2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
76 module_param(halt_poll_ns
, uint
, S_IRUGO
| S_IWUSR
);
77 EXPORT_SYMBOL_GPL(halt_poll_ns
);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow
= 2;
81 module_param(halt_poll_ns_grow
, uint
, S_IRUGO
| S_IWUSR
);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink
;
86 module_param(halt_poll_ns_shrink
, uint
, S_IRUGO
| S_IWUSR
);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock
);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
99 static cpumask_var_t cpus_hardware_enabled
;
100 static int kvm_usage_count
;
101 static atomic_t hardware_enable_failed
;
103 struct kmem_cache
*kvm_vcpu_cache
;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
106 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
108 struct dentry
*kvm_debugfs_dir
;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
111 static int kvm_debugfs_num_entries
;
112 static const struct file_operations
*stat_fops_per_vm
[];
114 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
125 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
);
126 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
128 __visible
bool kvm_rebooting
;
129 EXPORT_SYMBOL_GPL(kvm_rebooting
);
131 static bool largepages_enabled
= true;
133 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
136 return PageReserved(pfn_to_page(pfn
));
142 * Switches to specified vcpu, until a matching vcpu_put()
144 int vcpu_load(struct kvm_vcpu
*vcpu
)
148 if (mutex_lock_killable(&vcpu
->mutex
))
151 preempt_notifier_register(&vcpu
->preempt_notifier
);
152 kvm_arch_vcpu_load(vcpu
, cpu
);
156 EXPORT_SYMBOL_GPL(vcpu_load
);
158 void vcpu_put(struct kvm_vcpu
*vcpu
)
161 kvm_arch_vcpu_put(vcpu
);
162 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
164 mutex_unlock(&vcpu
->mutex
);
166 EXPORT_SYMBOL_GPL(vcpu_put
);
168 static void ack_flush(void *_completed
)
172 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
177 struct kvm_vcpu
*vcpu
;
179 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
182 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
183 kvm_make_request(req
, vcpu
);
186 /* Set ->requests bit before we read ->mode. */
187 smp_mb__after_atomic();
189 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
190 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
191 cpumask_set_cpu(cpu
, cpus
);
193 if (unlikely(cpus
== NULL
))
194 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
195 else if (!cpumask_empty(cpus
))
196 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
200 free_cpumask_var(cpus
);
204 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
205 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
208 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
209 * kvm_make_all_cpus_request.
211 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
214 * We want to publish modifications to the page tables before reading
215 * mode. Pairs with a memory barrier in arch-specific code.
216 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
217 * and smp_mb in walk_shadow_page_lockless_begin/end.
218 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
220 * There is already an smp_mb__after_atomic() before
221 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
224 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
225 ++kvm
->stat
.remote_tlb_flush
;
226 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
228 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
231 void kvm_reload_remote_mmus(struct kvm
*kvm
)
233 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
236 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
241 mutex_init(&vcpu
->mutex
);
246 init_swait_queue_head(&vcpu
->wq
);
247 kvm_async_pf_vcpu_init(vcpu
);
250 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
252 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
257 vcpu
->run
= page_address(page
);
259 kvm_vcpu_set_in_spin_loop(vcpu
, false);
260 kvm_vcpu_set_dy_eligible(vcpu
, false);
261 vcpu
->preempted
= false;
263 r
= kvm_arch_vcpu_init(vcpu
);
269 free_page((unsigned long)vcpu
->run
);
273 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
275 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
278 kvm_arch_vcpu_uninit(vcpu
);
279 free_page((unsigned long)vcpu
->run
);
281 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
283 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
284 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
286 return container_of(mn
, struct kvm
, mmu_notifier
);
289 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
290 struct mm_struct
*mm
,
291 unsigned long address
)
293 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
294 int need_tlb_flush
, idx
;
297 * When ->invalidate_page runs, the linux pte has been zapped
298 * already but the page is still allocated until
299 * ->invalidate_page returns. So if we increase the sequence
300 * here the kvm page fault will notice if the spte can't be
301 * established because the page is going to be freed. If
302 * instead the kvm page fault establishes the spte before
303 * ->invalidate_page runs, kvm_unmap_hva will release it
306 * The sequence increase only need to be seen at spin_unlock
307 * time, and not at spin_lock time.
309 * Increasing the sequence after the spin_unlock would be
310 * unsafe because the kvm page fault could then establish the
311 * pte after kvm_unmap_hva returned, without noticing the page
312 * is going to be freed.
314 idx
= srcu_read_lock(&kvm
->srcu
);
315 spin_lock(&kvm
->mmu_lock
);
317 kvm
->mmu_notifier_seq
++;
318 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
319 /* we've to flush the tlb before the pages can be freed */
321 kvm_flush_remote_tlbs(kvm
);
323 spin_unlock(&kvm
->mmu_lock
);
325 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
327 srcu_read_unlock(&kvm
->srcu
, idx
);
330 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
331 struct mm_struct
*mm
,
332 unsigned long address
,
335 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
338 idx
= srcu_read_lock(&kvm
->srcu
);
339 spin_lock(&kvm
->mmu_lock
);
340 kvm
->mmu_notifier_seq
++;
341 kvm_set_spte_hva(kvm
, address
, pte
);
342 spin_unlock(&kvm
->mmu_lock
);
343 srcu_read_unlock(&kvm
->srcu
, idx
);
346 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
347 struct mm_struct
*mm
,
351 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
352 int need_tlb_flush
= 0, idx
;
354 idx
= srcu_read_lock(&kvm
->srcu
);
355 spin_lock(&kvm
->mmu_lock
);
357 * The count increase must become visible at unlock time as no
358 * spte can be established without taking the mmu_lock and
359 * count is also read inside the mmu_lock critical section.
361 kvm
->mmu_notifier_count
++;
362 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
363 need_tlb_flush
|= kvm
->tlbs_dirty
;
364 /* we've to flush the tlb before the pages can be freed */
366 kvm_flush_remote_tlbs(kvm
);
368 spin_unlock(&kvm
->mmu_lock
);
369 srcu_read_unlock(&kvm
->srcu
, idx
);
372 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
373 struct mm_struct
*mm
,
377 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
379 spin_lock(&kvm
->mmu_lock
);
381 * This sequence increase will notify the kvm page fault that
382 * the page that is going to be mapped in the spte could have
385 kvm
->mmu_notifier_seq
++;
388 * The above sequence increase must be visible before the
389 * below count decrease, which is ensured by the smp_wmb above
390 * in conjunction with the smp_rmb in mmu_notifier_retry().
392 kvm
->mmu_notifier_count
--;
393 spin_unlock(&kvm
->mmu_lock
);
395 BUG_ON(kvm
->mmu_notifier_count
< 0);
398 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
399 struct mm_struct
*mm
,
403 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
406 idx
= srcu_read_lock(&kvm
->srcu
);
407 spin_lock(&kvm
->mmu_lock
);
409 young
= kvm_age_hva(kvm
, start
, end
);
411 kvm_flush_remote_tlbs(kvm
);
413 spin_unlock(&kvm
->mmu_lock
);
414 srcu_read_unlock(&kvm
->srcu
, idx
);
419 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
420 struct mm_struct
*mm
,
424 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
427 idx
= srcu_read_lock(&kvm
->srcu
);
428 spin_lock(&kvm
->mmu_lock
);
430 * Even though we do not flush TLB, this will still adversely
431 * affect performance on pre-Haswell Intel EPT, where there is
432 * no EPT Access Bit to clear so that we have to tear down EPT
433 * tables instead. If we find this unacceptable, we can always
434 * add a parameter to kvm_age_hva so that it effectively doesn't
435 * do anything on clear_young.
437 * Also note that currently we never issue secondary TLB flushes
438 * from clear_young, leaving this job up to the regular system
439 * cadence. If we find this inaccurate, we might come up with a
440 * more sophisticated heuristic later.
442 young
= kvm_age_hva(kvm
, start
, end
);
443 spin_unlock(&kvm
->mmu_lock
);
444 srcu_read_unlock(&kvm
->srcu
, idx
);
449 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
450 struct mm_struct
*mm
,
451 unsigned long address
)
453 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
456 idx
= srcu_read_lock(&kvm
->srcu
);
457 spin_lock(&kvm
->mmu_lock
);
458 young
= kvm_test_age_hva(kvm
, address
);
459 spin_unlock(&kvm
->mmu_lock
);
460 srcu_read_unlock(&kvm
->srcu
, idx
);
465 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
466 struct mm_struct
*mm
)
468 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
471 idx
= srcu_read_lock(&kvm
->srcu
);
472 kvm_arch_flush_shadow_all(kvm
);
473 srcu_read_unlock(&kvm
->srcu
, idx
);
476 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
477 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
478 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
479 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
480 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
481 .clear_young
= kvm_mmu_notifier_clear_young
,
482 .test_young
= kvm_mmu_notifier_test_young
,
483 .change_pte
= kvm_mmu_notifier_change_pte
,
484 .release
= kvm_mmu_notifier_release
,
487 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
489 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
490 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
493 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
495 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
500 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
502 static struct kvm_memslots
*kvm_alloc_memslots(void)
505 struct kvm_memslots
*slots
;
507 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
511 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
512 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
517 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
519 if (!memslot
->dirty_bitmap
)
522 kvfree(memslot
->dirty_bitmap
);
523 memslot
->dirty_bitmap
= NULL
;
527 * Free any memory in @free but not in @dont.
