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
, 0644);
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
, 0644);
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
, 0644);
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 #define KVM_EVENT_CREATE_VM 0
134 #define KVM_EVENT_DESTROY_VM 1
135 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
136 static unsigned long long kvm_createvm_count
;
137 static unsigned long long kvm_active_vms
;
139 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
142 return PageReserved(pfn_to_page(pfn
));
148 * Switches to specified vcpu, until a matching vcpu_put()
150 int vcpu_load(struct kvm_vcpu
*vcpu
)
154 if (mutex_lock_killable(&vcpu
->mutex
))
157 preempt_notifier_register(&vcpu
->preempt_notifier
);
158 kvm_arch_vcpu_load(vcpu
, cpu
);
162 EXPORT_SYMBOL_GPL(vcpu_load
);
164 void vcpu_put(struct kvm_vcpu
*vcpu
)
167 kvm_arch_vcpu_put(vcpu
);
168 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
170 mutex_unlock(&vcpu
->mutex
);
172 EXPORT_SYMBOL_GPL(vcpu_put
);
174 /* TODO: merge with kvm_arch_vcpu_should_kick */
175 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
177 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
180 * We need to wait for the VCPU to reenable interrupts and get out of
181 * READING_SHADOW_PAGE_TABLES mode.
183 if (req
& KVM_REQUEST_WAIT
)
184 return mode
!= OUTSIDE_GUEST_MODE
;
187 * Need to kick a running VCPU, but otherwise there is nothing to do.
189 return mode
== IN_GUEST_MODE
;
192 static void ack_flush(void *_completed
)
196 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
199 cpus
= cpu_online_mask
;
201 if (cpumask_empty(cpus
))
204 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
208 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
213 struct kvm_vcpu
*vcpu
;
215 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
218 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
219 kvm_make_request(req
, vcpu
);
222 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
225 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
226 kvm_request_needs_ipi(vcpu
, req
))
227 __cpumask_set_cpu(cpu
, cpus
);
229 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
231 free_cpumask_var(cpus
);
235 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
236 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
239 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
240 * kvm_make_all_cpus_request.
242 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
245 * We want to publish modifications to the page tables before reading
246 * mode. Pairs with a memory barrier in arch-specific code.
247 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
248 * and smp_mb in walk_shadow_page_lockless_begin/end.
249 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
251 * There is already an smp_mb__after_atomic() before
252 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
255 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
256 ++kvm
->stat
.remote_tlb_flush
;
257 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
259 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
262 void kvm_reload_remote_mmus(struct kvm
*kvm
)
264 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
267 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
272 mutex_init(&vcpu
->mutex
);
277 init_swait_queue_head(&vcpu
->wq
);
278 kvm_async_pf_vcpu_init(vcpu
);
281 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
283 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
288 vcpu
->run
= page_address(page
);
290 kvm_vcpu_set_in_spin_loop(vcpu
, false);
291 kvm_vcpu_set_dy_eligible(vcpu
, false);
292 vcpu
->preempted
= false;
294 r
= kvm_arch_vcpu_init(vcpu
);
300 free_page((unsigned long)vcpu
->run
);
304 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
306 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
309 * no need for rcu_read_lock as VCPU_RUN is the only place that
310 * will change the vcpu->pid pointer and on uninit all file
311 * descriptors are already gone.
313 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
314 kvm_arch_vcpu_uninit(vcpu
);
315 free_page((unsigned long)vcpu
->run
);
317 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
319 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
320 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
322 return container_of(mn
, struct kvm
, mmu_notifier
);
325 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
326 struct mm_struct
*mm
,
327 unsigned long address
,
330 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
333 idx
= srcu_read_lock(&kvm
->srcu
);
334 spin_lock(&kvm
->mmu_lock
);
335 kvm
->mmu_notifier_seq
++;
336 kvm_set_spte_hva(kvm
, address
, pte
);
337 spin_unlock(&kvm
->mmu_lock
);
338 srcu_read_unlock(&kvm
->srcu
, idx
);
341 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
342 struct mm_struct
*mm
,
346 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
347 int need_tlb_flush
= 0, idx
;
349 idx
= srcu_read_lock(&kvm
->srcu
);
350 spin_lock(&kvm
->mmu_lock
);
352 * The count increase must become visible at unlock time as no
353 * spte can be established without taking the mmu_lock and
354 * count is also read inside the mmu_lock critical section.
356 kvm
->mmu_notifier_count
++;
357 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
358 need_tlb_flush
|= kvm
->tlbs_dirty
;
359 /* we've to flush the tlb before the pages can be freed */
361 kvm_flush_remote_tlbs(kvm
);
363 spin_unlock(&kvm
->mmu_lock
);
364 srcu_read_unlock(&kvm
->srcu
, idx
);
367 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
368 struct mm_struct
*mm
,
372 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
374 spin_lock(&kvm
->mmu_lock
);
376 * This sequence increase will notify the kvm page fault that
377 * the page that is going to be mapped in the spte could have
380 kvm
->mmu_notifier_seq
++;
383 * The above sequence increase must be visible before the
384 * below count decrease, which is ensured by the smp_wmb above
385 * in conjunction with the smp_rmb in mmu_notifier_retry().
387 kvm
->mmu_notifier_count
--;
388 spin_unlock(&kvm
->mmu_lock
);
390 BUG_ON(kvm
->mmu_notifier_count
< 0);
393 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
394 struct mm_struct
*mm
,
398 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
401 idx
= srcu_read_lock(&kvm
->srcu
);
402 spin_lock(&kvm
->mmu_lock
);
404 young
= kvm_age_hva(kvm
, start
, end
);
406 kvm_flush_remote_tlbs(kvm
);
408 spin_unlock(&kvm
->mmu_lock
);
409 srcu_read_unlock(&kvm
->srcu
, idx
);
414 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
415 struct mm_struct
*mm
,
419 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
422 idx
= srcu_read_lock(&kvm
->srcu
);
423 spin_lock(&kvm
->mmu_lock
);
425 * Even though we do not flush TLB, this will still adversely
426 * affect performance on pre-Haswell Intel EPT, where there is
427 * no EPT Access Bit to clear so that we have to tear down EPT
428 * tables instead. If we find this unacceptable, we can always
429 * add a parameter to kvm_age_hva so that it effectively doesn't
430 * do anything on clear_young.
432 * Also note that currently we never issue secondary TLB flushes
433 * from clear_young, leaving this job up to the regular system
434 * cadence. If we find this inaccurate, we might come up with a
435 * more sophisticated heuristic later.
437 young
= kvm_age_hva(kvm
, start
, end
);
438 spin_unlock(&kvm
->mmu_lock
);
439 srcu_read_unlock(&kvm
->srcu
, idx
);
444 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
445 struct mm_struct
*mm
,
446 unsigned long address
)
448 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
451 idx
= srcu_read_lock(&kvm
->srcu
);
452 spin_lock(&kvm
->mmu_lock
);
453 young
= kvm_test_age_hva(kvm
, address
);
454 spin_unlock(&kvm
->mmu_lock
);
455 srcu_read_unlock(&kvm
->srcu
, idx
);
460 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
461 struct mm_struct
*mm
)
463 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
466 idx
= srcu_read_lock(&kvm
->srcu
);
467 kvm_arch_flush_shadow_all(kvm
);
468 srcu_read_unlock(&kvm
->srcu
, idx
);
471 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
472 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
473 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
474 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
475 .clear_young
= kvm_mmu_notifier_clear_young
,
476 .test_young
= kvm_mmu_notifier_test_young
,
477 .change_pte
= kvm_mmu_notifier_change_pte
,
478 .release
= kvm_mmu_notifier_release
,
481 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
483 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
484 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
487 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
489 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
494 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
496 static struct kvm_memslots
*kvm_alloc_memslots(void)
499 struct kvm_memslots
*slots
;
501 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
505 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
506 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
511 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
513 if (!memslot
->dirty_bitmap
)
516 kvfree(memslot
->dirty_bitmap
);
517 memslot
->dirty_bitmap
= NULL
;
521 * Free any memory in @free but not in @dont.
