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
);
98 static void update_memslots(struct kvm_memslots
*slots
,
99 struct kvm_memory_slot
*new, u64 last_generation
);
101 static void kvm_release_pfn_dirty(pfn_t pfn
);
102 static void mark_page_dirty_in_slot(struct kvm
*kvm
,
103 struct kvm_memory_slot
*memslot
, gfn_t gfn
);
105 __visible
bool kvm_rebooting
;
106 EXPORT_SYMBOL_GPL(kvm_rebooting
);
108 static bool largepages_enabled
= true;
110 bool kvm_is_mmio_pfn(pfn_t pfn
)
113 return PageReserved(pfn_to_page(pfn
));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu
*vcpu
)
125 if (mutex_lock_killable(&vcpu
->mutex
))
127 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
128 /* The thread running this VCPU changed. */
129 struct pid
*oldpid
= vcpu
->pid
;
130 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
131 rcu_assign_pointer(vcpu
->pid
, newpid
);
136 preempt_notifier_register(&vcpu
->preempt_notifier
);
137 kvm_arch_vcpu_load(vcpu
, cpu
);
142 void vcpu_put(struct kvm_vcpu
*vcpu
)
145 kvm_arch_vcpu_put(vcpu
);
146 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
148 mutex_unlock(&vcpu
->mutex
);
151 static void ack_flush(void *_completed
)
155 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
160 struct kvm_vcpu
*vcpu
;
162 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
165 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
166 kvm_make_request(req
, vcpu
);
169 /* Set ->requests bit before we read ->mode */
172 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
173 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
174 cpumask_set_cpu(cpu
, cpus
);
176 if (unlikely(cpus
== NULL
))
177 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
178 else if (!cpumask_empty(cpus
))
179 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
183 free_cpumask_var(cpus
);
187 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
189 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
190 ++kvm
->stat
.remote_tlb_flush
;
191 kvm
->tlbs_dirty
= false;
193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
195 void kvm_reload_remote_mmus(struct kvm
*kvm
)
197 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
200 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
202 make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
205 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
207 make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
210 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
215 mutex_init(&vcpu
->mutex
);
220 init_waitqueue_head(&vcpu
->wq
);
221 kvm_async_pf_vcpu_init(vcpu
);
223 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
228 vcpu
->run
= page_address(page
);
230 kvm_vcpu_set_in_spin_loop(vcpu
, false);
231 kvm_vcpu_set_dy_eligible(vcpu
, false);
232 vcpu
->preempted
= false;
234 r
= kvm_arch_vcpu_init(vcpu
);
240 free_page((unsigned long)vcpu
->run
);
244 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
246 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
249 kvm_arch_vcpu_uninit(vcpu
);
250 free_page((unsigned long)vcpu
->run
);
252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
257 return container_of(mn
, struct kvm
, mmu_notifier
);
260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
261 struct mm_struct
*mm
,
262 unsigned long address
)
264 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
265 int need_tlb_flush
, idx
;
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
285 idx
= srcu_read_lock(&kvm
->srcu
);
286 spin_lock(&kvm
->mmu_lock
);
288 kvm
->mmu_notifier_seq
++;
289 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm
);
294 spin_unlock(&kvm
->mmu_lock
);
295 srcu_read_unlock(&kvm
->srcu
, idx
);
298 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
299 struct mm_struct
*mm
,
300 unsigned long address
,
303 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
306 idx
= srcu_read_lock(&kvm
->srcu
);
307 spin_lock(&kvm
->mmu_lock
);
308 kvm
->mmu_notifier_seq
++;
309 kvm_set_spte_hva(kvm
, address
, pte
);
310 spin_unlock(&kvm
->mmu_lock
);
311 srcu_read_unlock(&kvm
->srcu
, idx
);
314 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
315 struct mm_struct
*mm
,
319 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
320 int need_tlb_flush
= 0, idx
;
322 idx
= srcu_read_lock(&kvm
->srcu
);
323 spin_lock(&kvm
->mmu_lock
);
325 * The count increase must become visible at unlock time as no
326 * spte can be established without taking the mmu_lock and
327 * count is also read inside the mmu_lock critical section.
329 kvm
->mmu_notifier_count
++;
330 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
331 need_tlb_flush
|= kvm
->tlbs_dirty
;
332 /* we've to flush the tlb before the pages can be freed */
334 kvm_flush_remote_tlbs(kvm
);
336 spin_unlock(&kvm
->mmu_lock
);
337 srcu_read_unlock(&kvm
->srcu
, idx
);
340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
341 struct mm_struct
*mm
,
345 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
347 spin_lock(&kvm
->mmu_lock
);
349 * This sequence increase will notify the kvm page fault that
350 * the page that is going to be mapped in the spte could have
353 kvm
->mmu_notifier_seq
++;
356 * The above sequence increase must be visible before the
357 * below count decrease, which is ensured by the smp_wmb above
358 * in conjunction with the smp_rmb in mmu_notifier_retry().
360 kvm
->mmu_notifier_count
--;
361 spin_unlock(&kvm
->mmu_lock
);
363 BUG_ON(kvm
->mmu_notifier_count
< 0);
366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
367 struct mm_struct
*mm
,
368 unsigned long address
)
370 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
373 idx
= srcu_read_lock(&kvm
->srcu
);
374 spin_lock(&kvm
->mmu_lock
);
376 young
= kvm_age_hva(kvm
, address
);
378 kvm_flush_remote_tlbs(kvm
);
380 spin_unlock(&kvm
->mmu_lock
);
381 srcu_read_unlock(&kvm
->srcu
, idx
);
386 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
387 struct mm_struct
*mm
,
388 unsigned long address
)
390 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
393 idx
= srcu_read_lock(&kvm
->srcu
);
394 spin_lock(&kvm
->mmu_lock
);
395 young
= kvm_test_age_hva(kvm
, address
);
396 spin_unlock(&kvm
->mmu_lock
);
397 srcu_read_unlock(&kvm
->srcu
, idx
);
402 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
403 struct mm_struct
*mm
)
405 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
408 idx
= srcu_read_lock(&kvm
->srcu
);
409 kvm_arch_flush_shadow_all(kvm
);
410 srcu_read_unlock(&kvm
->srcu
, idx
);
413 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
414 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
415 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
416 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
417 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
418 .test_young
= kvm_mmu_notifier_test_young
,
419 .change_pte
= kvm_mmu_notifier_change_pte
,
420 .release
= kvm_mmu_notifier_release
,
423 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
425 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
426 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
429 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
431 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
436 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
438 static void kvm_init_memslots_id(struct kvm
*kvm
)
441 struct kvm_memslots
*slots
= kvm
->memslots
;
443 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
444 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
447 static struct kvm
*kvm_create_vm(unsigned long type
)
450 struct kvm
*kvm
= kvm_arch_alloc_vm();
453 return ERR_PTR(-ENOMEM
);
455 r
= kvm_arch_init_vm(kvm
, type
);
457 goto out_err_nodisable
;
459 r
= hardware_enable_all();
461 goto out_err_nodisable
;
463 #ifdef CONFIG_HAVE_KVM_IRQCHIP
464 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
465 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
468 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
471 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
474 kvm_init_memslots_id(kvm
);
475 if (init_srcu_struct(&kvm
->srcu
))
477 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
478 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
484 spin_lock_init(&kvm
->mmu_lock
);
485 kvm
->mm
= current
->mm
;
486 atomic_inc(&kvm
->mm
->mm_count
);
487 kvm_eventfd_init(kvm
);
488 mutex_init(&kvm
->lock
);
489 mutex_init(&kvm
->irq_lock
);
490 mutex_init(&kvm
->slots_lock
);
491 atomic_set(&kvm
->users_count
, 1);
492 INIT_LIST_HEAD(&kvm
->devices
);
494 r
= kvm_init_mmu_notifier(kvm
);
498 spin_lock(&kvm_lock
);
499 list_add(&kvm
->vm_list
, &vm_list
);
500 spin_unlock(&kvm_lock
);
505 cleanup_srcu_struct(&kvm
->srcu
);
507 hardware_disable_all();
509 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
510 kfree(kvm
->buses
[i
]);
511 kfree(kvm
->memslots
);
512 kvm_arch_free_vm(kvm
);
517 * Avoid using vmalloc for a small buffer.
