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 mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
127 __visible
bool kvm_rebooting
;
128 EXPORT_SYMBOL_GPL(kvm_rebooting
);
130 static bool largepages_enabled
= true;
132 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
135 static unsigned long long kvm_createvm_count
;
136 static unsigned long long kvm_active_vms
;
138 __weak
void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
139 unsigned long start
, unsigned long end
)
143 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
146 return PageReserved(pfn_to_page(pfn
));
152 * Switches to specified vcpu, until a matching vcpu_put()
154 int vcpu_load(struct kvm_vcpu
*vcpu
)
158 if (mutex_lock_killable(&vcpu
->mutex
))
161 preempt_notifier_register(&vcpu
->preempt_notifier
);
162 kvm_arch_vcpu_load(vcpu
, cpu
);
166 EXPORT_SYMBOL_GPL(vcpu_load
);
168 void vcpu_put(struct kvm_vcpu
*vcpu
)
171 kvm_arch_vcpu_put(vcpu
);
172 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
174 mutex_unlock(&vcpu
->mutex
);
176 EXPORT_SYMBOL_GPL(vcpu_put
);
178 /* TODO: merge with kvm_arch_vcpu_should_kick */
179 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
181 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
184 * We need to wait for the VCPU to reenable interrupts and get out of
185 * READING_SHADOW_PAGE_TABLES mode.
187 if (req
& KVM_REQUEST_WAIT
)
188 return mode
!= OUTSIDE_GUEST_MODE
;
191 * Need to kick a running VCPU, but otherwise there is nothing to do.
193 return mode
== IN_GUEST_MODE
;
196 static void ack_flush(void *_completed
)
200 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
203 cpus
= cpu_online_mask
;
205 if (cpumask_empty(cpus
))
208 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
212 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
217 struct kvm_vcpu
*vcpu
;
219 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
222 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
223 kvm_make_request(req
, vcpu
);
226 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
229 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
230 kvm_request_needs_ipi(vcpu
, req
))
231 __cpumask_set_cpu(cpu
, cpus
);
233 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
235 free_cpumask_var(cpus
);
239 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
240 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
243 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
244 * kvm_make_all_cpus_request.
246 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
249 * We want to publish modifications to the page tables before reading
250 * mode. Pairs with a memory barrier in arch-specific code.
251 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
252 * and smp_mb in walk_shadow_page_lockless_begin/end.
253 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
255 * There is already an smp_mb__after_atomic() before
256 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
259 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
260 ++kvm
->stat
.remote_tlb_flush
;
261 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
263 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
266 void kvm_reload_remote_mmus(struct kvm
*kvm
)
268 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
271 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
276 mutex_init(&vcpu
->mutex
);
281 init_swait_queue_head(&vcpu
->wq
);
282 kvm_async_pf_vcpu_init(vcpu
);
285 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
287 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
292 vcpu
->run
= page_address(page
);
294 kvm_vcpu_set_in_spin_loop(vcpu
, false);
295 kvm_vcpu_set_dy_eligible(vcpu
, false);
296 vcpu
->preempted
= false;
298 r
= kvm_arch_vcpu_init(vcpu
);
304 free_page((unsigned long)vcpu
->run
);
308 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
310 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
313 * no need for rcu_read_lock as VCPU_RUN is the only place that
314 * will change the vcpu->pid pointer and on uninit all file
315 * descriptors are already gone.
317 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
318 kvm_arch_vcpu_uninit(vcpu
);
319 free_page((unsigned long)vcpu
->run
);
321 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
323 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
324 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
326 return container_of(mn
, struct kvm
, mmu_notifier
);
329 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
330 struct mm_struct
*mm
,
331 unsigned long address
,
334 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
337 idx
= srcu_read_lock(&kvm
->srcu
);
338 spin_lock(&kvm
->mmu_lock
);
339 kvm
->mmu_notifier_seq
++;
340 kvm_set_spte_hva(kvm
, address
, pte
);
341 spin_unlock(&kvm
->mmu_lock
);
342 srcu_read_unlock(&kvm
->srcu
, idx
);
345 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
346 struct mm_struct
*mm
,
350 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
351 int need_tlb_flush
= 0, idx
;
353 idx
= srcu_read_lock(&kvm
->srcu
);
354 spin_lock(&kvm
->mmu_lock
);
356 * The count increase must become visible at unlock time as no
357 * spte can be established without taking the mmu_lock and
358 * count is also read inside the mmu_lock critical section.
360 kvm
->mmu_notifier_count
++;
361 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
362 need_tlb_flush
|= kvm
->tlbs_dirty
;
363 /* we've to flush the tlb before the pages can be freed */
365 kvm_flush_remote_tlbs(kvm
);
367 spin_unlock(&kvm
->mmu_lock
);
369 kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
);
371 srcu_read_unlock(&kvm
->srcu
, idx
);
374 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
375 struct mm_struct
*mm
,
379 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
381 spin_lock(&kvm
->mmu_lock
);
383 * This sequence increase will notify the kvm page fault that
384 * the page that is going to be mapped in the spte could have
387 kvm
->mmu_notifier_seq
++;
390 * The above sequence increase must be visible before the
391 * below count decrease, which is ensured by the smp_wmb above
392 * in conjunction with the smp_rmb in mmu_notifier_retry().
394 kvm
->mmu_notifier_count
--;
395 spin_unlock(&kvm
->mmu_lock
);
397 BUG_ON(kvm
->mmu_notifier_count
< 0);
400 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
401 struct mm_struct
*mm
,
405 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
408 idx
= srcu_read_lock(&kvm
->srcu
);
409 spin_lock(&kvm
->mmu_lock
);
411 young
= kvm_age_hva(kvm
, start
, end
);
413 kvm_flush_remote_tlbs(kvm
);
415 spin_unlock(&kvm
->mmu_lock
);
416 srcu_read_unlock(&kvm
->srcu
, idx
);
421 static int kvm_mmu_notifier_clear_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 * Even though we do not flush TLB, this will still adversely
433 * affect performance on pre-Haswell Intel EPT, where there is
434 * no EPT Access Bit to clear so that we have to tear down EPT
435 * tables instead. If we find this unacceptable, we can always
436 * add a parameter to kvm_age_hva so that it effectively doesn't
437 * do anything on clear_young.
439 * Also note that currently we never issue secondary TLB flushes
440 * from clear_young, leaving this job up to the regular system
441 * cadence. If we find this inaccurate, we might come up with a
442 * more sophisticated heuristic later.
444 young
= kvm_age_hva(kvm
, start
, end
);
445 spin_unlock(&kvm
->mmu_lock
);
446 srcu_read_unlock(&kvm
->srcu
, idx
);
451 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
452 struct mm_struct
*mm
,
453 unsigned long address
)
455 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
458 idx
= srcu_read_lock(&kvm
->srcu
);
459 spin_lock(&kvm
->mmu_lock
);
460 young
= kvm_test_age_hva(kvm
, address
);
461 spin_unlock(&kvm
->mmu_lock
);
462 srcu_read_unlock(&kvm
->srcu
, idx
);
467 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
468 struct mm_struct
*mm
)
470 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
473 idx
= srcu_read_lock(&kvm
->srcu
);
474 kvm_arch_flush_shadow_all(kvm
);
475 srcu_read_unlock(&kvm
->srcu
, idx
);
478 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
479 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
480 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
481 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
482 .clear_young
= kvm_mmu_notifier_clear_young
,
483 .test_young
= kvm_mmu_notifier_test_young
,
484 .change_pte
= kvm_mmu_notifier_change_pte
,
485 .release
= kvm_mmu_notifier_release
,
488 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
490 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
491 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
494 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
496 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
501 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
503 static struct kvm_memslots
*kvm_alloc_memslots(void)
506 struct kvm_memslots
*slots
;
508 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
512 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
513 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
520 if (!memslot
->dirty_bitmap
)
523 kvfree(memslot
->dirty_bitmap
);
524 memslot
->dirty_bitmap
= NULL
;
528 * Free any memory in @free but not in @dont.
