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
++;
358 if (kvm_set_spte_hva(kvm
, address
, pte
))
359 kvm_flush_remote_tlbs(kvm
);
361 spin_unlock(&kvm
->mmu_lock
);
362 srcu_read_unlock(&kvm
->srcu
, idx
);
365 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
366 const struct mmu_notifier_range
*range
)
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
, range
->start
, range
->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
, range
->start
,
389 range
->end
, range
->blockable
);
391 srcu_read_unlock(&kvm
->srcu
, idx
);
396 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
397 const struct mmu_notifier_range
*range
)
399 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
401 spin_lock(&kvm
->mmu_lock
);
403 * This sequence increase will notify the kvm page fault that
404 * the page that is going to be mapped in the spte could have
407 kvm
->mmu_notifier_seq
++;
410 * The above sequence increase must be visible before the
411 * below count decrease, which is ensured by the smp_wmb above
412 * in conjunction with the smp_rmb in mmu_notifier_retry().
414 kvm
->mmu_notifier_count
--;
415 spin_unlock(&kvm
->mmu_lock
);
417 BUG_ON(kvm
->mmu_notifier_count
< 0);
420 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
421 struct mm_struct
*mm
,
425 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
428 idx
= srcu_read_lock(&kvm
->srcu
);
429 spin_lock(&kvm
->mmu_lock
);
431 young
= kvm_age_hva(kvm
, start
, end
);
433 kvm_flush_remote_tlbs(kvm
);
435 spin_unlock(&kvm
->mmu_lock
);
436 srcu_read_unlock(&kvm
->srcu
, idx
);
441 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
442 struct mm_struct
*mm
,
446 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
449 idx
= srcu_read_lock(&kvm
->srcu
);
450 spin_lock(&kvm
->mmu_lock
);
452 * Even though we do not flush TLB, this will still adversely
453 * affect performance on pre-Haswell Intel EPT, where there is
454 * no EPT Access Bit to clear so that we have to tear down EPT
455 * tables instead. If we find this unacceptable, we can always
456 * add a parameter to kvm_age_hva so that it effectively doesn't
457 * do anything on clear_young.
459 * Also note that currently we never issue secondary TLB flushes
460 * from clear_young, leaving this job up to the regular system
461 * cadence. If we find this inaccurate, we might come up with a
462 * more sophisticated heuristic later.
464 young
= kvm_age_hva(kvm
, start
, end
);
465 spin_unlock(&kvm
->mmu_lock
);
466 srcu_read_unlock(&kvm
->srcu
, idx
);
471 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
472 struct mm_struct
*mm
,
473 unsigned long address
)
475 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
478 idx
= srcu_read_lock(&kvm
->srcu
);
479 spin_lock(&kvm
->mmu_lock
);
480 young
= kvm_test_age_hva(kvm
, address
);
481 spin_unlock(&kvm
->mmu_lock
);
482 srcu_read_unlock(&kvm
->srcu
, idx
);
487 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
488 struct mm_struct
*mm
)
490 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
493 idx
= srcu_read_lock(&kvm
->srcu
);
494 kvm_arch_flush_shadow_all(kvm
);
495 srcu_read_unlock(&kvm
->srcu
, idx
);
498 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
499 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
500 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
501 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
502 .clear_young
= kvm_mmu_notifier_clear_young
,
503 .test_young
= kvm_mmu_notifier_test_young
,
504 .change_pte
= kvm_mmu_notifier_change_pte
,
505 .release
= kvm_mmu_notifier_release
,
508 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
510 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
511 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
514 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
516 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
521 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
523 static struct kvm_memslots
*kvm_alloc_memslots(void)
526 struct kvm_memslots
*slots
;
528 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
532 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
533 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
538 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
540 if (!memslot
->dirty_bitmap
)
543 kvfree(memslot
->dirty_bitmap
);
544 memslot
->dirty_bitmap
= NULL
;
548 * Free any memory in @free but not in @dont.
