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
,
119 #define KVM_COMPAT(c) .compat_ioctl = (c)
121 static long kvm_no_compat_ioctl(struct file
*file
, unsigned int ioctl
,
122 unsigned long arg
) { return -EINVAL
; }
123 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
125 static int hardware_enable_all(void);
126 static void hardware_disable_all(void);
128 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
130 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
132 __visible
bool kvm_rebooting
;
133 EXPORT_SYMBOL_GPL(kvm_rebooting
);
135 static bool largepages_enabled
= true;
137 #define KVM_EVENT_CREATE_VM 0
138 #define KVM_EVENT_DESTROY_VM 1
139 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
140 static unsigned long long kvm_createvm_count
;
141 static unsigned long long kvm_active_vms
;
143 __weak
int kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
144 unsigned long start
, unsigned long end
, bool blockable
)
149 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
152 return PageReserved(pfn_to_page(pfn
));
158 * Switches to specified vcpu, until a matching vcpu_put()
160 void vcpu_load(struct kvm_vcpu
*vcpu
)
163 preempt_notifier_register(&vcpu
->preempt_notifier
);
164 kvm_arch_vcpu_load(vcpu
, cpu
);
167 EXPORT_SYMBOL_GPL(vcpu_load
);
169 void vcpu_put(struct kvm_vcpu
*vcpu
)
172 kvm_arch_vcpu_put(vcpu
);
173 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
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_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
213 unsigned long *vcpu_bitmap
, cpumask_var_t tmp
)
216 struct kvm_vcpu
*vcpu
;
221 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
222 if (vcpu_bitmap
&& !test_bit(i
, vcpu_bitmap
))
225 kvm_make_request(req
, vcpu
);
228 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
231 if (tmp
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
232 kvm_request_needs_ipi(vcpu
, req
))
233 __cpumask_set_cpu(cpu
, tmp
);
236 called
= kvm_kick_many_cpus(tmp
, !!(req
& KVM_REQUEST_WAIT
));
242 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
247 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
249 called
= kvm_make_vcpus_request_mask(kvm
, req
, NULL
, cpus
);
251 free_cpumask_var(cpus
);
255 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
256 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
259 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
260 * kvm_make_all_cpus_request.
262 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
265 * We want to publish modifications to the page tables before reading
266 * mode. Pairs with a memory barrier in arch-specific code.
267 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
268 * and smp_mb in walk_shadow_page_lockless_begin/end.
269 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
271 * There is already an smp_mb__after_atomic() before
272 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
275 if (!kvm_arch_flush_remote_tlb(kvm
)
276 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
277 ++kvm
->stat
.remote_tlb_flush
;
278 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
280 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
283 void kvm_reload_remote_mmus(struct kvm
*kvm
)
285 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
288 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
293 mutex_init(&vcpu
->mutex
);
298 init_swait_queue_head(&vcpu
->wq
);
299 kvm_async_pf_vcpu_init(vcpu
);
302 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
304 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
309 vcpu
->run
= page_address(page
);
311 kvm_vcpu_set_in_spin_loop(vcpu
, false);
312 kvm_vcpu_set_dy_eligible(vcpu
, false);
313 vcpu
->preempted
= false;
315 r
= kvm_arch_vcpu_init(vcpu
);
321 free_page((unsigned long)vcpu
->run
);
325 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
327 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
330 * no need for rcu_read_lock as VCPU_RUN is the only place that
331 * will change the vcpu->pid pointer and on uninit all file
332 * descriptors are already gone.
334 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
335 kvm_arch_vcpu_uninit(vcpu
);
336 free_page((unsigned long)vcpu
->run
);
338 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
340 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
341 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
343 return container_of(mn
, struct kvm
, mmu_notifier
);
346 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
347 struct mm_struct
*mm
,
348 unsigned long address
,
351 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
354 idx
= srcu_read_lock(&kvm
->srcu
);
355 spin_lock(&kvm
->mmu_lock
);
356 kvm
->mmu_notifier_seq
++;
357 kvm_set_spte_hva(kvm
, address
, pte
);
358 spin_unlock(&kvm
->mmu_lock
);
359 srcu_read_unlock(&kvm
->srcu
, idx
);
362 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
363 struct mm_struct
*mm
,
368 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
369 int need_tlb_flush
= 0, idx
;
372 idx
= srcu_read_lock(&kvm
->srcu
);
373 spin_lock(&kvm
->mmu_lock
);
375 * The count increase must become visible at unlock time as no
376 * spte can be established without taking the mmu_lock and
377 * count is also read inside the mmu_lock critical section.
379 kvm
->mmu_notifier_count
++;
380 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
381 need_tlb_flush
|= kvm
->tlbs_dirty
;
382 /* we've to flush the tlb before the pages can be freed */
384 kvm_flush_remote_tlbs(kvm
);
386 spin_unlock(&kvm
->mmu_lock
);
388 ret
= kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
, blockable
);
390 srcu_read_unlock(&kvm
->srcu
, idx
);
395 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
396 struct mm_struct
*mm
,
400 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
402 spin_lock(&kvm
->mmu_lock
);
404 * This sequence increase will notify the kvm page fault that
405 * the page that is going to be mapped in the spte could have
408 kvm
->mmu_notifier_seq
++;
411 * The above sequence increase must be visible before the
412 * below count decrease, which is ensured by the smp_wmb above
413 * in conjunction with the smp_rmb in mmu_notifier_retry().
415 kvm
->mmu_notifier_count
--;
416 spin_unlock(&kvm
->mmu_lock
);
418 BUG_ON(kvm
->mmu_notifier_count
< 0);
421 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
422 struct mm_struct
*mm
,
426 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
429 idx
= srcu_read_lock(&kvm
->srcu
);
430 spin_lock(&kvm
->mmu_lock
);
432 young
= kvm_age_hva(kvm
, start
, end
);
434 kvm_flush_remote_tlbs(kvm
);
436 spin_unlock(&kvm
->mmu_lock
);
437 srcu_read_unlock(&kvm
->srcu
, idx
);
442 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
443 struct mm_struct
*mm
,
447 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
450 idx
= srcu_read_lock(&kvm
->srcu
);
451 spin_lock(&kvm
->mmu_lock
);
453 * Even though we do not flush TLB, this will still adversely
454 * affect performance on pre-Haswell Intel EPT, where there is
455 * no EPT Access Bit to clear so that we have to tear down EPT
456 * tables instead. If we find this unacceptable, we can always
457 * add a parameter to kvm_age_hva so that it effectively doesn't
458 * do anything on clear_young.
460 * Also note that currently we never issue secondary TLB flushes
461 * from clear_young, leaving this job up to the regular system
462 * cadence. If we find this inaccurate, we might come up with a
463 * more sophisticated heuristic later.
465 young
= kvm_age_hva(kvm
, start
, end
);
466 spin_unlock(&kvm
->mmu_lock
);
467 srcu_read_unlock(&kvm
->srcu
, idx
);
472 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
473 struct mm_struct
*mm
,
474 unsigned long address
)
476 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
479 idx
= srcu_read_lock(&kvm
->srcu
);
480 spin_lock(&kvm
->mmu_lock
);
481 young
= kvm_test_age_hva(kvm
, address
);
482 spin_unlock(&kvm
->mmu_lock
);
483 srcu_read_unlock(&kvm
->srcu
, idx
);
488 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
489 struct mm_struct
*mm
)
491 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
494 idx
= srcu_read_lock(&kvm
->srcu
);
495 kvm_arch_flush_shadow_all(kvm
);
496 srcu_read_unlock(&kvm
->srcu
, idx
);
499 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
500 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
501 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
502 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
503 .clear_young
= kvm_mmu_notifier_clear_young
,
504 .test_young
= kvm_mmu_notifier_test_young
,
505 .change_pte
= kvm_mmu_notifier_change_pte
,
506 .release
= kvm_mmu_notifier_release
,
509 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
511 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
512 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
515 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
517 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
522 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
524 static struct kvm_memslots
*kvm_alloc_memslots(void)
527 struct kvm_memslots
*slots
;
529 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
533 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
534 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
539 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
541 if (!memslot
->dirty_bitmap
)
544 kvfree(memslot
->dirty_bitmap
);
545 memslot
->dirty_bitmap
= NULL
;
549 * Free any memory in @free but not in @dont.
