2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_RAW_SPINLOCK(kvm_lock
);
76 static cpumask_var_t cpus_hardware_enabled
;
77 static int kvm_usage_count
= 0;
78 static atomic_t hardware_enable_failed
;
80 struct kmem_cache
*kvm_vcpu_cache
;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
83 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
85 struct dentry
*kvm_debugfs_dir
;
87 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
90 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
99 EXPORT_SYMBOL_GPL(kvm_rebooting
);
101 static bool largepages_enabled
= true;
103 bool kvm_is_mmio_pfn(pfn_t pfn
)
105 if (pfn_valid(pfn
)) {
107 struct page
*tail
= pfn_to_page(pfn
);
108 struct page
*head
= compound_trans_head(tail
);
109 reserved
= PageReserved(head
);
112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
125 return PageReserved(tail
);
132 * Switches to specified vcpu, until a matching vcpu_put()
134 int vcpu_load(struct kvm_vcpu
*vcpu
)
138 if (mutex_lock_killable(&vcpu
->mutex
))
140 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
141 /* The thread running this VCPU changed. */
142 struct pid
*oldpid
= vcpu
->pid
;
143 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
144 rcu_assign_pointer(vcpu
->pid
, newpid
);
149 preempt_notifier_register(&vcpu
->preempt_notifier
);
150 kvm_arch_vcpu_load(vcpu
, cpu
);
155 void vcpu_put(struct kvm_vcpu
*vcpu
)
158 kvm_arch_vcpu_put(vcpu
);
159 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
161 mutex_unlock(&vcpu
->mutex
);
164 static void ack_flush(void *_completed
)
168 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
173 struct kvm_vcpu
*vcpu
;
175 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
178 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
179 kvm_make_request(req
, vcpu
);
182 /* Set ->requests bit before we read ->mode */
185 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
186 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
187 cpumask_set_cpu(cpu
, cpus
);
189 if (unlikely(cpus
== NULL
))
190 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
191 else if (!cpumask_empty(cpus
))
192 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
196 free_cpumask_var(cpus
);
200 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
202 long dirty_count
= kvm
->tlbs_dirty
;
205 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
206 ++kvm
->stat
.remote_tlb_flush
;
207 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
210 void kvm_reload_remote_mmus(struct kvm
*kvm
)
212 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
215 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
217 make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
220 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
225 mutex_init(&vcpu
->mutex
);
230 init_waitqueue_head(&vcpu
->wq
);
231 kvm_async_pf_vcpu_init(vcpu
);
233 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
238 vcpu
->run
= page_address(page
);
240 kvm_vcpu_set_in_spin_loop(vcpu
, false);
241 kvm_vcpu_set_dy_eligible(vcpu
, false);
243 r
= kvm_arch_vcpu_init(vcpu
);
249 free_page((unsigned long)vcpu
->run
);
253 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
255 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
258 kvm_arch_vcpu_uninit(vcpu
);
259 free_page((unsigned long)vcpu
->run
);
261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
266 return container_of(mn
, struct kvm
, mmu_notifier
);
269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
270 struct mm_struct
*mm
,
271 unsigned long address
)
273 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
274 int need_tlb_flush
, idx
;
277 * When ->invalidate_page runs, the linux pte has been zapped
278 * already but the page is still allocated until
279 * ->invalidate_page returns. So if we increase the sequence
280 * here the kvm page fault will notice if the spte can't be
281 * established because the page is going to be freed. If
282 * instead the kvm page fault establishes the spte before
283 * ->invalidate_page runs, kvm_unmap_hva will release it
286 * The sequence increase only need to be seen at spin_unlock
287 * time, and not at spin_lock time.
289 * Increasing the sequence after the spin_unlock would be
290 * unsafe because the kvm page fault could then establish the
291 * pte after kvm_unmap_hva returned, without noticing the page
292 * is going to be freed.
294 idx
= srcu_read_lock(&kvm
->srcu
);
295 spin_lock(&kvm
->mmu_lock
);
297 kvm
->mmu_notifier_seq
++;
298 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
299 /* we've to flush the tlb before the pages can be freed */
301 kvm_flush_remote_tlbs(kvm
);
303 spin_unlock(&kvm
->mmu_lock
);
304 srcu_read_unlock(&kvm
->srcu
, idx
);
307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
308 struct mm_struct
*mm
,
309 unsigned long address
,
312 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
315 idx
= srcu_read_lock(&kvm
->srcu
);
316 spin_lock(&kvm
->mmu_lock
);
317 kvm
->mmu_notifier_seq
++;
318 kvm_set_spte_hva(kvm
, address
, pte
);
319 spin_unlock(&kvm
->mmu_lock
);
320 srcu_read_unlock(&kvm
->srcu
, idx
);
323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
324 struct mm_struct
*mm
,
328 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
329 int need_tlb_flush
= 0, idx
;
331 idx
= srcu_read_lock(&kvm
->srcu
);
332 spin_lock(&kvm
->mmu_lock
);
334 * The count increase must become visible at unlock time as no
335 * spte can be established without taking the mmu_lock and
336 * count is also read inside the mmu_lock critical section.
338 kvm
->mmu_notifier_count
++;
339 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
340 need_tlb_flush
|= kvm
->tlbs_dirty
;
341 /* we've to flush the tlb before the pages can be freed */
343 kvm_flush_remote_tlbs(kvm
);
345 spin_unlock(&kvm
->mmu_lock
);
346 srcu_read_unlock(&kvm
->srcu
, idx
);
349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
350 struct mm_struct
*mm
,
354 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
356 spin_lock(&kvm
->mmu_lock
);
358 * This sequence increase will notify the kvm page fault that
359 * the page that is going to be mapped in the spte could have
362 kvm
->mmu_notifier_seq
++;
365 * The above sequence increase must be visible before the
366 * below count decrease, which is ensured by the smp_wmb above
367 * in conjunction with the smp_rmb in mmu_notifier_retry().
