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 /* TODO: merge with kvm_arch_vcpu_should_kick */
169 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
171 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
174 * We need to wait for the VCPU to reenable interrupts and get out of
175 * READING_SHADOW_PAGE_TABLES mode.
177 if (req
& KVM_REQUEST_WAIT
)
178 return mode
!= OUTSIDE_GUEST_MODE
;
181 * Need to kick a running VCPU, but otherwise there is nothing to do.
183 return mode
== IN_GUEST_MODE
;
186 static void ack_flush(void *_completed
)
190 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
195 bool wait
= req
& KVM_REQUEST_WAIT
;
196 struct kvm_vcpu
*vcpu
;
198 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
201 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
202 kvm_make_request(req
, vcpu
);
205 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
208 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
209 kvm_request_needs_ipi(vcpu
, req
))
210 cpumask_set_cpu(cpu
, cpus
);
212 if (unlikely(cpus
== NULL
))
213 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, wait
);
214 else if (!cpumask_empty(cpus
))
215 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
219 free_cpumask_var(cpus
);
223 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
224 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
227 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
228 * kvm_make_all_cpus_request.
230 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
233 * We want to publish modifications to the page tables before reading
234 * mode. Pairs with a memory barrier in arch-specific code.
235 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
236 * and smp_mb in walk_shadow_page_lockless_begin/end.
237 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
239 * There is already an smp_mb__after_atomic() before
240 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
243 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
244 ++kvm
->stat
.remote_tlb_flush
;
245 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
247 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
250 void kvm_reload_remote_mmus(struct kvm
*kvm
)
252 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
255 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
260 mutex_init(&vcpu
->mutex
);
265 init_swait_queue_head(&vcpu
->wq
);
266 kvm_async_pf_vcpu_init(vcpu
);
269 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
271 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
276 vcpu
->run
= page_address(page
);
278 kvm_vcpu_set_in_spin_loop(vcpu
, false);
279 kvm_vcpu_set_dy_eligible(vcpu
, false);
280 vcpu
->preempted
= false;
282 r
= kvm_arch_vcpu_init(vcpu
);
288 free_page((unsigned long)vcpu
->run
);
292 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
294 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
297 kvm_arch_vcpu_uninit(vcpu
);
298 free_page((unsigned long)vcpu
->run
);
300 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
302 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
303 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
305 return container_of(mn
, struct kvm
, mmu_notifier
);
308 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
309 struct mm_struct
*mm
,
310 unsigned long address
)
312 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
313 int need_tlb_flush
, idx
;
316 * When ->invalidate_page runs, the linux pte has been zapped
317 * already but the page is still allocated until
318 * ->invalidate_page returns. So if we increase the sequence
319 * here the kvm page fault will notice if the spte can't be
320 * established because the page is going to be freed. If
321 * instead the kvm page fault establishes the spte before
322 * ->invalidate_page runs, kvm_unmap_hva will release it
325 * The sequence increase only need to be seen at spin_unlock
326 * time, and not at spin_lock time.
328 * Increasing the sequence after the spin_unlock would be
329 * unsafe because the kvm page fault could then establish the
330 * pte after kvm_unmap_hva returned, without noticing the page
331 * is going to be freed.
333 idx
= srcu_read_lock(&kvm
->srcu
);
334 spin_lock(&kvm
->mmu_lock
);
336 kvm
->mmu_notifier_seq
++;
337 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
338 /* we've to flush the tlb before the pages can be freed */
340 kvm_flush_remote_tlbs(kvm
);
342 spin_unlock(&kvm
->mmu_lock
);
344 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
346 srcu_read_unlock(&kvm
->srcu
, idx
);
349 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
350 struct mm_struct
*mm
,
351 unsigned long address
,
354 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
357 idx
= srcu_read_lock(&kvm
->srcu
);
358 spin_lock(&kvm
->mmu_lock
);
359 kvm
->mmu_notifier_seq
++;
360 kvm_set_spte_hva(kvm
, address
, pte
);
361 spin_unlock(&kvm
->mmu_lock
);
362 srcu_read_unlock(&kvm
->srcu
, idx
);
365 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
366 struct mm_struct
*mm
,
370 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
371 int need_tlb_flush
= 0, idx
;
373 idx
= srcu_read_lock(&kvm
->srcu
);
374 spin_lock(&kvm
->mmu_lock
);
376 * The count increase must become visible at unlock time as no
377 * spte can be established without taking the mmu_lock and
378 * count is also read inside the mmu_lock critical section.
380 kvm
->mmu_notifier_count
++;
381 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
382 need_tlb_flush
|= kvm
->tlbs_dirty
;
383 /* we've to flush the tlb before the pages can be freed */
385 kvm_flush_remote_tlbs(kvm
);
387 spin_unlock(&kvm
->mmu_lock
);
388 srcu_read_unlock(&kvm
->srcu
, idx
);
391 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
392 struct mm_struct
*mm
,
396 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
398 spin_lock(&kvm
->mmu_lock
);
400 * This sequence increase will notify the kvm page fault that
401 * the page that is going to be mapped in the spte could have
404 kvm
->mmu_notifier_seq
++;
407 * The above sequence increase must be visible before the
408 * below count decrease, which is ensured by the smp_wmb above
409 * in conjunction with the smp_rmb in mmu_notifier_retry().
411 kvm
->mmu_notifier_count
--;
412 spin_unlock(&kvm
->mmu_lock
);
414 BUG_ON(kvm
->mmu_notifier_count
< 0);
417 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
418 struct mm_struct
*mm
,
422 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
425 idx
= srcu_read_lock(&kvm
->srcu
);
426 spin_lock(&kvm
->mmu_lock
);
428 young
= kvm_age_hva(kvm
, start
, end
);
430 kvm_flush_remote_tlbs(kvm
);
432 spin_unlock(&kvm
->mmu_lock
);
433 srcu_read_unlock(&kvm
->srcu
, idx
);
438 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
439 struct mm_struct
*mm
,
443 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
446 idx
= srcu_read_lock(&kvm
->srcu
);
447 spin_lock(&kvm
->mmu_lock
);
449 * Even though we do not flush TLB, this will still adversely
450 * affect performance on pre-Haswell Intel EPT, where there is
451 * no EPT Access Bit to clear so that we have to tear down EPT
452 * tables instead. If we find this unacceptable, we can always
453 * add a parameter to kvm_age_hva so that it effectively doesn't
454 * do anything on clear_young.
456 * Also note that currently we never issue secondary TLB flushes
457 * from clear_young, leaving this job up to the regular system
458 * cadence. If we find this inaccurate, we might come up with a
459 * more sophisticated heuristic later.
461 young
= kvm_age_hva(kvm
, start
, end
);
462 spin_unlock(&kvm
->mmu_lock
);
463 srcu_read_unlock(&kvm
->srcu
, idx
);
468 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
469 struct mm_struct
*mm
,
470 unsigned long address
)
472 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
475 idx
= srcu_read_lock(&kvm
->srcu
);
476 spin_lock(&kvm
->mmu_lock
);
477 young
= kvm_test_age_hva(kvm
, address
);
478 spin_unlock(&kvm
->mmu_lock
);
479 srcu_read_unlock(&kvm
->srcu
, idx
);
484 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
485 struct mm_struct
*mm
)
487 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
490 idx
= srcu_read_lock(&kvm
->srcu
);
491 kvm_arch_flush_shadow_all(kvm
);
492 srcu_read_unlock(&kvm
->srcu
, idx
);
495 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
496 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
497 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
498 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
499 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
500 .clear_young
= kvm_mmu_notifier_clear_young
,
501 .test_young
= kvm_mmu_notifier_test_young
,
502 .change_pte
= kvm_mmu_notifier_change_pte
,
503 .release
= kvm_mmu_notifier_release
,
506 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
508 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
509 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
512 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
514 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
519 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
521 static struct kvm_memslots
*kvm_alloc_memslots(void)
524 struct kvm_memslots
*slots
;
526 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
530 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
531 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
536 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
538 if (!memslot
->dirty_bitmap
)
541 kvfree(memslot
->dirty_bitmap
);
542 memslot
->dirty_bitmap
= NULL
;
546 * Free any memory in @free but not in @dont.