529 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
530 struct kvm_memory_slot
*dont
)
532 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
533 kvm_destroy_dirty_bitmap(free
);
535 kvm_arch_free_memslot(kvm
, free
, dont
);
540 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
542 struct kvm_memory_slot
*memslot
;
547 kvm_for_each_memslot(memslot
, slots
)
548 kvm_free_memslot(kvm
, memslot
, NULL
);
553 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
557 if (!kvm
->debugfs_dentry
)
560 debugfs_remove_recursive(kvm
->debugfs_dentry
);
562 if (kvm
->debugfs_stat_data
) {
563 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
564 kfree(kvm
->debugfs_stat_data
[i
]);
565 kfree(kvm
->debugfs_stat_data
);
569 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
571 char dir_name
[ITOA_MAX_LEN
* 2];
572 struct kvm_stat_data
*stat_data
;
573 struct kvm_stats_debugfs_item
*p
;
575 if (!debugfs_initialized())
578 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
579 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
581 if (!kvm
->debugfs_dentry
)
584 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
585 sizeof(*kvm
->debugfs_stat_data
),
587 if (!kvm
->debugfs_stat_data
)
590 for (p
= debugfs_entries
; p
->name
; p
++) {
591 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
595 stat_data
->kvm
= kvm
;
596 stat_data
->offset
= p
->offset
;
597 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
598 if (!debugfs_create_file(p
->name
, 0644,
601 stat_fops_per_vm
[p
->kind
]))
607 static struct kvm
*kvm_create_vm(unsigned long type
)
610 struct kvm
*kvm
= kvm_arch_alloc_vm();
613 return ERR_PTR(-ENOMEM
);
615 spin_lock_init(&kvm
->mmu_lock
);
617 kvm
->mm
= current
->mm
;
618 kvm_eventfd_init(kvm
);
619 mutex_init(&kvm
->lock
);
620 mutex_init(&kvm
->irq_lock
);
621 mutex_init(&kvm
->slots_lock
);
622 refcount_set(&kvm
->users_count
, 1);
623 INIT_LIST_HEAD(&kvm
->devices
);
625 r
= kvm_arch_init_vm(kvm
, type
);
627 goto out_err_no_disable
;
629 r
= hardware_enable_all();
631 goto out_err_no_disable
;
633 #ifdef CONFIG_HAVE_KVM_IRQFD
634 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
637 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
640 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
641 struct kvm_memslots
*slots
= kvm_alloc_memslots();
643 goto out_err_no_srcu
;
645 * Generations must be different for each address space.
646 * Init kvm generation close to the maximum to easily test the
647 * code of handling generation number wrap-around.
649 slots
->generation
= i
* 2 - 150;
650 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
653 if (init_srcu_struct(&kvm
->srcu
))
654 goto out_err_no_srcu
;
655 if (init_srcu_struct(&kvm
->irq_srcu
))
656 goto out_err_no_irq_srcu
;
657 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
658 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
664 r
= kvm_init_mmu_notifier(kvm
);
668 spin_lock(&kvm_lock
);
669 list_add(&kvm
->vm_list
, &vm_list
);
670 spin_unlock(&kvm_lock
);
672 preempt_notifier_inc();
677 cleanup_srcu_struct(&kvm
->irq_srcu
);
679 cleanup_srcu_struct(&kvm
->srcu
);
681 hardware_disable_all();
683 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
684 kfree(kvm
->buses
[i
]);
685 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
686 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
687 kvm_arch_free_vm(kvm
);
693 * Avoid using vmalloc for a small buffer.
694 * Should not be used when the size is statically known.
696 void *kvm_kvzalloc(unsigned long size
)
698 if (size
> PAGE_SIZE
)
699 return vzalloc(size
);
701 return kzalloc(size
, GFP_KERNEL
);
704 static void kvm_destroy_devices(struct kvm
*kvm
)
706 struct kvm_device
*dev
, *tmp
;
709 * We do not need to take the kvm->lock here, because nobody else
710 * has a reference to the struct kvm at this point and therefore
711 * cannot access the devices list anyhow.
713 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
714 list_del(&dev
->vm_node
);
715 dev
->ops
->destroy(dev
);
719 static void kvm_destroy_vm(struct kvm
*kvm
)
722 struct mm_struct
*mm
= kvm
->mm
;
724 kvm_destroy_vm_debugfs(kvm
);
725 kvm_arch_sync_events(kvm
);
726 spin_lock(&kvm_lock
);
727 list_del(&kvm
->vm_list
);
728 spin_unlock(&kvm_lock
);
729 kvm_free_irq_routing(kvm
);
730 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
731 kvm_io_bus_destroy(kvm
->buses
[i
]);
732 kvm_coalesced_mmio_free(kvm
);
733 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
734 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
736 kvm_arch_flush_shadow_all(kvm
);
738 kvm_arch_destroy_vm(kvm
);
739 kvm_destroy_devices(kvm
);
740 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
741 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
742 cleanup_srcu_struct(&kvm
->irq_srcu
);
743 cleanup_srcu_struct(&kvm
->srcu
);
744 kvm_arch_free_vm(kvm
);
745 preempt_notifier_dec();
746 hardware_disable_all();
750 void kvm_get_kvm(struct kvm
*kvm
)
752 refcount_inc(&kvm
->users_count
);
754 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
756 void kvm_put_kvm(struct kvm
*kvm
)
758 if (refcount_dec_and_test(&kvm
->users_count
))
761 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
764 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
766 struct kvm
*kvm
= filp
->private_data
;
768 kvm_irqfd_release(kvm
);
775 * Allocation size is twice as large as the actual dirty bitmap size.
776 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
778 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
780 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
782 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
783 if (!memslot
->dirty_bitmap
)
790 * Insert memslot and re-sort memslots based on their GFN,
791 * so binary search could be used to lookup GFN.
792 * Sorting algorithm takes advantage of having initially
793 * sorted array and known changed memslot position.
795 static void update_memslots(struct kvm_memslots
*slots
,
796 struct kvm_memory_slot
*new)
799 int i
= slots
->id_to_index
[id
];
800 struct kvm_memory_slot
*mslots
= slots
->memslots
;
802 WARN_ON(mslots
[i
].id
!= id
);
804 WARN_ON(!mslots
[i
].npages
);
805 if (mslots
[i
].npages
)
808 if (!mslots
[i
].npages
)
812 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
813 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
814 if (!mslots
[i
+ 1].npages
)
816 mslots
[i
] = mslots
[i
+ 1];
817 slots
->id_to_index
[mslots
[i
].id
] = i
;
822 * The ">=" is needed when creating a slot with base_gfn == 0,
823 * so that it moves before all those with base_gfn == npages == 0.
825 * On the other hand, if new->npages is zero, the above loop has
826 * already left i pointing to the beginning of the empty part of
827 * mslots, and the ">=" would move the hole backwards in this
828 * case---which is wrong. So skip the loop when deleting a slot.
832 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
833 mslots
[i
] = mslots
[i
- 1];
834 slots
->id_to_index
[mslots
[i
].id
] = i
;
838 WARN_ON_ONCE(i
!= slots
->used_slots
);
841 slots
->id_to_index
[mslots
[i
].id
] = i
;
844 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
846 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
848 #ifdef __KVM_HAVE_READONLY_MEM
849 valid_flags
|= KVM_MEM_READONLY
;
852 if (mem
->flags
& ~valid_flags
)
858 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
859 int as_id
, struct kvm_memslots
*slots
)
861 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
864 * Set the low bit in the generation, which disables SPTE caching
865 * until the end of synchronize_srcu_expedited.
867 WARN_ON(old_memslots
->generation
& 1);
868 slots
->generation
= old_memslots
->generation
+ 1;
870 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
871 synchronize_srcu_expedited(&kvm
->srcu
);
874 * Increment the new memslot generation a second time. This prevents
875 * vm exits that race with memslot updates from caching a memslot
876 * generation that will (potentially) be valid forever.