530 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
531 struct kvm_memory_slot
*dont
)
533 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
534 kvm_destroy_dirty_bitmap(free
);
536 kvm_arch_free_memslot(kvm
, free
, dont
);
541 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
543 struct kvm_memory_slot
*memslot
;
548 kvm_for_each_memslot(memslot
, slots
)
549 kvm_free_memslot(kvm
, memslot
, NULL
);
554 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
558 if (!kvm
->debugfs_dentry
)
561 debugfs_remove_recursive(kvm
->debugfs_dentry
);
563 if (kvm
->debugfs_stat_data
) {
564 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
565 kfree(kvm
->debugfs_stat_data
[i
]);
566 kfree(kvm
->debugfs_stat_data
);
570 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
572 char dir_name
[ITOA_MAX_LEN
* 2];
573 struct kvm_stat_data
*stat_data
;
574 struct kvm_stats_debugfs_item
*p
;
576 if (!debugfs_initialized())
579 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
580 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
582 if (!kvm
->debugfs_dentry
)
585 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
586 sizeof(*kvm
->debugfs_stat_data
),
588 if (!kvm
->debugfs_stat_data
)
591 for (p
= debugfs_entries
; p
->name
; p
++) {
592 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
596 stat_data
->kvm
= kvm
;
597 stat_data
->offset
= p
->offset
;
598 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
599 if (!debugfs_create_file(p
->name
, 0644,
602 stat_fops_per_vm
[p
->kind
]))
608 static struct kvm
*kvm_create_vm(unsigned long type
)
611 struct kvm
*kvm
= kvm_arch_alloc_vm();
614 return ERR_PTR(-ENOMEM
);
616 spin_lock_init(&kvm
->mmu_lock
);
618 kvm
->mm
= current
->mm
;
619 kvm_eventfd_init(kvm
);
620 mutex_init(&kvm
->lock
);
621 mutex_init(&kvm
->irq_lock
);
622 mutex_init(&kvm
->slots_lock
);
623 refcount_set(&kvm
->users_count
, 1);
624 INIT_LIST_HEAD(&kvm
->devices
);
626 r
= kvm_arch_init_vm(kvm
, type
);
628 goto out_err_no_disable
;
630 r
= hardware_enable_all();
632 goto out_err_no_disable
;
634 #ifdef CONFIG_HAVE_KVM_IRQFD
635 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
638 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
641 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
642 struct kvm_memslots
*slots
= kvm_alloc_memslots();
644 goto out_err_no_srcu
;
646 * Generations must be different for each address space.
647 * Init kvm generation close to the maximum to easily test the
648 * code of handling generation number wrap-around.
650 slots
->generation
= i
* 2 - 150;
651 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
654 if (init_srcu_struct(&kvm
->srcu
))
655 goto out_err_no_srcu
;
656 if (init_srcu_struct(&kvm
->irq_srcu
))
657 goto out_err_no_irq_srcu
;
658 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
659 rcu_assign_pointer(kvm
->buses
[i
],
660 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
665 r
= kvm_init_mmu_notifier(kvm
);
669 spin_lock(&kvm_lock
);
670 list_add(&kvm
->vm_list
, &vm_list
);
671 spin_unlock(&kvm_lock
);
673 preempt_notifier_inc();
678 cleanup_srcu_struct(&kvm
->irq_srcu
);
680 cleanup_srcu_struct(&kvm
->srcu
);
682 hardware_disable_all();
684 refcount_set(&kvm
->users_count
, 0);
685 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
686 kfree(kvm_get_bus(kvm
, i
));
687 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
688 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
689 kvm_arch_free_vm(kvm
);
694 static void kvm_destroy_devices(struct kvm
*kvm
)
696 struct kvm_device
*dev
, *tmp
;
699 * We do not need to take the kvm->lock here, because nobody else
700 * has a reference to the struct kvm at this point and therefore
701 * cannot access the devices list anyhow.
703 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
704 list_del(&dev
->vm_node
);
705 dev
->ops
->destroy(dev
);
709 static void kvm_destroy_vm(struct kvm
*kvm
)
712 struct mm_struct
*mm
= kvm
->mm
;
714 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
715 kvm_destroy_vm_debugfs(kvm
);
716 kvm_arch_sync_events(kvm
);
717 spin_lock(&kvm_lock
);
718 list_del(&kvm
->vm_list
);
719 spin_unlock(&kvm_lock
);
720 kvm_free_irq_routing(kvm
);
721 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
722 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
725 kvm_io_bus_destroy(bus
);
726 kvm
->buses
[i
] = NULL
;
728 kvm_coalesced_mmio_free(kvm
);
729 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
730 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
732 kvm_arch_flush_shadow_all(kvm
);
734 kvm_arch_destroy_vm(kvm
);
735 kvm_destroy_devices(kvm
);
736 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
737 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
738 cleanup_srcu_struct(&kvm
->irq_srcu
);
739 cleanup_srcu_struct(&kvm
->srcu
);
740 kvm_arch_free_vm(kvm
);
741 preempt_notifier_dec();
742 hardware_disable_all();
746 void kvm_get_kvm(struct kvm
*kvm
)
748 refcount_inc(&kvm
->users_count
);
750 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
752 void kvm_put_kvm(struct kvm
*kvm
)
754 if (refcount_dec_and_test(&kvm
->users_count
))
757 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
760 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
762 struct kvm
*kvm
= filp
->private_data
;
764 kvm_irqfd_release(kvm
);
771 * Allocation size is twice as large as the actual dirty bitmap size.
772 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
774 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
776 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
778 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
779 if (!memslot
->dirty_bitmap
)
786 * Insert memslot and re-sort memslots based on their GFN,
787 * so binary search could be used to lookup GFN.
788 * Sorting algorithm takes advantage of having initially
789 * sorted array and known changed memslot position.
791 static void update_memslots(struct kvm_memslots
*slots
,
792 struct kvm_memory_slot
*new)
795 int i
= slots
->id_to_index
[id
];
796 struct kvm_memory_slot
*mslots
= slots
->memslots
;
798 WARN_ON(mslots
[i
].id
!= id
);
800 WARN_ON(!mslots
[i
].npages
);
801 if (mslots
[i
].npages
)
804 if (!mslots
[i
].npages
)
808 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
809 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
810 if (!mslots
[i
+ 1].npages
)
812 mslots
[i
] = mslots
[i
+ 1];
813 slots
->id_to_index
[mslots
[i
].id
] = i
;
818 * The ">=" is needed when creating a slot with base_gfn == 0,
819 * so that it moves before all those with base_gfn == npages == 0.
821 * On the other hand, if new->npages is zero, the above loop has
822 * already left i pointing to the beginning of the empty part of
823 * mslots, and the ">=" would move the hole backwards in this
824 * case---which is wrong. So skip the loop when deleting a slot.
828 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
829 mslots
[i
] = mslots
[i
- 1];
830 slots
->id_to_index
[mslots
[i
].id
] = i
;
834 WARN_ON_ONCE(i
!= slots
->used_slots
);
837 slots
->id_to_index
[mslots
[i
].id
] = i
;
840 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
842 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
844 #ifdef __KVM_HAVE_READONLY_MEM
845 valid_flags
|= KVM_MEM_READONLY
;
848 if (mem
->flags
& ~valid_flags
)
854 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
855 int as_id
, struct kvm_memslots
*slots
)
857 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
860 * Set the low bit in the generation, which disables SPTE caching
861 * until the end of synchronize_srcu_expedited.
863 WARN_ON(old_memslots
->generation
& 1);
864 slots
->generation
= old_memslots
->generation
+ 1;
866 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
867 synchronize_srcu_expedited(&kvm
->srcu
);
870 * Increment the new memslot generation a second time. This prevents
871 * vm exits that race with memslot updates from caching a memslot
872 * generation that will (potentially) be valid forever.
874 * Generations must be unique even across address spaces. We do not need
875 * a global counter for that, instead the generation space is evenly split
876 * across address spaces. For example, with two address spaces, address
877 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
878 * use generations 2, 6, 10, 14, ...
880 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
882 kvm_arch_memslots_updated(kvm
, slots
);
888 * Allocate some memory and give it an address in the guest physical address
891 * Discontiguous memory is allowed, mostly for framebuffers.
893 * Must be called holding kvm->slots_lock for write.
