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>
54 #include <linux/kthread.h>
56 #include <asm/processor.h>
58 #include <asm/ioctl.h>
59 #include <linux/uaccess.h>
60 #include <asm/pgtable.h>
62 #include "coalesced_mmio.h"
66 #define CREATE_TRACE_POINTS
67 #include <trace/events/kvm.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
77 module_param(halt_poll_ns
, uint
, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns
);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow
= 2;
82 module_param(halt_poll_ns_grow
, uint
, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
85 /* Default resets per-vcpu halt_poll_ns . */
86 unsigned int halt_poll_ns_shrink
;
87 module_param(halt_poll_ns_shrink
, uint
, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
93 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
96 DEFINE_MUTEX(kvm_lock
);
97 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
100 static cpumask_var_t cpus_hardware_enabled
;
101 static int kvm_usage_count
;
102 static atomic_t hardware_enable_failed
;
104 struct kmem_cache
*kvm_vcpu_cache
;
105 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
107 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
109 struct dentry
*kvm_debugfs_dir
;
110 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
112 static int kvm_debugfs_num_entries
;
113 static const struct file_operations
*stat_fops_per_vm
[];
115 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
117 #ifdef CONFIG_KVM_COMPAT
118 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
121 static int hardware_enable_all(void);
122 static void hardware_disable_all(void);
124 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
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 __weak
void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
140 unsigned long start
, unsigned long end
)
144 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn
)
147 * The metadata used by is_zone_device_page() to determine whether or
148 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
149 * the device has been pinned, e.g. by get_user_pages(). WARN if the
150 * page_count() is zero to help detect bad usage of this helper.
152 if (!pfn_valid(pfn
) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn
))))
155 return is_zone_device_page(pfn_to_page(pfn
));
158 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
161 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
162 * perspective they are "normal" pages, albeit with slightly different
166 return PageReserved(pfn_to_page(pfn
)) &&
167 !kvm_is_zone_device_pfn(pfn
);
173 * Switches to specified vcpu, until a matching vcpu_put()
175 int vcpu_load(struct kvm_vcpu
*vcpu
)
179 if (mutex_lock_killable(&vcpu
->mutex
))
182 preempt_notifier_register(&vcpu
->preempt_notifier
);
183 kvm_arch_vcpu_load(vcpu
, cpu
);
187 EXPORT_SYMBOL_GPL(vcpu_load
);
189 void vcpu_put(struct kvm_vcpu
*vcpu
)
192 kvm_arch_vcpu_put(vcpu
);
193 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
195 mutex_unlock(&vcpu
->mutex
);
197 EXPORT_SYMBOL_GPL(vcpu_put
);
199 /* TODO: merge with kvm_arch_vcpu_should_kick */
200 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
202 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
205 * We need to wait for the VCPU to reenable interrupts and get out of
206 * READING_SHADOW_PAGE_TABLES mode.
208 if (req
& KVM_REQUEST_WAIT
)
209 return mode
!= OUTSIDE_GUEST_MODE
;
212 * Need to kick a running VCPU, but otherwise there is nothing to do.
214 return mode
== IN_GUEST_MODE
;
217 static void ack_flush(void *_completed
)
221 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
224 cpus
= cpu_online_mask
;
226 if (cpumask_empty(cpus
))
229 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
233 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
238 struct kvm_vcpu
*vcpu
;
240 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
243 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
244 kvm_make_request(req
, vcpu
);
247 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
250 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
251 kvm_request_needs_ipi(vcpu
, req
))
252 __cpumask_set_cpu(cpu
, cpus
);
254 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
256 free_cpumask_var(cpus
);
260 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
261 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
264 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
265 * kvm_make_all_cpus_request.
267 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
270 * We want to publish modifications to the page tables before reading
271 * mode. Pairs with a memory barrier in arch-specific code.
272 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
273 * and smp_mb in walk_shadow_page_lockless_begin/end.
274 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
276 * There is already an smp_mb__after_atomic() before
277 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
280 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
281 ++kvm
->stat
.remote_tlb_flush
;
282 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
284 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
287 void kvm_reload_remote_mmus(struct kvm
*kvm
)
289 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
292 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
297 mutex_init(&vcpu
->mutex
);
302 init_swait_queue_head(&vcpu
->wq
);
303 kvm_async_pf_vcpu_init(vcpu
);
306 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
308 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
313 vcpu
->run
= page_address(page
);
315 kvm_vcpu_set_in_spin_loop(vcpu
, false);
316 kvm_vcpu_set_dy_eligible(vcpu
, false);
317 vcpu
->preempted
= false;
319 r
= kvm_arch_vcpu_init(vcpu
);
325 free_page((unsigned long)vcpu
->run
);
329 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
331 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
334 * no need for rcu_read_lock as VCPU_RUN is the only place that
335 * will change the vcpu->pid pointer and on uninit all file
336 * descriptors are already gone.
338 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
339 kvm_arch_vcpu_uninit(vcpu
);
340 free_page((unsigned long)vcpu
->run
);
342 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
344 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
345 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
347 return container_of(mn
, struct kvm
, mmu_notifier
);
350 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
351 struct mm_struct
*mm
,
352 unsigned long address
,
355 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
358 idx
= srcu_read_lock(&kvm
->srcu
);
359 spin_lock(&kvm
->mmu_lock
);
360 kvm
->mmu_notifier_seq
++;
361 kvm_set_spte_hva(kvm
, address
, pte
);
362 spin_unlock(&kvm
->mmu_lock
);
363 srcu_read_unlock(&kvm
->srcu
, idx
);
366 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
367 struct mm_struct
*mm
,
371 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
372 int need_tlb_flush
= 0, idx
;
374 idx
= srcu_read_lock(&kvm
->srcu
);
375 spin_lock(&kvm
->mmu_lock
);
377 * The count increase must become visible at unlock time as no
378 * spte can be established without taking the mmu_lock and
379 * count is also read inside the mmu_lock critical section.
381 kvm
->mmu_notifier_count
++;
382 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
383 need_tlb_flush
|= kvm
->tlbs_dirty
;
384 /* we've to flush the tlb before the pages can be freed */
386 kvm_flush_remote_tlbs(kvm
);
388 spin_unlock(&kvm
->mmu_lock
);
390 kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
);
392 srcu_read_unlock(&kvm
->srcu
, idx
);
395 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
396 struct mm_struct
*mm
,
400 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
402 spin_lock(&kvm
->mmu_lock
);
404 * This sequence increase will notify the kvm page fault that
405 * the page that is going to be mapped in the spte could have
408 kvm
->mmu_notifier_seq
++;
411 * The above sequence increase must be visible before the
412 * below count decrease, which is ensured by the smp_wmb above
413 * in conjunction with the smp_rmb in mmu_notifier_retry().
415 kvm
->mmu_notifier_count
--;
416 spin_unlock(&kvm
->mmu_lock
);
418 BUG_ON(kvm
->mmu_notifier_count
< 0);
421 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
422 struct mm_struct
*mm
,
426 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
429 idx
= srcu_read_lock(&kvm
->srcu
);
430 spin_lock(&kvm
->mmu_lock
);
432 young
= kvm_age_hva(kvm
, start
, end
);
434 kvm_flush_remote_tlbs(kvm
);
436 spin_unlock(&kvm
->mmu_lock
);
437 srcu_read_unlock(&kvm
->srcu
, idx
);
442 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
443 struct mm_struct
*mm
,
447 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
450 idx
= srcu_read_lock(&kvm
->srcu
);
451 spin_lock(&kvm
->mmu_lock
);
453 * Even though we do not flush TLB, this will still adversely
454 * affect performance on pre-Haswell Intel EPT, where there is
455 * no EPT Access Bit to clear so that we have to tear down EPT
456 * tables instead. If we find this unacceptable, we can always
457 * add a parameter to kvm_age_hva so that it effectively doesn't
458 * do anything on clear_young.
460 * Also note that currently we never issue secondary TLB flushes
461 * from clear_young, leaving this job up to the regular system
462 * cadence. If we find this inaccurate, we might come up with a
463 * more sophisticated heuristic later.
465 young
= kvm_age_hva(kvm
, start
, end
);
466 spin_unlock(&kvm
->mmu_lock
);
467 srcu_read_unlock(&kvm
->srcu
, idx
);
472 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
473 struct mm_struct
*mm
,
474 unsigned long address
)
476 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
479 idx
= srcu_read_lock(&kvm
->srcu
);
480 spin_lock(&kvm
->mmu_lock
);
481 young
= kvm_test_age_hva(kvm
, address
);
482 spin_unlock(&kvm
->mmu_lock
);
483 srcu_read_unlock(&kvm
->srcu
, idx
);
488 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
489 struct mm_struct
*mm
)
491 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
494 idx
= srcu_read_lock(&kvm
->srcu
);
495 kvm_arch_flush_shadow_all(kvm
);
496 srcu_read_unlock(&kvm
->srcu
, idx
);
499 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
500 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
501 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
502 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
503 .clear_young
= kvm_mmu_notifier_clear_young
,
504 .test_young
= kvm_mmu_notifier_test_young
,
505 .change_pte
= kvm_mmu_notifier_change_pte
,
506 .release
= kvm_mmu_notifier_release
,
509 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
511 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
512 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
515 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
517 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
522 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
524 static struct kvm_memslots
*kvm_alloc_memslots(void)
527 struct kvm_memslots
*slots
;
529 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
533 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
534 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
539 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
541 if (!memslot
->dirty_bitmap
)
544 kvfree(memslot
->dirty_bitmap
);
545 memslot
->dirty_bitmap
= NULL
;
549 * Free any memory in @free but not in @dont.