551 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
552 struct kvm_memory_slot
*dont
)
554 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
555 kvm_destroy_dirty_bitmap(free
);
557 kvm_arch_free_memslot(kvm
, free
, dont
);
562 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
564 struct kvm_memory_slot
*memslot
;
569 kvm_for_each_memslot(memslot
, slots
)
570 kvm_free_memslot(kvm
, memslot
, NULL
);
575 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
579 if (!kvm
->debugfs_dentry
)
582 debugfs_remove_recursive(kvm
->debugfs_dentry
);
584 if (kvm
->debugfs_stat_data
) {
585 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
586 kfree(kvm
->debugfs_stat_data
[i
]);
587 kfree(kvm
->debugfs_stat_data
);
591 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
593 char dir_name
[ITOA_MAX_LEN
* 2];
594 struct kvm_stat_data
*stat_data
;
595 struct kvm_stats_debugfs_item
*p
;
597 if (!debugfs_initialized())
600 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
601 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
603 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
604 sizeof(*kvm
->debugfs_stat_data
),
606 if (!kvm
->debugfs_stat_data
)
609 for (p
= debugfs_entries
; p
->name
; p
++) {
610 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
614 stat_data
->kvm
= kvm
;
615 stat_data
->offset
= p
->offset
;
616 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
617 debugfs_create_file(p
->name
, 0644, kvm
->debugfs_dentry
,
618 stat_data
, stat_fops_per_vm
[p
->kind
]);
623 static struct kvm
*kvm_create_vm(unsigned long type
)
626 struct kvm
*kvm
= kvm_arch_alloc_vm();
629 return ERR_PTR(-ENOMEM
);
631 spin_lock_init(&kvm
->mmu_lock
);
633 kvm
->mm
= current
->mm
;
634 kvm_eventfd_init(kvm
);
635 mutex_init(&kvm
->lock
);
636 mutex_init(&kvm
->irq_lock
);
637 mutex_init(&kvm
->slots_lock
);
638 refcount_set(&kvm
->users_count
, 1);
639 INIT_LIST_HEAD(&kvm
->devices
);
641 r
= kvm_arch_init_vm(kvm
, type
);
643 goto out_err_no_disable
;
645 r
= hardware_enable_all();
647 goto out_err_no_disable
;
649 #ifdef CONFIG_HAVE_KVM_IRQFD
650 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
653 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
656 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
657 struct kvm_memslots
*slots
= kvm_alloc_memslots();
659 goto out_err_no_srcu
;
661 * Generations must be different for each address space.
662 * Init kvm generation close to the maximum to easily test the
663 * code of handling generation number wrap-around.
665 slots
->generation
= i
* 2 - 150;
666 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
669 if (init_srcu_struct(&kvm
->srcu
))
670 goto out_err_no_srcu
;
671 if (init_srcu_struct(&kvm
->irq_srcu
))
672 goto out_err_no_irq_srcu
;
673 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
674 rcu_assign_pointer(kvm
->buses
[i
],
675 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
680 r
= kvm_init_mmu_notifier(kvm
);
684 spin_lock(&kvm_lock
);
685 list_add(&kvm
->vm_list
, &vm_list
);
686 spin_unlock(&kvm_lock
);
688 preempt_notifier_inc();
693 cleanup_srcu_struct(&kvm
->irq_srcu
);
695 cleanup_srcu_struct(&kvm
->srcu
);
697 hardware_disable_all();
699 refcount_set(&kvm
->users_count
, 0);
700 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
701 kfree(kvm_get_bus(kvm
, i
));
702 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
703 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
704 kvm_arch_free_vm(kvm
);
709 static void kvm_destroy_devices(struct kvm
*kvm
)
711 struct kvm_device
*dev
, *tmp
;
714 * We do not need to take the kvm->lock here, because nobody else
715 * has a reference to the struct kvm at this point and therefore
716 * cannot access the devices list anyhow.
718 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
719 list_del(&dev
->vm_node
);
720 dev
->ops
->destroy(dev
);
724 static void kvm_destroy_vm(struct kvm
*kvm
)
727 struct mm_struct
*mm
= kvm
->mm
;
729 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
730 kvm_destroy_vm_debugfs(kvm
);
731 kvm_arch_sync_events(kvm
);
732 spin_lock(&kvm_lock
);
733 list_del(&kvm
->vm_list
);
734 spin_unlock(&kvm_lock
);
735 kvm_free_irq_routing(kvm
);
736 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
737 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
740 kvm_io_bus_destroy(bus
);
741 kvm
->buses
[i
] = NULL
;
743 kvm_coalesced_mmio_free(kvm
);
744 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
745 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
747 kvm_arch_flush_shadow_all(kvm
);
749 kvm_arch_destroy_vm(kvm
);
750 kvm_destroy_devices(kvm
);
751 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
752 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
753 cleanup_srcu_struct(&kvm
->irq_srcu
);
754 cleanup_srcu_struct(&kvm
->srcu
);
755 kvm_arch_free_vm(kvm
);
756 preempt_notifier_dec();
757 hardware_disable_all();
761 void kvm_get_kvm(struct kvm
*kvm
)
763 refcount_inc(&kvm
->users_count
);
765 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
767 void kvm_put_kvm(struct kvm
*kvm
)
769 if (refcount_dec_and_test(&kvm
->users_count
))
772 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
775 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
777 struct kvm
*kvm
= filp
->private_data
;
779 kvm_irqfd_release(kvm
);
786 * Allocation size is twice as large as the actual dirty bitmap size.
787 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
789 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
791 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
793 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
794 if (!memslot
->dirty_bitmap
)
801 * Insert memslot and re-sort memslots based on their GFN,
802 * so binary search could be used to lookup GFN.
803 * Sorting algorithm takes advantage of having initially
804 * sorted array and known changed memslot position.
806 static void update_memslots(struct kvm_memslots
*slots
,
807 struct kvm_memory_slot
*new,
808 enum kvm_mr_change change
)
811 int i
= slots
->id_to_index
[id
];
812 struct kvm_memory_slot
*mslots
= slots
->memslots
;
814 WARN_ON(mslots
[i
].id
!= id
);
818 WARN_ON(mslots
[i
].npages
|| !new->npages
);
822 WARN_ON(new->npages
|| !mslots
[i
].npages
);
828 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
829 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
830 if (!mslots
[i
+ 1].npages
)
832 mslots
[i
] = mslots
[i
+ 1];
833 slots
->id_to_index
[mslots
[i
].id
] = i
;
838 * The ">=" is needed when creating a slot with base_gfn == 0,
839 * so that it moves before all those with base_gfn == npages == 0.
841 * On the other hand, if new->npages is zero, the above loop has
842 * already left i pointing to the beginning of the empty part of
843 * mslots, and the ">=" would move the hole backwards in this
844 * case---which is wrong. So skip the loop when deleting a slot.
848 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
849 mslots
[i
] = mslots
[i
- 1];
850 slots
->id_to_index
[mslots
[i
].id
] = i
;
854 WARN_ON_ONCE(i
!= slots
->used_slots
);
857 slots
->id_to_index
[mslots
[i
].id
] = i
;
860 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
862 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
864 #ifdef __KVM_HAVE_READONLY_MEM
865 valid_flags
|= KVM_MEM_READONLY
;
868 if (mem
->flags
& ~valid_flags
)
874 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
875 int as_id
, struct kvm_memslots
*slots
)
877 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
880 * Set the low bit in the generation, which disables SPTE caching
881 * until the end of synchronize_srcu_expedited.
883 WARN_ON(old_memslots
->generation
& 1);
884 slots
->generation
= old_memslots
->generation
+ 1;
886 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
887 synchronize_srcu_expedited(&kvm
->srcu
);
890 * Increment the new memslot generation a second time. This prevents
891 * vm exits that race with memslot updates from caching a memslot
892 * generation that will (potentially) be valid forever.
894 * Generations must be unique even across address spaces. We do not need
895 * a global counter for that, instead the generation space is evenly split
896 * across address spaces. For example, with two address spaces, address
897 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
898 * use generations 2, 6, 10, 14, ...
900 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
902 kvm_arch_memslots_updated(kvm
, slots
);
908 * Allocate some memory and give it an address in the guest physical address
911 * Discontiguous memory is allowed, mostly for framebuffers.
913 * Must be called holding kvm->slots_lock for write.