523 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
524 struct kvm_memory_slot
*dont
)
526 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
527 kvm_destroy_dirty_bitmap(free
);
529 kvm_arch_free_memslot(kvm
, free
, dont
);
534 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
536 struct kvm_memory_slot
*memslot
;
541 kvm_for_each_memslot(memslot
, slots
)
542 kvm_free_memslot(kvm
, memslot
, NULL
);
547 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
551 if (!kvm
->debugfs_dentry
)
554 debugfs_remove_recursive(kvm
->debugfs_dentry
);
556 if (kvm
->debugfs_stat_data
) {
557 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
558 kfree(kvm
->debugfs_stat_data
[i
]);
559 kfree(kvm
->debugfs_stat_data
);
563 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
565 char dir_name
[ITOA_MAX_LEN
* 2];
566 struct kvm_stat_data
*stat_data
;
567 struct kvm_stats_debugfs_item
*p
;
569 if (!debugfs_initialized())
572 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
573 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
575 if (!kvm
->debugfs_dentry
)
578 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
579 sizeof(*kvm
->debugfs_stat_data
),
581 if (!kvm
->debugfs_stat_data
)
584 for (p
= debugfs_entries
; p
->name
; p
++) {
585 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
589 stat_data
->kvm
= kvm
;
590 stat_data
->offset
= p
->offset
;
591 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
592 if (!debugfs_create_file(p
->name
, 0644,
595 stat_fops_per_vm
[p
->kind
]))
601 static struct kvm
*kvm_create_vm(unsigned long type
)
604 struct kvm
*kvm
= kvm_arch_alloc_vm();
607 return ERR_PTR(-ENOMEM
);
609 spin_lock_init(&kvm
->mmu_lock
);
611 kvm
->mm
= current
->mm
;
612 kvm_eventfd_init(kvm
);
613 mutex_init(&kvm
->lock
);
614 mutex_init(&kvm
->irq_lock
);
615 mutex_init(&kvm
->slots_lock
);
616 refcount_set(&kvm
->users_count
, 1);
617 INIT_LIST_HEAD(&kvm
->devices
);
619 r
= kvm_arch_init_vm(kvm
, type
);
621 goto out_err_no_disable
;
623 r
= hardware_enable_all();
625 goto out_err_no_disable
;
627 #ifdef CONFIG_HAVE_KVM_IRQFD
628 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
631 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
634 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
635 struct kvm_memslots
*slots
= kvm_alloc_memslots();
637 goto out_err_no_srcu
;
639 * Generations must be different for each address space.
640 * Init kvm generation close to the maximum to easily test the
641 * code of handling generation number wrap-around.
643 slots
->generation
= i
* 2 - 150;
644 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
647 if (init_srcu_struct(&kvm
->srcu
))
648 goto out_err_no_srcu
;
649 if (init_srcu_struct(&kvm
->irq_srcu
))
650 goto out_err_no_irq_srcu
;
651 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
652 rcu_assign_pointer(kvm
->buses
[i
],
653 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
658 r
= kvm_init_mmu_notifier(kvm
);
662 spin_lock(&kvm_lock
);
663 list_add(&kvm
->vm_list
, &vm_list
);
664 spin_unlock(&kvm_lock
);
666 preempt_notifier_inc();
671 cleanup_srcu_struct(&kvm
->irq_srcu
);
673 cleanup_srcu_struct(&kvm
->srcu
);
675 hardware_disable_all();
677 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
678 kfree(kvm_get_bus(kvm
, i
));
679 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
680 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
681 kvm_arch_free_vm(kvm
);
686 static void kvm_destroy_devices(struct kvm
*kvm
)
688 struct kvm_device
*dev
, *tmp
;
691 * We do not need to take the kvm->lock here, because nobody else
692 * has a reference to the struct kvm at this point and therefore
693 * cannot access the devices list anyhow.
695 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
696 list_del(&dev
->vm_node
);
697 dev
->ops
->destroy(dev
);
701 static void kvm_destroy_vm(struct kvm
*kvm
)
704 struct mm_struct
*mm
= kvm
->mm
;
706 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
707 kvm_destroy_vm_debugfs(kvm
);
708 kvm_arch_sync_events(kvm
);
709 spin_lock(&kvm_lock
);
710 list_del(&kvm
->vm_list
);
711 spin_unlock(&kvm_lock
);
712 kvm_free_irq_routing(kvm
);
713 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
714 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
717 kvm_io_bus_destroy(bus
);
718 kvm
->buses
[i
] = NULL
;
720 kvm_coalesced_mmio_free(kvm
);
721 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
722 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
724 kvm_arch_flush_shadow_all(kvm
);
726 kvm_arch_destroy_vm(kvm
);
727 kvm_destroy_devices(kvm
);
728 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
729 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
730 cleanup_srcu_struct(&kvm
->irq_srcu
);
731 cleanup_srcu_struct(&kvm
->srcu
);
732 kvm_arch_free_vm(kvm
);
733 preempt_notifier_dec();
734 hardware_disable_all();
738 void kvm_get_kvm(struct kvm
*kvm
)
740 refcount_inc(&kvm
->users_count
);
742 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
744 void kvm_put_kvm(struct kvm
*kvm
)
746 if (refcount_dec_and_test(&kvm
->users_count
))
749 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
752 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
754 struct kvm
*kvm
= filp
->private_data
;
756 kvm_irqfd_release(kvm
);
763 * Allocation size is twice as large as the actual dirty bitmap size.
764 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
766 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
768 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
770 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
771 if (!memslot
->dirty_bitmap
)
778 * Insert memslot and re-sort memslots based on their GFN,
779 * so binary search could be used to lookup GFN.
780 * Sorting algorithm takes advantage of having initially
781 * sorted array and known changed memslot position.
783 static void update_memslots(struct kvm_memslots
*slots
,
784 struct kvm_memory_slot
*new)
787 int i
= slots
->id_to_index
[id
];
788 struct kvm_memory_slot
*mslots
= slots
->memslots
;
790 WARN_ON(mslots
[i
].id
!= id
);
792 WARN_ON(!mslots
[i
].npages
);
793 if (mslots
[i
].npages
)
796 if (!mslots
[i
].npages
)
800 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
801 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
802 if (!mslots
[i
+ 1].npages
)
804 mslots
[i
] = mslots
[i
+ 1];
805 slots
->id_to_index
[mslots
[i
].id
] = i
;
810 * The ">=" is needed when creating a slot with base_gfn == 0,
811 * so that it moves before all those with base_gfn == npages == 0.
813 * On the other hand, if new->npages is zero, the above loop has
814 * already left i pointing to the beginning of the empty part of
815 * mslots, and the ">=" would move the hole backwards in this
816 * case---which is wrong. So skip the loop when deleting a slot.
820 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
821 mslots
[i
] = mslots
[i
- 1];
822 slots
->id_to_index
[mslots
[i
].id
] = i
;
826 WARN_ON_ONCE(i
!= slots
->used_slots
);
829 slots
->id_to_index
[mslots
[i
].id
] = i
;
832 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
834 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
836 #ifdef __KVM_HAVE_READONLY_MEM
837 valid_flags
|= KVM_MEM_READONLY
;
840 if (mem
->flags
& ~valid_flags
)
846 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
847 int as_id
, struct kvm_memslots
*slots
)
849 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
852 * Set the low bit in the generation, which disables SPTE caching
853 * until the end of synchronize_srcu_expedited.
855 WARN_ON(old_memslots
->generation
& 1);
856 slots
->generation
= old_memslots
->generation
+ 1;
858 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
859 synchronize_srcu_expedited(&kvm
->srcu
);
862 * Increment the new memslot generation a second time. This prevents
863 * vm exits that race with memslot updates from caching a memslot
864 * generation that will (potentially) be valid forever.
866 * Generations must be unique even across address spaces. We do not need
867 * a global counter for that, instead the generation space is evenly split
868 * across address spaces. For example, with two address spaces, address
869 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
870 * use generations 2, 6, 10, 14, ...
872 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
874 kvm_arch_memslots_updated(kvm
, slots
);
880 * Allocate some memory and give it an address in the guest physical address
883 * Discontiguous memory is allowed, mostly for framebuffers.
885 * Must be called holding kvm->slots_lock for write.