518 * Should not be used when the size is statically known.
520 void *kvm_kvzalloc(unsigned long size
)
522 if (size
> PAGE_SIZE
)
523 return vzalloc(size
);
525 return kzalloc(size
, GFP_KERNEL
);
528 void kvm_kvfree(const void *addr
)
530 if (is_vmalloc_addr(addr
))
536 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
538 if (!memslot
->dirty_bitmap
)
541 kvm_kvfree(memslot
->dirty_bitmap
);
542 memslot
->dirty_bitmap
= NULL
;
546 * Free any memory in @free but not in @dont.
548 static void kvm_free_physmem_slot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
549 struct kvm_memory_slot
*dont
)
551 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
552 kvm_destroy_dirty_bitmap(free
);
554 kvm_arch_free_memslot(kvm
, free
, dont
);
559 static void kvm_free_physmem(struct kvm
*kvm
)
561 struct kvm_memslots
*slots
= kvm
->memslots
;
562 struct kvm_memory_slot
*memslot
;
564 kvm_for_each_memslot(memslot
, slots
)
565 kvm_free_physmem_slot(kvm
, memslot
, NULL
);
567 kfree(kvm
->memslots
);
570 static void kvm_destroy_devices(struct kvm
*kvm
)
572 struct list_head
*node
, *tmp
;
574 list_for_each_safe(node
, tmp
, &kvm
->devices
) {
575 struct kvm_device
*dev
=
576 list_entry(node
, struct kvm_device
, vm_node
);
579 dev
->ops
->destroy(dev
);
583 static void kvm_destroy_vm(struct kvm
*kvm
)
586 struct mm_struct
*mm
= kvm
->mm
;
588 kvm_arch_sync_events(kvm
);
589 spin_lock(&kvm_lock
);
590 list_del(&kvm
->vm_list
);
591 spin_unlock(&kvm_lock
);
592 kvm_free_irq_routing(kvm
);
593 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
594 kvm_io_bus_destroy(kvm
->buses
[i
]);
595 kvm_coalesced_mmio_free(kvm
);
596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
599 kvm_arch_flush_shadow_all(kvm
);
601 kvm_arch_destroy_vm(kvm
);
602 kvm_destroy_devices(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 static void update_memslots(struct kvm_memslots
*slots
,
682 struct kvm_memory_slot
*new,
687 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
688 unsigned long npages
= old
->npages
;
691 if (new->npages
!= npages
)
692 sort_memslots(slots
);
695 slots
->generation
= last_generation
+ 1;
698 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
700 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
702 #ifdef KVM_CAP_READONLY_MEM
703 valid_flags
|= KVM_MEM_READONLY
;
706 if (mem
->flags
& ~valid_flags
)
712 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
713 struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
715 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
717 update_memslots(slots
, new, kvm
->memslots
->generation
);
718 rcu_assign_pointer(kvm
->memslots
, slots
);
719 synchronize_srcu_expedited(&kvm
->srcu
);
721 kvm_arch_memslots_updated(kvm
);
727 * Allocate some memory and give it an address in the guest physical address
730 * Discontiguous memory is allowed, mostly for framebuffers.
732 * Must be called holding mmap_sem for write.
734 int __kvm_set_memory_region(struct kvm
*kvm
,
735 struct kvm_userspace_memory_region
*mem
)
739 unsigned long npages
;
740 struct kvm_memory_slot
*slot
;
741 struct kvm_memory_slot old
, new;
742 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
743 enum kvm_mr_change change
;
745 r
= check_memory_region_flags(mem
);
750 /* General sanity checks */
751 if (mem
->memory_size
& (PAGE_SIZE
- 1))
753 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
755 /* We can read the guest memory with __xxx_user() later on. */
756 if ((mem
->slot
< KVM_USER_MEM_SLOTS
) &&
757 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
758 !access_ok(VERIFY_WRITE
,
759 (void __user
*)(unsigned long)mem
->userspace_addr
,
762 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
764 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
767 slot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
768 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
769 npages
= mem
->memory_size
>> PAGE_SHIFT
;
772 if (npages
> KVM_MEM_MAX_NR_PAGES
)
776 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
781 new.base_gfn
= base_gfn
;
783 new.flags
= mem
->flags
;
788 change
= KVM_MR_CREATE
;
789 else { /* Modify an existing slot. */
790 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
791 (npages
!= old
.npages
) ||
792 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
795 if (base_gfn
!= old
.base_gfn
)
796 change
= KVM_MR_MOVE
;
797 else if (new.flags
!= old
.flags
)
798 change
= KVM_MR_FLAGS_ONLY
;
799 else { /* Nothing to change. */
804 } else if (old
.npages
) {
805 change
= KVM_MR_DELETE
;
806 } else /* Modify a non-existent slot: disallowed. */
809 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
810 /* Check for overlaps */
812 kvm_for_each_memslot(slot
, kvm
->memslots
) {
813 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
814 (slot
->id
== mem
->slot
))
816 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
817 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
822 /* Free page dirty bitmap if unneeded */
823 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
824 new.dirty_bitmap
= NULL
;
827 if (change
== KVM_MR_CREATE
) {
828 new.userspace_addr
= mem
->userspace_addr
;
830 if (kvm_arch_create_memslot(kvm
, &new, npages
))
834 /* Allocate page dirty bitmap if needed */
835 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
836 if (kvm_create_dirty_bitmap(&new) < 0)
840 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
842 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
846 slot
= id_to_memslot(slots
, mem
->slot
);
847 slot
->flags
|= KVM_MEMSLOT_INVALID
;
849 old_memslots
= install_new_memslots(kvm
, slots
, NULL
);
851 /* slot was deleted or moved, clear iommu mapping */
852 kvm_iommu_unmap_pages(kvm
, &old
);
853 /* From this point no new shadow pages pointing to a deleted,
854 * or moved, memslot will be created.
