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.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
69 static unsigned int halt_poll_ns
;
70 module_param(halt_poll_ns
, uint
, S_IRUGO
| S_IWUSR
);
75 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
78 DEFINE_SPINLOCK(kvm_lock
);
79 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
82 static cpumask_var_t cpus_hardware_enabled
;
83 static int kvm_usage_count
;
84 static atomic_t hardware_enable_failed
;
86 struct kmem_cache
*kvm_vcpu_cache
;
87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
89 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
91 struct dentry
*kvm_debugfs_dir
;
92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
94 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
96 #ifdef CONFIG_KVM_COMPAT
97 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
100 static int hardware_enable_all(void);
101 static void hardware_disable_all(void);
103 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
105 static void kvm_release_pfn_dirty(pfn_t pfn
);
106 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
108 __visible
bool kvm_rebooting
;
109 EXPORT_SYMBOL_GPL(kvm_rebooting
);
111 static bool largepages_enabled
= true;
113 bool kvm_is_reserved_pfn(pfn_t pfn
)
116 return PageReserved(pfn_to_page(pfn
));
122 * Switches to specified vcpu, until a matching vcpu_put()
124 int vcpu_load(struct kvm_vcpu
*vcpu
)
128 if (mutex_lock_killable(&vcpu
->mutex
))
131 preempt_notifier_register(&vcpu
->preempt_notifier
);
132 kvm_arch_vcpu_load(vcpu
, cpu
);
137 void vcpu_put(struct kvm_vcpu
*vcpu
)
140 kvm_arch_vcpu_put(vcpu
);
141 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
143 mutex_unlock(&vcpu
->mutex
);
146 static void ack_flush(void *_completed
)
150 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
155 struct kvm_vcpu
*vcpu
;
157 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
160 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
161 kvm_make_request(req
, vcpu
);
164 /* Set ->requests bit before we read ->mode */
167 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
168 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
169 cpumask_set_cpu(cpu
, cpus
);
171 if (unlikely(cpus
== NULL
))
172 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
173 else if (!cpumask_empty(cpus
))
174 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
178 free_cpumask_var(cpus
);
182 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
183 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
185 long dirty_count
= kvm
->tlbs_dirty
;
188 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
189 ++kvm
->stat
.remote_tlb_flush
;
190 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
192 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
195 void kvm_reload_remote_mmus(struct kvm
*kvm
)
197 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
200 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
202 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
205 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
207 kvm_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
);
296 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
298 srcu_read_unlock(&kvm
->srcu
, idx
);
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
302 struct mm_struct
*mm
,
303 unsigned long address
,
306 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
309 idx
= srcu_read_lock(&kvm
->srcu
);
310 spin_lock(&kvm
->mmu_lock
);
311 kvm
->mmu_notifier_seq
++;
312 kvm_set_spte_hva(kvm
, address
, pte
);
313 spin_unlock(&kvm
->mmu_lock
);
314 srcu_read_unlock(&kvm
->srcu
, idx
);
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
318 struct mm_struct
*mm
,
322 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
323 int need_tlb_flush
= 0, idx
;
325 idx
= srcu_read_lock(&kvm
->srcu
);
326 spin_lock(&kvm
->mmu_lock
);
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
332 kvm
->mmu_notifier_count
++;
333 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
334 need_tlb_flush
|= kvm
->tlbs_dirty
;
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm
);
339 spin_unlock(&kvm
->mmu_lock
);
340 srcu_read_unlock(&kvm
->srcu
, idx
);
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
344 struct mm_struct
*mm
,
348 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
350 spin_lock(&kvm
->mmu_lock
);
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
356 kvm
->mmu_notifier_seq
++;
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 kvm
->mmu_notifier_count
--;
364 spin_unlock(&kvm
->mmu_lock
);
366 BUG_ON(kvm
->mmu_notifier_count
< 0);
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
370 struct mm_struct
*mm
,
374 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
377 idx
= srcu_read_lock(&kvm
->srcu
);
378 spin_lock(&kvm
->mmu_lock
);
380 young
= kvm_age_hva(kvm
, start
, end
);
382 kvm_flush_remote_tlbs(kvm
);
384 spin_unlock(&kvm
->mmu_lock
);
385 srcu_read_unlock(&kvm
->srcu
, idx
);
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
391 struct mm_struct
*mm
,
392 unsigned long address
)
394 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
397 idx
= srcu_read_lock(&kvm
->srcu
);
398 spin_lock(&kvm
->mmu_lock
);
399 young
= kvm_test_age_hva(kvm
, address
);
400 spin_unlock(&kvm
->mmu_lock
);
401 srcu_read_unlock(&kvm
->srcu
, idx
);
406 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
407 struct mm_struct
*mm
)
409 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
412 idx
= srcu_read_lock(&kvm
->srcu
);
413 kvm_arch_flush_shadow_all(kvm
);
414 srcu_read_unlock(&kvm
->srcu
, idx
);
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
418 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
419 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
420 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
421 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
422 .test_young
= kvm_mmu_notifier_test_young
,
423 .change_pte
= kvm_mmu_notifier_change_pte
,
424 .release
= kvm_mmu_notifier_release
,
427 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
429 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
430 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
433 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442 static struct kvm_memslots
*kvm_alloc_memslots(void)
445 struct kvm_memslots
*slots
;
447 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
452 * Init kvm generation close to the maximum to easily test the
453 * code of handling generation number wrap-around.
455 slots
->generation
= -150;
456 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
457 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
462 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
464 if (!memslot
->dirty_bitmap
)
467 kvfree(memslot
->dirty_bitmap
);
468 memslot
->dirty_bitmap
= NULL
;
472 * Free any memory in @free but not in @dont.
474 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
475 struct kvm_memory_slot
*dont
)
477 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
478 kvm_destroy_dirty_bitmap(free
);
480 kvm_arch_free_memslot(kvm
, free
, dont
);
485 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
487 struct kvm_memory_slot
*memslot
;
492 kvm_for_each_memslot(memslot
, slots
)
493 kvm_free_memslot(kvm
, memslot
, NULL
);
498 static struct kvm
*kvm_create_vm(unsigned long type
)
501 struct kvm
*kvm
= kvm_arch_alloc_vm();
504 return ERR_PTR(-ENOMEM
);
506 r
= kvm_arch_init_vm(kvm
, type
);
508 goto out_err_no_disable
;
510 r
= hardware_enable_all();
512 goto out_err_no_disable
;
514 #ifdef CONFIG_HAVE_KVM_IRQFD
515 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
518 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
521 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
522 kvm
->memslots
[i
] = kvm_alloc_memslots();
523 if (!kvm
->memslots
[i
])
524 goto out_err_no_srcu
;
527 if (init_srcu_struct(&kvm
->srcu
))
528 goto out_err_no_srcu
;
529 if (init_srcu_struct(&kvm
->irq_srcu
))
530 goto out_err_no_irq_srcu
;
531 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
532 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
538 spin_lock_init(&kvm
->mmu_lock
);
539 kvm
->mm
= current
->mm
;
540 atomic_inc(&kvm
->mm
->mm_count
);
541 kvm_eventfd_init(kvm
);
542 mutex_init(&kvm
->lock
);
543 mutex_init(&kvm
->irq_lock
);
544 mutex_init(&kvm
->slots_lock
);
545 atomic_set(&kvm
->users_count
, 1);
546 INIT_LIST_HEAD(&kvm
->devices
);
548 r
= kvm_init_mmu_notifier(kvm
);
552 spin_lock(&kvm_lock
);
553 list_add(&kvm
->vm_list
, &vm_list
);
554 spin_unlock(&kvm_lock
);
556 preempt_notifier_inc();
561 cleanup_srcu_struct(&kvm
->irq_srcu
);
563 cleanup_srcu_struct(&kvm
->srcu
);
565 hardware_disable_all();
567 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
568 kfree(kvm
->buses
[i
]);
569 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
570 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
571 kvm_arch_free_vm(kvm
);
576 * Avoid using vmalloc for a small buffer.