551 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
552 struct kvm_memory_slot
*dont
)
554 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
555 kvm_destroy_dirty_bitmap(free
);
557 kvm_arch_free_memslot(kvm
, free
, dont
);
562 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
564 struct kvm_memory_slot
*memslot
;
569 kvm_for_each_memslot(memslot
, slots
)
570 kvm_free_memslot(kvm
, memslot
, NULL
);
575 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
579 if (!kvm
->debugfs_dentry
)
582 debugfs_remove_recursive(kvm
->debugfs_dentry
);
584 if (kvm
->debugfs_stat_data
) {
585 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
586 kfree(kvm
->debugfs_stat_data
[i
]);
587 kfree(kvm
->debugfs_stat_data
);
591 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
593 char dir_name
[ITOA_MAX_LEN
* 2];
594 struct kvm_stat_data
*stat_data
;
595 struct kvm_stats_debugfs_item
*p
;
597 if (!debugfs_initialized())
600 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
601 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
603 if (!kvm
->debugfs_dentry
)
606 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
607 sizeof(*kvm
->debugfs_stat_data
),
609 if (!kvm
->debugfs_stat_data
)
612 for (p
= debugfs_entries
; p
->name
; p
++) {
613 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
617 stat_data
->kvm
= kvm
;
618 stat_data
->offset
= p
->offset
;
619 stat_data
->mode
= p
->mode
? p
->mode
: 0644;
620 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
621 if (!debugfs_create_file(p
->name
, stat_data
->mode
,
624 stat_fops_per_vm
[p
->kind
]))
631 * Called after the VM is otherwise initialized, but just before adding it to
634 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
640 * Called just after removing the VM from the vm_list, but before doing any
643 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
647 static struct kvm
*kvm_create_vm(unsigned long type
)
650 struct kvm
*kvm
= kvm_arch_alloc_vm();
653 return ERR_PTR(-ENOMEM
);
655 spin_lock_init(&kvm
->mmu_lock
);
657 kvm
->mm
= current
->mm
;
658 kvm_eventfd_init(kvm
);
659 mutex_init(&kvm
->lock
);
660 mutex_init(&kvm
->irq_lock
);
661 mutex_init(&kvm
->slots_lock
);
662 refcount_set(&kvm
->users_count
, 1);
663 INIT_LIST_HEAD(&kvm
->devices
);
665 r
= kvm_arch_init_vm(kvm
, type
);
667 goto out_err_no_disable
;
669 r
= hardware_enable_all();
671 goto out_err_no_disable
;
673 #ifdef CONFIG_HAVE_KVM_IRQFD
674 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
677 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
680 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
681 struct kvm_memslots
*slots
= kvm_alloc_memslots();
683 goto out_err_no_srcu
;
685 * Generations must be different for each address space.
686 * Init kvm generation close to the maximum to easily test the
687 * code of handling generation number wrap-around.
689 slots
->generation
= i
* 2 - 150;
690 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
693 if (init_srcu_struct(&kvm
->srcu
))
694 goto out_err_no_srcu
;
695 if (init_srcu_struct(&kvm
->irq_srcu
))
696 goto out_err_no_irq_srcu
;
697 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
698 rcu_assign_pointer(kvm
->buses
[i
],
699 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
701 goto out_err_no_mmu_notifier
;
704 r
= kvm_init_mmu_notifier(kvm
);
706 goto out_err_no_mmu_notifier
;
708 r
= kvm_arch_post_init_vm(kvm
);
712 mutex_lock(&kvm_lock
);
713 list_add(&kvm
->vm_list
, &vm_list
);
714 mutex_unlock(&kvm_lock
);
716 preempt_notifier_inc();
721 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
722 if (kvm
->mmu_notifier
.ops
)
723 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
725 out_err_no_mmu_notifier
:
726 cleanup_srcu_struct(&kvm
->irq_srcu
);
728 cleanup_srcu_struct(&kvm
->srcu
);
730 hardware_disable_all();
732 refcount_set(&kvm
->users_count
, 0);
733 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
734 kfree(kvm_get_bus(kvm
, i
));
735 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
736 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
737 kvm_arch_free_vm(kvm
);
742 static void kvm_destroy_devices(struct kvm
*kvm
)
744 struct kvm_device
*dev
, *tmp
;
747 * We do not need to take the kvm->lock here, because nobody else
748 * has a reference to the struct kvm at this point and therefore
749 * cannot access the devices list anyhow.
751 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
752 list_del(&dev
->vm_node
);
753 dev
->ops
->destroy(dev
);
757 static void kvm_destroy_vm(struct kvm
*kvm
)
760 struct mm_struct
*mm
= kvm
->mm
;
762 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
763 kvm_destroy_vm_debugfs(kvm
);
764 kvm_arch_sync_events(kvm
);
765 mutex_lock(&kvm_lock
);
766 list_del(&kvm
->vm_list
);
767 mutex_unlock(&kvm_lock
);
768 kvm_arch_pre_destroy_vm(kvm
);
770 kvm_free_irq_routing(kvm
);
771 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
772 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
775 kvm_io_bus_destroy(bus
);
776 kvm
->buses
[i
] = NULL
;
778 kvm_coalesced_mmio_free(kvm
);
779 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
780 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
782 kvm_arch_flush_shadow_all(kvm
);
784 kvm_arch_destroy_vm(kvm
);
785 kvm_destroy_devices(kvm
);
786 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
787 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
788 cleanup_srcu_struct(&kvm
->irq_srcu
);
789 cleanup_srcu_struct(&kvm
->srcu
);
790 kvm_arch_free_vm(kvm
);
791 preempt_notifier_dec();
792 hardware_disable_all();
796 void kvm_get_kvm(struct kvm
*kvm
)
798 refcount_inc(&kvm
->users_count
);
800 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
802 void kvm_put_kvm(struct kvm
*kvm
)
804 if (refcount_dec_and_test(&kvm
->users_count
))
807 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
810 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
812 struct kvm
*kvm
= filp
->private_data
;
814 kvm_irqfd_release(kvm
);
821 * Allocation size is twice as large as the actual dirty bitmap size.
822 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
824 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
826 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
828 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
829 if (!memslot
->dirty_bitmap
)
836 * Insert memslot and re-sort memslots based on their GFN,
837 * so binary search could be used to lookup GFN.
838 * Sorting algorithm takes advantage of having initially
839 * sorted array and known changed memslot position.
841 static void update_memslots(struct kvm_memslots
*slots
,
842 struct kvm_memory_slot
*new)
845 int i
= slots
->id_to_index
[id
];
846 struct kvm_memory_slot
*mslots
= slots
->memslots
;
848 WARN_ON(mslots
[i
].id
!= id
);
850 WARN_ON(!mslots
[i
].npages
);
851 if (mslots
[i
].npages
)
854 if (!mslots
[i
].npages
)
858 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
859 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
860 if (!mslots
[i
+ 1].npages
)
862 mslots
[i
] = mslots
[i
+ 1];
863 slots
->id_to_index
[mslots
[i
].id
] = i
;
868 * The ">=" is needed when creating a slot with base_gfn == 0,
869 * so that it moves before all those with base_gfn == npages == 0.
871 * On the other hand, if new->npages is zero, the above loop has
872 * already left i pointing to the beginning of the empty part of
873 * mslots, and the ">=" would move the hole backwards in this
874 * case---which is wrong. So skip the loop when deleting a slot.
878 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
879 mslots
[i
] = mslots
[i
- 1];
880 slots
->id_to_index
[mslots
[i
].id
] = i
;
884 WARN_ON_ONCE(i
!= slots
->used_slots
);
887 slots
->id_to_index
[mslots
[i
].id
] = i
;
890 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
892 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
894 #ifdef __KVM_HAVE_READONLY_MEM
895 valid_flags
|= KVM_MEM_READONLY
;
898 if (mem
->flags
& ~valid_flags
)
904 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
905 int as_id
, struct kvm_memslots
*slots
)
907 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
911 * Set the low bit in the generation, which disables SPTE caching
912 * until the end of synchronize_srcu_expedited.
914 WARN_ON(old_memslots
->generation
& 1);
915 slots
->generation
= old_memslots
->generation
+ 1;
917 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
918 synchronize_srcu_expedited(&kvm
->srcu
);
921 * Increment the new memslot generation a second time. This prevents
922 * vm exits that race with memslot updates from caching a memslot
923 * generation that will (potentially) be valid forever.
925 * Generations must be unique even across address spaces. We do not need
926 * a global counter for that, instead the generation space is evenly split
927 * across address spaces. For example, with two address spaces, address
928 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
929 * use generations 2, 6, 10, 14, ...
931 gen
= slots
->generation
+ KVM_ADDRESS_SPACE_NUM
* 2 - 1;
933 kvm_arch_memslots_updated(kvm
, gen
);
935 slots
->generation
= gen
;
941 * Allocate some memory and give it an address in the guest physical address
944 * Discontiguous memory is allowed, mostly for framebuffers.
946 * Must be called holding kvm->slots_lock for write.