887 int __kvm_set_memory_region(struct kvm
*kvm
,
888 const struct kvm_userspace_memory_region
*mem
)
892 unsigned long npages
;
893 struct kvm_memory_slot
*slot
;
894 struct kvm_memory_slot old
, new;
895 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
897 enum kvm_mr_change change
;
899 r
= check_memory_region_flags(mem
);
904 as_id
= mem
->slot
>> 16;
907 /* General sanity checks */
908 if (mem
->memory_size
& (PAGE_SIZE
- 1))
910 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
912 /* We can read the guest memory with __xxx_user() later on. */
913 if ((id
< KVM_USER_MEM_SLOTS
) &&
914 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
915 !access_ok(VERIFY_WRITE
,
916 (void __user
*)(unsigned long)mem
->userspace_addr
,
919 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
921 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
924 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
925 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
926 npages
= mem
->memory_size
>> PAGE_SHIFT
;
928 if (npages
> KVM_MEM_MAX_NR_PAGES
)
934 new.base_gfn
= base_gfn
;
936 new.flags
= mem
->flags
;
940 change
= KVM_MR_CREATE
;
941 else { /* Modify an existing slot. */
942 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
943 (npages
!= old
.npages
) ||
944 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
947 if (base_gfn
!= old
.base_gfn
)
948 change
= KVM_MR_MOVE
;
949 else if (new.flags
!= old
.flags
)
950 change
= KVM_MR_FLAGS_ONLY
;
951 else { /* Nothing to change. */
960 change
= KVM_MR_DELETE
;
965 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
966 /* Check for overlaps */
968 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
969 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
972 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
973 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
978 /* Free page dirty bitmap if unneeded */
979 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
980 new.dirty_bitmap
= NULL
;
983 if (change
== KVM_MR_CREATE
) {
984 new.userspace_addr
= mem
->userspace_addr
;
986 if (kvm_arch_create_memslot(kvm
, &new, npages
))
990 /* Allocate page dirty bitmap if needed */
991 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
992 if (kvm_create_dirty_bitmap(&new) < 0)
996 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
999 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1001 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1002 slot
= id_to_memslot(slots
, id
);
1003 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1005 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1007 /* From this point no new shadow pages pointing to a deleted,
1008 * or moved, memslot will be created.
1010 * validation of sp->gfn happens in:
1011 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1012 * - kvm_is_visible_gfn (mmu_check_roots)
1014 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1017 * We can re-use the old_memslots from above, the only difference
1018 * from the currently installed memslots is the invalid flag. This
1019 * will get overwritten by update_memslots anyway.
1021 slots
= old_memslots
;
1024 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1028 /* actual memory is freed via old in kvm_free_memslot below */
1029 if (change
== KVM_MR_DELETE
) {
1030 new.dirty_bitmap
= NULL
;
1031 memset(&new.arch
, 0, sizeof(new.arch
));
1034 update_memslots(slots
, &new);
1035 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1037 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1039 kvm_free_memslot(kvm
, &old
, &new);
1040 kvfree(old_memslots
);
1046 kvm_free_memslot(kvm
, &new, &old
);
1050 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1052 int kvm_set_memory_region(struct kvm
*kvm
,
1053 const struct kvm_userspace_memory_region
*mem
)
1057 mutex_lock(&kvm
->slots_lock
);
1058 r
= __kvm_set_memory_region(kvm
, mem
);
1059 mutex_unlock(&kvm
->slots_lock
);
1062 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1064 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1065 struct kvm_userspace_memory_region
*mem
)
1067 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1070 return kvm_set_memory_region(kvm
, mem
);
1073 int kvm_get_dirty_log(struct kvm
*kvm
,
1074 struct kvm_dirty_log
*log
, int *is_dirty
)
1076 struct kvm_memslots
*slots
;
1077 struct kvm_memory_slot
*memslot
;
1080 unsigned long any
= 0;
1082 as_id
= log
->slot
>> 16;
1083 id
= (u16
)log
->slot
;
1084 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1087 slots
= __kvm_memslots(kvm
, as_id
);
1088 memslot
= id_to_memslot(slots
, id
);
1089 if (!memslot
->dirty_bitmap
)
1092 n
= kvm_dirty_bitmap_bytes(memslot
);
1094 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1095 any
= memslot
->dirty_bitmap
[i
];
1097 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1104 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1106 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1108 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1109 * are dirty write protect them for next write.
1110 * @kvm: pointer to kvm instance
1111 * @log: slot id and address to which we copy the log
1112 * @is_dirty: flag set if any page is dirty
1114 * We need to keep it in mind that VCPU threads can write to the bitmap
1115 * concurrently. So, to avoid losing track of dirty pages we keep the
1118 * 1. Take a snapshot of the bit and clear it if needed.
1119 * 2. Write protect the corresponding page.
1120 * 3. Copy the snapshot to the userspace.
1121 * 4. Upon return caller flushes TLB's if needed.
1123 * Between 2 and 4, the guest may write to the page using the remaining TLB
1124 * entry. This is not a problem because the page is reported dirty using
1125 * the snapshot taken before and step 4 ensures that writes done after
1126 * exiting to userspace will be logged for the next call.
1129 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1130 struct kvm_dirty_log
*log
, bool *is_dirty
)
1132 struct kvm_memslots
*slots
;
1133 struct kvm_memory_slot
*memslot
;
1136 unsigned long *dirty_bitmap
;
1137 unsigned long *dirty_bitmap_buffer
;
1139 as_id
= log
->slot
>> 16;
1140 id
= (u16
)log
->slot
;
1141 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1144 slots
= __kvm_memslots(kvm
, as_id
);
1145 memslot
= id_to_memslot(slots
, id
);
1147 dirty_bitmap
= memslot
->dirty_bitmap
;
1151 n
= kvm_dirty_bitmap_bytes(memslot
);
1153 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1154 memset(dirty_bitmap_buffer
, 0, n
);
1156 spin_lock(&kvm
->mmu_lock
);
1158 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1162 if (!dirty_bitmap
[i
])
1167 mask
= xchg(&dirty_bitmap
[i
], 0);
1168 dirty_bitmap_buffer
[i
] = mask
;
1171 offset
= i
* BITS_PER_LONG
;
1172 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1177 spin_unlock(&kvm
->mmu_lock
);
1178 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1182 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1185 bool kvm_largepages_enabled(void)
1187 return largepages_enabled
;
1190 void kvm_disable_largepages(void)
1192 largepages_enabled
= false;
1194 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1196 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1198 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1200 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1202 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1204 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1207 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1209 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1211 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1212 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1217 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1219 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1221 struct vm_area_struct
*vma
;
1222 unsigned long addr
, size
;
1226 addr
= gfn_to_hva(kvm
, gfn
);
1227 if (kvm_is_error_hva(addr
))
1230 down_read(¤t
->mm
->mmap_sem
);
1231 vma
= find_vma(current
->mm
, addr
);
1235 size
= vma_kernel_pagesize(vma
);
1238 up_read(¤t
->mm
->mmap_sem
);
1243 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1245 return slot
->flags
& KVM_MEM_READONLY
;
1248 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1249 gfn_t
*nr_pages
, bool write
)
1251 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1252 return KVM_HVA_ERR_BAD
;
1254 if (memslot_is_readonly(slot
) && write
)
1255 return KVM_HVA_ERR_RO_BAD
;
1258 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1260 return __gfn_to_hva_memslot(slot
, gfn
);
1263 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1266 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1269 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1272 return gfn_to_hva_many(slot
, gfn
, NULL
);
1274 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1276 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1278 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1280 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1282 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1284 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1286 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1289 * If writable is set to false, the hva returned by this function is only
1290 * allowed to be read.
1292 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1293 gfn_t gfn
, bool *writable
)
1295 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1297 if (!kvm_is_error_hva(hva
) && writable
)
1298 *writable
= !memslot_is_readonly(slot
);
1303 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1305 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1307 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1310 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1312 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1314 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1317 static int get_user_page_nowait(unsigned long start
, int write
,
1320 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1323 flags
|= FOLL_WRITE
;
1325 return get_user_pages(start
, 1, flags
, page
, NULL
);
1328 static inline int check_user_page_hwpoison(unsigned long addr
)
1330 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1332 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1333 return rc
== -EHWPOISON
;
1337 * The atomic path to get the writable pfn which will be stored in @pfn,
1338 * true indicates success, otherwise false is returned.
1340 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1341 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1343 struct page
*page
[1];
1346 if (!(async
|| atomic
))
1350 * Fast pin a writable pfn only if it is a write fault request
1351 * or the caller allows to map a writable pfn for a read fault
1354 if (!(write_fault
|| writable
))
1357 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1359 *pfn
= page_to_pfn(page
[0]);
1370 * The slow path to get the pfn of the specified host virtual address,
1371 * 1 indicates success, -errno is returned if error is detected.
1373 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1374 bool *writable
, kvm_pfn_t
*pfn
)
1376 struct page
*page
[1];
1382 *writable
= write_fault
;
1385 down_read(¤t
->mm
->mmap_sem
);
1386 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1387 up_read(¤t
->mm
->mmap_sem
);
1389 unsigned int flags
= FOLL_HWPOISON
;
1392 flags
|= FOLL_WRITE
;
1394 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1399 /* map read fault as writable if possible */
1400 if (unlikely(!write_fault
) && writable
) {
1401 struct page
*wpage
[1];
1403 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1412 *pfn
= page_to_pfn(page
[0]);
1416 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1418 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1421 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1427 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1428 unsigned long addr
, bool *async
,
1429 bool write_fault
, kvm_pfn_t
*p_pfn
)
1434 r
= follow_pfn(vma
, addr
, &pfn
);
1437 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1438 * not call the fault handler, so do it here.