577 * Should not be used when the size is statically known.
579 void *kvm_kvzalloc(unsigned long size
)
581 if (size
> PAGE_SIZE
)
582 return vzalloc(size
);
584 return kzalloc(size
, GFP_KERNEL
);
587 static void kvm_destroy_devices(struct kvm
*kvm
)
589 struct list_head
*node
, *tmp
;
591 list_for_each_safe(node
, tmp
, &kvm
->devices
) {
592 struct kvm_device
*dev
=
593 list_entry(node
, struct kvm_device
, vm_node
);
596 dev
->ops
->destroy(dev
);
600 static void kvm_destroy_vm(struct kvm
*kvm
)
603 struct mm_struct
*mm
= kvm
->mm
;
605 kvm_arch_sync_events(kvm
);
606 spin_lock(&kvm_lock
);
607 list_del(&kvm
->vm_list
);
608 spin_unlock(&kvm_lock
);
609 kvm_free_irq_routing(kvm
);
610 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
611 kvm_io_bus_destroy(kvm
->buses
[i
]);
612 kvm_coalesced_mmio_free(kvm
);
613 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
614 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
616 kvm_arch_flush_shadow_all(kvm
);
618 kvm_arch_destroy_vm(kvm
);
619 kvm_destroy_devices(kvm
);
620 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
621 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
622 cleanup_srcu_struct(&kvm
->irq_srcu
);
623 cleanup_srcu_struct(&kvm
->srcu
);
624 kvm_arch_free_vm(kvm
);
625 preempt_notifier_dec();
626 hardware_disable_all();
630 void kvm_get_kvm(struct kvm
*kvm
)
632 atomic_inc(&kvm
->users_count
);
634 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
636 void kvm_put_kvm(struct kvm
*kvm
)
638 if (atomic_dec_and_test(&kvm
->users_count
))
641 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
644 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
646 struct kvm
*kvm
= filp
->private_data
;
648 kvm_irqfd_release(kvm
);
655 * Allocation size is twice as large as the actual dirty bitmap size.
656 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
658 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
660 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
662 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
663 if (!memslot
->dirty_bitmap
)
670 * Insert memslot and re-sort memslots based on their GFN,
671 * so binary search could be used to lookup GFN.
672 * Sorting algorithm takes advantage of having initially
673 * sorted array and known changed memslot position.
675 static void update_memslots(struct kvm_memslots
*slots
,
676 struct kvm_memory_slot
*new)
679 int i
= slots
->id_to_index
[id
];
680 struct kvm_memory_slot
*mslots
= slots
->memslots
;
682 WARN_ON(mslots
[i
].id
!= id
);
684 WARN_ON(!mslots
[i
].npages
);
685 if (mslots
[i
].npages
)
688 if (!mslots
[i
].npages
)
692 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
693 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
694 if (!mslots
[i
+ 1].npages
)
696 mslots
[i
] = mslots
[i
+ 1];
697 slots
->id_to_index
[mslots
[i
].id
] = i
;
702 * The ">=" is needed when creating a slot with base_gfn == 0,
703 * so that it moves before all those with base_gfn == npages == 0.
705 * On the other hand, if new->npages is zero, the above loop has
706 * already left i pointing to the beginning of the empty part of
707 * mslots, and the ">=" would move the hole backwards in this
708 * case---which is wrong. So skip the loop when deleting a slot.
712 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
713 mslots
[i
] = mslots
[i
- 1];
714 slots
->id_to_index
[mslots
[i
].id
] = i
;
718 WARN_ON_ONCE(i
!= slots
->used_slots
);
721 slots
->id_to_index
[mslots
[i
].id
] = i
;
724 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
726 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
728 #ifdef __KVM_HAVE_READONLY_MEM
729 valid_flags
|= KVM_MEM_READONLY
;
732 if (mem
->flags
& ~valid_flags
)
738 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
739 int as_id
, struct kvm_memslots
*slots
)
741 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
744 * Set the low bit in the generation, which disables SPTE caching
745 * until the end of synchronize_srcu_expedited.
747 WARN_ON(old_memslots
->generation
& 1);
748 slots
->generation
= old_memslots
->generation
+ 1;
750 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
751 synchronize_srcu_expedited(&kvm
->srcu
);
754 * Increment the new memslot generation a second time. This prevents
755 * vm exits that race with memslot updates from caching a memslot
756 * generation that will (potentially) be valid forever.
760 kvm_arch_memslots_updated(kvm
, slots
);
766 * Allocate some memory and give it an address in the guest physical address
769 * Discontiguous memory is allowed, mostly for framebuffers.
771 * Must be called holding kvm->slots_lock for write.
773 int __kvm_set_memory_region(struct kvm
*kvm
,
774 const struct kvm_userspace_memory_region
*mem
)
778 unsigned long npages
;
779 struct kvm_memory_slot
*slot
;
780 struct kvm_memory_slot old
, new;
781 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
783 enum kvm_mr_change change
;
785 r
= check_memory_region_flags(mem
);
790 as_id
= mem
->slot
>> 16;
793 /* General sanity checks */
794 if (mem
->memory_size
& (PAGE_SIZE
- 1))
796 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
798 /* We can read the guest memory with __xxx_user() later on. */
799 if ((id
< KVM_USER_MEM_SLOTS
) &&
800 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
801 !access_ok(VERIFY_WRITE
,
802 (void __user
*)(unsigned long)mem
->userspace_addr
,
805 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
807 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
810 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
811 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
812 npages
= mem
->memory_size
>> PAGE_SHIFT
;
814 if (npages
> KVM_MEM_MAX_NR_PAGES
)
820 new.base_gfn
= base_gfn
;
822 new.flags
= mem
->flags
;
826 change
= KVM_MR_CREATE
;
827 else { /* Modify an existing slot. */
828 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
829 (npages
!= old
.npages
) ||
830 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
833 if (base_gfn
!= old
.base_gfn
)
834 change
= KVM_MR_MOVE
;
835 else if (new.flags
!= old
.flags
)
836 change
= KVM_MR_FLAGS_ONLY
;
837 else { /* Nothing to change. */
846 change
= KVM_MR_DELETE
;
851 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
852 /* Check for overlaps */
854 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
855 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
858 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
859 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
864 /* Free page dirty bitmap if unneeded */
865 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
866 new.dirty_bitmap
= NULL
;
869 if (change
== KVM_MR_CREATE
) {
870 new.userspace_addr
= mem
->userspace_addr
;
872 if (kvm_arch_create_memslot(kvm
, &new, npages
))
876 /* Allocate page dirty bitmap if needed */
877 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
878 if (kvm_create_dirty_bitmap(&new) < 0)
882 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
885 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
887 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
888 slot
= id_to_memslot(slots
, id
);
889 slot
->flags
|= KVM_MEMSLOT_INVALID
;
891 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
893 /* slot was deleted or moved, clear iommu mapping */
894 kvm_iommu_unmap_pages(kvm
, &old
);
895 /* From this point no new shadow pages pointing to a deleted,
896 * or moved, memslot will be created.