550 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
551 struct kvm_memory_slot
*dont
)
553 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
554 kvm_destroy_dirty_bitmap(free
);
556 kvm_arch_free_memslot(kvm
, free
, dont
);
561 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
563 struct kvm_memory_slot
*memslot
;
568 kvm_for_each_memslot(memslot
, slots
)
569 kvm_free_memslot(kvm
, memslot
, NULL
);
574 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
578 if (!kvm
->debugfs_dentry
)
581 debugfs_remove_recursive(kvm
->debugfs_dentry
);
583 if (kvm
->debugfs_stat_data
) {
584 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
585 kfree(kvm
->debugfs_stat_data
[i
]);
586 kfree(kvm
->debugfs_stat_data
);
590 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
592 char dir_name
[ITOA_MAX_LEN
* 2];
593 struct kvm_stat_data
*stat_data
;
594 struct kvm_stats_debugfs_item
*p
;
596 if (!debugfs_initialized())
599 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
600 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
602 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
603 sizeof(*kvm
->debugfs_stat_data
),
605 if (!kvm
->debugfs_stat_data
)
608 for (p
= debugfs_entries
; p
->name
; p
++) {
609 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
613 stat_data
->kvm
= kvm
;
614 stat_data
->offset
= p
->offset
;
615 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
616 debugfs_create_file(p
->name
, 0644, kvm
->debugfs_dentry
,
617 stat_data
, stat_fops_per_vm
[p
->kind
]);
622 static struct kvm
*kvm_create_vm(unsigned long type
)
625 struct kvm
*kvm
= kvm_arch_alloc_vm();
628 return ERR_PTR(-ENOMEM
);
630 spin_lock_init(&kvm
->mmu_lock
);
632 kvm
->mm
= current
->mm
;
633 kvm_eventfd_init(kvm
);
634 mutex_init(&kvm
->lock
);
635 mutex_init(&kvm
->irq_lock
);
636 mutex_init(&kvm
->slots_lock
);
637 refcount_set(&kvm
->users_count
, 1);
638 INIT_LIST_HEAD(&kvm
->devices
);
640 r
= kvm_arch_init_vm(kvm
, type
);
642 goto out_err_no_disable
;
644 r
= hardware_enable_all();
646 goto out_err_no_disable
;
648 #ifdef CONFIG_HAVE_KVM_IRQFD
649 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
652 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
655 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
656 struct kvm_memslots
*slots
= kvm_alloc_memslots();
658 goto out_err_no_srcu
;
660 * Generations must be different for each address space.
661 * Init kvm generation close to the maximum to easily test the
662 * code of handling generation number wrap-around.
664 slots
->generation
= i
* 2 - 150;
665 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
668 if (init_srcu_struct(&kvm
->srcu
))
669 goto out_err_no_srcu
;
670 if (init_srcu_struct(&kvm
->irq_srcu
))
671 goto out_err_no_irq_srcu
;
672 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
673 rcu_assign_pointer(kvm
->buses
[i
],
674 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
679 r
= kvm_init_mmu_notifier(kvm
);
683 spin_lock(&kvm_lock
);
684 list_add(&kvm
->vm_list
, &vm_list
);
685 spin_unlock(&kvm_lock
);
687 preempt_notifier_inc();
692 cleanup_srcu_struct(&kvm
->irq_srcu
);
694 cleanup_srcu_struct(&kvm
->srcu
);
696 hardware_disable_all();
698 refcount_set(&kvm
->users_count
, 0);
699 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
700 kfree(kvm_get_bus(kvm
, i
));
701 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
702 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
703 kvm_arch_free_vm(kvm
);
708 static void kvm_destroy_devices(struct kvm
*kvm
)
710 struct kvm_device
*dev
, *tmp
;
713 * We do not need to take the kvm->lock here, because nobody else
714 * has a reference to the struct kvm at this point and therefore
715 * cannot access the devices list anyhow.