369 kvm
->mmu_notifier_count
--;
370 spin_unlock(&kvm
->mmu_lock
);
372 BUG_ON(kvm
->mmu_notifier_count
< 0);
375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
376 struct mm_struct
*mm
,
377 unsigned long address
)
379 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
382 idx
= srcu_read_lock(&kvm
->srcu
);
383 spin_lock(&kvm
->mmu_lock
);
385 young
= kvm_age_hva(kvm
, address
);
387 kvm_flush_remote_tlbs(kvm
);
389 spin_unlock(&kvm
->mmu_lock
);
390 srcu_read_unlock(&kvm
->srcu
, idx
);
395 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
396 struct mm_struct
*mm
,
397 unsigned long address
)
399 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
402 idx
= srcu_read_lock(&kvm
->srcu
);
403 spin_lock(&kvm
->mmu_lock
);
404 young
= kvm_test_age_hva(kvm
, address
);
405 spin_unlock(&kvm
->mmu_lock
);
406 srcu_read_unlock(&kvm
->srcu
, idx
);
411 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
412 struct mm_struct
*mm
)
414 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
417 idx
= srcu_read_lock(&kvm
->srcu
);
418 kvm_arch_flush_shadow_all(kvm
);
419 srcu_read_unlock(&kvm
->srcu
, idx
);
422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
423 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
424 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
425 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
426 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
427 .test_young
= kvm_mmu_notifier_test_young
,
428 .change_pte
= kvm_mmu_notifier_change_pte
,
429 .release
= kvm_mmu_notifier_release
,
432 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
434 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
435 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
438 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
440 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
447 static void kvm_init_memslots_id(struct kvm
*kvm
)
450 struct kvm_memslots
*slots
= kvm
->memslots
;
452 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
453 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
456 static struct kvm
*kvm_create_vm(unsigned long type
)
459 struct kvm
*kvm
= kvm_arch_alloc_vm();
462 return ERR_PTR(-ENOMEM
);
464 r
= kvm_arch_init_vm(kvm
, type
);
466 goto out_err_nodisable
;
468 r
= hardware_enable_all();
470 goto out_err_nodisable
;
472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
473 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
474 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
477 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
480 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
483 kvm_init_memslots_id(kvm
);
484 if (init_srcu_struct(&kvm
->srcu
))
486 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
487 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
493 spin_lock_init(&kvm
->mmu_lock
);
494 kvm
->mm
= current
->mm
;
495 atomic_inc(&kvm
->mm
->mm_count
);
496 kvm_eventfd_init(kvm
);
497 mutex_init(&kvm
->lock
);
498 mutex_init(&kvm
->irq_lock
);
499 mutex_init(&kvm
->slots_lock
);
500 atomic_set(&kvm
->users_count
, 1);
502 r
= kvm_init_mmu_notifier(kvm
);
506 raw_spin_lock(&kvm_lock
);
507 list_add(&kvm
->vm_list
, &vm_list
);
508 raw_spin_unlock(&kvm_lock
);
513 cleanup_srcu_struct(&kvm
->srcu
);
515 hardware_disable_all();
517 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
518 kfree(kvm
->buses
[i
]);
519 kfree(kvm
->memslots
);
520 kvm_arch_free_vm(kvm
);
525 * Avoid using vmalloc for a small buffer.
526 * Should not be used when the size is statically known.
528 void *kvm_kvzalloc(unsigned long size
)
530 if (size
> PAGE_SIZE
)
531 return vzalloc(size
);
533 return kzalloc(size
, GFP_KERNEL
);
536 void kvm_kvfree(const void *addr
)
538 if (is_vmalloc_addr(addr
))
544 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
546 if (!memslot
->dirty_bitmap
)
549 kvm_kvfree(memslot
->dirty_bitmap
);
550 memslot
->dirty_bitmap
= NULL
;
554 * Free any memory in @free but not in @dont.
556 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
557 struct kvm_memory_slot
*dont
)
559 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
560 kvm_destroy_dirty_bitmap(free
);
562 kvm_arch_free_memslot(free
, dont
);
567 void kvm_free_physmem(struct kvm
*kvm
)
569 struct kvm_memslots
*slots
= kvm
->memslots
;
570 struct kvm_memory_slot
*memslot
;
572 kvm_for_each_memslot(memslot
, slots
)
573 kvm_free_physmem_slot(memslot
, NULL
);
575 kfree(kvm
->memslots
);
578 static void kvm_destroy_vm(struct kvm
*kvm
)
581 struct mm_struct
*mm
= kvm
->mm
;
583 kvm_arch_sync_events(kvm
);
584 raw_spin_lock(&kvm_lock
);
585 list_del(&kvm
->vm_list
);
586 raw_spin_unlock(&kvm_lock
);
587 kvm_free_irq_routing(kvm
);
588 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
589 kvm_io_bus_destroy(kvm
->buses
[i
]);
590 kvm_coalesced_mmio_free(kvm
);
591 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
594 kvm_arch_flush_shadow_all(kvm
);
596 kvm_arch_destroy_vm(kvm
);
597 kvm_free_physmem(kvm
);
598 cleanup_srcu_struct(&kvm
->srcu
);
599 kvm_arch_free_vm(kvm
);
600 hardware_disable_all();
604 void kvm_get_kvm(struct kvm
*kvm
)
606 atomic_inc(&kvm
->users_count
);
608 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
610 void kvm_put_kvm(struct kvm
*kvm
)
612 if (atomic_dec_and_test(&kvm
->users_count
))
615 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
618 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
620 struct kvm
*kvm
= filp
->private_data
;
622 kvm_irqfd_release(kvm
);
629 * Allocation size is twice as large as the actual dirty bitmap size.
630 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
632 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
635 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
637 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
638 if (!memslot
->dirty_bitmap
)
641 #endif /* !CONFIG_S390 */
645 static int cmp_memslot(const void *slot1
, const void *slot2
)
647 struct kvm_memory_slot
*s1
, *s2
;
649 s1
= (struct kvm_memory_slot
*)slot1
;
650 s2
= (struct kvm_memory_slot
*)slot2
;
652 if (s1
->npages
< s2
->npages
)
654 if (s1
->npages
> s2
->npages
)
661 * Sort the memslots base on its size, so the larger slots
662 * will get better fit.
664 static void sort_memslots(struct kvm_memslots
*slots
)
668 sort(slots
->memslots
, KVM_MEM_SLOTS_NUM
,
669 sizeof(struct kvm_memory_slot
), cmp_memslot
, NULL
);
671 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
672 slots
->id_to_index
[slots
->memslots
[i
].id
] = i
;
675 void update_memslots(struct kvm_memslots
*slots
, struct kvm_memory_slot
*new,
680 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
681 unsigned long npages
= old
->npages
;
684 if (new->npages
!= npages
)
685 sort_memslots(slots
);
688 slots
->generation
= last_generation
+ 1;
691 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
693 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
695 #ifdef KVM_CAP_READONLY_MEM
696 valid_flags
|= KVM_MEM_READONLY
;
699 if (mem
->flags
& ~valid_flags
)
705 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
706 struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
708 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
710 update_memslots(slots
, new, kvm
->memslots
->generation
);
711 rcu_assign_pointer(kvm
->memslots
, slots
);
712 synchronize_srcu_expedited(&kvm
->srcu
);
717 * Allocate some memory and give it an address in the guest physical address
720 * Discontiguous memory is allowed, mostly for framebuffers.
722 * Must be called holding mmap_sem for write.