915 int __kvm_set_memory_region(struct kvm
*kvm
,
916 const struct kvm_userspace_memory_region
*mem
)
920 unsigned long npages
;
921 struct kvm_memory_slot
*slot
;
922 struct kvm_memory_slot old
, new;
923 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
925 enum kvm_mr_change change
;
927 r
= check_memory_region_flags(mem
);
932 as_id
= mem
->slot
>> 16;
935 /* General sanity checks */
936 if (mem
->memory_size
& (PAGE_SIZE
- 1))
938 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
940 /* We can read the guest memory with __xxx_user() later on. */
941 if ((id
< KVM_USER_MEM_SLOTS
) &&
942 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
943 !access_ok(VERIFY_WRITE
,
944 (void __user
*)(unsigned long)mem
->userspace_addr
,
947 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
949 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
952 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
953 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
954 npages
= mem
->memory_size
>> PAGE_SHIFT
;
956 if (npages
> KVM_MEM_MAX_NR_PAGES
)
962 new.base_gfn
= base_gfn
;
964 new.flags
= mem
->flags
;
968 change
= KVM_MR_CREATE
;
969 else { /* Modify an existing slot. */
970 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
971 (npages
!= old
.npages
) ||
972 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
975 if (base_gfn
!= old
.base_gfn
)
976 change
= KVM_MR_MOVE
;
977 else if (new.flags
!= old
.flags
)
978 change
= KVM_MR_FLAGS_ONLY
;
979 else { /* Nothing to change. */
988 change
= KVM_MR_DELETE
;
993 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
994 /* Check for overlaps */
996 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
999 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
1000 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
1005 /* Free page dirty bitmap if unneeded */
1006 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1007 new.dirty_bitmap
= NULL
;
1010 if (change
== KVM_MR_CREATE
) {
1011 new.userspace_addr
= mem
->userspace_addr
;
1013 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1017 /* Allocate page dirty bitmap if needed */
1018 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1019 if (kvm_create_dirty_bitmap(&new) < 0)
1023 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1026 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1028 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1029 slot
= id_to_memslot(slots
, id
);
1030 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1032 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1034 /* From this point no new shadow pages pointing to a deleted,
1035 * or moved, memslot will be created.
1037 * validation of sp->gfn happens in:
1038 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1039 * - kvm_is_visible_gfn (mmu_check_roots)
1041 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1044 * We can re-use the old_memslots from above, the only difference
1045 * from the currently installed memslots is the invalid flag. This
1046 * will get overwritten by update_memslots anyway.
1048 slots
= old_memslots
;
1051 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1055 /* actual memory is freed via old in kvm_free_memslot below */
1056 if (change
== KVM_MR_DELETE
) {
1057 new.dirty_bitmap
= NULL
;
1058 memset(&new.arch
, 0, sizeof(new.arch
));
1061 update_memslots(slots
, &new, change
);
1062 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1064 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1066 kvm_free_memslot(kvm
, &old
, &new);
1067 kvfree(old_memslots
);
1073 kvm_free_memslot(kvm
, &new, &old
);
1077 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1079 int kvm_set_memory_region(struct kvm
*kvm
,
1080 const struct kvm_userspace_memory_region
*mem
)
1084 mutex_lock(&kvm
->slots_lock
);
1085 r
= __kvm_set_memory_region(kvm
, mem
);
1086 mutex_unlock(&kvm
->slots_lock
);
1089 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1091 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1092 struct kvm_userspace_memory_region
*mem
)
1094 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1097 return kvm_set_memory_region(kvm
, mem
);
1100 int kvm_get_dirty_log(struct kvm
*kvm
,
1101 struct kvm_dirty_log
*log
, int *is_dirty
)
1103 struct kvm_memslots
*slots
;
1104 struct kvm_memory_slot
*memslot
;
1107 unsigned long any
= 0;
1109 as_id
= log
->slot
>> 16;
1110 id
= (u16
)log
->slot
;
1111 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1114 slots
= __kvm_memslots(kvm
, as_id
);
1115 memslot
= id_to_memslot(slots
, id
);
1116 if (!memslot
->dirty_bitmap
)
1119 n
= kvm_dirty_bitmap_bytes(memslot
);
1121 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1122 any
= memslot
->dirty_bitmap
[i
];
1124 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1131 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1133 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1135 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1136 * are dirty write protect them for next write.
1137 * @kvm: pointer to kvm instance
1138 * @log: slot id and address to which we copy the log
1139 * @is_dirty: flag set if any page is dirty
1141 * We need to keep it in mind that VCPU threads can write to the bitmap
1142 * concurrently. So, to avoid losing track of dirty pages we keep the
1145 * 1. Take a snapshot of the bit and clear it if needed.
1146 * 2. Write protect the corresponding page.
1147 * 3. Copy the snapshot to the userspace.
1148 * 4. Upon return caller flushes TLB's if needed.
1150 * Between 2 and 4, the guest may write to the page using the remaining TLB
1151 * entry. This is not a problem because the page is reported dirty using
1152 * the snapshot taken before and step 4 ensures that writes done after
1153 * exiting to userspace will be logged for the next call.
1156 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1157 struct kvm_dirty_log
*log
, bool *flush
)
1159 struct kvm_memslots
*slots
;
1160 struct kvm_memory_slot
*memslot
;
1163 unsigned long *dirty_bitmap
;
1164 unsigned long *dirty_bitmap_buffer
;
1166 as_id
= log
->slot
>> 16;
1167 id
= (u16
)log
->slot
;
1168 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1171 slots
= __kvm_memslots(kvm
, as_id
);
1172 memslot
= id_to_memslot(slots
, id
);
1174 dirty_bitmap
= memslot
->dirty_bitmap
;
1178 n
= kvm_dirty_bitmap_bytes(memslot
);
1180 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1181 memset(dirty_bitmap_buffer
, 0, n
);
1183 spin_lock(&kvm
->mmu_lock
);
1185 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1189 if (!dirty_bitmap
[i
])
1194 mask
= xchg(&dirty_bitmap
[i
], 0);
1195 dirty_bitmap_buffer
[i
] = mask
;
1198 offset
= i
* BITS_PER_LONG
;
1199 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1204 spin_unlock(&kvm
->mmu_lock
);
1205 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1209 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1212 bool kvm_largepages_enabled(void)
1214 return largepages_enabled
;
1217 void kvm_disable_largepages(void)
1219 largepages_enabled
= false;
1221 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1223 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1225 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1227 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1229 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1231 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1234 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1236 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1238 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1239 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1244 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1246 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1248 struct vm_area_struct
*vma
;
1249 unsigned long addr
, size
;
1253 addr
= gfn_to_hva(kvm
, gfn
);
1254 if (kvm_is_error_hva(addr
))
1257 down_read(¤t
->mm
->mmap_sem
);
1258 vma
= find_vma(current
->mm
, addr
);
1262 size
= vma_kernel_pagesize(vma
);
1265 up_read(¤t
->mm
->mmap_sem
);
1270 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1272 return slot
->flags
& KVM_MEM_READONLY
;
1275 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1276 gfn_t
*nr_pages
, bool write
)
1278 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1279 return KVM_HVA_ERR_BAD
;
1281 if (memslot_is_readonly(slot
) && write
)
1282 return KVM_HVA_ERR_RO_BAD
;
1285 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1287 return __gfn_to_hva_memslot(slot
, gfn
);
1290 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1293 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1296 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1299 return gfn_to_hva_many(slot
, gfn
, NULL
);
1301 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1303 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1305 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1307 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1309 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1311 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1313 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1316 * Return the hva of a @gfn and the R/W attribute if possible.
1318 * @slot: the kvm_memory_slot which contains @gfn
1319 * @gfn: the gfn to be translated
1320 * @writable: used to return the read/write attribute of the @slot if the hva
1321 * is valid and @writable is not NULL
1323 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1324 gfn_t gfn
, bool *writable
)
1326 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1328 if (!kvm_is_error_hva(hva
) && writable
)
1329 *writable
= !memslot_is_readonly(slot
);
1334 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1336 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1338 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1341 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1343 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1345 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1348 static inline int check_user_page_hwpoison(unsigned long addr
)
1350 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1352 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1353 return rc
== -EHWPOISON
;
1357 * The fast path to get the writable pfn which will be stored in @pfn,
1358 * true indicates success, otherwise false is returned. It's also the
1359 * only part that runs if we can are in atomic context.
1361 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
1362 bool *writable
, kvm_pfn_t
*pfn
)
1364 struct page
*page
[1];
1368 * Fast pin a writable pfn only if it is a write fault request
1369 * or the caller allows to map a writable pfn for a read fault
1372 if (!(write_fault
|| writable
))
1375 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1377 *pfn
= page_to_pfn(page
[0]);
1388 * The slow path to get the pfn of the specified host virtual address,
1389 * 1 indicates success, -errno is returned if error is detected.
1391 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1392 bool *writable
, kvm_pfn_t
*pfn
)
1394 unsigned int flags
= FOLL_HWPOISON
;
1401 *writable
= write_fault
;
1404 flags
|= FOLL_WRITE
;
1406 flags
|= FOLL_NOWAIT
;
1408 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
1412 /* map read fault as writable if possible */
1413 if (unlikely(!write_fault
) && writable
) {
1416 if (__get_user_pages_fast(addr
, 1, 1, &wpage
) == 1) {
1422 *pfn
= page_to_pfn(page
);
1426 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1428 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1431 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1437 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1438 unsigned long addr
, bool *async
,
1439 bool write_fault
, bool *writable
,
1445 r
= follow_pfn(vma
, addr
, &pfn
);
1448 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1449 * not call the fault handler, so do it here.