856 * validation of sp->gfn happens in:
857 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
858 * - kvm_is_visible_gfn (mmu_check_roots)
860 kvm_arch_flush_shadow_memslot(kvm
, slot
);
861 slots
= old_memslots
;
864 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
870 * We can re-use the old_memslots from above, the only difference
871 * from the currently installed memslots is the invalid flag. This
872 * will get overwritten by update_memslots anyway.
875 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
881 /* actual memory is freed via old in kvm_free_physmem_slot below */
882 if (change
== KVM_MR_DELETE
) {
883 new.dirty_bitmap
= NULL
;
884 memset(&new.arch
, 0, sizeof(new.arch
));
887 old_memslots
= install_new_memslots(kvm
, slots
, &new);
889 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
891 kvm_free_physmem_slot(kvm
, &old
, &new);
895 * IOMMU mapping: New slots need to be mapped. Old slots need to be
896 * un-mapped and re-mapped if their base changes. Since base change
897 * unmapping is handled above with slot deletion, mapping alone is
898 * needed here. Anything else the iommu might care about for existing
899 * slots (size changes, userspace addr changes and read-only flag
900 * changes) is disallowed above, so any other attribute changes getting
901 * here can be skipped.
903 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
904 r
= kvm_iommu_map_pages(kvm
, &new);
913 kvm_free_physmem_slot(kvm
, &new, &old
);
917 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
919 int kvm_set_memory_region(struct kvm
*kvm
,
920 struct kvm_userspace_memory_region
*mem
)
924 mutex_lock(&kvm
->slots_lock
);
925 r
= __kvm_set_memory_region(kvm
, mem
);
926 mutex_unlock(&kvm
->slots_lock
);
929 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
931 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
932 struct kvm_userspace_memory_region
*mem
)
934 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
936 return kvm_set_memory_region(kvm
, mem
);
939 int kvm_get_dirty_log(struct kvm
*kvm
,
940 struct kvm_dirty_log
*log
, int *is_dirty
)
942 struct kvm_memory_slot
*memslot
;
945 unsigned long any
= 0;
948 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
951 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
953 if (!memslot
->dirty_bitmap
)
956 n
= kvm_dirty_bitmap_bytes(memslot
);
958 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
959 any
= memslot
->dirty_bitmap
[i
];
962 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
972 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
974 bool kvm_largepages_enabled(void)
976 return largepages_enabled
;
979 void kvm_disable_largepages(void)
981 largepages_enabled
= false;
983 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
985 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
987 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
989 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
991 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
993 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
995 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
996 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1001 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1003 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1005 struct vm_area_struct
*vma
;
1006 unsigned long addr
, size
;
1010 addr
= gfn_to_hva(kvm
, gfn
);
1011 if (kvm_is_error_hva(addr
))
1014 down_read(¤t
->mm
->mmap_sem
);
1015 vma
= find_vma(current
->mm
, addr
);
1019 size
= vma_kernel_pagesize(vma
);
1022 up_read(¤t
->mm
->mmap_sem
);
1027 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1029 return slot
->flags
& KVM_MEM_READONLY
;
1032 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1033 gfn_t
*nr_pages
, bool write
)
1035 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1036 return KVM_HVA_ERR_BAD
;
1038 if (memslot_is_readonly(slot
) && write
)
1039 return KVM_HVA_ERR_RO_BAD
;
1042 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1044 return __gfn_to_hva_memslot(slot
, gfn
);
1047 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1050 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1053 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1056 return gfn_to_hva_many(slot
, gfn
, NULL
);
1058 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1060 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1062 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1064 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1067 * If writable is set to false, the hva returned by this function is only
1068 * allowed to be read.
1070 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1072 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1073 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1075 if (!kvm_is_error_hva(hva
) && writable
)
1076 *writable
= !memslot_is_readonly(slot
);
1081 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1083 return __copy_from_user(data
, hva
, len
);
1086 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1088 return __copy_from_user_inatomic(data
, hva
, len
);
1091 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1092 unsigned long start
, int write
, struct page
**page
)
1094 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1097 flags
|= FOLL_WRITE
;
1099 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1102 static inline int check_user_page_hwpoison(unsigned long addr
)
1104 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1106 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1107 flags
, NULL
, NULL
, NULL
);
1108 return rc
== -EHWPOISON
;
1112 * The atomic path to get the writable pfn which will be stored in @pfn,
1113 * true indicates success, otherwise false is returned.
1115 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1116 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1118 struct page
*page
[1];
1121 if (!(async
|| atomic
))
1125 * Fast pin a writable pfn only if it is a write fault request
1126 * or the caller allows to map a writable pfn for a read fault
1129 if (!(write_fault
|| writable
))
1132 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1134 *pfn
= page_to_pfn(page
[0]);
1145 * The slow path to get the pfn of the specified host virtual address,
1146 * 1 indicates success, -errno is returned if error is detected.
1148 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1149 bool *writable
, pfn_t
*pfn
)
1151 struct page
*page
[1];
1157 *writable
= write_fault
;
1160 down_read(¤t
->mm
->mmap_sem
);
1161 npages
= get_user_page_nowait(current
, current
->mm
,
1162 addr
, write_fault
, page
);
1163 up_read(¤t
->mm
->mmap_sem
);
1165 npages
= get_user_pages_fast(addr
, 1, write_fault
,
1170 /* map read fault as writable if possible */
1171 if (unlikely(!write_fault
) && writable
) {
1172 struct page
*wpage
[1];
1174 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1183 *pfn
= page_to_pfn(page
[0]);
1187 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1189 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1192 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1199 * Pin guest page in memory and return its pfn.
1200 * @addr: host virtual address which maps memory to the guest
1201 * @atomic: whether this function can sleep
1202 * @async: whether this function need to wait IO complete if the
1203 * host page is not in the memory
1204 * @write_fault: whether we should get a writable host page
1205 * @writable: whether it allows to map a writable host page for !@write_fault
1207 * The function will map a writable host page for these two cases:
1208 * 1): @write_fault = true
1209 * 2): @write_fault = false && @writable, @writable will tell the caller
1210 * whether the mapping is writable.