548 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
549 struct kvm_memory_slot
*dont
)
551 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
552 kvm_destroy_dirty_bitmap(free
);
554 kvm_arch_free_memslot(kvm
, free
, dont
);
559 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
561 struct kvm_memory_slot
*memslot
;
566 kvm_for_each_memslot(memslot
, slots
)
567 kvm_free_memslot(kvm
, memslot
, NULL
);
572 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
576 if (!kvm
->debugfs_dentry
)
579 debugfs_remove_recursive(kvm
->debugfs_dentry
);
581 if (kvm
->debugfs_stat_data
) {
582 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
583 kfree(kvm
->debugfs_stat_data
[i
]);
584 kfree(kvm
->debugfs_stat_data
);
588 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
590 char dir_name
[ITOA_MAX_LEN
* 2];
591 struct kvm_stat_data
*stat_data
;
592 struct kvm_stats_debugfs_item
*p
;
594 if (!debugfs_initialized())
597 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
598 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
600 if (!kvm
->debugfs_dentry
)
603 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
604 sizeof(*kvm
->debugfs_stat_data
),
606 if (!kvm
->debugfs_stat_data
)
609 for (p
= debugfs_entries
; p
->name
; p
++) {
610 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
614 stat_data
->kvm
= kvm
;
615 stat_data
->offset
= p
->offset
;
616 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
617 if (!debugfs_create_file(p
->name
, 0644,
620 stat_fops_per_vm
[p
->kind
]))
626 static struct kvm
*kvm_create_vm(unsigned long type
)
629 struct kvm
*kvm
= kvm_arch_alloc_vm();
632 return ERR_PTR(-ENOMEM
);
634 spin_lock_init(&kvm
->mmu_lock
);
636 kvm
->mm
= current
->mm
;
637 kvm_eventfd_init(kvm
);
638 mutex_init(&kvm
->lock
);
639 mutex_init(&kvm
->irq_lock
);
640 mutex_init(&kvm
->slots_lock
);
641 refcount_set(&kvm
->users_count
, 1);
642 INIT_LIST_HEAD(&kvm
->devices
);
644 r
= kvm_arch_init_vm(kvm
, type
);
646 goto out_err_no_disable
;
648 r
= hardware_enable_all();
650 goto out_err_no_disable
;
652 #ifdef CONFIG_HAVE_KVM_IRQFD
653 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
656 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
659 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
660 struct kvm_memslots
*slots
= kvm_alloc_memslots();
662 goto out_err_no_srcu
;
664 * Generations must be different for each address space.
665 * Init kvm generation close to the maximum to easily test the
666 * code of handling generation number wrap-around.
668 slots
->generation
= i
* 2 - 150;
669 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
672 if (init_srcu_struct(&kvm
->srcu
))
673 goto out_err_no_srcu
;
674 if (init_srcu_struct(&kvm
->irq_srcu
))
675 goto out_err_no_irq_srcu
;
676 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
677 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
683 r
= kvm_init_mmu_notifier(kvm
);
687 spin_lock(&kvm_lock
);
688 list_add(&kvm
->vm_list
, &vm_list
);
689 spin_unlock(&kvm_lock
);
691 preempt_notifier_inc();
696 cleanup_srcu_struct(&kvm
->irq_srcu
);
698 cleanup_srcu_struct(&kvm
->srcu
);
700 hardware_disable_all();
702 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
703 kfree(kvm
->buses
[i
]);
704 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
705 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
706 kvm_arch_free_vm(kvm
);
712 * Avoid using vmalloc for a small buffer.
713 * Should not be used when the size is statically known.
715 void *kvm_kvzalloc(unsigned long size
)
717 if (size
> PAGE_SIZE
)
718 return vzalloc(size
);
720 return kzalloc(size
, GFP_KERNEL
);
723 static void kvm_destroy_devices(struct kvm
*kvm
)
725 struct kvm_device
*dev
, *tmp
;
728 * We do not need to take the kvm->lock here, because nobody else
729 * has a reference to the struct kvm at this point and therefore
730 * cannot access the devices list anyhow.
732 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
733 list_del(&dev
->vm_node
);
734 dev
->ops
->destroy(dev
);
738 static void kvm_destroy_vm(struct kvm
*kvm
)
741 struct mm_struct
*mm
= kvm
->mm
;
743 kvm_destroy_vm_debugfs(kvm
);
744 kvm_arch_sync_events(kvm
);
745 spin_lock(&kvm_lock
);
746 list_del(&kvm
->vm_list
);
747 spin_unlock(&kvm_lock
);
748 kvm_free_irq_routing(kvm
);
749 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
751 kvm_io_bus_destroy(kvm
->buses
[i
]);
752 kvm
->buses
[i
] = NULL
;
754 kvm_coalesced_mmio_free(kvm
);
755 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
756 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
758 kvm_arch_flush_shadow_all(kvm
);
760 kvm_arch_destroy_vm(kvm
);
761 kvm_destroy_devices(kvm
);
762 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
763 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
764 cleanup_srcu_struct(&kvm
->irq_srcu
);
765 cleanup_srcu_struct(&kvm
->srcu
);
766 kvm_arch_free_vm(kvm
);
767 preempt_notifier_dec();
768 hardware_disable_all();
772 void kvm_get_kvm(struct kvm
*kvm
)
774 refcount_inc(&kvm
->users_count
);
776 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
778 void kvm_put_kvm(struct kvm
*kvm
)
780 if (refcount_dec_and_test(&kvm
->users_count
))
783 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
786 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
788 struct kvm
*kvm
= filp
->private_data
;
790 kvm_irqfd_release(kvm
);
797 * Allocation size is twice as large as the actual dirty bitmap size.
798 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
800 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
802 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
804 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
805 if (!memslot
->dirty_bitmap
)
812 * Insert memslot and re-sort memslots based on their GFN,
813 * so binary search could be used to lookup GFN.
814 * Sorting algorithm takes advantage of having initially
815 * sorted array and known changed memslot position.
817 static void update_memslots(struct kvm_memslots
*slots
,
818 struct kvm_memory_slot
*new)
821 int i
= slots
->id_to_index
[id
];
822 struct kvm_memory_slot
*mslots
= slots
->memslots
;
824 WARN_ON(mslots
[i
].id
!= id
);
826 WARN_ON(!mslots
[i
].npages
);
827 if (mslots
[i
].npages
)
830 if (!mslots
[i
].npages
)
834 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
835 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
836 if (!mslots
[i
+ 1].npages
)
838 mslots
[i
] = mslots
[i
+ 1];
839 slots
->id_to_index
[mslots
[i
].id
] = i
;
844 * The ">=" is needed when creating a slot with base_gfn == 0,
845 * so that it moves before all those with base_gfn == npages == 0.
847 * On the other hand, if new->npages is zero, the above loop has
848 * already left i pointing to the beginning of the empty part of
849 * mslots, and the ">=" would move the hole backwards in this
850 * case---which is wrong. So skip the loop when deleting a slot.
854 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
855 mslots
[i
] = mslots
[i
- 1];
856 slots
->id_to_index
[mslots
[i
].id
] = i
;
860 WARN_ON_ONCE(i
!= slots
->used_slots
);
863 slots
->id_to_index
[mslots
[i
].id
] = i
;
866 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
868 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
870 #ifdef __KVM_HAVE_READONLY_MEM
871 valid_flags
|= KVM_MEM_READONLY
;
874 if (mem
->flags
& ~valid_flags
)
880 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
881 int as_id
, struct kvm_memslots
*slots
)
883 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
886 * Set the low bit in the generation, which disables SPTE caching
887 * until the end of synchronize_srcu_expedited.
889 WARN_ON(old_memslots
->generation
& 1);
890 slots
->generation
= old_memslots
->generation
+ 1;
892 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
893 synchronize_srcu_expedited(&kvm
->srcu
);
896 * Increment the new memslot generation a second time. This prevents
897 * vm exits that race with memslot updates from caching a memslot
898 * generation that will (potentially) be valid forever.
900 * Generations must be unique even across address spaces. We do not need
901 * a global counter for that, instead the generation space is evenly split
902 * across address spaces. For example, with two address spaces, address
903 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
904 * use generations 2, 6, 10, 14, ...
906 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
908 kvm_arch_memslots_updated(kvm
, slots
);
914 * Allocate some memory and give it an address in the guest physical address
917 * Discontiguous memory is allowed, mostly for framebuffers.
919 * Must be called holding kvm->slots_lock for write.
