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/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/kvm.h>
65 MODULE_AUTHOR("Qumranet");
66 MODULE_LICENSE("GPL");
71 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
74 DEFINE_RAW_SPINLOCK(kvm_lock
);
77 static cpumask_var_t cpus_hardware_enabled
;
78 static int kvm_usage_count
= 0;
79 static atomic_t hardware_enable_failed
;
81 struct kmem_cache
*kvm_vcpu_cache
;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
84 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
86 struct dentry
*kvm_debugfs_dir
;
88 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
91 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
97 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
100 EXPORT_SYMBOL_GPL(kvm_rebooting
);
102 static bool largepages_enabled
= true;
104 bool kvm_is_mmio_pfn(pfn_t pfn
)
106 if (pfn_valid(pfn
)) {
108 struct page
*tail
= pfn_to_page(pfn
);
109 struct page
*head
= compound_head(tail
);
110 reserved
= PageReserved(head
);
113 * "head" is not a dangling pointer
114 * (compound_head takes care of that)
115 * but the hugepage may have been splitted
116 * from under us (and we may not hold a
117 * reference count on the head page so it can
118 * be reused before we run PageReferenced), so
119 * we've to check PageTail before returning
126 return PageReserved(tail
);
133 * Switches to specified vcpu, until a matching vcpu_put()
135 int vcpu_load(struct kvm_vcpu
*vcpu
)
139 if (mutex_lock_killable(&vcpu
->mutex
))
141 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
142 /* The thread running this VCPU changed. */
143 struct pid
*oldpid
= vcpu
->pid
;
144 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
145 rcu_assign_pointer(vcpu
->pid
, newpid
);
150 preempt_notifier_register(&vcpu
->preempt_notifier
);
151 kvm_arch_vcpu_load(vcpu
, cpu
);
156 void vcpu_put(struct kvm_vcpu
*vcpu
)
159 kvm_arch_vcpu_put(vcpu
);
160 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
162 mutex_unlock(&vcpu
->mutex
);
165 static void ack_flush(void *_completed
)
169 static bool make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
174 struct kvm_vcpu
*vcpu
;
176 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
179 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
180 kvm_make_request(req
, vcpu
);
183 /* Set ->requests bit before we read ->mode */
186 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
187 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
188 cpumask_set_cpu(cpu
, cpus
);
190 if (unlikely(cpus
== NULL
))
191 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
192 else if (!cpumask_empty(cpus
))
193 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
197 free_cpumask_var(cpus
);
201 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
203 long dirty_count
= kvm
->tlbs_dirty
;
206 if (make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
207 ++kvm
->stat
.remote_tlb_flush
;
208 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
211 void kvm_reload_remote_mmus(struct kvm
*kvm
)
213 make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
216 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
218 make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
221 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
223 make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
226 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
231 mutex_init(&vcpu
->mutex
);
236 init_waitqueue_head(&vcpu
->wq
);
237 kvm_async_pf_vcpu_init(vcpu
);
239 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
244 vcpu
->run
= page_address(page
);
246 kvm_vcpu_set_in_spin_loop(vcpu
, false);
247 kvm_vcpu_set_dy_eligible(vcpu
, false);
248 vcpu
->preempted
= false;
250 r
= kvm_arch_vcpu_init(vcpu
);
256 free_page((unsigned long)vcpu
->run
);
260 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
262 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
265 kvm_arch_vcpu_uninit(vcpu
);
266 free_page((unsigned long)vcpu
->run
);
268 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
270 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
271 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
273 return container_of(mn
, struct kvm
, mmu_notifier
);
276 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
277 struct mm_struct
*mm
,
278 unsigned long address
)
280 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
281 int need_tlb_flush
, idx
;
284 * When ->invalidate_page runs, the linux pte has been zapped
285 * already but the page is still allocated until
286 * ->invalidate_page returns. So if we increase the sequence
287 * here the kvm page fault will notice if the spte can't be
288 * established because the page is going to be freed. If
289 * instead the kvm page fault establishes the spte before
290 * ->invalidate_page runs, kvm_unmap_hva will release it
293 * The sequence increase only need to be seen at spin_unlock
294 * time, and not at spin_lock time.
296 * Increasing the sequence after the spin_unlock would be
297 * unsafe because the kvm page fault could then establish the
298 * pte after kvm_unmap_hva returned, without noticing the page
299 * is going to be freed.
301 idx
= srcu_read_lock(&kvm
->srcu
);
302 spin_lock(&kvm
->mmu_lock
);
304 kvm
->mmu_notifier_seq
++;
305 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
306 /* we've to flush the tlb before the pages can be freed */
308 kvm_flush_remote_tlbs(kvm
);
310 spin_unlock(&kvm
->mmu_lock
);
311 srcu_read_unlock(&kvm
->srcu
, idx
);
314 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
315 struct mm_struct
*mm
,
316 unsigned long address
,
319 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
322 idx
= srcu_read_lock(&kvm
->srcu
);
323 spin_lock(&kvm
->mmu_lock
);
324 kvm
->mmu_notifier_seq
++;
325 kvm_set_spte_hva(kvm
, address
, pte
);
326 spin_unlock(&kvm
->mmu_lock
);
327 srcu_read_unlock(&kvm
->srcu
, idx
);
330 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
331 struct mm_struct
*mm
,
335 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
336 int need_tlb_flush
= 0, idx
;
338 idx
= srcu_read_lock(&kvm
->srcu
);
339 spin_lock(&kvm
->mmu_lock
);
341 * The count increase must become visible at unlock time as no
342 * spte can be established without taking the mmu_lock and
343 * count is also read inside the mmu_lock critical section.
345 kvm
->mmu_notifier_count
++;
346 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
347 need_tlb_flush
|= kvm
->tlbs_dirty
;
348 /* we've to flush the tlb before the pages can be freed */
350 kvm_flush_remote_tlbs(kvm
);
352 spin_unlock(&kvm
->mmu_lock
);
353 srcu_read_unlock(&kvm
->srcu
, idx
);
356 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
357 struct mm_struct
*mm
,
361 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
363 spin_lock(&kvm
->mmu_lock
);
365 * This sequence increase will notify the kvm page fault that
366 * the page that is going to be mapped in the spte could have
369 kvm
->mmu_notifier_seq
++;
372 * The above sequence increase must be visible before the
373 * below count decrease, which is ensured by the smp_wmb above
374 * in conjunction with the smp_rmb in mmu_notifier_retry().