724 int __kvm_set_memory_region(struct kvm
*kvm
,
725 struct kvm_userspace_memory_region
*mem
,
730 unsigned long npages
;
731 struct kvm_memory_slot
*memslot
, *slot
;
732 struct kvm_memory_slot old
, new;
733 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
735 r
= check_memory_region_flags(mem
);
740 /* General sanity checks */
741 if (mem
->memory_size
& (PAGE_SIZE
- 1))
743 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
745 /* We can read the guest memory with __xxx_user() later on. */
747 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
748 !access_ok(VERIFY_WRITE
,
749 (void __user
*)(unsigned long)mem
->userspace_addr
,
752 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
754 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
757 memslot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
758 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
759 npages
= mem
->memory_size
>> PAGE_SHIFT
;
762 if (npages
> KVM_MEM_MAX_NR_PAGES
)
766 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
768 new = old
= *memslot
;
771 new.base_gfn
= base_gfn
;
773 new.flags
= mem
->flags
;
776 * Disallow changing a memory slot's size or changing anything about
777 * zero sized slots that doesn't involve making them non-zero.
780 if (npages
&& old
.npages
&& npages
!= old
.npages
)
782 if (!npages
&& !old
.npages
)
785 /* Check for overlaps */
787 kvm_for_each_memslot(slot
, kvm
->memslots
) {
788 if (slot
->id
>= KVM_USER_MEM_SLOTS
|| slot
== memslot
)
790 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
791 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
795 /* Free page dirty bitmap if unneeded */
796 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
797 new.dirty_bitmap
= NULL
;
802 * Allocate if a slot is being created. If modifying a slot,
803 * the userspace_addr cannot change.
806 new.user_alloc
= user_alloc
;
807 new.userspace_addr
= mem
->userspace_addr
;
809 if (kvm_arch_create_memslot(&new, npages
))
811 } else if (npages
&& mem
->userspace_addr
!= old
.userspace_addr
) {
816 /* Allocate page dirty bitmap if needed */
817 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
818 if (kvm_create_dirty_bitmap(&new) < 0)
822 if (!npages
|| base_gfn
!= old
.base_gfn
) {
823 struct kvm_memory_slot
*slot
;
826 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
830 slot
= id_to_memslot(slots
, mem
->slot
);
831 slot
->flags
|= KVM_MEMSLOT_INVALID
;
833 old_memslots
= install_new_memslots(kvm
, slots
, NULL
);
835 /* slot was deleted or moved, clear iommu mapping */
836 kvm_iommu_unmap_pages(kvm
, &old
);
837 /* From this point no new shadow pages pointing to a deleted,
838 * or moved, memslot will be created.
840 * validation of sp->gfn happens in:
841 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
842 * - kvm_is_visible_gfn (mmu_check_roots)
844 kvm_arch_flush_shadow_memslot(kvm
, slot
);
845 slots
= old_memslots
;
848 r
= kvm_arch_prepare_memory_region(kvm
, &new, old
, mem
, user_alloc
);
854 * We can re-use the old_memslots from above, the only difference
855 * from the currently installed memslots is the invalid flag. This
856 * will get overwritten by update_memslots anyway.
859 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
865 /* map new memory slot into the iommu */
867 r
= kvm_iommu_map_pages(kvm
, &new);
872 /* actual memory is freed via old in kvm_free_physmem_slot below */
874 new.dirty_bitmap
= NULL
;
875 memset(&new.arch
, 0, sizeof(new.arch
));
878 old_memslots
= install_new_memslots(kvm
, slots
, &new);
880 kvm_arch_commit_memory_region(kvm
, mem
, old
, user_alloc
);
882 kvm_free_physmem_slot(&old
, &new);
890 kvm_free_physmem_slot(&new, &old
);
895 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
897 int kvm_set_memory_region(struct kvm
*kvm
,
898 struct kvm_userspace_memory_region
*mem
,
903 mutex_lock(&kvm
->slots_lock
);
904 r
= __kvm_set_memory_region(kvm
, mem
, user_alloc
);
905 mutex_unlock(&kvm
->slots_lock
);
908 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
910 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
912 kvm_userspace_memory_region
*mem
,
915 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
917 return kvm_set_memory_region(kvm
, mem
, user_alloc
);
920 int kvm_get_dirty_log(struct kvm
*kvm
,
921 struct kvm_dirty_log
*log
, int *is_dirty
)
923 struct kvm_memory_slot
*memslot
;
926 unsigned long any
= 0;
929 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
932 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
934 if (!memslot
->dirty_bitmap
)
937 n
= kvm_dirty_bitmap_bytes(memslot
);
939 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
940 any
= memslot
->dirty_bitmap
[i
];
943 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
954 bool kvm_largepages_enabled(void)
956 return largepages_enabled
;
959 void kvm_disable_largepages(void)
961 largepages_enabled
= false;
963 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
965 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
967 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
969 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
971 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
973 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
975 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
976 memslot
->flags
& KVM_MEMSLOT_INVALID
)
981 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
983 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
985 struct vm_area_struct
*vma
;
986 unsigned long addr
, size
;
990 addr
= gfn_to_hva(kvm
, gfn
);
991 if (kvm_is_error_hva(addr
))
994 down_read(¤t
->mm
->mmap_sem
);
995 vma
= find_vma(current
->mm
, addr
);
999 size
= vma_kernel_pagesize(vma
);
1002 up_read(¤t
->mm
->mmap_sem
);
1007 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1009 return slot
->flags
& KVM_MEM_READONLY
;
1012 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1013 gfn_t
*nr_pages
, bool write
)
1015 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1016 return KVM_HVA_ERR_BAD
;
1018 if (memslot_is_readonly(slot
) && write
)
1019 return KVM_HVA_ERR_RO_BAD
;
1022 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1024 return __gfn_to_hva_memslot(slot
, gfn
);
1027 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1030 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1033 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1036 return gfn_to_hva_many(slot
, gfn
, NULL
);
1038 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1040 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1042 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1044 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1047 * The hva returned by this function is only allowed to be read.
1048 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1050 static unsigned long gfn_to_hva_read(struct kvm
*kvm
, gfn_t gfn
)
1052 return __gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
, false);
1055 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1057 return __copy_from_user(data
, hva
, len
);
1060 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1062 return __copy_from_user_inatomic(data
, hva
, len
);
1065 int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1066 unsigned long start
, int write
, struct page
**page
)
1068 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1071 flags
|= FOLL_WRITE
;
1073 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1076 static inline int check_user_page_hwpoison(unsigned long addr
)
1078 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1080 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1081 flags
, NULL
, NULL
, NULL
);
1082 return rc
== -EHWPOISON
;
1086 * The atomic path to get the writable pfn which will be stored in @pfn,
1087 * true indicates success, otherwise false is returned.