878 * Generations must be unique even across address spaces. We do not need
879 * a global counter for that, instead the generation space is evenly split
880 * across address spaces. For example, with two address spaces, address
881 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
882 * use generations 2, 6, 10, 14, ...
884 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
886 kvm_arch_memslots_updated(kvm
, slots
);
892 * Allocate some memory and give it an address in the guest physical address
895 * Discontiguous memory is allowed, mostly for framebuffers.
897 * Must be called holding kvm->slots_lock for write.
899 int __kvm_set_memory_region(struct kvm
*kvm
,
900 const struct kvm_userspace_memory_region
*mem
)
904 unsigned long npages
;
905 struct kvm_memory_slot
*slot
;
906 struct kvm_memory_slot old
, new;
907 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
909 enum kvm_mr_change change
;
911 r
= check_memory_region_flags(mem
);
916 as_id
= mem
->slot
>> 16;
919 /* General sanity checks */
920 if (mem
->memory_size
& (PAGE_SIZE
- 1))
922 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
924 /* We can read the guest memory with __xxx_user() later on. */
925 if ((id
< KVM_USER_MEM_SLOTS
) &&
926 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
927 !access_ok(VERIFY_WRITE
,
928 (void __user
*)(unsigned long)mem
->userspace_addr
,
931 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
933 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
936 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
937 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
938 npages
= mem
->memory_size
>> PAGE_SHIFT
;
940 if (npages
> KVM_MEM_MAX_NR_PAGES
)
946 new.base_gfn
= base_gfn
;
948 new.flags
= mem
->flags
;
952 change
= KVM_MR_CREATE
;
953 else { /* Modify an existing slot. */
954 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
955 (npages
!= old
.npages
) ||
956 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
959 if (base_gfn
!= old
.base_gfn
)
960 change
= KVM_MR_MOVE
;
961 else if (new.flags
!= old
.flags
)
962 change
= KVM_MR_FLAGS_ONLY
;
963 else { /* Nothing to change. */
972 change
= KVM_MR_DELETE
;
977 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
978 /* Check for overlaps */
980 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
981 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
984 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
985 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
990 /* Free page dirty bitmap if unneeded */
991 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
992 new.dirty_bitmap
= NULL
;
995 if (change
== KVM_MR_CREATE
) {
996 new.userspace_addr
= mem
->userspace_addr
;
998 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1002 /* Allocate page dirty bitmap if needed */
1003 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1004 if (kvm_create_dirty_bitmap(&new) < 0)
1008 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
1011 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1013 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1014 slot
= id_to_memslot(slots
, id
);
1015 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1017 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1019 /* slot was deleted or moved, clear iommu mapping */
1020 kvm_iommu_unmap_pages(kvm
, &old
);
1021 /* From this point no new shadow pages pointing to a deleted,
1022 * or moved, memslot will be created.
1024 * validation of sp->gfn happens in:
1025 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1026 * - kvm_is_visible_gfn (mmu_check_roots)
1028 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1031 * We can re-use the old_memslots from above, the only difference
1032 * from the currently installed memslots is the invalid flag. This
1033 * will get overwritten by update_memslots anyway.
1035 slots
= old_memslots
;
1038 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1042 /* actual memory is freed via old in kvm_free_memslot below */
1043 if (change
== KVM_MR_DELETE
) {
1044 new.dirty_bitmap
= NULL
;
1045 memset(&new.arch
, 0, sizeof(new.arch
));
1048 update_memslots(slots
, &new);
1049 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1051 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1053 kvm_free_memslot(kvm
, &old
, &new);
1054 kvfree(old_memslots
);
1057 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1058 * un-mapped and re-mapped if their base changes. Since base change
1059 * unmapping is handled above with slot deletion, mapping alone is
1060 * needed here. Anything else the iommu might care about for existing
1061 * slots (size changes, userspace addr changes and read-only flag
1062 * changes) is disallowed above, so any other attribute changes getting
1063 * here can be skipped.
1065 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1066 r
= kvm_iommu_map_pages(kvm
, &new);
1075 kvm_free_memslot(kvm
, &new, &old
);
1079 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1081 int kvm_set_memory_region(struct kvm
*kvm
,
1082 const struct kvm_userspace_memory_region
*mem
)
1086 mutex_lock(&kvm
->slots_lock
);
1087 r
= __kvm_set_memory_region(kvm
, mem
);
1088 mutex_unlock(&kvm
->slots_lock
);
1091 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1093 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1094 struct kvm_userspace_memory_region
*mem
)
1096 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1099 return kvm_set_memory_region(kvm
, mem
);
1102 int kvm_get_dirty_log(struct kvm
*kvm
,
1103 struct kvm_dirty_log
*log
, int *is_dirty
)
1105 struct kvm_memslots
*slots
;
1106 struct kvm_memory_slot
*memslot
;
1109 unsigned long any
= 0;
1111 as_id
= log
->slot
>> 16;
1112 id
= (u16
)log
->slot
;
1113 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1116 slots
= __kvm_memslots(kvm
, as_id
);
1117 memslot
= id_to_memslot(slots
, id
);
1118 if (!memslot
->dirty_bitmap
)
1121 n
= kvm_dirty_bitmap_bytes(memslot
);
1123 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1124 any
= memslot
->dirty_bitmap
[i
];
1126 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1133 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1135 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1137 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1138 * are dirty write protect them for next write.
1139 * @kvm: pointer to kvm instance
1140 * @log: slot id and address to which we copy the log
1141 * @is_dirty: flag set if any page is dirty
1143 * We need to keep it in mind that VCPU threads can write to the bitmap
1144 * concurrently. So, to avoid losing track of dirty pages we keep the
1147 * 1. Take a snapshot of the bit and clear it if needed.
1148 * 2. Write protect the corresponding page.
1149 * 3. Copy the snapshot to the userspace.
1150 * 4. Upon return caller flushes TLB's if needed.
1152 * Between 2 and 4, the guest may write to the page using the remaining TLB
1153 * entry. This is not a problem because the page is reported dirty using
1154 * the snapshot taken before and step 4 ensures that writes done after
1155 * exiting to userspace will be logged for the next call.
1158 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1159 struct kvm_dirty_log
*log
, bool *is_dirty
)
1161 struct kvm_memslots
*slots
;
1162 struct kvm_memory_slot
*memslot
;
1165 unsigned long *dirty_bitmap
;
1166 unsigned long *dirty_bitmap_buffer
;
1168 as_id
= log
->slot
>> 16;
1169 id
= (u16
)log
->slot
;
1170 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1173 slots
= __kvm_memslots(kvm
, as_id
);
1174 memslot
= id_to_memslot(slots
, id
);
1176 dirty_bitmap
= memslot
->dirty_bitmap
;
1180 n
= kvm_dirty_bitmap_bytes(memslot
);
1182 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1183 memset(dirty_bitmap_buffer
, 0, n
);
1185 spin_lock(&kvm
->mmu_lock
);
1187 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1191 if (!dirty_bitmap
[i
])
1196 mask
= xchg(&dirty_bitmap
[i
], 0);
1197 dirty_bitmap_buffer
[i
] = mask
;
1200 offset
= i
* BITS_PER_LONG
;
1201 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1206 spin_unlock(&kvm
->mmu_lock
);
1207 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1211 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1214 bool kvm_largepages_enabled(void)
1216 return largepages_enabled
;
1219 void kvm_disable_largepages(void)
1221 largepages_enabled
= false;
1223 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1225 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1227 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1229 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1231 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1233 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1236 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1238 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1240 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1241 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1246 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1248 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1250 struct vm_area_struct
*vma
;
1251 unsigned long addr
, size
;
1255 addr
= gfn_to_hva(kvm
, gfn
);
1256 if (kvm_is_error_hva(addr
))
1259 down_read(¤t
->mm
->mmap_sem
);
1260 vma
= find_vma(current
->mm
, addr
);
1264 size
= vma_kernel_pagesize(vma
);
1267 up_read(¤t
->mm
->mmap_sem
);
1272 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1274 return slot
->flags
& KVM_MEM_READONLY
;
1277 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1278 gfn_t
*nr_pages
, bool write
)
1280 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1281 return KVM_HVA_ERR_BAD
;
1283 if (memslot_is_readonly(slot
) && write
)
1284 return KVM_HVA_ERR_RO_BAD
;
1287 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1289 return __gfn_to_hva_memslot(slot
, gfn
);
1292 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1295 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1298 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1301 return gfn_to_hva_many(slot
, gfn
, NULL
);
1303 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1305 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1307 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1309 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1311 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1313 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1315 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1318 * If writable is set to false, the hva returned by this function is only
1319 * allowed to be read.