376 kvm
->mmu_notifier_count
--;
377 spin_unlock(&kvm
->mmu_lock
);
379 BUG_ON(kvm
->mmu_notifier_count
< 0);
382 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
383 struct mm_struct
*mm
,
384 unsigned long address
)
386 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
389 idx
= srcu_read_lock(&kvm
->srcu
);
390 spin_lock(&kvm
->mmu_lock
);
392 young
= kvm_age_hva(kvm
, address
);
394 kvm_flush_remote_tlbs(kvm
);
396 spin_unlock(&kvm
->mmu_lock
);
397 srcu_read_unlock(&kvm
->srcu
, idx
);
402 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
403 struct mm_struct
*mm
,
404 unsigned long address
)
406 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
409 idx
= srcu_read_lock(&kvm
->srcu
);
410 spin_lock(&kvm
->mmu_lock
);
411 young
= kvm_test_age_hva(kvm
, address
);
412 spin_unlock(&kvm
->mmu_lock
);
413 srcu_read_unlock(&kvm
->srcu
, idx
);
418 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
419 struct mm_struct
*mm
)
421 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
424 idx
= srcu_read_lock(&kvm
->srcu
);
425 kvm_arch_flush_shadow_all(kvm
);
426 srcu_read_unlock(&kvm
->srcu
, idx
);
429 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
430 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
431 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
432 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
433 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
434 .test_young
= kvm_mmu_notifier_test_young
,
435 .change_pte
= kvm_mmu_notifier_change_pte
,
436 .release
= kvm_mmu_notifier_release
,
439 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
441 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
442 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
445 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
447 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
452 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
454 static void kvm_init_memslots_id(struct kvm
*kvm
)
457 struct kvm_memslots
*slots
= kvm
->memslots
;
459 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
460 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
463 static struct kvm
*kvm_create_vm(unsigned long type
)
466 struct kvm
*kvm
= kvm_arch_alloc_vm();
469 return ERR_PTR(-ENOMEM
);
471 spin_lock_init(&kvm
->mmu_lock
);
472 atomic_inc(¤t
->mm
->mm_count
);
473 kvm
->mm
= current
->mm
;
474 kvm_eventfd_init(kvm
);
475 mutex_init(&kvm
->lock
);
476 mutex_init(&kvm
->irq_lock
);
477 mutex_init(&kvm
->slots_lock
);
478 atomic_set(&kvm
->users_count
, 1);
479 INIT_LIST_HEAD(&kvm
->devices
);
481 r
= kvm_arch_init_vm(kvm
, type
);
483 goto out_err_nodisable
;
485 r
= hardware_enable_all();
487 goto out_err_nodisable
;
489 #ifdef CONFIG_HAVE_KVM_IRQCHIP
490 INIT_HLIST_HEAD(&kvm
->mask_notifier_list
);
491 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
494 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
497 kvm
->memslots
= kzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
500 kvm_init_memslots_id(kvm
);
501 if (init_srcu_struct(&kvm
->srcu
))
503 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
504 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
510 r
= kvm_init_mmu_notifier(kvm
);
514 raw_spin_lock(&kvm_lock
);
515 list_add(&kvm
->vm_list
, &vm_list
);
516 raw_spin_unlock(&kvm_lock
);
521 cleanup_srcu_struct(&kvm
->srcu
);
523 hardware_disable_all();
525 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
526 kfree(kvm
->buses
[i
]);
527 kfree(kvm
->memslots
);
528 kvm_arch_free_vm(kvm
);
534 * Avoid using vmalloc for a small buffer.
535 * Should not be used when the size is statically known.
537 void *kvm_kvzalloc(unsigned long size
)
539 if (size
> PAGE_SIZE
)
540 return vzalloc(size
);
542 return kzalloc(size
, GFP_KERNEL
);
545 void kvm_kvfree(const void *addr
)
547 if (is_vmalloc_addr(addr
))
553 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
555 if (!memslot
->dirty_bitmap
)
558 kvm_kvfree(memslot
->dirty_bitmap
);
559 memslot
->dirty_bitmap
= NULL
;
563 * Free any memory in @free but not in @dont.
565 static void kvm_free_physmem_slot(struct kvm_memory_slot
*free
,
566 struct kvm_memory_slot
*dont
)
568 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
569 kvm_destroy_dirty_bitmap(free
);
571 kvm_arch_free_memslot(free
, dont
);
576 void kvm_free_physmem(struct kvm
*kvm
)
578 struct kvm_memslots
*slots
= kvm
->memslots
;
579 struct kvm_memory_slot
*memslot
;
581 kvm_for_each_memslot(memslot
, slots
)
582 kvm_free_physmem_slot(memslot
, NULL
);
584 kfree(kvm
->memslots
);
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 raw_spin_lock(&kvm_lock
);
607 list_del(&kvm
->vm_list
);
608 raw_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 kvm_free_physmem(kvm
);
621 cleanup_srcu_struct(&kvm
->srcu
);
622 kvm_arch_free_vm(kvm
);
623 hardware_disable_all();
627 void kvm_get_kvm(struct kvm
*kvm
)
629 atomic_inc(&kvm
->users_count
);
631 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
633 void kvm_put_kvm(struct kvm
*kvm
)
635 if (atomic_dec_and_test(&kvm
->users_count
))
638 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
641 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
643 struct kvm
*kvm
= filp
->private_data
;
645 kvm_irqfd_release(kvm
);
652 * Allocation size is twice as large as the actual dirty bitmap size.
653 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
655 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
658 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
660 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
661 if (!memslot
->dirty_bitmap
)
664 #endif /* !CONFIG_S390 */
668 static int cmp_memslot(const void *slot1
, const void *slot2
)
670 struct kvm_memory_slot
*s1
, *s2
;
672 s1
= (struct kvm_memory_slot
*)slot1
;
673 s2
= (struct kvm_memory_slot
*)slot2
;
675 if (s1
->npages
< s2
->npages
)
677 if (s1
->npages
> s2
->npages
)
684 * Sort the memslots base on its size, so the larger slots
685 * will get better fit.
687 static void sort_memslots(struct kvm_memslots
*slots
)
691 sort(slots
->memslots
, KVM_MEM_SLOTS_NUM
,
692 sizeof(struct kvm_memory_slot
), cmp_memslot
, NULL
);
694 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
695 slots
->id_to_index
[slots
->memslots
[i
].id
] = i
;
698 void update_memslots(struct kvm_memslots
*slots
, struct kvm_memory_slot
*new,
703 struct kvm_memory_slot
*old
= id_to_memslot(slots
, id
);
704 unsigned long npages
= old
->npages
;
707 if (new->npages
!= npages
)
708 sort_memslots(slots
);
711 slots
->generation
= last_generation
+ 1;
714 static int check_memory_region_flags(struct kvm_userspace_memory_region
*mem
)
716 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
718 #ifdef KVM_CAP_READONLY_MEM
719 valid_flags
|= KVM_MEM_READONLY
;
722 if (mem
->flags
& ~valid_flags
)
728 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
729 struct kvm_memslots
*slots
, struct kvm_memory_slot
*new)
731 struct kvm_memslots
*old_memslots
= kvm
->memslots
;
733 update_memslots(slots
, new, kvm
->memslots
->generation
);
734 rcu_assign_pointer(kvm
->memslots
, slots
);
735 synchronize_srcu_expedited(&kvm
->srcu
);
740 * Allocate some memory and give it an address in the guest physical address
743 * Discontiguous memory is allowed, mostly for framebuffers.
745 * Must be called holding mmap_sem for write.