948 int __kvm_set_memory_region(struct kvm
*kvm
,
949 const struct kvm_userspace_memory_region
*mem
)
953 unsigned long npages
;
954 struct kvm_memory_slot
*slot
;
955 struct kvm_memory_slot old
, new;
956 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
958 enum kvm_mr_change change
;
960 r
= check_memory_region_flags(mem
);
965 as_id
= mem
->slot
>> 16;
968 /* General sanity checks */
969 if (mem
->memory_size
& (PAGE_SIZE
- 1))
971 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
973 /* We can read the guest memory with __xxx_user() later on. */
974 if ((id
< KVM_USER_MEM_SLOTS
) &&
975 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
976 !access_ok(VERIFY_WRITE
,
977 (void __user
*)(unsigned long)mem
->userspace_addr
,
980 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
982 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
985 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
986 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
987 npages
= mem
->memory_size
>> PAGE_SHIFT
;
989 if (npages
> KVM_MEM_MAX_NR_PAGES
)
995 new.base_gfn
= base_gfn
;
997 new.flags
= mem
->flags
;
1001 change
= KVM_MR_CREATE
;
1002 else { /* Modify an existing slot. */
1003 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
1004 (npages
!= old
.npages
) ||
1005 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
1008 if (base_gfn
!= old
.base_gfn
)
1009 change
= KVM_MR_MOVE
;
1010 else if (new.flags
!= old
.flags
)
1011 change
= KVM_MR_FLAGS_ONLY
;
1012 else { /* Nothing to change. */
1021 change
= KVM_MR_DELETE
;
1026 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1027 /* Check for overlaps */
1029 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
1032 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
1033 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
1038 /* Free page dirty bitmap if unneeded */
1039 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1040 new.dirty_bitmap
= NULL
;
1043 if (change
== KVM_MR_CREATE
) {
1044 new.userspace_addr
= mem
->userspace_addr
;
1046 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1050 /* Allocate page dirty bitmap if needed */
1051 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1052 if (kvm_create_dirty_bitmap(&new) < 0)
1056 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1059 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1061 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1062 slot
= id_to_memslot(slots
, id
);
1063 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1065 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1067 /* From this point no new shadow pages pointing to a deleted,
1068 * or moved, memslot will be created.
1070 * validation of sp->gfn happens in:
1071 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1072 * - kvm_is_visible_gfn (mmu_check_roots)
1074 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1077 * We can re-use the old_memslots from above, the only difference
1078 * from the currently installed memslots is the invalid flag. This
1079 * will get overwritten by update_memslots anyway.
1081 slots
= old_memslots
;
1084 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1088 /* actual memory is freed via old in kvm_free_memslot below */
1089 if (change
== KVM_MR_DELETE
) {
1090 new.dirty_bitmap
= NULL
;
1091 memset(&new.arch
, 0, sizeof(new.arch
));
1094 update_memslots(slots
, &new);
1095 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1097 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1099 kvm_free_memslot(kvm
, &old
, &new);
1100 kvfree(old_memslots
);
1106 kvm_free_memslot(kvm
, &new, &old
);
1110 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1112 int kvm_set_memory_region(struct kvm
*kvm
,
1113 const struct kvm_userspace_memory_region
*mem
)
1117 mutex_lock(&kvm
->slots_lock
);
1118 r
= __kvm_set_memory_region(kvm
, mem
);
1119 mutex_unlock(&kvm
->slots_lock
);
1122 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1124 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1125 struct kvm_userspace_memory_region
*mem
)
1127 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1130 return kvm_set_memory_region(kvm
, mem
);
1133 int kvm_get_dirty_log(struct kvm
*kvm
,
1134 struct kvm_dirty_log
*log
, int *is_dirty
)
1136 struct kvm_memslots
*slots
;
1137 struct kvm_memory_slot
*memslot
;
1140 unsigned long any
= 0;
1142 as_id
= log
->slot
>> 16;
1143 id
= (u16
)log
->slot
;
1144 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1147 slots
= __kvm_memslots(kvm
, as_id
);
1148 memslot
= id_to_memslot(slots
, id
);
1149 if (!memslot
->dirty_bitmap
)
1152 n
= kvm_dirty_bitmap_bytes(memslot
);
1154 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1155 any
= memslot
->dirty_bitmap
[i
];
1157 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1164 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1166 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1168 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1169 * are dirty write protect them for next write.
1170 * @kvm: pointer to kvm instance
1171 * @log: slot id and address to which we copy the log
1172 * @is_dirty: flag set if any page is dirty
1174 * We need to keep it in mind that VCPU threads can write to the bitmap
1175 * concurrently. So, to avoid losing track of dirty pages we keep the
1178 * 1. Take a snapshot of the bit and clear it if needed.
1179 * 2. Write protect the corresponding page.
1180 * 3. Copy the snapshot to the userspace.
1181 * 4. Upon return caller flushes TLB's if needed.
1183 * Between 2 and 4, the guest may write to the page using the remaining TLB
1184 * entry. This is not a problem because the page is reported dirty using
1185 * the snapshot taken before and step 4 ensures that writes done after
1186 * exiting to userspace will be logged for the next call.
1189 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1190 struct kvm_dirty_log
*log
, bool *is_dirty
)
1192 struct kvm_memslots
*slots
;
1193 struct kvm_memory_slot
*memslot
;
1196 unsigned long *dirty_bitmap
;
1197 unsigned long *dirty_bitmap_buffer
;
1199 as_id
= log
->slot
>> 16;
1200 id
= (u16
)log
->slot
;
1201 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1204 slots
= __kvm_memslots(kvm
, as_id
);
1205 memslot
= id_to_memslot(slots
, id
);
1207 dirty_bitmap
= memslot
->dirty_bitmap
;
1211 n
= kvm_dirty_bitmap_bytes(memslot
);
1213 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1214 memset(dirty_bitmap_buffer
, 0, n
);
1216 spin_lock(&kvm
->mmu_lock
);
1218 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1222 if (!dirty_bitmap
[i
])
1227 mask
= xchg(&dirty_bitmap
[i
], 0);
1228 dirty_bitmap_buffer
[i
] = mask
;
1231 offset
= i
* BITS_PER_LONG
;
1232 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1237 spin_unlock(&kvm
->mmu_lock
);
1238 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1242 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1245 bool kvm_largepages_enabled(void)
1247 return largepages_enabled
;
1250 void kvm_disable_largepages(void)
1252 largepages_enabled
= false;
1254 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1256 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1258 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1260 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1262 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1264 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1267 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1269 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1271 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1272 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1277 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1279 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1281 struct vm_area_struct
*vma
;
1282 unsigned long addr
, size
;
1286 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
1287 if (kvm_is_error_hva(addr
))
1290 down_read(¤t
->mm
->mmap_sem
);
1291 vma
= find_vma(current
->mm
, addr
);
1295 size
= vma_kernel_pagesize(vma
);
1298 up_read(¤t
->mm
->mmap_sem
);
1303 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1305 return slot
->flags
& KVM_MEM_READONLY
;
1308 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1309 gfn_t
*nr_pages
, bool write
)
1311 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1312 return KVM_HVA_ERR_BAD
;
1314 if (memslot_is_readonly(slot
) && write
)
1315 return KVM_HVA_ERR_RO_BAD
;
1318 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1320 return __gfn_to_hva_memslot(slot
, gfn
);
1323 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1326 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1329 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1332 return gfn_to_hva_many(slot
, gfn
, NULL
);
1334 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1336 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1338 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1340 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1342 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1344 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1346 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1349 * If writable is set to false, the hva returned by this function is only
1350 * allowed to be read.
1352 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1353 gfn_t gfn
, bool *writable
)
1355 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1357 if (!kvm_is_error_hva(hva
) && writable
)
1358 *writable
= !memslot_is_readonly(slot
);
1363 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1365 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1367 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1370 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1372 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1374 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1377 static int get_user_page_nowait(unsigned long start
, int write
,
1380 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1383 flags
|= FOLL_WRITE
;
1385 return get_user_pages(start
, 1, flags
, page
, NULL
);
1388 static inline int check_user_page_hwpoison(unsigned long addr
)
1390 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1392 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1393 return rc
== -EHWPOISON
;
1397 * The atomic path to get the writable pfn which will be stored in @pfn,
1398 * true indicates success, otherwise false is returned.
1400 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1401 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1403 struct page
*page
[1];
1406 if (!(async
|| atomic
))
1410 * Fast pin a writable pfn only if it is a write fault request
1411 * or the caller allows to map a writable pfn for a read fault
1414 if (!(write_fault
|| writable
))
1417 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1419 *pfn
= page_to_pfn(page
[0]);
1430 * The slow path to get the pfn of the specified host virtual address,
1431 * 1 indicates success, -errno is returned if error is detected.
1433 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1434 bool *writable
, kvm_pfn_t
*pfn
)
1436 struct page
*page
[1];
1442 *writable
= write_fault
;
1445 down_read(¤t
->mm
->mmap_sem
);
1446 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1447 up_read(¤t
->mm
->mmap_sem
);
1449 unsigned int flags
= FOLL_HWPOISON
;
1452 flags
|= FOLL_WRITE
;
1454 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1459 /* map read fault as writable if possible */
1460 if (unlikely(!write_fault
) && writable
) {
1461 struct page
*wpage
[1];
1463 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1472 *pfn
= page_to_pfn(page
[0]);
1476 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1478 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1481 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1487 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1488 unsigned long addr
, bool *async
,
1489 bool write_fault
, bool *writable
,
1495 r
= follow_pfn(vma
, addr
, &pfn
);
1498 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1499 * not call the fault handler, so do it here.
