop
)
1777 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1783 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1785 struct kvm
*kvm
= me
->kvm
;
1786 struct kvm_vcpu
*vcpu
;
1787 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1793 kvm_vcpu_set_in_spin_loop(me
, true);
1795 * We boost the priority of a VCPU that is runnable but not
1796 * currently running, because it got preempted by something
1797 * else and called schedule in __vcpu_run. Hopefully that
1798 * VCPU is holding the lock that we need and will release it.
1799 * We approximate round-robin by starting at the last boosted VCPU.
1801 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1802 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1803 if (!pass
&& i
<= last_boosted_vcpu
) {
1804 i
= last_boosted_vcpu
;
1806 } else if (pass
&& i
> last_boosted_vcpu
)
1808 if (!ACCESS_ONCE(vcpu
->preempted
))
1812 if (waitqueue_active(&vcpu
->wq
))
1814 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1817 yielded
= kvm_vcpu_yield_to(vcpu
);
1819 kvm
->last_boosted_vcpu
= i
;
1821 } else if (yielded
< 0) {
1828 kvm_vcpu_set_in_spin_loop(me
, false);
1830 /* Ensure vcpu is not eligible during next spinloop */
1831 kvm_vcpu_set_dy_eligible(me
, false);
1833 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1835 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1837 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1840 if (vmf
->pgoff
== 0)
1841 page
= virt_to_page(vcpu
->run
);
1843 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1844 page
= virt_to_page(vcpu
->arch
.pio_data
);
1846 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1847 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1848 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1851 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1857 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1858 .fault
= kvm_vcpu_fault
,
1861 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1863 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1867 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1869 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1871 kvm_put_kvm(vcpu
->kvm
);
1875 static struct file_operations kvm_vcpu_fops
= {
1876 .release
= kvm_vcpu_release
,
1877 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1878 #ifdef CONFIG_COMPAT
1879 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1881 .mmap
= kvm_vcpu_mmap
,
1882 .llseek
= noop_llseek
,
1886 * Allocates an inode for the vcpu.
1888 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1890 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
1894 * Creates some virtual cpus. Good luck creating more than one.
1896 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1899 struct kvm_vcpu
*vcpu
, *v
;
1901 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1903 return PTR_ERR(vcpu
);
1905 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1907 r
= kvm_arch_vcpu_setup(vcpu
);
1911 mutex_lock(&kvm
->lock
);
1912 if (!kvm_vcpu_compatible(vcpu
)) {
1914 goto unlock_vcpu_destroy
;
1916 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1918 goto unlock_vcpu_destroy
;
1921 kvm_for_each_vcpu(r
, v
, kvm
)
1922 if (v
->vcpu_id
== id
) {
1924 goto unlock_vcpu_destroy
;
1927 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1929 /* Now it's all set up, let userspace reach it */
1931 r
= create_vcpu_fd(vcpu
);
1934 goto unlock_vcpu_destroy
;
1937 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1939 atomic_inc(&kvm
->online_vcpus
);
1941 mutex_unlock(&kvm
->lock
);
1942 kvm_arch_vcpu_postcreate(vcpu
);
1945 unlock_vcpu_destroy
:
1946 mutex_unlock(&kvm
->lock
);
1948 kvm_arch_vcpu_destroy(vcpu
);
1952 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1955 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1956 vcpu
->sigset_active
= 1;
1957 vcpu
->sigset
= *sigset
;
1959 vcpu
->sigset_active
= 0;
1963 static long kvm_vcpu_ioctl(struct file
*filp
,
1964 unsigned int ioctl
, unsigned long arg
)
1966 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1967 void __user
*argp
= (void __user
*)arg
;
1969 struct kvm_fpu
*fpu
= NULL
;
1970 struct kvm_sregs
*kvm_sregs
= NULL
;
1972 if (vcpu
->kvm
->mm
!= current
->mm
)
1975 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1977 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1978 * so vcpu_load() would break it.