1212 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1213 bool write_fault
, bool *writable
)
1215 struct vm_area_struct
*vma
;
1219 /* we can do it either atomically or asynchronously, not both */
1220 BUG_ON(atomic
&& async
);
1222 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1226 return KVM_PFN_ERR_FAULT
;
1228 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1232 down_read(¤t
->mm
->mmap_sem
);
1233 if (npages
== -EHWPOISON
||
1234 (!async
&& check_user_page_hwpoison(addr
))) {
1235 pfn
= KVM_PFN_ERR_HWPOISON
;
1239 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1242 pfn
= KVM_PFN_ERR_FAULT
;
1243 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1244 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1246 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1248 if (async
&& vma_is_valid(vma
, write_fault
))
1250 pfn
= KVM_PFN_ERR_FAULT
;
1253 up_read(¤t
->mm
->mmap_sem
);
1258 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1259 bool *async
, bool write_fault
, bool *writable
)
1261 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1263 if (addr
== KVM_HVA_ERR_RO_BAD
)
1264 return KVM_PFN_ERR_RO_FAULT
;
1266 if (kvm_is_error_hva(addr
))
1267 return KVM_PFN_NOSLOT
;
1269 /* Do not map writable pfn in the readonly memslot. */
1270 if (writable
&& memslot_is_readonly(slot
)) {
1275 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1279 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1280 bool write_fault
, bool *writable
)
1282 struct kvm_memory_slot
*slot
;
1287 slot
= gfn_to_memslot(kvm
, gfn
);
1289 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1293 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1295 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1297 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1299 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1300 bool write_fault
, bool *writable
)
1302 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1304 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1306 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1308 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1310 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1312 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1315 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1317 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1319 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1321 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1324 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1326 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1328 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1330 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1336 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1337 if (kvm_is_error_hva(addr
))
1340 if (entry
< nr_pages
)
1343 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1345 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1347 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1349 if (is_error_noslot_pfn(pfn
))
1350 return KVM_ERR_PTR_BAD_PAGE
;
1352 if (kvm_is_mmio_pfn(pfn
)) {
1354 return KVM_ERR_PTR_BAD_PAGE
;
1357 return pfn_to_page(pfn
);
1360 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1364 pfn
= gfn_to_pfn(kvm
, gfn
);
1366 return kvm_pfn_to_page(pfn
);
1369 EXPORT_SYMBOL_GPL(gfn_to_page
);
1371 void kvm_release_page_clean(struct page
*page
)
1373 WARN_ON(is_error_page(page
));
1375 kvm_release_pfn_clean(page_to_pfn(page
));
1377 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1379 void kvm_release_pfn_clean(pfn_t pfn
)
1381 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1382 put_page(pfn_to_page(pfn
));
1384 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1386 void kvm_release_page_dirty(struct page
*page
)
1388 WARN_ON(is_error_page(page
));
1390 kvm_release_pfn_dirty(page_to_pfn(page
));
1392 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1394 static void kvm_release_pfn_dirty(pfn_t pfn
)
1396 kvm_set_pfn_dirty(pfn
);
1397 kvm_release_pfn_clean(pfn
);
1400 void kvm_set_pfn_dirty(pfn_t pfn
)
1402 if (!kvm_is_mmio_pfn(pfn
)) {
1403 struct page
*page
= pfn_to_page(pfn
);
1404 if (!PageReserved(page
))
1408 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1410 void kvm_set_pfn_accessed(pfn_t pfn
)
1412 if (!kvm_is_mmio_pfn(pfn
))
1413 mark_page_accessed(pfn_to_page(pfn
));
1415 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1417 void kvm_get_pfn(pfn_t pfn
)
1419 if (!kvm_is_mmio_pfn(pfn
))
1420 get_page(pfn_to_page(pfn
));
1422 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1424 static int next_segment(unsigned long len
, int offset
)
1426 if (len
> PAGE_SIZE
- offset
)
1427 return PAGE_SIZE
- offset
;
1432 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1438 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1439 if (kvm_is_error_hva(addr
))
1441 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1446 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1448 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1450 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1452 int offset
= offset_in_page(gpa
);
1455 while ((seg
= next_segment(len
, offset
)) != 0) {
1456 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1466 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1468 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1473 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1474 int offset
= offset_in_page(gpa
);
1476 addr
= gfn_to_hva_prot(kvm
, gfn
, NULL
);
1477 if (kvm_is_error_hva(addr
))
1479 pagefault_disable();
1480 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1486 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1488 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1489 int offset
, int len
)
1494 addr
= gfn_to_hva(kvm
, gfn
);
1495 if (kvm_is_error_hva(addr
))
1497 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1500 mark_page_dirty(kvm
, gfn
);
1503 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1505 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1508 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1510 int offset
= offset_in_page(gpa
);
1513 while ((seg
= next_segment(len
, offset
)) != 0) {
1514 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1525 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1526 gpa_t gpa
, unsigned long len
)
1528 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1529 int offset
= offset_in_page(gpa
);
1530 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1531 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1532 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1533 gfn_t nr_pages_avail
;
1536 ghc
->generation
= slots
->generation
;
1538 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1539 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, &nr_pages_avail
);
1540 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_avail
>= nr_pages_needed
) {
1544 * If the requested region crosses two memslots, we still
1545 * verify that the entire region is valid here.
1547 while (start_gfn
<= end_gfn
) {
1548 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1549 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1551 if (kvm_is_error_hva(ghc
->hva
))
1553 start_gfn
+= nr_pages_avail
;
1555 /* Use the slow path for cross page reads and writes. */
1556 ghc
->memslot
= NULL
;
1560 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1562 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1563 void *data
, unsigned long len
)
1565 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1568 BUG_ON(len
> ghc
->len
);
1570 if (slots
->generation
!= ghc
->generation
)
1571 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1573 if (unlikely(!ghc
->memslot
))
1574 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1576 if (kvm_is_error_hva(ghc
->hva
))
1579 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1582 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1586 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1588 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1589 void *data
, unsigned long len
)
1591 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1594 BUG_ON(len
> ghc
->len
);
1596 if (slots
->generation
!= ghc
->generation
)
1597 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1599 if (unlikely(!ghc
->memslot
))
1600 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1602 if (kvm_is_error_hva(ghc
->hva
))
1605 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1611 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1613 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1615 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1617 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1619 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1621 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1623 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1625 int offset
= offset_in_page(gpa
);
1628 while ((seg
= next_segment(len
, offset
)) != 0) {
1629 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1638 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1640 static void mark_page_dirty_in_slot(struct kvm
*kvm
,
1641 struct kvm_memory_slot
*memslot
,
1644 if (memslot
&& memslot
->dirty_bitmap
) {
1645 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1647 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1651 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1653 struct kvm_memory_slot
*memslot
;
1655 memslot
= gfn_to_memslot(kvm
, gfn
);
1656 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1658 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1661 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1663 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1668 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1670 if (kvm_arch_vcpu_runnable(vcpu
)) {
1671 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1674 if (kvm_cpu_has_pending_timer(vcpu
))
1676 if (signal_pending(current
))
1682 finish_wait(&vcpu
->wq
, &wait
);
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
1688 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1690 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1693 int cpu
= vcpu
->cpu
;
1694 wait_queue_head_t
*wqp
;
1696 wqp
= kvm_arch_vcpu_wq(vcpu
);
1697 if (waitqueue_active(wqp
)) {
1698 wake_up_interruptible(wqp
);
1699 ++vcpu
->stat
.halt_wakeup
;
1703 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1704 if (kvm_arch_vcpu_should_kick(vcpu
))
1705 smp_send_reschedule(cpu
);
1708 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1709 #endif /* !CONFIG_S390 */
1711 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1714 struct task_struct
*task
= NULL
;
1718 pid
= rcu_dereference(target
->pid
);
1720 task
= get_pid_task(target
->pid
, PIDTYPE_PID
);
1724 if (task
->flags
& PF_VCPU
) {
1725 put_task_struct(task
);
1728 ret
= yield_to(task
, 1);
1729 put_task_struct(task
);
1733 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1736 * Helper that checks whether a VCPU is eligible for directed yield.