1089 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1090 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1092 struct page
*page
[1];
1095 if (!(async
|| atomic
))
1099 * Fast pin a writable pfn only if it is a write fault request
1100 * or the caller allows to map a writable pfn for a read fault
1103 if (!(write_fault
|| writable
))
1106 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1108 *pfn
= page_to_pfn(page
[0]);
1119 * The slow path to get the pfn of the specified host virtual address,
1120 * 1 indicates success, -errno is returned if error is detected.
1122 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1123 bool *writable
, pfn_t
*pfn
)
1125 struct page
*page
[1];
1131 *writable
= write_fault
;
1134 down_read(¤t
->mm
->mmap_sem
);
1135 npages
= get_user_page_nowait(current
, current
->mm
,
1136 addr
, write_fault
, page
);
1137 up_read(¤t
->mm
->mmap_sem
);
1139 npages
= get_user_pages_fast(addr
, 1, write_fault
,
1144 /* map read fault as writable if possible */
1145 if (unlikely(!write_fault
) && writable
) {
1146 struct page
*wpage
[1];
1148 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1157 *pfn
= page_to_pfn(page
[0]);
1161 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1163 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1166 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1173 * Pin guest page in memory and return its pfn.
1174 * @addr: host virtual address which maps memory to the guest
1175 * @atomic: whether this function can sleep
1176 * @async: whether this function need to wait IO complete if the
1177 * host page is not in the memory
1178 * @write_fault: whether we should get a writable host page
1179 * @writable: whether it allows to map a writable host page for !@write_fault
1181 * The function will map a writable host page for these two cases:
1182 * 1): @write_fault = true
1183 * 2): @write_fault = false && @writable, @writable will tell the caller
1184 * whether the mapping is writable.
1186 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1187 bool write_fault
, bool *writable
)
1189 struct vm_area_struct
*vma
;
1193 /* we can do it either atomically or asynchronously, not both */
1194 BUG_ON(atomic
&& async
);
1196 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1200 return KVM_PFN_ERR_FAULT
;
1202 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1206 down_read(¤t
->mm
->mmap_sem
);
1207 if (npages
== -EHWPOISON
||
1208 (!async
&& check_user_page_hwpoison(addr
))) {
1209 pfn
= KVM_PFN_ERR_HWPOISON
;
1213 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1216 pfn
= KVM_PFN_ERR_FAULT
;
1217 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1218 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1220 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1222 if (async
&& vma_is_valid(vma
, write_fault
))
1224 pfn
= KVM_PFN_ERR_FAULT
;
1227 up_read(¤t
->mm
->mmap_sem
);
1232 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1233 bool *async
, bool write_fault
, bool *writable
)
1235 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1237 if (addr
== KVM_HVA_ERR_RO_BAD
)
1238 return KVM_PFN_ERR_RO_FAULT
;
1240 if (kvm_is_error_hva(addr
))
1241 return KVM_PFN_NOSLOT
;
1243 /* Do not map writable pfn in the readonly memslot. */
1244 if (writable
&& memslot_is_readonly(slot
)) {
1249 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1253 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1254 bool write_fault
, bool *writable
)
1256 struct kvm_memory_slot
*slot
;
1261 slot
= gfn_to_memslot(kvm
, gfn
);
1263 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1267 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1269 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1271 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1273 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1274 bool write_fault
, bool *writable
)
1276 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1278 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1280 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1282 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1284 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1286 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1289 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1291 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1293 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1295 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1298 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1300 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1302 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1304 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1310 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1311 if (kvm_is_error_hva(addr
))
1314 if (entry
< nr_pages
)
1317 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1319 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1321 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1323 if (is_error_noslot_pfn(pfn
))
1324 return KVM_ERR_PTR_BAD_PAGE
;
1326 if (kvm_is_mmio_pfn(pfn
)) {
1328 return KVM_ERR_PTR_BAD_PAGE
;
1331 return pfn_to_page(pfn
);
1334 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1338 pfn
= gfn_to_pfn(kvm
, gfn
);
1340 return kvm_pfn_to_page(pfn
);
1343 EXPORT_SYMBOL_GPL(gfn_to_page
);
1345 void kvm_release_page_clean(struct page
*page
)
1347 WARN_ON(is_error_page(page
));
1349 kvm_release_pfn_clean(page_to_pfn(page
));
1351 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1353 void kvm_release_pfn_clean(pfn_t pfn
)
1355 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1356 put_page(pfn_to_page(pfn
));
1358 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1360 void kvm_release_page_dirty(struct page
*page
)
1362 WARN_ON(is_error_page(page
));
1364 kvm_release_pfn_dirty(page_to_pfn(page
));
1366 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1368 void kvm_release_pfn_dirty(pfn_t pfn
)
1370 kvm_set_pfn_dirty(pfn
);
1371 kvm_release_pfn_clean(pfn
);
1373 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1375 void kvm_set_page_dirty(struct page
*page
)
1377 kvm_set_pfn_dirty(page_to_pfn(page
));
1379 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1381 void kvm_set_pfn_dirty(pfn_t pfn
)
1383 if (!kvm_is_mmio_pfn(pfn
)) {
1384 struct page
*page
= pfn_to_page(pfn
);
1385 if (!PageReserved(page
))
1389 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1391 void kvm_set_pfn_accessed(pfn_t pfn
)
1393 if (!kvm_is_mmio_pfn(pfn
))
1394 mark_page_accessed(pfn_to_page(pfn
));
1396 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1398 void kvm_get_pfn(pfn_t pfn
)
1400 if (!kvm_is_mmio_pfn(pfn
))
1401 get_page(pfn_to_page(pfn
));
1403 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1405 static int next_segment(unsigned long len
, int offset
)
1407 if (len
> PAGE_SIZE
- offset
)
1408 return PAGE_SIZE
- offset
;
1413 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1419 addr
= gfn_to_hva_read(kvm
, gfn
);
1420 if (kvm_is_error_hva(addr
))
1422 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1427 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1429 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1431 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1433 int offset
= offset_in_page(gpa
);
1436 while ((seg
= next_segment(len
, offset
)) != 0) {
1437 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1447 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1449 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1454 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1455 int offset
= offset_in_page(gpa
);
1457 addr
= gfn_to_hva_read(kvm
, gfn
);
1458 if (kvm_is_error_hva(addr
))
1460 pagefault_disable();
1461 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1467 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1469 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1470 int offset
, int len
)
1475 addr
= gfn_to_hva(kvm
, gfn
);
1476 if (kvm_is_error_hva(addr
))
1478 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1481 mark_page_dirty(kvm