1321 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1322 gfn_t gfn
, bool *writable
)
1324 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1326 if (!kvm_is_error_hva(hva
) && writable
)
1327 *writable
= !memslot_is_readonly(slot
);
1332 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1334 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1336 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1339 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1341 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1343 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1346 static int get_user_page_nowait(unsigned long start
, int write
,
1349 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1352 flags
|= FOLL_WRITE
;
1354 return get_user_pages(start
, 1, flags
, page
, NULL
);
1357 static inline int check_user_page_hwpoison(unsigned long addr
)
1359 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1361 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1362 return rc
== -EHWPOISON
;
1366 * The atomic path to get the writable pfn which will be stored in @pfn,
1367 * true indicates success, otherwise false is returned.
1369 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1370 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1372 struct page
*page
[1];
1375 if (!(async
|| atomic
))
1379 * Fast pin a writable pfn only if it is a write fault request
1380 * or the caller allows to map a writable pfn for a read fault
1383 if (!(write_fault
|| writable
))
1386 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1388 *pfn
= page_to_pfn(page
[0]);
1399 * The slow path to get the pfn of the specified host virtual address,
1400 * 1 indicates success, -errno is returned if error is detected.
1402 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1403 bool *writable
, kvm_pfn_t
*pfn
)
1405 struct page
*page
[1];
1411 *writable
= write_fault
;
1414 down_read(¤t
->mm
->mmap_sem
);
1415 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1416 up_read(¤t
->mm
->mmap_sem
);
1418 unsigned int flags
= FOLL_HWPOISON
;
1421 flags
|= FOLL_WRITE
;
1423 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1428 /* map read fault as writable if possible */
1429 if (unlikely(!write_fault
) && writable
) {
1430 struct page
*wpage
[1];
1432 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1441 *pfn
= page_to_pfn(page
[0]);
1445 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1447 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1450 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1456 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1457 unsigned long addr
, bool *async
,
1458 bool write_fault
, kvm_pfn_t
*p_pfn
)
1463 r
= follow_pfn(vma
, addr
, &pfn
);
1466 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1467 * not call the fault handler, so do it here.
1469 bool unlocked
= false;
1470 r
= fixup_user_fault(current
, current
->mm
, addr
,
1471 (write_fault
? FAULT_FLAG_WRITE
: 0),
1478 r
= follow_pfn(vma
, addr
, &pfn
);
1486 * Get a reference here because callers of *hva_to_pfn* and
1487 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1488 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1489 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1490 * simply do nothing for reserved pfns.
1492 * Whoever called remap_pfn_range is also going to call e.g.
1493 * unmap_mapping_range before the underlying pages are freed,
1494 * causing a call to our MMU notifier.
1503 * Pin guest page in memory and return its pfn.
1504 * @addr: host virtual address which maps memory to the guest
1505 * @atomic: whether this function can sleep
1506 * @async: whether this function need to wait IO complete if the
1507 * host page is not in the memory
1508 * @write_fault: whether we should get a writable host page
1509 * @writable: whether it allows to map a writable host page for !@write_fault
1511 * The function will map a writable host page for these two cases:
1512 * 1): @write_fault = true
1513 * 2): @write_fault = false && @writable, @writable will tell the caller
1514 * whether the mapping is writable.
1516 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1517 bool write_fault
, bool *writable
)
1519 struct vm_area_struct
*vma
;
1523 /* we can do it either atomically or asynchronously, not both */
1524 BUG_ON(atomic
&& async
);
1526 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1530 return KVM_PFN_ERR_FAULT
;
1532 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1536 down_read(¤t
->mm
->mmap_sem
);
1537 if (npages
== -EHWPOISON
||
1538 (!async
&& check_user_page_hwpoison(addr
))) {
1539 pfn
= KVM_PFN_ERR_HWPOISON
;
1544 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1547 pfn
= KVM_PFN_ERR_FAULT
;
1548 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1549 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, &pfn
);
1553 pfn
= KVM_PFN_ERR_FAULT
;
1555 if (async
&& vma_is_valid(vma
, write_fault
))
1557 pfn
= KVM_PFN_ERR_FAULT
;
1560 up_read(¤t
->mm
->mmap_sem
);
1564 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1565 bool atomic
, bool *async
, bool write_fault
,
1568 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1570 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1573 return KVM_PFN_ERR_RO_FAULT
;
1576 if (kvm_is_error_hva(addr
)) {
1579 return KVM_PFN_NOSLOT
;
1582 /* Do not map writable pfn in the readonly memslot. */
1583 if (writable
&& memslot_is_readonly(slot
)) {
1588 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1591 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1593 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1596 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1597 write_fault
, writable
);
1599 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1601 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1603 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1605 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1607 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1609 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1611 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1613 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1615 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1617 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1619 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1621 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1623 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1625 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1627 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1629 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1631 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1633 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1635 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1637 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1638 struct page
**pages
, int nr_pages
)
1643 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1644 if (kvm_is_error_hva(addr
))
1647 if (entry
< nr_pages
)
1650 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1652 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1654 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1656 if (is_error_noslot_pfn(pfn
))
1657 return KVM_ERR_PTR_BAD_PAGE
;
1659 if (kvm_is_reserved_pfn(pfn
)) {
1661 return KVM_ERR_PTR_BAD_PAGE
;
1664 return pfn_to_page(pfn
);
1667 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1671 pfn
= gfn_to_pfn(kvm
, gfn
);
1673 return kvm_pfn_to_page(pfn
);
1675 EXPORT_SYMBOL_GPL(gfn_to_page
);
1677 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1681 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1683 return kvm_pfn_to_page(pfn
);
1685 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1687 void kvm_release_page_clean(struct page
*page
)
1689 WARN_ON(is_error_page(page
));
1691 kvm_release_pfn_clean(page_to_pfn(page
));
1693 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1695 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1697 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1698 put_page(pfn_to_page(pfn
));
1700 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1702 void kvm_release_page_dirty(struct page
*page
)
1704 WARN_ON(is_error_page(page
));
1706 kvm_release_pfn_dirty(page_to_pfn(page
));
1708 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1710 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1712 kvm_set_pfn_dirty(pfn
);
1713 kvm_release_pfn_clean(pfn
);
1716 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1718 if (!kvm_is_reserved_pfn(pfn
)) {
1719 struct page
*page
= pfn_to_page(pfn
);
1721 if (!PageReserved(page
))
1725 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1727 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1729 if (!kvm_is_reserved_pfn(pfn
))
1730 mark_page_accessed(pfn_to_page(pfn
));
1732 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1734 void kvm_get_pfn(kvm_pfn_t pfn
)
1736 if (!kvm_is_reserved_pfn(pfn
))
1737 get_page(pfn_to_page(pfn
));
1739 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1741 static int next_segment(unsigned long len
, int offset
)
1743 if (len
> PAGE_SIZE
- offset
)
1744 return PAGE_SIZE
- offset
;
1749 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1750 void *data
, int offset
, int len
)
1755 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1756 if (kvm_is_error_hva(addr
))
1758 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1764 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1767 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1769 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1771 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1773 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1774 int offset
, int len
)
1776 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1778 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1780 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1782 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1784 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1786 int offset
= offset_in_page(gpa
);
1789 while ((seg
= next_segment(len
, offset
)) != 0) {
1790 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1800 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1802 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1804 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1806 int offset
= offset_in_page(gpa
);
1809 while ((seg
= next_segment(len
, offset
)) != 0) {
1810 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1820 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1822 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1823 void *data
, int offset
, unsigned long len
)
1828 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1829 if (kvm_is_error_hva(addr
))
1831 pagefault_disable();
1832 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1839 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1842 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1843 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1844 int offset
= offset_in_page(gpa
);
1846 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1848 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1850 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1851 void *data
, unsigned long len
)
1853 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1854 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1855 int offset
= offset_in_page(gpa
);
1857 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1859 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1861 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1862 const void *data
, int offset
, int len
)
1867 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1868 if (kvm_is_error_hva(addr
))
1870 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1873 mark_page_dirty_in_slot(memslot
, gfn
);
1877 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1878 const void *data
, int offset
, int len
)
1880 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1882 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1884 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1886 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1887 const void *data
, int offset
, int len
)
1889 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1891 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1893 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1895 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1898 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1900 int offset
= offset_in_page(gpa
);
1903 while ((seg
= next_segment(len
, offset
)) != 0) {
1904 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1914 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1916 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1919 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1921 int offset
= offset_in_page(gpa
);
1924 while ((seg
= next_segment(len
, offset
)) != 0) {
1925 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1935 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1937 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1938 struct gfn_to_hva_cache
*ghc
,
1939 gpa_t gpa
, unsigned long len
)
1941 int offset
= offset_in_page(gpa
);
1942 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1943 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1944 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1945 gfn_t nr_pages_avail
;
1948 ghc
->generation
= slots
->generation
;
1950 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1951 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1952 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1956 * If the requested region crosses two memslots, we still
1957 * verify that the entire region is valid here.