1737 * Most eligible candidate to yield is decided by following heuristics:
1739 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1740 * (preempted lock holder), indicated by @in_spin_loop.
1741 * Set at the beiginning and cleared at the end of interception/PLE handler.
1743 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1744 * chance last time (mostly it has become eligible now since we have probably
1745 * yielded to lockholder in last iteration. This is done by toggling
1746 * @dy_eligible each time a VCPU checked for eligibility.)
1748 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1749 * to preempted lock-holder could result in wrong VCPU selection and CPU
1750 * burning. Giving priority for a potential lock-holder increases lock
1753 * Since algorithm is based on heuristics, accessing another VCPU data without
1754 * locking does not harm. It may result in trying to yield to same VCPU, fail
1755 * and continue with next VCPU and so on.
1757 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1759 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1762 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1763 (vcpu
->spin_loop
.in_spin_loop
&&
1764 vcpu
->spin_loop
.dy_eligible
);
1766 if (vcpu
->spin_loop
.in_spin_loop
)
1767 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1775 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1777 struct kvm
*kvm
= me
->kvm
;
1778 struct kvm_vcpu
*vcpu
;
1779 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1785 kvm_vcpu_set_in_spin_loop(me
, true);
1787 * We boost the priority of a VCPU that is runnable but not
1788 * currently running, because it got preempted by something
1789 * else and called schedule in __vcpu_run. Hopefully that
1790 * VCPU is holding the lock that we need and will release it.
1791 * We approximate round-robin by starting at the last boosted VCPU.
1793 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1794 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1795 if (!pass
&& i
<= last_boosted_vcpu
) {
1796 i
= last_boosted_vcpu
;
1798 } else if (pass
&& i
> last_boosted_vcpu
)
1800 if (!ACCESS_ONCE(vcpu
->preempted
))
1804 if (waitqueue_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
1806 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1809 yielded
= kvm_vcpu_yield_to(vcpu
);
1811 kvm
->last_boosted_vcpu
= i
;
1813 } else if (yielded
< 0) {
1820 kvm_vcpu_set_in_spin_loop(me
, false);
1822 /* Ensure vcpu is not eligible during next spinloop */
1823 kvm_vcpu_set_dy_eligible(me
, false);
1825 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1827 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1829 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1832 if (vmf
->pgoff
== 0)
1833 page
= virt_to_page(vcpu
->run
);
1835 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1836 page
= virt_to_page(vcpu
->arch
.pio_data
);
1838 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1839 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1840 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1843 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1849 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1850 .fault
= kvm_vcpu_fault
,
1853 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1855 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1859 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1861 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1863 kvm_put_kvm(vcpu
->kvm
);
1867 static struct file_operations kvm_vcpu_fops
= {
1868 .release
= kvm_vcpu_release
,
1869 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1870 #ifdef CONFIG_COMPAT
1871 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1873 .mmap
= kvm_vcpu_mmap
,
1874 .llseek
= noop_llseek
,
1878 * Allocates an inode for the vcpu.
1880 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1882 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
1886 * Creates some virtual cpus. Good luck creating more than one.
1888 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1891 struct kvm_vcpu
*vcpu
, *v
;
1893 if (id
>= KVM_MAX_VCPUS
)
1896 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1898 return PTR_ERR(vcpu
);
1900 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1902 r
= kvm_arch_vcpu_setup(vcpu
);
1906 mutex_lock(&kvm
->lock
);
1907 if (!kvm_vcpu_compatible(vcpu
)) {
1909 goto unlock_vcpu_destroy
;
1911 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1913 goto unlock_vcpu_destroy
;
1916 kvm_for_each_vcpu(r
, v
, kvm
)
1917 if (v
->vcpu_id
== id
) {
1919 goto unlock_vcpu_destroy
;
1922 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1924 /* Now it's all set up, let userspace reach it */
1926 r
= create_vcpu_fd(vcpu
);
1929 goto unlock_vcpu_destroy
;
1932 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1934 atomic_inc(&kvm
->online_vcpus
);
1936 mutex_unlock(&kvm
->lock
);
1937 kvm_arch_vcpu_postcreate(vcpu
);
1940 unlock_vcpu_destroy
:
1941 mutex_unlock(&kvm
->lock
);
1943 kvm_arch_vcpu_destroy(vcpu
);
1947 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1950 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1951 vcpu
->sigset_active
= 1;
1952 vcpu
->sigset
= *sigset
;
1954 vcpu
->sigset_active
= 0;
1958 static long kvm_vcpu_ioctl(struct file
*filp
,
1959 unsigned int ioctl
, unsigned long arg
)
1961 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1962 void __user
*argp
= (void __user
*)arg
;
1964 struct kvm_fpu
*fpu
= NULL
;
1965 struct kvm_sregs
*kvm_sregs
= NULL
;
1967 if (vcpu
->kvm
->mm
!= current
->mm
)
1970 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1972 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1973 * so vcpu_load() would break it.
