2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
76 module_param(halt_poll_ns
, uint
, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns
);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow
= 2;
81 module_param(halt_poll_ns_grow
, uint
, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink
;
86 module_param(halt_poll_ns_shrink
, uint
, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock
);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
99 static cpumask_var_t cpus_hardware_enabled
;
100 static int kvm_usage_count
;
101 static atomic_t hardware_enable_failed
;
103 struct kmem_cache
*kvm_vcpu_cache
;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
106 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
108 struct dentry
*kvm_debugfs_dir
;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
111 static int kvm_debugfs_num_entries
;
112 static const struct file_operations
*stat_fops_per_vm
[];
114 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
125 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
);
126 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
128 __visible
bool kvm_rebooting
;
129 EXPORT_SYMBOL_GPL(kvm_rebooting
);
131 static bool largepages_enabled
= true;
133 #define KVM_EVENT_CREATE_VM 0
134 #define KVM_EVENT_DESTROY_VM 1
135 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
136 static unsigned long long kvm_createvm_count
;
137 static unsigned long long kvm_active_vms
;
139 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
142 return PageReserved(pfn_to_page(pfn
));
148 * Switches to specified vcpu, until a matching vcpu_put()
150 int vcpu_load(struct kvm_vcpu
*vcpu
)
154 if (mutex_lock_killable(&vcpu
->mutex
))
157 preempt_notifier_register(&vcpu
->preempt_notifier
);
158 kvm_arch_vcpu_load(vcpu
, cpu
);
162 EXPORT_SYMBOL_GPL(vcpu_load
);
164 void vcpu_put(struct kvm_vcpu
*vcpu
)
167 kvm_arch_vcpu_put(vcpu
);
168 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
170 mutex_unlock(&vcpu
->mutex
);
172 EXPORT_SYMBOL_GPL(vcpu_put
);
174 /* TODO: merge with kvm_arch_vcpu_should_kick */
175 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
177 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
180 * We need to wait for the VCPU to reenable interrupts and get out of
181 * READING_SHADOW_PAGE_TABLES mode.
183 if (req
& KVM_REQUEST_WAIT
)
184 return mode
!= OUTSIDE_GUEST_MODE
;
187 * Need to kick a running VCPU, but otherwise there is nothing to do.
189 return mode
== IN_GUEST_MODE
;
192 static void ack_flush(void *_completed
)
196 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
199 cpus
= cpu_online_mask
;
201 if (cpumask_empty(cpus
))
204 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
208 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
213 struct kvm_vcpu
*vcpu
;
215 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
218 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
219 kvm_make_request(req
, vcpu
);
222 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
225 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
226 kvm_request_needs_ipi(vcpu
, req
))
227 __cpumask_set_cpu(cpu
, cpus
);
229 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
231 free_cpumask_var(cpus
);
235 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
236 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
239 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
240 * kvm_make_all_cpus_request.
242 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
245 * We want to publish modifications to the page tables before reading
246 * mode. Pairs with a memory barrier in arch-specific code.
247 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
248 * and smp_mb in walk_shadow_page_lockless_begin/end.
249 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
251 * There is already an smp_mb__after_atomic() before
252 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
255 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
256 ++kvm
->stat
.remote_tlb_flush
;
257 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
259 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
262 void kvm_reload_remote_mmus(struct kvm
*kvm
)
264 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
267 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
272 mutex_init(&vcpu
->mutex
);
277 init_swait_queue_head(&vcpu
->wq
);
278 kvm_async_pf_vcpu_init(vcpu
);
281 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
283 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
288 vcpu
->run
= page_address(page
);
290 kvm_vcpu_set_in_spin_loop(vcpu
, false);
291 kvm_vcpu_set_dy_eligible(vcpu
, false);
292 vcpu
->preempted
= false;
294 r
= kvm_arch_vcpu_init(vcpu
);
300 free_page((unsigned long)vcpu
->run
);
304 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
306 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
309 * no need for rcu_read_lock as VCPU_RUN is the only place that
310 * will change the vcpu->pid pointer and on uninit all file
311 * descriptors are already gone.
313 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
314 kvm_arch_vcpu_uninit(vcpu
);
315 free_page((unsigned long)vcpu
->run
);
317 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
319 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
320 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
322 return container_of(mn
, struct kvm
, mmu_notifier
);
325 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
326 struct mm_struct
*mm
,
327 unsigned long address
,
330 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
333 idx
= srcu_read_lock(&kvm
->srcu
);
334 spin_lock(&kvm
->mmu_lock
);
335 kvm
->mmu_notifier_seq
++;
336 kvm_set_spte_hva(kvm
, address
, pte
);
337 spin_unlock(&kvm
->mmu_lock
);
338 srcu_read_unlock(&kvm
->srcu
, idx
);
341 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
342 struct mm_struct
*mm
,
346 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
347 int need_tlb_flush
= 0, idx
;
349 idx
= srcu_read_lock(&kvm
->srcu
);
350 spin_lock(&kvm
->mmu_lock
);
352 * The count increase must become visible at unlock time as no
353 * spte can be established without taking the mmu_lock and
354 * count is also read inside the mmu_lock critical section.
356 kvm
->mmu_notifier_count
++;
357 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
358 need_tlb_flush
|= kvm
->tlbs_dirty
;
359 /* we've to flush the tlb before the pages can be freed */
361 kvm_flush_remote_tlbs(kvm
);
363 spin_unlock(&kvm
->mmu_lock
);
364 srcu_read_unlock(&kvm
->srcu
, idx
);
367 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
368 struct mm_struct
*mm
,
372 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
374 spin_lock(&kvm
->mmu_lock
);
376 * This sequence increase will notify the kvm page fault that
377 * the page that is going to be mapped in the spte could have
380 kvm
->mmu_notifier_seq
++;
383 * The above sequence increase must be visible before the
384 * below count decrease, which is ensured by the smp_wmb above
385 * in conjunction with the smp_rmb in mmu_notifier_retry().
387 kvm
->mmu_notifier_count
--;
388 spin_unlock(&kvm
->mmu_lock
);
390 BUG_ON(kvm
->mmu_notifier_count
< 0);
393 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
394 struct mm_struct
*mm
,
398 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
401 idx
= srcu_read_lock(&kvm
->srcu
);
402 spin_lock(&kvm
->mmu_lock
);
404 young
= kvm_age_hva(kvm
, start
, end
);
406 kvm_flush_remote_tlbs(kvm
);
408 spin_unlock(&kvm
->mmu_lock
);
409 srcu_read_unlock(&kvm
->srcu
, idx
);
414 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
415 struct mm_struct
*mm
,
419 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
422 idx
= srcu_read_lock(&kvm
->srcu
);
423 spin_lock(&kvm
->mmu_lock
);
425 * Even though we do not flush TLB, this will still adversely
426 * affect performance on pre-Haswell Intel EPT, where there is
427 * no EPT Access Bit to clear so that we have to tear down EPT
428 * tables instead. If we find this unacceptable, we can always
429 * add a parameter to kvm_age_hva so that it effectively doesn't
430 * do anything on clear_young.
432 * Also note that currently we never issue secondary TLB flushes
433 * from clear_young, leaving this job up to the regular system
434 * cadence. If we find this inaccurate, we might come up with a
435 * more sophisticated heuristic later.
437 young
= kvm_age_hva(kvm
, start
, end
);
438 spin_unlock(&kvm
->mmu_lock
);
439 srcu_read_unlock(&kvm
->srcu
, idx
);
444 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
445 struct mm_struct
*mm
,
446 unsigned long address
)
448 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
451 idx
= srcu_read_lock(&kvm
->srcu
);
452 spin_lock(&kvm
->mmu_lock
);
453 young
= kvm_test_age_hva(kvm
, address
);
454 spin_unlock(&kvm
->mmu_lock
);
455 srcu_read_unlock(&kvm
->srcu
, idx
);
460 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
461 struct mm_struct
*mm
)
463 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
466 idx
= srcu_read_lock(&kvm
->srcu
);
467 kvm_arch_flush_shadow_all(kvm
);
468 srcu_read_unlock(&kvm
->srcu
, idx
);
471 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
472 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
473 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
474 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
475 .clear_young
= kvm_mmu_notifier_clear_young
,
476 .test_young
= kvm_mmu_notifier_test_young
,
477 .change_pte
= kvm_mmu_notifier_change_pte
,
478 .release
= kvm_mmu_notifier_release
,
481 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
483 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
484 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
487 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
489 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
494 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
496 static struct kvm_memslots
*kvm_alloc_memslots(void)
499 struct kvm_memslots
*slots
;
501 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
505 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
506 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
511 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
513 if (!memslot
->dirty_bitmap
)
516 kvfree(memslot
->dirty_bitmap
);
517 memslot
->dirty_bitmap
= NULL
;
521 * Free any memory in @free but not in @dont.