1451 bool unlocked
= false;
1452 r
= fixup_user_fault(current
, current
->mm
, addr
,
1453 (write_fault
? FAULT_FLAG_WRITE
: 0),
1460 r
= follow_pfn(vma
, addr
, &pfn
);
1470 * Get a reference here because callers of *hva_to_pfn* and
1471 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1472 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1473 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1474 * simply do nothing for reserved pfns.
1476 * Whoever called remap_pfn_range is also going to call e.g.
1477 * unmap_mapping_range before the underlying pages are freed,
1478 * causing a call to our MMU notifier.
1487 * Pin guest page in memory and return its pfn.
1488 * @addr: host virtual address which maps memory to the guest
1489 * @atomic: whether this function can sleep
1490 * @async: whether this function need to wait IO complete if the
1491 * host page is not in the memory
1492 * @write_fault: whether we should get a writable host page
1493 * @writable: whether it allows to map a writable host page for !@write_fault
1495 * The function will map a writable host page for these two cases:
1496 * 1): @write_fault = true
1497 * 2): @write_fault = false && @writable, @writable will tell the caller
1498 * whether the mapping is writable.
1500 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1501 bool write_fault
, bool *writable
)
1503 struct vm_area_struct
*vma
;
1507 /* we can do it either atomically or asynchronously, not both */
1508 BUG_ON(atomic
&& async
);
1510 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
1514 return KVM_PFN_ERR_FAULT
;
1516 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1520 down_read(¤t
->mm
->mmap_sem
);
1521 if (npages
== -EHWPOISON
||
1522 (!async
&& check_user_page_hwpoison(addr
))) {
1523 pfn
= KVM_PFN_ERR_HWPOISON
;
1528 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1531 pfn
= KVM_PFN_ERR_FAULT
;
1532 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1533 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1537 pfn
= KVM_PFN_ERR_FAULT
;
1539 if (async
&& vma_is_valid(vma
, write_fault
))
1541 pfn
= KVM_PFN_ERR_FAULT
;
1544 up_read(¤t
->mm
->mmap_sem
);
1548 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1549 bool atomic
, bool *async
, bool write_fault
,
1552 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1554 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1557 return KVM_PFN_ERR_RO_FAULT
;
1560 if (kvm_is_error_hva(addr
)) {
1563 return KVM_PFN_NOSLOT
;
1566 /* Do not map writable pfn in the readonly memslot. */
1567 if (writable
&& memslot_is_readonly(slot
)) {
1572 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1575 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1577 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1580 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1581 write_fault
, writable
);
1583 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1585 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1587 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1589 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1591 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1593 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1595 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1597 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1599 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1601 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1603 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1605 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1607 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1609 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1611 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1613 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1615 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1617 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1619 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1621 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1622 struct page
**pages
, int nr_pages
)
1627 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1628 if (kvm_is_error_hva(addr
))
1631 if (entry
< nr_pages
)
1634 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1636 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1638 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1640 if (is_error_noslot_pfn(pfn
))
1641 return KVM_ERR_PTR_BAD_PAGE
;
1643 if (kvm_is_reserved_pfn(pfn
)) {
1645 return KVM_ERR_PTR_BAD_PAGE
;
1648 return pfn_to_page(pfn
);
1651 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1655 pfn
= gfn_to_pfn(kvm
, gfn
);
1657 return kvm_pfn_to_page(pfn
);
1659 EXPORT_SYMBOL_GPL(gfn_to_page
);
1661 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1665 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1667 return kvm_pfn_to_page(pfn
);
1669 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1671 void kvm_release_page_clean(struct page
*page
)
1673 WARN_ON(is_error_page(page
));
1675 kvm_release_pfn_clean(page_to_pfn(page
));
1677 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1679 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1681 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1682 put_page(pfn_to_page(pfn
));
1684 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1686 void kvm_release_page_dirty(struct page
*page
)
1688 WARN_ON(is_error_page(page
));
1690 kvm_release_pfn_dirty(page_to_pfn(page
));
1692 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1694 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1696 kvm_set_pfn_dirty(pfn
);
1697 kvm_release_pfn_clean(pfn
);
1699 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1701 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1703 if (!kvm_is_reserved_pfn(pfn
)) {
1704 struct page
*page
= pfn_to_page(pfn
);
1706 if (!PageReserved(page
))
1710 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1712 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1714 if (!kvm_is_reserved_pfn(pfn
))
1715 mark_page_accessed(pfn_to_page(pfn
));
1717 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1719 void kvm_get_pfn(kvm_pfn_t pfn
)
1721 if (!kvm_is_reserved_pfn(pfn
))
1722 get_page(pfn_to_page(pfn
));
1724 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1726 static int next_segment(unsigned long len
, int offset
)
1728 if (len
> PAGE_SIZE
- offset
)
1729 return PAGE_SIZE
- offset
;
1734 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1735 void *data
, int offset
, int len
)
1740 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1741 if (kvm_is_error_hva(addr
))
1743 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1749 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1752 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1754 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1756 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1758 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1759 int offset
, int len
)
1761 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1763 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1765 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1767 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1769 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1771 int offset
= offset_in_page(gpa
);
1774 while ((seg
= next_segment(len
, offset
)) != 0) {
1775 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1785 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1787 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1789 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1791 int offset
= offset_in_page(gpa
);
1794 while ((seg
= next_segment(len
, offset
)) != 0) {
1795 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1805 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1807 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1808 void *data
, int offset
, unsigned long len
)
1813 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1814 if (kvm_is_error_hva(addr
))
1816 pagefault_disable();
1817 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1824 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1827 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1828 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1829 int offset
= offset_in_page(gpa
);
1831 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1833 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1835 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1836 void *data
, unsigned long len
)
1838 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1839 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1840 int offset
= offset_in_page(gpa
);
1842 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1844 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1846 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1847 const void *data
, int offset
, int len
)
1852 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1853 if (kvm_is_error_hva(addr
))
1855 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1858 mark_page_dirty_in_slot(memslot
, gfn
);
1862 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1863 const void *data
, int offset
, int len
)
1865 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1867 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1869 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1871 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1872 const void *data
, int offset
, int len
)
1874 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1876 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1878 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1880 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1883 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1885 int offset
= offset_in_page(gpa
);
1888 while ((seg
= next_segment(len
, offset
)) != 0) {
1889 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1899 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1901 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1904 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1906 int offset
= offset_in_page(gpa
);
1909 while ((seg
= next_segment(len
, offset
)) != 0) {
1910 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1920 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1922 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1923 struct gfn_to_hva_cache
*ghc
,
1924 gpa_t gpa
, unsigned long len
)
1926 int offset
= offset_in_page(gpa
);
1927 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1928 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1929 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1930 gfn_t nr_pages_avail
;
1933 ghc
->generation
= slots
->generation
;
1935 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1936 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1937 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1941 * If the requested region crosses two memslots, we still
1942 * verify that the entire region is valid here.