898 * validation of sp->gfn happens in:
899 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
900 * - kvm_is_visible_gfn (mmu_check_roots)
902 kvm_arch_flush_shadow_memslot(kvm
, slot
);
905 * We can re-use the old_memslots from above, the only difference
906 * from the currently installed memslots is the invalid flag. This
907 * will get overwritten by update_memslots anyway.
909 slots
= old_memslots
;
912 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
916 /* actual memory is freed via old in kvm_free_memslot below */
917 if (change
== KVM_MR_DELETE
) {
918 new.dirty_bitmap
= NULL
;
919 memset(&new.arch
, 0, sizeof(new.arch
));
922 update_memslots(slots
, &new);
923 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
925 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
927 kvm_free_memslot(kvm
, &old
, &new);
928 kvfree(old_memslots
);
931 * IOMMU mapping: New slots need to be mapped. Old slots need to be
932 * un-mapped and re-mapped if their base changes. Since base change
933 * unmapping is handled above with slot deletion, mapping alone is
934 * needed here. Anything else the iommu might care about for existing
935 * slots (size changes, userspace addr changes and read-only flag
936 * changes) is disallowed above, so any other attribute changes getting
937 * here can be skipped.
939 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
940 r
= kvm_iommu_map_pages(kvm
, &new);
949 kvm_free_memslot(kvm
, &new, &old
);
953 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
955 int kvm_set_memory_region(struct kvm
*kvm
,
956 const struct kvm_userspace_memory_region
*mem
)
960 mutex_lock(&kvm
->slots_lock
);
961 r
= __kvm_set_memory_region(kvm
, mem
);
962 mutex_unlock(&kvm
->slots_lock
);
965 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
967 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
968 struct kvm_userspace_memory_region
*mem
)
970 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
973 return kvm_set_memory_region(kvm
, mem
);
976 int kvm_get_dirty_log(struct kvm
*kvm
,
977 struct kvm_dirty_log
*log
, int *is_dirty
)
979 struct kvm_memslots
*slots
;
980 struct kvm_memory_slot
*memslot
;
983 unsigned long any
= 0;
986 as_id
= log
->slot
>> 16;
988 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
991 slots
= __kvm_memslots(kvm
, as_id
);
992 memslot
= id_to_memslot(slots
, id
);
994 if (!memslot
->dirty_bitmap
)
997 n
= kvm_dirty_bitmap_bytes(memslot
);
999 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1000 any
= memslot
->dirty_bitmap
[i
];
1003 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1013 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1015 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1017 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1018 * are dirty write protect them for next write.
1019 * @kvm: pointer to kvm instance
1020 * @log: slot id and address to which we copy the log
1021 * @is_dirty: flag set if any page is dirty
1023 * We need to keep it in mind that VCPU threads can write to the bitmap
1024 * concurrently. So, to avoid losing track of dirty pages we keep the
1027 * 1. Take a snapshot of the bit and clear it if needed.
1028 * 2. Write protect the corresponding page.
1029 * 3. Copy the snapshot to the userspace.
1030 * 4. Upon return caller flushes TLB's if needed.
1032 * Between 2 and 4, the guest may write to the page using the remaining TLB
1033 * entry. This is not a problem because the page is reported dirty using
1034 * the snapshot taken before and step 4 ensures that writes done after
1035 * exiting to userspace will be logged for the next call.
1038 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1039 struct kvm_dirty_log
*log
, bool *is_dirty
)
1041 struct kvm_memslots
*slots
;
1042 struct kvm_memory_slot
*memslot
;
1043 int r
, i
, as_id
, id
;
1045 unsigned long *dirty_bitmap
;
1046 unsigned long *dirty_bitmap_buffer
;
1049 as_id
= log
->slot
>> 16;
1050 id
= (u16
)log
->slot
;
1051 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1054 slots
= __kvm_memslots(kvm
, as_id
);
1055 memslot
= id_to_memslot(slots
, id
);
1057 dirty_bitmap
= memslot
->dirty_bitmap
;
1062 n
= kvm_dirty_bitmap_bytes(memslot
);
1064 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1065 memset(dirty_bitmap_buffer
, 0, n
);
1067 spin_lock(&kvm
->mmu_lock
);
1069 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1073 if (!dirty_bitmap
[i
])
1078 mask
= xchg(&dirty_bitmap
[i
], 0);
1079 dirty_bitmap_buffer
[i
] = mask
;
1082 offset
= i
* BITS_PER_LONG
;
1083 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1088 spin_unlock(&kvm
->mmu_lock
);
1091 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1098 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1101 bool kvm_largepages_enabled(void)
1103 return largepages_enabled
;
1106 void kvm_disable_largepages(void)
1108 largepages_enabled
= false;
1110 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1112 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1114 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1116 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1118 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1120 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1123 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1125 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1127 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1128 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1133 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1135 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1137 struct vm_area_struct
*vma
;
1138 unsigned long addr
, size
;
1142 addr
= gfn_to_hva(kvm
, gfn
);
1143 if (kvm_is_error_hva(addr
))
1146 down_read(¤t
->mm
->mmap_sem
);
1147 vma
= find_vma(current
->mm
, addr
);
1151 size
= vma_kernel_pagesize(vma
);
1154 up_read(¤t
->mm
->mmap_sem
);
1159 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1161 return slot
->flags
& KVM_MEM_READONLY
;
1164 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1165 gfn_t
*nr_pages
, bool write
)
1167 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1168 return KVM_HVA_ERR_BAD
;
1170 if (memslot_is_readonly(slot
) && write
)
1171 return KVM_HVA_ERR_RO_BAD
;
1174 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1176 return __gfn_to_hva_memslot(slot
, gfn
);
1179 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1182 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1185 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1188 return gfn_to_hva_many(slot
, gfn
, NULL
);
1190 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1192 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1194 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1196 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1198 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1200 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1202 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1205 * If writable is set to false, the hva returned by this function is only
1206 * allowed to be read.
1208 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1209 gfn_t gfn
, bool *writable
)
1211 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1213 if (!kvm_is_error_hva(hva
) && writable
)
1214 *writable
= !memslot_is_readonly(slot
);
1219 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1221 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1223 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1226 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1228 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1230 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1233 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1234 unsigned long start
, int write
, struct page
**page
)
1236 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1239 flags
|= FOLL_WRITE
;
1241 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1244 static inline int check_user_page_hwpoison(unsigned long addr
)
1246 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1248 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1249 flags
, NULL
, NULL
, NULL
);
1250 return rc
== -EHWPOISON
;
1254 * The atomic path to get the writable pfn which will be stored in @pfn,
1255 * true indicates success, otherwise false is returned.
1257 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1258 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1260 struct page
*page
[1];
1263 if (!(async
|| atomic
))
1267 * Fast pin a writable pfn only if it is a write fault request
1268 * or the caller allows to map a writable pfn for a read fault
1271 if (!(write_fault
|| writable
))
1274 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1276 *pfn
= page_to_pfn(page
[0]);
1287 * The slow path to get the pfn of the specified host virtual address,
1288 * 1 indicates success, -errno is returned if error is detected.