523 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
524 struct kvm_memory_slot
*dont
)
526 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
527 kvm_destroy_dirty_bitmap(free
);
529 kvm_arch_free_memslot(kvm
, free
, dont
);
534 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
536 struct kvm_memory_slot
*memslot
;
541 kvm_for_each_memslot(memslot
, slots
)
542 kvm_free_memslot(kvm
, memslot
, NULL
);
547 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
551 if (!kvm
->debugfs_dentry
)
554 debugfs_remove_recursive(kvm
->debugfs_dentry
);
556 if (kvm
->debugfs_stat_data
) {
557 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
558 kfree(kvm
->debugfs_stat_data
[i
]);
559 kfree(kvm
->debugfs_stat_data
);
563 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
565 char dir_name
[ITOA_MAX_LEN
* 2];
566 struct kvm_stat_data
*stat_data
;
567 struct kvm_stats_debugfs_item
*p
;
569 if (!debugfs_initialized())
572 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
573 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
575 if (!kvm
->debugfs_dentry
)
578 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
579 sizeof(*kvm
->debugfs_stat_data
),
581 if (!kvm
->debugfs_stat_data
)
584 for (p
= debugfs_entries
; p
->name
; p
++) {
585 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
589 stat_data
->kvm
= kvm
;
590 stat_data
->offset
= p
->offset
;
591 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
592 if (!debugfs_create_file(p
->name
, 0644,
595 stat_fops_per_vm
[p
->kind
]))
601 static struct kvm
*kvm_create_vm(unsigned long type
)
604 struct kvm
*kvm
= kvm_arch_alloc_vm();
607 return ERR_PTR(-ENOMEM
);
609 spin_lock_init(&kvm
->mmu_lock
);
611 kvm
->mm
= current
->mm
;
612 kvm_eventfd_init(kvm
);
613 mutex_init(&kvm
->lock
);
614 mutex_init(&kvm
->irq_lock
);
615 mutex_init(&kvm
->slots_lock
);
616 refcount_set(&kvm
->users_count
, 1);
617 INIT_LIST_HEAD(&kvm
->devices
);
619 r
= kvm_arch_init_vm(kvm
, type
);
621 goto out_err_no_disable
;
623 r
= hardware_enable_all();
625 goto out_err_no_disable
;
627 #ifdef CONFIG_HAVE_KVM_IRQFD
628 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
631 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
634 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
635 struct kvm_memslots
*slots
= kvm_alloc_memslots();
637 goto out_err_no_srcu
;
639 * Generations must be different for each address space.
640 * Init kvm generation close to the maximum to easily test the
641 * code of handling generation number wrap-around.
643 slots
->generation
= i
* 2 - 150;
644 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
647 if (init_srcu_struct(&kvm
->srcu
))
648 goto out_err_no_srcu
;
649 if (init_srcu_struct(&kvm
->irq_srcu
))
650 goto out_err_no_irq_srcu
;
651 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
652 rcu_assign_pointer(kvm
->buses
[i
],
653 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
658 r
= kvm_init_mmu_notifier(kvm
);
662 spin_lock(&kvm_lock
);
663 list_add(&kvm
->vm_list
, &vm_list
);
664 spin_unlock(&kvm_lock
);
666 preempt_notifier_inc();
671 cleanup_srcu_struct(&kvm
->irq_srcu
);
673 cleanup_srcu_struct(&kvm
->srcu
);
675 hardware_disable_all();
677 refcount_set(&kvm
->users_count
, 0);
678 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
679 kfree(kvm_get_bus(kvm
, i
));
680 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
681 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
682 kvm_arch_free_vm(kvm
);
687 static void kvm_destroy_devices(struct kvm
*kvm
)
689 struct kvm_device
*dev
, *tmp
;
692 * We do not need to take the kvm->lock here, because nobody else
693 * has a reference to the struct kvm at this point and therefore
694 * cannot access the devices list anyhow.
696 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
697 list_del(&dev
->vm_node
);
698 dev
->ops
->destroy(dev
);
702 static void kvm_destroy_vm(struct kvm
*kvm
)
705 struct mm_struct
*mm
= kvm
->mm
;
707 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
708 kvm_destroy_vm_debugfs(kvm
);
709 kvm_arch_sync_events(kvm
);
710 spin_lock(&kvm_lock
);
711 list_del(&kvm
->vm_list
);
712 spin_unlock(&kvm_lock
);
713 kvm_free_irq_routing(kvm
);
714 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
715 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
718 kvm_io_bus_destroy(bus
);
719 kvm
->buses
[i
] = NULL
;
721 kvm_coalesced_mmio_free(kvm
);
722 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
723 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
725 kvm_arch_flush_shadow_all(kvm
);
727 kvm_arch_destroy_vm(kvm
);
728 kvm_destroy_devices(kvm
);
729 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
730 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
731 cleanup_srcu_struct(&kvm
->irq_srcu
);
732 cleanup_srcu_struct(&kvm
->srcu
);
733 kvm_arch_free_vm(kvm
);
734 preempt_notifier_dec();
735 hardware_disable_all();
739 void kvm_get_kvm(struct kvm
*kvm
)
741 refcount_inc(&kvm
->users_count
);
743 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
745 void kvm_put_kvm(struct kvm
*kvm
)
747 if (refcount_dec_and_test(&kvm
->users_count
))
750 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
753 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
755 struct kvm
*kvm
= filp
->private_data
;
757 kvm_irqfd_release(kvm
);
764 * Allocation size is twice as large as the actual dirty bitmap size.
765 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
767 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
769 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
771 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
772 if (!memslot
->dirty_bitmap
)
779 * Insert memslot and re-sort memslots based on their GFN,
780 * so binary search could be used to lookup GFN.
781 * Sorting algorithm takes advantage of having initially
782 * sorted array and known changed memslot position.
784 static void update_memslots(struct kvm_memslots
*slots
,
785 struct kvm_memory_slot
*new)
788 int i
= slots
->id_to_index
[id
];
789 struct kvm_memory_slot
*mslots
= slots
->memslots
;
791 WARN_ON(mslots
[i
].id
!= id
);
793 WARN_ON(!mslots
[i
].npages
);
794 if (mslots
[i
].npages
)
797 if (!mslots
[i
].npages
)
801 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
802 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
803 if (!mslots
[i
+ 1].npages
)
805 mslots
[i
] = mslots
[i
+ 1];
806 slots
->id_to_index
[mslots
[i
].id
] = i
;
811 * The ">=" is needed when creating a slot with base_gfn == 0,
812 * so that it moves before all those with base_gfn == npages == 0.
814 * On the other hand, if new->npages is zero, the above loop has
815 * already left i pointing to the beginning of the empty part of
816 * mslots, and the ">=" would move the hole backwards in this
817 * case---which is wrong. So skip the loop when deleting a slot.
821 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
822 mslots
[i
] = mslots
[i
- 1];
823 slots
->id_to_index
[mslots
[i
].id
] = i
;
827 WARN_ON_ONCE(i
!= slots
->used_slots
);
830 slots
->id_to_index
[mslots
[i
].id
] = i
;
833 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
835 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
837 #ifdef __KVM_HAVE_READONLY_MEM
838 valid_flags
|= KVM_MEM_READONLY
;
841 if (mem
->flags
& ~valid_flags
)
847 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
848 int as_id
, struct kvm_memslots
*slots
)
850 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
853 * Set the low bit in the generation, which disables SPTE caching
854 * until the end of synchronize_srcu_expedited.
856 WARN_ON(old_memslots
->generation
& 1);
857 slots
->generation
= old_memslots
->generation
+ 1;
859 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
860 synchronize_srcu_expedited(&kvm
->srcu
);
863 * Increment the new memslot generation a second time. This prevents
864 * vm exits that race with memslot updates from caching a memslot
865 * generation that will (potentially) be valid forever.
867 * Generations must be unique even across address spaces. We do not need
868 * a global counter for that, instead the generation space is evenly split
869 * across address spaces. For example, with two address spaces, address
870 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
871 * use generations 2, 6, 10, 14, ...
873 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
875 kvm_arch_memslots_updated(kvm
, slots
);
881 * Allocate some memory and give it an address in the guest physical address
884 * Discontiguous memory is allowed, mostly for framebuffers.
886 * Must be called holding kvm->slots_lock for write.