1944 while (start_gfn
<= end_gfn
) {
1946 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1947 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1949 if (kvm_is_error_hva(ghc
->hva
))
1951 start_gfn
+= nr_pages_avail
;
1953 /* Use the slow path for cross page reads and writes. */
1954 ghc
->memslot
= NULL
;
1959 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1960 gpa_t gpa
, unsigned long len
)
1962 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1963 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1965 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1967 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1968 void *data
, int offset
, unsigned long len
)
1970 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1972 gpa_t gpa
= ghc
->gpa
+ offset
;
1974 BUG_ON(len
+ offset
> ghc
->len
);
1976 if (slots
->generation
!= ghc
->generation
)
1977 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1979 if (unlikely(!ghc
->memslot
))
1980 return kvm_write_guest(kvm
, gpa
, data
, len
);
1982 if (kvm_is_error_hva(ghc
->hva
))
1985 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
1988 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
1992 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
1994 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1995 void *data
, unsigned long len
)
1997 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
1999 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2001 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2002 void *data
, unsigned long len
)
2004 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2007 BUG_ON(len
> ghc
->len
);
2009 if (slots
->generation
!= ghc
->generation
)
2010 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2012 if (unlikely(!ghc
->memslot
))
2013 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2015 if (kvm_is_error_hva(ghc
->hva
))
2018 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2024 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2026 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2028 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2030 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2032 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2034 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2036 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2038 int offset
= offset_in_page(gpa
);
2041 while ((seg
= next_segment(len
, offset
)) != 0) {
2042 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2051 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2053 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2056 if (memslot
&& memslot
->dirty_bitmap
) {
2057 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2059 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2063 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2065 struct kvm_memory_slot
*memslot
;
2067 memslot
= gfn_to_memslot(kvm
, gfn
);
2068 mark_page_dirty_in_slot(memslot
, gfn
);
2070 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2072 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2074 struct kvm_memory_slot
*memslot
;
2076 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2077 mark_page_dirty_in_slot(memslot
, gfn
);
2079 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2081 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2083 if (!vcpu
->sigset_active
)
2087 * This does a lockless modification of ->real_blocked, which is fine
2088 * because, only current can change ->real_blocked and all readers of
2089 * ->real_blocked don't care as long ->real_blocked is always a subset
2092 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2095 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2097 if (!vcpu
->sigset_active
)
2100 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2101 sigemptyset(¤t
->real_blocked
);
2104 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2106 unsigned int old
, val
, grow
;
2108 old
= val
= vcpu
->halt_poll_ns
;
2109 grow
= READ_ONCE(halt_poll_ns_grow
);
2111 if (val
== 0 && grow
)
2116 if (val
> halt_poll_ns
)
2119 vcpu
->halt_poll_ns
= val
;
2120 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2123 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2125 unsigned int old
, val
, shrink
;
2127 old
= val
= vcpu
->halt_poll_ns
;
2128 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2134 vcpu
->halt_poll_ns
= val
;
2135 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2138 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2141 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2143 if (kvm_arch_vcpu_runnable(vcpu
)) {
2144 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2147 if (kvm_cpu_has_pending_timer(vcpu
))
2149 if (signal_pending(current
))
2154 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2159 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2161 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2164 DECLARE_SWAITQUEUE(wait
);
2165 bool waited
= false;
2168 start
= cur
= ktime_get();
2169 if (vcpu
->halt_poll_ns
) {
2170 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2172 ++vcpu
->stat
.halt_attempted_poll
;
2175 * This sets KVM_REQ_UNHALT if an interrupt
2178 if (kvm_vcpu_check_block(vcpu
) < 0) {
2179 ++vcpu
->stat
.halt_successful_poll
;
2180 if (!vcpu_valid_wakeup(vcpu
))
2181 ++vcpu
->stat
.halt_poll_invalid
;
2185 } while (single_task_running() && ktime_before(cur
, stop
));
2188 kvm_arch_vcpu_blocking(vcpu
);
2191 prepare_to_swait_exclusive(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2193 if (kvm_vcpu_check_block(vcpu
) < 0)
2200 finish_swait(&vcpu
->wq
, &wait
);
2203 kvm_arch_vcpu_unblocking(vcpu
);
2205 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2207 if (!vcpu_valid_wakeup(vcpu
))
2208 shrink_halt_poll_ns(vcpu
);
2209 else if (halt_poll_ns
) {
2210 if (block_ns
<= vcpu
->halt_poll_ns
)
2212 /* we had a long block, shrink polling */
2213 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2214 shrink_halt_poll_ns(vcpu
);
2215 /* we had a short halt and our poll time is too small */
2216 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2217 block_ns
< halt_poll_ns
)
2218 grow_halt_poll_ns(vcpu
);
2220 vcpu
->halt_poll_ns
= 0;
2222 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2223 kvm_arch_vcpu_block_finish(vcpu
);
2225 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2227 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2229 struct swait_queue_head
*wqp
;
2231 wqp
= kvm_arch_vcpu_wq(vcpu
);
2232 if (swq_has_sleeper(wqp
)) {
2234 ++vcpu
->stat
.halt_wakeup
;
2240 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2244 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2246 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2249 int cpu
= vcpu
->cpu
;
2251 if (kvm_vcpu_wake_up(vcpu
))
2255 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2256 if (kvm_arch_vcpu_should_kick(vcpu
))
2257 smp_send_reschedule(cpu
);
2260 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2261 #endif /* !CONFIG_S390 */
2263 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2266 struct task_struct
*task
= NULL
;
2270 pid
= rcu_dereference(target
->pid
);
2272 task
= get_pid_task(pid
, PIDTYPE_PID
);
2276 ret
= yield_to(task
, 1);
2277 put_task_struct(task
);
2281 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2284 * Helper that checks whether a VCPU is eligible for directed yield.
2285 * Most eligible candidate to yield is decided by following heuristics:
2287 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2288 * (preempted lock holder), indicated by @in_spin_loop.
2289 * Set at the beiginning and cleared at the end of interception/PLE handler.
2291 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2292 * chance last time (mostly it has become eligible now since we have probably
2293 * yielded to lockholder in last iteration. This is done by toggling
2294 * @dy_eligible each time a VCPU checked for eligibility.)
2296 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2297 * to preempted lock-holder could result in wrong VCPU selection and CPU
2298 * burning. Giving priority for a potential lock-holder increases lock
2301 * Since algorithm is based on heuristics, accessing another VCPU data without
2302 * locking does not harm. It may result in trying to yield to same VCPU, fail
2303 * and continue with next VCPU and so on.
2305 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2307 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2310 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2311 vcpu
->spin_loop
.dy_eligible
;
2313 if (vcpu
->spin_loop
.in_spin_loop
)
2314 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2322 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2324 struct kvm
*kvm
= me
->kvm
;
2325 struct kvm_vcpu
*vcpu
;
2326 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2332 kvm_vcpu_set_in_spin_loop(me
, true);
2334 * We boost the priority of a VCPU that is runnable but not
2335 * currently running, because it got preempted by something
2336 * else and called schedule in __vcpu_run. Hopefully that
2337 * VCPU is holding the lock that we need and will release it.
2338 * We approximate round-robin by starting at the last boosted VCPU.
2340 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2341 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2342 if (!pass
&& i
<= last_boosted_vcpu
) {
2343 i
= last_boosted_vcpu
;
2345 } else if (pass
&& i
> last_boosted_vcpu
)
2347 if (!READ_ONCE(vcpu
->preempted
))
2351 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2353 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2355 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2358 yielded
= kvm_vcpu_yield_to(vcpu
);
2360 kvm
->last_boosted_vcpu
= i
;
2362 } else if (yielded
< 0) {
2369 kvm_vcpu_set_in_spin_loop(me
, false);
2371 /* Ensure vcpu is not eligible during next spinloop */
2372 kvm_vcpu_set_dy_eligible(me
, false);
2374 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2376 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
2378 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2381 if (vmf
->pgoff
== 0)
2382 page
= virt_to_page(vcpu
->run
);
2384 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2385 page
= virt_to_page(vcpu
->arch
.pio_data
);
2387 #ifdef CONFIG_KVM_MMIO
2388 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2389 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2392 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2398 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2399 .fault
= kvm_vcpu_fault
,
2402 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2404 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2408 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2410 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2412 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2413 kvm_put_kvm(vcpu
->kvm
);
2417 static struct file_operations kvm_vcpu_fops
= {
2418 .release
= kvm_vcpu_release
,
2419 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2420 .mmap
= kvm_vcpu_mmap
,
2421 .llseek
= noop_llseek
,
2422 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
2426 * Allocates an inode for the vcpu.
2428 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2430 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
2432 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
2433 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2436 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2438 char dir_name
[ITOA_MAX_LEN
* 2];
2441 if (!kvm_arch_has_vcpu_debugfs())
2444 if (!debugfs_initialized())
2447 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2448 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2449 vcpu
->kvm
->debugfs_dentry
);
2450 if (!vcpu
->debugfs_dentry
)
2453 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2455 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2463 * Creates some virtual cpus. Good luck creating more than one.
2465 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2468 struct kvm_vcpu
*vcpu
;
2470 if (id
>= KVM_MAX_VCPU_ID
)
2473 mutex_lock(&kvm
->lock
);
2474 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2475 mutex_unlock(&kvm
->lock
);
2479 kvm
->created_vcpus
++;
2480 mutex_unlock(&kvm
->lock
);
2482 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2485 goto vcpu_decrement
;
2488 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2490 r
= kvm_arch_vcpu_setup(vcpu
);
2494 r
= kvm_create_vcpu_debugfs(vcpu
);
2498 mutex_lock(&kvm
->lock
);
2499 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2501 goto unlock_vcpu_destroy
;
2504 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2506 /* Now it's all set up, let userspace reach it */
2508 r
= create_vcpu_fd(vcpu
);
2511 goto unlock_vcpu_destroy
;
2514 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2517 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2518 * before kvm->online_vcpu's incremented value.
2521 atomic_inc(&kvm
->online_vcpus
);
2523 mutex_unlock(&kvm
->lock
);
2524 kvm_arch_vcpu_postcreate(vcpu
);
2527 unlock_vcpu_destroy
:
2528 mutex_unlock(&kvm
->lock
);
2529 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2531 kvm_arch_vcpu_destroy(vcpu
);
2533 mutex_lock(&kvm
->lock
);
2534 kvm
->created_vcpus
--;
2535 mutex_unlock(&kvm
->lock
);
2539 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2542 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2543 vcpu
->sigset_active
= 1;
2544 vcpu
->sigset
= *sigset
;
2546 vcpu
->sigset_active
= 0;
2550 static long kvm_vcpu_ioctl(struct file
*filp
,
2551 unsigned int ioctl
, unsigned long arg
)
2553 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2554 void __user
*argp
= (void __user
*)arg
;
2556 struct kvm_fpu
*fpu
= NULL
;
2557 struct kvm_sregs
*kvm_sregs
= NULL
;
2559 if (vcpu
->kvm
->mm
!= current
->mm
)
2562 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2566 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2567 * execution; mutex_lock() would break them.