895 int __kvm_set_memory_region(struct kvm
*kvm
,
896 const struct kvm_userspace_memory_region
*mem
)
900 unsigned long npages
;
901 struct kvm_memory_slot
*slot
;
902 struct kvm_memory_slot old
, new;
903 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
905 enum kvm_mr_change change
;
907 r
= check_memory_region_flags(mem
);
912 as_id
= mem
->slot
>> 16;
915 /* General sanity checks */
916 if (mem
->memory_size
& (PAGE_SIZE
- 1))
918 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
920 /* We can read the guest memory with __xxx_user() later on. */
921 if ((id
< KVM_USER_MEM_SLOTS
) &&
922 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
923 !access_ok(VERIFY_WRITE
,
924 (void __user
*)(unsigned long)mem
->userspace_addr
,
927 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
929 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
932 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
933 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
934 npages
= mem
->memory_size
>> PAGE_SHIFT
;
936 if (npages
> KVM_MEM_MAX_NR_PAGES
)
942 new.base_gfn
= base_gfn
;
944 new.flags
= mem
->flags
;
948 change
= KVM_MR_CREATE
;
949 else { /* Modify an existing slot. */
950 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
951 (npages
!= old
.npages
) ||
952 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
955 if (base_gfn
!= old
.base_gfn
)
956 change
= KVM_MR_MOVE
;
957 else if (new.flags
!= old
.flags
)
958 change
= KVM_MR_FLAGS_ONLY
;
959 else { /* Nothing to change. */
968 change
= KVM_MR_DELETE
;
973 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
974 /* Check for overlaps */
976 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
979 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
980 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
985 /* Free page dirty bitmap if unneeded */
986 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
987 new.dirty_bitmap
= NULL
;
990 if (change
== KVM_MR_CREATE
) {
991 new.userspace_addr
= mem
->userspace_addr
;
993 if (kvm_arch_create_memslot(kvm
, &new, npages
))
997 /* Allocate page dirty bitmap if needed */
998 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
999 if (kvm_create_dirty_bitmap(&new) < 0)
1003 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1006 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1008 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1009 slot
= id_to_memslot(slots
, id
);
1010 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1012 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1014 /* From this point no new shadow pages pointing to a deleted,
1015 * or moved, memslot will be created.
1017 * validation of sp->gfn happens in:
1018 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1019 * - kvm_is_visible_gfn (mmu_check_roots)
1021 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1024 * We can re-use the old_memslots from above, the only difference
1025 * from the currently installed memslots is the invalid flag. This
1026 * will get overwritten by update_memslots anyway.
1028 slots
= old_memslots
;
1031 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1035 /* actual memory is freed via old in kvm_free_memslot below */
1036 if (change
== KVM_MR_DELETE
) {
1037 new.dirty_bitmap
= NULL
;
1038 memset(&new.arch
, 0, sizeof(new.arch
));
1041 update_memslots(slots
, &new);
1042 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1044 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1046 kvm_free_memslot(kvm
, &old
, &new);
1047 kvfree(old_memslots
);
1053 kvm_free_memslot(kvm
, &new, &old
);
1057 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1059 int kvm_set_memory_region(struct kvm
*kvm
,
1060 const struct kvm_userspace_memory_region
*mem
)
1064 mutex_lock(&kvm
->slots_lock
);
1065 r
= __kvm_set_memory_region(kvm
, mem
);
1066 mutex_unlock(&kvm
->slots_lock
);
1069 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1071 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1072 struct kvm_userspace_memory_region
*mem
)
1074 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1077 return kvm_set_memory_region(kvm
, mem
);
1080 int kvm_get_dirty_log(struct kvm
*kvm
,
1081 struct kvm_dirty_log
*log
, int *is_dirty
)
1083 struct kvm_memslots
*slots
;
1084 struct kvm_memory_slot
*memslot
;
1087 unsigned long any
= 0;
1089 as_id
= log
->slot
>> 16;
1090 id
= (u16
)log
->slot
;
1091 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1094 slots
= __kvm_memslots(kvm
, as_id
);
1095 memslot
= id_to_memslot(slots
, id
);
1096 if (!memslot
->dirty_bitmap
)
1099 n
= kvm_dirty_bitmap_bytes(memslot
);
1101 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1102 any
= memslot
->dirty_bitmap
[i
];
1104 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1111 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1113 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1115 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1116 * are dirty write protect them for next write.
1117 * @kvm: pointer to kvm instance
1118 * @log: slot id and address to which we copy the log
1119 * @is_dirty: flag set if any page is dirty
1121 * We need to keep it in mind that VCPU threads can write to the bitmap
1122 * concurrently. So, to avoid losing track of dirty pages we keep the
1125 * 1. Take a snapshot of the bit and clear it if needed.
1126 * 2. Write protect the corresponding page.
1127 * 3. Copy the snapshot to the userspace.
1128 * 4. Upon return caller flushes TLB's if needed.
1130 * Between 2 and 4, the guest may write to the page using the remaining TLB
1131 * entry. This is not a problem because the page is reported dirty using
1132 * the snapshot taken before and step 4 ensures that writes done after
1133 * exiting to userspace will be logged for the next call.
1136 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1137 struct kvm_dirty_log
*log
, bool *is_dirty
)
1139 struct kvm_memslots
*slots
;
1140 struct kvm_memory_slot
*memslot
;
1143 unsigned long *dirty_bitmap
;
1144 unsigned long *dirty_bitmap_buffer
;
1146 as_id
= log
->slot
>> 16;
1147 id
= (u16
)log
->slot
;
1148 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1151 slots
= __kvm_memslots(kvm
, as_id
);
1152 memslot
= id_to_memslot(slots
, id
);
1154 dirty_bitmap
= memslot
->dirty_bitmap
;
1158 n
= kvm_dirty_bitmap_bytes(memslot
);
1160 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1161 memset(dirty_bitmap_buffer
, 0, n
);
1163 spin_lock(&kvm
->mmu_lock
);
1165 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1169 if (!dirty_bitmap
[i
])
1174 mask
= xchg(&dirty_bitmap
[i
], 0);
1175 dirty_bitmap_buffer
[i
] = mask
;
1178 offset
= i
* BITS_PER_LONG
;
1179 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1184 spin_unlock(&kvm
->mmu_lock
);
1185 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1189 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1192 bool kvm_largepages_enabled(void)
1194 return largepages_enabled
;
1197 void kvm_disable_largepages(void)
1199 largepages_enabled
= false;
1201 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1203 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1205 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1207 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1209 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1211 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1214 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1216 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1218 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1219 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1224 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1226 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1228 struct vm_area_struct
*vma
;
1229 unsigned long addr
, size
;
1233 addr
= gfn_to_hva(kvm
, gfn
);
1234 if (kvm_is_error_hva(addr
))
1237 down_read(¤t
->mm
->mmap_sem
);
1238 vma
= find_vma(current
->mm
, addr
);
1242 size
= vma_kernel_pagesize(vma
);
1245 up_read(¤t
->mm
->mmap_sem
);
1250 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1252 return slot
->flags
& KVM_MEM_READONLY
;
1255 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1256 gfn_t
*nr_pages
, bool write
)
1258 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1259 return KVM_HVA_ERR_BAD
;
1261 if (memslot_is_readonly(slot
) && write
)
1262 return KVM_HVA_ERR_RO_BAD
;
1265 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1267 return __gfn_to_hva_memslot(slot
, gfn
);
1270 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1273 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1276 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1279 return gfn_to_hva_many(slot
, gfn
, NULL
);
1281 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1283 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1285 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1287 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1289 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1291 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1293 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1296 * If writable is set to false, the hva returned by this function is only
1297 * allowed to be read.
1299 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1300 gfn_t gfn
, bool *writable
)
1302 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1304 if (!kvm_is_error_hva(hva
) && writable
)
1305 *writable
= !memslot_is_readonly(slot
);
1310 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1312 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1314 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1317 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1319 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1321 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1324 static int get_user_page_nowait(unsigned long start
, int write
,
1327 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1330 flags
|= FOLL_WRITE
;
1332 return get_user_pages(start
, 1, flags
, page
, NULL
);
1335 static inline int check_user_page_hwpoison(unsigned long addr
)
1337 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1339 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1340 return rc
== -EHWPOISON
;
1344 * The atomic path to get the writable pfn which will be stored in @pfn,
1345 * true indicates success, otherwise false is returned.
1347 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1348 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1350 struct page
*page
[1];
1353 if (!(async
|| atomic
))
1357 * Fast pin a writable pfn only if it is a write fault request
1358 * or the caller allows to map a writable pfn for a read fault
1361 if (!(write_fault
|| writable
))
1364 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1366 *pfn
= page_to_pfn(page
[0]);
1377 * The slow path to get the pfn of the specified host virtual address,
1378 * 1 indicates success, -errno is returned if error is detected.
1380 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1381 bool *writable
, kvm_pfn_t
*pfn
)
1383 struct page
*page
[1];
1389 *writable
= write_fault
;
1392 down_read(¤t
->mm
->mmap_sem
);
1393 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1394 up_read(¤t
->mm
->mmap_sem
);
1396 unsigned int flags
= FOLL_HWPOISON
;
1399 flags
|= FOLL_WRITE
;
1401 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1406 /* map read fault as writable if possible */
1407 if (unlikely(!write_fault
) && writable
) {
1408 struct page
*wpage
[1];
1410 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1419 *pfn
= page_to_pfn(page
[0]);
1423 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1425 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1428 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1434 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1435 unsigned long addr
, bool *async
,
1436 bool write_fault
, bool *writable
,
1442 r
= follow_pfn(vma
, addr
, &pfn
);
1445 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1446 * not call the fault handler, so do it here.