888 int __kvm_set_memory_region(struct kvm
*kvm
,
889 const struct kvm_userspace_memory_region
*mem
)
893 unsigned long npages
;
894 struct kvm_memory_slot
*slot
;
895 struct kvm_memory_slot old
, new;
896 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
898 enum kvm_mr_change change
;
900 r
= check_memory_region_flags(mem
);
905 as_id
= mem
->slot
>> 16;
908 /* General sanity checks */
909 if (mem
->memory_size
& (PAGE_SIZE
- 1))
911 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
913 /* We can read the guest memory with __xxx_user() later on. */
914 if ((id
< KVM_USER_MEM_SLOTS
) &&
915 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
916 !access_ok(VERIFY_WRITE
,
917 (void __user
*)(unsigned long)mem
->userspace_addr
,
920 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
922 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
925 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
926 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
927 npages
= mem
->memory_size
>> PAGE_SHIFT
;
929 if (npages
> KVM_MEM_MAX_NR_PAGES
)
935 new.base_gfn
= base_gfn
;
937 new.flags
= mem
->flags
;
941 change
= KVM_MR_CREATE
;
942 else { /* Modify an existing slot. */
943 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
944 (npages
!= old
.npages
) ||
945 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
948 if (base_gfn
!= old
.base_gfn
)
949 change
= KVM_MR_MOVE
;
950 else if (new.flags
!= old
.flags
)
951 change
= KVM_MR_FLAGS_ONLY
;
952 else { /* Nothing to change. */
961 change
= KVM_MR_DELETE
;
966 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
967 /* Check for overlaps */
969 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
970 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
973 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
974 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
979 /* Free page dirty bitmap if unneeded */
980 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
981 new.dirty_bitmap
= NULL
;
984 if (change
== KVM_MR_CREATE
) {
985 new.userspace_addr
= mem
->userspace_addr
;
987 if (kvm_arch_create_memslot(kvm
, &new, npages
))
991 /* Allocate page dirty bitmap if needed */
992 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
993 if (kvm_create_dirty_bitmap(&new) < 0)
997 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1000 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1002 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1003 slot
= id_to_memslot(slots
, id
);
1004 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1006 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1008 /* From this point no new shadow pages pointing to a deleted,
1009 * or moved, memslot will be created.
1011 * validation of sp->gfn happens in:
1012 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1013 * - kvm_is_visible_gfn (mmu_check_roots)
1015 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1018 * We can re-use the old_memslots from above, the only difference
1019 * from the currently installed memslots is the invalid flag. This
1020 * will get overwritten by update_memslots anyway.
1022 slots
= old_memslots
;
1025 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1029 /* actual memory is freed via old in kvm_free_memslot below */
1030 if (change
== KVM_MR_DELETE
) {
1031 new.dirty_bitmap
= NULL
;
1032 memset(&new.arch
, 0, sizeof(new.arch
));
1035 update_memslots(slots
, &new);
1036 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1038 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1040 kvm_free_memslot(kvm
, &old
, &new);
1041 kvfree(old_memslots
);
1047 kvm_free_memslot(kvm
, &new, &old
);
1051 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1053 int kvm_set_memory_region(struct kvm
*kvm
,
1054 const struct kvm_userspace_memory_region
*mem
)
1058 mutex_lock(&kvm
->slots_lock
);
1059 r
= __kvm_set_memory_region(kvm
, mem
);
1060 mutex_unlock(&kvm
->slots_lock
);
1063 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1065 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1066 struct kvm_userspace_memory_region
*mem
)
1068 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1071 return kvm_set_memory_region(kvm
, mem
);
1074 int kvm_get_dirty_log(struct kvm
*kvm
,
1075 struct kvm_dirty_log
*log
, int *is_dirty
)
1077 struct kvm_memslots
*slots
;
1078 struct kvm_memory_slot
*memslot
;
1081 unsigned long any
= 0;
1083 as_id
= log
->slot
>> 16;
1084 id
= (u16
)log
->slot
;
1085 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1088 slots
= __kvm_memslots(kvm
, as_id
);
1089 memslot
= id_to_memslot(slots
, id
);
1090 if (!memslot
->dirty_bitmap
)
1093 n
= kvm_dirty_bitmap_bytes(memslot
);
1095 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1096 any
= memslot
->dirty_bitmap
[i
];
1098 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1105 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1107 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1109 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1110 * are dirty write protect them for next write.
1111 * @kvm: pointer to kvm instance
1112 * @log: slot id and address to which we copy the log
1113 * @is_dirty: flag set if any page is dirty
1115 * We need to keep it in mind that VCPU threads can write to the bitmap
1116 * concurrently. So, to avoid losing track of dirty pages we keep the
1119 * 1. Take a snapshot of the bit and clear it if needed.
1120 * 2. Write protect the corresponding page.
1121 * 3. Copy the snapshot to the userspace.
1122 * 4. Upon return caller flushes TLB's if needed.
1124 * Between 2 and 4, the guest may write to the page using the remaining TLB
1125 * entry. This is not a problem because the page is reported dirty using
1126 * the snapshot taken before and step 4 ensures that writes done after
1127 * exiting to userspace will be logged for the next call.
1130 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1131 struct kvm_dirty_log
*log
, bool *is_dirty
)
1133 struct kvm_memslots
*slots
;
1134 struct kvm_memory_slot
*memslot
;
1137 unsigned long *dirty_bitmap
;
1138 unsigned long *dirty_bitmap_buffer
;
1140 as_id
= log
->slot
>> 16;
1141 id
= (u16
)log
->slot
;
1142 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1145 slots
= __kvm_memslots(kvm
, as_id
);
1146 memslot
= id_to_memslot(slots
, id
);
1148 dirty_bitmap
= memslot
->dirty_bitmap
;
1152 n
= kvm_dirty_bitmap_bytes(memslot
);
1154 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1155 memset(dirty_bitmap_buffer
, 0, n
);
1157 spin_lock(&kvm
->mmu_lock
);
1159 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1163 if (!dirty_bitmap
[i
])
1168 mask
= xchg(&dirty_bitmap
[i
], 0);
1169 dirty_bitmap_buffer
[i
] = mask
;
1172 offset
= i
* BITS_PER_LONG
;
1173 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1178 spin_unlock(&kvm
->mmu_lock
);
1179 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1183 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1186 bool kvm_largepages_enabled(void)
1188 return largepages_enabled
;
1191 void kvm_disable_largepages(void)
1193 largepages_enabled
= false;
1195 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1197 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1199 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1201 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1203 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1205 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1208 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1210 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1212 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1213 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1218 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1220 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1222 struct vm_area_struct
*vma
;
1223 unsigned long addr
, size
;
1227 addr
= gfn_to_hva(kvm
, gfn
);
1228 if (kvm_is_error_hva(addr
))
1231 down_read(¤t
->mm
->mmap_sem
);
1232 vma
= find_vma(current
->mm
, addr
);
1236 size
= vma_kernel_pagesize(vma
);
1239 up_read(¤t
->mm
->mmap_sem
);
1244 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1246 return slot
->flags
& KVM_MEM_READONLY
;
1249 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1250 gfn_t
*nr_pages
, bool write
)
1252 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1253 return KVM_HVA_ERR_BAD
;
1255 if (memslot_is_readonly(slot
) && write
)
1256 return KVM_HVA_ERR_RO_BAD
;
1259 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1261 return __gfn_to_hva_memslot(slot
, gfn
);
1264 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1267 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1270 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1273 return gfn_to_hva_many(slot
, gfn
, NULL
);
1275 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1277 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1279 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1281 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1283 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1285 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1287 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1290 * If writable is set to false, the hva returned by this function is only
1291 * allowed to be read.
1293 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1294 gfn_t gfn
, bool *writable
)
1296 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1298 if (!kvm_is_error_hva(hva
) && writable
)
1299 *writable
= !memslot_is_readonly(slot
);
1304 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1306 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1308 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1311 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1313 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1315 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1318 static int get_user_page_nowait(unsigned long start
, int write
,
1321 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1324 flags
|= FOLL_WRITE
;
1326 return get_user_pages(start
, 1, flags
, page
, NULL
);
1329 static inline int check_user_page_hwpoison(unsigned long addr
)
1331 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1333 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1334 return rc
== -EHWPOISON
;
1338 * The atomic path to get the writable pfn which will be stored in @pfn,
1339 * true indicates success, otherwise false is returned.
1341 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1342 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1344 struct page
*page
[1];
1347 if (!(async
|| atomic
))
1351 * Fast pin a writable pfn only if it is a write fault request
1352 * or the caller allows to map a writable pfn for a read fault
1355 if (!(write_fault
|| writable
))
1358 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1360 *pfn
= page_to_pfn(page
[0]);
1371 * The slow path to get the pfn of the specified host virtual address,
1372 * 1 indicates success, -errno is returned if error is detected.