1290 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1291 bool *writable
, pfn_t
*pfn
)
1293 struct page
*page
[1];
1299 *writable
= write_fault
;
1302 down_read(¤t
->mm
->mmap_sem
);
1303 npages
= get_user_page_nowait(current
, current
->mm
,
1304 addr
, write_fault
, page
);
1305 up_read(¤t
->mm
->mmap_sem
);
1307 npages
= __get_user_pages_unlocked(current
, current
->mm
, addr
, 1,
1308 write_fault
, 0, page
,
1309 FOLL_TOUCH
|FOLL_HWPOISON
);
1313 /* map read fault as writable if possible */
1314 if (unlikely(!write_fault
) && writable
) {
1315 struct page
*wpage
[1];
1317 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1326 *pfn
= page_to_pfn(page
[0]);
1330 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1332 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1335 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1342 * Pin guest page in memory and return its pfn.
1343 * @addr: host virtual address which maps memory to the guest
1344 * @atomic: whether this function can sleep
1345 * @async: whether this function need to wait IO complete if the
1346 * host page is not in the memory
1347 * @write_fault: whether we should get a writable host page
1348 * @writable: whether it allows to map a writable host page for !@write_fault
1350 * The function will map a writable host page for these two cases:
1351 * 1): @write_fault = true
1352 * 2): @write_fault = false && @writable, @writable will tell the caller
1353 * whether the mapping is writable.
1355 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1356 bool write_fault
, bool *writable
)
1358 struct vm_area_struct
*vma
;
1362 /* we can do it either atomically or asynchronously, not both */
1363 BUG_ON(atomic
&& async
);
1365 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1369 return KVM_PFN_ERR_FAULT
;
1371 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1375 down_read(¤t
->mm
->mmap_sem
);
1376 if (npages
== -EHWPOISON
||
1377 (!async
&& check_user_page_hwpoison(addr
))) {
1378 pfn
= KVM_PFN_ERR_HWPOISON
;
1382 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1385 pfn
= KVM_PFN_ERR_FAULT
;
1386 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1387 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1389 BUG_ON(!kvm_is_reserved_pfn(pfn
));
1391 if (async
&& vma_is_valid(vma
, write_fault
))
1393 pfn
= KVM_PFN_ERR_FAULT
;
1396 up_read(¤t
->mm
->mmap_sem
);
1400 pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1401 bool *async
, bool write_fault
, bool *writable
)
1403 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1405 if (addr
== KVM_HVA_ERR_RO_BAD
)
1406 return KVM_PFN_ERR_RO_FAULT
;
1408 if (kvm_is_error_hva(addr
))
1409 return KVM_PFN_NOSLOT
;
1411 /* Do not map writable pfn in the readonly memslot. */
1412 if (writable
&& memslot_is_readonly(slot
)) {
1417 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1420 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1422 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1425 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1426 write_fault
, writable
);
1428 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1430 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1432 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1434 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1436 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1438 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1440 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1442 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1444 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1446 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1448 pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1450 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1452 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1454 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1456 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1458 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1460 pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1462 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1464 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1466 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1467 struct page
**pages
, int nr_pages
)
1472 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1473 if (kvm_is_error_hva(addr
))
1476 if (entry
< nr_pages
)
1479 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1481 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1483 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1485 if (is_error_noslot_pfn(pfn
))
1486 return KVM_ERR_PTR_BAD_PAGE
;
1488 if (kvm_is_reserved_pfn(pfn
)) {
1490 return KVM_ERR_PTR_BAD_PAGE
;
1493 return pfn_to_page(pfn
);
1496 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1500 pfn
= gfn_to_pfn(kvm
, gfn
);
1502 return kvm_pfn_to_page(pfn
);
1504 EXPORT_SYMBOL_GPL(gfn_to_page
);
1506 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1510 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1512 return kvm_pfn_to_page(pfn
);
1514 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1516 void kvm_release_page_clean(struct page
*page
)
1518 WARN_ON(is_error_page(page
));
1520 kvm_release_pfn_clean(page_to_pfn(page
));
1522 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1524 void kvm_release_pfn_clean(pfn_t pfn
)
1526 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1527 put_page(pfn_to_page(pfn
));
1529 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1531 void kvm_release_page_dirty(struct page
*page
)
1533 WARN_ON(is_error_page(page
));
1535 kvm_release_pfn_dirty(page_to_pfn(page
));
1537 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1539 static void kvm_release_pfn_dirty(pfn_t pfn
)
1541 kvm_set_pfn_dirty(pfn
);
1542 kvm_release_pfn_clean(pfn
);
1545 void kvm_set_pfn_dirty(pfn_t pfn
)
1547 if (!kvm_is_reserved_pfn(pfn
)) {
1548 struct page
*page
= pfn_to_page(pfn
);
1550 if (!PageReserved(page
))
1554 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1556 void kvm_set_pfn_accessed(pfn_t pfn
)
1558 if (!kvm_is_reserved_pfn(pfn
))
1559 mark_page_accessed(pfn_to_page(pfn
));
1561 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1563 void kvm_get_pfn(pfn_t pfn
)
1565 if (!kvm_is_reserved_pfn(pfn
))
1566 get_page(pfn_to_page(pfn
));
1568 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1570 static int next_segment(unsigned long len
, int offset
)
1572 if (len
> PAGE_SIZE
- offset
)
1573 return PAGE_SIZE
- offset
;
1578 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1579 void *data
, int offset
, int len
)
1584 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1585 if (kvm_is_error_hva(addr
))
1587 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1593 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1596 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1598 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1600 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1602 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1603 int offset
, int len
)
1605 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1607 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1609 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1611 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1613 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1615 int offset
= offset_in_page(gpa
);
1618 while ((seg
= next_segment(len
, offset
)) != 0) {
1619 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1629 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1631 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1633 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1635 int offset
= offset_in_page(gpa
);
1638 while ((seg
= next_segment(len
, offset
)) != 0) {
1639 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1649 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1651 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1652 void *data
, int offset
, unsigned long len
)
1657 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1658 if (kvm_is_error_hva(addr
))
1660 pagefault_disable();
1661 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1668 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1671 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1672 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1673 int offset
= offset_in_page(gpa
);
1675 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1677 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1679 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1680 void *data
, unsigned long len
)
1682 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1683 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1684 int offset
= offset_in_page(gpa
);
1686 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1688 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1690 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1691 const void *data
, int offset
, int len
)
1696 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1697 if (kvm_is_error_hva(addr
))
1699 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1702 mark_page_dirty_in_slot(memslot
, gfn
);
1706 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1707 const void *data
, int offset
, int len
)
1709 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1711 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1713 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1715 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1716 const void *data
, int offset
, int len
)
1718 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1720 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1722 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1724 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1727 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1729 int offset
= offset_in_page(gpa
);
1732 while ((seg
= next_segment(len
, offset
)) != 0) {
1733 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1743 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1745 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1748 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1750 int offset
= offset_in_page(gpa
);
1753 while ((seg
= next_segment(len
, offset
)) != 0) {
1754 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1764 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1766 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1767 gpa_t gpa
, unsigned long len
)
1769 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1770 int offset
= offset_in_page(gpa
);
1771 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1772 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1773 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1774 gfn_t nr_pages_avail
;
1777 ghc
->generation
= slots
->generation
;
1779 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1780 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1781 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1785 * If the requested region crosses two memslots, we still
1786 * verify that the entire region is valid here.