1448 bool unlocked
= false;
1449 r
= fixup_user_fault(current
, current
->mm
, addr
,
1450 (write_fault
? FAULT_FLAG_WRITE
: 0),
1457 r
= follow_pfn(vma
, addr
, &pfn
);
1467 * Get a reference here because callers of *hva_to_pfn* and
1468 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1469 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1470 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1471 * simply do nothing for reserved pfns.
1473 * Whoever called remap_pfn_range is also going to call e.g.
1474 * unmap_mapping_range before the underlying pages are freed,
1475 * causing a call to our MMU notifier.
1484 * Pin guest page in memory and return its pfn.
1485 * @addr: host virtual address which maps memory to the guest
1486 * @atomic: whether this function can sleep
1487 * @async: whether this function need to wait IO complete if the
1488 * host page is not in the memory
1489 * @write_fault: whether we should get a writable host page
1490 * @writable: whether it allows to map a writable host page for !@write_fault
1492 * The function will map a writable host page for these two cases:
1493 * 1): @write_fault = true
1494 * 2): @write_fault = false && @writable, @writable will tell the caller
1495 * whether the mapping is writable.
1497 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1498 bool write_fault
, bool *writable
)
1500 struct vm_area_struct
*vma
;
1504 /* we can do it either atomically or asynchronously, not both */
1505 BUG_ON(atomic
&& async
);
1507 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1511 return KVM_PFN_ERR_FAULT
;
1513 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1517 down_read(¤t
->mm
->mmap_sem
);
1518 if (npages
== -EHWPOISON
||
1519 (!async
&& check_user_page_hwpoison(addr
))) {
1520 pfn
= KVM_PFN_ERR_HWPOISON
;
1525 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1528 pfn
= KVM_PFN_ERR_FAULT
;
1529 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1530 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1534 pfn
= KVM_PFN_ERR_FAULT
;
1536 if (async
&& vma_is_valid(vma
, write_fault
))
1538 pfn
= KVM_PFN_ERR_FAULT
;
1541 up_read(¤t
->mm
->mmap_sem
);
1545 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1546 bool atomic
, bool *async
, bool write_fault
,
1549 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1551 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1554 return KVM_PFN_ERR_RO_FAULT
;
1557 if (kvm_is_error_hva(addr
)) {
1560 return KVM_PFN_NOSLOT
;
1563 /* Do not map writable pfn in the readonly memslot. */
1564 if (writable
&& memslot_is_readonly(slot
)) {
1569 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1572 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1574 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1577 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1578 write_fault
, writable
);
1580 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1582 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1584 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1586 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1588 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1590 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1592 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1594 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1596 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1598 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1600 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1602 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1604 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1606 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1608 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1610 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1612 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1614 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1616 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1618 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1619 struct page
**pages
, int nr_pages
)
1624 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1625 if (kvm_is_error_hva(addr
))
1628 if (entry
< nr_pages
)
1631 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1633 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1635 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1637 if (is_error_noslot_pfn(pfn
))
1638 return KVM_ERR_PTR_BAD_PAGE
;
1640 if (kvm_is_reserved_pfn(pfn
)) {
1642 return KVM_ERR_PTR_BAD_PAGE
;
1645 return pfn_to_page(pfn
);
1648 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1652 pfn
= gfn_to_pfn(kvm
, gfn
);
1654 return kvm_pfn_to_page(pfn
);
1656 EXPORT_SYMBOL_GPL(gfn_to_page
);
1658 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1662 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1664 return kvm_pfn_to_page(pfn
);
1666 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1668 void kvm_release_page_clean(struct page
*page
)
1670 WARN_ON(is_error_page(page
));
1672 kvm_release_pfn_clean(page_to_pfn(page
));
1674 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1676 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1678 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1679 put_page(pfn_to_page(pfn
));
1681 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1683 void kvm_release_page_dirty(struct page
*page
)
1685 WARN_ON(is_error_page(page
));
1687 kvm_release_pfn_dirty(page_to_pfn(page
));
1689 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1691 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1693 kvm_set_pfn_dirty(pfn
);
1694 kvm_release_pfn_clean(pfn
);
1696 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1698 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1700 if (!kvm_is_reserved_pfn(pfn
)) {
1701 struct page
*page
= pfn_to_page(pfn
);
1703 if (!PageReserved(page
))
1707 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1709 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1711 if (!kvm_is_reserved_pfn(pfn
))
1712 mark_page_accessed(pfn_to_page(pfn
));
1714 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1716 void kvm_get_pfn(kvm_pfn_t pfn
)
1718 if (!kvm_is_reserved_pfn(pfn
))
1719 get_page(pfn_to_page(pfn
));
1721 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1723 static int next_segment(unsigned long len
, int offset
)
1725 if (len
> PAGE_SIZE
- offset
)
1726 return PAGE_SIZE
- offset
;
1731 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1732 void *data
, int offset
, int len
)
1737 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1738 if (kvm_is_error_hva(addr
))
1740 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1746 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1749 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1751 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1753 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1755 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1756 int offset
, int len
)
1758 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1760 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1762 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1764 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1766 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1768 int offset
= offset_in_page(gpa
);
1771 while ((seg
= next_segment(len
, offset
)) != 0) {
1772 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1782 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1784 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1786 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1788 int offset
= offset_in_page(gpa
);
1791 while ((seg
= next_segment(len
, offset
)) != 0) {
1792 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1802 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1804 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1805 void *data
, int offset
, unsigned long len
)
1810 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1811 if (kvm_is_error_hva(addr
))
1813 pagefault_disable();
1814 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1821 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1824 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1825 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1826 int offset
= offset_in_page(gpa
);
1828 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1830 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1832 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1833 void *data
, unsigned long len
)
1835 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1836 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1837 int offset
= offset_in_page(gpa
);
1839 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1841 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1843 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1844 const void *data
, int offset
, int len
)
1849 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1850 if (kvm_is_error_hva(addr
))
1852 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1855 mark_page_dirty_in_slot(memslot
, gfn
);
1859 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1860 const void *data
, int offset
, int len
)
1862 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1864 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1866 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1868 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1869 const void *data
, int offset
, int len
)
1871 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1873 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1875 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1877 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1880 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1882 int offset
= offset_in_page(gpa
);
1885 while ((seg
= next_segment(len
, offset
)) != 0) {
1886 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1896 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1898 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1901 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1903 int offset
= offset_in_page(gpa
);
1906 while ((seg
= next_segment(len
, offset
)) != 0) {
1907 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1917 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1919 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1920 struct gfn_to_hva_cache
*ghc
,
1921 gpa_t gpa
, unsigned long len
)
1923 int offset
= offset_in_page(gpa
);
1924 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1925 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1926 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1927 gfn_t nr_pages_avail
;
1930 ghc
->generation
= slots
->generation
;
1932 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1933 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1934 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1938 * If the requested region crosses two memslots, we still
1939 * verify that the entire region is valid here.