747 int __kvm_set_memory_region(struct kvm
*kvm
,
748 struct kvm_userspace_memory_region
*mem
)
752 unsigned long npages
;
753 struct kvm_memory_slot
*slot
;
754 struct kvm_memory_slot old
, new;
755 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
756 enum kvm_mr_change change
;
758 r
= check_memory_region_flags(mem
);
763 /* General sanity checks */
764 if (mem
->memory_size
& (PAGE_SIZE
- 1))
766 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
768 /* We can read the guest memory with __xxx_user() later on. */
769 if ((mem
->slot
< KVM_USER_MEM_SLOTS
) &&
770 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
771 !access_ok(VERIFY_WRITE
,
772 (void __user
*)(unsigned long)mem
->userspace_addr
,
775 if (mem
->slot
>= KVM_MEM_SLOTS_NUM
)
777 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
780 slot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
781 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
782 npages
= mem
->memory_size
>> PAGE_SHIFT
;
785 if (npages
> KVM_MEM_MAX_NR_PAGES
)
789 mem
->flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
794 new.base_gfn
= base_gfn
;
796 new.flags
= mem
->flags
;
801 change
= KVM_MR_CREATE
;
802 else { /* Modify an existing slot. */
803 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
804 (npages
!= old
.npages
) ||
805 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
808 if (base_gfn
!= old
.base_gfn
)
809 change
= KVM_MR_MOVE
;
810 else if (new.flags
!= old
.flags
)
811 change
= KVM_MR_FLAGS_ONLY
;
812 else { /* Nothing to change. */
817 } else if (old
.npages
) {
818 change
= KVM_MR_DELETE
;
819 } else /* Modify a non-existent slot: disallowed. */
822 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
823 /* Check for overlaps */
825 kvm_for_each_memslot(slot
, kvm
->memslots
) {
826 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
827 (slot
->id
== mem
->slot
))
829 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
830 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
835 /* Free page dirty bitmap if unneeded */
836 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
837 new.dirty_bitmap
= NULL
;
840 if (change
== KVM_MR_CREATE
) {
841 new.userspace_addr
= mem
->userspace_addr
;
843 if (kvm_arch_create_memslot(&new, npages
))
847 /* Allocate page dirty bitmap if needed */
848 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
849 if (kvm_create_dirty_bitmap(&new) < 0)
853 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
855 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
859 slot
= id_to_memslot(slots
, mem
->slot
);
860 slot
->flags
|= KVM_MEMSLOT_INVALID
;
862 old_memslots
= install_new_memslots(kvm
, slots
, NULL
);
864 /* slot was deleted or moved, clear iommu mapping */
865 kvm_iommu_unmap_pages(kvm
, &old
);
866 /* From this point no new shadow pages pointing to a deleted,
867 * or moved, memslot will be created.
869 * validation of sp->gfn happens in:
870 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
871 * - kvm_is_visible_gfn (mmu_check_roots)
873 kvm_arch_flush_shadow_memslot(kvm
, slot
);
874 slots
= old_memslots
;
877 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
883 * We can re-use the old_memslots from above, the only difference
884 * from the currently installed memslots is the invalid flag. This
885 * will get overwritten by update_memslots anyway.
888 slots
= kmemdup(kvm
->memslots
, sizeof(struct kvm_memslots
),
895 * IOMMU mapping: New slots need to be mapped. Old slots need to be
896 * un-mapped and re-mapped if their base changes. Since base change
897 * unmapping is handled above with slot deletion, mapping alone is
898 * needed here. Anything else the iommu might care about for existing
899 * slots (size changes, userspace addr changes and read-only flag
900 * changes) is disallowed above, so any other attribute changes getting
901 * here can be skipped.
903 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
904 r
= kvm_iommu_map_pages(kvm
, &new);
909 /* actual memory is freed via old in kvm_free_physmem_slot below */
910 if (change
== KVM_MR_DELETE
) {
911 new.dirty_bitmap
= NULL
;
912 memset(&new.arch
, 0, sizeof(new.arch
));
915 old_memslots
= install_new_memslots(kvm
, slots
, &new);
917 kvm_arch_commit_memory_region(kvm
, mem
, &old
, change
);
919 kvm_free_physmem_slot(&old
, &new);
927 kvm_free_physmem_slot(&new, &old
);
931 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
933 int kvm_set_memory_region(struct kvm
*kvm
,
934 struct kvm_userspace_memory_region
*mem
)
938 mutex_lock(&kvm
->slots_lock
);
939 r
= __kvm_set_memory_region(kvm
, mem
);
940 mutex_unlock(&kvm
->slots_lock
);
943 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
945 int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
946 struct kvm_userspace_memory_region
*mem
)
948 if (mem
->slot
>= KVM_USER_MEM_SLOTS
)
950 return kvm_set_memory_region(kvm
, mem
);
953 int kvm_get_dirty_log(struct kvm
*kvm
,
954 struct kvm_dirty_log
*log
, int *is_dirty
)
956 struct kvm_memory_slot
*memslot
;
959 unsigned long any
= 0;
962 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
965 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
967 if (!memslot
->dirty_bitmap
)
970 n
= kvm_dirty_bitmap_bytes(memslot
);
972 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
973 any
= memslot
->dirty_bitmap
[i
];
976 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
987 bool kvm_largepages_enabled(void)
989 return largepages_enabled
;
992 void kvm_disable_largepages(void)
994 largepages_enabled
= false;
996 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
998 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1000 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1002 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1004 int kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1006 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1008 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1009 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1014 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1016 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1018 struct vm_area_struct
*vma
;
1019 unsigned long addr
, size
;
1023 addr
= gfn_to_hva(kvm
, gfn
);
1024 if (kvm_is_error_hva(addr
))
1027 down_read(¤t
->mm
->mmap_sem
);
1028 vma
= find_vma(current
->mm
, addr
);
1032 size
= vma_kernel_pagesize(vma
);
1035 up_read(¤t
->mm
->mmap_sem
);
1040 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1042 return slot
->flags
& KVM_MEM_READONLY
;
1045 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1046 gfn_t
*nr_pages
, bool write
)
1048 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1049 return KVM_HVA_ERR_BAD
;
1051 if (memslot_is_readonly(slot
) && write
)
1052 return KVM_HVA_ERR_RO_BAD
;
1055 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1057 return __gfn_to_hva_memslot(slot
, gfn
);
1060 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1063 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1066 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1069 return gfn_to_hva_many(slot
, gfn
, NULL
);
1071 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1073 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1075 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1077 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1080 * The hva returned by this function is only allowed to be read.
1081 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1083 static unsigned long gfn_to_hva_read(struct kvm
*kvm
, gfn_t gfn
)
1085 return __gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
, false);
1088 static int kvm_read_hva(void *data
, void __user
*hva
, int len
)
1090 return __copy_from_user(data
, hva
, len
);
1093 static int kvm_read_hva_atomic(void *data
, void __user
*hva
, int len
)
1095 return __copy_from_user_inatomic(data
, hva
, len
);
1098 static int get_user_page_nowait(struct task_struct
*tsk
, struct mm_struct
*mm
,
1099 unsigned long start
, int write
, struct page
**page
)
1101 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1104 flags
|= FOLL_WRITE
;
1106 return __get_user_pages(tsk
, mm
, start
, 1, flags
, page
, NULL
, NULL
);
1109 static inline int check_user_page_hwpoison(unsigned long addr
)
1111 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1113 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1114 flags
, NULL
, NULL
, NULL
);
1115 return rc
== -EHWPOISON
;
1119 * The atomic path to get the writable pfn which will be stored in @pfn,
1120 * true indicates success, otherwise false is returned.
1122 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1123 bool write_fault
, bool *writable
, pfn_t
*pfn
)
1125 struct page
*page
[1];
1128 if (!(async
|| atomic
))
1132 * Fast pin a writable pfn only if it is a write fault request
1133 * or the caller allows to map a writable pfn for a read fault
1136 if (!(write_fault
|| writable
))
1139 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1141 *pfn
= page_to_pfn(page
[0]);
1152 * The slow path to get the pfn of the specified host virtual address,
1153 * 1 indicates success, -errno is returned if error is detected.
1155 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1156 bool *writable
, pfn_t
*pfn
)
1158 struct page
*page
[1];
1164 *writable
= write_fault
;
1167 down_read(¤t
->mm
->mmap_sem
);
1168 npages
= get_user_page_nowait(current
, current
->mm
,
1169 addr
, write_fault
, page
);
1170 up_read(¤t
->mm
->mmap_sem
);
1172 npages
= get_user_pages_fast(addr
, 1, write_fault
,
1177 /* map read fault as writable if possible */
1178 if (unlikely(!write_fault
) && writable
) {
1179 struct page
*wpage
[1];
1181 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1190 *pfn
= page_to_pfn(page
[0]);
1194 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1196 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1199 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1206 * Pin guest page in memory and return its pfn.