921 int __kvm_set_memory_region(struct kvm
*kvm
,
922 const struct kvm_userspace_memory_region
*mem
)
926 unsigned long npages
;
927 struct kvm_memory_slot
*slot
;
928 struct kvm_memory_slot old
, new;
929 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
931 enum kvm_mr_change change
;
933 r
= check_memory_region_flags(mem
);
938 as_id
= mem
->slot
>> 16;
941 /* General sanity checks */
942 if (mem
->memory_size
& (PAGE_SIZE
- 1))
944 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
946 /* We can read the guest memory with __xxx_user() later on. */
947 if ((id
< KVM_USER_MEM_SLOTS
) &&
948 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
949 !access_ok(VERIFY_WRITE
,
950 (void __user
*)(unsigned long)mem
->userspace_addr
,
953 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
955 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
958 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
959 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
960 npages
= mem
->memory_size
>> PAGE_SHIFT
;
962 if (npages
> KVM_MEM_MAX_NR_PAGES
)
968 new.base_gfn
= base_gfn
;
970 new.flags
= mem
->flags
;
974 change
= KVM_MR_CREATE
;
975 else { /* Modify an existing slot. */
976 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
977 (npages
!= old
.npages
) ||
978 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
981 if (base_gfn
!= old
.base_gfn
)
982 change
= KVM_MR_MOVE
;
983 else if (new.flags
!= old
.flags
)
984 change
= KVM_MR_FLAGS_ONLY
;
985 else { /* Nothing to change. */
994 change
= KVM_MR_DELETE
;
999 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1000 /* Check for overlaps */
1002 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
1003 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
1006 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
1007 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
1012 /* Free page dirty bitmap if unneeded */
1013 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1014 new.dirty_bitmap
= NULL
;
1017 if (change
== KVM_MR_CREATE
) {
1018 new.userspace_addr
= mem
->userspace_addr
;
1020 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1024 /* Allocate page dirty bitmap if needed */
1025 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1026 if (kvm_create_dirty_bitmap(&new) < 0)
1030 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
1033 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1035 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1036 slot
= id_to_memslot(slots
, id
);
1037 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1039 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1041 /* From this point no new shadow pages pointing to a deleted,
1042 * or moved, memslot will be created.
1044 * validation of sp->gfn happens in:
1045 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1046 * - kvm_is_visible_gfn (mmu_check_roots)
1048 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1051 * We can re-use the old_memslots from above, the only difference
1052 * from the currently installed memslots is the invalid flag. This
1053 * will get overwritten by update_memslots anyway.
1055 slots
= old_memslots
;
1058 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1062 /* actual memory is freed via old in kvm_free_memslot below */
1063 if (change
== KVM_MR_DELETE
) {
1064 new.dirty_bitmap
= NULL
;
1065 memset(&new.arch
, 0, sizeof(new.arch
));
1068 update_memslots(slots
, &new);
1069 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1071 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1073 kvm_free_memslot(kvm
, &old
, &new);
1074 kvfree(old_memslots
);
1080 kvm_free_memslot(kvm
, &new, &old
);
1084 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1086 int kvm_set_memory_region(struct kvm
*kvm
,
1087 const struct kvm_userspace_memory_region
*mem
)
1091 mutex_lock(&kvm
->slots_lock
);
1092 r
= __kvm_set_memory_region(kvm
, mem
);
1093 mutex_unlock(&kvm
->slots_lock
);
1096 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1098 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1099 struct kvm_userspace_memory_region
*mem
)
1101 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1104 return kvm_set_memory_region(kvm
, mem
);
1107 int kvm_get_dirty_log(struct kvm
*kvm
,
1108 struct kvm_dirty_log
*log
, int *is_dirty
)
1110 struct kvm_memslots
*slots
;
1111 struct kvm_memory_slot
*memslot
;
1114 unsigned long any
= 0;
1116 as_id
= log
->slot
>> 16;
1117 id
= (u16
)log
->slot
;
1118 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1121 slots
= __kvm_memslots(kvm
, as_id
);
1122 memslot
= id_to_memslot(slots
, id
);
1123 if (!memslot
->dirty_bitmap
)
1126 n
= kvm_dirty_bitmap_bytes(memslot
);
1128 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1129 any
= memslot
->dirty_bitmap
[i
];
1131 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1138 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1140 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1142 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1143 * are dirty write protect them for next write.
1144 * @kvm: pointer to kvm instance
1145 * @log: slot id and address to which we copy the log
1146 * @is_dirty: flag set if any page is dirty
1148 * We need to keep it in mind that VCPU threads can write to the bitmap
1149 * concurrently. So, to avoid losing track of dirty pages we keep the
1152 * 1. Take a snapshot of the bit and clear it if needed.
1153 * 2. Write protect the corresponding page.
1154 * 3. Copy the snapshot to the userspace.
1155 * 4. Upon return caller flushes TLB's if needed.
1157 * Between 2 and 4, the guest may write to the page using the remaining TLB
1158 * entry. This is not a problem because the page is reported dirty using
1159 * the snapshot taken before and step 4 ensures that writes done after
1160 * exiting to userspace will be logged for the next call.
1163 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1164 struct kvm_dirty_log
*log
, bool *is_dirty
)
1166 struct kvm_memslots
*slots
;
1167 struct kvm_memory_slot
*memslot
;
1170 unsigned long *dirty_bitmap
;
1171 unsigned long *dirty_bitmap_buffer
;
1173 as_id
= log
->slot
>> 16;
1174 id
= (u16
)log
->slot
;
1175 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1178 slots
= __kvm_memslots(kvm
, as_id
);
1179 memslot
= id_to_memslot(slots
, id
);
1181 dirty_bitmap
= memslot
->dirty_bitmap
;
1185 n
= kvm_dirty_bitmap_bytes(memslot
);
1187 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1188 memset(dirty_bitmap_buffer
, 0, n
);
1190 spin_lock(&kvm
->mmu_lock
);
1192 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1196 if (!dirty_bitmap
[i
])
1201 mask
= xchg(&dirty_bitmap
[i
], 0);
1202 dirty_bitmap_buffer
[i
] = mask
;
1205 offset
= i
* BITS_PER_LONG
;
1206 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1211 spin_unlock(&kvm
->mmu_lock
);
1212 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1216 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1219 bool kvm_largepages_enabled(void)
1221 return largepages_enabled
;
1224 void kvm_disable_largepages(void)
1226 largepages_enabled
= false;
1228 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1230 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1232 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1234 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1236 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1238 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1241 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1243 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1245 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1246 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1251 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1253 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1255 struct vm_area_struct
*vma
;
1256 unsigned long addr
, size
;
1260 addr
= gfn_to_hva(kvm
, gfn
);
1261 if (kvm_is_error_hva(addr
))
1264 down_read(¤t
->mm
->mmap_sem
);
1265 vma
= find_vma(current
->mm
, addr
);
1269 size
= vma_kernel_pagesize(vma
);
1272 up_read(¤t
->mm
->mmap_sem
);
1277 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1279 return slot
->flags
& KVM_MEM_READONLY
;
1282 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1283 gfn_t
*nr_pages
, bool write
)
1285 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1286 return KVM_HVA_ERR_BAD
;
1288 if (memslot_is_readonly(slot
) && write
)
1289 return KVM_HVA_ERR_RO_BAD
;
1292 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1294 return __gfn_to_hva_memslot(slot
, gfn
);
1297 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1300 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1303 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1306 return gfn_to_hva_many(slot
, gfn
, NULL
);
1308 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1310 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1312 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1314 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1316 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1318 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1320 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1323 * If writable is set to false, the hva returned by this function is only
1324 * allowed to be read.
1326 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1327 gfn_t gfn
, bool *writable
)
1329 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1331 if (!kvm_is_error_hva(hva
) && writable
)
1332 *writable
= !memslot_is_readonly(slot
);
1337 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1339 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1341 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1344 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1346 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1348 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1351 static int get_user_page_nowait(unsigned long start
, int write
,
1354 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1357 flags
|= FOLL_WRITE
;
1359 return get_user_pages(start
, 1, flags
, page
, NULL
);
1362 static inline int check_user_page_hwpoison(unsigned long addr
)
1364 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1366 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1367 return rc
== -EHWPOISON
;
1371 * The atomic path to get the writable pfn which will be stored in @pfn,
1372 * true indicates success, otherwise false is returned.
1374 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1375 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1377 struct page
*page
[1];
1380 if (!(async
|| atomic
))
1384 * Fast pin a writable pfn only if it is a write fault request
1385 * or the caller allows to map a writable pfn for a read fault
1388 if (!(write_fault
|| writable
))
1391 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1393 *pfn
= page_to_pfn(page
[0]);
1404 * The slow path to get the pfn of the specified host virtual address,
1405 * 1 indicates success, -errno is returned if error is detected.