1975 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1976 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1980 r
= vcpu_load(vcpu
);
1988 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1989 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
1991 case KVM_GET_REGS
: {
1992 struct kvm_regs
*kvm_regs
;
1995 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1998 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2002 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2009 case KVM_SET_REGS
: {
2010 struct kvm_regs
*kvm_regs
;
2013 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2014 if (IS_ERR(kvm_regs
)) {
2015 r
= PTR_ERR(kvm_regs
);
2018 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2022 case KVM_GET_SREGS
: {
2023 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2027 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2031 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2036 case KVM_SET_SREGS
: {
2037 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2038 if (IS_ERR(kvm_sregs
)) {
2039 r
= PTR_ERR(kvm_sregs
);
2043 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2046 case KVM_GET_MP_STATE
: {
2047 struct kvm_mp_state mp_state
;
2049 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2053 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2058 case KVM_SET_MP_STATE
: {
2059 struct kvm_mp_state mp_state
;
2062 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2064 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2067 case KVM_TRANSLATE
: {
2068 struct kvm_translation tr
;
2071 if (copy_from_user(&tr
, argp
, sizeof tr
))
2073 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2077 if (copy_to_user(argp
, &tr
, sizeof tr
))
2082 case KVM_SET_GUEST_DEBUG
: {
2083 struct kvm_guest_debug dbg
;
2086 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2088 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2091 case KVM_SET_SIGNAL_MASK
: {
2092 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2093 struct kvm_signal_mask kvm_sigmask
;
2094 sigset_t sigset
, *p
;
2099 if (copy_from_user(&kvm_sigmask
, argp
,
2100 sizeof kvm_sigmask
))
2103 if (kvm_sigmask
.len
!= sizeof sigset
)
2106 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2111 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2115 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2119 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2123 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2129 fpu
= memdup_user(argp
, sizeof(*fpu
));
2135 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2139 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2148 #ifdef CONFIG_COMPAT
2149 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2150 unsigned int ioctl
, unsigned long arg
)
2152 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2153 void __user
*argp
= compat_ptr(arg
);
2156 if (vcpu
->kvm
->mm
!= current
->mm
)
2160 case KVM_SET_SIGNAL_MASK
: {
2161 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2162 struct kvm_signal_mask kvm_sigmask
;
2163 compat_sigset_t csigset
;
2168 if (copy_from_user(&kvm_sigmask
, argp
,
2169 sizeof kvm_sigmask
))
2172 if (kvm_sigmask
.len
!= sizeof csigset
)
2175 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2178 sigset_from_compat(&sigset
, &csigset
);
2179 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2181 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2185 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2193 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2194 int (*accessor
)(struct kvm_device
*dev
,
2195 struct kvm_device_attr
*attr
),
2198 struct kvm_device_attr attr
;
2203 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2206 return accessor(dev
, &attr
);
2209 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2212 struct kvm_device
*dev
= filp
->private_data
;
2215 case KVM_SET_DEVICE_ATTR
:
2216 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2217 case KVM_GET_DEVICE_ATTR
:
2218 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2219 case KVM_HAS_DEVICE_ATTR
:
2220 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2222 if (dev
->ops
->ioctl
)
2223 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2229 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2231 struct kvm_device
*dev
= filp
->private_data
;
2232 struct kvm
*kvm
= dev
->kvm
;
2238 static const struct file_operations kvm_device_fops
= {
2239 .unlocked_ioctl
= kvm_device_ioctl
,
2240 #ifdef CONFIG_COMPAT
2241 .compat_ioctl
= kvm_device_ioctl
,
2243 .release
= kvm_device_release
,
2246 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2248 if (filp
->f_op
!= &kvm_device_fops
)
2251 return filp
->private_data
;
2254 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2255 struct kvm_create_device
*cd
)
2257 struct kvm_device_ops
*ops
= NULL
;
2258 struct kvm_device
*dev
;
2259 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2263 #ifdef CONFIG_KVM_MPIC
2264 case KVM_DEV_TYPE_FSL_MPIC_20
:
2265 case KVM_DEV_TYPE_FSL_MPIC_42
:
2266 ops
= &kvm_mpic_ops
;
2269 #ifdef CONFIG_KVM_XICS
2270 case KVM_DEV_TYPE_XICS
:
2271 ops
= &kvm_xics_ops
;
2274 #ifdef CONFIG_KVM_VFIO
2275 case KVM_DEV_TYPE_VFIO
:
2276 ops
= &kvm_vfio_ops
;
2279 #ifdef CONFIG_KVM_ARM_VGIC
2280 case KVM_DEV_TYPE_ARM_VGIC_V2
:
2281 ops
= &kvm_arm_vgic_v2_ops
;
2285 case KVM_DEV_TYPE_FLIC
:
2286 ops
= &kvm_flic_ops
;
2296 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2303 ret
= ops
->create(dev
, cd
->type
);
2309 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2315 list_add(&dev
->vm_node
, &kvm
->devices
);
2321 static long kvm_vm_ioctl(struct file
*filp
,
2322 unsigned int ioctl
, unsigned long arg
)
2324 struct kvm
*kvm
= filp
->private_data
;
2325 void __user
*argp
= (void __user
*)arg
;
2328 if (kvm
->mm
!= current
->mm
)
2331 case KVM_CREATE_VCPU
:
2332 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2334 case KVM_SET_USER_MEMORY_REGION
: {
2335 struct kvm_userspace_memory_region kvm_userspace_mem
;
2338 if (copy_from_user(&kvm_userspace_mem
, argp
,
2339 sizeof kvm_userspace_mem
))
2342 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2345 case KVM_GET_DIRTY_LOG
: {
2346 struct kvm_dirty_log log
;
2349 if (copy_from_user(&log
, argp
, sizeof log
))
2351 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2354 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2355 case KVM_REGISTER_COALESCED_MMIO
: {
2356 struct kvm_coalesced_mmio_zone zone
;
2358 if (copy_from_user(&zone
, argp
, sizeof zone
))
2360 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2363 case KVM_UNREGISTER_COALESCED_MMIO
: {
2364 struct kvm_coalesced_mmio_zone zone
;
2366 if (copy_from_user(&zone
, argp
, sizeof zone
))
2368 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2373 struct kvm_irqfd data
;
2376 if (copy_from_user(&data
, argp
, sizeof data
))
2378 r
= kvm_irqfd(kvm
, &data
);
2381 case KVM_IOEVENTFD
: {
2382 struct kvm_ioeventfd data
;
2385 if (copy_from_user(&data
, argp
, sizeof data
))
2387 r
= kvm_ioeventfd(kvm
, &data
);
2390 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2391 case KVM_SET_BOOT_CPU_ID
:
2393 mutex_lock(&kvm
->lock
);
2394 if (atomic_read(&kvm
->online_vcpus
) != 0)
2397 kvm
->bsp_vcpu_id
= arg
;
2398 mutex_unlock(&kvm
->lock
);
2401 #ifdef CONFIG_HAVE_KVM_MSI
2402 case KVM_SIGNAL_MSI
: {
2406 if (copy_from_user(&msi
, argp
, sizeof msi
))
2408 r
= kvm_send_userspace_msi(kvm
, &msi
);
2412 #ifdef __KVM_HAVE_IRQ_LINE
2413 case KVM_IRQ_LINE_STATUS
:
2414 case KVM_IRQ_LINE
: {
2415 struct kvm_irq_level irq_event
;
2418 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2421 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2422 ioctl
== KVM_IRQ_LINE_STATUS
);
2427 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2428 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2436 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2437 case KVM_SET_GSI_ROUTING
: {
2438 struct kvm_irq_routing routing
;
2439 struct kvm_irq_routing __user
*urouting
;
2440 struct kvm_irq_routing_entry
*entries
;
2443 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2446 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2451 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2456 if (copy_from_user(entries
, urouting