1941 while (start_gfn
<= end_gfn
) {
1943 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1944 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1946 if (kvm_is_error_hva(ghc
->hva
))
1948 start_gfn
+= nr_pages_avail
;
1950 /* Use the slow path for cross page reads and writes. */
1951 ghc
->memslot
= NULL
;
1956 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1957 gpa_t gpa
, unsigned long len
)
1959 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1960 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1962 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1964 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1965 void *data
, int offset
, unsigned long len
)
1967 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1969 gpa_t gpa
= ghc
->gpa
+ offset
;
1971 BUG_ON(len
+ offset
> ghc
->len
);
1973 if (slots
->generation
!= ghc
->generation
)
1974 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1976 if (unlikely(!ghc
->memslot
))
1977 return kvm_write_guest(kvm
, gpa
, data
, len
);
1979 if (kvm_is_error_hva(ghc
->hva
))
1982 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
1985 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
1989 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
1991 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1992 void *data
, unsigned long len
)
1994 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
1996 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1998 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1999 void *data
, unsigned long len
)
2001 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2004 BUG_ON(len
> ghc
->len
);
2006 if (slots
->generation
!= ghc
->generation
)
2007 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2009 if (unlikely(!ghc
->memslot
))
2010 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2012 if (kvm_is_error_hva(ghc
->hva
))
2015 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2021 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2023 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2025 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2027 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2029 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2031 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2033 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2035 int offset
= offset_in_page(gpa
);
2038 while ((seg
= next_segment(len
, offset
)) != 0) {
2039 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2048 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2050 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2053 if (memslot
&& memslot
->dirty_bitmap
) {
2054 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2056 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2060 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2062 struct kvm_memory_slot
*memslot
;
2064 memslot
= gfn_to_memslot(kvm
, gfn
);
2065 mark_page_dirty_in_slot(memslot
, gfn
);
2067 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2069 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2071 struct kvm_memory_slot
*memslot
;
2073 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2074 mark_page_dirty_in_slot(memslot
, gfn
);
2076 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2078 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2080 if (!vcpu
->sigset_active
)
2084 * This does a lockless modification of ->real_blocked, which is fine
2085 * because, only current can change ->real_blocked and all readers of
2086 * ->real_blocked don't care as long ->real_blocked is always a subset
2089 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2092 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2094 if (!vcpu
->sigset_active
)
2097 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2098 sigemptyset(¤t
->real_blocked
);
2101 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2103 unsigned int old
, val
, grow
;
2105 old
= val
= vcpu
->halt_poll_ns
;
2106 grow
= READ_ONCE(halt_poll_ns_grow
);
2108 if (val
== 0 && grow
)
2113 if (val
> halt_poll_ns
)
2116 vcpu
->halt_poll_ns
= val
;
2117 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2120 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2122 unsigned int old
, val
, shrink
;
2124 old
= val
= vcpu
->halt_poll_ns
;
2125 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2131 vcpu
->halt_poll_ns
= val
;
2132 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2135 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2137 if (kvm_arch_vcpu_runnable(vcpu
)) {
2138 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2141 if (kvm_cpu_has_pending_timer(vcpu
))
2143 if (signal_pending(current
))
2150 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2152 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2155 DECLARE_SWAITQUEUE(wait
);
2156 bool waited
= false;
2159 start
= cur
= ktime_get();
2160 if (vcpu
->halt_poll_ns
) {
2161 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2163 ++vcpu
->stat
.halt_attempted_poll
;
2166 * This sets KVM_REQ_UNHALT if an interrupt
2169 if (kvm_vcpu_check_block(vcpu
) < 0) {
2170 ++vcpu
->stat
.halt_successful_poll
;
2171 if (!vcpu_valid_wakeup(vcpu
))
2172 ++vcpu
->stat
.halt_poll_invalid
;
2176 } while (single_task_running() && ktime_before(cur
, stop
));
2179 kvm_arch_vcpu_blocking(vcpu
);
2182 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2184 if (kvm_vcpu_check_block(vcpu
) < 0)
2191 finish_swait(&vcpu
->wq
, &wait
);
2194 kvm_arch_vcpu_unblocking(vcpu
);
2196 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2198 if (!vcpu_valid_wakeup(vcpu
))
2199 shrink_halt_poll_ns(vcpu
);
2200 else if (halt_poll_ns
) {
2201 if (block_ns
<= vcpu
->halt_poll_ns
)
2203 /* we had a long block, shrink polling */
2204 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2205 shrink_halt_poll_ns(vcpu
);
2206 /* we had a short halt and our poll time is too small */
2207 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2208 block_ns
< halt_poll_ns
)
2209 grow_halt_poll_ns(vcpu
);
2211 vcpu
->halt_poll_ns
= 0;
2213 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2214 kvm_arch_vcpu_block_finish(vcpu
);
2216 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2218 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2220 struct swait_queue_head
*wqp
;
2222 wqp
= kvm_arch_vcpu_wq(vcpu
);
2223 if (swq_has_sleeper(wqp
)) {
2225 ++vcpu
->stat
.halt_wakeup
;
2231 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2235 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2237 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2240 int cpu
= vcpu
->cpu
;
2242 if (kvm_vcpu_wake_up(vcpu
))
2246 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2247 if (kvm_arch_vcpu_should_kick(vcpu
))
2248 smp_send_reschedule(cpu
);
2251 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2252 #endif /* !CONFIG_S390 */
2254 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2257 struct task_struct
*task
= NULL
;
2261 pid
= rcu_dereference(target
->pid
);
2263 task
= get_pid_task(pid
, PIDTYPE_PID
);
2267 ret
= yield_to(task
, 1);
2268 put_task_struct(task
);
2272 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2275 * Helper that checks whether a VCPU is eligible for directed yield.
2276 * Most eligible candidate to yield is decided by following heuristics:
2278 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2279 * (preempted lock holder), indicated by @in_spin_loop.
2280 * Set at the beiginning and cleared at the end of interception/PLE handler.
2282 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2283 * chance last time (mostly it has become eligible now since we have probably
2284 * yielded to lockholder in last iteration. This is done by toggling
2285 * @dy_eligible each time a VCPU checked for eligibility.)
2287 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2288 * to preempted lock-holder could result in wrong VCPU selection and CPU
2289 * burning. Giving priority for a potential lock-holder increases lock
2292 * Since algorithm is based on heuristics, accessing another VCPU data without
2293 * locking does not harm. It may result in trying to yield to same VCPU, fail
2294 * and continue with next VCPU and so on.
2296 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2298 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2301 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2302 vcpu
->spin_loop
.dy_eligible
;
2304 if (vcpu
->spin_loop
.in_spin_loop
)
2305 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2313 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2315 struct kvm
*kvm
= me
->kvm
;
2316 struct kvm_vcpu
*vcpu
;
2317 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2323 kvm_vcpu_set_in_spin_loop(me
, true);
2325 * We boost the priority of a VCPU that is runnable but not
2326 * currently running, because it got preempted by something
2327 * else and called schedule in __vcpu_run. Hopefully that
2328 * VCPU is holding the lock that we need and will release it.
2329 * We approximate round-robin by starting at the last boosted VCPU.
2331 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2332 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2333 if (!pass
&& i
<= last_boosted_vcpu
) {
2334 i
= last_boosted_vcpu
;
2336 } else if (pass
&& i
> last_boosted_vcpu
)
2338 if (!READ_ONCE(vcpu
->preempted
))
2342 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2344 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2346 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2349 yielded
= kvm_vcpu_yield_to(vcpu
);
2351 kvm
->last_boosted_vcpu
= i
;
2353 } else if (yielded
< 0) {
2360 kvm_vcpu_set_in_spin_loop(me
, false);
2362 /* Ensure vcpu is not eligible during next spinloop */
2363 kvm_vcpu_set_dy_eligible(me
, false);
2365 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2367 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2369 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2372 if (vmf
->pgoff
== 0)
2373 page
= virt_to_page(vcpu
->run
);
2375 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2376 page
= virt_to_page(vcpu
->arch
.pio_data
);
2378 #ifdef CONFIG_KVM_MMIO
2379 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2380 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2383 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2389 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2390 .fault
= kvm_vcpu_fault
,
2393 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2395 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2399 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2401 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2403 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2404 kvm_put_kvm(vcpu
->kvm
);
2408 static struct file_operations kvm_vcpu_fops
= {
2409 .release
= kvm_vcpu_release
,
2410 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2411 #ifdef CONFIG_KVM_COMPAT
2412 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2414 .mmap
= kvm_vcpu_mmap
,
2415 .llseek
= noop_llseek
,
2419 * Allocates an inode for the vcpu.
2421 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2423 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2426 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2428 char dir_name
[ITOA_MAX_LEN
* 2];
2431 if (!kvm_arch_has_vcpu_debugfs())
2434 if (!debugfs_initialized())
2437 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2438 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2439 vcpu
->kvm
->debugfs_dentry
);
2440 if (!vcpu
->debugfs_dentry
)
2443 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2445 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2453 * Creates some virtual cpus. Good luck creating more than one.
2455 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2458 struct kvm_vcpu
*vcpu
;
2460 if (id
>= KVM_MAX_VCPU_ID
)
2463 mutex_lock(&kvm
->lock
);
2464 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2465 mutex_unlock(&kvm
->lock
);
2469 kvm
->created_vcpus
++;
2470 mutex_unlock(&kvm
->lock
);
2472 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2475 goto vcpu_decrement
;
2478 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2480 r
= kvm_arch_vcpu_setup(vcpu
);
2484 r
= kvm_create_vcpu_debugfs(vcpu
);
2488 mutex_lock(&kvm
->lock
);
2489 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2491 goto unlock_vcpu_destroy
;
2494 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2496 /* Now it's all set up, let userspace reach it */
2498 r
= create_vcpu_fd(vcpu
);
2501 goto unlock_vcpu_destroy
;
2504 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2507 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2508 * before kvm->online_vcpu's incremented value.
2511 atomic_inc(&kvm
->online_vcpus
);
2513 mutex_unlock(&kvm
->lock
);
2514 kvm_arch_vcpu_postcreate(vcpu
);
2517 unlock_vcpu_destroy
:
2518 mutex_unlock(&kvm
->lock
);
2519 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2521 kvm_arch_vcpu_destroy(vcpu
);
2523 mutex_lock(&kvm
->lock
);
2524 kvm
->created_vcpus
--;
2525 mutex_unlock(&kvm
->lock
);
2529 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2532 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2533 vcpu
->sigset_active
= 1;
2534 vcpu
->sigset
= *sigset
;
2536 vcpu
->sigset_active
= 0;
2540 static long kvm_vcpu_ioctl(struct file
*filp
,
2541 unsigned int ioctl
, unsigned long arg
)
2543 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2544 void __user
*argp
= (void __user
*)arg
;
2546 struct kvm_fpu
*fpu
= NULL
;
2547 struct kvm_sregs
*kvm_sregs
= NULL
;
2549 if (vcpu
->kvm
->mm
!= current
->mm
)
2552 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2555 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2557 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2558 * so vcpu_load() would break it.