717 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
718 list_del(&dev
->vm_node
);
719 dev
->ops
->destroy(dev
);
723 static void kvm_destroy_vm(struct kvm
*kvm
)
726 struct mm_struct
*mm
= kvm
->mm
;
728 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
729 kvm_destroy_vm_debugfs(kvm
);
730 kvm_arch_sync_events(kvm
);
731 spin_lock(&kvm_lock
);
732 list_del(&kvm
->vm_list
);
733 spin_unlock(&kvm_lock
);
734 kvm_free_irq_routing(kvm
);
735 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
736 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
739 kvm_io_bus_destroy(bus
);
740 kvm
->buses
[i
] = NULL
;
742 kvm_coalesced_mmio_free(kvm
);
743 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
744 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
746 kvm_arch_flush_shadow_all(kvm
);
748 kvm_arch_destroy_vm(kvm
);
749 kvm_destroy_devices(kvm
);
750 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
751 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
752 cleanup_srcu_struct(&kvm
->irq_srcu
);
753 cleanup_srcu_struct(&kvm
->srcu
);
754 kvm_arch_free_vm(kvm
);
755 preempt_notifier_dec();
756 hardware_disable_all();
760 void kvm_get_kvm(struct kvm
*kvm
)
762 refcount_inc(&kvm
->users_count
);
764 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
766 void kvm_put_kvm(struct kvm
*kvm
)
768 if (refcount_dec_and_test(&kvm
->users_count
))
771 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
774 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
776 struct kvm
*kvm
= filp
->private_data
;
778 kvm_irqfd_release(kvm
);
785 * Allocation size is twice as large as the actual dirty bitmap size.
786 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
788 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
790 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
792 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
793 if (!memslot
->dirty_bitmap
)
800 * Insert memslot and re-sort memslots based on their GFN,
801 * so binary search could be used to lookup GFN.
802 * Sorting algorithm takes advantage of having initially
803 * sorted array and known changed memslot position.
805 static void update_memslots(struct kvm_memslots
*slots
,
806 struct kvm_memory_slot
*new,
807 enum kvm_mr_change change
)
810 int i
= slots
->id_to_index
[id
];
811 struct kvm_memory_slot
*mslots
= slots
->memslots
;
813 WARN_ON(mslots
[i
].id
!= id
);
817 WARN_ON(mslots
[i
].npages
|| !new->npages
);
821 WARN_ON(new->npages
|| !mslots
[i
].npages
);
827 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
828 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
829 if (!mslots
[i
+ 1].npages
)
831 mslots
[i
] = mslots
[i
+ 1];
832 slots
->id_to_index
[mslots
[i
].id
] = i
;
837 * The ">=" is needed when creating a slot with base_gfn == 0,
838 * so that it moves before all those with base_gfn == npages == 0.
840 * On the other hand, if new->npages is zero, the above loop has
841 * already left i pointing to the beginning of the empty part of
842 * mslots, and the ">=" would move the hole backwards in this
843 * case---which is wrong. So skip the loop when deleting a slot.
847 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
848 mslots
[i
] = mslots
[i
- 1];
849 slots
->id_to_index
[mslots
[i
].id
] = i
;
853 WARN_ON_ONCE(i
!= slots
->used_slots
);
856 slots
->id_to_index
[mslots
[i
].id
] = i
;
859 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
861 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
863 #ifdef __KVM_HAVE_READONLY_MEM
864 valid_flags
|= KVM_MEM_READONLY
;
867 if (mem
->flags
& ~valid_flags
)
873 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
874 int as_id
, struct kvm_memslots
*slots
)
876 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
879 * Set the low bit in the generation, which disables SPTE caching
880 * until the end of synchronize_srcu_expedited.
882 WARN_ON(old_memslots
->generation
& 1);
883 slots
->generation
= old_memslots
->generation
+ 1;
885 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
886 synchronize_srcu_expedited(&kvm
->srcu
);
889 * Increment the new memslot generation a second time. This prevents
890 * vm exits that race with memslot updates from caching a memslot
891 * generation that will (potentially) be valid forever.