1440 bool unlocked
= false;
1441 r
= fixup_user_fault(current
, current
->mm
, addr
,
1442 (write_fault
? FAULT_FLAG_WRITE
: 0),
1449 r
= follow_pfn(vma
, addr
, &pfn
);
1457 * Get a reference here because callers of *hva_to_pfn* and
1458 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1459 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1460 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1461 * simply do nothing for reserved pfns.
1463 * Whoever called remap_pfn_range is also going to call e.g.
1464 * unmap_mapping_range before the underlying pages are freed,
1465 * causing a call to our MMU notifier.
1474 * Pin guest page in memory and return its pfn.
1475 * @addr: host virtual address which maps memory to the guest
1476 * @atomic: whether this function can sleep
1477 * @async: whether this function need to wait IO complete if the
1478 * host page is not in the memory
1479 * @write_fault: whether we should get a writable host page
1480 * @writable: whether it allows to map a writable host page for !@write_fault
1482 * The function will map a writable host page for these two cases:
1483 * 1): @write_fault = true
1484 * 2): @write_fault = false && @writable, @writable will tell the caller
1485 * whether the mapping is writable.
1487 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1488 bool write_fault
, bool *writable
)
1490 struct vm_area_struct
*vma
;
1494 /* we can do it either atomically or asynchronously, not both */
1495 BUG_ON(atomic
&& async
);
1497 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1501 return KVM_PFN_ERR_FAULT
;
1503 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1507 down_read(¤t
->mm
->mmap_sem
);
1508 if (npages
== -EHWPOISON
||
1509 (!async
&& check_user_page_hwpoison(addr
))) {
1510 pfn
= KVM_PFN_ERR_HWPOISON
;
1515 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1518 pfn
= KVM_PFN_ERR_FAULT
;
1519 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1520 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, &pfn
);
1524 pfn
= KVM_PFN_ERR_FAULT
;
1526 if (async
&& vma_is_valid(vma
, write_fault
))
1528 pfn
= KVM_PFN_ERR_FAULT
;
1531 up_read(¤t
->mm
->mmap_sem
);
1535 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1536 bool atomic
, bool *async
, bool write_fault
,
1539 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1541 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1544 return KVM_PFN_ERR_RO_FAULT
;
1547 if (kvm_is_error_hva(addr
)) {
1550 return KVM_PFN_NOSLOT
;
1553 /* Do not map writable pfn in the readonly memslot. */
1554 if (writable
&& memslot_is_readonly(slot
)) {
1559 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1562 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1564 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1567 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1568 write_fault
, writable
);
1570 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1572 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1574 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1576 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1578 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1580 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1582 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1584 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1586 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1588 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1590 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1592 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1594 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1596 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1598 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1600 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1602 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1604 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1606 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1608 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1609 struct page
**pages
, int nr_pages
)
1614 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1615 if (kvm_is_error_hva(addr
))
1618 if (entry
< nr_pages
)
1621 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1623 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1625 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1627 if (is_error_noslot_pfn(pfn
))
1628 return KVM_ERR_PTR_BAD_PAGE
;
1630 if (kvm_is_reserved_pfn(pfn
)) {
1632 return KVM_ERR_PTR_BAD_PAGE
;
1635 return pfn_to_page(pfn
);
1638 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1642 pfn
= gfn_to_pfn(kvm
, gfn
);
1644 return kvm_pfn_to_page(pfn
);
1646 EXPORT_SYMBOL_GPL(gfn_to_page
);
1648 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1652 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1654 return kvm_pfn_to_page(pfn
);
1656 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1658 void kvm_release_page_clean(struct page
*page
)
1660 WARN_ON(is_error_page(page
));
1662 kvm_release_pfn_clean(page_to_pfn(page
));
1664 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1666 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1668 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1669 put_page(pfn_to_page(pfn
));
1671 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1673 void kvm_release_page_dirty(struct page
*page
)
1675 WARN_ON(is_error_page(page
));
1677 kvm_release_pfn_dirty(page_to_pfn(page
));
1679 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1681 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1683 kvm_set_pfn_dirty(pfn
);
1684 kvm_release_pfn_clean(pfn
);
1687 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1689 if (!kvm_is_reserved_pfn(pfn
)) {
1690 struct page
*page
= pfn_to_page(pfn
);
1692 if (!PageReserved(page
))
1696 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1698 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1700 if (!kvm_is_reserved_pfn(pfn
))
1701 mark_page_accessed(pfn_to_page(pfn
));
1703 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1705 void kvm_get_pfn(kvm_pfn_t pfn
)
1707 if (!kvm_is_reserved_pfn(pfn
))
1708 get_page(pfn_to_page(pfn
));
1710 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1712 static int next_segment(unsigned long len
, int offset
)
1714 if (len
> PAGE_SIZE
- offset
)
1715 return PAGE_SIZE
- offset
;
1720 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1721 void *data
, int offset
, int len
)
1726 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1727 if (kvm_is_error_hva(addr
))
1729 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1735 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1738 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1740 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1742 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1744 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1745 int offset
, int len
)
1747 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1749 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1753 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1755 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1757 int offset
= offset_in_page(gpa
);
1760 while ((seg
= next_segment(len
, offset
)) != 0) {
1761 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1771 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1773 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1775 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1777 int offset
= offset_in_page(gpa
);
1780 while ((seg
= next_segment(len
, offset
)) != 0) {
1781 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1791 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1793 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1794 void *data
, int offset
, unsigned long len
)
1799 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1800 if (kvm_is_error_hva(addr
))
1802 pagefault_disable();
1803 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1810 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1813 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1814 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1815 int offset
= offset_in_page(gpa
);
1817 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1819 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1821 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1822 void *data
, unsigned long len
)
1824 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1825 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1826 int offset
= offset_in_page(gpa
);
1828 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1830 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1832 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1833 const void *data
, int offset
, int len
)
1838 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1839 if (kvm_is_error_hva(addr
))
1841 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1844 mark_page_dirty_in_slot(memslot
, gfn
);
1848 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1849 const void *data
, int offset
, int len
)
1851 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1853 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1855 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1857 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1858 const void *data
, int offset
, int len
)
1860 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1862 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1864 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1866 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1869 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1871 int offset
= offset_in_page(gpa
);
1874 while ((seg
= next_segment(len
, offset
)) != 0) {
1875 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1885 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1887 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1890 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1892 int offset
= offset_in_page(gpa
);
1895 while ((seg
= next_segment(len
, offset
)) != 0) {
1896 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1906 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1908 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1909 struct gfn_to_hva_cache
*ghc
,
1910 gpa_t gpa
, unsigned long len
)
1912 int offset
= offset_in_page(gpa
);
1913 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1914 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1915 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1916 gfn_t nr_pages_avail
;
1919 ghc
->generation
= slots
->generation
;
1921 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1922 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1923 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1927 * If the requested region crosses two memslots, we still
1928 * verify that the entire region is valid here.