2560 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2561 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2565 r
= vcpu_load(vcpu
);
2574 oldpid
= rcu_access_pointer(vcpu
->pid
);
2575 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2576 /* The thread running this VCPU changed. */
2577 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2579 rcu_assign_pointer(vcpu
->pid
, newpid
);
2584 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2585 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2588 case KVM_GET_REGS
: {
2589 struct kvm_regs
*kvm_regs
;
2592 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2595 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2599 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2606 case KVM_SET_REGS
: {
2607 struct kvm_regs
*kvm_regs
;
2610 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2611 if (IS_ERR(kvm_regs
)) {
2612 r
= PTR_ERR(kvm_regs
);
2615 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2619 case KVM_GET_SREGS
: {
2620 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2624 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2628 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2633 case KVM_SET_SREGS
: {
2634 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2635 if (IS_ERR(kvm_sregs
)) {
2636 r
= PTR_ERR(kvm_sregs
);
2640 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2643 case KVM_GET_MP_STATE
: {
2644 struct kvm_mp_state mp_state
;
2646 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2650 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2655 case KVM_SET_MP_STATE
: {
2656 struct kvm_mp_state mp_state
;
2659 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2661 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2664 case KVM_TRANSLATE
: {
2665 struct kvm_translation tr
;
2668 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2670 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2674 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2679 case KVM_SET_GUEST_DEBUG
: {
2680 struct kvm_guest_debug dbg
;
2683 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2685 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2688 case KVM_SET_SIGNAL_MASK
: {
2689 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2690 struct kvm_signal_mask kvm_sigmask
;
2691 sigset_t sigset
, *p
;
2696 if (copy_from_user(&kvm_sigmask
, argp
,
2697 sizeof(kvm_sigmask
)))
2700 if (kvm_sigmask
.len
!= sizeof(sigset
))
2703 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2708 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2712 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2716 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2720 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2726 fpu
= memdup_user(argp
, sizeof(*fpu
));
2732 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2736 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2745 #ifdef CONFIG_KVM_COMPAT
2746 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2747 unsigned int ioctl
, unsigned long arg
)
2749 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2750 void __user
*argp
= compat_ptr(arg
);
2753 if (vcpu
->kvm
->mm
!= current
->mm
)
2757 case KVM_SET_SIGNAL_MASK
: {
2758 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2759 struct kvm_signal_mask kvm_sigmask
;
2764 if (copy_from_user(&kvm_sigmask
, argp
,
2765 sizeof(kvm_sigmask
)))
2768 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2771 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2773 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2775 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2779 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2787 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2788 int (*accessor
)(struct kvm_device
*dev
,
2789 struct kvm_device_attr
*attr
),
2792 struct kvm_device_attr attr
;
2797 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2800 return accessor(dev
, &attr
);
2803 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2806 struct kvm_device
*dev
= filp
->private_data
;
2809 case KVM_SET_DEVICE_ATTR
:
2810 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2811 case KVM_GET_DEVICE_ATTR
:
2812 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2813 case KVM_HAS_DEVICE_ATTR
:
2814 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2816 if (dev
->ops
->ioctl
)
2817 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2823 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2825 struct kvm_device
*dev
= filp
->private_data
;
2826 struct kvm
*kvm
= dev
->kvm
;
2832 static const struct file_operations kvm_device_fops
= {
2833 .unlocked_ioctl
= kvm_device_ioctl
,
2834 #ifdef CONFIG_KVM_COMPAT
2835 .compat_ioctl
= kvm_device_ioctl
,
2837 .release
= kvm_device_release
,
2840 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2842 if (filp
->f_op
!= &kvm_device_fops
)
2845 return filp
->private_data
;
2848 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2849 #ifdef CONFIG_KVM_MPIC
2850 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2851 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2855 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2857 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2860 if (kvm_device_ops_table
[type
] != NULL
)
2863 kvm_device_ops_table
[type
] = ops
;
2867 void kvm_unregister_device_ops(u32 type
)
2869 if (kvm_device_ops_table
[type
] != NULL
)
2870 kvm_device_ops_table
[type
] = NULL
;
2873 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2874 struct kvm_create_device
*cd
)
2876 struct kvm_device_ops
*ops
= NULL
;
2877 struct kvm_device
*dev
;
2878 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2881 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2884 ops
= kvm_device_ops_table
[cd
->type
];
2891 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2898 mutex_lock(&kvm
->lock
);
2899 ret
= ops
->create(dev
, cd
->type
);
2901 mutex_unlock(&kvm
->lock
);
2905 list_add(&dev
->vm_node
, &kvm
->devices
);
2906 mutex_unlock(&kvm
->lock
);
2911 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2913 mutex_lock(&kvm
->lock
);
2914 list_del(&dev
->vm_node
);
2915 mutex_unlock(&kvm
->lock
);
2925 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2928 case KVM_CAP_USER_MEMORY
:
2929 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2930 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2931 case KVM_CAP_INTERNAL_ERROR_DATA
:
2932 #ifdef CONFIG_HAVE_KVM_MSI
2933 case KVM_CAP_SIGNAL_MSI
:
2935 #ifdef CONFIG_HAVE_KVM_IRQFD
2937 case KVM_CAP_IRQFD_RESAMPLE
:
2939 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2940 case KVM_CAP_CHECK_EXTENSION_VM
:
2942 #ifdef CONFIG_KVM_MMIO
2943 case KVM_CAP_COALESCED_MMIO
:
2944 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2946 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2947 case KVM_CAP_IRQ_ROUTING
:
2948 return KVM_MAX_IRQ_ROUTES
;
2950 #if KVM_ADDRESS_SPACE_NUM > 1
2951 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2952 return KVM_ADDRESS_SPACE_NUM
;
2954 case KVM_CAP_MAX_VCPU_ID
:
2955 return KVM_MAX_VCPU_ID
;
2959 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2962 static long kvm_vm_ioctl(struct file
*filp
,
2963 unsigned int ioctl
, unsigned long arg
)
2965 struct kvm
*kvm
= filp
->private_data
;
2966 void __user
*argp
= (void __user
*)arg
;
2969 if (kvm
->mm
!= current
->mm
)
2972 case KVM_CREATE_VCPU
:
2973 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2975 case KVM_SET_USER_MEMORY_REGION
: {
2976 struct kvm_userspace_memory_region kvm_userspace_mem
;
2979 if (copy_from_user(&kvm_userspace_mem
, argp
,
2980 sizeof(kvm_userspace_mem
)))
2983 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2986 case KVM_GET_DIRTY_LOG
: {
2987 struct kvm_dirty_log log
;
2990 if (copy_from_user(&log
, argp
, sizeof(log
)))
2992 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2995 #ifdef CONFIG_KVM_MMIO
2996 case KVM_REGISTER_COALESCED_MMIO
: {
2997 struct kvm_coalesced_mmio_zone zone
;
3000 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3002 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3005 case KVM_UNREGISTER_COALESCED_MMIO
: {
3006 struct kvm_coalesced_mmio_zone zone
;
3009 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3011 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3016 struct kvm_irqfd data
;
3019 if (copy_from_user(&data
, argp
, sizeof(data
)))
3021 r
= kvm_irqfd(kvm
, &data
);
3024 case KVM_IOEVENTFD
: {
3025 struct kvm_ioeventfd data
;
3028 if (copy_from_user(&data
, argp
, sizeof(data
)))
3030 r
= kvm_ioeventfd(kvm
, &data
);
3033 #ifdef CONFIG_HAVE_KVM_MSI
3034 case KVM_SIGNAL_MSI
: {
3038 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3040 r
= kvm_send_userspace_msi(kvm
, &msi
);
3044 #ifdef __KVM_HAVE_IRQ_LINE
3045 case KVM_IRQ_LINE_STATUS
:
3046 case KVM_IRQ_LINE
: {
3047 struct kvm_irq_level irq_event
;
3050 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3053 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3054 ioctl
== KVM_IRQ_LINE_STATUS
);
3059 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3060 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3068 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3069 case KVM_SET_GSI_ROUTING
: {
3070 struct kvm_irq_routing routing
;
3071 struct kvm_irq_routing __user
*urouting
;
3072 struct kvm_irq_routing_entry
*entries
= NULL
;
3075 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3078 if (!kvm_arch_can_set_irq_routing(kvm
))
3080 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3086 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3091 if (copy_from_user(entries
, urouting
->entries
,
3092 routing
.nr
* sizeof(*entries
)))
3093 goto out_free_irq_routing
;
3095 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3097 out_free_irq_routing
:
3101 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3102 case KVM_CREATE_DEVICE
: {
3103 struct kvm_create_device cd
;
3106 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3109 r
= kvm_ioctl_create_device(kvm
, &cd
);
3114 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3120 case KVM_CHECK_EXTENSION
:
3121 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3124 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3130 #ifdef CONFIG_KVM_COMPAT
3131 struct compat_kvm_dirty_log
{
3135 compat_uptr_t dirty_bitmap
; /* one bit per page */
3140 static long kvm_vm_compat_ioctl(struct file
*filp
,
3141 unsigned int ioctl
, unsigned long arg
)
3143 struct kvm
*kvm
= filp
->private_data
;
3146 if (kvm
->mm
!= current
->mm
)
3149 case KVM_GET_DIRTY_LOG
: {
3150 struct compat_kvm_dirty_log compat_log
;
3151 struct kvm_dirty_log log
;
3153 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3154 sizeof(compat_log
)))
3156 log
.slot
= compat_log
.slot
;
3157 log
.padding1
= compat_log
.padding1
;
3158 log
.padding2
= compat_log
.padding2
;
3159 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3161 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3165 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3171 static struct file_operations kvm_vm_fops
= {
3172 .release
= kvm_vm_release
,
3173 .unlocked_ioctl
= kvm_vm_ioctl
,
3174 #ifdef CONFIG_KVM_COMPAT
3175 .compat_ioctl
= kvm_vm_compat_ioctl
,
3177 .llseek
= noop_llseek
,
3180 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3186 kvm
= kvm_create_vm(type
);
3188 return PTR_ERR(kvm
);
3189 #ifdef CONFIG_KVM_MMIO
3190 r
= kvm_coalesced_mmio_init(kvm
);
3196 r
= get_unused_fd_flags(O_CLOEXEC
);
3201 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3205 return PTR_ERR(file
);
3209 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3210 * already set, with ->release() being kvm_vm_release(). In error
3211 * cases it will be called by the final fput(file) and will take
3212 * care of doing kvm_put_kvm(kvm).