1980 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1981 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1985 r
= vcpu_load(vcpu
);
1993 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1994 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
1996 case KVM_GET_REGS
: {
1997 struct kvm_regs
*kvm_regs
;
2000 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2003 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2007 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2014 case KVM_SET_REGS
: {
2015 struct kvm_regs
*kvm_regs
;
2018 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2019 if (IS_ERR(kvm_regs
)) {
2020 r
= PTR_ERR(kvm_regs
);
2023 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2027 case KVM_GET_SREGS
: {
2028 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2032 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2036 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2041 case KVM_SET_SREGS
: {
2042 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2043 if (IS_ERR(kvm_sregs
)) {
2044 r
= PTR_ERR(kvm_sregs
);
2048 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2051 case KVM_GET_MP_STATE
: {
2052 struct kvm_mp_state mp_state
;
2054 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2058 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2063 case KVM_SET_MP_STATE
: {
2064 struct kvm_mp_state mp_state
;
2067 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2069 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2072 case KVM_TRANSLATE
: {
2073 struct kvm_translation tr
;
2076 if (copy_from_user(&tr
, argp
, sizeof tr
))
2078 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2082 if (copy_to_user(argp
, &tr
, sizeof tr
))
2087 case KVM_SET_GUEST_DEBUG
: {
2088 struct kvm_guest_debug dbg
;
2091 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2093 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2096 case KVM_SET_SIGNAL_MASK
: {
2097 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2098 struct kvm_signal_mask kvm_sigmask
;
2099 sigset_t sigset
, *p
;
2104 if (copy_from_user(&kvm_sigmask
, argp
,
2105 sizeof kvm_sigmask
))
2108 if (kvm_sigmask
.len
!= sizeof sigset
)
2111 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2116 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2120 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2124 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2128 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2134 fpu
= memdup_user(argp
, sizeof(*fpu
));
2140 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2144 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2153 #ifdef CONFIG_COMPAT
2154 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2155 unsigned int ioctl
, unsigned long arg
)
2157 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2158 void __user
*argp
= compat_ptr(arg
);
2161 if (vcpu
->kvm
->mm
!= current
->mm
)
2165 case KVM_SET_SIGNAL_MASK
: {
2166 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2167 struct kvm_signal_mask kvm_sigmask
;
2168 compat_sigset_t csigset
;
2173 if (copy_from_user(&kvm_sigmask
, argp
,
2174 sizeof kvm_sigmask
))
2177 if (kvm_sigmask
.len
!= sizeof csigset
)
2180 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2183 sigset_from_compat(&sigset
, &csigset
);
2184 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2186 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2190 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2198 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2199 int (*accessor
)(struct kvm_device
*dev
,
2200 struct kvm_device_attr
*attr
),
2203 struct kvm_device_attr attr
;
2208 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2211 return accessor(dev
, &attr
);
2214 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2217 struct kvm_device
*dev
= filp
->private_data
;
2220 case KVM_SET_DEVICE_ATTR
:
2221 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2222 case KVM_GET_DEVICE_ATTR
:
2223 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2224 case KVM_HAS_DEVICE_ATTR
:
2225 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2227 if (dev
->ops
->ioctl
)
2228 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2234 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2236 struct kvm_device
*dev
= filp
->private_data
;
2237 struct kvm
*kvm
= dev
->kvm
;
2243 static const struct file_operations kvm_device_fops
= {
2244 .unlocked_ioctl
= kvm_device_ioctl
,
2245 #ifdef CONFIG_COMPAT
2246 .compat_ioctl
= kvm_device_ioctl
,
2248 .release