1501 bool unlocked
= false;
1502 r
= fixup_user_fault(current
, current
->mm
, addr
,
1503 (write_fault
? FAULT_FLAG_WRITE
: 0),
1510 r
= follow_pfn(vma
, addr
, &pfn
);
1520 * Get a reference here because callers of *hva_to_pfn* and
1521 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1522 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1523 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1524 * simply do nothing for reserved pfns.
1526 * Whoever called remap_pfn_range is also going to call e.g.
1527 * unmap_mapping_range before the underlying pages are freed,
1528 * causing a call to our MMU notifier.
1537 * Pin guest page in memory and return its pfn.
1538 * @addr: host virtual address which maps memory to the guest
1539 * @atomic: whether this function can sleep
1540 * @async: whether this function need to wait IO complete if the
1541 * host page is not in the memory
1542 * @write_fault: whether we should get a writable host page
1543 * @writable: whether it allows to map a writable host page for !@write_fault
1545 * The function will map a writable host page for these two cases:
1546 * 1): @write_fault = true
1547 * 2): @write_fault = false && @writable, @writable will tell the caller
1548 * whether the mapping is writable.
1550 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1551 bool write_fault
, bool *writable
)
1553 struct vm_area_struct
*vma
;
1557 /* we can do it either atomically or asynchronously, not both */
1558 BUG_ON(atomic
&& async
);
1560 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1564 return KVM_PFN_ERR_FAULT
;
1566 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1570 down_read(¤t
->mm
->mmap_sem
);
1571 if (npages
== -EHWPOISON
||
1572 (!async
&& check_user_page_hwpoison(addr
))) {
1573 pfn
= KVM_PFN_ERR_HWPOISON
;
1578 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1581 pfn
= KVM_PFN_ERR_FAULT
;
1582 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1583 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1587 pfn
= KVM_PFN_ERR_FAULT
;
1589 if (async
&& vma_is_valid(vma
, write_fault
))
1591 pfn
= KVM_PFN_ERR_FAULT
;
1594 up_read(¤t
->mm
->mmap_sem
);
1598 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1599 bool atomic
, bool *async
, bool write_fault
,
1602 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1604 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1607 return KVM_PFN_ERR_RO_FAULT
;
1610 if (kvm_is_error_hva(addr
)) {
1613 return KVM_PFN_NOSLOT
;
1616 /* Do not map writable pfn in the readonly memslot. */
1617 if (writable
&& memslot_is_readonly(slot
)) {
1622 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1625 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1627 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1630 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1631 write_fault
, writable
);
1633 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1635 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1637 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1639 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1641 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1643 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1645 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1647 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1649 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1651 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1653 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1655 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1657 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1659 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1661 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1663 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1665 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1667 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1669 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1671 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1672 struct page
**pages
, int nr_pages
)
1677 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1678 if (kvm_is_error_hva(addr
))
1681 if (entry
< nr_pages
)
1684 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1686 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1688 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1690 if (is_error_noslot_pfn(pfn
))
1691 return KVM_ERR_PTR_BAD_PAGE
;
1693 if (kvm_is_reserved_pfn(pfn
)) {
1695 return KVM_ERR_PTR_BAD_PAGE
;
1698 return pfn_to_page(pfn
);
1701 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1705 pfn
= gfn_to_pfn(kvm
, gfn
);
1707 return kvm_pfn_to_page(pfn
);
1709 EXPORT_SYMBOL_GPL(gfn_to_page
);
1711 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1715 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1717 return kvm_pfn_to_page(pfn
);
1719 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1721 void kvm_release_page_clean(struct page
*page
)
1723 WARN_ON(is_error_page(page
));
1725 kvm_release_pfn_clean(page_to_pfn(page
));
1727 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1729 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1731 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1732 put_page(pfn_to_page(pfn
));
1734 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1736 void kvm_release_page_dirty(struct page
*page
)
1738 WARN_ON(is_error_page(page
));
1740 kvm_release_pfn_dirty(page_to_pfn(page
));
1742 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1744 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1746 kvm_set_pfn_dirty(pfn
);
1747 kvm_release_pfn_clean(pfn
);
1749 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1751 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1753 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
)) {
1754 struct page
*page
= pfn_to_page(pfn
);
1756 if (!PageReserved(page
))
1760 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1762 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1764 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
1765 mark_page_accessed(pfn_to_page(pfn
));
1767 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1769 void kvm_get_pfn(kvm_pfn_t pfn
)
1771 if (!kvm_is_reserved_pfn(pfn
))
1772 get_page(pfn_to_page(pfn
));
1774 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1776 static int next_segment(unsigned long len
, int offset
)
1778 if (len
> PAGE_SIZE
- offset
)
1779 return PAGE_SIZE
- offset
;
1784 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1785 void *data
, int offset
, int len
)
1790 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1791 if (kvm_is_error_hva(addr
))
1793 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1799 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1802 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1804 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1806 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1808 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1809 int offset
, int len
)
1811 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1813 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1815 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1817 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1819 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1821 int offset
= offset_in_page(gpa
);
1824 while ((seg
= next_segment(len
, offset
)) != 0) {
1825 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1835 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1837 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1839 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1841 int offset
= offset_in_page(gpa
);
1844 while ((seg
= next_segment(len
, offset
)) != 0) {
1845 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1855 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1857 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1858 void *data
, int offset
, unsigned long len
)
1863 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1864 if (kvm_is_error_hva(addr
))
1866 pagefault_disable();
1867 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1874 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1877 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1878 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1879 int offset
= offset_in_page(gpa
);
1881 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1883 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1885 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1886 void *data
, unsigned long len
)
1888 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1889 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1890 int offset
= offset_in_page(gpa
);
1892 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1894 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1896 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1897 const void *data
, int offset
, int len
)
1902 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1903 if (kvm_is_error_hva(addr
))
1905 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1908 mark_page_dirty_in_slot(memslot
, gfn
);
1912 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1913 const void *data
, int offset
, int len
)
1915 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1917 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1919 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1921 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1922 const void *data
, int offset
, int len
)
1924 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1926 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1928 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1930 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1933 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1935 int offset
= offset_in_page(gpa
);
1938 while ((seg
= next_segment(len
, offset
)) != 0) {
1939 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1949 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1951 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1954 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1956 int offset
= offset_in_page(gpa
);
1959 while ((seg
= next_segment(len
, offset
)) != 0) {
1960 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1970 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1972 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1973 struct gfn_to_hva_cache
*ghc
,
1974 gpa_t gpa
, unsigned long len
)
1976 int offset
= offset_in_page(gpa
);
1977 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1978 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1979 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1980 gfn_t nr_pages_avail
;
1983 ghc
->generation
= slots
->generation
;
1985 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1986 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1987 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1991 * If the requested region crosses two memslots, we still
1992 * verify that the entire region is valid here.