893 * Generations must be unique even across address spaces. We do not need
894 * a global counter for that, instead the generation space is evenly split
895 * across address spaces. For example, with two address spaces, address
896 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
897 * use generations 2, 6, 10, 14, ...
899 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
901 kvm_arch_memslots_updated(kvm
, slots
);
907 * Allocate some memory and give it an address in the guest physical address
910 * Discontiguous memory is allowed, mostly for framebuffers.
912 * Must be called holding kvm->slots_lock for write.
914 int __kvm_set_memory_region(struct kvm
*kvm
,
915 const struct kvm_userspace_memory_region
*mem
)
919 unsigned long npages
;
920 struct kvm_memory_slot
*slot
;
921 struct kvm_memory_slot old
, new;
922 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
924 enum kvm_mr_change change
;
926 r
= check_memory_region_flags(mem
);
931 as_id
= mem
->slot
>> 16;
934 /* General sanity checks */
935 if (mem
->memory_size
& (PAGE_SIZE
- 1))
937 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
939 /* We can read the guest memory with __xxx_user() later on. */
940 if ((id
< KVM_USER_MEM_SLOTS
) &&
941 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
942 !access_ok(VERIFY_WRITE
,
943 (void __user
*)(unsigned long)mem
->userspace_addr
,
946 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
948 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
951 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
952 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
953 npages
= mem
->memory_size
>> PAGE_SHIFT
;
955 if (npages
> KVM_MEM_MAX_NR_PAGES
)
961 new.base_gfn
= base_gfn
;
963 new.flags
= mem
->flags
;
967 change
= KVM_MR_CREATE
;
968 else { /* Modify an existing slot. */
969 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
970 (npages
!= old
.npages
) ||
971 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
974 if (base_gfn
!= old
.base_gfn
)
975 change
= KVM_MR_MOVE
;
976 else if (new.flags
!= old
.flags
)
977 change
= KVM_MR_FLAGS_ONLY
;
978 else { /* Nothing to change. */
987 change
= KVM_MR_DELETE
;
992 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
993 /* Check for overlaps */
995 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
998 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
999 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
1004 /* Free page dirty bitmap if unneeded */
1005 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1006 new.dirty_bitmap
= NULL
;
1009 if (change
== KVM_MR_CREATE
) {
1010 new.userspace_addr
= mem
->userspace_addr
;
1012 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1016 /* Allocate page dirty bitmap if needed */
1017 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1018 if (kvm_create_dirty_bitmap(&new) < 0)
1022 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1025 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1027 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1028 slot
= id_to_memslot(slots
, id
);
1029 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1031 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1033 /* From this point no new shadow pages pointing to a deleted,
1034 * or moved, memslot will be created.
1036 * validation of sp->gfn happens in:
1037 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1038 * - kvm_is_visible_gfn (mmu_check_roots)
1040 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1043 * We can re-use the old_memslots from above, the only difference
1044 * from the currently installed memslots is the invalid flag. This
1045 * will get overwritten by update_memslots anyway.
1047 slots
= old_memslots
;
1050 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1054 /* actual memory is freed via old in kvm_free_memslot below */
1055 if (change
== KVM_MR_DELETE
) {
1056 new.dirty_bitmap
= NULL
;
1057 memset(&new.arch
, 0, sizeof(new.arch
));
1060 update_memslots(slots
, &new, change
);
1061 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1063 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1065 kvm_free_memslot(kvm
, &old
, &new);
1066 kvfree(old_memslots
);
1072 kvm_free_memslot(kvm
, &new, &old
);
1076 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1078 int kvm_set_memory_region(struct kvm
*kvm
,
1079 const struct kvm_userspace_memory_region
*mem
)
1083 mutex_lock(&kvm
->slots_lock
);
1084 r
= __kvm_set_memory_region(kvm
, mem
);
1085 mutex_unlock(&kvm
->slots_lock
);
1088 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1090 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1091 struct kvm_userspace_memory_region
*mem
)
1093 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1096 return kvm_set_memory_region(kvm
, mem
);
1099 int kvm_get_dirty_log(struct kvm
*kvm
,
1100 struct kvm_dirty_log
*log
, int *is_dirty
)
1102 struct kvm_memslots
*slots
;
1103 struct kvm_memory_slot
*memslot
;
1106 unsigned long any
= 0;
1108 as_id
= log
->slot
>> 16;
1109 id
= (u16
)log
->slot
;
1110 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1113 slots
= __kvm_memslots(kvm
, as_id
);
1114 memslot
= id_to_memslot(slots
, id
);
1115 if (!memslot
->dirty_bitmap
)
1118 n
= kvm_dirty_bitmap_bytes(memslot
);
1120 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1121 any
= memslot
->dirty_bitmap
[i
];
1123 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1130 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1132 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1134 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1135 * and reenable dirty page tracking for the corresponding pages.