1407 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1408 bool *writable
, kvm_pfn_t
*pfn
)
1410 struct page
*page
[1];
1416 *writable
= write_fault
;
1419 down_read(¤t
->mm
->mmap_sem
);
1420 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1421 up_read(¤t
->mm
->mmap_sem
);
1423 unsigned int flags
= FOLL_HWPOISON
;
1426 flags
|= FOLL_WRITE
;
1428 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1433 /* map read fault as writable if possible */
1434 if (unlikely(!write_fault
) && writable
) {
1435 struct page
*wpage
[1];
1437 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1446 *pfn
= page_to_pfn(page
[0]);
1450 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1452 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1455 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1461 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1462 unsigned long addr
, bool *async
,
1463 bool write_fault
, kvm_pfn_t
*p_pfn
)
1468 r
= follow_pfn(vma
, addr
, &pfn
);
1471 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1472 * not call the fault handler, so do it here.
1474 bool unlocked
= false;
1475 r
= fixup_user_fault(current
, current
->mm
, addr
,
1476 (write_fault
? FAULT_FLAG_WRITE
: 0),
1483 r
= follow_pfn(vma
, addr
, &pfn
);
1491 * Get a reference here because callers of *hva_to_pfn* and
1492 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1493 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1494 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1495 * simply do nothing for reserved pfns.
1497 * Whoever called remap_pfn_range is also going to call e.g.
1498 * unmap_mapping_range before the underlying pages are freed,
1499 * causing a call to our MMU notifier.
1508 * Pin guest page in memory and return its pfn.
1509 * @addr: host virtual address which maps memory to the guest
1510 * @atomic: whether this function can sleep
1511 * @async: whether this function need to wait IO complete if the
1512 * host page is not in the memory
1513 * @write_fault: whether we should get a writable host page
1514 * @writable: whether it allows to map a writable host page for !@write_fault
1516 * The function will map a writable host page for these two cases:
1517 * 1): @write_fault = true
1518 * 2): @write_fault = false && @writable, @writable will tell the caller
1519 * whether the mapping is writable.
1521 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1522 bool write_fault
, bool *writable
)
1524 struct vm_area_struct
*vma
;
1528 /* we can do it either atomically or asynchronously, not both */
1529 BUG_ON(atomic
&& async
);
1531 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1535 return KVM_PFN_ERR_FAULT
;
1537 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1541 down_read(¤t
->mm
->mmap_sem
);
1542 if (npages
== -EHWPOISON
||
1543 (!async
&& check_user_page_hwpoison(addr
))) {
1544 pfn
= KVM_PFN_ERR_HWPOISON
;
1549 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1552 pfn
= KVM_PFN_ERR_FAULT
;
1553 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1554 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, &pfn
);
1558 pfn
= KVM_PFN_ERR_FAULT
;
1560 if (async
&& vma_is_valid(vma
, write_fault
))
1562 pfn
= KVM_PFN_ERR_FAULT
;
1565 up_read(¤t
->mm
->mmap_sem
);
1569 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1570 bool atomic
, bool *async
, bool write_fault
,
1573 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1575 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1578 return KVM_PFN_ERR_RO_FAULT
;
1581 if (kvm_is_error_hva(addr
)) {
1584 return KVM_PFN_NOSLOT
;
1587 /* Do not map writable pfn in the readonly memslot. */
1588 if (writable
&& memslot_is_readonly(slot
)) {
1593 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1596 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1598 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1601 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1602 write_fault
, writable
);
1604 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1606 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1608 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1610 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1612 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1614 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1616 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1618 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1620 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1622 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1624 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1626 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1628 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1630 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1632 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1634 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1636 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1638 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1640 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1642 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1643 struct page
**pages
, int nr_pages
)
1648 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1649 if (kvm_is_error_hva(addr
))
1652 if (entry
< nr_pages
)
1655 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1657 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1659 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1661 if (is_error_noslot_pfn(pfn
))
1662 return KVM_ERR_PTR_BAD_PAGE
;
1664 if (kvm_is_reserved_pfn(pfn
)) {
1666 return KVM_ERR_PTR_BAD_PAGE
;
1669 return pfn_to_page(pfn
);
1672 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1676 pfn
= gfn_to_pfn(kvm
, gfn
);
1678 return kvm_pfn_to_page(pfn
);
1680 EXPORT_SYMBOL_GPL(gfn_to_page
);
1682 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1686 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1688 return kvm_pfn_to_page(pfn
);
1690 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1692 void kvm_release_page_clean(struct page
*page
)
1694 WARN_ON(is_error_page(page
));
1696 kvm_release_pfn_clean(page_to_pfn(page
));
1698 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1700 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1702 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1703 put_page(pfn_to_page(pfn
));
1705 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1707 void kvm_release_page_dirty(struct page
*page
)
1709 WARN_ON(is_error_page(page
));
1711 kvm_release_pfn_dirty(page_to_pfn(page
));
1713 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1715 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1717 kvm_set_pfn_dirty(pfn
);
1718 kvm_release_pfn_clean(pfn
);
1721 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1723 if (!kvm_is_reserved_pfn(pfn
)) {
1724 struct page
*page
= pfn_to_page(pfn
);
1726 if (!PageReserved(page
))
1730 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1732 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1734 if (!kvm_is_reserved_pfn(pfn
))
1735 mark_page_accessed(pfn_to_page(pfn
));
1737 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1739 void kvm_get_pfn(kvm_pfn_t pfn
)
1741 if (!kvm_is_reserved_pfn(pfn
))
1742 get_page(pfn_to_page(pfn
));
1744 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1746 static int next_segment(unsigned long len
, int offset
)
1748 if (len
> PAGE_SIZE
- offset
)
1749 return PAGE_SIZE
- offset
;
1754 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1755 void *data
, int offset
, int len
)
1760 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1761 if (kvm_is_error_hva(addr
))
1763 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1769 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1772 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1774 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1776 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1778 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1779 int offset
, int len
)
1781 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1783 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1785 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1787 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1789 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1791 int offset
= offset_in_page(gpa
);
1794 while ((seg
= next_segment(len
, offset
)) != 0) {
1795 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1805 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1807 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1809 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1811 int offset
= offset_in_page(gpa
);
1814 while ((seg
= next_segment(len
, offset
)) != 0) {
1815 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1825 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1827 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1828 void *data
, int offset
, unsigned long len
)
1833 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1834 if (kvm_is_error_hva(addr
))
1836 pagefault_disable();
1837 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1844 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1847 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1848 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1849 int offset
= offset_in_page(gpa
);
1851 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1853 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1855 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1856 void *data
, unsigned long len
)
1858 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1859 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1860 int offset
= offset_in_page(gpa
);
1862 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1864 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1866 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1867 const void *data
, int offset
, int len
)
1872 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1873 if (kvm_is_error_hva(addr
))
1875 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1878 mark_page_dirty_in_slot(memslot
, gfn
);
1882 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1883 const void *data
, int offset
, int len
)
1885 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1887 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1889 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1891 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1892 const void *data
, int offset
, int len
)
1894 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1896 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1898 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1900 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1903 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1905 int offset
= offset_in_page(gpa
);
1908 while ((seg
= next_segment(len
, offset
)) != 0) {
1909 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1919 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1921 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1924 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1926 int offset
= offset_in_page(gpa
);
1929 while ((seg
= next_segment(len
, offset
)) != 0) {
1930 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1940 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1942 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1943 struct gfn_to_hva_cache
*ghc
,
1944 gpa_t gpa
, unsigned long len
)
1946 int offset
= offset_in_page(gpa
);
1947 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1948 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1949 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1950 gfn_t nr_pages_avail
;
1953 ghc
->generation
= slots
->generation
;
1955 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1956 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1957 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1961 * If the requested region crosses two memslots, we still
1962 * verify that the entire region is valid here.