3214 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3219 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3221 fd_install(r
, file
);
3225 static long kvm_dev_ioctl(struct file
*filp
,
3226 unsigned int ioctl
, unsigned long arg
)
3231 case KVM_GET_API_VERSION
:
3234 r
= KVM_API_VERSION
;
3237 r
= kvm_dev_ioctl_create_vm(arg
);
3239 case KVM_CHECK_EXTENSION
:
3240 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3242 case KVM_GET_VCPU_MMAP_SIZE
:
3245 r
= PAGE_SIZE
; /* struct kvm_run */
3247 r
+= PAGE_SIZE
; /* pio data page */
3249 #ifdef CONFIG_KVM_MMIO
3250 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3253 case KVM_TRACE_ENABLE
:
3254 case KVM_TRACE_PAUSE
:
3255 case KVM_TRACE_DISABLE
:
3259 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3265 static struct file_operations kvm_chardev_ops
= {
3266 .unlocked_ioctl
= kvm_dev_ioctl
,
3267 .compat_ioctl
= kvm_dev_ioctl
,
3268 .llseek
= noop_llseek
,
3271 static struct miscdevice kvm_dev
= {
3277 static void hardware_enable_nolock(void *junk
)
3279 int cpu
= raw_smp_processor_id();
3282 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3285 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3287 r
= kvm_arch_hardware_enable();
3290 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3291 atomic_inc(&hardware_enable_failed
);
3292 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3296 static int kvm_starting_cpu(unsigned int cpu
)
3298 raw_spin_lock(&kvm_count_lock
);
3299 if (kvm_usage_count
)
3300 hardware_enable_nolock(NULL
);
3301 raw_spin_unlock(&kvm_count_lock
);
3305 static void hardware_disable_nolock(void *junk
)
3307 int cpu
= raw_smp_processor_id();
3309 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3311 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3312 kvm_arch_hardware_disable();
3315 static int kvm_dying_cpu(unsigned int cpu
)
3317 raw_spin_lock(&kvm_count_lock
);
3318 if (kvm_usage_count
)
3319 hardware_disable_nolock(NULL
);
3320 raw_spin_unlock(&kvm_count_lock
);
3324 static void hardware_disable_all_nolock(void)
3326 BUG_ON(!kvm_usage_count
);
3329 if (!kvm_usage_count
)
3330 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3333 static void hardware_disable_all(void)
3335 raw_spin_lock(&kvm_count_lock
);
3336 hardware_disable_all_nolock();
3337 raw_spin_unlock(&kvm_count_lock
);
3340 static int hardware_enable_all(void)
3344 raw_spin_lock(&kvm_count_lock
);
3347 if (kvm_usage_count
== 1) {
3348 atomic_set(&hardware_enable_failed
, 0);
3349 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3351 if (atomic_read(&hardware_enable_failed
)) {
3352 hardware_disable_all_nolock();
3357 raw_spin_unlock(&kvm_count_lock
);
3362 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3366 * Some (well, at least mine) BIOSes hang on reboot if
3369 * And Intel TXT required VMX off for all cpu when system shutdown.
3371 pr_info("kvm: exiting hardware virtualization\n");
3372 kvm_rebooting
= true;
3373 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3377 static struct notifier_block kvm_reboot_notifier
= {
3378 .notifier_call
= kvm_reboot
,
3382 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3386 for (i
= 0; i
< bus
->dev_count
; i
++) {
3387 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3389 kvm_iodevice_destructor(pos
);
3394 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3395 const struct kvm_io_range
*r2
)
3397 gpa_t addr1
= r1
->addr
;
3398 gpa_t addr2
= r2
->addr
;
3403 /* If r2->len == 0, match the exact address. If r2->len != 0,
3404 * accept any overlapping write. Any order is acceptable for
3405 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3406 * we process all of them.
3419 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3421 return kvm_io_bus_cmp(p1
, p2
);
3424 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3425 gpa_t addr
, int len
)
3427 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3433 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3434 kvm_io_bus_sort_cmp
, NULL
);
3439 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3440 gpa_t addr
, int len
)
3442 struct kvm_io_range
*range
, key
;
3445 key
= (struct kvm_io_range
) {
3450 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3451 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3455 off
= range
- bus
->range
;
3457 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3463 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3464 struct kvm_io_range
*range
, const void *val
)
3468 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3472 while (idx
< bus
->dev_count
&&
3473 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3474 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3483 /* kvm_io_bus_write - called under kvm->slots_lock */
3484 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3485 int len
, const void *val
)
3487 struct kvm_io_bus
*bus
;
3488 struct kvm_io_range range
;
3491 range
= (struct kvm_io_range
) {
3496 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3499 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3500 return r
< 0 ? r
: 0;
3503 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3504 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3505 gpa_t addr
, int len
, const void *val
, long cookie
)
3507 struct kvm_io_bus
*bus
;
3508 struct kvm_io_range range
;
3510 range
= (struct kvm_io_range
) {
3515 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3519 /* First try the device referenced by cookie. */
3520 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3521 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3522 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3527 * cookie contained garbage; fall back to search and return the
3528 * correct cookie value.
3530 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3533 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3534 struct kvm_io_range
*range
, void *val
)
3538 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3542 while (idx
< bus
->dev_count
&&
3543 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3544 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3552 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3554 /* kvm_io_bus_read - called under kvm->slots_lock */
3555 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3558 struct kvm_io_bus
*bus
;
3559 struct kvm_io_range range
;
3562 range
= (struct kvm_io_range
) {
3567 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3570 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3571 return r
< 0 ? r
: 0;
3575 /* Caller must hold slots_lock. */
3576 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3577 int len
, struct kvm_io_device
*dev
)
3579 struct kvm_io_bus
*new_bus
, *bus
;
3581 bus
= kvm_get_bus(kvm
, bus_idx
);
3585 /* exclude ioeventfd which is limited by maximum fd */
3586 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3589 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3590 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3593 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3594 sizeof(struct kvm_io_range
)));
3595 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3596 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3597 synchronize_srcu_expedited(&kvm
->srcu
);
3603 /* Caller must hold slots_lock. */
3604 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3605 struct kvm_io_device
*dev
)
3608 struct kvm_io_bus
*new_bus
, *bus
;
3610 bus
= kvm_get_bus(kvm
, bus_idx
);
3614 for (i
= 0; i
< bus
->dev_count
; i
++)
3615 if (bus
->range
[i
].dev
== dev
) {
3619 if (i
== bus
->dev_count
)
3622 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3623 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3625 pr_err("kvm: failed to shrink bus, removing it completely\n");
3629 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3630 new_bus
->dev_count
--;
3631 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3632 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3635 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3636 synchronize_srcu_expedited(&kvm
->srcu
);
3641 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3644 struct kvm_io_bus
*bus
;
3645 int dev_idx
, srcu_idx
;
3646 struct kvm_io_device
*iodev
= NULL
;
3648 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3650 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3654 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3658 iodev
= bus
->range
[dev_idx
].dev
;
3661 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3665 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3667 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3668 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3671 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3674 /* The debugfs files are a reference to the kvm struct which
3675 * is still valid when kvm_destroy_vm is called.