, gfn
);
1484 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1486 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1489 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1491 int offset
= offset_in_page(gpa
);
1494 while ((seg
= next_segment(len
, offset
)) != 0) {
1495 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1506 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1509 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1510 int offset
= offset_in_page(gpa
);
1511 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1514 ghc
->generation
= slots
->generation
;
1515 ghc
->memslot
= gfn_to_memslot(kvm
, gfn
);
1516 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, gfn
, NULL
);
1517 if (!kvm_is_error_hva(ghc
->hva
))
1524 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1526 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1527 void *data
, unsigned long len
)
1529 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1532 if (slots
->generation
!= ghc
->generation
)
1533 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1535 if (kvm_is_error_hva(ghc
->hva
))
1538 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1541 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1545 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1547 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1548 void *data
, unsigned long len
)
1550 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1553 if (slots
->generation
!= ghc
->generation
)
1554 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
);
1556 if (kvm_is_error_hva(ghc
->hva
))
1559 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1565 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1567 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1569 return kvm_write_guest_page(kvm
, gfn
, (const void *) empty_zero_page
,
1572 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1574 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1576 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1578 int offset
= offset_in_page(gpa
);
1581 while ((seg
= next_segment(len
, offset
)) != 0) {
1582 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1591 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1593 void mark_page_dirty_in_slot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
,
1596 if (memslot
&& memslot
->dirty_bitmap
) {
1597 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1599 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1603 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1605 struct kvm_memory_slot
*memslot
;
1607 memslot
= gfn_to_memslot(kvm
, gfn
);
1608 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1612 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1614 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1619 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1621 if (kvm_arch_vcpu_runnable(vcpu
)) {
1622 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1625 if (kvm_cpu_has_pending_timer(vcpu
))
1627 if (signal_pending(current
))
1633 finish_wait(&vcpu
->wq
, &wait
);
1638 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1640 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1643 int cpu
= vcpu
->cpu
;
1644 wait_queue_head_t
*wqp
;
1646 wqp
= kvm_arch_vcpu_wq(vcpu
);
1647 if (waitqueue_active(wqp
)) {
1648 wake_up_interruptible(wqp
);
1649 ++vcpu
->stat
.halt_wakeup
;
1653 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1654 if (kvm_arch_vcpu_should_kick(vcpu
))
1655 smp_send_reschedule(cpu
);
1658 #endif /* !CONFIG_S390 */
1660 void kvm_resched(struct kvm_vcpu
*vcpu
)
1662 if (!need_resched())
1666 EXPORT_SYMBOL_GPL(kvm_resched
);
1668 bool kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1671 struct task_struct
*task
= NULL
;
1674 pid
= rcu_dereference(target
->pid
);
1676 task
= get_pid_task(target
->pid
, PIDTYPE_PID
);
1680 if (task
->flags
& PF_VCPU
) {
1681 put_task_struct(task
);
1684 if (yield_to(task
, 1)) {
1685 put_task_struct(task
);
1688 put_task_struct(task
);
1691 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1693 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1695 * Helper that checks whether a VCPU is eligible for directed yield.
1696 * Most eligible candidate to yield is decided by following heuristics:
1698 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1699 * (preempted lock holder), indicated by @in_spin_loop.
1700 * Set at the beiginning and cleared at the end of interception/PLE handler.
1702 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1703 * chance last time (mostly it has become eligible now since we have probably
1704 * yielded to lockholder in last iteration. This is done by toggling
1705 * @dy_eligible each time a VCPU checked for eligibility.)
1707 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1708 * to preempted lock-holder could result in wrong VCPU selection and CPU
1709 * burning. Giving priority for a potential lock-holder increases lock
1712 * Since algorithm is based on heuristics, accessing another VCPU data without
1713 * locking does not harm. It may result in trying to yield to same VCPU, fail
1714 * and continue with next VCPU and so on.
1716 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1720 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1721 (vcpu
->spin_loop
.in_spin_loop
&&
1722 vcpu
->spin_loop
.dy_eligible
);
1724 if (vcpu
->spin_loop
.in_spin_loop
)
1725 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1730 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1732 struct kvm
*kvm
= me
->kvm
;
1733 struct kvm_vcpu
*vcpu
;
1734 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1739 kvm_vcpu_set_in_spin_loop(me
, true);
1741 * We boost the priority of a VCPU that is runnable but not
1742 * currently running, because it got preempted by something
1743 * else and called schedule in __vcpu_run. Hopefully that
1744 * VCPU is holding the lock that we need and will release it.
1745 * We approximate round-robin by starting at the last boosted VCPU.
1747 for (pass
= 0; pass
< 2 && !yielded
; pass
++) {
1748 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1749 if (!pass
&& i
<= last_boosted_vcpu
) {
1750 i
= last_boosted_vcpu
;
1752 } else if (pass
&& i
> last_boosted_vcpu
)
1756 if (waitqueue_active(&vcpu
->wq
))
1758 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1760 if (kvm_vcpu_yield_to(vcpu
)) {
1761 kvm
->last_boosted_vcpu
= i
;
1767 kvm_vcpu_set_in_spin_loop(me
, false);
1769 /* Ensure vcpu is not eligible during next spinloop */
1770 kvm_vcpu_set_dy_eligible(me
, false);
1772 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1774 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1776 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1779 if (vmf
->pgoff
== 0)
1780 page
= virt_to_page(vcpu
->run
);
1782 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1783 page
= virt_to_page(vcpu
->arch
.pio_data
);
1785 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1786 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1787 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1790 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1796 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1797 .fault
= kvm_vcpu_fault
,
1800 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1802 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1806 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1808 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1810 kvm_put_kvm(vcpu
->kvm
);
1814 static struct file_operations kvm_vcpu_fops
= {
1815 .release
= kvm_vcpu_release
,
1816 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1817 #ifdef CONFIG_COMPAT
1818 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1820 .mmap
= kvm_vcpu_mmap
,
1821 .llseek
= noop_llseek
,
1825 * Allocates an inode for the vcpu.
1827 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1829 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
);
1833 * Creates some virtual cpus. Good luck creating more than one.