1930 while (start_gfn
<= end_gfn
) {
1931 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1932 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1934 if (kvm_is_error_hva(ghc
->hva
))
1936 start_gfn
+= nr_pages_avail
;
1938 /* Use the slow path for cross page reads and writes. */
1939 ghc
->memslot
= NULL
;
1944 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1945 gpa_t gpa
, unsigned long len
)
1947 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1948 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1950 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1952 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1953 void *data
, int offset
, unsigned long len
)
1955 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1957 gpa_t gpa
= ghc
->gpa
+ offset
;
1959 BUG_ON(len
+ offset
> ghc
->len
);
1961 if (slots
->generation
!= ghc
->generation
)
1962 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1964 if (unlikely(!ghc
->memslot
))
1965 return kvm_write_guest(kvm
, gpa
, data
, len
);
1967 if (kvm_is_error_hva(ghc
->hva
))
1970 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
1973 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
1977 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
1979 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1980 void *data
, unsigned long len
)
1982 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
1984 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1986 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1987 void *data
, unsigned long len
)
1989 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1992 BUG_ON(len
> ghc
->len
);
1994 if (slots
->generation
!= ghc
->generation
)
1995 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1997 if (unlikely(!ghc
->memslot
))
1998 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2000 if (kvm_is_error_hva(ghc
->hva
))
2003 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2009 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2011 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2013 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2015 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2017 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2019 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2021 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2023 int offset
= offset_in_page(gpa
);
2026 while ((seg
= next_segment(len
, offset
)) != 0) {
2027 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2036 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2038 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2041 if (memslot
&& memslot
->dirty_bitmap
) {
2042 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2044 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2048 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2050 struct kvm_memory_slot
*memslot
;
2052 memslot
= gfn_to_memslot(kvm
, gfn
);
2053 mark_page_dirty_in_slot(memslot
, gfn
);
2055 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2057 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2059 struct kvm_memory_slot
*memslot
;
2061 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2062 mark_page_dirty_in_slot(memslot
, gfn
);
2064 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2066 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2068 unsigned int old
, val
, grow
;
2070 old
= val
= vcpu
->halt_poll_ns
;
2071 grow
= READ_ONCE(halt_poll_ns_grow
);
2073 if (val
== 0 && grow
)
2078 if (val
> halt_poll_ns
)
2081 vcpu
->halt_poll_ns
= val
;
2082 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2085 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2087 unsigned int old
, val
, shrink
;
2089 old
= val
= vcpu
->halt_poll_ns
;
2090 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2096 vcpu
->halt_poll_ns
= val
;
2097 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2100 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2102 if (kvm_arch_vcpu_runnable(vcpu
)) {
2103 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2106 if (kvm_cpu_has_pending_timer(vcpu
))
2108 if (signal_pending(current
))
2115 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2117 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2120 DECLARE_SWAITQUEUE(wait
);
2121 bool waited
= false;
2124 start
= cur
= ktime_get();
2125 if (vcpu
->halt_poll_ns
) {
2126 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2128 ++vcpu
->stat
.halt_attempted_poll
;
2131 * This sets KVM_REQ_UNHALT if an interrupt
2134 if (kvm_vcpu_check_block(vcpu
) < 0) {
2135 ++vcpu
->stat
.halt_successful_poll
;
2136 if (!vcpu_valid_wakeup(vcpu
))
2137 ++vcpu
->stat
.halt_poll_invalid
;
2141 } while (single_task_running() && ktime_before(cur
, stop
));
2144 kvm_arch_vcpu_blocking(vcpu
);
2147 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2149 if (kvm_vcpu_check_block(vcpu
) < 0)
2156 finish_swait(&vcpu
->wq
, &wait
);
2159 kvm_arch_vcpu_unblocking(vcpu
);
2161 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2163 if (!vcpu_valid_wakeup(vcpu
))
2164 shrink_halt_poll_ns(vcpu
);
2165 else if (halt_poll_ns
) {
2166 if (block_ns
<= vcpu
->halt_poll_ns
)
2168 /* we had a long block, shrink polling */
2169 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2170 shrink_halt_poll_ns(vcpu
);
2171 /* we had a short halt and our poll time is too small */
2172 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2173 block_ns
< halt_poll_ns
)
2174 grow_halt_poll_ns(vcpu
);
2176 vcpu
->halt_poll_ns
= 0;
2178 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2179 kvm_arch_vcpu_block_finish(vcpu
);
2181 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2183 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2185 struct swait_queue_head
*wqp
;
2187 wqp
= kvm_arch_vcpu_wq(vcpu
);
2188 if (swait_active(wqp
)) {
2190 ++vcpu
->stat
.halt_wakeup
;
2196 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2200 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2202 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2205 int cpu
= vcpu
->cpu
;
2207 if (kvm_vcpu_wake_up(vcpu
))
2211 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2212 if (kvm_arch_vcpu_should_kick(vcpu
))
2213 smp_send_reschedule(cpu
);
2216 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2217 #endif /* !CONFIG_S390 */
2219 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2222 struct task_struct
*task
= NULL
;
2226 pid
= rcu_dereference(target
->pid
);
2228 task
= get_pid_task(pid
, PIDTYPE_PID
);
2232 ret
= yield_to(task
, 1);
2233 put_task_struct(task
);
2237 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2240 * Helper that checks whether a VCPU is eligible for directed yield.
2241 * Most eligible candidate to yield is decided by following heuristics:
2243 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2244 * (preempted lock holder), indicated by @in_spin_loop.
2245 * Set at the beiginning and cleared at the end of interception/PLE handler.
2247 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2248 * chance last time (mostly it has become eligible now since we have probably
2249 * yielded to lockholder in last iteration. This is done by toggling
2250 * @dy_eligible each time a VCPU checked for eligibility.)
2252 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2253 * to preempted lock-holder could result in wrong VCPU selection and CPU
2254 * burning. Giving priority for a potential lock-holder increases lock
2257 * Since algorithm is based on heuristics, accessing another VCPU data without
2258 * locking does not harm. It may result in trying to yield to same VCPU, fail
2259 * and continue with next VCPU and so on.
2261 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2263 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2266 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2267 vcpu
->spin_loop
.dy_eligible
;
2269 if (vcpu
->spin_loop
.in_spin_loop
)
2270 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2278 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2280 struct kvm
*kvm
= me
->kvm
;
2281 struct kvm_vcpu
*vcpu
;
2282 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2288 kvm_vcpu_set_in_spin_loop(me
, true);
2290 * We boost the priority of a VCPU that is runnable but not
2291 * currently running, because it got preempted by something
2292 * else and called schedule in __vcpu_run. Hopefully that
2293 * VCPU is holding the lock that we need and will release it.
2294 * We approximate round-robin by starting at the last boosted VCPU.
2296 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2297 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2298 if (!pass
&& i
<= last_boosted_vcpu
) {
2299 i
= last_boosted_vcpu
;
2301 } else if (pass
&& i
> last_boosted_vcpu
)
2303 if (!ACCESS_ONCE(vcpu
->preempted
))
2307 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2309 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2312 yielded
= kvm_vcpu_yield_to(vcpu
);
2314 kvm
->last_boosted_vcpu
= i
;
2316 } else if (yielded
< 0) {
2323 kvm_vcpu_set_in_spin_loop(me
, false);
2325 /* Ensure vcpu is not eligible during next spinloop */
2326 kvm_vcpu_set_dy_eligible(me
, false);
2328 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2330 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2332 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2335 if (vmf
->pgoff
== 0)
2336 page
= virt_to_page(vcpu
->run
);
2338 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2339 page
= virt_to_page(vcpu
->arch
.pio_data
);
2341 #ifdef CONFIG_KVM_MMIO
2342 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2343 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2346 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2352 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2353 .fault
= kvm_vcpu_fault
,
2356 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2358 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2362 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2364 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2366 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2367 kvm_put_kvm(vcpu
->kvm
);
2371 static struct file_operations kvm_vcpu_fops
= {
2372 .release
= kvm_vcpu_release
,
2373 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2374 #ifdef CONFIG_KVM_COMPAT
2375 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2377 .mmap
= kvm_vcpu_mmap
,
2378 .llseek
= noop_llseek
,
2382 * Allocates an inode for the vcpu.
2384 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2386 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2389 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2391 char dir_name
[ITOA_MAX_LEN
* 2];
2394 if (!kvm_arch_has_vcpu_debugfs())
2397 if (!debugfs_initialized())
2400 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2401 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2402 vcpu
->kvm
->debugfs_dentry
);
2403 if (!vcpu
->debugfs_dentry
)
2406 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2408 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2416 * Creates some virtual cpus. Good luck creating more than one.
2418 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2421 struct kvm_vcpu
*vcpu
;
2423 if (id
>= KVM_MAX_VCPU_ID
)
2426 mutex_lock(&kvm
->lock
);
2427 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2428 mutex_unlock(&kvm
->lock
);
2432 kvm
->created_vcpus
++;
2433 mutex_unlock(&kvm
->lock
);
2435 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2438 goto vcpu_decrement
;
2441 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2443 r
= kvm_arch_vcpu_setup(vcpu
);
2447 r
= kvm_create_vcpu_debugfs(vcpu
);
2451 mutex_lock(&kvm
->lock
);
2452 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2454 goto unlock_vcpu_destroy
;
2457 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2459 /* Now it's all set up, let userspace reach it */
2461 r
= create_vcpu_fd(vcpu
);
2464 goto unlock_vcpu_destroy
;
2467 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2470 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2471 * before kvm->online_vcpu's incremented value.
2474 atomic_inc(&kvm
->online_vcpus
);
2476 mutex_unlock(&kvm
->lock
);
2477 kvm_arch_vcpu_postcreate(vcpu
);
2480 unlock_vcpu_destroy
:
2481 mutex_unlock(&kvm
->lock
);
2482 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2484 kvm_arch_vcpu_destroy(vcpu
);
2486 mutex_lock(&kvm
->lock
);
2487 kvm
->created_vcpus
--;
2488 mutex_unlock(&kvm
->lock
);
2492 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2495 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2496 vcpu
->sigset_active
= 1;
2497 vcpu
->sigset
= *sigset
;
2499 vcpu
->sigset_active
= 0;
2503 static long kvm_vcpu_ioctl(struct file
*filp
,
2504 unsigned int ioctl
, unsigned long arg
)
2506 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2507 void __user
*argp
= (void __user
*)arg
;
2509 struct kvm_fpu
*fpu
= NULL
;
2510 struct kvm_sregs
*kvm_sregs
= NULL
;
2512 if (vcpu
->kvm
->mm
!= current
->mm
)
2515 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2518 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2520 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2521 * so vcpu_load() would break it.