= kvm_device_release
,
2251 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2253 if (filp
->f_op
!= &kvm_device_fops
)
2256 return filp
->private_data
;
2259 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2260 struct kvm_create_device
*cd
)
2262 struct kvm_device_ops
*ops
= NULL
;
2263 struct kvm_device
*dev
;
2264 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2268 #ifdef CONFIG_KVM_MPIC
2269 case KVM_DEV_TYPE_FSL_MPIC_20
:
2270 case KVM_DEV_TYPE_FSL_MPIC_42
:
2271 ops
= &kvm_mpic_ops
;
2274 #ifdef CONFIG_KVM_XICS
2275 case KVM_DEV_TYPE_XICS
:
2276 ops
= &kvm_xics_ops
;
2279 #ifdef CONFIG_KVM_VFIO
2280 case KVM_DEV_TYPE_VFIO
:
2281 ops
= &kvm_vfio_ops
;
2291 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2298 ret
= ops
->create(dev
, cd
->type
);
2304 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2310 list_add(&dev
->vm_node
, &kvm
->devices
);
2316 static long kvm_vm_ioctl(struct file
*filp
,
2317 unsigned int ioctl
, unsigned long arg
)
2319 struct kvm
*kvm
= filp
->private_data
;
2320 void __user
*argp
= (void __user
*)arg
;
2323 if (kvm
->mm
!= current
->mm
)
2326 case KVM_CREATE_VCPU
:
2327 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2329 case KVM_SET_USER_MEMORY_REGION
: {
2330 struct kvm_userspace_memory_region kvm_userspace_mem
;
2333 if (copy_from_user(&kvm_userspace_mem
, argp
,
2334 sizeof kvm_userspace_mem
))
2337 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2340 case KVM_GET_DIRTY_LOG
: {
2341 struct kvm_dirty_log log
;
2344 if (copy_from_user(&log
, argp
, sizeof log
))
2346 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2349 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2350 case KVM_REGISTER_COALESCED_MMIO
: {
2351 struct kvm_coalesced_mmio_zone zone
;
2353 if (copy_from_user(&zone
, argp
, sizeof zone
))
2355 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2358 case KVM_UNREGISTER_COALESCED_MMIO
: {
2359 struct kvm_coalesced_mmio_zone zone
;
2361 if (copy_from_user(&zone
, argp
, sizeof zone
))
2363 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2368 struct kvm_irqfd data
;
2371 if (copy_from_user(&data
, argp
, sizeof data
))
2373 r
= kvm_irqfd(kvm
, &data
);
2376 case KVM_IOEVENTFD
: {
2377 struct kvm_ioeventfd data
;
2380 if (copy_from_user(&data
, argp
, sizeof data
))
2382 r
= kvm_ioeventfd(kvm
, &data
);
2385 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2386 case KVM_SET_BOOT_CPU_ID
:
2388 mutex_lock(&kvm
->lock
);
2389 if (atomic_read(&kvm
->online_vcpus
) != 0)
2392 kvm
->bsp_vcpu_id
= arg
;
2393 mutex_unlock(&kvm
->lock
);
2396 #ifdef CONFIG_HAVE_KVM_MSI
2397 case KVM_SIGNAL_MSI
: {
2401 if (copy_from_user(&msi
, argp
, sizeof msi
))
2403 r
= kvm_send_userspace_msi(kvm
, &msi
);
2407 #ifdef __KVM_HAVE_IRQ_LINE
2408 case KVM_IRQ_LINE_STATUS
:
2409 case KVM_IRQ_LINE
: {
2410 struct kvm_irq_level irq_event
;
2413 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2416 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2417 ioctl
== KVM_IRQ_LINE_STATUS
);
2422 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2423 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2431 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2432 case KVM_SET_GSI_ROUTING
: {
2433 struct kvm_irq_routing routing
;
2434 struct kvm_irq_routing __user
*urouting
;
2435 struct kvm_irq_routing_entry
*entries
;
2438 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2441 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2446 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2451 if (copy_from_user(entries
, urouting
->entries
,
2452 routing
.nr
* sizeof(*entries
)))
2453 goto out_free_irq_routing
;
2454 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2456 out_free_irq_routing
:
2460 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2461 case KVM_CREATE_DEVICE
: {
2462 struct kvm_create_device cd
;
2465 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2468 r
= kvm_ioctl_create_device(kvm
, &cd
);
2473 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2480 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2482 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2488 #ifdef CONFIG_COMPAT
2489 struct compat_kvm_dirty_log
{
2493 compat_uptr_t dirty_bitmap
; /* one bit per page */
2498 static long kvm_vm_compat_ioctl(struct file
*filp
,
2499 unsigned int ioctl
, unsigned long arg
)
2501 struct kvm
*kvm
= filp
->private_data
;
2504 if (kvm
->mm
!= current
->mm
)
2507 case KVM_GET_DIRTY_LOG
: {
2508 struct compat_kvm_dirty_log compat_log
;
2509 struct kvm_dirty_log log