2569 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
2570 if (r
!= -ENOIOCTLCMD
)
2573 if (mutex_lock_killable(&vcpu
->mutex
))
2581 oldpid
= rcu_access_pointer(vcpu
->pid
);
2582 if (unlikely(oldpid
!= task_pid(current
))) {
2583 /* The thread running this VCPU changed. */
2586 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
2590 newpid
= get_task_pid(current
, PIDTYPE_PID
);
2591 rcu_assign_pointer(vcpu
->pid
, newpid
);
2596 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2597 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2600 case KVM_GET_REGS
: {
2601 struct kvm_regs
*kvm_regs
;
2604 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2607 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2611 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2618 case KVM_SET_REGS
: {
2619 struct kvm_regs
*kvm_regs
;
2622 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2623 if (IS_ERR(kvm_regs
)) {
2624 r
= PTR_ERR(kvm_regs
);
2627 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2631 case KVM_GET_SREGS
: {
2632 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2636 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2640 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2645 case KVM_SET_SREGS
: {
2646 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2647 if (IS_ERR(kvm_sregs
)) {
2648 r
= PTR_ERR(kvm_sregs
);
2652 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2655 case KVM_GET_MP_STATE
: {
2656 struct kvm_mp_state mp_state
;
2658 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2662 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2667 case KVM_SET_MP_STATE
: {
2668 struct kvm_mp_state mp_state
;
2671 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2673 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2676 case KVM_TRANSLATE
: {
2677 struct kvm_translation tr
;
2680 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2682 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2686 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2691 case KVM_SET_GUEST_DEBUG
: {
2692 struct kvm_guest_debug dbg
;
2695 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2697 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2700 case KVM_SET_SIGNAL_MASK
: {
2701 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2702 struct kvm_signal_mask kvm_sigmask
;
2703 sigset_t sigset
, *p
;
2708 if (copy_from_user(&kvm_sigmask
, argp
,
2709 sizeof(kvm_sigmask
)))
2712 if (kvm_sigmask
.len
!= sizeof(sigset
))
2715 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2720 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2724 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2728 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2732 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2738 fpu
= memdup_user(argp
, sizeof(*fpu
));
2744 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2748 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2751 mutex_unlock(&vcpu
->mutex
);
2757 #ifdef CONFIG_KVM_COMPAT
2758 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2759 unsigned int ioctl
, unsigned long arg
)
2761 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2762 void __user
*argp
= compat_ptr(arg
);
2765 if (vcpu
->kvm
->mm
!= current
->mm
)
2769 case KVM_SET_SIGNAL_MASK
: {
2770 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2771 struct kvm_signal_mask kvm_sigmask
;
2776 if (copy_from_user(&kvm_sigmask
, argp
,
2777 sizeof(kvm_sigmask
)))
2780 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2783 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2785 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2787 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2791 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2799 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2800 int (*accessor
)(struct kvm_device
*dev
,
2801 struct kvm_device_attr
*attr
),
2804 struct kvm_device_attr attr
;
2809 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2812 return accessor(dev
, &attr
);
2815 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2818 struct kvm_device
*dev
= filp
->private_data
;
2821 case KVM_SET_DEVICE_ATTR
:
2822 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2823 case KVM_GET_DEVICE_ATTR
:
2824 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2825 case KVM_HAS_DEVICE_ATTR
:
2826 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2828 if (dev
->ops
->ioctl
)
2829 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2835 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2837 struct kvm_device
*dev
= filp
->private_data
;
2838 struct kvm
*kvm
= dev
->kvm
;
2844 static const struct file_operations kvm_device_fops
= {
2845 .unlocked_ioctl
= kvm_device_ioctl
,
2846 .release
= kvm_device_release
,
2847 KVM_COMPAT(kvm_device_ioctl
),
2850 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2852 if (filp
->f_op
!= &kvm_device_fops
)
2855 return filp
->private_data
;
2858 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2859 #ifdef CONFIG_KVM_MPIC
2860 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2861 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2865 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2867 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2870 if (kvm_device_ops_table
[type
] != NULL
)
2873 kvm_device_ops_table
[type
] = ops
;
2877 void kvm_unregister_device_ops(u32 type
)
2879 if (kvm_device_ops_table
[type
] != NULL
)
2880 kvm_device_ops_table
[type
] = NULL
;
2883 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2884 struct kvm_create_device
*cd
)
2886 struct kvm_device_ops
*ops
= NULL
;
2887 struct kvm_device
*dev
;
2888 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2891 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2894 ops
= kvm_device_ops_table
[cd
->type
];
2901 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2908 mutex_lock(&kvm
->lock
);
2909 ret
= ops
->create(dev
, cd
->type
);
2911 mutex_unlock(&kvm
->lock
);
2915 list_add(&dev
->vm_node
, &kvm
->devices
);
2916 mutex_unlock(&kvm
->lock
);
2921 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2923 mutex_lock(&kvm
->lock
);
2924 list_del(&dev
->vm_node
);
2925 mutex_unlock(&kvm
->lock
);
2935 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2938 case KVM_CAP_USER_MEMORY
:
2939 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2940 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2941 case KVM_CAP_INTERNAL_ERROR_DATA
:
2942 #ifdef CONFIG_HAVE_KVM_MSI
2943 case KVM_CAP_SIGNAL_MSI
:
2945 #ifdef CONFIG_HAVE_KVM_IRQFD
2947 case KVM_CAP_IRQFD_RESAMPLE
:
2949 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2950 case KVM_CAP_CHECK_EXTENSION_VM
:
2951 case KVM_CAP_ENABLE_CAP_VM
:
2953 #ifdef CONFIG_KVM_MMIO
2954 case KVM_CAP_COALESCED_MMIO
:
2955 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2956 case KVM_CAP_COALESCED_PIO
:
2959 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2960 case KVM_CAP_IRQ_ROUTING
:
2961 return KVM_MAX_IRQ_ROUTES
;
2963 #if KVM_ADDRESS_SPACE_NUM > 1
2964 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2965 return KVM_ADDRESS_SPACE_NUM
;
2967 case KVM_CAP_MAX_VCPU_ID
:
2968 return KVM_MAX_VCPU_ID
;
2972 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2975 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
2976 struct kvm_enable_cap
*cap
)
2981 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
2982 struct kvm_enable_cap
*cap
)
2986 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
2990 static long kvm_vm_ioctl(struct file
*filp
,
2991 unsigned int ioctl
, unsigned long arg
)
2993 struct kvm
*kvm
= filp
->private_data
;
2994 void __user
*argp
= (void __user
*)arg
;
2997 if (kvm
->mm
!= current
->mm
)
3000 case KVM_CREATE_VCPU
:
3001 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3003 case KVM_ENABLE_CAP
: {
3004 struct kvm_enable_cap cap
;
3007 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3009 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
3012 case KVM_SET_USER_MEMORY_REGION
: {
3013 struct kvm_userspace_memory_region kvm_userspace_mem
;
3016 if (copy_from_user(&kvm_userspace_mem
, argp
,
3017 sizeof(kvm_userspace_mem
)))
3020 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3023 case KVM_GET_DIRTY_LOG
: {
3024 struct kvm_dirty_log log
;
3027 if (copy_from_user(&log
, argp
, sizeof(log
)))
3029 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3032 #ifdef CONFIG_KVM_MMIO
3033 case KVM_REGISTER_COALESCED_MMIO
: {
3034 struct kvm_coalesced_mmio_zone zone
;
3037 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3039 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3042 case KVM_UNREGISTER_COALESCED_MMIO
: {
3043 struct kvm_coalesced_mmio_zone zone
;
3046 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3048 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3053 struct kvm_irqfd data
;
3056 if (copy_from_user(&data