1994 while (start_gfn
<= end_gfn
) {
1996 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1997 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1999 if (kvm_is_error_hva(ghc
->hva
))
2001 start_gfn
+= nr_pages_avail
;
2003 /* Use the slow path for cross page reads and writes. */
2004 ghc
->memslot
= NULL
;
2009 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2010 gpa_t gpa
, unsigned long len
)
2012 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2013 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
2015 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
2017 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2018 void *data
, unsigned int offset
,
2021 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2023 gpa_t gpa
= ghc
->gpa
+ offset
;
2025 BUG_ON(len
+ offset
> ghc
->len
);
2027 if (slots
->generation
!= ghc
->generation
)
2028 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2030 if (unlikely(!ghc
->memslot
))
2031 return kvm_write_guest(kvm
, gpa
, data
, len
);
2033 if (kvm_is_error_hva(ghc
->hva
))
2036 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2039 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2043 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2045 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2046 void *data
, unsigned long len
)
2048 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2050 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2052 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2053 void *data
, unsigned long len
)
2055 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2058 BUG_ON(len
> ghc
->len
);
2060 if (slots
->generation
!= ghc
->generation
)
2061 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2063 if (unlikely(!ghc
->memslot
))
2064 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2066 if (kvm_is_error_hva(ghc
->hva
))
2069 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2075 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2077 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2079 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2081 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2083 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2085 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2087 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2089 int offset
= offset_in_page(gpa
);
2092 while ((seg
= next_segment(len
, offset
)) != 0) {
2093 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2102 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2104 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2107 if (memslot
&& memslot
->dirty_bitmap
) {
2108 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2110 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2114 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2116 struct kvm_memory_slot
*memslot
;
2118 memslot
= gfn_to_memslot(kvm
, gfn
);
2119 mark_page_dirty_in_slot(memslot
, gfn
);
2121 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2123 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2125 struct kvm_memory_slot
*memslot
;
2127 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2128 mark_page_dirty_in_slot(memslot
, gfn
);
2130 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2132 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2134 if (!vcpu
->sigset_active
)
2138 * This does a lockless modification of ->real_blocked, which is fine
2139 * because, only current can change ->real_blocked and all readers of
2140 * ->real_blocked don't care as long ->real_blocked is always a subset
2143 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2146 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2148 if (!vcpu
->sigset_active
)
2151 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2152 sigemptyset(¤t
->real_blocked
);
2155 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2157 unsigned int old
, val
, grow
;
2159 old
= val
= vcpu
->halt_poll_ns
;
2160 grow
= READ_ONCE(halt_poll_ns_grow
);
2162 if (val
== 0 && grow
)
2167 if (val
> halt_poll_ns
)
2170 vcpu
->halt_poll_ns
= val
;
2171 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2174 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2176 unsigned int old
, val
, shrink
;
2178 old
= val
= vcpu
->halt_poll_ns
;
2179 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2185 vcpu
->halt_poll_ns
= val
;
2186 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2189 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2191 if (kvm_arch_vcpu_runnable(vcpu
)) {
2192 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2195 if (kvm_cpu_has_pending_timer(vcpu
))
2197 if (signal_pending(current
))
2204 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2206 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2209 DECLARE_SWAITQUEUE(wait
);
2210 bool waited
= false;
2213 start
= cur
= ktime_get();
2214 if (vcpu
->halt_poll_ns
) {
2215 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2217 ++vcpu
->stat
.halt_attempted_poll
;
2220 * This sets KVM_REQ_UNHALT if an interrupt
2223 if (kvm_vcpu_check_block(vcpu
) < 0) {
2224 ++vcpu
->stat
.halt_successful_poll
;
2225 if (!vcpu_valid_wakeup(vcpu
))
2226 ++vcpu
->stat
.halt_poll_invalid
;
2230 } while (single_task_running() && ktime_before(cur
, stop
));
2233 kvm_arch_vcpu_blocking(vcpu
);
2236 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2238 if (kvm_vcpu_check_block(vcpu
) < 0)
2245 finish_swait(&vcpu
->wq
, &wait
);
2248 kvm_arch_vcpu_unblocking(vcpu
);
2250 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2252 if (!vcpu_valid_wakeup(vcpu
))
2253 shrink_halt_poll_ns(vcpu
);
2254 else if (halt_poll_ns
) {
2255 if (block_ns
<= vcpu
->halt_poll_ns
)
2257 /* we had a long block, shrink polling */
2258 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2259 shrink_halt_poll_ns(vcpu
);
2260 /* we had a short halt and our poll time is too small */
2261 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2262 block_ns
< halt_poll_ns
)
2263 grow_halt_poll_ns(vcpu
);
2265 vcpu
->halt_poll_ns
= 0;
2267 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2268 kvm_arch_vcpu_block_finish(vcpu
);
2270 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2272 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2274 struct swait_queue_head
*wqp
;
2276 wqp
= kvm_arch_vcpu_wq(vcpu
);
2277 if (swq_has_sleeper(wqp
)) {
2279 ++vcpu
->stat
.halt_wakeup
;
2285 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2289 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2291 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2294 int cpu
= vcpu
->cpu
;
2296 if (kvm_vcpu_wake_up(vcpu
))
2300 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2301 if (kvm_arch_vcpu_should_kick(vcpu
))
2302 smp_send_reschedule(cpu
);
2305 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2306 #endif /* !CONFIG_S390 */
2308 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2311 struct task_struct
*task
= NULL
;
2315 pid
= rcu_dereference(target
->pid
);
2317 task
= get_pid_task(pid
, PIDTYPE_PID
);
2321 ret
= yield_to(task
, 1);
2322 put_task_struct(task
);
2326 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2329 * Helper that checks whether a VCPU is eligible for directed yield.
2330 * Most eligible candidate to yield is decided by following heuristics:
2332 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2333 * (preempted lock holder), indicated by @in_spin_loop.
2334 * Set at the beiginning and cleared at the end of interception/PLE handler.
2336 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2337 * chance last time (mostly it has become eligible now since we have probably
2338 * yielded to lockholder in last iteration. This is done by toggling
2339 * @dy_eligible each time a VCPU checked for eligibility.)
2341 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2342 * to preempted lock-holder could result in wrong VCPU selection and CPU
2343 * burning. Giving priority for a potential lock-holder increases lock
2346 * Since algorithm is based on heuristics, accessing another VCPU data without
2347 * locking does not harm. It may result in trying to yield to same VCPU, fail
2348 * and continue with next VCPU and so on.
2350 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2352 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2355 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2356 vcpu
->spin_loop
.dy_eligible
;
2358 if (vcpu
->spin_loop
.in_spin_loop
)
2359 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2368 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2369 * a vcpu_load/vcpu_put pair. However, for most architectures
2370 * kvm_arch_vcpu_runnable does not require vcpu_load.
2372 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
2374 return kvm_arch_vcpu_runnable(vcpu
);
2377 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
2379 if (kvm_arch_dy_runnable(vcpu
))
2382 #ifdef CONFIG_KVM_ASYNC_PF
2383 if (!list_empty_careful(&vcpu
->async_pf
.done
))
2390 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2392 struct kvm
*kvm
= me
->kvm
;
2393 struct kvm_vcpu
*vcpu
;
2394 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2400 kvm_vcpu_set_in_spin_loop(me
, true);
2402 * We boost the priority of a VCPU that is runnable but not
2403 * currently running, because it got preempted by something
2404 * else and called schedule in __vcpu_run. Hopefully that
2405 * VCPU is holding the lock that we need and will release it.
2406 * We approximate round-robin by starting at the last boosted VCPU.
2408 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2409 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2410 if (!pass
&& i
<= last_boosted_vcpu
) {
2411 i
= last_boosted_vcpu
;
2413 } else if (pass
&& i
> last_boosted_vcpu
)
2415 if (!READ_ONCE(vcpu
->preempted
))
2419 if (swait_active(&vcpu
->wq
) && !vcpu_dy_runnable(vcpu
))
2421 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2423 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2426 yielded
= kvm_vcpu_yield_to(vcpu
);
2428 kvm
->last_boosted_vcpu
= i
;
2430 } else if (yielded
< 0) {
2437 kvm_vcpu_set_in_spin_loop(me
, false);
2439 /* Ensure vcpu is not eligible during next spinloop */
2440 kvm_vcpu_set_dy_eligible(me
, false);
2442 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2444 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2446 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2449 if (vmf
->pgoff
== 0)
2450 page
= virt_to_page(vcpu
->run
);
2452 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2453 page
= virt_to_page(vcpu
->arch
.pio_data
);
2455 #ifdef CONFIG_KVM_MMIO
2456 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2457 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2460 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2466 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2467 .fault
= kvm_vcpu_fault
,
2470 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2472 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2476 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2478 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2480 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2481 kvm_put_kvm(vcpu
->kvm
);
2485 static struct file_operations kvm_vcpu_fops
= {
2486 .release
= kvm_vcpu_release
,
2487 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2488 #ifdef CONFIG_KVM_COMPAT
2489 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2491 .mmap
= kvm_vcpu_mmap
,
2492 .llseek
= noop_llseek
,
2496 * Allocates an inode for the vcpu.
2498 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2500 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2503 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2505 char dir_name
[ITOA_MAX_LEN
* 2];
2508 if (!kvm_arch_has_vcpu_debugfs())
2511 if (!debugfs_initialized())
2514 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2515 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2516 vcpu
->kvm
->debugfs_dentry
);
2517 if (!vcpu
->debugfs_dentry
)
2520 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2522 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2530 * Creates some virtual cpus. Good luck creating more than one.
2532 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2535 struct kvm_vcpu
*vcpu
;
2537 if (id
>= KVM_MAX_VCPU_ID
)
2540 mutex_lock(&kvm
->lock
);
2541 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2542 mutex_unlock(&kvm
->lock
);
2546 kvm
->created_vcpus
++;
2547 mutex_unlock(&kvm
->lock
);
2549 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2552 goto vcpu_decrement
;
2555 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2557 r
= kvm_arch_vcpu_setup(vcpu
);
2561 r
= kvm_create_vcpu_debugfs(vcpu
);
2565 mutex_lock(&kvm
->lock
);
2566 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2568 goto unlock_vcpu_destroy
;
2571 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2573 /* Now it's all set up, let userspace reach it */
2575 r
= create_vcpu_fd(vcpu
);
2578 goto unlock_vcpu_destroy
;
2581 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2584 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2585 * before kvm->online_vcpu's incremented value.
2588 atomic_inc(&kvm
->online_vcpus
);
2590 mutex_unlock(&kvm
->lock
);
2591 kvm_arch_vcpu_postcreate(vcpu
);
2594 unlock_vcpu_destroy
:
2595 mutex_unlock(&kvm
->lock
);
2596 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2598 kvm_arch_vcpu_destroy(vcpu
);
2600 mutex_lock(&kvm
->lock
);
2601 kvm
->created_vcpus
--;
2602 mutex_unlock(&kvm
->lock
);
2606 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2609 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2610 vcpu
->sigset_active
= 1;
2611 vcpu
->sigset
= *sigset
;
2613 vcpu
->sigset_active
= 0;
2617 static long kvm_vcpu_ioctl(struct file
*filp
,
2618 unsigned int ioctl
, unsigned long arg
)
2620 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2621 void __user
*argp
= (void __user
*)arg
;
2623 struct kvm_fpu
*fpu
= NULL
;
2624 struct kvm_sregs
*kvm_sregs
= NULL
;
2626 if (vcpu
->kvm
->mm
!= current
->mm
)
2629 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2632 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2634 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2635 * so vcpu_load() would break it.