1964 while (start_gfn
<= end_gfn
) {
1965 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1966 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1968 if (kvm_is_error_hva(ghc
->hva
))
1970 start_gfn
+= nr_pages_avail
;
1972 /* Use the slow path for cross page reads and writes. */
1973 ghc
->memslot
= NULL
;
1978 int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
1979 gpa_t gpa
, unsigned long len
)
1981 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
1982 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1984 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init
);
1986 int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
1987 void *data
, int offset
, unsigned long len
)
1989 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
1991 gpa_t gpa
= ghc
->gpa
+ offset
;
1993 BUG_ON(len
+ offset
> ghc
->len
);
1995 if (slots
->generation
!= ghc
->generation
)
1996 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1998 if (unlikely(!ghc
->memslot
))
1999 return kvm_vcpu_write_guest(vcpu
, gpa
, data
, len
);
2001 if (kvm_is_error_hva(ghc
->hva
))
2004 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2007 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2011 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached
);
2013 int kvm_vcpu_write_guest_cached(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
2014 void *data
, unsigned long len
)
2016 return kvm_vcpu_write_guest_offset_cached(vcpu
, ghc
, data
, 0, len
);
2018 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached
);
2020 int kvm_vcpu_read_guest_cached(struct kvm_vcpu
*vcpu
, struct gfn_to_hva_cache
*ghc
,
2021 void *data
, unsigned long len
)
2023 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2026 BUG_ON(len
> ghc
->len
);
2028 if (slots
->generation
!= ghc
->generation
)
2029 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2031 if (unlikely(!ghc
->memslot
))
2032 return kvm_vcpu_read_guest(vcpu
, ghc
->gpa
, data
, len
);
2034 if (kvm_is_error_hva(ghc
->hva
))
2037 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2043 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached
);
2045 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2047 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2049 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2051 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2053 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2055 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2057 int offset
= offset_in_page(gpa
);
2060 while ((seg
= next_segment(len
, offset
)) != 0) {
2061 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2070 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2072 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2075 if (memslot
&& memslot
->dirty_bitmap
) {
2076 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2078 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2082 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2084 struct kvm_memory_slot
*memslot
;
2086 memslot
= gfn_to_memslot(kvm
, gfn
);
2087 mark_page_dirty_in_slot(memslot
, gfn
);
2089 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2091 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2093 struct kvm_memory_slot
*memslot
;
2095 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2096 mark_page_dirty_in_slot(memslot
, gfn
);
2098 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2100 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2102 unsigned int old
, val
, grow
;
2104 old
= val
= vcpu
->halt_poll_ns
;
2105 grow
= READ_ONCE(halt_poll_ns_grow
);
2107 if (val
== 0 && grow
)
2112 if (val
> halt_poll_ns
)
2115 vcpu
->halt_poll_ns
= val
;
2116 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2119 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2121 unsigned int old
, val
, shrink
;
2123 old
= val
= vcpu
->halt_poll_ns
;
2124 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2130 vcpu
->halt_poll_ns
= val
;
2131 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2134 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2136 if (kvm_arch_vcpu_runnable(vcpu
)) {
2137 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2140 if (kvm_cpu_has_pending_timer(vcpu
))
2142 if (signal_pending(current
))
2149 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2151 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2154 DECLARE_SWAITQUEUE(wait
);
2155 bool waited
= false;
2158 start
= cur
= ktime_get();
2159 if (vcpu
->halt_poll_ns
) {
2160 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2162 ++vcpu
->stat
.halt_attempted_poll
;
2165 * This sets KVM_REQ_UNHALT if an interrupt
2168 if (kvm_vcpu_check_block(vcpu
) < 0) {
2169 ++vcpu
->stat
.halt_successful_poll
;
2170 if (!vcpu_valid_wakeup(vcpu
))
2171 ++vcpu
->stat
.halt_poll_invalid
;
2175 } while (single_task_running() && ktime_before(cur
, stop
));
2178 kvm_arch_vcpu_blocking(vcpu
);
2181 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2183 if (kvm_vcpu_check_block(vcpu
) < 0)
2190 finish_swait(&vcpu
->wq
, &wait
);
2193 kvm_arch_vcpu_unblocking(vcpu
);
2195 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2197 if (!vcpu_valid_wakeup(vcpu
))
2198 shrink_halt_poll_ns(vcpu
);
2199 else if (halt_poll_ns
) {
2200 if (block_ns
<= vcpu
->halt_poll_ns
)
2202 /* we had a long block, shrink polling */
2203 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2204 shrink_halt_poll_ns(vcpu
);
2205 /* we had a short halt and our poll time is too small */
2206 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2207 block_ns
< halt_poll_ns
)
2208 grow_halt_poll_ns(vcpu
);
2210 vcpu
->halt_poll_ns
= 0;
2212 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2213 kvm_arch_vcpu_block_finish(vcpu
);
2215 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2217 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2219 struct swait_queue_head
*wqp
;
2221 wqp
= kvm_arch_vcpu_wq(vcpu
);
2222 if (swait_active(wqp
)) {
2224 ++vcpu
->stat
.halt_wakeup
;
2230 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2233 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2235 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2238 int cpu
= vcpu
->cpu
;
2240 if (kvm_vcpu_wake_up(vcpu
))
2244 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2245 if (kvm_arch_vcpu_should_kick(vcpu
))
2246 smp_send_reschedule(cpu
);
2249 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2251 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2254 struct task_struct
*task
= NULL
;
2258 pid
= rcu_dereference(target
->pid
);
2260 task
= get_pid_task(pid
, PIDTYPE_PID
);
2264 ret
= yield_to(task
, 1);
2265 put_task_struct(task
);
2269 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2272 * Helper that checks whether a VCPU is eligible for directed yield.
2273 * Most eligible candidate to yield is decided by following heuristics:
2275 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2276 * (preempted lock holder), indicated by @in_spin_loop.
2277 * Set at the beiginning and cleared at the end of interception/PLE handler.
2279 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2280 * chance last time (mostly it has become eligible now since we have probably
2281 * yielded to lockholder in last iteration. This is done by toggling
2282 * @dy_eligible each time a VCPU checked for eligibility.)
2284 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2285 * to preempted lock-holder could result in wrong VCPU selection and CPU
2286 * burning. Giving priority for a potential lock-holder increases lock
2289 * Since algorithm is based on heuristics, accessing another VCPU data without
2290 * locking does not harm. It may result in trying to yield to same VCPU, fail
2291 * and continue with next VCPU and so on.
2293 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2295 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2298 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2299 vcpu
->spin_loop
.dy_eligible
;
2301 if (vcpu
->spin_loop
.in_spin_loop
)
2302 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2310 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2312 struct kvm
*kvm
= me
->kvm
;
2313 struct kvm_vcpu
*vcpu
;
2314 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2320 kvm_vcpu_set_in_spin_loop(me
, true);
2322 * We boost the priority of a VCPU that is runnable but not
2323 * currently running, because it got preempted by something
2324 * else and called schedule in __vcpu_run. Hopefully that
2325 * VCPU is holding the lock that we need and will release it.
2326 * We approximate round-robin by starting at the last boosted VCPU.
2328 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2329 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2330 if (!pass
&& i
<= last_boosted_vcpu
) {
2331 i
= last_boosted_vcpu
;
2333 } else if (pass
&& i
> last_boosted_vcpu
)
2335 if (!ACCESS_ONCE(vcpu
->preempted
))
2339 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2341 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2344 yielded
= kvm_vcpu_yield_to(vcpu
);
2346 kvm
->last_boosted_vcpu
= i
;
2348 } else if (yielded
< 0) {
2355 kvm_vcpu_set_in_spin_loop(me
, false);
2357 /* Ensure vcpu is not eligible during next spinloop */
2358 kvm_vcpu_set_dy_eligible(me
, false);
2360 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2362 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2364 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2367 if (vmf
->pgoff
== 0)
2368 page
= virt_to_page(vcpu
->run
);
2370 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2371 page
= virt_to_page(vcpu
->arch
.pio_data
);
2373 #ifdef CONFIG_KVM_MMIO
2374 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2375 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2378 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2384 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2385 .fault
= kvm_vcpu_fault
,
2388 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2390 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2394 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2396 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2398 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2399 kvm_put_kvm(vcpu
->kvm
);
2403 static struct file_operations kvm_vcpu_fops
= {
2404 .release
= kvm_vcpu_release
,
2405 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2406 #ifdef CONFIG_KVM_COMPAT
2407 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2409 .mmap
= kvm_vcpu_mmap
,
2410 .llseek
= noop_llseek
,
2414 * Allocates an inode for the vcpu.
2416 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2418 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2421 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2423 char dir_name
[ITOA_MAX_LEN
* 2];
2426 if (!kvm_arch_has_vcpu_debugfs())
2429 if (!debugfs_initialized())
2432 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2433 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2434 vcpu
->kvm
->debugfs_dentry
);
2435 if (!vcpu
->debugfs_dentry
)
2438 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2440 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2448 * Creates some virtual cpus. Good luck creating more than one.
2450 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2453 struct kvm_vcpu
*vcpu
;
2455 if (id
>= KVM_MAX_VCPU_ID
)
2458 mutex_lock(&kvm
->lock
);
2459 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2460 mutex_unlock(&kvm
->lock
);
2464 kvm
->created_vcpus
++;
2465 mutex_unlock(&kvm
->lock
);
2467 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2470 goto vcpu_decrement
;
2473 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2475 r
= kvm_arch_vcpu_setup(vcpu
);
2479 r
= kvm_create_vcpu_debugfs(vcpu
);
2483 mutex_lock(&kvm
->lock
);
2484 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2486 goto unlock_vcpu_destroy
;
2489 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2491 /* Now it's all set up, let userspace reach it */
2493 r
= create_vcpu_fd(vcpu
);
2496 goto unlock_vcpu_destroy
;
2499 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2502 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2503 * before kvm->online_vcpu's incremented value.