3676 * To avoid the race between open and the removal of the debugfs
3677 * directory we test against the users count.
3679 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3682 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3683 kvm_put_kvm(stat_data
->kvm
);
3690 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3692 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3695 simple_attr_release(inode
, file
);
3696 kvm_put_kvm(stat_data
->kvm
);
3701 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3703 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3705 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3710 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3712 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3717 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3722 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3724 __simple_attr_check_format("%llu\n", 0ull);
3725 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3726 vm_stat_clear_per_vm
, "%llu\n");
3729 static const struct file_operations vm_stat_get_per_vm_fops
= {
3730 .owner
= THIS_MODULE
,
3731 .open
= vm_stat_get_per_vm_open
,
3732 .release
= kvm_debugfs_release
,
3733 .read
= simple_attr_read
,
3734 .write
= simple_attr_write
,
3735 .llseek
= no_llseek
,
3738 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3741 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3742 struct kvm_vcpu
*vcpu
;
3746 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3747 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3752 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3755 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3756 struct kvm_vcpu
*vcpu
;
3761 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3762 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3767 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3769 __simple_attr_check_format("%llu\n", 0ull);
3770 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3771 vcpu_stat_clear_per_vm
, "%llu\n");
3774 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3775 .owner
= THIS_MODULE
,
3776 .open
= vcpu_stat_get_per_vm_open
,
3777 .release
= kvm_debugfs_release
,
3778 .read
= simple_attr_read
,
3779 .write
= simple_attr_write
,
3780 .llseek
= no_llseek
,
3783 static const struct file_operations
*stat_fops_per_vm
[] = {
3784 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3785 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3788 static int vm_stat_get(void *_offset
, u64
*val
)
3790 unsigned offset
= (long)_offset
;
3792 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3796 spin_lock(&kvm_lock
);
3797 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3799 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3802 spin_unlock(&kvm_lock
);
3806 static int vm_stat_clear(void *_offset
, u64 val
)
3808 unsigned offset
= (long)_offset
;
3810 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3815 spin_lock(&kvm_lock
);
3816 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3818 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3820 spin_unlock(&kvm_lock
);
3825 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3827 static int vcpu_stat_get(void *_offset
, u64
*val
)
3829 unsigned offset
= (long)_offset
;
3831 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3835 spin_lock(&kvm_lock
);
3836 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3838 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3841 spin_unlock(&kvm_lock
);
3845 static int vcpu_stat_clear(void *_offset
, u64 val
)
3847 unsigned offset
= (long)_offset
;
3849 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3854 spin_lock(&kvm_lock
);
3855 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3857 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3859 spin_unlock(&kvm_lock
);
3864 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3867 static const struct file_operations
*stat_fops
[] = {
3868 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3869 [KVM_STAT_VM
] = &vm_stat_fops
,
3872 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3874 struct kobj_uevent_env
*env
;
3875 unsigned long long created
, active
;
3877 if (!kvm_dev
.this_device
|| !kvm
)
3880 spin_lock(&kvm_lock
);
3881 if (type
== KVM_EVENT_CREATE_VM
) {
3882 kvm_createvm_count
++;
3884 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3887 created
= kvm_createvm_count
;
3888 active
= kvm_active_vms
;
3889 spin_unlock(&kvm_lock
);
3891 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3895 add_uevent_var(env
, "CREATED=%llu", created
);
3896 add_uevent_var(env
, "COUNT=%llu", active
);
3898 if (type
== KVM_EVENT_CREATE_VM
) {
3899 add_uevent_var(env
, "EVENT=create");
3900 kvm
->userspace_pid
= task_pid_nr(current
);
3901 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3902 add_uevent_var(env
, "EVENT=destroy");
3904 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3906 if (kvm
->debugfs_dentry
) {
3907 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3910 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3912 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3916 /* no need for checks, since we are adding at most only 5 keys */
3917 env
->envp
[env
->envp_idx
++] = NULL
;
3918 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3922 static int kvm_init_debug(void)
3925 struct kvm_stats_debugfs_item
*p
;
3927 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3928 if (kvm_debugfs_dir
== NULL
)
3931 kvm_debugfs_num_entries
= 0;
3932 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3933 if (!debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3934 (void *)(long)p
->offset
,
3935 stat_fops
[p
->kind
]))
3942 debugfs_remove_recursive(kvm_debugfs_dir
);
3947 static int kvm_suspend(void)
3949 if (kvm_usage_count
)
3950 hardware_disable_nolock(NULL
);
3954 static void kvm_resume(void)
3956 if (kvm_usage_count
) {
3957 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3958 hardware_enable_nolock(NULL
);
3962 static struct syscore_ops kvm_syscore_ops
= {
3963 .suspend
= kvm_suspend
,
3964 .resume
= kvm_resume
,
3968 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3970 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3973 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3975 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3977 if (vcpu
->preempted
)
3978 vcpu
->preempted
= false;
3980 kvm_arch_sched_in(vcpu
, cpu
);
3982 kvm_arch_vcpu_load(vcpu
, cpu
);
3985 static void kvm_sched_out(struct preempt_notifier
*pn
,
3986 struct task_struct
*next
)
3988 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3990 if (current
->state
== TASK_RUNNING
)
3991 vcpu
->preempted
= true;
3992 kvm_arch_vcpu_put(vcpu
);
3995 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3996 struct module
*module
)
4001 r
= kvm_arch_init(opaque
);
4006 * kvm_arch_init makes sure there's at most one caller
4007 * for architectures that support multiple implementations,
4008 * like intel and amd on x86.
4009 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4010 * conflicts in case kvm is already setup for another implementation.
4012 r
= kvm_irqfd_init();
4016 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4021 r
= kvm_arch_hardware_setup();
4025 for_each_online_cpu(cpu
) {
4026 smp_call_function_single(cpu
,
4027 kvm_arch_check_processor_compat
,
4033 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4034 kvm_starting_cpu
, kvm_dying_cpu
);
4037 register_reboot_notifier(&kvm_reboot_notifier
);
4039 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4041 vcpu_align
= __alignof__(struct kvm_vcpu
);
4042 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4043 SLAB_ACCOUNT
, NULL
);
4044 if (!kvm_vcpu_cache
) {
4049 r
= kvm_async_pf_init();
4053 kvm_chardev_ops
.owner
= module
;
4054 kvm_vm_fops
.owner
= module
;
4055 kvm_vcpu_fops
.owner
= module
;
4057 r
= misc_register(&kvm_dev
);
4059 pr_err("kvm: misc device register failed\n");
4063 register_syscore_ops(&kvm_syscore_ops
);
4065 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4066 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4068 r
= kvm_init_debug();
4070 pr_err("kvm: create debugfs files failed\n");
4074 r
= kvm_vfio_ops_init();
4080 unregister_syscore_ops(&kvm_syscore_ops
);
4081 misc_deregister(&kvm_dev
);
4083 kvm_async_pf_deinit();
4085 kmem_cache_destroy(kvm_vcpu_cache
);
4087 unregister_reboot_notifier(&kvm_reboot_notifier
);
4088 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4091 kvm_arch_hardware_unsetup();
4093 free_cpumask_var(cpus_hardware_enabled
);
4101 EXPORT_SYMBOL_GPL(kvm_init
);
4105 debugfs_remove_recursive(kvm_debugfs_dir
);
4106 misc_deregister(&kvm_dev
);
4107 kmem_cache_destroy(kvm_vcpu_cache
);
4108 kvm_async_pf_deinit();
4109 unregister_syscore_ops(&kvm_syscore_ops
);
4110 unregister_reboot_notifier(&kvm_reboot_notifier
);
4111 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4112 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4113 kvm_arch_hardware_unsetup();
4116 free_cpumask_var(cpus_hardware_enabled
);
4117 kvm_vfio_ops_exit();
4119 EXPORT_SYMBOL_GPL(kvm_exit
);