1835 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1838 struct kvm_vcpu
*vcpu
, *v
;
1840 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1842 return PTR_ERR(vcpu
);
1844 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1846 r
= kvm_arch_vcpu_setup(vcpu
);
1850 mutex_lock(&kvm
->lock
);
1851 if (!kvm_vcpu_compatible(vcpu
)) {
1853 goto unlock_vcpu_destroy
;
1855 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1857 goto unlock_vcpu_destroy
;
1860 kvm_for_each_vcpu(r
, v
, kvm
)
1861 if (v
->vcpu_id
== id
) {
1863 goto unlock_vcpu_destroy
;
1866 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1868 /* Now it's all set up, let userspace reach it */
1870 r
= create_vcpu_fd(vcpu
);
1873 goto unlock_vcpu_destroy
;
1876 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1878 atomic_inc(&kvm
->online_vcpus
);
1880 mutex_unlock(&kvm
->lock
);
1881 kvm_arch_vcpu_postcreate(vcpu
);
1884 unlock_vcpu_destroy
:
1885 mutex_unlock(&kvm
->lock
);
1887 kvm_arch_vcpu_destroy(vcpu
);
1891 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1894 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1895 vcpu
->sigset_active
= 1;
1896 vcpu
->sigset
= *sigset
;
1898 vcpu
->sigset_active
= 0;
1902 static long kvm_vcpu_ioctl(struct file
*filp
,
1903 unsigned int ioctl
, unsigned long arg
)
1905 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1906 void __user
*argp
= (void __user
*)arg
;
1908 struct kvm_fpu
*fpu
= NULL
;
1909 struct kvm_sregs
*kvm_sregs
= NULL
;
1911 if (vcpu
->kvm
->mm
!= current
->mm
)
1914 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1916 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1917 * so vcpu_load() would break it.
1919 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1920 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1924 r
= vcpu_load(vcpu
);
1932 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
1933 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
1935 case KVM_GET_REGS
: {
1936 struct kvm_regs
*kvm_regs
;
1939 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
1942 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
1946 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
1953 case KVM_SET_REGS
: {
1954 struct kvm_regs
*kvm_regs
;
1957 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
1958 if (IS_ERR(kvm_regs
)) {
1959 r
= PTR_ERR(kvm_regs
);
1962 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
1966 case KVM_GET_SREGS
: {
1967 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
1971 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
1975 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
1980 case KVM_SET_SREGS
: {
1981 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
1982 if (IS_ERR(kvm_sregs
)) {
1983 r
= PTR_ERR(kvm_sregs
);
1987 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
1990 case KVM_GET_MP_STATE
: {
1991 struct kvm_mp_state mp_state
;
1993 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
1997 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2002 case KVM_SET_MP_STATE
: {
2003 struct kvm_mp_state mp_state
;
2006 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2008 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2011 case KVM_TRANSLATE
: {
2012 struct kvm_translation tr
;
2015 if (copy_from_user(&tr
, argp
, sizeof tr
))
2017 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2021 if (copy_to_user(argp
, &tr
, sizeof tr
))
2026 case KVM_SET_GUEST_DEBUG
: {
2027 struct kvm_guest_debug dbg
;
2030 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2032 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2035 case KVM_SET_SIGNAL_MASK
: {
2036 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2037 struct kvm_signal_mask kvm_sigmask
;
2038 sigset_t sigset
, *p
;
2043 if (copy_from_user(&kvm_sigmask
, argp
,
2044 sizeof kvm_sigmask
))
2047 if (kvm_sigmask
.len
!= sizeof sigset
)
2050 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2055 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2059 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2063 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2067 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2073 fpu
= memdup_user(argp
, sizeof(*fpu
));
2079 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2083 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2092 #ifdef CONFIG_COMPAT
2093 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2094 unsigned int ioctl
, unsigned long arg
)
2096 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2097 void __user
*argp
= compat_ptr(arg
);
2100 if (vcpu
->kvm
->mm
!= current
->mm
)
2104 case KVM_SET_SIGNAL_MASK
: {
2105 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2106 struct kvm_signal_mask kvm_sigmask
;
2107 compat_sigset_t csigset
;
2112 if (copy_from_user(&kvm_sigmask
, argp
,
2113 sizeof kvm_sigmask
))
2116 if (kvm_sigmask
.len
!= sizeof csigset
)
2119 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2122 sigset_from_compat(&sigset
, &csigset
);
2123 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2125 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2129 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2137 static long kvm_vm_ioctl(struct file
*filp
,
2138 unsigned int ioctl
, unsigned long arg
)
2140 struct kvm
*kvm
= filp
->private_data
;
2141 void __user
*argp
= (void __user
*)arg
;
2144 if (kvm
->mm
!= current
->mm
)
2147 case KVM_CREATE_VCPU
:
2148 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2150 case KVM_SET_USER_MEMORY_REGION
: {
2151 struct kvm_userspace_memory_region kvm_userspace_mem
;
2154 if (copy_from_user(&kvm_userspace_mem
, argp
,
2155 sizeof kvm_userspace_mem
))
2158 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, true);
2161 case KVM_GET_DIRTY_LOG
: {
2162 struct kvm_dirty_log log
;
2165 if (copy_from_user(&log
, argp
, sizeof log
))
2167 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2170 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2171 case KVM_REGISTER_COALESCED_MMIO
: {
2172 struct kvm_coalesced_mmio_zone zone
;
2174 if (copy_from_user(&zone
, argp
, sizeof zone
))
2176 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2179 case KVM_UNREGISTER_COALESCED_MMIO
: {
2180 struct kvm_coalesced_mmio_zone zone
;
2182 if (copy_from_user(&zone
, argp
, sizeof zone
))
2184 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2189 struct kvm_irqfd data
;
2192 if (copy_from_user(&data
, argp
, sizeof data
))
2194 r
= kvm_irqfd(kvm
, &data
);
2197 case KVM_IOEVENTFD
: {
2198 struct kvm_ioeventfd data
;
2201 if (copy_from_user(&data