1788 while (start_gfn
<= end_gfn
) {
1789 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1790 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1792 if (kvm_is_error_hva(ghc
->hva
))
1794 start_gfn
+= nr_pages_avail
;
1796 /* Use the slow path for cross page reads and writes. */
1797 ghc
->memslot
= NULL
;
1801 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1803 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1804 void *data
, unsigned long len
)
1806 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1809 BUG_ON(len
> ghc
->len
);
1811 if (slots
->generation
!= ghc
->generation
)
1812 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1814 if (unlikely(!ghc
->memslot
))
1815 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1817 if (kvm_is_error_hva(ghc
->hva
))
1820 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1823 mark_page_dirty_in_slot(ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1827 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1829 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1830 void *data
, unsigned long len
)
1832 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1835 BUG_ON(len
> ghc
->len
);
1837 if (slots
->generation
!= ghc
->generation
)
1838 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1840 if (unlikely(!ghc
->memslot
))
1841 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1843 if (kvm_is_error_hva(ghc
->hva
))
1846 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1852 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1854 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1856 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1858 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1860 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1862 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1864 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1866 int offset
= offset_in_page(gpa
);
1869 while ((seg
= next_segment(len
, offset
)) != 0) {
1870 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1879 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1881 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
1884 if (memslot
&& memslot
->dirty_bitmap
) {
1885 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1887 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1891 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1893 struct kvm_memory_slot
*memslot
;
1895 memslot
= gfn_to_memslot(kvm
, gfn
);
1896 mark_page_dirty_in_slot(memslot
, gfn
);
1898 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1900 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1902 struct kvm_memory_slot
*memslot
;
1904 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1905 mark_page_dirty_in_slot(memslot
, gfn
);
1907 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
1909 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
1911 if (kvm_arch_vcpu_runnable(vcpu
)) {
1912 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1915 if (kvm_cpu_has_pending_timer(vcpu
))
1917 if (signal_pending(current
))
1924 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1926 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1930 bool waited
= false;
1932 start
= cur
= ktime_get();
1934 ktime_t stop
= ktime_add_ns(ktime_get(), halt_poll_ns
);
1938 * This sets KVM_REQ_UNHALT if an interrupt
1941 if (kvm_vcpu_check_block(vcpu
) < 0) {
1942 ++vcpu
->stat
.halt_successful_poll
;
1946 } while (single_task_running() && ktime_before(cur
, stop
));
1950 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1952 if (kvm_vcpu_check_block(vcpu
) < 0)
1959 finish_wait(&vcpu
->wq
, &wait
);
1963 trace_kvm_vcpu_wakeup(ktime_to_ns(cur
) - ktime_to_ns(start
), waited
);
1965 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
1969 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1971 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1974 int cpu
= vcpu
->cpu
;
1975 wait_queue_head_t
*wqp
;
1977 wqp
= kvm_arch_vcpu_wq(vcpu
);
1978 if (waitqueue_active(wqp
)) {
1979 wake_up_interruptible(wqp
);
1980 ++vcpu
->stat
.halt_wakeup
;
1984 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1985 if (kvm_arch_vcpu_should_kick(vcpu
))
1986 smp_send_reschedule(cpu
);
1989 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1990 #endif /* !CONFIG_S390 */
1992 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1995 struct task_struct
*task
= NULL
;
1999 pid
= rcu_dereference(target
->pid
);
2001 task
= get_pid_task(pid
, PIDTYPE_PID
);
2005 ret
= yield_to(task
, 1);
2006 put_task_struct(task
);
2010 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2013 * Helper that checks whether a VCPU is eligible for directed yield.
2014 * Most eligible candidate to yield is decided by following heuristics:
2016 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2017 * (preempted lock holder), indicated by @in_spin_loop.
2018 * Set at the beiginning and cleared at the end of interception/PLE handler.
2020 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2021 * chance last time (mostly it has become eligible now since we have probably
2022 * yielded to lockholder in last iteration. This is done by toggling
2023 * @dy_eligible each time a VCPU checked for eligibility.)
2025 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2026 * to preempted lock-holder could result in wrong VCPU selection and CPU
2027 * burning. Giving priority for a potential lock-holder increases lock
2030 * Since algorithm is based on heuristics, accessing another VCPU data without
2031 * locking does not harm. It may result in trying to yield to same VCPU, fail
2032 * and continue with next VCPU and so on.
2034 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2036 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2039 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2040 vcpu
->spin_loop
.dy_eligible
;
2042 if (vcpu
->spin_loop
.in_spin_loop
)
2043 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2051 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2053 struct kvm
*kvm
= me
->kvm
;
2054 struct kvm_vcpu
*vcpu
;
2055 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2061 kvm_vcpu_set_in_spin_loop(me
, true);
2063 * We boost the priority of a VCPU that is runnable but not
2064 * currently running, because it got preempted by something
2065 * else and called schedule in __vcpu_run. Hopefully that
2066 * VCPU is holding the lock that we need and will release it.
2067 * We approximate round-robin by starting at the last boosted VCPU.
2069 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2070 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2071 if (!pass
&& i
<= last_boosted_vcpu
) {
2072 i
= last_boosted_vcpu
;
2074 } else if (pass
&& i
> last_boosted_vcpu
)
2076 if (!ACCESS_ONCE(vcpu
->preempted
))
2080 if (waitqueue_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2082 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2085 yielded
= kvm_vcpu_yield_to(vcpu
);
2087 kvm
->last_boosted_vcpu
= i
;
2089 } else if (yielded
< 0) {
2096 kvm_vcpu_set_in_spin_loop(me
, false);
2098 /* Ensure vcpu is not eligible during next spinloop */
2099 kvm_vcpu_set_dy_eligible(me
, false);
2101 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2103 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2105 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2108 if (vmf
->pgoff
== 0)
2109 page
= virt_to_page(vcpu
->run
);
2111 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2112 page
= virt_to_page(vcpu
->arch
.pio_data
);
2114 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2115 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2116 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2119 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2125 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2126 .fault
= kvm_vcpu_fault
,
2129 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2131 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2135 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2137 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2139 kvm_put_kvm(vcpu
->kvm
);
2143 static struct file_operations kvm_vcpu_fops
= {
2144 .release
= kvm_vcpu_release
,
2145 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2146 #ifdef CONFIG_KVM_COMPAT
2147 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2149 .mmap
= kvm_vcpu_mmap
,
2150 .llseek
= noop_llseek
,
2154 * Allocates an inode for the vcpu.
2156 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2158 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2162 * Creates some virtual cpus. Good luck creating more than one.
2164 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2167 struct kvm_vcpu
*vcpu
, *v
;
2169 if (id
>= KVM_MAX_VCPUS
)
2172 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2174 return PTR_ERR(vcpu
);
2176 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2178 r
= kvm_arch_vcpu_setup(vcpu
);
2182 mutex_lock(&kvm
->lock
);
2183 if (!kvm_vcpu_compatible(vcpu
)) {
2185 goto unlock_vcpu_destroy
;
2187 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
2189 goto unlock_vcpu_destroy
;
2192 kvm_for_each_vcpu(r
, v
, kvm
)
2193 if (v
->vcpu_id
== id
) {
2195 goto unlock_vcpu_destroy
;
2198 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2200 /* Now it's all set up, let userspace reach it */
2202 r
= create_vcpu_fd(vcpu
);
2205 goto unlock_vcpu_destroy
;
2208 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2211 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2212 * before kvm->online_vcpu's incremented value.