->entries
,
2457 routing
.nr
* sizeof(*entries
)))
2458 goto out_free_irq_routing
;
2459 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2461 out_free_irq_routing
:
2465 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2466 case KVM_CREATE_DEVICE
: {
2467 struct kvm_create_device cd
;
2470 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2473 r
= kvm_ioctl_create_device(kvm
, &cd
);
2478 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2485 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2487 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2493 #ifdef CONFIG_COMPAT
2494 struct compat_kvm_dirty_log
{
2498 compat_uptr_t dirty_bitmap
; /* one bit per page */
2503 static long kvm_vm_compat_ioctl(struct file
*filp
,
2504 unsigned int ioctl
, unsigned long arg
)
2506 struct kvm
*kvm
= filp
->private_data
;
2509 if (kvm
->mm
!= current
->mm
)
2512 case KVM_GET_DIRTY_LOG
: {
2513 struct compat_kvm_dirty_log compat_log
;
2514 struct kvm_dirty_log log
;
2517 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2518 sizeof(compat_log
)))
2520 log
.slot
= compat_log
.slot
;
2521 log
.padding1
= compat_log
.padding1
;
2522 log
.padding2
= compat_log
.padding2
;
2523 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2525 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2529 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2537 static struct file_operations kvm_vm_fops
= {
2538 .release
= kvm_vm_release
,
2539 .unlocked_ioctl
= kvm_vm_ioctl
,
2540 #ifdef CONFIG_COMPAT
2541 .compat_ioctl
= kvm_vm_compat_ioctl
,
2543 .llseek
= noop_llseek
,
2546 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2551 kvm
= kvm_create_vm(type
);
2553 return PTR_ERR(kvm
);
2554 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2555 r
= kvm_coalesced_mmio_init(kvm
);
2561 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2568 static long kvm_dev_ioctl_check_extension_generic(long arg
)
2571 case KVM_CAP_USER_MEMORY
:
2572 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2573 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2574 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2575 case KVM_CAP_SET_BOOT_CPU_ID
:
2577 case KVM_CAP_INTERNAL_ERROR_DATA
:
2578 #ifdef CONFIG_HAVE_KVM_MSI
2579 case KVM_CAP_SIGNAL_MSI
:
2581 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2582 case KVM_CAP_IRQFD_RESAMPLE
:
2585 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2586 case KVM_CAP_IRQ_ROUTING
:
2587 return KVM_MAX_IRQ_ROUTES
;
2592 return kvm_dev_ioctl_check_extension(arg
);
2595 static long kvm_dev_ioctl(struct file
*filp
,
2596 unsigned int ioctl
, unsigned long arg
)
2601 case KVM_GET_API_VERSION
:
2605 r
= KVM_API_VERSION
;
2608 r
= kvm_dev_ioctl_create_vm(arg
);
2610 case KVM_CHECK_EXTENSION
:
2611 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2613 case KVM_GET_VCPU_MMAP_SIZE
:
2617 r
= PAGE_SIZE
; /* struct kvm_run */
2619 r
+= PAGE_SIZE
; /* pio data page */
2621 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2622 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2625 case KVM_TRACE_ENABLE
:
2626 case KVM_TRACE_PAUSE
:
2627 case KVM_TRACE_DISABLE
:
2631 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2637 static struct file_operations kvm_chardev_ops
= {
2638 .unlocked_ioctl
= kvm_dev_ioctl
,
2639 .compat_ioctl
= kvm_dev_ioctl
,
2640 .llseek
= noop_llseek
,
2643 static struct miscdevice kvm_dev
= {
2649 static void hardware_enable_nolock(void *junk
)
2651 int cpu
= raw_smp_processor_id();
2654 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2657 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2659 r
= kvm_arch_hardware_enable(NULL
);
2662 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2663 atomic_inc(&hardware_enable_failed
);
2664 printk(KERN_INFO
"kvm: enabling virtualization on "
2665 "CPU%d failed\n", cpu
);
2669 static void hardware_enable(void)
2671 raw_spin_lock(&kvm_count_lock
);
2672 if (kvm_usage_count
)
2673 hardware_enable_nolock(NULL
);
2674 raw_spin_unlock(&kvm_count_lock
);
2677 static void hardware_disable_nolock(void *junk
)
2679 int cpu
= raw_smp_processor_id();
2681 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2683 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2684 kvm_arch_hardware_disable(NULL
);
2687 static void hardware_disable(void)
2689 raw_spin_lock(&kvm_count_lock
);
2690 if (kvm_usage_count
)
2691 hardware_disable_nolock(NULL
);
2692 raw_spin_unlock(&kvm_count_lock
);
2695 static void hardware_disable_all_nolock(void)
2697 BUG_ON(!kvm_usage_count
);
2700 if (!kvm_usage_count
)
2701 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2704 static void hardware_disable_all(void)
2706 raw_spin_lock(&kvm_count_lock
);
2707 hardware_disable_all_nolock();
2708 raw_spin_unlock(&kvm_count_lock
);
2711 static int hardware_enable_all(void)
2715 raw_spin_lock(&kvm_count_lock
);
2718 if (kvm_usage_count
== 1) {
2719 atomic_set(&hardware_enable_failed
, 0);
2720 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2722 if (atomic_read(&hardware_enable_failed
)) {
2723 hardware_disable_all_nolock();
2728 raw_spin_unlock(&kvm_count_lock
);
2733 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2738 val
&= ~CPU_TASKS_FROZEN
;
2741 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2746 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2754 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2758 * Some (well, at least mine) BIOSes hang on reboot if
2761 * And Intel TXT required VMX off for all cpu when system shutdown.
2763 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2764 kvm_rebooting
= true;
2765 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2769 static struct notifier_block kvm_reboot_notifier
= {
2770 .notifier_call
= kvm_reboot
,
2774 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2778 for (i
= 0; i
< bus
->dev_count
; i
++) {
2779 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2781 kvm_iodevice_destructor(pos
);
2786 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
2787 const struct kvm_io_range
*r2
)
2789 if (r1
->addr
< r2
->addr
)
2791 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2796 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2798 return kvm_io_bus_cmp(p1
, p2
);
2801 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2802 gpa_t addr
, int len
)
2804 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2810 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2811 kvm_io_bus_sort_cmp
, NULL
);
2816 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2817 gpa_t addr
, int len
)
2819 struct kvm_io_range
*range
, key
;
2822 key
= (struct kvm_io_range
) {
2827 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2828 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2832 off
= range
- bus
->range
;
2834 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
2840 static int __kvm_io_bus_write(struct kvm_io_bus
*bus
,
2841 struct kvm_io_range
*range
, const void *val
)
2845 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2849 while (idx
< bus
->dev_count
&&
2850 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2851 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, range
->addr
,
2860 /* kvm_io_bus_write - called under kvm->slots_lock */
2861 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2862 int len
, const void *val
)
2864 struct kvm_io_bus
*bus
;
2865 struct kvm_io_range range
;
2868 range
= (struct kvm_io_range
) {
2873 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2874 r
= __kvm_io_bus_write(bus
, &range
, val
);
2875 return r
< 0 ? r
: 0;
2878 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2879 int kvm_io_bus_write_cookie(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2880 int len
, const void *val
, long cookie
)
2882 struct kvm_io_bus
*bus
;
2883 struct kvm_io_range range
;
2885 range
= (struct kvm_io_range
) {
2890 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2892 /* First try the device referenced by cookie. */
2893 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
2894 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
2895 if (!kvm_iodevice_write(bus
->range
[cookie
].dev
, addr
, len
,
2900 * cookie contained garbage; fall back to search and return the
2901 * correct cookie value.