1207 * @addr: host virtual address which maps memory to the guest
1208 * @atomic: whether this function can sleep
1209 * @async: whether this function need to wait IO complete if the
1210 * host page is not in the memory
1211 * @write_fault: whether we should get a writable host page
1212 * @writable: whether it allows to map a writable host page for !@write_fault
1214 * The function will map a writable host page for these two cases:
1215 * 1): @write_fault = true
1216 * 2): @write_fault = false && @writable, @writable will tell the caller
1217 * whether the mapping is writable.
1219 static pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1220 bool write_fault
, bool *writable
)
1222 struct vm_area_struct
*vma
;
1226 /* we can do it either atomically or asynchronously, not both */
1227 BUG_ON(atomic
&& async
);
1229 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1233 return KVM_PFN_ERR_FAULT
;
1235 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1239 down_read(¤t
->mm
->mmap_sem
);
1240 if (npages
== -EHWPOISON
||
1241 (!async
&& check_user_page_hwpoison(addr
))) {
1242 pfn
= KVM_PFN_ERR_HWPOISON
;
1246 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1249 pfn
= KVM_PFN_ERR_FAULT
;
1250 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1251 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1253 BUG_ON(!kvm_is_mmio_pfn(pfn
));
1255 if (async
&& vma_is_valid(vma
, write_fault
))
1257 pfn
= KVM_PFN_ERR_FAULT
;
1260 up_read(¤t
->mm
->mmap_sem
);
1265 __gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
, bool atomic
,
1266 bool *async
, bool write_fault
, bool *writable
)
1268 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1270 if (addr
== KVM_HVA_ERR_RO_BAD
)
1271 return KVM_PFN_ERR_RO_FAULT
;
1273 if (kvm_is_error_hva(addr
))
1274 return KVM_PFN_NOSLOT
;
1276 /* Do not map writable pfn in the readonly memslot. */
1277 if (writable
&& memslot_is_readonly(slot
)) {
1282 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1286 static pfn_t
__gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
, bool atomic
, bool *async
,
1287 bool write_fault
, bool *writable
)
1289 struct kvm_memory_slot
*slot
;
1294 slot
= gfn_to_memslot(kvm
, gfn
);
1296 return __gfn_to_pfn_memslot(slot
, gfn
, atomic
, async
, write_fault
,
1300 pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1302 return __gfn_to_pfn(kvm
, gfn
, true, NULL
, true, NULL
);
1304 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1306 pfn_t
gfn_to_pfn_async(struct kvm
*kvm
, gfn_t gfn
, bool *async
,
1307 bool write_fault
, bool *writable
)
1309 return __gfn_to_pfn(kvm
, gfn
, false, async
, write_fault
, writable
);
1311 EXPORT_SYMBOL_GPL(gfn_to_pfn_async
);
1313 pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1315 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, true, NULL
);
1317 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1319 pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1322 return __gfn_to_pfn(kvm
, gfn
, false, NULL
, write_fault
, writable
);
1324 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1326 pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1328 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1331 pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1333 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1335 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1337 int gfn_to_page_many_atomic(struct kvm
*kvm
, gfn_t gfn
, struct page
**pages
,
1343 addr
= gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, &entry
);
1344 if (kvm_is_error_hva(addr
))
1347 if (entry
< nr_pages
)
1350 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1352 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1354 static struct page
*kvm_pfn_to_page(pfn_t pfn
)
1356 if (is_error_noslot_pfn(pfn
))
1357 return KVM_ERR_PTR_BAD_PAGE
;
1359 if (kvm_is_mmio_pfn(pfn
)) {
1361 return KVM_ERR_PTR_BAD_PAGE
;
1364 return pfn_to_page(pfn
);
1367 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1371 pfn
= gfn_to_pfn(kvm
, gfn
);
1373 return kvm_pfn_to_page(pfn
);
1376 EXPORT_SYMBOL_GPL(gfn_to_page
);
1378 void kvm_release_page_clean(struct page
*page
)
1380 WARN_ON(is_error_page(page
));
1382 kvm_release_pfn_clean(page_to_pfn(page
));
1384 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1386 void kvm_release_pfn_clean(pfn_t pfn
)
1388 if (!is_error_noslot_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
))
1389 put_page(pfn_to_page(pfn
));
1391 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1393 void kvm_release_page_dirty(struct page
*page
)
1395 WARN_ON(is_error_page(page
));
1397 kvm_release_pfn_dirty(page_to_pfn(page
));
1399 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1401 void kvm_release_pfn_dirty(pfn_t pfn
)
1403 kvm_set_pfn_dirty(pfn
);
1404 kvm_release_pfn_clean(pfn
);
1406 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1408 void kvm_set_page_dirty(struct page
*page
)
1410 kvm_set_pfn_dirty(page_to_pfn(page
));
1412 EXPORT_SYMBOL_GPL(kvm_set_page_dirty
);
1414 void kvm_set_pfn_dirty(pfn_t pfn
)
1416 if (!kvm_is_mmio_pfn(pfn
)) {
1417 struct page
*page
= pfn_to_page(pfn
);
1418 if (!PageReserved(page
))
1422 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1424 void kvm_set_pfn_accessed(pfn_t pfn
)
1426 if (!kvm_is_mmio_pfn(pfn
))
1427 mark_page_accessed(pfn_to_page(pfn
));
1429 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1431 void kvm_get_pfn(pfn_t pfn
)
1433 if (!kvm_is_mmio_pfn(pfn
))
1434 get_page(pfn_to_page(pfn
));
1436 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1438 static int next_segment(unsigned long len
, int offset
)
1440 if (len
> PAGE_SIZE
- offset
)
1441 return PAGE_SIZE
- offset
;
1446 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1452 addr
= gfn_to_hva_read(kvm
, gfn
);
1453 if (kvm_is_error_hva(addr
))
1455 r
= kvm_read_hva(data
, (void __user
*)addr
+ offset
, len
);
1460 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1462 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1464 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1466 int offset
= offset_in_page(gpa
);
1469 while ((seg
= next_segment(len
, offset
)) != 0) {
1470 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1480 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1482 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1487 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1488 int offset
= offset_in_page(gpa
);
1490 addr
= gfn_to_hva_read(kvm
, gfn
);
1491 if (kvm_is_error_hva(addr
))
1493 pagefault_disable();
1494 r
= kvm_read_hva_atomic(data
, (void __user
*)addr
+ offset
, len
);
1500 EXPORT_SYMBOL(kvm_read_guest_atomic
);
1502 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
, const void *data
,
1503 int offset
, int len
)
1508 addr
= gfn_to_hva(kvm
, gfn
);
1509 if (kvm_is_error_hva(addr
))
1511 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1514 mark_page_dirty(kvm
, gfn
);
1517 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1519 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1522 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1524 int offset
= offset_in_page(gpa
);
1527 while ((seg
= next_segment(len
, offset
)) != 0) {
1528 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1539 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1540 gpa_t gpa
, unsigned long len
)
1542 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1543 int offset
= offset_in_page(gpa
);
1544 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1545 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1546 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1547 gfn_t nr_pages_avail
;
1550 ghc
->generation
= slots
->generation
;
1552 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1553 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1554 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1558 * If the requested region crosses two memslots, we still
1559 * verify that the entire region is valid here.