, argp
, sizeof(data
)))
3058 r
= kvm_irqfd(kvm
, &data
);
3061 case KVM_IOEVENTFD
: {
3062 struct kvm_ioeventfd data
;
3065 if (copy_from_user(&data
, argp
, sizeof(data
)))
3067 r
= kvm_ioeventfd(kvm
, &data
);
3070 #ifdef CONFIG_HAVE_KVM_MSI
3071 case KVM_SIGNAL_MSI
: {
3075 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3077 r
= kvm_send_userspace_msi(kvm
, &msi
);
3081 #ifdef __KVM_HAVE_IRQ_LINE
3082 case KVM_IRQ_LINE_STATUS
:
3083 case KVM_IRQ_LINE
: {
3084 struct kvm_irq_level irq_event
;
3087 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3090 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3091 ioctl
== KVM_IRQ_LINE_STATUS
);
3096 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3097 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3105 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3106 case KVM_SET_GSI_ROUTING
: {
3107 struct kvm_irq_routing routing
;
3108 struct kvm_irq_routing __user
*urouting
;
3109 struct kvm_irq_routing_entry
*entries
= NULL
;
3112 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3115 if (!kvm_arch_can_set_irq_routing(kvm
))
3117 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3123 entries
= vmalloc(array_size(sizeof(*entries
),
3129 if (copy_from_user(entries
, urouting
->entries
,
3130 routing
.nr
* sizeof(*entries
)))
3131 goto out_free_irq_routing
;
3133 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3135 out_free_irq_routing
:
3139 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3140 case KVM_CREATE_DEVICE
: {
3141 struct kvm_create_device cd
;
3144 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3147 r
= kvm_ioctl_create_device(kvm
, &cd
);
3152 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3158 case KVM_CHECK_EXTENSION
:
3159 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3162 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3168 #ifdef CONFIG_KVM_COMPAT
3169 struct compat_kvm_dirty_log
{
3173 compat_uptr_t dirty_bitmap
; /* one bit per page */
3178 static long kvm_vm_compat_ioctl(struct file
*filp
,
3179 unsigned int ioctl
, unsigned long arg
)
3181 struct kvm
*kvm
= filp
->private_data
;
3184 if (kvm
->mm
!= current
->mm
)
3187 case KVM_GET_DIRTY_LOG
: {
3188 struct compat_kvm_dirty_log compat_log
;
3189 struct kvm_dirty_log log
;
3191 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3192 sizeof(compat_log
)))
3194 log
.slot
= compat_log
.slot
;
3195 log
.padding1
= compat_log
.padding1
;
3196 log
.padding2
= compat_log
.padding2
;
3197 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3199 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3203 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3209 static struct file_operations kvm_vm_fops
= {
3210 .release
= kvm_vm_release
,
3211 .unlocked_ioctl
= kvm_vm_ioctl
,
3212 .llseek
= noop_llseek
,
3213 KVM_COMPAT(kvm_vm_compat_ioctl
),
3216 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3222 kvm
= kvm_create_vm(type
);
3224 return PTR_ERR(kvm
);
3225 #ifdef CONFIG_KVM_MMIO
3226 r
= kvm_coalesced_mmio_init(kvm
);
3230 r
= get_unused_fd_flags(O_CLOEXEC
);
3234 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3242 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3243 * already set, with ->release() being kvm_vm_release(). In error
3244 * cases it will be called by the final fput(file) and will take
3245 * care of doing kvm_put_kvm(kvm).
3247 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3252 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3254 fd_install(r
, file
);
3262 static long kvm_dev_ioctl(struct file
*filp
,
3263 unsigned int ioctl
, unsigned long arg
)
3268 case KVM_GET_API_VERSION
:
3271 r
= KVM_API_VERSION
;
3274 r
= kvm_dev_ioctl_create_vm(arg
);
3276 case KVM_CHECK_EXTENSION
:
3277 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3279 case KVM_GET_VCPU_MMAP_SIZE
:
3282 r
= PAGE_SIZE
; /* struct kvm_run */
3284 r
+= PAGE_SIZE
; /* pio data page */
3286 #ifdef CONFIG_KVM_MMIO
3287 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3290 case KVM_TRACE_ENABLE
:
3291 case KVM_TRACE_PAUSE
:
3292 case KVM_TRACE_DISABLE
:
3296 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3302 static struct file_operations kvm_chardev_ops
= {
3303 .unlocked_ioctl
= kvm_dev_ioctl
,
3304 .llseek
= noop_llseek
,
3305 KVM_COMPAT(kvm_dev_ioctl
),
3308 static struct miscdevice kvm_dev
= {
3314 static void hardware_enable_nolock(void *junk
)
3316 int cpu
= raw_smp_processor_id();
3319 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3322 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3324 r
= kvm_arch_hardware_enable();
3327 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3328 atomic_inc(&hardware_enable_failed
);
3329 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3333 static int kvm_starting_cpu(unsigned int cpu
)
3335 raw_spin_lock(&kvm_count_lock
);
3336 if (kvm_usage_count
)
3337 hardware_enable_nolock(NULL
);
3338 raw_spin_unlock(&kvm_count_lock
);
3342 static void hardware_disable_nolock(void *junk
)
3344 int cpu
= raw_smp_processor_id();
3346 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3348 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3349 kvm_arch_hardware_disable();
3352 static int kvm_dying_cpu(unsigned int cpu
)
3354 raw_spin_lock(&kvm_count_lock
);
3355 if (kvm_usage_count
)
3356 hardware_disable_nolock(NULL
);
3357 raw_spin_unlock(&kvm_count_lock
);
3361 static void hardware_disable_all_nolock(void)
3363 BUG_ON(!kvm_usage_count
);
3366 if (!kvm_usage_count
)
3367 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3370 static void hardware_disable_all(void)
3372 raw_spin_lock(&kvm_count_lock
);
3373 hardware_disable_all_nolock();
3374 raw_spin_unlock(&kvm_count_lock
);
3377 static int hardware_enable_all(void)
3381 raw_spin_lock(&kvm_count_lock
);
3384 if (kvm_usage_count
== 1) {
3385 atomic_set(&hardware_enable_failed
, 0);
3386 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3388 if (atomic_read(&hardware_enable_failed
)) {
3389 hardware_disable_all_nolock();
3394 raw_spin_unlock(&kvm_count_lock
);
3399 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3403 * Some (well, at least mine) BIOSes hang on reboot if
3406 * And Intel TXT required VMX off for all cpu when system shutdown.
3408 pr_info("kvm: exiting hardware virtualization\n");
3409 kvm_rebooting
= true;
3410 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3414 static struct notifier_block kvm_reboot_notifier
= {
3415 .notifier_call
= kvm_reboot
,
3419 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3423 for (i
= 0; i
< bus
->dev_count
; i
++) {
3424 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3426 kvm_iodevice_destructor(pos
);
3431 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3432 const struct kvm_io_range
*r2
)
3434 gpa_t addr1
= r1
->addr
;
3435 gpa_t addr2
= r2
->addr
;
3440 /* If r2->len == 0, match the exact address. If r2->len != 0,
3441 * accept any overlapping write. Any order is acceptable for
3442 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3443 * we process all of them.
3456 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3458 return kvm_io_bus_cmp(p1
, p2
);
3461 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3462 gpa_t addr
, int len
)
3464 struct kvm_io_range
*range
, key
;
3467 key
= (struct kvm_io_range
) {
3472 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3473 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3477 off
= range
- bus
->range
;
3479 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3485 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3486 struct kvm_io_range
*range
, const void *val
)
3490 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3494 while (idx
< bus
->dev_count
&&
3495 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3496 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3505 /* kvm_io_bus_write - called under kvm->slots_lock */
3506 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3507 int len
, const void *val
)
3509 struct kvm_io_bus
*bus
;
3510 struct kvm_io_range range
;
3513 range
= (struct kvm_io_range
) {
3518 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3521 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3522 return r
< 0 ? r
: 0;
3525 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3526 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3527 gpa_t addr
, int len
, const void *val
, long cookie
)
3529 struct kvm_io_bus
*bus
;
3530 struct kvm_io_range range
;
3532 range
= (struct kvm_io_range
) {
3537 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3541 /* First try the device referenced by cookie. */
3542 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3543 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3544 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3549 * cookie contained garbage; fall back to search and return the
3550 * correct cookie value.