2215 atomic_inc(&kvm
->online_vcpus
);
2217 mutex_unlock(&kvm
->lock
);
2218 kvm_arch_vcpu_postcreate(vcpu
);
2221 unlock_vcpu_destroy
:
2222 mutex_unlock(&kvm
->lock
);
2224 kvm_arch_vcpu_destroy(vcpu
);
2228 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2231 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2232 vcpu
->sigset_active
= 1;
2233 vcpu
->sigset
= *sigset
;
2235 vcpu
->sigset_active
= 0;
2239 static long kvm_vcpu_ioctl(struct file
*filp
,
2240 unsigned int ioctl
, unsigned long arg
)
2242 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2243 void __user
*argp
= (void __user
*)arg
;
2245 struct kvm_fpu
*fpu
= NULL
;
2246 struct kvm_sregs
*kvm_sregs
= NULL
;
2248 if (vcpu
->kvm
->mm
!= current
->mm
)
2251 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2254 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2256 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2257 * so vcpu_load() would break it.
2259 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2260 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2264 r
= vcpu_load(vcpu
);
2272 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2273 /* The thread running this VCPU changed. */
2274 struct pid
*oldpid
= vcpu
->pid
;
2275 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2277 rcu_assign_pointer(vcpu
->pid
, newpid
);
2282 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2283 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2285 case KVM_GET_REGS
: {
2286 struct kvm_regs
*kvm_regs
;
2289 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2292 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2296 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2303 case KVM_SET_REGS
: {
2304 struct kvm_regs
*kvm_regs
;
2307 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2308 if (IS_ERR(kvm_regs
)) {
2309 r
= PTR_ERR(kvm_regs
);
2312 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2316 case KVM_GET_SREGS
: {
2317 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2321 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2325 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2330 case KVM_SET_SREGS
: {
2331 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2332 if (IS_ERR(kvm_sregs
)) {
2333 r
= PTR_ERR(kvm_sregs
);
2337 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2340 case KVM_GET_MP_STATE
: {
2341 struct kvm_mp_state mp_state
;
2343 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2347 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2352 case KVM_SET_MP_STATE
: {
2353 struct kvm_mp_state mp_state
;
2356 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2358 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2361 case KVM_TRANSLATE
: {
2362 struct kvm_translation tr
;
2365 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2367 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2371 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2376 case KVM_SET_GUEST_DEBUG
: {
2377 struct kvm_guest_debug dbg
;
2380 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2382 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2385 case KVM_SET_SIGNAL_MASK
: {
2386 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2387 struct kvm_signal_mask kvm_sigmask
;
2388 sigset_t sigset
, *p
;
2393 if (copy_from_user(&kvm_sigmask
, argp
,
2394 sizeof(kvm_sigmask
)))
2397 if (kvm_sigmask
.len
!= sizeof(sigset
))
2400 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2405 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2409 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2413 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2417 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2423 fpu
= memdup_user(argp
, sizeof(*fpu
));
2429 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2433 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2442 #ifdef CONFIG_KVM_COMPAT
2443 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2444 unsigned int ioctl
, unsigned long arg
)
2446 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2447 void __user
*argp
= compat_ptr(arg
);
2450 if (vcpu
->kvm
->mm
!= current
->mm
)
2454 case KVM_SET_SIGNAL_MASK
: {
2455 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2456 struct kvm_signal_mask kvm_sigmask
;
2457 compat_sigset_t csigset
;
2462 if (copy_from_user(&kvm_sigmask
, argp
,
2463 sizeof(kvm_sigmask
)))
2466 if (kvm_sigmask
.len
!= sizeof(csigset
))
2469 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2472 sigset_from_compat(&sigset
, &csigset
);
2473 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2475 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2479 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2487 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2488 int (*accessor
)(struct kvm_device
*dev
,
2489 struct kvm_device_attr
*attr
),
2492 struct kvm_device_attr attr
;
2497 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2500 return accessor(dev
, &attr
);
2503 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2506 struct kvm_device
*dev
= filp
->private_data
;
2509 case KVM_SET_DEVICE_ATTR
:
2510 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2511 case KVM_GET_DEVICE_ATTR
:
2512 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2513 case KVM_HAS_DEVICE_ATTR
:
2514 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2516 if (dev
->ops
->ioctl
)
2517 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2523 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2525 struct kvm_device
*dev
= filp
->private_data
;
2526 struct kvm
*kvm
= dev
->kvm
;
2532 static const struct file_operations kvm_device_fops
= {
2533 .unlocked_ioctl
= kvm_device_ioctl
,
2534 #ifdef CONFIG_KVM_COMPAT
2535 .compat_ioctl
= kvm_device_ioctl
,
2537 .release
= kvm_device_release
,
2540 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2542 if (filp
->f_op
!= &kvm_device_fops
)
2545 return filp
->private_data
;
2548 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2549 #ifdef CONFIG_KVM_MPIC
2550 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2551 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2554 #ifdef CONFIG_KVM_XICS
2555 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2559 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2561 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2564 if (kvm_device_ops_table
[type
] != NULL
)
2567 kvm_device_ops_table
[type
] = ops
;
2571 void kvm_unregister_device_ops(u32 type
)
2573 if (kvm_device_ops_table
[type
] != NULL
)
2574 kvm_device_ops_table
[type
] = NULL
;
2577 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2578 struct kvm_create_device
*cd
)
2580 struct kvm_device_ops
*ops
= NULL
;
2581 struct kvm_device
*dev
;
2582 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2585 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2588 ops
= kvm_device_ops_table
[cd
->type
];
2595 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2602 ret
= ops
->create(dev
, cd
->type
);
2608 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2614 list_add(&dev
->vm_node
, &kvm
->devices
);
2620 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2623 case KVM_CAP_USER_MEMORY
:
2624 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2625 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2626 case KVM_CAP_INTERNAL_ERROR_DATA
:
2627 #ifdef CONFIG_HAVE_KVM_MSI
2628 case KVM_CAP_SIGNAL_MSI
:
2630 #ifdef CONFIG_HAVE_KVM_IRQFD
2632 case KVM_CAP_IRQFD_RESAMPLE
:
2634 case KVM_CAP_CHECK_EXTENSION_VM
:
2636 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2637 case KVM_CAP_IRQ_ROUTING
:
2638 return KVM_MAX_IRQ_ROUTES
;
2640 #if KVM_ADDRESS_SPACE_NUM > 1
2641 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2642 return KVM_ADDRESS_SPACE_NUM
;
2647 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2650 static long kvm_vm_ioctl(struct file
*filp
,
2651 unsigned int ioctl
, unsigned long arg
)
2653 struct kvm
*kvm
= filp
->private_data
;
2654 void __user
*argp
= (void __user
*)arg
;
2657 if (kvm
->mm
!= current
->mm
)
2660 case KVM_CREATE_VCPU
:
2661 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2663 case KVM_SET_USER_MEMORY_REGION
: {
2664 struct kvm_userspace_memory_region kvm_userspace_mem
;
2667 if (copy_from_user(&kvm_userspace_mem
, argp
,
2668 sizeof(kvm_userspace_mem
)))
2671 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2674 case KVM_GET_DIRTY_LOG
: {
2675 struct kvm_dirty_log log
;
2678 if (copy_from_user(&log
, argp
, sizeof(log
)))
2680 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2683 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2684 case KVM_REGISTER_COALESCED_MMIO
: {
2685 struct kvm_coalesced_mmio_zone zone
;
2688 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2690 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2693 case KVM_UNREGISTER_COALESCED_MMIO
: {