;
2512 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2513 sizeof(compat_log
)))
2515 log
.slot
= compat_log
.slot
;
2516 log
.padding1
= compat_log
.padding1
;
2517 log
.padding2
= compat_log
.padding2
;
2518 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2520 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2524 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2532 static struct file_operations kvm_vm_fops
= {
2533 .release
= kvm_vm_release
,
2534 .unlocked_ioctl
= kvm_vm_ioctl
,
2535 #ifdef CONFIG_COMPAT
2536 .compat_ioctl
= kvm_vm_compat_ioctl
,
2538 .llseek
= noop_llseek
,
2541 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2546 kvm
= kvm_create_vm(type
);
2548 return PTR_ERR(kvm
);
2549 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2550 r
= kvm_coalesced_mmio_init(kvm
);
2556 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2563 static long kvm_dev_ioctl_check_extension_generic(long arg
)
2566 case KVM_CAP_USER_MEMORY
:
2567 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2568 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2569 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2570 case KVM_CAP_SET_BOOT_CPU_ID
:
2572 case KVM_CAP_INTERNAL_ERROR_DATA
:
2573 #ifdef CONFIG_HAVE_KVM_MSI
2574 case KVM_CAP_SIGNAL_MSI
:
2576 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2577 case KVM_CAP_IRQFD_RESAMPLE
:
2580 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2581 case KVM_CAP_IRQ_ROUTING
:
2582 return KVM_MAX_IRQ_ROUTES
;
2587 return kvm_dev_ioctl_check_extension(arg
);
2590 static long kvm_dev_ioctl(struct file
*filp
,
2591 unsigned int ioctl
, unsigned long arg
)
2596 case KVM_GET_API_VERSION
:
2600 r
= KVM_API_VERSION
;
2603 r
= kvm_dev_ioctl_create_vm(arg
);
2605 case KVM_CHECK_EXTENSION
:
2606 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2608 case KVM_GET_VCPU_MMAP_SIZE
:
2612 r
= PAGE_SIZE
; /* struct kvm_run */
2614 r
+= PAGE_SIZE
; /* pio data page */
2616 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2617 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2620 case KVM_TRACE_ENABLE
:
2621 case KVM_TRACE_PAUSE
:
2622 case KVM_TRACE_DISABLE
:
2626 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2632 static struct file_operations kvm_chardev_ops
= {
2633 .unlocked_ioctl
= kvm_dev_ioctl
,
2634 .compat_ioctl
= kvm_dev_ioctl
,
2635 .llseek
= noop_llseek
,
2638 static struct miscdevice kvm_dev
= {
2644 static void hardware_enable_nolock(void *junk
)
2646 int cpu
= raw_smp_processor_id();
2649 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2652 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2654 r
= kvm_arch_hardware_enable(NULL
);
2657 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2658 atomic_inc(&hardware_enable_failed
);
2659 printk(KERN_INFO
"kvm: enabling virtualization on "
2660 "CPU%d failed\n", cpu
);
2664 static void hardware_enable(void)
2666 raw_spin_lock(&kvm_count_lock
);
2667 if (kvm_usage_count
)
2668 hardware_enable_nolock(NULL
);
2669 raw_spin_unlock(&kvm_count_lock
);
2672 static void hardware_disable_nolock(void *junk
)
2674 int cpu
= raw_smp_processor_id();
2676 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2678 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2679 kvm_arch_hardware_disable(NULL
);
2682 static void hardware_disable(void)
2684 raw_spin_lock(&kvm_count_lock
);
2685 if (kvm_usage_count
)
2686 hardware_disable_nolock(NULL
);
2687 raw_spin_unlock(&kvm_count_lock
);
2690 static void hardware_disable_all_nolock(void)
2692 BUG_ON(!kvm_usage_count
);
2695 if (!kvm_usage_count
)
2696 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2699 static void hardware_disable_all(void)
2701 raw_spin_lock(&kvm_count_lock
);
2702 hardware_disable_all_nolock();
2703 raw_spin_unlock(&kvm_count_lock
);
2706 static int hardware_enable_all(void)
2710 raw_spin_lock(&kvm_count_lock
);
2713 if (kvm_usage_count
== 1) {
2714 atomic_set(&hardware_enable_failed
, 0);
2715 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2717 if (atomic_read(&hardware_enable_failed
)) {
2718 hardware_disable_all_nolock();
2723 raw_spin_unlock(&kvm_count_lock
);
2728 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2733 val
&= ~CPU_TASKS_FROZEN
;
2736 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2741 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2749 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2753 * Some (well, at least mine) BIOSes hang on reboot if
2756 * And Intel TXT required VMX off for all cpu when system shutdown.