1374 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1375 bool *writable
, kvm_pfn_t
*pfn
)
1377 struct page
*page
[1];
1383 *writable
= write_fault
;
1386 down_read(¤t
->mm
->mmap_sem
);
1387 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1388 up_read(¤t
->mm
->mmap_sem
);
1390 unsigned int flags
= FOLL_HWPOISON
;
1393 flags
|= FOLL_WRITE
;
1395 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1400 /* map read fault as writable if possible */
1401 if (unlikely(!write_fault
) && writable
) {
1402 struct page
*wpage
[1];
1404 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1413 *pfn
= page_to_pfn(page
[0]);
1417 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1419 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1422 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1428 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1429 unsigned long addr
, bool *async
,
1430 bool write_fault
, kvm_pfn_t
*p_pfn
)
1435 r
= follow_pfn(vma
, addr
, &pfn
);
1438 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1439 * not call the fault handler, so do it here.
1441 bool unlocked
= false;
1442 r
= fixup_user_fault(current
, current
->mm
, addr
,
1443 (write_fault
? FAULT_FLAG_WRITE
: 0),
1450 r
= follow_pfn(vma
, addr
, &pfn
);
1458 * Get a reference here because callers of *hva_to_pfn* and
1459 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1460 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1461 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1462 * simply do nothing for reserved pfns.
1464 * Whoever called remap_pfn_range is also going to call e.g.
1465 * unmap_mapping_range before the underlying pages are freed,
1466 * causing a call to our MMU notifier.
1475 * Pin guest page in memory and return its pfn.
1476 * @addr: host virtual address which maps memory to the guest
1477 * @atomic: whether this function can sleep
1478 * @async: whether this function need to wait IO complete if the
1479 * host page is not in the memory
1480 * @write_fault: whether we should get a writable host page
1481 * @writable: whether it allows to map a writable host page for !@write_fault
1483 * The function will map a writable host page for these two cases:
1484 * 1): @write_fault = true
1485 * 2): @write_fault = false && @writable, @writable will tell the caller
1486 * whether the mapping is writable.
1488 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1489 bool write_fault
, bool *writable
)
1491 struct vm_area_struct
*vma
;
1495 /* we can do it either atomically or asynchronously, not both */
1496 BUG_ON(atomic
&& async
);
1498 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1502 return KVM_PFN_ERR_FAULT
;
1504 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1508 down_read(¤t
->mm
->mmap_sem
);
1509 if (npages
== -EHWPOISON
||
1510 (!async
&& check_user_page_hwpoison(addr
))) {
1511 pfn
= KVM_PFN_ERR_HWPOISON
;
1516 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1519 pfn
= KVM_PFN_ERR_FAULT
;
1520 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1521 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, &pfn
);
1525 pfn
= KVM_PFN_ERR_FAULT
;
1527 if (async
&& vma_is_valid(vma
, write_fault
))
1529 pfn
= KVM_PFN_ERR_FAULT
;
1532 up_read(¤t
->mm
->mmap_sem
);
1536 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1537 bool atomic
, bool *async
, bool write_fault
,
1540 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1542 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1545 return KVM_PFN_ERR_RO_FAULT
;
1548 if (kvm_is_error_hva(addr
)) {
1551 return KVM_PFN_NOSLOT
;
1554 /* Do not map writable pfn in the readonly memslot. */
1555 if (writable
&& memslot_is_readonly(slot
)) {
1560 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1563 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1565 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1568 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1569 write_fault
, writable
);
1571 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1573 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1575 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1577 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1579 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1581 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1583 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1585 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1587 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1589 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1591 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1593 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1595 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1597 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1599 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1601 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1603 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1605 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1607 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1609 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1610 struct page
**pages
, int nr_pages
)
1615 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1616 if (kvm_is_error_hva(addr
))
1619 if (entry
< nr_pages
)
1622 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1624 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1626 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1628 if (is_error_noslot_pfn(pfn
))
1629 return KVM_ERR_PTR_BAD_PAGE
;
1631 if (kvm_is_reserved_pfn(pfn
)) {
1633 return KVM_ERR_PTR_BAD_PAGE
;
1636 return pfn_to_page(pfn
);
1639 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1643 pfn
= gfn_to_pfn(kvm
, gfn
);
1645 return kvm_pfn_to_page(pfn
);
1647 EXPORT_SYMBOL_GPL(gfn_to_page
);
1649 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1653 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1655 return kvm_pfn_to_page(pfn
);
1657 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1659 void kvm_release_page_clean(struct page
*page
)
1661 WARN_ON(is_error_page(page
));
1663 kvm_release_pfn_clean(page_to_pfn(page
));
1665 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1667 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1669 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1670 put_page(pfn_to_page(pfn
));
1672 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1674 void kvm_release_page_dirty(struct page
*page
)
1676 WARN_ON(is_error_page(page
));
1678 kvm_release_pfn_dirty(page_to_pfn(page
));
1680 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1682 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1684 kvm_set_pfn_dirty(pfn
);
1685 kvm_release_pfn_clean(pfn
);
1688 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1690 if (!kvm_is_reserved_pfn(pfn
)) {
1691 struct page
*page
= pfn_to_page(pfn
);
1693 if (!PageReserved(page
))
1697 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1699 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1701 if (!kvm_is_reserved_pfn(pfn
))
1702 mark_page_accessed(pfn_to_page(pfn
));
1704 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1706 void kvm_get_pfn(kvm_pfn_t pfn
)
1708 if (!kvm_is_reserved_pfn(pfn
))
1709 get_page(pfn_to_page(pfn
));
1711 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1713 static int next_segment(unsigned long len
, int offset
)
1715 if (len
> PAGE_SIZE
- offset
)
1716 return PAGE_SIZE
- offset
;
1721 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1722 void *data
, int offset
, int len
)
1727 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1728 if (kvm_is_error_hva(addr
))
1730 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1736 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1739 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1741 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1743 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1745 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1746 int offset
, int len
)
1748 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1750 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1752 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1754 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1756 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1758 int offset
= offset_in_page(gpa
);
1761 while ((seg
= next_segment(len
, offset
)) != 0) {
1762 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1772 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1774 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1776 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1778 int offset
= offset_in_page(gpa
);
1781 while ((seg
= next_segment(len
, offset
)) != 0) {
1782 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1792 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1794 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1795 void *data
, int offset
, unsigned long len
)
1800 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1801 if (kvm_is_error_hva(addr
))
1803 pagefault_disable();
1804 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1811 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1814 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1815 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1816 int offset
= offset_in_page(gpa
);
1818 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1820 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1822 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1823 void *data
, unsigned long len
)
1825 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1826 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1827 int offset
= offset_in_page(gpa
);
1829 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1831 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1833 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1834 const void *data
, int offset
, int len
)
1839 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1840 if (kvm_is_error_hva(addr
))
1842 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1845 mark_page_dirty_in_slot(memslot
, gfn
);
1849 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1850 const void *data
, int offset
, int len
)
1852 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1854 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1856 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1858 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1859 const void *data
, int offset
, int len
)
1861 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1863 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1865 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1867 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1870 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1872 int offset
= offset_in_page(gpa
);
1875 while ((seg
= next_segment(len
, offset
)) != 0) {
1876 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1886 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1888 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1891 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1893 int offset
= offset_in_page(gpa
);
1896 while ((seg
= next_segment(len
, offset
)) != 0) {
1897 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1907 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1909 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1910 struct gfn_to_hva_cache
*ghc
,
1911 gpa_t gpa
, unsigned long len
)
1913 int offset
= offset_in_page(gpa
);
1914 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1915 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1916 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1917 gfn_t nr_pages_avail
;
1920 ghc
->generation
= slots
->generation
;
1922 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1923 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1924 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1928 * If the requested region crosses two memslots, we still
1929 * verify that the entire region is valid here.