1959 while (start_gfn
<= end_gfn
) {
1960 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1961 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1963 if (kvm_is_error_hva(ghc
->hva
))
1965 start_gfn
+= nr_pages_avail
;
1967 /* Use the slow path for cross page reads and writes. */
1968 ghc
->memslot
= NULL
;
1973 int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
1974 gpa_t gpa
, unsigned long len
)
1976 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
1977 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1979 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init
);
1981 int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
1982 void *data
, int offset
, unsigned long len
)
1984 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
1986 gpa_t gpa
= ghc
->gpa
+ offset
;
1988 BUG_ON(len
+ offset
> ghc
->len
);
1990 if (slots
->generation
!= ghc
->generation
)
1991 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1993 if (unlikely(!ghc
->memslot
))
1994 return kvm_vcpu_write_guest(vcpu
, gpa
, data
, len
);
1996 if (kvm_is_error_hva(ghc
->hva
))
1999 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2002 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2006 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached
);
2008 int kvm_vcpu_write_guest_cached(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
2009 void *data
, unsigned long len
)
2011 return kvm_vcpu_write_guest_offset_cached(vcpu
, ghc
, data
, 0, len
);
2013 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached
);
2015 int kvm_vcpu_read_guest_cached(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
2016 void *data
, unsigned long len
)
2018 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2021 BUG_ON(len
> ghc
->len
);
2023 if (slots
->generation
!= ghc
->generation
)
2024 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2026 if (unlikely(!ghc
->memslot
))
2027 return kvm_vcpu_read_guest(vcpu
, ghc
->gpa
, data
, len
);
2029 if (kvm_is_error_hva(ghc
->hva
))
2032 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2038 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached
);
2040 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2042 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2044 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2046 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2048 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2050 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2052 int offset
= offset_in_page(gpa
);
2055 while ((seg
= next_segment(len
, offset
)) != 0) {
2056 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2065 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2067 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2070 if (memslot
&& memslot
->dirty_bitmap
) {
2071 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2073 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2077 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2079 struct kvm_memory_slot
*memslot
;
2081 memslot
= gfn_to_memslot(kvm
, gfn
);
2082 mark_page_dirty_in_slot(memslot
, gfn
);
2084 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2086 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2088 struct kvm_memory_slot
*memslot
;
2090 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2091 mark_page_dirty_in_slot(memslot
, gfn
);
2093 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2095 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2097 unsigned int old
, val
, grow
;
2099 old
= val
= vcpu
->halt_poll_ns
;
2100 grow
= READ_ONCE(halt_poll_ns_grow
);
2102 if (val
== 0 && grow
)
2107 if (val
> halt_poll_ns
)
2110 vcpu
->halt_poll_ns
= val
;
2111 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2114 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2116 unsigned int old
, val
, shrink
;
2118 old
= val
= vcpu
->halt_poll_ns
;
2119 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2125 vcpu
->halt_poll_ns
= val
;
2126 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2129 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2131 if (kvm_arch_vcpu_runnable(vcpu
)) {
2132 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2135 if (kvm_cpu_has_pending_timer(vcpu
))
2137 if (signal_pending(current
))
2144 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2146 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2149 DECLARE_SWAITQUEUE(wait
);
2150 bool waited
= false;
2153 start
= cur
= ktime_get();
2154 if (vcpu
->halt_poll_ns
) {
2155 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2157 ++vcpu
->stat
.halt_attempted_poll
;
2160 * This sets KVM_REQ_UNHALT if an interrupt
2163 if (kvm_vcpu_check_block(vcpu
) < 0) {
2164 ++vcpu
->stat
.halt_successful_poll
;
2165 if (!vcpu_valid_wakeup(vcpu
))
2166 ++vcpu
->stat
.halt_poll_invalid
;
2170 } while (single_task_running() && ktime_before(cur
, stop
));
2173 kvm_arch_vcpu_blocking(vcpu
);
2176 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2178 if (kvm_vcpu_check_block(vcpu
) < 0)
2185 finish_swait(&vcpu
->wq
, &wait
);
2188 kvm_arch_vcpu_unblocking(vcpu
);
2190 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2192 if (!vcpu_valid_wakeup(vcpu
))
2193 shrink_halt_poll_ns(vcpu
);
2194 else if (halt_poll_ns
) {
2195 if (block_ns
<= vcpu
->halt_poll_ns
)
2197 /* we had a long block, shrink polling */
2198 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2199 shrink_halt_poll_ns(vcpu
);
2200 /* we had a short halt and our poll time is too small */
2201 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2202 block_ns
< halt_poll_ns
)
2203 grow_halt_poll_ns(vcpu
);
2205 vcpu
->halt_poll_ns
= 0;
2207 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2208 kvm_arch_vcpu_block_finish(vcpu
);
2210 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2213 void kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2215 struct swait_queue_head
*wqp
;
2217 wqp
= kvm_arch_vcpu_wq(vcpu
);
2218 if (swait_active(wqp
)) {
2220 ++vcpu
->stat
.halt_wakeup
;
2224 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2227 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2229 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2232 int cpu
= vcpu
->cpu
;
2234 kvm_vcpu_wake_up(vcpu
);
2236 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2237 if (kvm_arch_vcpu_should_kick(vcpu
))
2238 smp_send_reschedule(cpu
);
2241 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2242 #endif /* !CONFIG_S390 */
2244 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2247 struct task_struct
*task
= NULL
;
2251 pid
= rcu_dereference(target
->pid
);
2253 task
= get_pid_task(pid
, PIDTYPE_PID
);
2257 ret
= yield_to(task
, 1);
2258 put_task_struct(task
);
2262 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2265 * Helper that checks whether a VCPU is eligible for directed yield.
2266 * Most eligible candidate to yield is decided by following heuristics:
2268 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2269 * (preempted lock holder), indicated by @in_spin_loop.
2270 * Set at the beiginning and cleared at the end of interception/PLE handler.
2272 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2273 * chance last time (mostly it has become eligible now since we have probably
2274 * yielded to lockholder in last iteration. This is done by toggling
2275 * @dy_eligible each time a VCPU checked for eligibility.)
2277 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2278 * to preempted lock-holder could result in wrong VCPU selection and CPU
2279 * burning. Giving priority for a potential lock-holder increases lock
2282 * Since algorithm is based on heuristics, accessing another VCPU data without
2283 * locking does not harm. It may result in trying to yield to same VCPU, fail
2284 * and continue with next VCPU and so on.
2286 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2288 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2291 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2292 vcpu
->spin_loop
.dy_eligible
;
2294 if (vcpu
->spin_loop
.in_spin_loop
)
2295 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2303 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2305 struct kvm
*kvm
= me
->kvm
;
2306 struct kvm_vcpu
*vcpu
;
2307 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2313 kvm_vcpu_set_in_spin_loop(me
, true);
2315 * We boost the priority of a VCPU that is runnable but not
2316 * currently running, because it got preempted by something
2317 * else and called schedule in __vcpu_run. Hopefully that
2318 * VCPU is holding the lock that we need and will release it.
2319 * We approximate round-robin by starting at the last boosted VCPU.
2321 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2322 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2323 if (!pass
&& i
<= last_boosted_vcpu
) {
2324 i
= last_boosted_vcpu
;
2326 } else if (pass
&& i
> last_boosted_vcpu
)
2328 if (!ACCESS_ONCE(vcpu
->preempted
))
2332 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2334 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2337 yielded
= kvm_vcpu_yield_to(vcpu
);
2339 kvm
->last_boosted_vcpu
= i
;
2341 } else if (yielded
< 0) {
2348 kvm_vcpu_set_in_spin_loop(me
, false);
2350 /* Ensure vcpu is not eligible during next spinloop */
2351 kvm_vcpu_set_dy_eligible(me
, false);
2353 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2355 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2357 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2360 if (vmf
->pgoff
== 0)
2361 page
= virt_to_page(vcpu
->run
);
2363 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2364 page
= virt_to_page(vcpu
->arch
.pio_data
);
2366 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2367 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2368 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2371 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2377 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2378 .fault
= kvm_vcpu_fault
,
2381 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2383 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2387 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2389 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2391 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2392 kvm_put_kvm(vcpu
->kvm
);
2396 static struct file_operations kvm_vcpu_fops
= {
2397 .release
= kvm_vcpu_release
,
2398 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2399 #ifdef CONFIG_KVM_COMPAT
2400 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2402 .mmap
= kvm_vcpu_mmap
,
2403 .llseek
= noop_llseek
,
2407 * Allocates an inode for the vcpu.