2758 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2759 kvm_rebooting
= true;
2760 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2764 static struct notifier_block kvm_reboot_notifier
= {
2765 .notifier_call
= kvm_reboot
,
2769 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2773 for (i
= 0; i
< bus
->dev_count
; i
++) {
2774 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2776 kvm_iodevice_destructor(pos
);
2781 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
2782 const struct kvm_io_range
*r2
)
2784 if (r1
->addr
< r2
->addr
)
2786 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2791 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2793 return kvm_io_bus_cmp(p1
, p2
);
2796 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2797 gpa_t addr
, int len
)
2799 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2805 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2806 kvm_io_bus_sort_cmp
, NULL
);
2811 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2812 gpa_t addr
, int len
)
2814 struct kvm_io_range
*range
, key
;
2817 key
= (struct kvm_io_range
) {
2822 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2823 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2827 off
= range
- bus
->range
;
2829 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
2835 static int __kvm_io_bus_write(struct kvm_io_bus
*bus
,
2836 struct kvm_io_range
*range
, const void *val
)
2840 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2844 while (idx
< bus
->dev_count
&&
2845 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2846 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, range
->addr
,
2855 /* kvm_io_bus_write - called under kvm->slots_lock */
2856 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2857 int len
, const void *val
)
2859 struct kvm_io_bus
*bus
;
2860 struct kvm_io_range range
;
2863 range
= (struct kvm_io_range
) {
2868 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2869 r
= __kvm_io_bus_write(bus
, &range
, val
);
2870 return r
< 0 ? r
: 0;
2873 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2874 int kvm_io_bus_write_cookie(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2875 int len
, const void *val
, long cookie
)
2877 struct kvm_io_bus
*bus
;
2878 struct kvm_io_range range
;
2880 range
= (struct kvm_io_range
) {
2885 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2887 /* First try the device referenced by cookie. */
2888 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
2889 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
2890 if (!kvm_iodevice_write(bus
->range
[cookie
].dev
, addr
, len
,
2895 * cookie contained garbage; fall back to search and return the
2896 * correct cookie value.
2898 return __kvm_io_bus_write(bus
, &range
, val
);
2901 static int __kvm_io_bus_read(struct kvm_io_bus
*bus
, struct kvm_io_range
*range
,
2906 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2910 while (idx
< bus
->dev_count
&&
2911 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2912 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, range
->addr
,
2921 /* kvm_io_bus_read - called under kvm->slots_lock */
2922 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2925 struct kvm_io_bus
*bus
;
2926 struct kvm_io_range range
;
2929 range
= (struct kvm_io_range
) {
2934 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2935 r
= __kvm_io_bus_read(bus
, &range
, val
);
2936 return r
< 0 ? r
: 0;
2939 /* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2940 int kvm_io_bus_read_cookie(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2941 int len
, void *val
, long cookie
)
2943 struct kvm_io_bus
*bus
;
2944 struct kvm_io_range range
;
2946 range
= (struct kvm_io_range
) {
2951 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2953 /* First try the device referenced by cookie. */
2954 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
2955 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
2956 if (!kvm_iodevice_read(bus
->range
[cookie
].dev
, addr
, len
,
2961 * cookie contained garbage; fall back to search and return the
2962 * correct cookie value.
2964 return __kvm_io_bus_read(bus
, &range
, val
);
2967 /* Caller must hold slots_lock. */
2968 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2969 int len
, struct kvm_io_device
*dev
)
2971 struct kvm_io_bus
*new_bus
, *bus
;
2973 bus
= kvm
->buses
[bus_idx
];
2974 /* exclude ioeventfd which is limited by maximum fd */
2975 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
2978 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
2979 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2982 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
2983 sizeof(struct kvm_io_range
)));
2984 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
2985 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2986 synchronize_srcu_expedited(&kvm
->srcu
);
2992 /* Caller must hold slots_lock. */
2993 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2994 struct kvm_io_device
*dev
)
2997 struct kvm_io_bus
*new_bus
, *bus
;
2999 bus
= kvm
->buses
[bus_idx
];
3001 for (i
= 0; i
< bus
->dev_count
; i
++)
3002 if (bus
->range
[i
].dev
== dev
) {