3552 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3555 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3556 struct kvm_io_range
*range
, void *val
)
3560 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3564 while (idx
< bus
->dev_count
&&
3565 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3566 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3574 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3576 /* kvm_io_bus_read - called under kvm->slots_lock */
3577 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3580 struct kvm_io_bus
*bus
;
3581 struct kvm_io_range range
;
3584 range
= (struct kvm_io_range
) {
3589 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3592 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3593 return r
< 0 ? r
: 0;
3597 /* Caller must hold slots_lock. */
3598 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3599 int len
, struct kvm_io_device
*dev
)
3602 struct kvm_io_bus
*new_bus
, *bus
;
3603 struct kvm_io_range range
;
3605 bus
= kvm_get_bus(kvm
, bus_idx
);
3609 /* exclude ioeventfd which is limited by maximum fd */
3610 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3613 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3614 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3618 range
= (struct kvm_io_range
) {
3624 for (i
= 0; i
< bus
->dev_count
; i
++)
3625 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
3628 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3629 new_bus
->dev_count
++;
3630 new_bus
->range
[i
] = range
;
3631 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
3632 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3633 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3634 synchronize_srcu_expedited(&kvm
->srcu
);
3640 /* Caller must hold slots_lock. */
3641 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3642 struct kvm_io_device
*dev
)
3645 struct kvm_io_bus
*new_bus
, *bus
;
3647 bus
= kvm_get_bus(kvm
, bus_idx
);
3651 for (i
= 0; i
< bus
->dev_count
; i
++)
3652 if (bus
->range
[i
].dev
== dev
) {
3656 if (i
== bus
->dev_count
)
3659 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3660 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3662 pr_err("kvm: failed to shrink bus, removing it completely\n");
3666 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3667 new_bus
->dev_count
--;
3668 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3669 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3672 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3673 synchronize_srcu_expedited(&kvm
->srcu
);
3678 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3681 struct kvm_io_bus
*bus
;
3682 int dev_idx
, srcu_idx
;
3683 struct kvm_io_device
*iodev
= NULL
;
3685 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3687 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3691 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3695 iodev
= bus
->range
[dev_idx
].dev
;
3698 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3702 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3704 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3705 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3708 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3711 /* The debugfs files are a reference to the kvm struct which
3712 * is still valid when kvm_destroy_vm is called.
3713 * To avoid the race between open and the removal of the debugfs
3714 * directory we test against the users count.
3716 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3719 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3720 kvm_put_kvm(stat_data
->kvm
);
3727 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3729 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3732 simple_attr_release(inode
, file
);
3733 kvm_put_kvm(stat_data
->kvm
);
3738 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3740 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3742 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3747 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3749 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3754 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3759 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3761 __simple_attr_check_format("%llu\n", 0ull);
3762 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3763 vm_stat_clear_per_vm
, "%llu\n");
3766 static const struct file_operations vm_stat_get_per_vm_fops
= {
3767 .owner
= THIS_MODULE
,
3768 .open
= vm_stat_get_per_vm_open
,
3769 .release
= kvm_debugfs_release
,
3770 .read
= simple_attr_read
,
3771 .write
= simple_attr_write
,
3772 .llseek
= no_llseek
,
3775 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3778 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3779 struct kvm_vcpu
*vcpu
;
3783 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3784 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3789 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3792 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3793 struct kvm_vcpu
*vcpu
;
3798 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3799 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3804 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3806 __simple_attr_check_format("%llu\n", 0ull);
3807 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3808 vcpu_stat_clear_per_vm
, "%llu\n");
3811 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3812 .owner
= THIS_MODULE
,
3813 .open
= vcpu_stat_get_per_vm_open
,
3814 .release
= kvm_debugfs_release
,
3815 .read
= simple_attr_read
,
3816 .write
= simple_attr_write
,
3817 .llseek
= no_llseek
,
3820 static const struct file_operations
*stat_fops_per_vm
[] = {
3821 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3822 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3825 static int vm_stat_get(void *_offset
, u64
*val
)
3827 unsigned offset
= (long)_offset
;
3829 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3833 spin_lock(&kvm_lock
);
3834 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3836 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3839 spin_unlock(&kvm_lock
);
3843 static int vm_stat_clear(void *_offset
, u64 val
)
3845 unsigned offset
= (long)_offset
;
3847 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3852 spin_lock(&kvm_lock
);
3853 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3855 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3857 spin_unlock(&kvm_lock
);
3862 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3864 static int vcpu_stat_get(void *_offset
, u64
*val
)
3866 unsigned offset
= (long)_offset
;
3868 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3872 spin_lock(&kvm_lock
);
3873 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3875 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3878 spin_unlock(&kvm_lock
);
3882 static int vcpu_stat_clear(void *_offset
, u64 val
)
3884 unsigned offset
= (long)_offset
;
3886 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3891 spin_lock(&kvm_lock
);
3892 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3894 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3896 spin_unlock(&kvm_lock
);
3901 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3904 static const struct file_operations
*stat_fops
[] = {
3905 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3906 [KVM_STAT_VM
] = &vm_stat_fops
,
3909 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3911 struct kobj_uevent_env
*env
;
3912 unsigned long long created
, active
;
3914 if (!kvm_dev
.this_device
|| !kvm
)
3917 spin_lock(&kvm_lock
);
3918 if (type
== KVM_EVENT_CREATE_VM
) {
3919 kvm_createvm_count
++;
3921 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3924 created
= kvm_createvm_count
;
3925 active
= kvm_active_vms
;
3926 spin_unlock(&kvm_lock
);
3928 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3932 add_uevent_var(env
, "CREATED=%llu", created
);
3933 add_uevent_var(env
, "COUNT=%llu", active
);
3935 if (type
== KVM_EVENT_CREATE_VM
) {
3936 add_uevent_var(env
, "EVENT=create");
3937 kvm
->userspace_pid
= task_pid_nr(current
);
3938 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3939 add_uevent_var(env
, "EVENT=destroy");
3941 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3943 if (kvm
->debugfs_dentry
) {
3944 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3947 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3949 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3953 /* no need for checks, since we are adding at most only 5 keys */
3954 env
->envp
[env
->envp_idx
++] = NULL
;
3955 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3959 static void kvm_init_debug(void)
3961 struct kvm_stats_debugfs_item
*p
;
3963 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3965 kvm_debugfs_num_entries
= 0;
3966 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3967 debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3968 (void *)(long)p
->offset
,
3969 stat_fops
[p
->kind
]);
3973 static int kvm_suspend(void)
3975 if (kvm_usage_count
)
3976 hardware_disable_nolock(NULL
);
3980 static void kvm_resume(void)
3982 if (kvm_usage_count
) {
3983 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3984 hardware_enable_nolock(NULL
);
3988 static struct syscore_ops kvm_syscore_ops
= {
3989 .suspend
= kvm_suspend
,
3990 .resume
= kvm_resume
,
3994 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3996 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3999 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4001 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4003 if (vcpu
->preempted
)
4004 vcpu
->preempted
= false;
4006 kvm_arch_sched_in(vcpu
, cpu
);
4008 kvm_arch_vcpu_load(vcpu
, cpu
);
4011 static void kvm_sched_out(struct preempt_notifier
*pn
,
4012 struct task_struct
*next
)
4014 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4016 if (current
->state
== TASK_RUNNING
)
4017 vcpu
->preempted
= true;
4018 kvm_arch_vcpu_put(vcpu
);
4021 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4022 struct module
*module
)
4027 r
= kvm_arch_init(opaque
);
4032 * kvm_arch_init makes sure there's at most one caller
4033 * for architectures that support multiple implementations,
4034 * like intel and amd on x86.
4035 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4036 * conflicts in case kvm is already setup for another implementation.
4038 r
= kvm_irqfd_init();
4042 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4047 r
= kvm_arch_hardware_setup();
4051 for_each_online_cpu(cpu
) {
4052 smp_call_function_single(cpu
,
4053 kvm_arch_check_processor_compat
,
4059 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4060 kvm_starting_cpu
, kvm_dying_cpu
);
4063 register_reboot_notifier(&kvm_reboot_notifier
);
4065 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4067 vcpu_align
= __alignof__(struct kvm_vcpu
);
4069 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
4071 offsetof(struct kvm_vcpu
, arch
),
4072 sizeof_field(struct kvm_vcpu
, arch
),
4074 if (!kvm_vcpu_cache
) {
4079 r
= kvm_async_pf_init();
4083 kvm_chardev_ops
.owner
= module
;
4084 kvm_vm_fops
.owner
= module
;
4085 kvm_vcpu_fops
.owner
= module
;
4087 r
= misc_register(&kvm_dev
);
4089 pr_err("kvm: misc device register failed\n");
4093 register_syscore_ops(&kvm_syscore_ops
);
4095 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4096 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4100 r
= kvm_vfio_ops_init();
4106 kvm_async_pf_deinit();
4108 kmem_cache_destroy(kvm_vcpu_cache
);
4110 unregister_reboot_notifier(&kvm_reboot_notifier
);
4111 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4114 kvm_arch_hardware_unsetup();
4116 free_cpumask_var(cpus_hardware_enabled
);
4124 EXPORT_SYMBOL_GPL(kvm_init
);
4128 debugfs_remove_recursive(kvm_debugfs_dir
);
4129 misc_deregister(&kvm_dev
);
4130 kmem_cache_destroy(kvm_vcpu_cache
);
4131 kvm_async_pf_deinit();
4132 unregister_syscore_ops(&kvm_syscore_ops
);
4133 unregister_reboot_notifier(&kvm_reboot_notifier
);
4134 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4135 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4136 kvm_arch_hardware_unsetup();
4139 free_cpumask_var(cpus_hardware_enabled
);
4140 kvm_vfio_ops_exit();
4142 EXPORT_SYMBOL_GPL(kvm_exit
);