2506 atomic_inc(&kvm
->online_vcpus
);
2508 mutex_unlock(&kvm
->lock
);
2509 kvm_arch_vcpu_postcreate(vcpu
);
2512 unlock_vcpu_destroy
:
2513 mutex_unlock(&kvm
->lock
);
2514 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2516 kvm_arch_vcpu_destroy(vcpu
);
2518 mutex_lock(&kvm
->lock
);
2519 kvm
->created_vcpus
--;
2520 mutex_unlock(&kvm
->lock
);
2524 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2527 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2528 vcpu
->sigset_active
= 1;
2529 vcpu
->sigset
= *sigset
;
2531 vcpu
->sigset_active
= 0;
2535 static long kvm_vcpu_ioctl(struct file
*filp
,
2536 unsigned int ioctl
, unsigned long arg
)
2538 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2539 void __user
*argp
= (void __user
*)arg
;
2541 struct kvm_fpu
*fpu
= NULL
;
2542 struct kvm_sregs
*kvm_sregs
= NULL
;
2544 if (vcpu
->kvm
->mm
!= current
->mm
)
2547 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2550 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2552 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2553 * so vcpu_load() would break it.
2555 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2556 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2560 r
= vcpu_load(vcpu
);
2568 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2569 /* The thread running this VCPU changed. */
2570 struct pid
*oldpid
= vcpu
->pid
;
2571 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2573 rcu_assign_pointer(vcpu
->pid
, newpid
);
2578 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2579 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2581 case KVM_GET_REGS
: {
2582 struct kvm_regs
*kvm_regs
;
2585 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2588 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2592 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2599 case KVM_SET_REGS
: {
2600 struct kvm_regs
*kvm_regs
;
2603 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2604 if (IS_ERR(kvm_regs
)) {
2605 r
= PTR_ERR(kvm_regs
);
2608 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2612 case KVM_GET_SREGS
: {
2613 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2617 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2621 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2626 case KVM_SET_SREGS
: {
2627 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2628 if (IS_ERR(kvm_sregs
)) {
2629 r
= PTR_ERR(kvm_sregs
);
2633 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2636 case KVM_GET_MP_STATE
: {
2637 struct kvm_mp_state mp_state
;
2639 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2643 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2648 case KVM_SET_MP_STATE
: {
2649 struct kvm_mp_state mp_state
;
2652 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2654 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2657 case KVM_TRANSLATE
: {
2658 struct kvm_translation tr
;
2661 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2663 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2667 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2672 case KVM_SET_GUEST_DEBUG
: {
2673 struct kvm_guest_debug dbg
;
2676 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2678 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2681 case KVM_SET_SIGNAL_MASK
: {
2682 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2683 struct kvm_signal_mask kvm_sigmask
;
2684 sigset_t sigset
, *p
;
2689 if (copy_from_user(&kvm_sigmask
, argp
,
2690 sizeof(kvm_sigmask
)))
2693 if (kvm_sigmask
.len
!= sizeof(sigset
))
2696 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2701 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2705 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2709 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2713 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2719 fpu
= memdup_user(argp
, sizeof(*fpu
));
2725 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2729 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2738 #ifdef CONFIG_KVM_COMPAT
2739 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2740 unsigned int ioctl
, unsigned long arg
)
2742 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2743 void __user
*argp
= compat_ptr(arg
);
2746 if (vcpu
->kvm
->mm
!= current
->mm
)
2750 case KVM_SET_SIGNAL_MASK
: {
2751 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2752 struct kvm_signal_mask kvm_sigmask
;
2753 compat_sigset_t csigset
;
2758 if (copy_from_user(&kvm_sigmask
, argp
,
2759 sizeof(kvm_sigmask
)))
2762 if (kvm_sigmask
.len
!= sizeof(csigset
))
2765 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2768 sigset_from_compat(&sigset
, &csigset
);
2769 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2771 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2775 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2783 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2784 int (*accessor
)(struct kvm_device
*dev
,
2785 struct kvm_device_attr
*attr
),
2788 struct kvm_device_attr attr
;
2793 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2796 return accessor(dev
, &attr
);
2799 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2802 struct kvm_device
*dev
= filp
->private_data
;
2805 case KVM_SET_DEVICE_ATTR
:
2806 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2807 case KVM_GET_DEVICE_ATTR
:
2808 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2809 case KVM_HAS_DEVICE_ATTR
:
2810 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2812 if (dev
->ops
->ioctl
)
2813 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2819 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2821 struct kvm_device
*dev
= filp
->private_data
;
2822 struct kvm
*kvm
= dev
->kvm
;
2828 static const struct file_operations kvm_device_fops
= {
2829 .unlocked_ioctl
= kvm_device_ioctl
,
2830 #ifdef CONFIG_KVM_COMPAT
2831 .compat_ioctl
= kvm_device_ioctl
,
2833 .release
= kvm_device_release
,
2836 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2838 if (filp
->f_op
!= &kvm_device_fops
)
2841 return filp
->private_data
;
2844 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2845 #ifdef CONFIG_KVM_MPIC
2846 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2847 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2850 #ifdef CONFIG_KVM_XICS
2851 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2855 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2857 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2860 if (kvm_device_ops_table
[type
] != NULL
)
2863 kvm_device_ops_table
[type
] = ops
;
2867 void kvm_unregister_device_ops(u32 type
)
2869 if (kvm_device_ops_table
[type
] != NULL
)
2870 kvm_device_ops_table
[type
] = NULL
;
2873 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2874 struct kvm_create_device
*cd
)
2876 struct kvm_device_ops
*ops
= NULL
;
2877 struct kvm_device
*dev
;
2878 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2881 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2884 ops
= kvm_device_ops_table
[cd
->type
];
2891 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2898 mutex_lock(&kvm
->lock
);
2899 ret
= ops
->create(dev
, cd
->type
);
2901 mutex_unlock(&kvm
->lock
);
2905 list_add(&dev
->vm_node
, &kvm
->devices
);
2906 mutex_unlock(&kvm
->lock
);
2911 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2913 mutex_lock(&kvm
->lock
);
2914 list_del(&dev
->vm_node
);
2915 mutex_unlock(&kvm
->lock
);
2925 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2928 case KVM_CAP_USER_MEMORY
:
2929 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2930 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2931 case KVM_CAP_INTERNAL_ERROR_DATA
:
2932 #ifdef CONFIG_HAVE_KVM_MSI
2933 case KVM_CAP_SIGNAL_MSI
:
2935 #ifdef CONFIG_HAVE_KVM_IRQFD
2937 case KVM_CAP_IRQFD_RESAMPLE
:
2939 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2940 case KVM_CAP_CHECK_EXTENSION_VM
:
2942 #ifdef CONFIG_KVM_MMIO
2943 case KVM_CAP_COALESCED_MMIO
:
2944 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2946 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2947 case KVM_CAP_IRQ_ROUTING
:
2948 return KVM_MAX_IRQ_ROUTES
;
2950 #if KVM_ADDRESS_SPACE_NUM > 1
2951 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2952 return KVM_ADDRESS_SPACE_NUM
;
2954 case KVM_CAP_MAX_VCPU_ID
:
2955 return KVM_MAX_VCPU_ID
;
2959 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2962 static long kvm_vm_ioctl(struct file
*filp
,
2963 unsigned int ioctl
, unsigned long arg
)
2965 struct kvm
*kvm
= filp
->private_data
;
2966 void __user
*argp
= (void __user
*)arg
;
2969 if (kvm
->mm
!= current
->mm
)
2972 case KVM_CREATE_VCPU
:
2973 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2975 case KVM_SET_USER_MEMORY_REGION
: {
2976 struct kvm_userspace_memory_region kvm_userspace_mem
;
2979 if (copy_from_user(&kvm_userspace_mem
, argp
,
2980 sizeof(kvm_userspace_mem
)))
2983 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2986 case KVM_GET_DIRTY_LOG
: {
2987 struct kvm_dirty_log log
;
2990 if (copy_from_user(&log
, argp
, sizeof(log
)))
2992 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2995 #ifdef CONFIG_KVM_MMIO
2996 case KVM_REGISTER_COALESCED_MMIO
: {
2997 struct kvm_coalesced_mmio_zone zone
;