1931 while (start_gfn
<= end_gfn
) {
1933 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1934 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1936 if (kvm_is_error_hva(ghc
->hva
))
1938 start_gfn
+= nr_pages_avail
;
1940 /* Use the slow path for cross page reads and writes. */
1941 ghc
->memslot
= NULL
;
1946 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1947 gpa_t gpa
, unsigned long len
)
1949 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1950 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1952 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1954 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1955 void *data
, int offset
, unsigned long len
)
1957 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1959 gpa_t gpa
= ghc
->gpa
+ offset
;
1961 BUG_ON(len
+ offset
> ghc
->len
);
1963 if (slots
->generation
!= ghc
->generation
)
1964 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1966 if (unlikely(!ghc
->memslot
))
1967 return kvm_write_guest(kvm
, gpa
, data
, len
);
1969 if (kvm_is_error_hva(ghc
->hva
))
1972 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
1975 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
1979 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
1981 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1982 void *data
, unsigned long len
)
1984 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
1986 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1988 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1989 void *data
, unsigned long len
)
1991 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1994 BUG_ON(len
> ghc
->len
);
1996 if (slots
->generation
!= ghc
->generation
)
1997 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1999 if (unlikely(!ghc
->memslot
))
2000 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2002 if (kvm_is_error_hva(ghc
->hva
))
2005 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2011 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2013 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2015 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2017 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2019 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2021 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2023 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2025 int offset
= offset_in_page(gpa
);
2028 while ((seg
= next_segment(len
, offset
)) != 0) {
2029 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2038 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2040 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2043 if (memslot
&& memslot
->dirty_bitmap
) {
2044 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2046 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2050 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2052 struct kvm_memory_slot
*memslot
;
2054 memslot
= gfn_to_memslot(kvm
, gfn
);
2055 mark_page_dirty_in_slot(memslot
, gfn
);
2057 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2059 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2061 struct kvm_memory_slot
*memslot
;
2063 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2064 mark_page_dirty_in_slot(memslot
, gfn
);
2066 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2068 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2070 unsigned int old
, val
, grow
;
2072 old
= val
= vcpu
->halt_poll_ns
;
2073 grow
= READ_ONCE(halt_poll_ns_grow
);
2075 if (val
== 0 && grow
)
2080 if (val
> halt_poll_ns
)
2083 vcpu
->halt_poll_ns
= val
;
2084 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2087 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2089 unsigned int old
, val
, shrink
;
2091 old
= val
= vcpu
->halt_poll_ns
;
2092 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2098 vcpu
->halt_poll_ns
= val
;
2099 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2102 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2104 if (kvm_arch_vcpu_runnable(vcpu
)) {
2105 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2108 if (kvm_cpu_has_pending_timer(vcpu
))
2110 if (signal_pending(current
))
2117 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2119 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2122 DECLARE_SWAITQUEUE(wait
);
2123 bool waited
= false;
2126 start
= cur
= ktime_get();
2127 if (vcpu
->halt_poll_ns
) {
2128 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2130 ++vcpu
->stat
.halt_attempted_poll
;
2133 * This sets KVM_REQ_UNHALT if an interrupt
2136 if (kvm_vcpu_check_block(vcpu
) < 0) {
2137 ++vcpu
->stat
.halt_successful_poll
;
2138 if (!vcpu_valid_wakeup(vcpu
))
2139 ++vcpu
->stat
.halt_poll_invalid
;
2143 } while (single_task_running() && ktime_before(cur
, stop
));
2146 kvm_arch_vcpu_blocking(vcpu
);
2149 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2151 if (kvm_vcpu_check_block(vcpu
) < 0)
2158 finish_swait(&vcpu
->wq
, &wait
);
2161 kvm_arch_vcpu_unblocking(vcpu
);
2163 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2165 if (!vcpu_valid_wakeup(vcpu
))
2166 shrink_halt_poll_ns(vcpu
);
2167 else if (halt_poll_ns
) {
2168 if (block_ns
<= vcpu
->halt_poll_ns
)
2170 /* we had a long block, shrink polling */
2171 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2172 shrink_halt_poll_ns(vcpu
);
2173 /* we had a short halt and our poll time is too small */
2174 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2175 block_ns
< halt_poll_ns
)
2176 grow_halt_poll_ns(vcpu
);
2178 vcpu
->halt_poll_ns
= 0;
2180 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2181 kvm_arch_vcpu_block_finish(vcpu
);
2183 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2185 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2187 struct swait_queue_head
*wqp
;
2189 wqp
= kvm_arch_vcpu_wq(vcpu
);
2190 if (swq_has_sleeper(wqp
)) {
2192 ++vcpu
->stat
.halt_wakeup
;
2198 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2202 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2204 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2207 int cpu
= vcpu
->cpu
;
2209 if (kvm_vcpu_wake_up(vcpu
))
2213 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2214 if (kvm_arch_vcpu_should_kick(vcpu
))
2215 smp_send_reschedule(cpu
);
2218 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2219 #endif /* !CONFIG_S390 */
2221 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2224 struct task_struct
*task
= NULL
;
2228 pid
= rcu_dereference(target
->pid
);
2230 task
= get_pid_task(pid
, PIDTYPE_PID
);
2234 ret
= yield_to(task
, 1);
2235 put_task_struct(task
);
2239 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2242 * Helper that checks whether a VCPU is eligible for directed yield.
2243 * Most eligible candidate to yield is decided by following heuristics:
2245 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2246 * (preempted lock holder), indicated by @in_spin_loop.
2247 * Set at the beiginning and cleared at the end of interception/PLE handler.
2249 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2250 * chance last time (mostly it has become eligible now since we have probably
2251 * yielded to lockholder in last iteration. This is done by toggling
2252 * @dy_eligible each time a VCPU checked for eligibility.)
2254 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2255 * to preempted lock-holder could result in wrong VCPU selection and CPU
2256 * burning. Giving priority for a potential lock-holder increases lock
2259 * Since algorithm is based on heuristics, accessing another VCPU data without
2260 * locking does not harm. It may result in trying to yield to same VCPU, fail
2261 * and continue with next VCPU and so on.
2263 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2265 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2268 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2269 vcpu
->spin_loop
.dy_eligible
;
2271 if (vcpu
->spin_loop
.in_spin_loop
)
2272 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2280 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2282 struct kvm
*kvm
= me
->kvm
;
2283 struct kvm_vcpu
*vcpu
;
2284 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2290 kvm_vcpu_set_in_spin_loop(me
, true);
2292 * We boost the priority of a VCPU that is runnable but not
2293 * currently running, because it got preempted by something
2294 * else and called schedule in __vcpu_run. Hopefully that
2295 * VCPU is holding the lock that we need and will release it.
2296 * We approximate round-robin by starting at the last boosted VCPU.
2298 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2299 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2300 if (!pass
&& i
<= last_boosted_vcpu
) {
2301 i
= last_boosted_vcpu
;
2303 } else if (pass
&& i
> last_boosted_vcpu
)
2305 if (!READ_ONCE(vcpu
->preempted
))
2309 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2311 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2313 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2316 yielded
= kvm_vcpu_yield_to(vcpu
);
2318 kvm
->last_boosted_vcpu
= i
;
2320 } else if (yielded
< 0) {
2327 kvm_vcpu_set_in_spin_loop(me
, false);
2329 /* Ensure vcpu is not eligible during next spinloop */
2330 kvm_vcpu_set_dy_eligible(me
, false);
2332 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2334 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2336 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2339 if (vmf
->pgoff
== 0)
2340 page
= virt_to_page(vcpu
->run
);
2342 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2343 page
= virt_to_page(vcpu
->arch
.pio_data
);
2345 #ifdef CONFIG_KVM_MMIO
2346 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2347 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2350 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2356 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2357 .fault
= kvm_vcpu_fault
,
2360 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2362 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2366 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2368 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2370 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2371 kvm_put_kvm(vcpu
->kvm
);
2375 static struct file_operations kvm_vcpu_fops
= {
2376 .release
= kvm_vcpu_release
,
2377 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2378 #ifdef CONFIG_KVM_COMPAT
2379 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2381 .mmap
= kvm_vcpu_mmap
,
2382 .llseek
= noop_llseek
,
2386 * Allocates an inode for the vcpu.
2388 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2390 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2393 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2395 char dir_name
[ITOA_MAX_LEN
* 2];
2398 if (!kvm_arch_has_vcpu_debugfs())
2401 if (!debugfs_initialized())
2404 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2405 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2406 vcpu
->kvm
->debugfs_dentry
);
2407 if (!vcpu
->debugfs_dentry
)
2410 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2412 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2420 * Creates some virtual cpus. Good luck creating more than one.
2422 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2425 struct kvm_vcpu
*vcpu
;
2427 if (id
>= KVM_MAX_VCPU_ID
)
2430 mutex_lock(&kvm
->lock
);
2431 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2432 mutex_unlock(&kvm
->lock
);
2436 kvm
->created_vcpus
++;
2437 mutex_unlock(&kvm
->lock
);
2439 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2442 goto vcpu_decrement
;
2445 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2447 r
= kvm_arch_vcpu_setup(vcpu
);
2451 r
= kvm_create_vcpu_debugfs(vcpu
);
2455 mutex_lock(&kvm
->lock
);
2456 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2458 goto unlock_vcpu_destroy
;
2461 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2463 /* Now it's all set up, let userspace reach it */
2465 r
= create_vcpu_fd(vcpu
);
2468 goto unlock_vcpu_destroy
;
2471 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2474 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2475 * before kvm->online_vcpu's incremented value.