2637 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2638 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2642 r
= vcpu_load(vcpu
);
2651 oldpid
= rcu_access_pointer(vcpu
->pid
);
2652 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2653 /* The thread running this VCPU changed. */
2654 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2656 rcu_assign_pointer(vcpu
->pid
, newpid
);
2661 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2662 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2665 case KVM_GET_REGS
: {
2666 struct kvm_regs
*kvm_regs
;
2669 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2672 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2676 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2683 case KVM_SET_REGS
: {
2684 struct kvm_regs
*kvm_regs
;
2687 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2688 if (IS_ERR(kvm_regs
)) {
2689 r
= PTR_ERR(kvm_regs
);
2692 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2696 case KVM_GET_SREGS
: {
2697 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2701 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2705 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2710 case KVM_SET_SREGS
: {
2711 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2712 if (IS_ERR(kvm_sregs
)) {
2713 r
= PTR_ERR(kvm_sregs
);
2717 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2720 case KVM_GET_MP_STATE
: {
2721 struct kvm_mp_state mp_state
;
2723 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2727 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2732 case KVM_SET_MP_STATE
: {
2733 struct kvm_mp_state mp_state
;
2736 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2738 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2741 case KVM_TRANSLATE
: {
2742 struct kvm_translation tr
;
2745 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2747 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2751 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2756 case KVM_SET_GUEST_DEBUG
: {
2757 struct kvm_guest_debug dbg
;
2760 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2762 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2765 case KVM_SET_SIGNAL_MASK
: {
2766 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2767 struct kvm_signal_mask kvm_sigmask
;
2768 sigset_t sigset
, *p
;
2773 if (copy_from_user(&kvm_sigmask
, argp
,
2774 sizeof(kvm_sigmask
)))
2777 if (kvm_sigmask
.len
!= sizeof(sigset
))
2780 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2785 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2789 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2793 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2797 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2803 fpu
= memdup_user(argp
, sizeof(*fpu
));
2809 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2813 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2822 #ifdef CONFIG_KVM_COMPAT
2823 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2824 unsigned int ioctl
, unsigned long arg
)
2826 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2827 void __user
*argp
= compat_ptr(arg
);
2830 if (vcpu
->kvm
->mm
!= current
->mm
)
2834 case KVM_SET_SIGNAL_MASK
: {
2835 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2836 struct kvm_signal_mask kvm_sigmask
;
2841 if (copy_from_user(&kvm_sigmask
, argp
,
2842 sizeof(kvm_sigmask
)))
2845 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2848 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2850 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2852 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2856 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2864 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2865 int (*accessor
)(struct kvm_device
*dev
,
2866 struct kvm_device_attr
*attr
),
2869 struct kvm_device_attr attr
;
2874 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2877 return accessor(dev
, &attr
);
2880 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2883 struct kvm_device
*dev
= filp
->private_data
;
2885 if (dev
->kvm
->mm
!= current
->mm
)
2889 case KVM_SET_DEVICE_ATTR
:
2890 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2891 case KVM_GET_DEVICE_ATTR
:
2892 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2893 case KVM_HAS_DEVICE_ATTR
:
2894 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2896 if (dev
->ops
->ioctl
)
2897 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2903 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2905 struct kvm_device
*dev
= filp
->private_data
;
2906 struct kvm
*kvm
= dev
->kvm
;
2912 static const struct file_operations kvm_device_fops
= {
2913 .unlocked_ioctl
= kvm_device_ioctl
,
2914 #ifdef CONFIG_KVM_COMPAT
2915 .compat_ioctl
= kvm_device_ioctl
,
2917 .release
= kvm_device_release
,
2920 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2922 if (filp
->f_op
!= &kvm_device_fops
)
2925 return filp
->private_data
;
2928 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2929 #ifdef CONFIG_KVM_MPIC
2930 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2931 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2935 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2937 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2940 if (kvm_device_ops_table
[type
] != NULL
)
2943 kvm_device_ops_table
[type
] = ops
;
2947 void kvm_unregister_device_ops(u32 type
)
2949 if (kvm_device_ops_table
[type
] != NULL
)
2950 kvm_device_ops_table
[type
] = NULL
;
2953 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2954 struct kvm_create_device
*cd
)
2956 struct kvm_device_ops
*ops
= NULL
;
2957 struct kvm_device
*dev
;
2958 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2962 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2965 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
2966 ops
= kvm_device_ops_table
[type
];
2973 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2980 mutex_lock(&kvm
->lock
);
2981 ret
= ops
->create(dev
, type
);
2983 mutex_unlock(&kvm
->lock
);
2987 list_add(&dev
->vm_node
, &kvm
->devices
);
2988 mutex_unlock(&kvm
->lock
);
2994 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2997 mutex_lock(&kvm
->lock
);
2998 list_del(&dev
->vm_node
);
2999 mutex_unlock(&kvm
->lock
);
3008 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
3011 case KVM_CAP_USER_MEMORY
:
3012 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
3013 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
3014 case KVM_CAP_INTERNAL_ERROR_DATA
:
3015 #ifdef CONFIG_HAVE_KVM_MSI
3016 case KVM_CAP_SIGNAL_MSI
:
3018 #ifdef CONFIG_HAVE_KVM_IRQFD
3020 case KVM_CAP_IRQFD_RESAMPLE
:
3022 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
3023 case KVM_CAP_CHECK_EXTENSION_VM
:
3025 #ifdef CONFIG_KVM_MMIO
3026 case KVM_CAP_COALESCED_MMIO
:
3027 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
3029 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3030 case KVM_CAP_IRQ_ROUTING
:
3031 return KVM_MAX_IRQ_ROUTES
;
3033 #if KVM_ADDRESS_SPACE_NUM > 1
3034 case KVM_CAP_MULTI_ADDRESS_SPACE
:
3035 return KVM_ADDRESS_SPACE_NUM
;
3040 return kvm_vm_ioctl_check_extension(kvm
, arg
);
3043 static long kvm_vm_ioctl(struct file
*filp
,
3044 unsigned int ioctl
, unsigned long arg
)
3046 struct kvm
*kvm
= filp
->private_data
;
3047 void __user
*argp
= (void __user
*)arg
;
3050 if (kvm
->mm
!= current
->mm
)
3053 case KVM_CREATE_VCPU
:
3054 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3056 case KVM_SET_USER_MEMORY_REGION
: {
3057 struct kvm_userspace_memory_region kvm_userspace_mem
;
3060 if (copy_from_user(&kvm_userspace_mem
, argp
,
3061 sizeof(kvm_userspace_mem
)))
3064 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3067 case KVM_GET_DIRTY_LOG
: {
3068 struct kvm_dirty_log log
;
3071 if (copy_from_user(&log
, argp
, sizeof(log
)))
3073 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3076 #ifdef CONFIG_KVM_MMIO
3077 case KVM_REGISTER_COALESCED_MMIO
: {
3078 struct kvm_coalesced_mmio_zone zone
;
3081 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3083 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3086 case KVM_UNREGISTER_COALESCED_MMIO
: {
3087 struct kvm_coalesced_mmio_zone zone
;
3090 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3092 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3097 struct kvm_irqfd data
;
3100 if (copy_from_user(&data
, argp
, sizeof(data
)))
3102 r
= kvm_irqfd(kvm
, &data
);
3105 case KVM_IOEVENTFD
: {
3106 struct kvm_ioeventfd data
;
3109 if (copy_from_user(&data
, argp
, sizeof(data
)))
3111 r
= kvm_ioeventfd(kvm
, &data
);
3114 #ifdef CONFIG_HAVE_KVM_MSI
3115 case KVM_SIGNAL_MSI
: {
3119 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3121 r
= kvm_send_userspace_msi(kvm
, &msi
);
3125 #ifdef __KVM_HAVE_IRQ_LINE
3126 case KVM_IRQ_LINE_STATUS
:
3127 case KVM_IRQ_LINE
: {
3128 struct kvm_irq_level irq_event
;
3131 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3134 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3135 ioctl
== KVM_IRQ_LINE_STATUS
);
3140 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3141 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3149 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3150 case KVM_SET_GSI_ROUTING
: {
3151 struct kvm_irq_routing routing
;
3152 struct kvm_irq_routing __user
*urouting
;
3153 struct kvm_irq_routing_entry
*entries
= NULL
;
3156 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3159 if (!kvm_arch_can_set_irq_routing(kvm
))
3161 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3167 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3172 if (copy_from_user(entries
, urouting
->entries
,
3173 routing
.nr
* sizeof(*entries
)))
3174 goto out_free_irq_routing
;
3176 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3178 out_free_irq_routing
:
3182 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3183 case KVM_CREATE_DEVICE
: {
3184 struct kvm_create_device cd
;
3187 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3190 r
= kvm_ioctl_create_device(kvm
, &cd
);
3195 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3201 case KVM_CHECK_EXTENSION
:
3202 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3205 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3211 #ifdef CONFIG_KVM_COMPAT
3212 struct compat_kvm_dirty_log
{
3216 compat_uptr_t dirty_bitmap
; /* one bit per page */
3221 static long kvm_vm_compat_ioctl(struct file
*filp
,
3222 unsigned int ioctl
, unsigned long arg
)
3224 struct kvm
*kvm
= filp
->private_data
;
3227 if (kvm
->mm
!= current
->mm
)
3230 case KVM_GET_DIRTY_LOG
: {
3231 struct compat_kvm_dirty_log compat_log
;
3232 struct kvm_dirty_log log
;
3234 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3235 sizeof(compat_log
)))
3237 log
.slot
= compat_log
.slot
;
3238 log
.padding1
= compat_log
.padding1
;
3239 log
.padding2
= compat_log
.padding2
;
3240 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3242 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3246 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3252 static struct file_operations kvm_vm_fops
= {
3253 .release
= kvm_vm_release
,
3254 .unlocked_ioctl
= kvm_vm_ioctl
,
3255 #ifdef CONFIG_KVM_COMPAT
3256 .compat_ioctl
= kvm_vm_compat_ioctl
,
3258 .llseek
= noop_llseek
,
3261 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3267 kvm
= kvm_create_vm(type
);
3269 return PTR_ERR(kvm
);
3270 #ifdef CONFIG_KVM_MMIO
3271 r
= kvm_coalesced_mmio_init(kvm
);
3277 r
= get_unused_fd_flags(O_CLOEXEC
);
3282 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3286 return PTR_ERR(file
);
3290 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3291 * already set, with ->release() being kvm_vm_release(). In error
3292 * cases it will be called by the final fput(file) and will take
3293 * care of doing kvm_put_kvm(kvm).