3000 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3002 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3005 case KVM_UNREGISTER_COALESCED_MMIO
: {
3006 struct kvm_coalesced_mmio_zone zone
;
3009 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3011 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3016 struct kvm_irqfd data
;
3019 if (copy_from_user(&data
, argp
, sizeof(data
)))
3021 r
= kvm_irqfd(kvm
, &data
);
3024 case KVM_IOEVENTFD
: {
3025 struct kvm_ioeventfd data
;
3028 if (copy_from_user(&data
, argp
, sizeof(data
)))
3030 r
= kvm_ioeventfd(kvm
, &data
);
3033 #ifdef CONFIG_HAVE_KVM_MSI
3034 case KVM_SIGNAL_MSI
: {
3038 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3040 r
= kvm_send_userspace_msi(kvm
, &msi
);
3044 #ifdef __KVM_HAVE_IRQ_LINE
3045 case KVM_IRQ_LINE_STATUS
:
3046 case KVM_IRQ_LINE
: {
3047 struct kvm_irq_level irq_event
;
3050 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3053 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3054 ioctl
== KVM_IRQ_LINE_STATUS
);
3059 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3060 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3068 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3069 case KVM_SET_GSI_ROUTING
: {
3070 struct kvm_irq_routing routing
;
3071 struct kvm_irq_routing __user
*urouting
;
3072 struct kvm_irq_routing_entry
*entries
= NULL
;
3075 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3078 if (!kvm_arch_can_set_irq_routing(kvm
))
3080 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3086 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3091 if (copy_from_user(entries
, urouting
->entries
,
3092 routing
.nr
* sizeof(*entries
)))
3093 goto out_free_irq_routing
;
3095 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3097 out_free_irq_routing
:
3101 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3102 case KVM_CREATE_DEVICE
: {
3103 struct kvm_create_device cd
;
3106 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3109 r
= kvm_ioctl_create_device(kvm
, &cd
);
3114 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3120 case KVM_CHECK_EXTENSION
:
3121 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3124 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3130 #ifdef CONFIG_KVM_COMPAT
3131 struct compat_kvm_dirty_log
{
3135 compat_uptr_t dirty_bitmap
; /* one bit per page */
3140 static long kvm_vm_compat_ioctl(struct file
*filp
,
3141 unsigned int ioctl
, unsigned long arg
)
3143 struct kvm
*kvm
= filp
->private_data
;
3146 if (kvm
->mm
!= current
->mm
)
3149 case KVM_GET_DIRTY_LOG
: {
3150 struct compat_kvm_dirty_log compat_log
;
3151 struct kvm_dirty_log log
;
3153 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3154 sizeof(compat_log
)))
3156 log
.slot
= compat_log
.slot
;
3157 log
.padding1
= compat_log
.padding1
;
3158 log
.padding2
= compat_log
.padding2
;
3159 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3161 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3165 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3171 static struct file_operations kvm_vm_fops
= {
3172 .release
= kvm_vm_release
,
3173 .unlocked_ioctl
= kvm_vm_ioctl
,
3174 #ifdef CONFIG_KVM_COMPAT
3175 .compat_ioctl
= kvm_vm_compat_ioctl
,
3177 .llseek
= noop_llseek
,
3180 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3186 kvm
= kvm_create_vm(type
);
3188 return PTR_ERR(kvm
);
3189 #ifdef CONFIG_KVM_MMIO
3190 r
= kvm_coalesced_mmio_init(kvm
);
3196 r
= get_unused_fd_flags(O_CLOEXEC
);
3201 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3205 return PTR_ERR(file
);
3208 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3214 fd_install(r
, file
);
3218 static long kvm_dev_ioctl(struct file
*filp
,
3219 unsigned int ioctl
, unsigned long arg
)
3224 case KVM_GET_API_VERSION
:
3227 r
= KVM_API_VERSION
;
3230 r
= kvm_dev_ioctl_create_vm(arg
);
3232 case KVM_CHECK_EXTENSION
:
3233 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3235 case KVM_GET_VCPU_MMAP_SIZE
:
3238 r
= PAGE_SIZE
; /* struct kvm_run */
3240 r
+= PAGE_SIZE
; /* pio data page */
3242 #ifdef CONFIG_KVM_MMIO
3243 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3246 case KVM_TRACE_ENABLE
:
3247 case KVM_TRACE_PAUSE
:
3248 case KVM_TRACE_DISABLE
:
3252 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3258 static struct file_operations kvm_chardev_ops
= {
3259 .unlocked_ioctl
= kvm_dev_ioctl
,
3260 .compat_ioctl
= kvm_dev_ioctl
,
3261 .llseek
= noop_llseek
,
3264 static struct miscdevice kvm_dev
= {
3270 static void hardware_enable_nolock(void *junk
)
3272 int cpu
= raw_smp_processor_id();
3275 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3278 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3280 r
= kvm_arch_hardware_enable();
3283 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3284 atomic_inc(&hardware_enable_failed
);
3285 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3289 static int kvm_starting_cpu(unsigned int cpu
)
3291 raw_spin_lock(&kvm_count_lock
);
3292 if (kvm_usage_count
)
3293 hardware_enable_nolock(NULL
);
3294 raw_spin_unlock(&kvm_count_lock
);
3298 static void hardware_disable_nolock(void *junk
)
3300 int cpu
= raw_smp_processor_id();
3302 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3304 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3305 kvm_arch_hardware_disable();
3308 static int kvm_dying_cpu(unsigned int cpu
)
3310 raw_spin_lock(&kvm_count_lock
);
3311 if (kvm_usage_count
)
3312 hardware_disable_nolock(NULL
);
3313 raw_spin_unlock(&kvm_count_lock
);
3317 static void hardware_disable_all_nolock(void)
3319 BUG_ON(!kvm_usage_count
);
3322 if (!kvm_usage_count
)
3323 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3326 static void hardware_disable_all(void)
3328 raw_spin_lock(&kvm_count_lock
);
3329 hardware_disable_all_nolock();
3330 raw_spin_unlock(&kvm_count_lock
);
3333 static int hardware_enable_all(void)
3337 raw_spin_lock(&kvm_count_lock
);
3340 if (kvm_usage_count
== 1) {
3341 atomic_set(&hardware_enable_failed
, 0);
3342 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3344 if (atomic_read(&hardware_enable_failed
)) {
3345 hardware_disable_all_nolock();
3350 raw_spin_unlock(&kvm_count_lock
);
3355 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3359 * Some (well, at least mine) BIOSes hang on reboot if
3362 * And Intel TXT required VMX off for all cpu when system shutdown.
3364 pr_info("kvm: exiting hardware virtualization\n");
3365 kvm_rebooting
= true;
3366 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3370 static struct notifier_block kvm_reboot_notifier
= {
3371 .notifier_call
= kvm_reboot
,
3375 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3379 for (i
= 0; i
< bus
->dev_count
; i
++) {
3380 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3382 kvm_iodevice_destructor(pos
);
3387 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3388 const struct kvm_io_range
*r2
)
3390 gpa_t addr1
= r1
->addr
;
3391 gpa_t addr2
= r2
->addr
;
3396 /* If r2->len == 0, match the exact address. If r2->len != 0,
3397 * accept any overlapping write. Any order is acceptable for
3398 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3399 * we process all of them.
3412 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3414 return kvm_io_bus_cmp(p1
, p2
);
3417 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3418 gpa_t addr
, int len
)
3420 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3426 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3427 kvm_io_bus_sort_cmp
, NULL
);
3432 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3433 gpa_t addr
, int len
)
3435 struct kvm_io_range
*range
, key
;
3438 key
= (struct kvm_io_range
) {
3443 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3444 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3448 off
= range
- bus
->range
;
3450 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3456 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3457 struct kvm_io_range
*range
, const void *val
)
3461 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3465 while (idx
< bus
->dev_count
&&
3466 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3467 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3476 /* kvm_io_bus_write - called under kvm->slots_lock */
3477 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3478 int len
, const void *val
)
3480 struct kvm_io_bus
*bus
;
3481 struct kvm_io_range range
;
3484 range
= (struct kvm_io_range
) {
3489 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3492 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3493 return r
< 0 ? r
: 0;
3496 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3497 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3498 gpa_t addr
, int len
, const void *val
, long cookie
)
3500 struct kvm_io_bus
*bus
;
3501 struct kvm_io_range range
;
3503 range
= (struct kvm_io_range
) {
3508 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3512 /* First try the device referenced by cookie. */
3513 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3514 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3515 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3520 * cookie contained garbage; fall back to search and return the
3521 * correct cookie value.