1561 while (start_gfn
<= end_gfn
) {
1562 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1563 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1565 if (kvm_is_error_hva(ghc
->hva
))
1567 start_gfn
+= nr_pages_avail
;
1569 /* Use the slow path for cross page reads and writes. */
1570 ghc
->memslot
= NULL
;
1574 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1576 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1577 void *data
, unsigned long len
)
1579 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1582 BUG_ON(len
> ghc
->len
);
1584 if (slots
->generation
!= ghc
->generation
)
1585 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1587 if (unlikely(!ghc
->memslot
))
1588 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1590 if (kvm_is_error_hva(ghc
->hva
))
1593 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1596 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1600 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1602 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1603 void *data
, unsigned long len
)
1605 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1608 BUG_ON(len
> ghc
->len
);
1610 if (slots
->generation
!= ghc
->generation
)
1611 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1613 if (unlikely(!ghc
->memslot
))
1614 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1616 if (kvm_is_error_hva(ghc
->hva
))
1619 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1625 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1627 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1629 return kvm_write_guest_page(kvm
, gfn
, (const void *) empty_zero_page
,
1632 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1634 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1636 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1638 int offset
= offset_in_page(gpa
);
1641 while ((seg
= next_segment(len
, offset
)) != 0) {
1642 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1651 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1653 void mark_page_dirty_in_slot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
,
1656 if (memslot
&& memslot
->dirty_bitmap
) {
1657 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1659 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1663 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1665 struct kvm_memory_slot
*memslot
;
1667 memslot
= gfn_to_memslot(kvm
, gfn
);
1668 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
1672 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1674 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
1679 prepare_to_wait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1681 if (kvm_arch_vcpu_runnable(vcpu
)) {
1682 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1685 if (kvm_cpu_has_pending_timer(vcpu
))
1687 if (signal_pending(current
))
1693 finish_wait(&vcpu
->wq
, &wait
);
1698 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1700 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
1703 int cpu
= vcpu
->cpu
;
1704 wait_queue_head_t
*wqp
;
1706 wqp
= kvm_arch_vcpu_wq(vcpu
);
1707 if (waitqueue_active(wqp
)) {
1708 wake_up_interruptible(wqp
);
1709 ++vcpu
->stat
.halt_wakeup
;
1713 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
1714 if (kvm_arch_vcpu_should_kick(vcpu
))
1715 smp_send_reschedule(cpu
);
1718 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
1719 #endif /* !CONFIG_S390 */
1721 void kvm_resched(struct kvm_vcpu
*vcpu
)
1723 if (!need_resched())
1727 EXPORT_SYMBOL_GPL(kvm_resched
);
1729 bool kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
1732 struct task_struct
*task
= NULL
;
1736 pid
= rcu_dereference(target
->pid
);
1738 task
= get_pid_task(target
->pid
, PIDTYPE_PID
);
1742 if (task
->flags
& PF_VCPU
) {
1743 put_task_struct(task
);
1746 ret
= yield_to(task
, 1);
1747 put_task_struct(task
);
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
1753 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1755 * Helper that checks whether a VCPU is eligible for directed yield.
1756 * Most eligible candidate to yield is decided by following heuristics:
1758 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1759 * (preempted lock holder), indicated by @in_spin_loop.
1760 * Set at the beiginning and cleared at the end of interception/PLE handler.
1762 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1763 * chance last time (mostly it has become eligible now since we have probably
1764 * yielded to lockholder in last iteration. This is done by toggling
1765 * @dy_eligible each time a VCPU checked for eligibility.)
1767 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1768 * to preempted lock-holder could result in wrong VCPU selection and CPU
1769 * burning. Giving priority for a potential lock-holder increases lock
1772 * Since algorithm is based on heuristics, accessing another VCPU data without
1773 * locking does not harm. It may result in trying to yield to same VCPU, fail
1774 * and continue with next VCPU and so on.
1776 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
1780 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
1781 (vcpu
->spin_loop
.in_spin_loop
&&
1782 vcpu
->spin_loop
.dy_eligible
);
1784 if (vcpu
->spin_loop
.in_spin_loop
)
1785 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
1791 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
1793 struct kvm
*kvm
= me
->kvm
;
1794 struct kvm_vcpu
*vcpu
;
1795 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
1801 kvm_vcpu_set_in_spin_loop(me
, true);
1803 * We boost the priority of a VCPU that is runnable but not
1804 * currently running, because it got preempted by something
1805 * else and called schedule in __vcpu_run. Hopefully that
1806 * VCPU is holding the lock that we need and will release it.
1807 * We approximate round-robin by starting at the last boosted VCPU.
1809 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
1810 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1811 if (!pass
&& i
<= last_boosted_vcpu
) {
1812 i
= last_boosted_vcpu
;
1814 } else if (pass
&& i
> last_boosted_vcpu
)
1816 if (!ACCESS_ONCE(vcpu
->preempted
))
1820 if (waitqueue_active(&vcpu
->wq
))
1822 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
1825 yielded
= kvm_vcpu_yield_to(vcpu
);
1827 kvm
->last_boosted_vcpu
= i
;
1829 } else if (yielded
< 0) {
1836 kvm_vcpu_set_in_spin_loop(me
, false);
1838 /* Ensure vcpu is not eligible during next spinloop */
1839 kvm_vcpu_set_dy_eligible(me
, false);
1841 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
1843 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1845 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
1848 if (vmf
->pgoff
== 0)
1849 page
= virt_to_page(vcpu
->run
);
1851 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
1852 page
= virt_to_page(vcpu
->arch
.pio_data
);
1854 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1855 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
1856 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
1859 return kvm_arch_vcpu_fault(vcpu
, vmf
);
1865 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
1866 .fault
= kvm_vcpu_fault
,
1869 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1871 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
1875 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
1877 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1879 kvm_put_kvm(vcpu
->kvm
);
1883 static struct file_operations kvm_vcpu_fops
= {
1884 .release
= kvm_vcpu_release
,
1885 .unlocked_ioctl
= kvm_vcpu_ioctl
,
1886 #ifdef CONFIG_COMPAT
1887 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
1889 .mmap
= kvm_vcpu_mmap
,
1890 .llseek
= noop_llseek
,
1894 * Allocates an inode for the vcpu.
1896 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
1898 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
);
1902 * Creates some virtual cpus. Good luck creating more than one.
1904 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
1907 struct kvm_vcpu
*vcpu
, *v
;
1909 if (id
>= KVM_MAX_VCPUS
)
1912 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
1914 return PTR_ERR(vcpu
);
1916 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
1918 r
= kvm_arch_vcpu_setup(vcpu
);
1922 mutex_lock(&kvm
->lock
);
1923 if (!kvm_vcpu_compatible(vcpu
)) {
1925 goto unlock_vcpu_destroy
;
1927 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
1929 goto unlock_vcpu_destroy
;
1932 kvm_for_each_vcpu(r
, v
, kvm
)
1933 if (v
->vcpu_id
== id
) {
1935 goto unlock_vcpu_destroy
;
1938 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
1940 /* Now it's all set up, let userspace reach it */
1942 r
= create_vcpu_fd(vcpu
);
1945 goto unlock_vcpu_destroy
;
1948 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
1950 atomic_inc(&kvm
->online_vcpus
);
1952 mutex_unlock(&kvm
->lock
);
1953 kvm_arch_vcpu_postcreate(vcpu
);
1956 unlock_vcpu_destroy
:
1957 mutex_unlock(&kvm
->lock
);
1959 kvm_arch_vcpu_destroy(vcpu
);
1963 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
1966 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
1967 vcpu
->sigset_active
= 1;
1968 vcpu
->sigset
= *sigset
;
1970 vcpu
->sigset_active
= 0;
1974 static long kvm_vcpu_ioctl(struct file
*filp
,
1975 unsigned int ioctl
, unsigned long arg
)
1977 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1978 void __user
*argp
= (void __user
*)arg
;
1980 struct kvm_fpu
*fpu
= NULL
;
1981 struct kvm_sregs
*kvm_sregs
= NULL
;
1983 if (vcpu
->kvm
->mm
!= current
->mm
)
1986 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
1989 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1991 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1992 * so vcpu_load() would break it.