2523 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2524 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2528 r
= vcpu_load(vcpu
);
2537 oldpid
= rcu_access_pointer(vcpu
->pid
);
2538 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2539 /* The thread running this VCPU changed. */
2540 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2542 rcu_assign_pointer(vcpu
->pid
, newpid
);
2547 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2548 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2551 case KVM_GET_REGS
: {
2552 struct kvm_regs
*kvm_regs
;
2555 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2558 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2562 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2569 case KVM_SET_REGS
: {
2570 struct kvm_regs
*kvm_regs
;
2573 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2574 if (IS_ERR(kvm_regs
)) {
2575 r
= PTR_ERR(kvm_regs
);
2578 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2582 case KVM_GET_SREGS
: {
2583 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2587 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2591 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2596 case KVM_SET_SREGS
: {
2597 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2598 if (IS_ERR(kvm_sregs
)) {
2599 r
= PTR_ERR(kvm_sregs
);
2603 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2606 case KVM_GET_MP_STATE
: {
2607 struct kvm_mp_state mp_state
;
2609 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2613 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2618 case KVM_SET_MP_STATE
: {
2619 struct kvm_mp_state mp_state
;
2622 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2624 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2627 case KVM_TRANSLATE
: {
2628 struct kvm_translation tr
;
2631 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2633 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2637 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2642 case KVM_SET_GUEST_DEBUG
: {
2643 struct kvm_guest_debug dbg
;
2646 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2648 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2651 case KVM_SET_SIGNAL_MASK
: {
2652 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2653 struct kvm_signal_mask kvm_sigmask
;
2654 sigset_t sigset
, *p
;
2659 if (copy_from_user(&kvm_sigmask
, argp
,
2660 sizeof(kvm_sigmask
)))
2663 if (kvm_sigmask
.len
!= sizeof(sigset
))
2666 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2671 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2675 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2679 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2683 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2689 fpu
= memdup_user(argp
, sizeof(*fpu
));
2695 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2699 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2708 #ifdef CONFIG_KVM_COMPAT
2709 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2710 unsigned int ioctl
, unsigned long arg
)
2712 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2713 void __user
*argp
= compat_ptr(arg
);
2716 if (vcpu
->kvm
->mm
!= current
->mm
)
2720 case KVM_SET_SIGNAL_MASK
: {
2721 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2722 struct kvm_signal_mask kvm_sigmask
;
2723 compat_sigset_t csigset
;
2728 if (copy_from_user(&kvm_sigmask
, argp
,
2729 sizeof(kvm_sigmask
)))
2732 if (kvm_sigmask
.len
!= sizeof(csigset
))
2735 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2738 sigset_from_compat(&sigset
, &csigset
);
2739 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2741 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2745 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2753 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2754 int (*accessor
)(struct kvm_device
*dev
,
2755 struct kvm_device_attr
*attr
),
2758 struct kvm_device_attr attr
;
2763 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2766 return accessor(dev
, &attr
);
2769 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2772 struct kvm_device
*dev
= filp
->private_data
;
2775 case KVM_SET_DEVICE_ATTR
:
2776 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2777 case KVM_GET_DEVICE_ATTR
:
2778 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2779 case KVM_HAS_DEVICE_ATTR
:
2780 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2782 if (dev
->ops
->ioctl
)
2783 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2789 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2791 struct kvm_device
*dev
= filp
->private_data
;
2792 struct kvm
*kvm
= dev
->kvm
;
2798 static const struct file_operations kvm_device_fops
= {
2799 .unlocked_ioctl
= kvm_device_ioctl
,
2800 #ifdef CONFIG_KVM_COMPAT
2801 .compat_ioctl
= kvm_device_ioctl
,
2803 .release
= kvm_device_release
,
2806 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2808 if (filp
->f_op
!= &kvm_device_fops
)
2811 return filp
->private_data
;
2814 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2815 #ifdef CONFIG_KVM_MPIC
2816 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2817 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2821 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2823 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2826 if (kvm_device_ops_table
[type
] != NULL
)
2829 kvm_device_ops_table
[type
] = ops
;
2833 void kvm_unregister_device_ops(u32 type
)
2835 if (kvm_device_ops_table
[type
] != NULL
)
2836 kvm_device_ops_table
[type
] = NULL
;
2839 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2840 struct kvm_create_device
*cd
)
2842 struct kvm_device_ops
*ops
= NULL
;
2843 struct kvm_device
*dev
;
2844 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2847 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2850 ops
= kvm_device_ops_table
[cd
->type
];
2857 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2864 mutex_lock(&kvm
->lock
);
2865 ret
= ops
->create(dev
, cd
->type
);
2867 mutex_unlock(&kvm
->lock
);
2871 list_add(&dev
->vm_node
, &kvm
->devices
);
2872 mutex_unlock(&kvm
->lock
);
2877 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2879 mutex_lock(&kvm
->lock
);
2880 list_del(&dev
->vm_node
);
2881 mutex_unlock(&kvm
->lock
);
2891 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2894 case KVM_CAP_USER_MEMORY
:
2895 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2896 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2897 case KVM_CAP_INTERNAL_ERROR_DATA
:
2898 #ifdef CONFIG_HAVE_KVM_MSI
2899 case KVM_CAP_SIGNAL_MSI
:
2901 #ifdef CONFIG_HAVE_KVM_IRQFD
2903 case KVM_CAP_IRQFD_RESAMPLE
:
2905 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2906 case KVM_CAP_CHECK_EXTENSION_VM
:
2908 #ifdef CONFIG_KVM_MMIO
2909 case KVM_CAP_COALESCED_MMIO
:
2910 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2912 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2913 case KVM_CAP_IRQ_ROUTING
:
2914 return KVM_MAX_IRQ_ROUTES
;
2916 #if KVM_ADDRESS_SPACE_NUM > 1
2917 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2918 return KVM_ADDRESS_SPACE_NUM
;
2920 case KVM_CAP_MAX_VCPU_ID
:
2921 return KVM_MAX_VCPU_ID
;
2925 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2928 static long kvm_vm_ioctl(struct file
*filp
,
2929 unsigned int ioctl
, unsigned long arg
)
2931 struct kvm
*kvm
= filp
->private_data
;
2932 void __user
*argp
= (void __user
*)arg
;
2935 if (kvm
->mm
!= current
->mm
)
2938 case KVM_CREATE_VCPU
:
2939 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2941 case KVM_SET_USER_MEMORY_REGION
: {
2942 struct kvm_userspace_memory_region kvm_userspace_mem
;
2945 if (copy_from_user(&kvm_userspace_mem
, argp
,
2946 sizeof(kvm_userspace_mem
)))
2949 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2952 case KVM_GET_DIRTY_LOG
: {
2953 struct kvm_dirty_log log
;
2956 if (copy_from_user(&log
, argp
, sizeof(log
)))
2958 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2961 #ifdef CONFIG_KVM_MMIO
2962 case KVM_REGISTER_COALESCED_MMIO
: {
2963 struct kvm_coalesced_mmio_zone zone
;
2966 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2968 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2971 case KVM_UNREGISTER_COALESCED_MMIO
: {
2972 struct kvm_coalesced_mmio_zone zone
;
2975 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2977 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2982 struct kvm_irqfd data
;
2985 if (copy_from_user(&data
, argp
, sizeof(data
)))
2987 r
= kvm_irqfd(kvm
, &data
);
2990 case KVM_IOEVENTFD
: {
2991 struct kvm_ioeventfd data
;
2994 if (copy_from_user(&data
, argp
, sizeof(data
)))
2996 r
= kvm_ioeventfd(kvm
, &data
);
2999 #ifdef CONFIG_HAVE_KVM_MSI
3000 case KVM_SIGNAL_MSI
: {
3004 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3006 r
= kvm_send_userspace_msi(kvm
, &msi
);
3010 #ifdef __KVM_HAVE_IRQ_LINE
3011 case KVM_IRQ_LINE_STATUS
:
3012 case KVM_IRQ_LINE
: {
3013 struct kvm_irq_level irq_event
;
3016 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3019 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3020 ioctl
== KVM_IRQ_LINE_STATUS
);
3025 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3026 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3034 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3035 case KVM_SET_GSI_ROUTING
: {
3036 struct kvm_irq_routing routing
;
3037 struct kvm_irq_routing __user
*urouting
;
3038 struct kvm_irq_routing_entry
*entries
= NULL
;
3041 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3044 if (!kvm_arch_can_set_irq_routing(kvm
))
3046 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3052 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3057 if (copy_from_user(entries
, urouting
->entries
,
3058 routing
.nr
* sizeof(*entries
)))
3059 goto out_free_irq_routing
;
3061 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3063 out_free_irq_routing
:
3067 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3068 case KVM_CREATE_DEVICE
: {
3069 struct kvm_create_device cd
;
3072 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3075 r
= kvm_ioctl_create_device(kvm
, &cd
);
3080 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3086 case KVM_CHECK_EXTENSION
:
3087 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3090 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3096 #ifdef CONFIG_KVM_COMPAT
3097 struct compat_kvm_dirty_log
{
3101 compat_uptr_t dirty_bitmap
; /* one bit per page */
3106 static long kvm_vm_compat_ioctl(struct file
*filp
,
3107 unsigned int ioctl
, unsigned long arg
)
3109 struct kvm
*kvm
= filp
->private_data
;
3112 if (kvm
->mm
!= current
->mm
)
3115 case KVM_GET_DIRTY_LOG
: {
3116 struct compat_kvm_dirty_log compat_log
;
3117 struct kvm_dirty_log log
;
3119 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3120 sizeof(compat_log
)))
3122 log
.slot
= compat_log
.slot
;
3123 log
.padding1
= compat_log
.padding1
;
3124 log
.padding2
= compat_log
.padding2
;
3125 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3127 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3131 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3137 static struct file_operations kvm_vm_fops
= {
3138 .release
= kvm_vm_release
,
3139 .unlocked_ioctl
= kvm_vm_ioctl
,
3140 #ifdef CONFIG_KVM_COMPAT
3141 .compat_ioctl
= kvm_vm_compat_ioctl
,
3143 .llseek
= noop_llseek
,
3146 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3152 kvm
= kvm_create_vm(type
);
3154 return PTR_ERR(kvm
);
3155 #ifdef CONFIG_KVM_MMIO
3156 r
= kvm_coalesced_mmio_init(kvm
);
3162 r
= get_unused_fd_flags(O_CLOEXEC
);
3167 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3171 return PTR_ERR(file
);
3175 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3176 * already set, with ->release() being kvm_vm_release(). In error
3177 * cases it will be called by the final fput(file) and will take
3178 * care of doing kvm_put_kvm(kvm).