2409 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2411 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2414 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2416 char dir_name
[ITOA_MAX_LEN
* 2];
2419 if (!kvm_arch_has_vcpu_debugfs())
2422 if (!debugfs_initialized())
2425 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2426 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2427 vcpu
->kvm
->debugfs_dentry
);
2428 if (!vcpu
->debugfs_dentry
)
2431 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2433 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2441 * Creates some virtual cpus. Good luck creating more than one.
2443 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2446 struct kvm_vcpu
*vcpu
;
2448 if (id
>= KVM_MAX_VCPU_ID
)
2451 mutex_lock(&kvm
->lock
);
2452 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2453 mutex_unlock(&kvm
->lock
);
2457 kvm
->created_vcpus
++;
2458 mutex_unlock(&kvm
->lock
);
2460 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2463 goto vcpu_decrement
;
2466 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2468 r
= kvm_arch_vcpu_setup(vcpu
);
2472 r
= kvm_create_vcpu_debugfs(vcpu
);
2476 mutex_lock(&kvm
->lock
);
2477 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2479 goto unlock_vcpu_destroy
;
2482 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2484 /* Now it's all set up, let userspace reach it */
2486 r
= create_vcpu_fd(vcpu
);
2489 goto unlock_vcpu_destroy
;
2492 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2495 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2496 * before kvm->online_vcpu's incremented value.
2499 atomic_inc(&kvm
->online_vcpus
);
2501 mutex_unlock(&kvm
->lock
);
2502 kvm_arch_vcpu_postcreate(vcpu
);
2505 unlock_vcpu_destroy
:
2506 mutex_unlock(&kvm
->lock
);
2507 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2509 kvm_arch_vcpu_destroy(vcpu
);
2511 mutex_lock(&kvm
->lock
);
2512 kvm
->created_vcpus
--;
2513 mutex_unlock(&kvm
->lock
);
2517 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2520 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2521 vcpu
->sigset_active
= 1;
2522 vcpu
->sigset
= *sigset
;
2524 vcpu
->sigset_active
= 0;
2528 static long kvm_vcpu_ioctl(struct file
*filp
,
2529 unsigned int ioctl
, unsigned long arg
)
2531 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2532 void __user
*argp
= (void __user
*)arg
;
2534 struct kvm_fpu
*fpu
= NULL
;
2535 struct kvm_sregs
*kvm_sregs
= NULL
;
2537 if (vcpu
->kvm
->mm
!= current
->mm
)
2540 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2543 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2545 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2546 * so vcpu_load() would break it.
2548 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2549 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2553 r
= vcpu_load(vcpu
);
2561 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2562 /* The thread running this VCPU changed. */
2563 struct pid
*oldpid
= vcpu
->pid
;
2564 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2566 rcu_assign_pointer(vcpu
->pid
, newpid
);
2571 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2572 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2574 case KVM_GET_REGS
: {
2575 struct kvm_regs
*kvm_regs
;
2578 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2581 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2585 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2592 case KVM_SET_REGS
: {
2593 struct kvm_regs
*kvm_regs
;
2596 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2597 if (IS_ERR(kvm_regs
)) {
2598 r
= PTR_ERR(kvm_regs
);
2601 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2605 case KVM_GET_SREGS
: {
2606 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2610 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2614 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2619 case KVM_SET_SREGS
: {
2620 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2621 if (IS_ERR(kvm_sregs
)) {
2622 r
= PTR_ERR(kvm_sregs
);
2626 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2629 case KVM_GET_MP_STATE
: {
2630 struct kvm_mp_state mp_state
;
2632 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2636 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2641 case KVM_SET_MP_STATE
: {
2642 struct kvm_mp_state mp_state
;
2645 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2647 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2650 case KVM_TRANSLATE
: {
2651 struct kvm_translation tr
;
2654 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2656 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2660 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2665 case KVM_SET_GUEST_DEBUG
: {
2666 struct kvm_guest_debug dbg
;
2669 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2671 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2674 case KVM_SET_SIGNAL_MASK
: {
2675 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2676 struct kvm_signal_mask kvm_sigmask
;
2677 sigset_t sigset
, *p
;
2682 if (copy_from_user(&kvm_sigmask
, argp
,
2683 sizeof(kvm_sigmask
)))
2686 if (kvm_sigmask
.len
!= sizeof(sigset
))
2689 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2694 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2698 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2702 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2706 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2712 fpu
= memdup_user(argp
, sizeof(*fpu
));
2718 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2722 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2731 #ifdef CONFIG_KVM_COMPAT
2732 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2733 unsigned int ioctl
, unsigned long arg
)
2735 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2736 void __user
*argp
= compat_ptr(arg
);
2739 if (vcpu
->kvm
->mm
!= current
->mm
)
2743 case KVM_SET_SIGNAL_MASK
: {
2744 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2745 struct kvm_signal_mask kvm_sigmask
;
2746 compat_sigset_t csigset
;
2751 if (copy_from_user(&kvm_sigmask
, argp
,
2752 sizeof(kvm_sigmask
)))
2755 if (kvm_sigmask
.len
!= sizeof(csigset
))
2758 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2761 sigset_from_compat(&sigset
, &csigset
);
2762 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2764 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2768 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2776 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2777 int (*accessor
)(struct kvm_device
*dev
,
2778 struct kvm_device_attr
*attr
),
2781 struct kvm_device_attr attr
;
2786 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2789 return accessor(dev
, &attr
);
2792 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2795 struct kvm_device
*dev
= filp
->private_data
;
2798 case KVM_SET_DEVICE_ATTR
:
2799 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2800 case KVM_GET_DEVICE_ATTR
:
2801 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2802 case KVM_HAS_DEVICE_ATTR
:
2803 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2805 if (dev
->ops
->ioctl
)
2806 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2812 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2814 struct kvm_device
*dev
= filp
->private_data
;
2815 struct kvm
*kvm
= dev
->kvm
;
2821 static const struct file_operations kvm_device_fops
= {
2822 .unlocked_ioctl
= kvm_device_ioctl
,
2823 #ifdef CONFIG_KVM_COMPAT
2824 .compat_ioctl
= kvm_device_ioctl
,
2826 .release
= kvm_device_release
,
2829 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2831 if (filp
->f_op
!= &kvm_device_fops
)
2834 return filp
->private_data
;
2837 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2838 #ifdef CONFIG_KVM_MPIC
2839 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2840 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2843 #ifdef CONFIG_KVM_XICS
2844 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2848 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2850 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2853 if (kvm_device_ops_table
[type
] != NULL
)
2856 kvm_device_ops_table
[type
] = ops
;
2860 void kvm_unregister_device_ops(u32 type
)
2862 if (kvm_device_ops_table
[type
] != NULL
)
2863 kvm_device_ops_table
[type
] = NULL
;
2866 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2867 struct kvm_create_device
*cd
)
2869 struct kvm_device_ops
*ops
= NULL
;
2870 struct kvm_device
*dev
;
2871 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2874 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2877 ops
= kvm_device_ops_table
[cd
->type
];
2884 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2891 mutex_lock(&kvm
->lock
);
2892 ret
= ops
->create(dev
, cd
->type
);
2894 mutex_unlock(&kvm
->lock
);
2898 list_add(&dev
->vm_node
, &kvm
->devices
);
2899 mutex_unlock(&kvm
->lock
);
2904 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2906 mutex_lock(&kvm
->lock
);
2907 list_del(&dev
->vm_node
);
2908 mutex_unlock(&kvm
->lock
);
2918 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2921 case KVM_CAP_USER_MEMORY
:
2922 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2923 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2924 case KVM_CAP_INTERNAL_ERROR_DATA
:
2925 #ifdef CONFIG_HAVE_KVM_MSI
2926 case KVM_CAP_SIGNAL_MSI
:
2928 #ifdef CONFIG_HAVE_KVM_IRQFD
2930 case KVM_CAP_IRQFD_RESAMPLE
:
2932 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2933 case KVM_CAP_CHECK_EXTENSION_VM
:
2935 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2936 case KVM_CAP_IRQ_ROUTING
:
2937 return KVM_MAX_IRQ_ROUTES
;
2939 #if KVM_ADDRESS_SPACE_NUM > 1
2940 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2941 return KVM_ADDRESS_SPACE_NUM
;
2943 case KVM_CAP_MAX_VCPU_ID
:
2944 return KVM_MAX_VCPU_ID
;
2948 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2951 static long kvm_vm_ioctl(struct file
*filp
,
2952 unsigned int ioctl
, unsigned long arg
)
2954 struct kvm
*kvm
= filp