1994 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_INTERRUPT
)
1995 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
1999 r
= vcpu_load(vcpu
);
2007 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2008 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2010 case KVM_GET_REGS
: {
2011 struct kvm_regs
*kvm_regs
;
2014 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2017 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2021 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2028 case KVM_SET_REGS
: {
2029 struct kvm_regs
*kvm_regs
;
2032 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2033 if (IS_ERR(kvm_regs
)) {
2034 r
= PTR_ERR(kvm_regs
);
2037 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2041 case KVM_GET_SREGS
: {
2042 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2046 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2050 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2055 case KVM_SET_SREGS
: {
2056 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2057 if (IS_ERR(kvm_sregs
)) {
2058 r
= PTR_ERR(kvm_sregs
);
2062 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2065 case KVM_GET_MP_STATE
: {
2066 struct kvm_mp_state mp_state
;
2068 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2072 if (copy_to_user(argp
, &mp_state
, sizeof mp_state
))
2077 case KVM_SET_MP_STATE
: {
2078 struct kvm_mp_state mp_state
;
2081 if (copy_from_user(&mp_state
, argp
, sizeof mp_state
))
2083 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2086 case KVM_TRANSLATE
: {
2087 struct kvm_translation tr
;
2090 if (copy_from_user(&tr
, argp
, sizeof tr
))
2092 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2096 if (copy_to_user(argp
, &tr
, sizeof tr
))
2101 case KVM_SET_GUEST_DEBUG
: {
2102 struct kvm_guest_debug dbg
;
2105 if (copy_from_user(&dbg
, argp
, sizeof dbg
))
2107 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2110 case KVM_SET_SIGNAL_MASK
: {
2111 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2112 struct kvm_signal_mask kvm_sigmask
;
2113 sigset_t sigset
, *p
;
2118 if (copy_from_user(&kvm_sigmask
, argp
,
2119 sizeof kvm_sigmask
))
2122 if (kvm_sigmask
.len
!= sizeof sigset
)
2125 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2130 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2134 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2138 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2142 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2148 fpu
= memdup_user(argp
, sizeof(*fpu
));
2154 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2158 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2167 #ifdef CONFIG_COMPAT
2168 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2169 unsigned int ioctl
, unsigned long arg
)
2171 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2172 void __user
*argp
= compat_ptr(arg
);
2175 if (vcpu
->kvm
->mm
!= current
->mm
)
2179 case KVM_SET_SIGNAL_MASK
: {
2180 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2181 struct kvm_signal_mask kvm_sigmask
;
2182 compat_sigset_t csigset
;
2187 if (copy_from_user(&kvm_sigmask
, argp
,
2188 sizeof kvm_sigmask
))
2191 if (kvm_sigmask
.len
!= sizeof csigset
)
2194 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2197 sigset_from_compat(&sigset
, &csigset
);
2198 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2200 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2204 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2212 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2213 int (*accessor
)(struct kvm_device
*dev
,
2214 struct kvm_device_attr
*attr
),
2217 struct kvm_device_attr attr
;
2222 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2225 return accessor(dev
, &attr
);
2228 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2231 struct kvm_device
*dev
= filp
->private_data
;
2234 case KVM_SET_DEVICE_ATTR
:
2235 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2236 case KVM_GET_DEVICE_ATTR
:
2237 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2238 case KVM_HAS_DEVICE_ATTR
:
2239 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2241 if (dev
->ops
->ioctl
)
2242 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2248 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2250 struct kvm_device
*dev
= filp
->private_data
;
2251 struct kvm
*kvm
= dev
->kvm
;
2257 static const struct file_operations kvm_device_fops
= {
2258 .unlocked_ioctl
= kvm_device_ioctl
,
2259 #ifdef CONFIG_COMPAT
2260 .compat_ioctl
= kvm_device_ioctl
,
2262 .release
= kvm_device_release
,
2265 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2267 if (filp
->f_op
!= &kvm_device_fops
)
2270 return filp
->private_data
;
2273 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2274 struct kvm_create_device
*cd
)
2276 struct kvm_device_ops
*ops
= NULL
;
2277 struct kvm_device
*dev
;
2278 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2282 #ifdef CONFIG_KVM_MPIC
2283 case KVM_DEV_TYPE_FSL_MPIC_20
:
2284 case KVM_DEV_TYPE_FSL_MPIC_42
:
2285 ops
= &kvm_mpic_ops
;
2288 #ifdef CONFIG_KVM_XICS
2289 case KVM_DEV_TYPE_XICS
:
2290 ops
= &kvm_xics_ops
;
2300 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2307 ret
= ops
->create(dev
, cd
->type
);
2313 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
);
2319 list_add(&dev
->vm_node
, &kvm
->devices
);
2325 static long kvm_vm_ioctl(struct file
*filp
,
2326 unsigned int ioctl
, unsigned long arg
)
2328 struct kvm
*kvm
= filp
->private_data
;
2329 void __user
*argp
= (void __user
*)arg
;
2332 if (kvm
->mm
!= current
->mm
)
2335 case KVM_CREATE_VCPU
:
2336 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2338 case KVM_SET_USER_MEMORY_REGION
: {
2339 struct kvm_userspace_memory_region kvm_userspace_mem
;
2342 if (copy_from_user(&kvm_userspace_mem
, argp
,
2343 sizeof kvm_userspace_mem
))
2346 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2349 case KVM_GET_DIRTY_LOG
: {
2350 struct kvm_dirty_log log
;
2353 if (copy_from_user(&log
, argp
, sizeof log
))
2355 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2358 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2359 case KVM_REGISTER_COALESCED_MMIO
: {
2360 struct kvm_coalesced_mmio_zone zone
;
2362 if (copy_from_user(&zone
, argp
, sizeof zone
))
2364 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2367 case KVM_UNREGISTER_COALESCED_MMIO
: {
2368 struct kvm_coalesced_mmio_zone zone
;
2370 if (copy_from_user(&zone
, argp
, sizeof zone
))
2372 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2377 struct kvm_irqfd data
;
2380 if (copy_from_user(&data
, argp
, sizeof data
))
2382 r
= kvm_irqfd(kvm
, &data
);
2385 case KVM_IOEVENTFD
: {
2386 struct kvm_ioeventfd data
;
2389 if (copy_from_user(&data
, argp
, sizeof data
))
2391 r
= kvm_ioeventfd(kvm
, &data
);
2394 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2395 case KVM_SET_BOOT_CPU_ID
:
2397 mutex_lock(&kvm
->lock
);
2398 if (atomic_read(&kvm
->online_vcpus
) != 0)
2401 kvm
->bsp_vcpu_id
= arg
;
2402 mutex_unlock(&kvm
->lock
);
2405 #ifdef CONFIG_HAVE_KVM_MSI
2406 case KVM_SIGNAL_MSI
: {
2410 if (copy_from_user(&msi