3180 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3185 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3187 fd_install(r
, file
);
3191 static long kvm_dev_ioctl(struct file
*filp
,
3192 unsigned int ioctl
, unsigned long arg
)
3197 case KVM_GET_API_VERSION
:
3200 r
= KVM_API_VERSION
;
3203 r
= kvm_dev_ioctl_create_vm(arg
);
3205 case KVM_CHECK_EXTENSION
:
3206 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3208 case KVM_GET_VCPU_MMAP_SIZE
:
3211 r
= PAGE_SIZE
; /* struct kvm_run */
3213 r
+= PAGE_SIZE
; /* pio data page */
3215 #ifdef CONFIG_KVM_MMIO
3216 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3219 case KVM_TRACE_ENABLE
:
3220 case KVM_TRACE_PAUSE
:
3221 case KVM_TRACE_DISABLE
:
3225 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3231 static struct file_operations kvm_chardev_ops
= {
3232 .unlocked_ioctl
= kvm_dev_ioctl
,
3233 .compat_ioctl
= kvm_dev_ioctl
,
3234 .llseek
= noop_llseek
,
3237 static struct miscdevice kvm_dev
= {
3243 static void hardware_enable_nolock(void *junk
)
3245 int cpu
= raw_smp_processor_id();
3248 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3251 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3253 r
= kvm_arch_hardware_enable();
3256 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3257 atomic_inc(&hardware_enable_failed
);
3258 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3262 static int kvm_starting_cpu(unsigned int cpu
)
3264 raw_spin_lock(&kvm_count_lock
);
3265 if (kvm_usage_count
)
3266 hardware_enable_nolock(NULL
);
3267 raw_spin_unlock(&kvm_count_lock
);
3271 static void hardware_disable_nolock(void *junk
)
3273 int cpu
= raw_smp_processor_id();
3275 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3277 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3278 kvm_arch_hardware_disable();
3281 static int kvm_dying_cpu(unsigned int cpu
)
3283 raw_spin_lock(&kvm_count_lock
);
3284 if (kvm_usage_count
)
3285 hardware_disable_nolock(NULL
);
3286 raw_spin_unlock(&kvm_count_lock
);
3290 static void hardware_disable_all_nolock(void)
3292 BUG_ON(!kvm_usage_count
);
3295 if (!kvm_usage_count
)
3296 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3299 static void hardware_disable_all(void)
3301 raw_spin_lock(&kvm_count_lock
);
3302 hardware_disable_all_nolock();
3303 raw_spin_unlock(&kvm_count_lock
);
3306 static int hardware_enable_all(void)
3310 raw_spin_lock(&kvm_count_lock
);
3313 if (kvm_usage_count
== 1) {
3314 atomic_set(&hardware_enable_failed
, 0);
3315 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3317 if (atomic_read(&hardware_enable_failed
)) {
3318 hardware_disable_all_nolock();
3323 raw_spin_unlock(&kvm_count_lock
);
3328 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3332 * Some (well, at least mine) BIOSes hang on reboot if
3335 * And Intel TXT required VMX off for all cpu when system shutdown.
3337 pr_info("kvm: exiting hardware virtualization\n");
3338 kvm_rebooting
= true;
3339 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3343 static struct notifier_block kvm_reboot_notifier
= {
3344 .notifier_call
= kvm_reboot
,
3348 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3352 for (i
= 0; i
< bus
->dev_count
; i
++) {
3353 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3355 kvm_iodevice_destructor(pos
);
3360 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3361 const struct kvm_io_range
*r2
)
3363 gpa_t addr1
= r1
->addr
;
3364 gpa_t addr2
= r2
->addr
;
3369 /* If r2->len == 0, match the exact address. If r2->len != 0,
3370 * accept any overlapping write. Any order is acceptable for
3371 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3372 * we process all of them.
3385 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3387 return kvm_io_bus_cmp(p1
, p2
);
3390 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3391 gpa_t addr
, int len
)
3393 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3399 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3400 kvm_io_bus_sort_cmp
, NULL
);
3405 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3406 gpa_t addr
, int len
)
3408 struct kvm_io_range
*range
, key
;
3411 key
= (struct kvm_io_range
) {
3416 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3417 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3421 off
= range
- bus
->range
;
3423 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3429 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3430 struct kvm_io_range
*range
, const void *val
)
3434 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3438 while (idx
< bus
->dev_count
&&
3439 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3440 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3449 /* kvm_io_bus_write - called under kvm->slots_lock */
3450 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3451 int len
, const void *val
)
3453 struct kvm_io_bus
*bus
;
3454 struct kvm_io_range range
;
3457 range
= (struct kvm_io_range
) {
3462 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3465 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3466 return r
< 0 ? r
: 0;
3469 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3470 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3471 gpa_t addr
, int len
, const void *val
, long cookie
)
3473 struct kvm_io_bus
*bus
;
3474 struct kvm_io_range range
;
3476 range
= (struct kvm_io_range
) {
3481 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3485 /* First try the device referenced by cookie. */
3486 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3487 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3488 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3493 * cookie contained garbage; fall back to search and return the
3494 * correct cookie value.
3496 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3499 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3500 struct kvm_io_range
*range
, void *val
)
3504 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3508 while (idx
< bus
->dev_count
&&
3509 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3510 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3518 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3520 /* kvm_io_bus_read - called under kvm->slots_lock */
3521 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3524 struct kvm_io_bus
*bus
;
3525 struct kvm_io_range range
;
3528 range
= (struct kvm_io_range
) {
3533 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3536 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3537 return r
< 0 ? r
: 0;
3541 /* Caller must hold slots_lock. */
3542 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3543 int len
, struct kvm_io_device
*dev
)
3545 struct kvm_io_bus
*new_bus
, *bus
;
3547 bus
= kvm_get_bus(kvm
, bus_idx
);
3551 /* exclude ioeventfd which is limited by maximum fd */
3552 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3555 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3556 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3559 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3560 sizeof(struct kvm_io_range
)));
3561 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3562 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3563 synchronize_srcu_expedited(&kvm
->srcu
);
3569 /* Caller must hold slots_lock. */
3570 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3571 struct kvm_io_device
*dev
)
3574 struct kvm_io_bus
*new_bus
, *bus
;
3576 bus
= kvm_get_bus(kvm
, bus_idx
);
3580 for (i
= 0; i
< bus
->dev_count
; i
++)
3581 if (bus
->range
[i
].dev
== dev
) {
3585 if (i
== bus
->dev_count
)
3588 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3589 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3591 pr_err("kvm: failed to shrink bus, removing it completely\n");
3595 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3596 new_bus
->dev_count
--;
3597 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3598 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3601 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3602 synchronize_srcu_expedited(&kvm
->srcu
);
3607 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3610 struct kvm_io_bus
*bus
;
3611 int dev_idx
, srcu_idx
;
3612 struct kvm_io_device
*iodev
= NULL
;
3614 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3616 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3620 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3624 iodev
= bus
->range
[dev_idx
].dev
;
3627 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3631 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3633 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3634 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3637 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3640 /* The debugfs files are a reference to the kvm struct which
3641 * is still valid when kvm_destroy_vm is called.