3010 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3011 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3015 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3016 new_bus
->dev_count
--;
3017 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3018 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3020 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3021 synchronize_srcu_expedited(&kvm
->srcu
);
3026 static struct notifier_block kvm_cpu_notifier
= {
3027 .notifier_call
= kvm_cpu_hotplug
,
3030 static int vm_stat_get(void *_offset
, u64
*val
)
3032 unsigned offset
= (long)_offset
;
3036 spin_lock(&kvm_lock
);
3037 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3038 *val
+= *(u32
*)((void *)kvm
+ offset
);
3039 spin_unlock(&kvm_lock
);
3043 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3045 static int vcpu_stat_get(void *_offset
, u64
*val
)
3047 unsigned offset
= (long)_offset
;
3049 struct kvm_vcpu
*vcpu
;
3053 spin_lock(&kvm_lock
);
3054 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3055 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3056 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3058 spin_unlock(&kvm_lock
);
3062 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3064 static const struct file_operations
*stat_fops
[] = {
3065 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3066 [KVM_STAT_VM
] = &vm_stat_fops
,
3069 static int kvm_init_debug(void)
3072 struct kvm_stats_debugfs_item
*p
;
3074 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3075 if (kvm_debugfs_dir
== NULL
)
3078 for (p
= debugfs_entries
; p
->name
; ++p
) {
3079 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3080 (void *)(long)p
->offset
,
3081 stat_fops
[p
->kind
]);
3082 if (p
->dentry
== NULL
)
3089 debugfs_remove_recursive(kvm_debugfs_dir
);
3094 static void kvm_exit_debug(void)
3096 struct kvm_stats_debugfs_item
*p
;
3098 for (p
= debugfs_entries
; p
->name
; ++p
)
3099 debugfs_remove(p
->dentry
);
3100 debugfs_remove(kvm_debugfs_dir
);
3103 static int kvm_suspend(void)
3105 if (kvm_usage_count
)
3106 hardware_disable_nolock(NULL
);
3110 static void kvm_resume(void)
3112 if (kvm_usage_count
) {
3113 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3114 hardware_enable_nolock(NULL
);
3118 static struct syscore_ops kvm_syscore_ops
= {
3119 .suspend
= kvm_suspend
,
3120 .resume
= kvm_resume
,
3124 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3126 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3129 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3131 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3132 if (vcpu
->preempted
)
3133 vcpu
->preempted
= false;
3135 kvm_arch_vcpu_load(vcpu
, cpu
);
3138 static void kvm_sched_out(struct preempt_notifier
*pn
,
3139 struct task_struct
*next
)
3141 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3143 if (current
->state
== TASK_RUNNING
)
3144 vcpu
->preempted
= true;
3145 kvm_arch_vcpu_put(vcpu
);
3148 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3149 struct module
*module
)
3154 r
= kvm_arch_init(opaque
);
3159 * kvm_arch_init makes sure there's at most one caller
3160 * for architectures that support multiple implementations,
3161 * like intel and amd on x86.
3162 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3163 * conflicts in case kvm is already setup for another implementation.
3165 r
= kvm_irqfd_init();
3169 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3174 r
= kvm_arch_hardware_setup();
3178 for_each_online_cpu(cpu
) {
3179 smp_call_function_single(cpu
,
3180 kvm_arch_check_processor_compat
,
3186 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3189 register_reboot_notifier(&kvm_reboot_notifier
);
3191 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3193 vcpu_align
= __alignof__(struct kvm_vcpu
);
3194 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3196 if (!kvm_vcpu_cache
) {
3201 r
= kvm_async_pf_init();
3205 kvm_chardev_ops
.owner
= module
;
3206 kvm_vm_fops
.owner
= module
;
3207 kvm_vcpu_fops
.owner
= module
;
3209 r
= misc_register(&kvm_dev
);
3211 printk(KERN_ERR
"kvm: misc device register failed\n");
3215 register_syscore_ops(&kvm_syscore_ops
);
3217 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3218 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3220 r
= kvm_init_debug();
3222 printk(KERN_ERR
"kvm: create debugfs files failed\n");
3229 unregister_syscore_ops(&kvm_syscore_ops
);
3230 misc_deregister(&kvm_dev
);
3232 kvm_async_pf_deinit();
3234 kmem_cache_destroy(kvm_vcpu_cache
);
3236 unregister_reboot_notifier(&kvm_reboot_notifier
);
3237 unregister_cpu_notifier(&kvm_cpu_notifier
);
3240 kvm_arch_hardware_unsetup();
3242 free_cpumask_var(cpus_hardware_enabled
);
3250 EXPORT_SYMBOL_GPL(kvm_init
);
3255 misc_deregister(&kvm_dev
);
3256 kmem_cache_destroy(kvm_vcpu_cache
);
3257 kvm_async_pf_deinit();
3258 unregister_syscore_ops(&kvm_syscore_ops
);
3259 unregister_reboot_notifier(&kvm_reboot_notifier
);
3260 unregister_cpu_notifier(&kvm_cpu_notifier
);
3261 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3262 kvm_arch_hardware_unsetup();
3265 free_cpumask_var(cpus_hardware_enabled
);
3267 EXPORT_SYMBOL_GPL(kvm_exit
);