, argp
, sizeof msi
))
2412 r
= kvm_send_userspace_msi(kvm
, &msi
);
2416 #ifdef __KVM_HAVE_IRQ_LINE
2417 case KVM_IRQ_LINE_STATUS
:
2418 case KVM_IRQ_LINE
: {
2419 struct kvm_irq_level irq_event
;
2422 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2425 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2426 ioctl
== KVM_IRQ_LINE_STATUS
);
2431 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2432 if (copy_to_user(argp
, &irq_event
, sizeof irq_event
))
2440 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2441 case KVM_SET_GSI_ROUTING
: {
2442 struct kvm_irq_routing routing
;
2443 struct kvm_irq_routing __user
*urouting
;
2444 struct kvm_irq_routing_entry
*entries
;
2447 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2450 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
2455 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2460 if (copy_from_user(entries
, urouting
->entries
,
2461 routing
.nr
* sizeof(*entries
)))
2462 goto out_free_irq_routing
;
2463 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2465 out_free_irq_routing
:
2469 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2470 case KVM_CREATE_DEVICE
: {
2471 struct kvm_create_device cd
;
2474 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2477 r
= kvm_ioctl_create_device(kvm
, &cd
);
2482 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2489 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2491 r
= kvm_vm_ioctl_assigned_device(kvm
, ioctl
, arg
);
2497 #ifdef CONFIG_COMPAT
2498 struct compat_kvm_dirty_log
{
2502 compat_uptr_t dirty_bitmap
; /* one bit per page */
2507 static long kvm_vm_compat_ioctl(struct file
*filp
,
2508 unsigned int ioctl
, unsigned long arg
)
2510 struct kvm
*kvm
= filp
->private_data
;
2513 if (kvm
->mm
!= current
->mm
)
2516 case KVM_GET_DIRTY_LOG
: {
2517 struct compat_kvm_dirty_log compat_log
;
2518 struct kvm_dirty_log log
;
2521 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2522 sizeof(compat_log
)))
2524 log
.slot
= compat_log
.slot
;
2525 log
.padding1
= compat_log
.padding1
;
2526 log
.padding2
= compat_log
.padding2
;
2527 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2529 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2533 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2541 static int kvm_vm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2543 struct page
*page
[1];
2546 gfn_t gfn
= vmf
->pgoff
;
2547 struct kvm
*kvm
= vma
->vm_file
->private_data
;
2549 addr
= gfn_to_hva(kvm
, gfn
);
2550 if (kvm_is_error_hva(addr
))
2551 return VM_FAULT_SIGBUS
;
2553 npages
= get_user_pages(current
, current
->mm
, addr
, 1, 1, 0, page
,
2555 if (unlikely(npages
!= 1))
2556 return VM_FAULT_SIGBUS
;
2558 vmf
->page
= page
[0];
2562 static const struct vm_operations_struct kvm_vm_vm_ops
= {
2563 .fault
= kvm_vm_fault
,
2566 static int kvm_vm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2568 vma
->vm_ops
= &kvm_vm_vm_ops
;
2572 static struct file_operations kvm_vm_fops
= {
2573 .release
= kvm_vm_release
,
2574 .unlocked_ioctl
= kvm_vm_ioctl
,
2575 #ifdef CONFIG_COMPAT
2576 .compat_ioctl
= kvm_vm_compat_ioctl
,
2578 .mmap
= kvm_vm_mmap
,
2579 .llseek
= noop_llseek
,
2582 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2587 kvm
= kvm_create_vm(type
);
2589 return PTR_ERR(kvm
);
2590 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2591 r
= kvm_coalesced_mmio_init(kvm
);
2597 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
2604 static long kvm_dev_ioctl_check_extension_generic(long arg
)
2607 case KVM_CAP_USER_MEMORY
:
2608 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2609 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2610 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2611 case KVM_CAP_SET_BOOT_CPU_ID
:
2613 case KVM_CAP_INTERNAL_ERROR_DATA
:
2614 #ifdef CONFIG_HAVE_KVM_MSI
2615 case KVM_CAP_SIGNAL_MSI
:
2617 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2618 case KVM_CAP_IRQFD_RESAMPLE
:
2621 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2622 case KVM_CAP_IRQ_ROUTING
:
2623 return KVM_MAX_IRQ_ROUTES
;
2628 return kvm_dev_ioctl_check_extension(arg
);
2631 static long kvm_dev_ioctl(struct file
*filp
,
2632 unsigned int ioctl
, unsigned long arg
)
2637 case KVM_GET_API_VERSION
:
2641 r
= KVM_API_VERSION
;
2644 r
= kvm_dev_ioctl_create_vm(arg
);
2646 case KVM_CHECK_EXTENSION
:
2647 r
= kvm_dev_ioctl_check_extension_generic(arg
);
2649 case KVM_GET_VCPU_MMAP_SIZE
:
2653 r
= PAGE_SIZE
; /* struct kvm_run */
2655 r
+= PAGE_SIZE
; /* pio data page */
2657 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2658 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
2661 case KVM_TRACE_ENABLE
:
2662 case KVM_TRACE_PAUSE
:
2663 case KVM_TRACE_DISABLE
:
2667 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
2673 static struct file_operations kvm_chardev_ops
= {
2674 .unlocked_ioctl
= kvm_dev_ioctl
,
2675 .compat_ioctl
= kvm_dev_ioctl
,
2676 .llseek
= noop_llseek
,
2679 static struct miscdevice kvm_dev
= {
2685 static void hardware_enable_nolock(void *junk
)
2687 int cpu
= raw_smp_processor_id();
2690 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2693 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
2695 r
= kvm_arch_hardware_enable(NULL
);
2698 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2699 atomic_inc(&hardware_enable_failed
);
2700 printk(KERN_INFO
"kvm: enabling virtualization on "
2701 "CPU%d failed\n", cpu
);
2705 static void hardware_enable(void *junk
)
2707 raw_spin_lock(&kvm_lock
);
2708 hardware_enable_nolock(junk
);
2709 raw_spin_unlock(&kvm_lock
);
2712 static void hardware_disable_nolock(void *junk
)
2714 int cpu
= raw_smp_processor_id();
2716 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
2718 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
2719 kvm_arch_hardware_disable(NULL
);
2722 static void hardware_disable(void *junk
)
2724 raw_spin_lock(&kvm_lock
);
2725 hardware_disable_nolock(junk
);
2726 raw_spin_unlock(&kvm_lock
);
2729 static void hardware_disable_all_nolock(void)
2731 BUG_ON(!kvm_usage_count
);
2734 if (!kvm_usage_count
)
2735 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2738 static void hardware_disable_all(void)
2740 raw_spin_lock(&kvm_lock
);
2741 hardware_disable_all_nolock();
2742 raw_spin_unlock(&kvm_lock
);
2745 static int hardware_enable_all(void)
2749 raw_spin_lock(&kvm_lock
);
2752 if (kvm_usage_count
== 1) {
2753 atomic_set(&hardware_enable_failed
, 0);
2754 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
2756 if (atomic_read(&hardware_enable_failed
)) {
2757 hardware_disable_all_nolock();
2762 raw_spin_unlock(&kvm_lock
);
2767 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
2772 if (!kvm_usage_count
)
2775 val
&= ~CPU_TASKS_FROZEN
;
2778 printk(KERN_INFO
"kvm: disabling virtualization on CPU%d\n",
2780 hardware_disable(NULL
);
2783 printk(KERN_INFO
"kvm: enabling virtualization on CPU%d\n",
2785 hardware_enable(NULL
);
2791 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
2795 * Some (well, at least mine) BIOSes hang on reboot if
2798 * And Intel TXT required VMX off for all cpu when system shutdown.