->private_data
;
2955 void __user
*argp
= (void __user
*)arg
;
2958 if (kvm
->mm
!= current
->mm
)
2961 case KVM_CREATE_VCPU
:
2962 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2964 case KVM_SET_USER_MEMORY_REGION
: {
2965 struct kvm_userspace_memory_region kvm_userspace_mem
;
2968 if (copy_from_user(&kvm_userspace_mem
, argp
,
2969 sizeof(kvm_userspace_mem
)))
2972 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2975 case KVM_GET_DIRTY_LOG
: {
2976 struct kvm_dirty_log log
;
2979 if (copy_from_user(&log
, argp
, sizeof(log
)))
2981 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2984 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2985 case KVM_REGISTER_COALESCED_MMIO
: {
2986 struct kvm_coalesced_mmio_zone zone
;
2989 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2991 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2994 case KVM_UNREGISTER_COALESCED_MMIO
: {
2995 struct kvm_coalesced_mmio_zone zone
;
2998 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3000 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3005 struct kvm_irqfd data
;
3008 if (copy_from_user(&data
, argp
, sizeof(data
)))
3010 r
= kvm_irqfd(kvm
, &data
);
3013 case KVM_IOEVENTFD
: {
3014 struct kvm_ioeventfd data
;
3017 if (copy_from_user(&data
, argp
, sizeof(data
)))
3019 r
= kvm_ioeventfd(kvm
, &data
);
3022 #ifdef CONFIG_HAVE_KVM_MSI
3023 case KVM_SIGNAL_MSI
: {
3027 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3029 r
= kvm_send_userspace_msi(kvm
, &msi
);
3033 #ifdef __KVM_HAVE_IRQ_LINE
3034 case KVM_IRQ_LINE_STATUS
:
3035 case KVM_IRQ_LINE
: {
3036 struct kvm_irq_level irq_event
;
3039 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3042 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3043 ioctl
== KVM_IRQ_LINE_STATUS
);
3048 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3049 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3057 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3058 case KVM_SET_GSI_ROUTING
: {
3059 struct kvm_irq_routing routing
;
3060 struct kvm_irq_routing __user
*urouting
;
3061 struct kvm_irq_routing_entry
*entries
= NULL
;
3064 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3067 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3073 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3078 if (copy_from_user(entries
, urouting
->entries
,
3079 routing
.nr
* sizeof(*entries
)))
3080 goto out_free_irq_routing
;
3082 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3084 out_free_irq_routing
:
3088 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3089 case KVM_CREATE_DEVICE
: {
3090 struct kvm_create_device cd
;
3093 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3096 r
= kvm_ioctl_create_device(kvm
, &cd
);
3101 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3107 case KVM_CHECK_EXTENSION
:
3108 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3111 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3117 #ifdef CONFIG_KVM_COMPAT
3118 struct compat_kvm_dirty_log
{
3122 compat_uptr_t dirty_bitmap
; /* one bit per page */
3127 static long kvm_vm_compat_ioctl(struct file
*filp
,
3128 unsigned int ioctl
, unsigned long arg
)
3130 struct kvm
*kvm
= filp
->private_data
;
3133 if (kvm
->mm
!= current
->mm
)
3136 case KVM_GET_DIRTY_LOG
: {
3137 struct compat_kvm_dirty_log compat_log
;
3138 struct kvm_dirty_log log
;
3140 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3141 sizeof(compat_log
)))
3143 log
.slot
= compat_log
.slot
;
3144 log
.padding1
= compat_log
.padding1
;
3145 log
.padding2
= compat_log
.padding2
;
3146 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3148 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3152 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3158 static struct file_operations kvm_vm_fops
= {
3159 .release
= kvm_vm_release
,
3160 .unlocked_ioctl
= kvm_vm_ioctl
,
3161 #ifdef CONFIG_KVM_COMPAT
3162 .compat_ioctl
= kvm_vm_compat_ioctl
,
3164 .llseek
= noop_llseek
,
3167 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3173 kvm
= kvm_create_vm(type
);
3175 return PTR_ERR(kvm
);
3176 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3177 r
= kvm_coalesced_mmio_init(kvm
);
3183 r
= get_unused_fd_flags(O_CLOEXEC
);
3188 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3192 return PTR_ERR(file
);
3195 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3201 fd_install(r
, file
);
3205 static long kvm_dev_ioctl(struct file
*filp
,
3206 unsigned int ioctl
, unsigned long arg
)
3211 case KVM_GET_API_VERSION
:
3214 r
= KVM_API_VERSION
;
3217 r
= kvm_dev_ioctl_create_vm(arg
);
3219 case KVM_CHECK_EXTENSION
:
3220 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3222 case KVM_GET_VCPU_MMAP_SIZE
:
3225 r
= PAGE_SIZE
; /* struct kvm_run */
3227 r
+= PAGE_SIZE
; /* pio data page */
3229 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3230 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3233 case KVM_TRACE_ENABLE
:
3234 case KVM_TRACE_PAUSE
:
3235 case KVM_TRACE_DISABLE
:
3239 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3245 static struct file_operations kvm_chardev_ops
= {
3246 .unlocked_ioctl
= kvm_dev_ioctl
,
3247 .compat_ioctl
= kvm_dev_ioctl
,
3248 .llseek
= noop_llseek
,
3251 static struct miscdevice kvm_dev
= {
3257 static void hardware_enable_nolock(void *junk
)
3259 int cpu
= raw_smp_processor_id();
3262 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3265 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3267 r
= kvm_arch_hardware_enable();
3270 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3271 atomic_inc(&hardware_enable_failed
);
3272 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3276 static int kvm_starting_cpu(unsigned int cpu
)
3278 raw_spin_lock(&kvm_count_lock
);
3279 if (kvm_usage_count
)
3280 hardware_enable_nolock(NULL
);
3281 raw_spin_unlock(&kvm_count_lock
);
3285 static void hardware_disable_nolock(void *junk
)
3287 int cpu
= raw_smp_processor_id();
3289 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3291 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3292 kvm_arch_hardware_disable();
3295 static int kvm_dying_cpu(unsigned int cpu
)
3297 raw_spin_lock(&kvm_count_lock
);
3298 if (kvm_usage_count
)
3299 hardware_disable_nolock(NULL
);
3300 raw_spin_unlock(&kvm_count_lock
);
3304 static void hardware_disable_all_nolock(void)
3306 BUG_ON(!kvm_usage_count
);
3309 if (!kvm_usage_count
)
3310 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3313 static void hardware_disable_all(void)
3315 raw_spin_lock(&kvm_count_lock
);
3316 hardware_disable_all_nolock();
3317 raw_spin_unlock(&kvm_count_lock
);
3320 static int hardware_enable_all(void)
3324 raw_spin_lock(&kvm_count_lock
);
3327 if (kvm_usage_count
== 1) {
3328 atomic_set(&hardware_enable_failed
, 0);
3329 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3331 if (atomic_read(&hardware_enable_failed
)) {
3332 hardware_disable_all_nolock();
3337 raw_spin_unlock(&kvm_count_lock
);
3342 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3346 * Some (well, at least mine) BIOSes hang on reboot if
3349 * And Intel TXT required VMX off for all cpu when system shutdown.
3351 pr_info("kvm: exiting hardware virtualization\n");
3352 kvm_rebooting
= true;
3353 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3357 static struct notifier_block kvm_reboot_notifier
= {
3358 .notifier_call
= kvm_reboot
,
3362 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3366 for (i
= 0; i
< bus
->dev_count
; i
++) {
3367 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3369 kvm_iodevice_destructor(pos
);
3374 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3375 const struct kvm_io_range
*r2
)
3377 gpa_t addr1
= r1
->addr
;
3378 gpa_t addr2
= r2
->addr
;
3383 /* If r2->len == 0, match the exact address. If r2->len != 0,
3384 * accept any overlapping write. Any order is acceptable for
3385 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3386 * we process all of them.
3399 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3401 return kvm_io_bus_cmp(p1
, p2
);
3404 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3405 gpa_t addr
, int len
)
3407 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3413 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3414 kvm_io_bus_sort_cmp
, NULL
);
3419 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3420 gpa_t addr
, int len
)
3422 struct kvm_io_range
*range
, key
;
3425 key
= (struct kvm_io_range
) {
3430 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3431 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3435 off
= range
- bus
->range
;
3437 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3443 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3444 struct kvm_io_range
*range
, const void *val
)
3448 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3452 while (idx
< bus
->dev_count
&&
3453 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3454 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3463 /* kvm_io_bus_write - called under kvm->slots_lock */
3464 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3465 int len
, const void *val
)
3467 struct kvm_io_bus
*bus
;
3468 struct kvm_io_range range
;
3471 range
= (struct kvm_io_range
) {
3476 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3477 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3478 return r
< 0 ? r
: 0;
3481 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3482 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3483 gpa_t addr
, int len
, const void *val
, long cookie
)
3485 struct kvm_io_bus
*bus
;
3486 struct kvm_io_range range
;
3488 range
= (struct kvm_io_range
) {
3493 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3495 /* First try the device referenced by cookie. */
3496 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3497 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3498 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3503 * cookie contained garbage; fall back to search and return the
3504 * correct cookie value.