2478 atomic_inc(&kvm
->online_vcpus
);
2480 mutex_unlock(&kvm
->lock
);
2481 kvm_arch_vcpu_postcreate(vcpu
);
2484 unlock_vcpu_destroy
:
2485 mutex_unlock(&kvm
->lock
);
2486 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2488 kvm_arch_vcpu_destroy(vcpu
);
2490 mutex_lock(&kvm
->lock
);
2491 kvm
->created_vcpus
--;
2492 mutex_unlock(&kvm
->lock
);
2496 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2499 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2500 vcpu
->sigset_active
= 1;
2501 vcpu
->sigset
= *sigset
;
2503 vcpu
->sigset_active
= 0;
2507 static long kvm_vcpu_ioctl(struct file
*filp
,
2508 unsigned int ioctl
, unsigned long arg
)
2510 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2511 void __user
*argp
= (void __user
*)arg
;
2513 struct kvm_fpu
*fpu
= NULL
;
2514 struct kvm_sregs
*kvm_sregs
= NULL
;
2516 if (vcpu
->kvm
->mm
!= current
->mm
)
2519 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2522 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2524 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2525 * so vcpu_load() would break it.
2527 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2528 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2532 r
= vcpu_load(vcpu
);
2541 oldpid
= rcu_access_pointer(vcpu
->pid
);
2542 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2543 /* The thread running this VCPU changed. */
2544 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2546 rcu_assign_pointer(vcpu
->pid
, newpid
);
2551 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2552 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2555 case KVM_GET_REGS
: {
2556 struct kvm_regs
*kvm_regs
;
2559 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2562 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2566 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2573 case KVM_SET_REGS
: {
2574 struct kvm_regs
*kvm_regs
;
2577 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2578 if (IS_ERR(kvm_regs
)) {
2579 r
= PTR_ERR(kvm_regs
);
2582 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2586 case KVM_GET_SREGS
: {
2587 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2591 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2595 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2600 case KVM_SET_SREGS
: {
2601 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2602 if (IS_ERR(kvm_sregs
)) {
2603 r
= PTR_ERR(kvm_sregs
);
2607 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2610 case KVM_GET_MP_STATE
: {
2611 struct kvm_mp_state mp_state
;
2613 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2617 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2622 case KVM_SET_MP_STATE
: {
2623 struct kvm_mp_state mp_state
;
2626 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2628 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2631 case KVM_TRANSLATE
: {
2632 struct kvm_translation tr
;
2635 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2637 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2641 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2646 case KVM_SET_GUEST_DEBUG
: {
2647 struct kvm_guest_debug dbg
;
2650 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2652 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2655 case KVM_SET_SIGNAL_MASK
: {
2656 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2657 struct kvm_signal_mask kvm_sigmask
;
2658 sigset_t sigset
, *p
;
2663 if (copy_from_user(&kvm_sigmask
, argp
,
2664 sizeof(kvm_sigmask
)))
2667 if (kvm_sigmask
.len
!= sizeof(sigset
))
2670 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2675 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2679 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2683 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2687 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2693 fpu
= memdup_user(argp
, sizeof(*fpu
));
2699 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2703 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2712 #ifdef CONFIG_KVM_COMPAT
2713 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2714 unsigned int ioctl
, unsigned long arg
)
2716 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2717 void __user
*argp
= compat_ptr(arg
);
2720 if (vcpu
->kvm
->mm
!= current
->mm
)
2724 case KVM_SET_SIGNAL_MASK
: {
2725 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2726 struct kvm_signal_mask kvm_sigmask
;
2727 compat_sigset_t csigset
;
2732 if (copy_from_user(&kvm_sigmask
, argp
,
2733 sizeof(kvm_sigmask
)))
2736 if (kvm_sigmask
.len
!= sizeof(csigset
))
2739 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2742 sigset_from_compat(&sigset
, &csigset
);
2743 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2745 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2749 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2757 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2758 int (*accessor
)(struct kvm_device
*dev
,
2759 struct kvm_device_attr
*attr
),
2762 struct kvm_device_attr attr
;
2767 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2770 return accessor(dev
, &attr
);
2773 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2776 struct kvm_device
*dev
= filp
->private_data
;
2779 case KVM_SET_DEVICE_ATTR
:
2780 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2781 case KVM_GET_DEVICE_ATTR
:
2782 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2783 case KVM_HAS_DEVICE_ATTR
:
2784 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2786 if (dev
->ops
->ioctl
)
2787 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2793 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2795 struct kvm_device
*dev
= filp
->private_data
;
2796 struct kvm
*kvm
= dev
->kvm
;
2802 static const struct file_operations kvm_device_fops
= {
2803 .unlocked_ioctl
= kvm_device_ioctl
,
2804 #ifdef CONFIG_KVM_COMPAT
2805 .compat_ioctl
= kvm_device_ioctl
,
2807 .release
= kvm_device_release
,
2810 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2812 if (filp
->f_op
!= &kvm_device_fops
)
2815 return filp
->private_data
;
2818 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2819 #ifdef CONFIG_KVM_MPIC
2820 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2821 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2825 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2827 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2830 if (kvm_device_ops_table
[type
] != NULL
)
2833 kvm_device_ops_table
[type
] = ops
;
2837 void kvm_unregister_device_ops(u32 type
)
2839 if (kvm_device_ops_table
[type
] != NULL
)
2840 kvm_device_ops_table
[type
] = NULL
;
2843 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2844 struct kvm_create_device
*cd
)
2846 struct kvm_device_ops
*ops
= NULL
;
2847 struct kvm_device
*dev
;
2848 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2851 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2854 ops
= kvm_device_ops_table
[cd
->type
];
2861 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2868 mutex_lock(&kvm
->lock
);
2869 ret
= ops
->create(dev
, cd
->type
);
2871 mutex_unlock(&kvm
->lock
);
2875 list_add(&dev
->vm_node
, &kvm
->devices
);
2876 mutex_unlock(&kvm
->lock
);
2881 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2883 mutex_lock(&kvm
->lock
);
2884 list_del(&dev
->vm_node
);
2885 mutex_unlock(&kvm
->lock
);
2895 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2898 case KVM_CAP_USER_MEMORY
:
2899 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2900 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2901 case KVM_CAP_INTERNAL_ERROR_DATA
:
2902 #ifdef CONFIG_HAVE_KVM_MSI
2903 case KVM_CAP_SIGNAL_MSI
:
2905 #ifdef CONFIG_HAVE_KVM_IRQFD
2907 case KVM_CAP_IRQFD_RESAMPLE
:
2909 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2910 case KVM_CAP_CHECK_EXTENSION_VM
:
2912 #ifdef CONFIG_KVM_MMIO
2913 case KVM_CAP_COALESCED_MMIO
:
2914 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2916 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2917 case KVM_CAP_IRQ_ROUTING
:
2918 return KVM_MAX_IRQ_ROUTES
;
2920 #if KVM_ADDRESS_SPACE_NUM > 1
2921 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2922 return KVM_ADDRESS_SPACE_NUM
;
2924 case KVM_CAP_MAX_VCPU_ID
:
2925 return KVM_MAX_VCPU_ID
;
2929 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2932 static long kvm_vm_ioctl(struct file
*filp
,
2933 unsigned int ioctl
, unsigned long arg
)
2935 struct kvm
*kvm
= filp
->private_data
;
2936 void __user
*argp
= (void __user
*)arg
;
2939 if (kvm
->mm
!= current
->mm
)
2942 case KVM_CREATE_VCPU
:
2943 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2945 case KVM_SET_USER_MEMORY_REGION
: {
2946 struct kvm_userspace_memory_region kvm_userspace_mem
;
2949 if (copy_from_user(&kvm_userspace_mem
, argp
,
2950 sizeof(kvm_userspace_mem
)))
2953 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2956 case KVM_GET_DIRTY_LOG
: {
2957 struct kvm_dirty_log log
;
2960 if (copy_from_user(&log
, argp
, sizeof(log
)))
2962 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2965 #ifdef CONFIG_KVM_MMIO
2966 case KVM_REGISTER_COALESCED_MMIO
: {
2967 struct kvm_coalesced_mmio_zone zone
;
2970 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2972 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2975 case KVM_UNREGISTER_COALESCED_MMIO
: {
2976 struct kvm_coalesced_mmio_zone zone
;
2979 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2981 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2986 struct kvm_irqfd data
;
2989 if (copy_from_user(&data
, argp
, sizeof(data
)))
2991 r
= kvm_irqfd(kvm
, &data
);
2994 case KVM_IOEVENTFD
: {
2995 struct kvm_ioeventfd data
;
2998 if (copy_from_user(&data
, argp
, sizeof(data
)))
3000 r
= kvm_ioeventfd(kvm
, &data
);
3003 #ifdef CONFIG_HAVE_KVM_MSI
3004 case KVM_SIGNAL_MSI
: {
3008 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3010 r
= kvm_send_userspace_msi(kvm
, &msi
);
3014 #ifdef __KVM_HAVE_IRQ_LINE
3015 case KVM_IRQ_LINE_STATUS
:
3016 case KVM_IRQ_LINE
: {
3017 struct kvm_irq_level irq_event
;
3020 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3023 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3024 ioctl
== KVM_IRQ_LINE_STATUS
);
3029 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3030 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3038 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3039 case KVM_SET_GSI_ROUTING
: {
3040 struct kvm_irq_routing routing
;
3041 struct kvm_irq_routing __user
*urouting
;
3042 struct kvm_irq_routing_entry
*entries
= NULL
;
3045 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3048 if (!kvm_arch_can_set_irq_routing(kvm
))
3050 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3056 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3061 if (copy_from_user(entries
, urouting
->entries
,
3062 routing
.nr
* sizeof(*entries
)))
3063 goto out_free_irq_routing
;
3065 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3067 out_free_irq_routing
:
3071 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3072 case KVM_CREATE_DEVICE
: {
3073 struct kvm_create_device cd
;
3076 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3079 r
= kvm_ioctl_create_device(kvm
, &cd
);
3084 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3090 case KVM_CHECK_EXTENSION
:
3091 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3094 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3100 #ifdef CONFIG_KVM_COMPAT
3101 struct compat_kvm_dirty_log
{
3105 compat_uptr_t dirty_bitmap
; /* one bit per page */
3110 static long kvm_vm_compat_ioctl(struct file
*filp
,
3111 unsigned int ioctl
, unsigned long arg
)
3113 struct kvm
*kvm
= filp
->private_data
;
3116 if (kvm
->mm
!= current
->mm
)
3119 case KVM_GET_DIRTY_LOG
: {
3120 struct compat_kvm_dirty_log compat_log
;
3121 struct kvm_dirty_log log
;
3123 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3124 sizeof(compat_log
)))
3126 log
.slot
= compat_log
.slot
;
3127 log
.padding1
= compat_log
.padding1
;
3128 log
.padding2
= compat_log
.padding2
;
3129 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3131 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3135 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3141 static struct file_operations kvm_vm_fops
= {
3142 .release
= kvm_vm_release
,
3143 .unlocked_ioctl
= kvm_vm_ioctl
,
3144 #ifdef CONFIG_KVM_COMPAT
3145 .compat_ioctl
= kvm_vm_compat_ioctl
,
3147 .llseek
= noop_llseek
,
3150 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3156 kvm
= kvm_create_vm(type
);
3158 return PTR_ERR(kvm
);
3159 #ifdef CONFIG_KVM_MMIO
3160 r
= kvm_coalesced_mmio_init(kvm
);
3166 r
= get_unused_fd_flags(O_CLOEXEC
);
3171 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3175 return PTR_ERR(file
);
3179 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3180 * already set, with ->release() being kvm_vm_release(). In error
3181 * cases it will be called by the final fput(file) and will take
3182 * care of doing kvm_put_kvm(kvm).