3295 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3300 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3302 fd_install(r
, file
);
3306 static long kvm_dev_ioctl(struct file
*filp
,
3307 unsigned int ioctl
, unsigned long arg
)
3312 case KVM_GET_API_VERSION
:
3315 r
= KVM_API_VERSION
;
3318 r
= kvm_dev_ioctl_create_vm(arg
);
3320 case KVM_CHECK_EXTENSION
:
3321 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3323 case KVM_GET_VCPU_MMAP_SIZE
:
3326 r
= PAGE_SIZE
; /* struct kvm_run */
3328 r
+= PAGE_SIZE
; /* pio data page */
3330 #ifdef CONFIG_KVM_MMIO
3331 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3334 case KVM_TRACE_ENABLE
:
3335 case KVM_TRACE_PAUSE
:
3336 case KVM_TRACE_DISABLE
:
3340 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3346 static struct file_operations kvm_chardev_ops
= {
3347 .unlocked_ioctl
= kvm_dev_ioctl
,
3348 .compat_ioctl
= kvm_dev_ioctl
,
3349 .llseek
= noop_llseek
,
3352 static struct miscdevice kvm_dev
= {
3358 static void hardware_enable_nolock(void *junk
)
3360 int cpu
= raw_smp_processor_id();
3363 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3366 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3368 r
= kvm_arch_hardware_enable();
3371 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3372 atomic_inc(&hardware_enable_failed
);
3373 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3377 static int kvm_starting_cpu(unsigned int cpu
)
3379 raw_spin_lock(&kvm_count_lock
);
3380 if (kvm_usage_count
)
3381 hardware_enable_nolock(NULL
);
3382 raw_spin_unlock(&kvm_count_lock
);
3386 static void hardware_disable_nolock(void *junk
)
3388 int cpu
= raw_smp_processor_id();
3390 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3392 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3393 kvm_arch_hardware_disable();
3396 static int kvm_dying_cpu(unsigned int cpu
)
3398 raw_spin_lock(&kvm_count_lock
);
3399 if (kvm_usage_count
)
3400 hardware_disable_nolock(NULL
);
3401 raw_spin_unlock(&kvm_count_lock
);
3405 static void hardware_disable_all_nolock(void)
3407 BUG_ON(!kvm_usage_count
);
3410 if (!kvm_usage_count
)
3411 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3414 static void hardware_disable_all(void)
3416 raw_spin_lock(&kvm_count_lock
);
3417 hardware_disable_all_nolock();
3418 raw_spin_unlock(&kvm_count_lock
);
3421 static int hardware_enable_all(void)
3425 raw_spin_lock(&kvm_count_lock
);
3428 if (kvm_usage_count
== 1) {
3429 atomic_set(&hardware_enable_failed
, 0);
3430 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3432 if (atomic_read(&hardware_enable_failed
)) {
3433 hardware_disable_all_nolock();
3438 raw_spin_unlock(&kvm_count_lock
);
3443 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3447 * Some (well, at least mine) BIOSes hang on reboot if
3450 * And Intel TXT required VMX off for all cpu when system shutdown.
3452 pr_info("kvm: exiting hardware virtualization\n");
3453 kvm_rebooting
= true;
3454 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3458 static struct notifier_block kvm_reboot_notifier
= {
3459 .notifier_call
= kvm_reboot
,
3463 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3467 for (i
= 0; i
< bus
->dev_count
; i
++) {
3468 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3470 kvm_iodevice_destructor(pos
);
3475 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3476 const struct kvm_io_range
*r2
)
3478 gpa_t addr1
= r1
->addr
;
3479 gpa_t addr2
= r2
->addr
;
3484 /* If r2->len == 0, match the exact address. If r2->len != 0,
3485 * accept any overlapping write. Any order is acceptable for
3486 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3487 * we process all of them.
3500 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3502 return kvm_io_bus_cmp(p1
, p2
);
3505 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3506 gpa_t addr
, int len
)
3508 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3514 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3515 kvm_io_bus_sort_cmp
, NULL
);
3520 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3521 gpa_t addr
, int len
)
3523 struct kvm_io_range
*range
, key
;
3526 key
= (struct kvm_io_range
) {
3531 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3532 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3536 off
= range
- bus
->range
;
3538 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3544 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3545 struct kvm_io_range
*range
, const void *val
)
3549 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3553 while (idx
< bus
->dev_count
&&
3554 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3555 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3564 /* kvm_io_bus_write - called under kvm->slots_lock */
3565 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3566 int len
, const void *val
)
3568 struct kvm_io_bus
*bus
;
3569 struct kvm_io_range range
;
3572 range
= (struct kvm_io_range
) {
3577 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3580 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3581 return r
< 0 ? r
: 0;
3584 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3585 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3586 gpa_t addr
, int len
, const void *val
, long cookie
)
3588 struct kvm_io_bus
*bus
;
3589 struct kvm_io_range range
;
3591 range
= (struct kvm_io_range
) {
3596 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3600 /* First try the device referenced by cookie. */
3601 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3602 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3603 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3608 * cookie contained garbage; fall back to search and return the
3609 * correct cookie value.
3611 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3614 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3615 struct kvm_io_range
*range
, void *val
)
3619 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3623 while (idx
< bus
->dev_count
&&
3624 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3625 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3633 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3635 /* kvm_io_bus_read - called under kvm->slots_lock */
3636 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3639 struct kvm_io_bus
*bus
;
3640 struct kvm_io_range range
;
3643 range
= (struct kvm_io_range
) {
3648 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3651 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3652 return r
< 0 ? r
: 0;
3656 /* Caller must hold slots_lock. */
3657 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3658 int len
, struct kvm_io_device
*dev
)
3660 struct kvm_io_bus
*new_bus
, *bus
;
3662 bus
= kvm_get_bus(kvm
, bus_idx
);
3666 /* exclude ioeventfd which is limited by maximum fd */
3667 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3670 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3671 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3674 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3675 sizeof(struct kvm_io_range
)));
3676 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3677 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3678 synchronize_srcu_expedited(&kvm
->srcu
);
3684 /* Caller must hold slots_lock. */
3685 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3686 struct kvm_io_device
*dev
)
3689 struct kvm_io_bus
*new_bus
, *bus
;
3691 bus
= kvm_get_bus(kvm
, bus_idx
);
3695 for (i
= 0; i
< bus
->dev_count
; i
++)
3696 if (bus
->range
[i
].dev
== dev
) {
3700 if (i
== bus
->dev_count
)
3703 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3704 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3706 pr_err("kvm: failed to shrink bus, removing it completely\n");
3710 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3711 new_bus
->dev_count
--;
3712 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3713 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3716 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3717 synchronize_srcu_expedited(&kvm
->srcu
);
3722 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3725 struct kvm_io_bus
*bus
;
3726 int dev_idx
, srcu_idx
;
3727 struct kvm_io_device
*iodev
= NULL
;
3729 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3731 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3735 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3739 iodev
= bus
->range
[dev_idx
].dev
;
3742 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3746 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3748 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3749 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3752 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3755 /* The debugfs files are a reference to the kvm struct which
3756 * is still valid when kvm_destroy_vm is called.
3757 * To avoid the race between open and the removal of the debugfs
3758 * directory we test against the users count.