2903 return __kvm_io_bus_write(bus
, &range
, val
);
2906 static int __kvm_io_bus_read(struct kvm_io_bus
*bus
, struct kvm_io_range
*range
,
2911 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
2915 while (idx
< bus
->dev_count
&&
2916 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
2917 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, range
->addr
,
2926 /* kvm_io_bus_read - called under kvm->slots_lock */
2927 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2930 struct kvm_io_bus
*bus
;
2931 struct kvm_io_range range
;
2934 range
= (struct kvm_io_range
) {
2939 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2940 r
= __kvm_io_bus_read(bus
, &range
, val
);
2941 return r
< 0 ? r
: 0;
2945 /* Caller must hold slots_lock. */
2946 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2947 int len
, struct kvm_io_device
*dev
)
2949 struct kvm_io_bus
*new_bus
, *bus
;
2951 bus
= kvm
->buses
[bus_idx
];
2952 /* exclude ioeventfd which is limited by maximum fd */
2953 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
2956 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
2957 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2960 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
2961 sizeof(struct kvm_io_range
)));
2962 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
2963 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2964 synchronize_srcu_expedited(&kvm
->srcu
);
2970 /* Caller must hold slots_lock. */
2971 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2972 struct kvm_io_device
*dev
)
2975 struct kvm_io_bus
*new_bus
, *bus
;
2977 bus
= kvm
->buses
[bus_idx
];
2979 for (i
= 0; i
< bus
->dev_count
; i
++)
2980 if (bus
->range
[i
].dev
== dev
) {
2988 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
2989 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2993 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
2994 new_bus
->dev_count
--;
2995 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
2996 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
2998 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2999 synchronize_srcu_expedited(&kvm
->srcu
);
3004 static struct notifier_block kvm_cpu_notifier
= {
3005 .notifier_call
= kvm_cpu_hotplug
,
3008 static int vm_stat_get(void *_offset
, u64
*val
)
3010 unsigned offset
= (long)_offset
;
3014 spin_lock(&kvm_lock
);
3015 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3016 *val
+= *(u32
*)((void *)kvm
+ offset
);
3017 spin_unlock(&kvm_lock
);
3021 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3023 static int vcpu_stat_get(void *_offset
, u64
*val
)
3025 unsigned offset
= (long)_offset
;
3027 struct kvm_vcpu
*vcpu
;
3031 spin_lock(&kvm_lock
);
3032 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3033 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3034 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3036 spin_unlock(&kvm_lock
);
3040 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3042 static const struct file_operations
*stat_fops
[] = {
3043 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3044 [KVM_STAT_VM
] = &vm_stat_fops
,
3047 static int kvm_init_debug(void)
3050 struct kvm_stats_debugfs_item
*p
;
3052 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3053 if (kvm_debugfs_dir
== NULL
)
3056 for (p
= debugfs_entries
; p
->name
; ++p
) {
3057 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3058 (void *)(long)p
->offset
,
3059 stat_fops
[p
->kind
]);
3060 if (p
->dentry
== NULL
)
3067 debugfs_remove_recursive(kvm_debugfs_dir
);
3072 static void kvm_exit_debug(void)
3074 struct kvm_stats_debugfs_item
*p
;
3076 for (p
= debugfs_entries
; p
->name
; ++p
)
3077 debugfs_remove(p
->dentry
);
3078 debugfs_remove(kvm_debugfs_dir
);
3081 static int kvm_suspend(void)
3083 if (kvm_usage_count
)
3084 hardware_disable_nolock(NULL
);
3088 static void kvm_resume(void)
3090 if (kvm_usage_count
) {
3091 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3092 hardware_enable_nolock(NULL
);
3096 static struct syscore_ops kvm_syscore_ops
= {
3097 .suspend
= kvm_suspend
,
3098 .resume
= kvm_resume
,
3102 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3104 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3107 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3109 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3110 if (vcpu
->preempted
)
3111 vcpu
->preempted
= false;
3113 kvm_arch_vcpu_load(vcpu
, cpu
);
3116 static void kvm_sched_out(struct preempt_notifier
*pn
,
3117 struct task_struct
*next
)
3119 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3121 if (current
->state
== TASK_RUNNING
)
3122 vcpu
->preempted
= true;
3123 kvm_arch_vcpu_put(vcpu
);
3126 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3127 struct module
*module
)
3132 r
= kvm_arch_init(opaque
);
3137 * kvm_arch_init makes sure there's at most one caller
3138 * for architectures that support multiple implementations,
3139 * like intel and amd on x86.
3140 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3141 * conflicts in case kvm is already setup for another implementation.
3143 r
= kvm_irqfd_init();
3147 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3152 r
= kvm_arch_hardware_setup();
3156 for_each_online_cpu(cpu
) {
3157 smp_call_function_single(cpu
,
3158 kvm_arch_check_processor_compat
,
3164 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3167 register_reboot_notifier(&kvm_reboot_notifier
);
3169 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3171 vcpu_align
= __alignof__(struct kvm_vcpu
);
3172 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3174 if (!kvm_vcpu_cache
) {
3179 r
= kvm_async_pf_init();
3183 kvm_chardev_ops
.owner
= module
;
3184 kvm_vm_fops
.owner
= module
;
3185 kvm_vcpu_fops
.owner
= module
;
3187 r
= misc_register(&kvm_dev
);
3189 printk(KERN_ERR
"kvm: misc device register failed\n");
3193 register_syscore_ops(&kvm_syscore_ops
);
3195 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3196 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3198 r
= kvm_init_debug();
3200 printk(KERN_ERR
"kvm: create debugfs files failed\n");
3207 unregister_syscore_ops(&kvm_syscore_ops
);
3208 misc_deregister(&kvm_dev
);
3210 kvm_async_pf_deinit();
3212 kmem_cache_destroy(kvm_vcpu_cache
);
3214 unregister_reboot_notifier(&kvm_reboot_notifier
);
3215 unregister_cpu_notifier(&kvm_cpu_notifier
);
3218 kvm_arch_hardware_unsetup();
3220 free_cpumask_var(cpus_hardware_enabled
);
3228 EXPORT_SYMBOL_GPL(kvm_init
);
3233 misc_deregister(&kvm_dev
);
3234 kmem_cache_destroy(kvm_vcpu_cache
);
3235 kvm_async_pf_deinit();
3236 unregister_syscore_ops(&kvm_syscore_ops
);
3237 unregister_reboot_notifier(&kvm_reboot_notifier
);
3238 unregister_cpu_notifier(&kvm_cpu_notifier
);
3239 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3240 kvm_arch_hardware_unsetup();
3243 free_cpumask_var(cpus_hardware_enabled
);
3245 EXPORT_SYMBOL_GPL(kvm_exit
);