1136 * @kvm: pointer to kvm instance
1137 * @log: slot id and address to which we copy the log
1138 * @is_dirty: flag set if any page is dirty
1140 * We need to keep it in mind that VCPU threads can write to the bitmap
1141 * concurrently. So, to avoid losing track of dirty pages we keep the
1144 * 1. Take a snapshot of the bit and clear it if needed.
1145 * 2. Write protect the corresponding page.
1146 * 3. Copy the snapshot to the userspace.
1147 * 4. Upon return caller flushes TLB's if needed.
1149 * Between 2 and 4, the guest may write to the page using the remaining TLB
1150 * entry. This is not a problem because the page is reported dirty using
1151 * the snapshot taken before and step 4 ensures that writes done after
1152 * exiting to userspace will be logged for the next call.
1155 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1156 struct kvm_dirty_log
*log
, bool *flush
)
1158 struct kvm_memslots
*slots
;
1159 struct kvm_memory_slot
*memslot
;
1162 unsigned long *dirty_bitmap
;
1163 unsigned long *dirty_bitmap_buffer
;
1165 as_id
= log
->slot
>> 16;
1166 id
= (u16
)log
->slot
;
1167 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1170 slots
= __kvm_memslots(kvm
, as_id
);
1171 memslot
= id_to_memslot(slots
, id
);
1173 dirty_bitmap
= memslot
->dirty_bitmap
;
1177 n
= kvm_dirty_bitmap_bytes(memslot
);
1179 if (kvm
->manual_dirty_log_protect
) {
1181 * Unlike kvm_get_dirty_log, we always return false in *flush,
1182 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1183 * is some code duplication between this function and
1184 * kvm_get_dirty_log, but hopefully all architecture
1185 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1186 * can be eliminated.
1188 dirty_bitmap_buffer
= dirty_bitmap
;
1190 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1191 memset(dirty_bitmap_buffer
, 0, n
);
1193 spin_lock(&kvm
->mmu_lock
);
1194 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1198 if (!dirty_bitmap
[i
])
1202 mask
= xchg(&dirty_bitmap
[i
], 0);
1203 dirty_bitmap_buffer
[i
] = mask
;
1206 offset
= i
* BITS_PER_LONG
;
1207 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1211 spin_unlock(&kvm
->mmu_lock
);
1214 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1218 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1221 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1222 * and reenable dirty page tracking for the corresponding pages.