3506 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3509 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3510 struct kvm_io_range
*range
, void *val
)
3514 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3518 while (idx
< bus
->dev_count
&&
3519 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3520 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3528 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3530 /* kvm_io_bus_read - called under kvm->slots_lock */
3531 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3534 struct kvm_io_bus
*bus
;
3535 struct kvm_io_range range
;
3538 range
= (struct kvm_io_range
) {
3543 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3544 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3545 return r
< 0 ? r
: 0;
3549 /* Caller must hold slots_lock. */
3550 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3551 int len
, struct kvm_io_device
*dev
)
3553 struct kvm_io_bus
*new_bus
, *bus
;
3555 bus
= kvm
->buses
[bus_idx
];
3556 /* exclude ioeventfd which is limited by maximum fd */
3557 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3560 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3561 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3564 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3565 sizeof(struct kvm_io_range
)));
3566 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3567 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3568 synchronize_srcu_expedited(&kvm
->srcu
);
3574 /* Caller must hold slots_lock. */
3575 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3576 struct kvm_io_device
*dev
)
3579 struct kvm_io_bus
*new_bus
, *bus
;
3581 bus
= kvm
->buses
[bus_idx
];
3583 for (i
= 0; i
< bus
->dev_count
; i
++)
3584 if (bus
->range
[i
].dev
== dev
) {
3592 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3593 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3597 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3598 new_bus
->dev_count
--;
3599 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3600 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3602 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3603 synchronize_srcu_expedited(&kvm
->srcu
);
3608 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3611 struct kvm_io_bus
*bus
;
3612 int dev_idx
, srcu_idx
;
3613 struct kvm_io_device
*iodev
= NULL
;
3615 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3617 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3619 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3623 iodev
= bus
->range
[dev_idx
].dev
;
3626 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3630 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3632 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3633 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3636 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3639 /* The debugfs files are a reference to the kvm struct which
3640 * is still valid when kvm_destroy_vm is called.
3641 * To avoid the race between open and the removal of the debugfs
3642 * directory we test against the users count.
3644 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3647 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3648 kvm_put_kvm(stat_data
->kvm
);
3655 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3657 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3660 simple_attr_release(inode
, file
);
3661 kvm_put_kvm(stat_data
->kvm
);
3666 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3668 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3670 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3675 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3677 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3682 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3687 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3689 __simple_attr_check_format("%llu\n", 0ull);
3690 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3691 vm_stat_clear_per_vm
, "%llu\n");
3694 static const struct file_operations vm_stat_get_per_vm_fops
= {
3695 .owner
= THIS_MODULE
,
3696 .open
= vm_stat_get_per_vm_open
,
3697 .release
= kvm_debugfs_release
,
3698 .read
= simple_attr_read
,
3699 .write
= simple_attr_write
,
3700 .llseek
= generic_file_llseek
,
3703 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3706 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3707 struct kvm_vcpu
*vcpu
;
3711 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3712 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3717 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3720 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3721 struct kvm_vcpu
*vcpu
;
3726 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3727 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3732 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3734 __simple_attr_check_format("%llu\n", 0ull);
3735 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3736 vcpu_stat_clear_per_vm
, "%llu\n");
3739 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3740 .owner
= THIS_MODULE
,
3741 .open
= vcpu_stat_get_per_vm_open
,
3742 .release
= kvm_debugfs_release
,
3743 .read
= simple_attr_read
,
3744 .write
= simple_attr_write
,
3745 .llseek
= generic_file_llseek
,
3748 static const struct file_operations
*stat_fops_per_vm
[] = {
3749 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3750 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3753 static int vm_stat_get(void *_offset
, u64
*val
)
3755 unsigned offset
= (long)_offset
;
3757 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3761 spin_lock(&kvm_lock
);
3762 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3764 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3767 spin_unlock(&kvm_lock
);
3771 static int vm_stat_clear(void *_offset
, u64 val
)
3773 unsigned offset
= (long)_offset
;
3775 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3780 spin_lock(&kvm_lock
);
3781 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3783 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3785 spin_unlock(&kvm_lock
);
3790 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3792 static int vcpu_stat_get(void *_offset
, u64
*val
)
3794 unsigned offset
= (long)_offset
;
3796 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3800 spin_lock(&kvm_lock
);
3801 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3803 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3806 spin_unlock(&kvm_lock
);
3810 static int vcpu_stat_clear(void *_offset
, u64 val
)
3812 unsigned offset
= (long)_offset
;
3814 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3819 spin_lock(&kvm_lock
);
3820 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3822 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3824 spin_unlock(&kvm_lock
);
3829 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3832 static const struct file_operations
*stat_fops
[] = {
3833 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3834 [KVM_STAT_VM
] = &vm_stat_fops
,
3837 static int kvm_init_debug(void)
3840 struct kvm_stats_debugfs_item
*p
;
3842 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3843 if (kvm_debugfs_dir
== NULL
)
3846 kvm_debugfs_num_entries
= 0;
3847 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3848 if (!debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3849 (void *)(long)p
->offset
,
3850 stat_fops
[p
->kind
]))
3857 debugfs_remove_recursive(kvm_debugfs_dir
);
3862 static int kvm_suspend(void)
3864 if (kvm_usage_count
)
3865 hardware_disable_nolock(NULL
);
3869 static void kvm_resume(void)
3871 if (kvm_usage_count
) {
3872 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3873 hardware_enable_nolock(NULL
);
3877 static struct syscore_ops kvm_syscore_ops
= {
3878 .suspend
= kvm_suspend
,
3879 .resume
= kvm_resume
,
3883 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3885 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3888 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3890 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3892 if (vcpu
->preempted
)
3893 vcpu
->preempted
= false;
3895 kvm_arch_sched_in(vcpu
, cpu
);
3897 kvm_arch_vcpu_load(vcpu
, cpu
);
3900 static void kvm_sched_out(struct preempt_notifier
*pn
,
3901 struct task_struct
*next
)
3903 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3905 if (current
->state
== TASK_RUNNING
)
3906 vcpu
->preempted
= true;
3907 kvm_arch_vcpu_put(vcpu
);
3910 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3911 struct module
*module
)
3916 r
= kvm_arch_init(opaque
);
3921 * kvm_arch_init makes sure there's at most one caller
3922 * for architectures that support multiple implementations,
3923 * like intel and amd on x86.
3924 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3925 * conflicts in case kvm is already setup for another implementation.
3927 r
= kvm_irqfd_init();
3931 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3936 r
= kvm_arch_hardware_setup();
3940 for_each_online_cpu(cpu
) {
3941 smp_call_function_single(cpu
,
3942 kvm_arch_check_processor_compat
,
3948 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
3949 kvm_starting_cpu
, kvm_dying_cpu
);
3952 register_reboot_notifier(&kvm_reboot_notifier
);
3954 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3956 vcpu_align
= __alignof__(struct kvm_vcpu
);
3957 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3959 if (!kvm_vcpu_cache
) {
3964 r
= kvm_async_pf_init();
3968 kvm_chardev_ops
.owner
= module
;
3969 kvm_vm_fops
.owner
= module
;
3970 kvm_vcpu_fops
.owner
= module
;
3972 r
= misc_register(&kvm_dev
);
3974 pr_err("kvm: misc device register failed\n");
3978 register_syscore_ops(&kvm_syscore_ops
);
3980 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3981 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3983 r
= kvm_init_debug();
3985 pr_err("kvm: create debugfs files failed\n");
3989 r
= kvm_vfio_ops_init();
3995 unregister_syscore_ops(&kvm_syscore_ops
);
3996 misc_deregister(&kvm_dev
);
3998 kvm_async_pf_deinit();
4000 kmem_cache_destroy(kvm_vcpu_cache
);
4002 unregister_reboot_notifier(&kvm_reboot_notifier
);
4003 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4006 kvm_arch_hardware_unsetup();
4008 free_cpumask_var(cpus_hardware_enabled
);
4016 EXPORT_SYMBOL_GPL(kvm_init
);
4020 debugfs_remove_recursive(kvm_debugfs_dir
);
4021 misc_deregister(&kvm_dev
);
4022 kmem_cache_destroy(kvm_vcpu_cache
);
4023 kvm_async_pf_deinit();
4024 unregister_syscore_ops(&kvm_syscore_ops
);
4025 unregister_reboot_notifier(&kvm_reboot_notifier
);
4026 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4027 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4028 kvm_arch_hardware_unsetup();
4031 free_cpumask_var(cpus_hardware_enabled
);
4032 kvm_vfio_ops_exit();
4034 EXPORT_SYMBOL_GPL(kvm_exit
);