2694 struct kvm_coalesced_mmio_zone zone
;
2697 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2699 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2704 struct kvm_irqfd data
;
2707 if (copy_from_user(&data
, argp
, sizeof(data
)))
2709 r
= kvm_irqfd(kvm
, &data
);
2712 case KVM_IOEVENTFD
: {
2713 struct kvm_ioeventfd data
;
2716 if (copy_from_user(&data
, argp
, sizeof(data
)))
2718 r
= kvm_ioeventfd(kvm
, &data
);
2721 #ifdef CONFIG_HAVE_KVM_MSI
2722 case KVM_SIGNAL_MSI
: {
2726 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
2728 r
= kvm_send_userspace_msi(kvm
, &msi
);
2732 #ifdef __KVM_HAVE_IRQ_LINE
2733 case KVM_IRQ_LINE_STATUS
:
2734 case KVM_IRQ_LINE
: {
2735 struct kvm_irq_level irq_event
;
2738 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
2741 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2742 ioctl
== KVM_IRQ_LINE_STATUS
);
2747 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2748 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
2756 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2757 case KVM_SET_GSI_ROUTING
: {
2758 struct kvm_irq_routing routing
;
2759 struct kvm_irq_routing __user
*urouting
;
2760 struct kvm_irq_routing_entry
*entries
;
2763 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2766 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2771 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2776 if (copy_from_user(entries
, urouting
->entries
,
2777 routing
.nr
* sizeof(*entries
)))
2778 goto out_free_irq_routing
;
2779 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2781 out_free_irq_routing
:
2785 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2786 case KVM_CREATE_DEVICE
: {
2787 struct kvm_create_device cd
;
2790 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2793 r
= kvm_ioctl_create_device(kvm
, &cd
);
2798 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2804 case KVM_CHECK_EXTENSION
:
2805 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
2808 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2814 #ifdef CONFIG_KVM_COMPAT
2815 struct compat_kvm_dirty_log
{
2819 compat_uptr_t dirty_bitmap
; /* one bit per page */
2824 static long kvm_vm_compat_ioctl(struct file
*filp
,
2825 unsigned int ioctl
, unsigned long arg
)
2827 struct kvm
*kvm
= filp
->private_data
;
2830 if (kvm
->mm
!= current
->mm
)
2833 case KVM_GET_DIRTY_LOG
: {
2834 struct compat_kvm_dirty_log compat_log
;
2835 struct kvm_dirty_log log
;
2838 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2839 sizeof(compat_log
)))
2841 log
.slot
= compat_log
.slot
;
2842 log
.padding1
= compat_log
.padding1
;
2843 log
.padding2
= compat_log
.padding2
;
2844 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2846 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2850 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2858 static struct file_operations kvm_vm_fops
= {
2859 .release
= kvm_vm_release
,
2860 .unlocked_ioctl
= kvm_vm_ioctl
,
2861 #ifdef CONFIG_KVM_COMPAT
2862 .compat_ioctl
= kvm_vm_compat_ioctl
,
2864 .llseek
= noop_llseek
,
2867 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2872 kvm
= kvm_create_vm(type
);
2874 return PTR_ERR(kvm
);
2875 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2876 r
= kvm_coalesced_mmio_init(kvm
);
2882 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2889 static long kvm_dev_ioctl(struct file
*filp
,
2890 unsigned int ioctl
, unsigned long arg
)
2895 case KVM_GET_API_VERSION
:
2898 r
= KVM_API_VERSION
;
2901 r
= kvm_dev_ioctl_create_vm(arg
);
2903 case KVM_CHECK_EXTENSION
:
2904 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
2906 case KVM_GET_VCPU_MMAP_SIZE
:
2909 r
= PAGE_SIZE
; /* struct kvm_run */
2911 r
+= PAGE_SIZE
; /* pio data page */
2913 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2914 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2917 case KVM_TRACE_ENABLE
:
2918 case KVM_TRACE_PAUSE
:
2919 case KVM_TRACE_DISABLE
:
2923 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2929 static struct file_operations kvm_chardev_ops
= {
2930 .unlocked_ioctl
= kvm_dev_ioctl
,
2931 .compat_ioctl
= kvm_dev_ioctl
,
2932 .llseek
= noop_llseek
,
2935 static struct miscdevice kvm_dev
= {
2941 static void hardware_enable_nolock(void *junk
)
2943 int cpu
= raw_smp_processor_id();
2946 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2949 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2951 r
= kvm_arch_hardware_enable();
2954 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2955 atomic_inc(&hardware_enable_failed
);
2956 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
2960 static void hardware_enable(void)
2962 raw_spin_lock(&kvm_count_lock
);
2963 if (kvm_usage_count
)
2964 hardware_enable_nolock(NULL
);
2965 raw_spin_unlock(&kvm_count_lock
);
2968 static void hardware_disable_nolock(void *junk
)
2970 int cpu
= raw_smp_processor_id();
2972 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2974 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2975 kvm_arch_hardware_disable();
2978 static void hardware_disable(void)
2980 raw_spin_lock(&kvm_count_lock
);
2981 if (kvm_usage_count
)
2982 hardware_disable_nolock(NULL
);
2983 raw_spin_unlock(&kvm_count_lock
);
2986 static void hardware_disable_all_nolock(void)
2988 BUG_ON(!kvm_usage_count
);
2991 if (!kvm_usage_count
)
2992 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2995 static void hardware_disable_all(void)
2997 raw_spin_lock(&kvm_count_lock
);
2998 hardware_disable_all_nolock();
2999 raw_spin_unlock(&kvm_count_lock
);
3002 static int hardware_enable_all(void)
3006 raw_spin_lock(&kvm_count_lock
);
3009 if (kvm_usage_count
== 1) {
3010 atomic_set(&hardware_enable_failed
, 0);
3011 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3013 if (atomic_read(&hardware_enable_failed
)) {
3014 hardware_disable_all_nolock();
3019 raw_spin_unlock(&kvm_count_lock
);
3024 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
3027 val
&= ~CPU_TASKS_FROZEN
;
3039 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3043 * Some (well, at least mine) BIOSes hang on reboot if
3046 * And Intel TXT required VMX off for all cpu when system shutdown.
3048 pr_info("kvm: exiting hardware virtualization\n");
3049 kvm_rebooting
= true;
3050 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3054 static struct notifier_block kvm_reboot_notifier
= {
3055 .notifier_call
= kvm_reboot
,
3059 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3063 for (i
= 0; i
< bus
->dev_count
; i
++) {
3064 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3066 kvm_iodevice_destructor(pos
);
3071 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3072 const struct kvm_io_range
*r2
)
3074 if (r1
->addr
< r2
->addr
)
3076 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
3081 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3083 return kvm_io_bus_cmp(p1
, p2
);
3086 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3087 gpa_t addr
, int len
)
3089 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3095 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3096 kvm_io_bus_sort_cmp
, NULL
);
3101 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3102 gpa_t addr
, int len
)
3104 struct kvm_io_range
*range
, key
;
3107 key
= (struct kvm_io_range
) {
3112 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3113 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3117 off
= range
- bus
->range
;
3119 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3125 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3126 struct kvm_io_range
*range
, const void *val
)
3130 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3134 while (idx
< bus
->dev_count
&&
3135 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3136 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3145 /* kvm_io_bus_write - called under kvm->slots_lock */
3146 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3147 int len
, const void *val
)
3149 struct kvm_io_bus
*bus
;
3150 struct kvm_io_range range
;
3153 range
= (struct kvm_io_range
) {
3158 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3159 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3160 return r
< 0 ? r
: 0;
3163 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3164 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3165 gpa_t addr
, int len
, const void *val
, long cookie
)
3167 struct kvm_io_bus
*bus
;
3168 struct kvm_io_range range
;
3170 range
= (struct kvm_io_range
) {
3175 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3177 /* First try the device referenced by cookie. */
3178 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3179 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3180 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3185 * cookie contained garbage; fall back to search and return the
3186 * correct cookie value.