3184 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3189 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3191 fd_install(r
, file
);
3195 static long kvm_dev_ioctl(struct file
*filp
,
3196 unsigned int ioctl
, unsigned long arg
)
3201 case KVM_GET_API_VERSION
:
3204 r
= KVM_API_VERSION
;
3207 r
= kvm_dev_ioctl_create_vm(arg
);
3209 case KVM_CHECK_EXTENSION
:
3210 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3212 case KVM_GET_VCPU_MMAP_SIZE
:
3215 r
= PAGE_SIZE
; /* struct kvm_run */
3217 r
+= PAGE_SIZE
; /* pio data page */
3219 #ifdef CONFIG_KVM_MMIO
3220 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3223 case KVM_TRACE_ENABLE
:
3224 case KVM_TRACE_PAUSE
:
3225 case KVM_TRACE_DISABLE
:
3229 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3235 static struct file_operations kvm_chardev_ops
= {
3236 .unlocked_ioctl
= kvm_dev_ioctl
,
3237 .compat_ioctl
= kvm_dev_ioctl
,
3238 .llseek
= noop_llseek
,
3241 static struct miscdevice kvm_dev
= {
3247 static void hardware_enable_nolock(void *junk
)
3249 int cpu
= raw_smp_processor_id();
3252 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3255 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3257 r
= kvm_arch_hardware_enable();
3260 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3261 atomic_inc(&hardware_enable_failed
);
3262 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3266 static int kvm_starting_cpu(unsigned int cpu
)
3268 raw_spin_lock(&kvm_count_lock
);
3269 if (kvm_usage_count
)
3270 hardware_enable_nolock(NULL
);
3271 raw_spin_unlock(&kvm_count_lock
);
3275 static void hardware_disable_nolock(void *junk
)
3277 int cpu
= raw_smp_processor_id();
3279 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3281 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3282 kvm_arch_hardware_disable();
3285 static int kvm_dying_cpu(unsigned int cpu
)
3287 raw_spin_lock(&kvm_count_lock
);
3288 if (kvm_usage_count
)
3289 hardware_disable_nolock(NULL
);
3290 raw_spin_unlock(&kvm_count_lock
);
3294 static void hardware_disable_all_nolock(void)
3296 BUG_ON(!kvm_usage_count
);
3299 if (!kvm_usage_count
)
3300 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3303 static void hardware_disable_all(void)
3305 raw_spin_lock(&kvm_count_lock
);
3306 hardware_disable_all_nolock();
3307 raw_spin_unlock(&kvm_count_lock
);
3310 static int hardware_enable_all(void)
3314 raw_spin_lock(&kvm_count_lock
);
3317 if (kvm_usage_count
== 1) {
3318 atomic_set(&hardware_enable_failed
, 0);
3319 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3321 if (atomic_read(&hardware_enable_failed
)) {
3322 hardware_disable_all_nolock();
3327 raw_spin_unlock(&kvm_count_lock
);
3332 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3336 * Some (well, at least mine) BIOSes hang on reboot if
3339 * And Intel TXT required VMX off for all cpu when system shutdown.
3341 pr_info("kvm: exiting hardware virtualization\n");
3342 kvm_rebooting
= true;
3343 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3347 static struct notifier_block kvm_reboot_notifier
= {
3348 .notifier_call
= kvm_reboot
,
3352 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3356 for (i
= 0; i
< bus
->dev_count
; i
++) {
3357 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3359 kvm_iodevice_destructor(pos
);
3364 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3365 const struct kvm_io_range
*r2
)
3367 gpa_t addr1
= r1
->addr
;
3368 gpa_t addr2
= r2
->addr
;
3373 /* If r2->len == 0, match the exact address. If r2->len != 0,
3374 * accept any overlapping write. Any order is acceptable for
3375 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3376 * we process all of them.
3389 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3391 return kvm_io_bus_cmp(p1
, p2
);
3394 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3395 gpa_t addr
, int len
)
3397 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3403 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3404 kvm_io_bus_sort_cmp
, NULL
);
3409 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3410 gpa_t addr
, int len
)
3412 struct kvm_io_range
*range
, key
;
3415 key
= (struct kvm_io_range
) {
3420 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3421 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3425 off
= range
- bus
->range
;
3427 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3433 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3434 struct kvm_io_range
*range
, const void *val
)
3438 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3442 while (idx
< bus
->dev_count
&&
3443 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3444 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3453 /* kvm_io_bus_write - called under kvm->slots_lock */
3454 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3455 int len
, const void *val
)
3457 struct kvm_io_bus
*bus
;
3458 struct kvm_io_range range
;
3461 range
= (struct kvm_io_range
) {
3466 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3469 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3470 return r
< 0 ? r
: 0;
3473 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3474 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3475 gpa_t addr
, int len
, const void *val
, long cookie
)
3477 struct kvm_io_bus
*bus
;
3478 struct kvm_io_range range
;
3480 range
= (struct kvm_io_range
) {
3485 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3489 /* First try the device referenced by cookie. */
3490 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3491 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3492 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3497 * cookie contained garbage; fall back to search and return the
3498 * correct cookie value.
3500 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3503 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3504 struct kvm_io_range
*range
, void *val
)
3508 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3512 while (idx
< bus
->dev_count
&&
3513 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3514 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3522 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3524 /* kvm_io_bus_read - called under kvm->slots_lock */
3525 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3528 struct kvm_io_bus
*bus
;
3529 struct kvm_io_range range
;
3532 range
= (struct kvm_io_range
) {
3537 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3540 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3541 return r
< 0 ? r
: 0;
3545 /* Caller must hold slots_lock. */
3546 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3547 int len
, struct kvm_io_device
*dev
)
3549 struct kvm_io_bus
*new_bus
, *bus
;
3551 bus
= kvm_get_bus(kvm
, bus_idx
);
3555 /* exclude ioeventfd which is limited by maximum fd */
3556 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3559 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3560 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3563 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3564 sizeof(struct kvm_io_range
)));
3565 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3566 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3567 synchronize_srcu_expedited(&kvm
->srcu
);
3573 /* Caller must hold slots_lock. */
3574 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3575 struct kvm_io_device
*dev
)
3578 struct kvm_io_bus
*new_bus
, *bus
;
3580 bus
= kvm_get_bus(kvm
, bus_idx
);
3584 for (i
= 0; i
< bus
->dev_count
; i
++)
3585 if (bus
->range
[i
].dev
== dev
) {
3589 if (i
== bus
->dev_count
)
3592 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3593 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3595 pr_err("kvm: failed to shrink bus, removing it completely\n");
3599 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3600 new_bus
->dev_count
--;
3601 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3602 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3605 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3606 synchronize_srcu_expedited(&kvm
->srcu
);
3611 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3614 struct kvm_io_bus
*bus
;
3615 int dev_idx
, srcu_idx
;
3616 struct kvm_io_device
*iodev
= NULL
;
3618 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3620 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3624 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3628 iodev
= bus
->range
[dev_idx
].dev
;
3631 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3635 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3637 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3638 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3641 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3644 /* The debugfs files are a reference to the kvm struct which
3645 * is still valid when kvm_destroy_vm is called.