, argp
, sizeof data
))
2203 r
= kvm_ioeventfd(kvm
, &data
);
2206 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2207 case KVM_SET_BOOT_CPU_ID
:
2209 mutex_lock(&kvm
->lock
);
2210 if (atomic_read(&kvm
->online_vcpus
) != 0)
2213 kvm
->bsp_vcpu_id
= arg
;
2214 mutex_unlock(&kvm
->lock
);
2217 #ifdef CONFIG_HAVE_KVM_MSI
2218 case KVM_SIGNAL_MSI
: {
2222 if (copy_from_user(&msi
, argp
, sizeof msi
))
2224 r
= kvm_send_userspace_msi(kvm
, &msi
);
2228 #ifdef __KVM_HAVE_IRQ_LINE
2229 case KVM_IRQ_LINE_STATUS
:
2230 case KVM_IRQ_LINE
: {
2231 struct kvm_irq_level irq_event
;
2234 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2237 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
);
2242 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2243 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2252 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2254 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2260 #ifdef CONFIG_COMPAT
2261 struct compat_kvm_dirty_log
{
2265 compat_uptr_t dirty_bitmap
; /* one bit per page */
2270 static long kvm_vm_compat_ioctl(struct file
*filp
,
2271 unsigned int ioctl
, unsigned long arg
)
2273 struct kvm
*kvm
= filp
->private_data
;
2276 if (kvm
->mm
!= current
->mm
)
2279 case KVM_GET_DIRTY_LOG
: {
2280 struct compat_kvm_dirty_log compat_log
;
2281 struct kvm_dirty_log log
;
2284 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2285 sizeof(compat_log
)))
2287 log
.slot
= compat_log
.slot
;
2288 log
.padding1
= compat_log
.padding1
;
2289 log
.padding2
= compat_log
.padding2
;
2290 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2292 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2296 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2304 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2306 struct page
*page
[1];
2309 gfn_t gfn
= vmf
->pgoff
;
2310 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2312 addr
= gfn_to_hva(kvm
, gfn
);
2313 if (kvm_is_error_hva(addr
))
2314 return VM_FAULT_SIGBUS
;
2316 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
2318 if (unlikely(npages
!= 1))
2319 return VM_FAULT_SIGBUS
;
2321 vmf
->page
= page
[0];
2325 static const struct vm_operations_struct kvm_vm_vm_ops
= {
2326 .fault
= kvm_vm_fault
,
2329 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2331 vma
->vm_ops
= &kvm_vm_vm_ops
;
2335 static struct file_operations kvm_vm_fops
= {
2336 .release
= kvm_vm_release
,
2337 .unlocked_ioctl
= kvm_vm_ioctl
,
2338 #ifdef CONFIG_COMPAT
2339 .compat_ioctl
= kvm_vm_compat_ioctl
,
2341 .mmap
= kvm_vm_mmap
,
2342 .llseek
= noop_llseek
,
2345 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2350 kvm
= kvm_create_vm(type
);
2352 return PTR_ERR(kvm
);
2353 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2354 r
= kvm_coalesced_mmio_init(kvm
);
2360 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
2367 static long kvm_dev_ioctl_check_extension_generic(long arg
)
2370 case KVM_CAP_USER_MEMORY
:
2371 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2372 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2373 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2374 case KVM_CAP_SET_BOOT_CPU_ID
:
2376 case KVM_CAP_INTERNAL_ERROR_DATA
:
2377 #ifdef CONFIG_HAVE_KVM_MSI
2378 case KVM_CAP_SIGNAL_MSI
:
2381 #ifdef KVM_CAP_IRQ_ROUTING
2382 case KVM_CAP_IRQ_ROUTING
:
2383 return KVM_MAX_IRQ_ROUTES
;
2388 return kvm_dev_ioctl_check_extension(arg
);
2391 static long kvm_dev_ioctl(struct file
*filp
,
2392 unsigned int ioctl
, unsigned long arg
)
2397 case KVM_GET_API_VERSION
:
2401 r
= KVM_API_VERSION
;
2404 r
= kvm_dev_ioctl_create_vm(arg
);
2406 case KVM_CHECK_EXTENSION
:
2407 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2409 case KVM_GET_VCPU_MMAP_SIZE
:
2413 r
= PAGE_SIZE
; /* struct kvm_run */
2415 r
+= PAGE_SIZE
; /* pio data page */
2417 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2418 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2421 case KVM_TRACE_ENABLE
:
2422 case KVM_TRACE_PAUSE
:
2423 case KVM_TRACE_DISABLE
:
2427 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2433 static struct file_operations kvm_chardev_ops
= {
2434 .unlocked_ioctl
= kvm_dev_ioctl
,
2435 .compat_ioctl
= kvm_dev_ioctl
,
2436 .llseek
= noop_llseek
,
2439 static struct miscdevice kvm_dev
= {
2445 static void hardware_enable_nolock(void *junk
)
2447 int cpu
= raw_smp_processor_id();
2450 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2453 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2455 r
= kvm_arch_hardware_enable(NULL
);
2458 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2459 atomic_inc(&hardware_enable_failed
);
2460 printk(KERN_INFO
"kvm: enabling virtualization on "
2461 "CPU%d failed\n", cpu
);
2465 static void hardware_enable(void *junk
)
2467 raw_spin_lock(&kvm_lock
);
2468 hardware_enable_nolock(junk
);
2469 raw_spin_unlock(&kvm_lock
);
2472 static void hardware_disable_nolock(void *junk
)
2474 int cpu
= raw_smp_processor_id();
2476 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2478 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2479 kvm_arch_hardware_disable(NULL
);
2482 static void hardware_disable(void *junk
)
2484 raw_spin_lock(&kvm_lock
);
2485 hardware_disable_nolock(junk
);
2486 raw_spin_unlock(&kvm_lock
);
2489 static void hardware_disable_all_nolock(void)
2491 BUG_ON(!kvm_usage_count
);
2494 if (!kvm_usage_count
)
2495 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2498 static void hardware_disable_all(void)
2500 raw_spin_lock(&kvm_lock
);
2501 hardware_disable_all_nolock();
2502 raw_spin_unlock(&kvm_lock
);
2505 static int hardware_enable_all(void)
2509 raw_spin_lock(&kvm_lock
);
2512 if (kvm_usage_count
== 1) {
2513 atomic_set(&hardware_enable_failed
, 0);
2514 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2516 if (atomic_read(&hardware_enable_failed
)) {
2517 hardware_disable_all_nolock();
2522 raw_spin_unlock(&kvm_lock
);
2527 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2532 if (!kvm_usage_count
)
2535 val
&= ~CPU_TASKS_FROZEN
;
2538 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2540 hardware_disable(NULL
);
2543 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2545 hardware_enable(NULL
);
2552 asmlinkage
void kvm_spurious_fault(void)
2554 /* Fault while not rebooting. We want the trace. */
2557 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
2559 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2563 * Some (well, at least mine) BIOSes hang on reboot if
2566 * And Intel TXT required VMX off for all cpu when system shutdown.