3523 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3526 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3527 struct kvm_io_range
*range
, void *val
)
3531 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3535 while (idx
< bus
->dev_count
&&
3536 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3537 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3545 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3547 /* kvm_io_bus_read - called under kvm->slots_lock */
3548 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3551 struct kvm_io_bus
*bus
;
3552 struct kvm_io_range range
;
3555 range
= (struct kvm_io_range
) {
3560 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3563 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3564 return r
< 0 ? r
: 0;
3568 /* Caller must hold slots_lock. */
3569 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3570 int len
, struct kvm_io_device
*dev
)
3572 struct kvm_io_bus
*new_bus
, *bus
;
3574 bus
= kvm
->buses
[bus_idx
];
3578 /* exclude ioeventfd which is limited by maximum fd */
3579 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3582 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3583 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3586 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3587 sizeof(struct kvm_io_range
)));
3588 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3589 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3590 synchronize_srcu_expedited(&kvm
->srcu
);
3596 /* Caller must hold slots_lock. */
3597 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3598 struct kvm_io_device
*dev
)
3601 struct kvm_io_bus
*new_bus
, *bus
;
3603 bus
= kvm
->buses
[bus_idx
];
3607 for (i
= 0; i
< bus
->dev_count
; i
++)
3608 if (bus
->range
[i
].dev
== dev
) {
3612 if (i
== bus
->dev_count
)
3615 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3616 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3618 pr_err("kvm: failed to shrink bus, removing it completely\n");
3622 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3623 new_bus
->dev_count
--;
3624 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3625 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3628 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3629 synchronize_srcu_expedited(&kvm
->srcu
);
3634 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3637 struct kvm_io_bus
*bus
;
3638 int dev_idx
, srcu_idx
;
3639 struct kvm_io_device
*iodev
= NULL
;
3641 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3643 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3647 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3651 iodev
= bus
->range
[dev_idx
].dev
;
3654 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3658 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3660 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3661 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3664 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3667 /* The debugfs files are a reference to the kvm struct which
3668 * is still valid when kvm_destroy_vm is called.
3669 * To avoid the race between open and the removal of the debugfs
3670 * directory we test against the users count.
3672 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3675 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3676 kvm_put_kvm(stat_data
->kvm
);
3683 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3685 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3688 simple_attr_release(inode
, file
);
3689 kvm_put_kvm(stat_data
->kvm
);
3694 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3696 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3698 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3703 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3705 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3710 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3715 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3717 __simple_attr_check_format("%llu\n", 0ull);
3718 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3719 vm_stat_clear_per_vm
, "%llu\n");
3722 static const struct file_operations vm_stat_get_per_vm_fops
= {
3723 .owner
= THIS_MODULE
,
3724 .open
= vm_stat_get_per_vm_open
,
3725 .release
= kvm_debugfs_release
,
3726 .read
= simple_attr_read
,
3727 .write
= simple_attr_write
,
3728 .llseek
= generic_file_llseek
,
3731 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3734 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3735 struct kvm_vcpu
*vcpu
;
3739 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3740 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3745 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3748 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3749 struct kvm_vcpu
*vcpu
;
3754 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3755 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3760 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3762 __simple_attr_check_format("%llu\n", 0ull);
3763 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3764 vcpu_stat_clear_per_vm
, "%llu\n");
3767 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3768 .owner
= THIS_MODULE
,
3769 .open
= vcpu_stat_get_per_vm_open
,
3770 .release
= kvm_debugfs_release
,
3771 .read
= simple_attr_read
,
3772 .write
= simple_attr_write
,
3773 .llseek
= generic_file_llseek
,
3776 static const struct file_operations
*stat_fops_per_vm
[] = {
3777 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3778 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3781 static int vm_stat_get(void *_offset
, u64
*val
)
3783 unsigned offset
= (long)_offset
;
3785 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3789 spin_lock(&kvm_lock
);
3790 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3792 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3795 spin_unlock(&kvm_lock
);
3799 static int vm_stat_clear(void *_offset
, u64 val
)
3801 unsigned offset
= (long)_offset
;
3803 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3808 spin_lock(&kvm_lock
);
3809 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3811 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3813 spin_unlock(&kvm_lock
);
3818 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3820 static int vcpu_stat_get(void *_offset
, u64
*val
)
3822 unsigned offset
= (long)_offset
;
3824 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3828 spin_lock(&kvm_lock
);
3829 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3831 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3834 spin_unlock(&kvm_lock
);
3838 static int vcpu_stat_clear(void *_offset
, u64 val
)
3840 unsigned offset
= (long)_offset
;
3842 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3847 spin_lock(&kvm_lock
);
3848 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3850 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3852 spin_unlock(&kvm_lock
);
3857 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3860 static const struct file_operations
*stat_fops
[] = {
3861 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3862 [KVM_STAT_VM
] = &vm_stat_fops
,
3865 static int kvm_init_debug(void)
3868 struct kvm_stats_debugfs_item
*p
;
3870 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3871 if (kvm_debugfs_dir
== NULL
)
3874 kvm_debugfs_num_entries
= 0;
3875 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3876 if (!debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3877 (void *)(long)p
->offset
,
3878 stat_fops
[p
->kind
]))
3885 debugfs_remove_recursive(kvm_debugfs_dir
);
3890 static int kvm_suspend(void)
3892 if (kvm_usage_count
)
3893 hardware_disable_nolock(NULL
);
3897 static void kvm_resume(void)
3899 if (kvm_usage_count
) {
3900 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3901 hardware_enable_nolock(NULL
);
3905 static struct syscore_ops kvm_syscore_ops
= {
3906 .suspend
= kvm_suspend
,
3907 .resume
= kvm_resume
,
3911 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3913 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3916 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3918 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3920 if (vcpu
->preempted
)
3921 vcpu
->preempted
= false;
3923 kvm_arch_sched_in(vcpu
, cpu
);
3925 kvm_arch_vcpu_load(vcpu
, cpu
);
3928 static void kvm_sched_out(struct preempt_notifier
*pn
,
3929 struct task_struct
*next
)
3931 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3933 if (current
->state
== TASK_RUNNING
)
3934 vcpu
->preempted
= true;
3935 kvm_arch_vcpu_put(vcpu
);
3938 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3939 struct module
*module
)
3944 r
= kvm_arch_init(opaque
);
3949 * kvm_arch_init makes sure there's at most one caller
3950 * for architectures that support multiple implementations,
3951 * like intel and amd on x86.
3952 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3953 * conflicts in case kvm is already setup for another implementation.
3955 r
= kvm_irqfd_init();
3959 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3964 r
= kvm_arch_hardware_setup();
3968 for_each_online_cpu(cpu
) {
3969 smp_call_function_single(cpu
,
3970 kvm_arch_check_processor_compat
,
3976 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
3977 kvm_starting_cpu
, kvm_dying_cpu
);
3980 register_reboot_notifier(&kvm_reboot_notifier
);
3982 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3984 vcpu_align
= __alignof__(struct kvm_vcpu
);
3985 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3987 if (!kvm_vcpu_cache
) {
3992 r
= kvm_async_pf_init();
3996 kvm_chardev_ops
.owner
= module
;
3997 kvm_vm_fops
.owner
= module
;
3998 kvm_vcpu_fops
.owner
= module
;
4000 r
= misc_register(&kvm_dev
);
4002 pr_err("kvm: misc device register failed\n");
4006 register_syscore_ops(&kvm_syscore_ops
);
4008 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4009 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4011 r
= kvm_init_debug();
4013 pr_err("kvm: create debugfs files failed\n");
4017 r
= kvm_vfio_ops_init();
4023 unregister_syscore_ops(&kvm_syscore_ops
);
4024 misc_deregister(&kvm_dev
);
4026 kvm_async_pf_deinit();
4028 kmem_cache_destroy(kvm_vcpu_cache
);
4030 unregister_reboot_notifier(&kvm_reboot_notifier
);
4031 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4034 kvm_arch_hardware_unsetup();
4036 free_cpumask_var(cpus_hardware_enabled
);
4044 EXPORT_SYMBOL_GPL(kvm_init
);
4048 debugfs_remove_recursive(kvm_debugfs_dir
);
4049 misc_deregister(&kvm_dev
);
4050 kmem_cache_destroy(kvm_vcpu_cache
);
4051 kvm_async_pf_deinit();
4052 unregister_syscore_ops(&kvm_syscore_ops
);
4053 unregister_reboot_notifier(&kvm_reboot_notifier
);
4054 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4055 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4056 kvm_arch_hardware_unsetup();
4059 free_cpumask_var(cpus_hardware_enabled
);
4060 kvm_vfio_ops_exit();
4062 EXPORT_SYMBOL_GPL(kvm_exit
);