3642 * To avoid the race between open and the removal of the debugfs
3643 * directory we test against the users count.
3645 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3648 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3649 kvm_put_kvm(stat_data
->kvm
);
3656 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3658 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3661 simple_attr_release(inode
, file
);
3662 kvm_put_kvm(stat_data
->kvm
);
3667 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3669 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3671 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3676 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3678 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3683 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3688 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3690 __simple_attr_check_format("%llu\n", 0ull);
3691 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3692 vm_stat_clear_per_vm
, "%llu\n");
3695 static const struct file_operations vm_stat_get_per_vm_fops
= {
3696 .owner
= THIS_MODULE
,
3697 .open
= vm_stat_get_per_vm_open
,
3698 .release
= kvm_debugfs_release
,
3699 .read
= simple_attr_read
,
3700 .write
= simple_attr_write
,
3701 .llseek
= no_llseek
,
3704 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3707 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3708 struct kvm_vcpu
*vcpu
;
3712 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3713 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3718 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3721 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3722 struct kvm_vcpu
*vcpu
;
3727 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3728 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3733 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3735 __simple_attr_check_format("%llu\n", 0ull);
3736 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3737 vcpu_stat_clear_per_vm
, "%llu\n");
3740 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3741 .owner
= THIS_MODULE
,
3742 .open
= vcpu_stat_get_per_vm_open
,
3743 .release
= kvm_debugfs_release
,
3744 .read
= simple_attr_read
,
3745 .write
= simple_attr_write
,
3746 .llseek
= no_llseek
,
3749 static const struct file_operations
*stat_fops_per_vm
[] = {
3750 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3751 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3754 static int vm_stat_get(void *_offset
, u64
*val
)
3756 unsigned offset
= (long)_offset
;
3758 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3762 spin_lock(&kvm_lock
);
3763 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3765 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3768 spin_unlock(&kvm_lock
);
3772 static int vm_stat_clear(void *_offset
, u64 val
)
3774 unsigned offset
= (long)_offset
;
3776 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3781 spin_lock(&kvm_lock
);
3782 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3784 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3786 spin_unlock(&kvm_lock
);
3791 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3793 static int vcpu_stat_get(void *_offset
, u64
*val
)
3795 unsigned offset
= (long)_offset
;
3797 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3801 spin_lock(&kvm_lock
);
3802 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3804 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3807 spin_unlock(&kvm_lock
);
3811 static int vcpu_stat_clear(void *_offset
, u64 val
)
3813 unsigned offset
= (long)_offset
;
3815 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3820 spin_lock(&kvm_lock
);
3821 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3823 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3825 spin_unlock(&kvm_lock
);
3830 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3833 static const struct file_operations
*stat_fops
[] = {
3834 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3835 [KVM_STAT_VM
] = &vm_stat_fops
,
3838 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3840 struct kobj_uevent_env
*env
;
3841 unsigned long long created
, active
;
3843 if (!kvm_dev
.this_device
|| !kvm
)
3846 spin_lock(&kvm_lock
);
3847 if (type
== KVM_EVENT_CREATE_VM
) {
3848 kvm_createvm_count
++;
3850 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3853 created
= kvm_createvm_count
;
3854 active
= kvm_active_vms
;
3855 spin_unlock(&kvm_lock
);
3857 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3861 add_uevent_var(env
, "CREATED=%llu", created
);
3862 add_uevent_var(env
, "COUNT=%llu", active
);
3864 if (type
== KVM_EVENT_CREATE_VM
) {
3865 add_uevent_var(env
, "EVENT=create");
3866 kvm
->userspace_pid
= task_pid_nr(current
);
3867 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3868 add_uevent_var(env
, "EVENT=destroy");
3870 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3872 if (kvm
->debugfs_dentry
) {
3873 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3876 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3878 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3882 /* no need for checks, since we are adding at most only 5 keys */
3883 env
->envp
[env
->envp_idx
++] = NULL
;
3884 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3888 static int kvm_init_debug(void)
3891 struct kvm_stats_debugfs_item
*p
;
3893 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3894 if (kvm_debugfs_dir
== NULL
)
3897 kvm_debugfs_num_entries
= 0;
3898 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3899 if (!debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3900 (void *)(long)p
->offset
,
3901 stat_fops
[p
->kind
]))
3908 debugfs_remove_recursive(kvm_debugfs_dir
);
3913 static int kvm_suspend(void)
3915 if (kvm_usage_count
)
3916 hardware_disable_nolock(NULL
);
3920 static void kvm_resume(void)
3922 if (kvm_usage_count
) {
3923 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3924 hardware_enable_nolock(NULL
);
3928 static struct syscore_ops kvm_syscore_ops
= {
3929 .suspend
= kvm_suspend
,
3930 .resume
= kvm_resume
,
3934 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3936 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3939 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3941 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3943 if (vcpu
->preempted
)
3944 vcpu
->preempted
= false;
3946 kvm_arch_sched_in(vcpu
, cpu
);
3948 kvm_arch_vcpu_load(vcpu
, cpu
);
3951 static void kvm_sched_out(struct preempt_notifier
*pn
,
3952 struct task_struct
*next
)
3954 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3956 if (current
->state
== TASK_RUNNING
)
3957 vcpu
->preempted
= true;
3958 kvm_arch_vcpu_put(vcpu
);
3961 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3962 struct module
*module
)
3967 r
= kvm_arch_init(opaque
);
3972 * kvm_arch_init makes sure there's at most one caller
3973 * for architectures that support multiple implementations,
3974 * like intel and amd on x86.
3975 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3976 * conflicts in case kvm is already setup for another implementation.
3978 r
= kvm_irqfd_init();
3982 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3987 r
= kvm_arch_hardware_setup();
3991 for_each_online_cpu(cpu
) {
3992 smp_call_function_single(cpu
,
3993 kvm_arch_check_processor_compat
,
3999 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4000 kvm_starting_cpu
, kvm_dying_cpu
);
4003 register_reboot_notifier(&kvm_reboot_notifier
);
4005 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4007 vcpu_align
= __alignof__(struct kvm_vcpu
);
4008 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4010 if (!kvm_vcpu_cache
) {
4015 r
= kvm_async_pf_init();
4019 kvm_chardev_ops
.owner
= module
;
4020 kvm_vm_fops
.owner
= module
;
4021 kvm_vcpu_fops
.owner
= module
;
4023 r
= misc_register(&kvm_dev
);
4025 pr_err("kvm: misc device register failed\n");
4029 register_syscore_ops(&kvm_syscore_ops
);
4031 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4032 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4034 r
= kvm_init_debug();
4036 pr_err("kvm: create debugfs files failed\n");
4040 r
= kvm_vfio_ops_init();
4046 unregister_syscore_ops(&kvm_syscore_ops
);
4047 misc_deregister(&kvm_dev
);
4049 kvm_async_pf_deinit();
4051 kmem_cache_destroy(kvm_vcpu_cache
);
4053 unregister_reboot_notifier(&kvm_reboot_notifier
);
4054 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4057 kvm_arch_hardware_unsetup();
4059 free_cpumask_var(cpus_hardware_enabled
);
4067 EXPORT_SYMBOL_GPL(kvm_init
);
4071 debugfs_remove_recursive(kvm_debugfs_dir
);
4072 misc_deregister(&kvm_dev
);
4073 kmem_cache_destroy(kvm_vcpu_cache
);
4074 kvm_async_pf_deinit();
4075 unregister_syscore_ops(&kvm_syscore_ops
);
4076 unregister_reboot_notifier(&kvm_reboot_notifier
);
4077 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4078 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4079 kvm_arch_hardware_unsetup();
4082 free_cpumask_var(cpus_hardware_enabled
);
4083 kvm_vfio_ops_exit();
4085 EXPORT_SYMBOL_GPL(kvm_exit
);