3646 * To avoid the race between open and the removal of the debugfs
3647 * directory we test against the users count.
3649 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3652 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3653 kvm_put_kvm(stat_data
->kvm
);
3660 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3662 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3665 simple_attr_release(inode
, file
);
3666 kvm_put_kvm(stat_data
->kvm
);
3671 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3673 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3675 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3680 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3682 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3687 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3692 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3694 __simple_attr_check_format("%llu\n", 0ull);
3695 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3696 vm_stat_clear_per_vm
, "%llu\n");
3699 static const struct file_operations vm_stat_get_per_vm_fops
= {
3700 .owner
= THIS_MODULE
,
3701 .open
= vm_stat_get_per_vm_open
,
3702 .release
= kvm_debugfs_release
,
3703 .read
= simple_attr_read
,
3704 .write
= simple_attr_write
,
3705 .llseek
= no_llseek
,
3708 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3711 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3712 struct kvm_vcpu
*vcpu
;
3716 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3717 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3722 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3725 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3726 struct kvm_vcpu
*vcpu
;
3731 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3732 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3737 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3739 __simple_attr_check_format("%llu\n", 0ull);
3740 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3741 vcpu_stat_clear_per_vm
, "%llu\n");
3744 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3745 .owner
= THIS_MODULE
,
3746 .open
= vcpu_stat_get_per_vm_open
,
3747 .release
= kvm_debugfs_release
,
3748 .read
= simple_attr_read
,
3749 .write
= simple_attr_write
,
3750 .llseek
= no_llseek
,
3753 static const struct file_operations
*stat_fops_per_vm
[] = {
3754 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3755 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3758 static int vm_stat_get(void *_offset
, u64
*val
)
3760 unsigned offset
= (long)_offset
;
3762 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3766 spin_lock(&kvm_lock
);
3767 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3769 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3772 spin_unlock(&kvm_lock
);
3776 static int vm_stat_clear(void *_offset
, u64 val
)
3778 unsigned offset
= (long)_offset
;
3780 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3785 spin_lock(&kvm_lock
);
3786 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3788 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3790 spin_unlock(&kvm_lock
);
3795 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3797 static int vcpu_stat_get(void *_offset
, u64
*val
)
3799 unsigned offset
= (long)_offset
;
3801 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3805 spin_lock(&kvm_lock
);
3806 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3808 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3811 spin_unlock(&kvm_lock
);
3815 static int vcpu_stat_clear(void *_offset
, u64 val
)
3817 unsigned offset
= (long)_offset
;
3819 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3824 spin_lock(&kvm_lock
);
3825 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3827 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3829 spin_unlock(&kvm_lock
);
3834 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3837 static const struct file_operations
*stat_fops
[] = {
3838 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3839 [KVM_STAT_VM
] = &vm_stat_fops
,
3842 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3844 struct kobj_uevent_env
*env
;
3845 unsigned long long created
, active
;
3847 if (!kvm_dev
.this_device
|| !kvm
)
3850 spin_lock(&kvm_lock
);
3851 if (type
== KVM_EVENT_CREATE_VM
) {
3852 kvm_createvm_count
++;
3854 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3857 created
= kvm_createvm_count
;
3858 active
= kvm_active_vms
;
3859 spin_unlock(&kvm_lock
);
3861 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3865 add_uevent_var(env
, "CREATED=%llu", created
);
3866 add_uevent_var(env
, "COUNT=%llu", active
);
3868 if (type
== KVM_EVENT_CREATE_VM
) {
3869 add_uevent_var(env
, "EVENT=create");
3870 kvm
->userspace_pid
= task_pid_nr(current
);
3871 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3872 add_uevent_var(env
, "EVENT=destroy");
3874 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3876 if (kvm
->debugfs_dentry
) {
3877 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3880 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3882 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3886 /* no need for checks, since we are adding at most only 5 keys */
3887 env
->envp
[env
->envp_idx
++] = NULL
;
3888 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3892 static int kvm_init_debug(void)
3895 struct kvm_stats_debugfs_item
*p
;
3897 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3898 if (kvm_debugfs_dir
== NULL
)
3901 kvm_debugfs_num_entries
= 0;
3902 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3903 if (!debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3904 (void *)(long)p
->offset
,
3905 stat_fops
[p
->kind
]))
3912 debugfs_remove_recursive(kvm_debugfs_dir
);
3917 static int kvm_suspend(void)
3919 if (kvm_usage_count
)
3920 hardware_disable_nolock(NULL
);
3924 static void kvm_resume(void)
3926 if (kvm_usage_count
) {
3927 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3928 hardware_enable_nolock(NULL
);
3932 static struct syscore_ops kvm_syscore_ops
= {
3933 .suspend
= kvm_suspend
,
3934 .resume
= kvm_resume
,
3938 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3940 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3943 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3945 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3947 if (vcpu
->preempted
)
3948 vcpu
->preempted
= false;
3950 kvm_arch_sched_in(vcpu
, cpu
);
3952 kvm_arch_vcpu_load(vcpu
, cpu
);
3955 static void kvm_sched_out(struct preempt_notifier
*pn
,
3956 struct task_struct
*next
)
3958 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3960 if (current
->state
== TASK_RUNNING
)
3961 vcpu
->preempted
= true;
3962 kvm_arch_vcpu_put(vcpu
);
3965 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3966 struct module
*module
)
3971 r
= kvm_arch_init(opaque
);
3976 * kvm_arch_init makes sure there's at most one caller
3977 * for architectures that support multiple implementations,
3978 * like intel and amd on x86.
3979 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3980 * conflicts in case kvm is already setup for another implementation.
3982 r
= kvm_irqfd_init();
3986 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3991 r
= kvm_arch_hardware_setup();
3995 for_each_online_cpu(cpu
) {
3996 smp_call_function_single(cpu
,
3997 kvm_arch_check_processor_compat
,
4003 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4004 kvm_starting_cpu
, kvm_dying_cpu
);
4007 register_reboot_notifier(&kvm_reboot_notifier
);
4009 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4011 vcpu_align
= __alignof__(struct kvm_vcpu
);
4012 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4014 if (!kvm_vcpu_cache
) {
4019 r
= kvm_async_pf_init();
4023 kvm_chardev_ops
.owner
= module
;
4024 kvm_vm_fops
.owner
= module
;
4025 kvm_vcpu_fops
.owner
= module
;
4027 r
= misc_register(&kvm_dev
);
4029 pr_err("kvm: misc device register failed\n");
4033 register_syscore_ops(&kvm_syscore_ops
);
4035 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4036 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4038 r
= kvm_init_debug();
4040 pr_err("kvm: create debugfs files failed\n");
4044 r
= kvm_vfio_ops_init();
4050 unregister_syscore_ops(&kvm_syscore_ops
);
4051 misc_deregister(&kvm_dev
);
4053 kvm_async_pf_deinit();
4055 kmem_cache_destroy(kvm_vcpu_cache
);
4057 unregister_reboot_notifier(&kvm_reboot_notifier
);
4058 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4061 kvm_arch_hardware_unsetup();
4063 free_cpumask_var(cpus_hardware_enabled
);
4071 EXPORT_SYMBOL_GPL(kvm_init
);
4075 debugfs_remove_recursive(kvm_debugfs_dir
);
4076 misc_deregister(&kvm_dev
);
4077 kmem_cache_destroy(kvm_vcpu_cache
);
4078 kvm_async_pf_deinit();
4079 unregister_syscore_ops(&kvm_syscore_ops
);
4080 unregister_reboot_notifier(&kvm_reboot_notifier
);
4081 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4082 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4083 kvm_arch_hardware_unsetup();
4086 free_cpumask_var(cpus_hardware_enabled
);
4087 kvm_vfio_ops_exit();
4089 EXPORT_SYMBOL_GPL(kvm_exit
);