2568 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2569 kvm_rebooting
= true;
2570 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2574 static struct notifier_block kvm_reboot_notifier
= {
2575 .notifier_call
= kvm_reboot
,
2579 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2583 for (i
= 0; i
< bus
->dev_count
; i
++) {
2584 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2586 kvm_iodevice_destructor(pos
);
2591 int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2593 const struct kvm_io_range
*r1
= p1
;
2594 const struct kvm_io_range
*r2
= p2
;
2596 if (r1
->addr
< r2
->addr
)
2598 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2603 int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2604 gpa_t addr
, int len
)
2606 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2612 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2613 kvm_io_bus_sort_cmp
, NULL
);
2618 int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2619 gpa_t addr
, int len
)
2621 struct kvm_io_range
*range
, key
;
2624 key
= (struct kvm_io_range
) {
2629 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2630 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2634 off
= range
- bus
->range
;
2636 while (off
> 0 && kvm_io_bus_sort_cmp(&key
, &bus
->range
[off
-1]) == 0)
2642 /* kvm_io_bus_write - called under kvm->slots_lock */
2643 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2644 int len
, const void *val
)
2647 struct kvm_io_bus
*bus
;
2648 struct kvm_io_range range
;
2650 range
= (struct kvm_io_range
) {
2655 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2656 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2660 while (idx
< bus
->dev_count
&&
2661 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2662 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, addr
, len
, val
))
2670 /* kvm_io_bus_read - called under kvm->slots_lock */
2671 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2675 struct kvm_io_bus
*bus
;
2676 struct kvm_io_range range
;
2678 range
= (struct kvm_io_range
) {
2683 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2684 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2688 while (idx
< bus
->dev_count
&&
2689 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2690 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, addr
, len
, val
))
2698 /* Caller must hold slots_lock. */
2699 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2700 int len
, struct kvm_io_device
*dev
)
2702 struct kvm_io_bus
*new_bus
, *bus
;
2704 bus
= kvm
->buses
[bus_idx
];
2705 if (bus
->dev_count
> NR_IOBUS_DEVS
- 1)
2708 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
2709 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2712 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
2713 sizeof(struct kvm_io_range
)));
2714 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
2715 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2716 synchronize_srcu_expedited(&kvm
->srcu
);
2722 /* Caller must hold slots_lock. */
2723 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2724 struct kvm_io_device
*dev
)
2727 struct kvm_io_bus
*new_bus
, *bus
;
2729 bus
= kvm
->buses
[bus_idx
];
2731 for (i
= 0; i
< bus
->dev_count
; i
++)
2732 if (bus
->range
[i
].dev
== dev
) {
2740 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
2741 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2745 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
2746 new_bus
->dev_count
--;
2747 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
2748 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
2750 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2751 synchronize_srcu_expedited(&kvm
->srcu
);
2756 static struct notifier_block kvm_cpu_notifier
= {
2757 .notifier_call
= kvm_cpu_hotplug
,
2760 static int vm_stat_get(void *_offset
, u64
*val
)
2762 unsigned offset
= (long)_offset
;
2766 raw_spin_lock(&kvm_lock
);
2767 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2768 *val
+= *(u32
*)((void *)kvm
+ offset
);
2769 raw_spin_unlock(&kvm_lock
);
2773 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
2775 static int vcpu_stat_get(void *_offset
, u64
*val
)
2777 unsigned offset
= (long)_offset
;
2779 struct kvm_vcpu
*vcpu
;
2783 raw_spin_lock(&kvm_lock
);
2784 list_for_each_entry(kvm
, &vm_list
, vm_list
)
2785 kvm_for_each_vcpu(i
, vcpu
, kvm
)
2786 *val
+= *(u32
*)((void *)vcpu
+ offset
);
2788 raw_spin_unlock(&kvm_lock
);
2792 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
2794 static const struct file_operations
*stat_fops
[] = {
2795 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
2796 [KVM_STAT_VM
] = &vm_stat_fops
,
2799 static int kvm_init_debug(void)
2802 struct kvm_stats_debugfs_item
*p
;
2804 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
2805 if (kvm_debugfs_dir
== NULL
)
2808 for (p
= debugfs_entries
; p
->name
; ++p
) {
2809 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
2810 (void *)(long)p
->offset
,
2811 stat_fops
[p
->kind
]);
2812 if (p
->dentry
== NULL
)
2819 debugfs_remove_recursive(kvm_debugfs_dir
);
2824 static void kvm_exit_debug(void)
2826 struct kvm_stats_debugfs_item
*p
;
2828 for (p
= debugfs_entries
; p
->name
; ++p
)
2829 debugfs_remove(p
->dentry
);
2830 debugfs_remove(kvm_debugfs_dir
);
2833 static int kvm_suspend(void)
2835 if (kvm_usage_count
)
2836 hardware_disable_nolock(NULL
);
2840 static void kvm_resume(void)
2842 if (kvm_usage_count
) {
2843 WARN_ON(raw_spin_is_locked(&kvm_lock
));
2844 hardware_enable_nolock(NULL
);
2848 static struct syscore_ops kvm_syscore_ops
= {
2849 .suspend
= kvm_suspend
,
2850 .resume
= kvm_resume
,
2854 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
2856 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
2859 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
2861 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2863 kvm_arch_vcpu_load(vcpu
, cpu
);
2866 static void kvm_sched_out(struct preempt_notifier
*pn
,
2867 struct task_struct
*next
)
2869 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
2871 kvm_arch_vcpu_put(vcpu
);
2874 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
2875 struct module
*module
)
2880 r
= kvm_arch_init(opaque
);
2884 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
2889 r
= kvm_arch_hardware_setup();
2893 for_each_online_cpu(cpu
) {
2894 smp_call_function_single(cpu
,
2895 kvm_arch_check_processor_compat
,
2901 r
= register_cpu_notifier(&kvm_cpu_notifier
);
2904 register_reboot_notifier(&kvm_reboot_notifier
);
2906 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2908 vcpu_align
= __alignof__(struct kvm_vcpu
);
2909 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
2911 if (!kvm_vcpu_cache
) {
2916 r
= kvm_async_pf_init();
2920 kvm_chardev_ops
.owner
= module
;
2921 kvm_vm_fops
.owner
= module
;
2922 kvm_vcpu_fops
.owner
= module
;
2924 r
= misc_register(&kvm_dev
);
2926 printk(KERN_ERR
"kvm: misc device register failed\n");
2930 register_syscore_ops(&kvm_syscore_ops
);
2932 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
2933 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
2935 r
= kvm_init_debug();
2937 printk(KERN_ERR
"kvm: create debugfs files failed\n");
2944 unregister_syscore_ops(&kvm_syscore_ops
);
2946 kvm_async_pf_deinit();
2948 kmem_cache_destroy(kvm_vcpu_cache
);
2950 unregister_reboot_notifier(&kvm_reboot_notifier
);
2951 unregister_cpu_notifier(&kvm_cpu_notifier
);
2954 kvm_arch_hardware_unsetup();
2956 free_cpumask_var(cpus_hardware_enabled
);
2962 EXPORT_SYMBOL_GPL(kvm_init
);
2967 misc_deregister(&kvm_dev
);
2968 kmem_cache_destroy(kvm_vcpu_cache
);
2969 kvm_async_pf_deinit();
2970 unregister_syscore_ops(&kvm_syscore_ops
);
2971 unregister_reboot_notifier(&kvm_reboot_notifier
);
2972 unregister_cpu_notifier(&kvm_cpu_notifier
);
2973 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2974 kvm_arch_hardware_unsetup();
2976 free_cpumask_var(cpus_hardware_enabled
);
2978 EXPORT_SYMBOL_GPL(kvm_exit
);