3760 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3763 if (simple_attr_open(inode
, file
, get
,
3764 stat_data
->mode
& S_IWUGO
? set
: NULL
,
3766 kvm_put_kvm(stat_data
->kvm
);
3773 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3775 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3778 simple_attr_release(inode
, file
);
3779 kvm_put_kvm(stat_data
->kvm
);
3784 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3786 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3788 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3793 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3795 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3800 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3805 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3807 __simple_attr_check_format("%llu\n", 0ull);
3808 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3809 vm_stat_clear_per_vm
, "%llu\n");
3812 static const struct file_operations vm_stat_get_per_vm_fops
= {
3813 .owner
= THIS_MODULE
,
3814 .open
= vm_stat_get_per_vm_open
,
3815 .release
= kvm_debugfs_release
,
3816 .read
= simple_attr_read
,
3817 .write
= simple_attr_write
,
3818 .llseek
= no_llseek
,
3821 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3824 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3825 struct kvm_vcpu
*vcpu
;
3829 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3830 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3835 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3838 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3839 struct kvm_vcpu
*vcpu
;
3844 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3845 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3850 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3852 __simple_attr_check_format("%llu\n", 0ull);
3853 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3854 vcpu_stat_clear_per_vm
, "%llu\n");
3857 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3858 .owner
= THIS_MODULE
,
3859 .open
= vcpu_stat_get_per_vm_open
,
3860 .release
= kvm_debugfs_release
,
3861 .read
= simple_attr_read
,
3862 .write
= simple_attr_write
,
3863 .llseek
= no_llseek
,
3866 static const struct file_operations
*stat_fops_per_vm
[] = {
3867 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3868 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3871 static int vm_stat_get(void *_offset
, u64
*val
)
3873 unsigned offset
= (long)_offset
;
3875 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3879 mutex_lock(&kvm_lock
);
3880 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3882 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3885 mutex_unlock(&kvm_lock
);
3889 static int vm_stat_clear(void *_offset
, u64 val
)
3891 unsigned offset
= (long)_offset
;
3893 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3898 mutex_lock(&kvm_lock
);
3899 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3901 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3903 mutex_unlock(&kvm_lock
);
3908 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3910 static int vcpu_stat_get(void *_offset
, u64
*val
)
3912 unsigned offset
= (long)_offset
;
3914 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3918 mutex_lock(&kvm_lock
);
3919 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3921 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3924 mutex_unlock(&kvm_lock
);
3928 static int vcpu_stat_clear(void *_offset
, u64 val
)
3930 unsigned offset
= (long)_offset
;
3932 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3937 mutex_lock(&kvm_lock
);
3938 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3940 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3942 mutex_unlock(&kvm_lock
);
3947 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3950 static const struct file_operations
*stat_fops
[] = {
3951 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3952 [KVM_STAT_VM
] = &vm_stat_fops
,
3955 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3957 struct kobj_uevent_env
*env
;
3958 unsigned long long created
, active
;
3960 if (!kvm_dev
.this_device
|| !kvm
)
3963 mutex_lock(&kvm_lock
);
3964 if (type
== KVM_EVENT_CREATE_VM
) {
3965 kvm_createvm_count
++;
3967 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3970 created
= kvm_createvm_count
;
3971 active
= kvm_active_vms
;
3972 mutex_unlock(&kvm_lock
);
3974 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3978 add_uevent_var(env
, "CREATED=%llu", created
);
3979 add_uevent_var(env
, "COUNT=%llu", active
);
3981 if (type
== KVM_EVENT_CREATE_VM
) {
3982 add_uevent_var(env
, "EVENT=create");
3983 kvm
->userspace_pid
= task_pid_nr(current
);
3984 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3985 add_uevent_var(env
, "EVENT=destroy");
3987 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3989 if (!IS_ERR_OR_NULL(kvm
->debugfs_dentry
)) {
3990 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3993 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3995 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3999 /* no need for checks, since we are adding at most only 5 keys */
4000 env
->envp
[env
->envp_idx
++] = NULL
;
4001 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
4005 static int kvm_init_debug(void)
4008 struct kvm_stats_debugfs_item
*p
;
4010 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
4011 if (kvm_debugfs_dir
== NULL
)
4014 kvm_debugfs_num_entries
= 0;
4015 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
4016 int mode
= p
->mode
? p
->mode
: 0644;
4017 if (!debugfs_create_file(p
->name
, mode
, kvm_debugfs_dir
,
4018 (void *)(long)p
->offset
,
4019 stat_fops
[p
->kind
]))
4026 debugfs_remove_recursive(kvm_debugfs_dir
);
4031 static int kvm_suspend(void)
4033 if (kvm_usage_count
)
4034 hardware_disable_nolock(NULL
);
4038 static void kvm_resume(void)
4040 if (kvm_usage_count
) {
4041 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
4042 hardware_enable_nolock(NULL
);
4046 static struct syscore_ops kvm_syscore_ops
= {
4047 .suspend
= kvm_suspend
,
4048 .resume
= kvm_resume
,
4052 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4054 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4057 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4059 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4061 if (vcpu
->preempted
)
4062 vcpu
->preempted
= false;
4064 kvm_arch_sched_in(vcpu
, cpu
);
4066 kvm_arch_vcpu_load(vcpu
, cpu
);
4069 static void kvm_sched_out(struct preempt_notifier
*pn
,
4070 struct task_struct
*next
)
4072 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4074 if (current
->state
== TASK_RUNNING
)
4075 vcpu
->preempted
= true;
4076 kvm_arch_vcpu_put(vcpu
);
4079 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4080 struct module
*module
)
4085 r
= kvm_arch_init(opaque
);
4090 * kvm_arch_init makes sure there's at most one caller
4091 * for architectures that support multiple implementations,
4092 * like intel and amd on x86.
4093 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4094 * conflicts in case kvm is already setup for another implementation.
4096 r
= kvm_irqfd_init();
4100 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4105 r
= kvm_arch_hardware_setup();
4109 for_each_online_cpu(cpu
) {
4110 smp_call_function_single(cpu
,
4111 kvm_arch_check_processor_compat
,
4117 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4118 kvm_starting_cpu
, kvm_dying_cpu
);
4121 register_reboot_notifier(&kvm_reboot_notifier
);
4123 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4125 vcpu_align
= __alignof__(struct kvm_vcpu
);
4126 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4127 SLAB_ACCOUNT
, NULL
);
4128 if (!kvm_vcpu_cache
) {
4133 r
= kvm_async_pf_init();
4137 kvm_chardev_ops
.owner
= module
;
4138 kvm_vm_fops
.owner
= module
;
4139 kvm_vcpu_fops
.owner
= module
;
4141 r
= misc_register(&kvm_dev
);
4143 pr_err("kvm: misc device register failed\n");
4147 register_syscore_ops(&kvm_syscore_ops
);
4149 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4150 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4152 r
= kvm_init_debug();
4154 pr_err("kvm: create debugfs files failed\n");
4158 r
= kvm_vfio_ops_init();
4164 unregister_syscore_ops(&kvm_syscore_ops
);
4165 misc_deregister(&kvm_dev
);
4167 kvm_async_pf_deinit();
4169 kmem_cache_destroy(kvm_vcpu_cache
);
4171 unregister_reboot_notifier(&kvm_reboot_notifier
);
4172 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4175 kvm_arch_hardware_unsetup();
4177 free_cpumask_var(cpus_hardware_enabled
);
4185 EXPORT_SYMBOL_GPL(kvm_init
);
4189 debugfs_remove_recursive(kvm_debugfs_dir
);
4190 misc_deregister(&kvm_dev
);
4191 kmem_cache_destroy(kvm_vcpu_cache
);
4192 kvm_async_pf_deinit();
4193 unregister_syscore_ops(&kvm_syscore_ops
);
4194 unregister_reboot_notifier(&kvm_reboot_notifier
);
4195 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4196 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4197 kvm_arch_hardware_unsetup();
4200 free_cpumask_var(cpus_hardware_enabled
);
4201 kvm_vfio_ops_exit();
4203 EXPORT_SYMBOL_GPL(kvm_exit
);
4205 struct kvm_vm_worker_thread_context
{
4207 struct task_struct
*parent
;
4208 struct completion init_done
;
4209 kvm_vm_thread_fn_t thread_fn
;
4214 static int kvm_vm_worker_thread(void *context
)
4217 * The init_context is allocated on the stack of the parent thread, so
4218 * we have to locally copy anything that is needed beyond initialization
4220 struct kvm_vm_worker_thread_context
*init_context
= context
;
4221 struct kvm
*kvm
= init_context
->kvm
;
4222 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
4223 uintptr_t data
= init_context
->data
;
4226 err
= kthread_park(current
);
4227 /* kthread_park(current) is never supposed to return an error */
4232 err
= cgroup_attach_task_all(init_context
->parent
, current
);
4234 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4239 set_user_nice(current
, task_nice(init_context
->parent
));
4242 init_context
->err
= err
;
4243 complete(&init_context
->init_done
);
4244 init_context
= NULL
;
4249 /* Wait to be woken up by the spawner before proceeding. */
4252 if (!kthread_should_stop())
4253 err
= thread_fn(kvm
, data
);
4258 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
4259 uintptr_t data
, const char *name
,
4260 struct task_struct
**thread_ptr
)
4262 struct kvm_vm_worker_thread_context init_context
= {};
4263 struct task_struct
*thread
;
4266 init_context
.kvm
= kvm
;
4267 init_context
.parent
= current
;
4268 init_context
.thread_fn
= thread_fn
;
4269 init_context
.data
= data
;
4270 init_completion(&init_context
.init_done
);
4272 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
4273 "%s-%d", name
, task_pid_nr(current
));
4275 return PTR_ERR(thread
);
4277 /* kthread_run is never supposed to return NULL */
4278 WARN_ON(thread
== NULL
);
4280 wait_for_completion(&init_context
.init_done
);
4282 if (!init_context
.err
)
4283 *thread_ptr
= thread
;
4285 return init_context
.err
;