3188 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3191 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3192 struct kvm_io_range
*range
, void *val
)
3196 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3200 while (idx
< bus
->dev_count
&&
3201 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3202 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3210 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3212 /* kvm_io_bus_read - called under kvm->slots_lock */
3213 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3216 struct kvm_io_bus
*bus
;
3217 struct kvm_io_range range
;
3220 range
= (struct kvm_io_range
) {
3225 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3226 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3227 return r
< 0 ? r
: 0;
3231 /* Caller must hold slots_lock. */
3232 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3233 int len
, struct kvm_io_device
*dev
)
3235 struct kvm_io_bus
*new_bus
, *bus
;
3237 bus
= kvm
->buses
[bus_idx
];
3238 /* exclude ioeventfd which is limited by maximum fd */
3239 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3242 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3243 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3246 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3247 sizeof(struct kvm_io_range
)));
3248 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3249 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3250 synchronize_srcu_expedited(&kvm
->srcu
);
3256 /* Caller must hold slots_lock. */
3257 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3258 struct kvm_io_device
*dev
)
3261 struct kvm_io_bus
*new_bus
, *bus
;
3263 bus
= kvm
->buses
[bus_idx
];
3265 for (i
= 0; i
< bus
->dev_count
; i
++)
3266 if (bus
->range
[i
].dev
== dev
) {
3274 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3275 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3279 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3280 new_bus
->dev_count
--;
3281 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3282 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3284 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3285 synchronize_srcu_expedited(&kvm
->srcu
);
3290 static struct notifier_block kvm_cpu_notifier
= {
3291 .notifier_call
= kvm_cpu_hotplug
,
3294 static int vm_stat_get(void *_offset
, u64
*val
)
3296 unsigned offset
= (long)_offset
;
3300 spin_lock(&kvm_lock
);
3301 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3302 *val
+= *(u32
*)((void *)kvm
+ offset
);
3303 spin_unlock(&kvm_lock
);
3307 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3309 static int vcpu_stat_get(void *_offset
, u64
*val
)
3311 unsigned offset
= (long)_offset
;
3313 struct kvm_vcpu
*vcpu
;
3317 spin_lock(&kvm_lock
);
3318 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3319 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3320 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3322 spin_unlock(&kvm_lock
);
3326 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3328 static const struct file_operations
*stat_fops
[] = {
3329 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3330 [KVM_STAT_VM
] = &vm_stat_fops
,
3333 static int kvm_init_debug(void)
3336 struct kvm_stats_debugfs_item
*p
;
3338 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3339 if (kvm_debugfs_dir
== NULL
)
3342 for (p
= debugfs_entries
; p
->name
; ++p
) {
3343 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3344 (void *)(long)p
->offset
,
3345 stat_fops
[p
->kind
]);
3346 if (p
->dentry
== NULL
)
3353 debugfs_remove_recursive(kvm_debugfs_dir
);
3358 static void kvm_exit_debug(void)
3360 struct kvm_stats_debugfs_item
*p
;
3362 for (p
= debugfs_entries
; p
->name
; ++p
)
3363 debugfs_remove(p
->dentry
);
3364 debugfs_remove(kvm_debugfs_dir
);
3367 static int kvm_suspend(void)
3369 if (kvm_usage_count
)
3370 hardware_disable_nolock(NULL
);
3374 static void kvm_resume(void)
3376 if (kvm_usage_count
) {
3377 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3378 hardware_enable_nolock(NULL
);
3382 static struct syscore_ops kvm_syscore_ops
= {
3383 .suspend
= kvm_suspend
,
3384 .resume
= kvm_resume
,
3388 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3390 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3393 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3395 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3397 if (vcpu
->preempted
)
3398 vcpu
->preempted
= false;
3400 kvm_arch_sched_in(vcpu
, cpu
);
3402 kvm_arch_vcpu_load(vcpu
, cpu
);
3405 static void kvm_sched_out(struct preempt_notifier
*pn
,
3406 struct task_struct
*next
)
3408 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3410 if (current
->state
== TASK_RUNNING
)
3411 vcpu
->preempted
= true;
3412 kvm_arch_vcpu_put(vcpu
);
3415 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3416 struct module
*module
)
3421 r
= kvm_arch_init(opaque
);
3426 * kvm_arch_init makes sure there's at most one caller
3427 * for architectures that support multiple implementations,
3428 * like intel and amd on x86.
3429 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3430 * conflicts in case kvm is already setup for another implementation.
3432 r
= kvm_irqfd_init();
3436 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3441 r
= kvm_arch_hardware_setup();
3445 for_each_online_cpu(cpu
) {
3446 smp_call_function_single(cpu
,
3447 kvm_arch_check_processor_compat
,
3453 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3456 register_reboot_notifier(&kvm_reboot_notifier
);
3458 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3460 vcpu_align
= __alignof__(struct kvm_vcpu
);
3461 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3463 if (!kvm_vcpu_cache
) {
3468 r
= kvm_async_pf_init();
3472 kvm_chardev_ops
.owner
= module
;
3473 kvm_vm_fops
.owner
= module
;
3474 kvm_vcpu_fops
.owner
= module
;
3476 r
= misc_register(&kvm_dev
);
3478 pr_err("kvm: misc device register failed\n");
3482 register_syscore_ops(&kvm_syscore_ops
);
3484 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3485 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3487 r
= kvm_init_debug();
3489 pr_err("kvm: create debugfs files failed\n");
3493 r
= kvm_vfio_ops_init();
3499 unregister_syscore_ops(&kvm_syscore_ops
);
3500 misc_deregister(&kvm_dev
);
3502 kvm_async_pf_deinit();
3504 kmem_cache_destroy(kvm_vcpu_cache
);
3506 unregister_reboot_notifier(&kvm_reboot_notifier
);
3507 unregister_cpu_notifier(&kvm_cpu_notifier
);
3510 kvm_arch_hardware_unsetup();
3512 free_cpumask_var(cpus_hardware_enabled
);
3520 EXPORT_SYMBOL_GPL(kvm_init
);
3525 misc_deregister(&kvm_dev
);
3526 kmem_cache_destroy(kvm_vcpu_cache
);
3527 kvm_async_pf_deinit();
3528 unregister_syscore_ops(&kvm_syscore_ops
);
3529 unregister_reboot_notifier(&kvm_reboot_notifier
);
3530 unregister_cpu_notifier(&kvm_cpu_notifier
);
3531 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3532 kvm_arch_hardware_unsetup();
3535 free_cpumask_var(cpus_hardware_enabled
);
3536 kvm_vfio_ops_exit();
3538 EXPORT_SYMBOL_GPL(kvm_exit
);