2800 printk(KERN_INFO
"kvm: exiting hardware virtualization\n");
2801 kvm_rebooting
= true;
2802 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
2806 static struct notifier_block kvm_reboot_notifier
= {
2807 .notifier_call
= kvm_reboot
,
2811 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
2815 for (i
= 0; i
< bus
->dev_count
; i
++) {
2816 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
2818 kvm_iodevice_destructor(pos
);
2823 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
2825 const struct kvm_io_range
*r1
= p1
;
2826 const struct kvm_io_range
*r2
= p2
;
2828 if (r1
->addr
< r2
->addr
)
2830 if (r1
->addr
+ r1
->len
> r2
->addr
+ r2
->len
)
2835 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
2836 gpa_t addr
, int len
)
2838 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
2844 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
2845 kvm_io_bus_sort_cmp
, NULL
);
2850 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
2851 gpa_t addr
, int len
)
2853 struct kvm_io_range
*range
, key
;
2856 key
= (struct kvm_io_range
) {
2861 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
2862 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
2866 off
= range
- bus
->range
;
2868 while (off
> 0 && kvm_io_bus_sort_cmp(&key
, &bus
->range
[off
-1]) == 0)
2874 /* kvm_io_bus_write - called under kvm->slots_lock */
2875 int kvm_io_bus_write(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2876 int len
, const void *val
)
2879 struct kvm_io_bus
*bus
;
2880 struct kvm_io_range range
;
2882 range
= (struct kvm_io_range
) {
2887 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2888 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2892 while (idx
< bus
->dev_count
&&
2893 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2894 if (!kvm_iodevice_write(bus
->range
[idx
].dev
, addr
, len
, val
))
2902 /* kvm_io_bus_read - called under kvm->slots_lock */
2903 int kvm_io_bus_read(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2907 struct kvm_io_bus
*bus
;
2908 struct kvm_io_range range
;
2910 range
= (struct kvm_io_range
) {
2915 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
2916 idx
= kvm_io_bus_get_first_dev(bus
, addr
, len
);
2920 while (idx
< bus
->dev_count
&&
2921 kvm_io_bus_sort_cmp(&range
, &bus
->range
[idx
]) == 0) {
2922 if (!kvm_iodevice_read(bus
->range
[idx
].dev
, addr
, len
, val
))
2930 /* Caller must hold slots_lock. */
2931 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
2932 int len
, struct kvm_io_device
*dev
)
2934 struct kvm_io_bus
*new_bus
, *bus
;
2936 bus
= kvm
->buses
[bus_idx
];
2937 if (bus
->dev_count
> NR_IOBUS_DEVS
- 1)
2940 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
2941 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2944 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
2945 sizeof(struct kvm_io_range
)));
2946 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
2947 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2948 synchronize_srcu_expedited(&kvm
->srcu
);
2954 /* Caller must hold slots_lock. */
2955 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
2956 struct kvm_io_device
*dev
)
2959 struct kvm_io_bus
*new_bus
, *bus
;
2961 bus
= kvm
->buses
[bus_idx
];
2963 for (i
= 0; i
< bus
->dev_count
; i
++)
2964 if (bus
->range
[i
].dev
== dev
) {
2972 new_bus
= kzalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
2973 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
2977 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
2978 new_bus
->dev_count
--;
2979 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
2980 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
2982 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
2983 synchronize_srcu_expedited(&kvm
->srcu
);
2988 static struct notifier_block kvm_cpu_notifier
= {
2989 .notifier_call
= kvm_cpu_hotplug
,
2992 static int vm_stat_get(void *_offset
, u64
*val
)
2994 unsigned offset
= (long)_offset
;
2998 raw_spin_lock(&kvm_lock
);
2999 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3000 *val
+= *(u32
*)((void *)kvm
+ offset
);
3001 raw_spin_unlock(&kvm_lock
);
3005 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3007 static int vcpu_stat_get(void *_offset
, u64
*val
)
3009 unsigned offset
= (long)_offset
;
3011 struct kvm_vcpu
*vcpu
;
3015 raw_spin_lock(&kvm_lock
);
3016 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3017 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3018 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3020 raw_spin_unlock(&kvm_lock
);
3024 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3026 static const struct file_operations
*stat_fops
[] = {
3027 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3028 [KVM_STAT_VM
] = &vm_stat_fops
,
3031 static int kvm_init_debug(void)
3034 struct kvm_stats_debugfs_item
*p
;
3036 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3037 if (kvm_debugfs_dir
== NULL
)
3040 for (p
= debugfs_entries
; p
->name
; ++p
) {
3041 p
->dentry
= debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3042 (void *)(long)p
->offset
,
3043 stat_fops
[p
->kind
]);
3044 if (p
->dentry
== NULL
)
3051 debugfs_remove_recursive(kvm_debugfs_dir
);
3056 static void kvm_exit_debug(void)
3058 struct kvm_stats_debugfs_item
*p
;
3060 for (p
= debugfs_entries
; p
->name
; ++p
)
3061 debugfs_remove(p
->dentry
);
3062 debugfs_remove(kvm_debugfs_dir
);
3065 static int kvm_suspend(void)
3067 if (kvm_usage_count
)
3068 hardware_disable_nolock(NULL
);
3072 static void kvm_resume(void)
3074 if (kvm_usage_count
) {
3075 WARN_ON(raw_spin_is_locked(&kvm_lock
));
3076 hardware_enable_nolock(NULL
);
3080 static struct syscore_ops kvm_syscore_ops
= {
3081 .suspend
= kvm_suspend
,
3082 .resume
= kvm_resume
,
3086 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3088 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3091 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3093 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3094 if (vcpu
->preempted
)
3095 vcpu
->preempted
= false;
3097 kvm_arch_vcpu_load(vcpu
, cpu
);
3100 static void kvm_sched_out(struct preempt_notifier
*pn
,
3101 struct task_struct
*next
)
3103 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3105 if (current
->state
== TASK_RUNNING
)
3106 vcpu
->preempted
= true;
3107 kvm_arch_vcpu_put(vcpu
);
3110 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3111 struct module
*module
)
3116 r
= kvm_arch_init(opaque
);
3121 * kvm_arch_init makes sure there's at most one caller
3122 * for architectures that support multiple implementations,
3123 * like intel and amd on x86.
3124 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3125 * conflicts in case kvm is already setup for another implementation.
3127 r
= kvm_irqfd_init();
3131 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3136 r
= kvm_arch_hardware_setup();
3140 for_each_online_cpu(cpu
) {
3141 smp_call_function_single(cpu
,
3142 kvm_arch_check_processor_compat
,
3148 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3151 register_reboot_notifier(&kvm_reboot_notifier
);
3153 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3155 vcpu_align
= __alignof__(struct kvm_vcpu
);
3156 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3158 if (!kvm_vcpu_cache
) {
3163 r
= kvm_async_pf_init();
3167 kvm_chardev_ops
.owner
= module
;
3168 kvm_vm_fops
.owner
= module
;
3169 kvm_vcpu_fops
.owner
= module
;
3171 r
= misc_register(&kvm_dev
);
3173 printk(KERN_ERR
"kvm: misc device register failed\n");
3177 register_syscore_ops(&kvm_syscore_ops
);
3179 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3180 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3182 r
= kvm_init_debug();
3184 printk(KERN_ERR
"kvm: create debugfs files failed\n");
3191 unregister_syscore_ops(&kvm_syscore_ops
);
3193 kvm_async_pf_deinit();
3195 kmem_cache_destroy(kvm_vcpu_cache
);
3197 unregister_reboot_notifier(&kvm_reboot_notifier
);
3198 unregister_cpu_notifier(&kvm_cpu_notifier
);
3201 kvm_arch_hardware_unsetup();
3203 free_cpumask_var(cpus_hardware_enabled
);
3211 EXPORT_SYMBOL_GPL(kvm_init
);
3216 misc_deregister(&kvm_dev
);
3217 kmem_cache_destroy(kvm_vcpu_cache
);
3218 kvm_async_pf_deinit();
3219 unregister_syscore_ops(&kvm_syscore_ops
);
3220 unregister_reboot_notifier(&kvm_reboot_notifier
);
3221 unregister_cpu_notifier(&kvm_cpu_notifier
);
3222 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3223 kvm_arch_hardware_unsetup();
3226 free_cpumask_var(cpus_hardware_enabled
);
3228 EXPORT_SYMBOL_GPL(kvm_exit
);