1223 * @kvm: pointer to kvm instance
1224 * @log: slot id and address from which to fetch the bitmap of dirty pages
1226 int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
1227 struct kvm_clear_dirty_log
*log
, bool *flush
)
1229 struct kvm_memslots
*slots
;
1230 struct kvm_memory_slot
*memslot
;
1234 unsigned long *dirty_bitmap
;
1235 unsigned long *dirty_bitmap_buffer
;
1237 as_id
= log
->slot
>> 16;
1238 id
= (u16
)log
->slot
;
1239 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1242 if ((log
->first_page
& 63) || (log
->num_pages
& 63))
1245 slots
= __kvm_memslots(kvm
, as_id
);
1246 memslot
= id_to_memslot(slots
, id
);
1248 dirty_bitmap
= memslot
->dirty_bitmap
;
1252 n
= kvm_dirty_bitmap_bytes(memslot
);
1254 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1255 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
1258 spin_lock(&kvm
->mmu_lock
);
1259 for (offset
= log
->first_page
,
1260 i
= offset
/ BITS_PER_LONG
, n
= log
->num_pages
/ BITS_PER_LONG
; n
--;
1261 i
++, offset
+= BITS_PER_LONG
) {
1262 unsigned long mask
= *dirty_bitmap_buffer
++;
1263 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
1267 mask
&= atomic_long_fetch_andnot(mask
, p
);
1270 * mask contains the bits that really have been cleared. This
1271 * never includes any bits beyond the length of the memslot (if
1272 * the length is not aligned to 64 pages), therefore it is not
1273 * a problem if userspace sets them in log->dirty_bitmap.
1277 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1281 spin_unlock(&kvm
->mmu_lock
);
1285 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect
);
1288 bool kvm_largepages_enabled(void)
1290 return largepages_enabled
;
1293 void kvm_disable_largepages(void)
1295 largepages_enabled
= false;
1297 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1299 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1301 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1303 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1305 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1307 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1310 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1312 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1314 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1315 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1320 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1322 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1324 struct vm_area_struct
*vma
;
1325 unsigned long addr
, size
;
1329 addr
= gfn_to_hva(kvm
, gfn
);
1330 if (kvm_is_error_hva(addr
))
1333 down_read(¤t
->mm
->mmap_sem
);
1334 vma
= find_vma(current
->mm
, addr
);
1338 size
= vma_kernel_pagesize(vma
);
1341 up_read(¤t
->mm
->mmap_sem
);
1346 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1348 return slot
->flags
& KVM_MEM_READONLY
;
1351 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1352 gfn_t
*nr_pages
, bool write
)
1354 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1355 return KVM_HVA_ERR_BAD
;
1357 if (memslot_is_readonly(slot
) && write
)
1358 return KVM_HVA_ERR_RO_BAD
;
1361 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1363 return __gfn_to_hva_memslot(slot
, gfn
);
1366 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1369 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1372 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1375 return gfn_to_hva_many(slot
, gfn
, NULL
);
1377 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1379 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1381 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1383 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1385 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1387 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1389 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1392 * Return the hva of a @gfn and the R/W attribute if possible.
1394 * @slot: the kvm_memory_slot which contains @gfn
1395 * @gfn: the gfn to be translated
1396 * @writable: used to return the read/write attribute of the @slot if the hva
1397 * is valid and @writable is not NULL
1399 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1400 gfn_t gfn
, bool *writable
)
1402 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1404 if (!kvm_is_error_hva(hva
) && writable
)
1405 *writable
= !memslot_is_readonly(slot
);
1410 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1412 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1414 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1417 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1419 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1421 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1424 static inline int check_user_page_hwpoison(unsigned long addr
)
1426 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1428 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1429 return rc
== -EHWPOISON
;
1433 * The fast path to get the writable pfn which will be stored in @pfn,
1434 * true indicates success, otherwise false is returned. It's also the
1435 * only part that runs if we can are in atomic context.
1437 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
1438 bool *writable
, kvm_pfn_t
*pfn
)
1440 struct page
*page
[1];
1444 * Fast pin a writable pfn only if it is a write fault request
1445 * or the caller allows to map a writable pfn for a read fault
1448 if (!(write_fault
|| writable
))
1451 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1453 *pfn
= page_to_pfn(page
[0]);
1464 * The slow path to get the pfn of the specified host virtual address,
1465 * 1 indicates success, -errno is returned if error is detected.
1467 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1468 bool *writable
, kvm_pfn_t
*pfn
)
1470 unsigned int flags
= FOLL_HWPOISON
;
1477 *writable
= write_fault
;
1480 flags
|= FOLL_WRITE
;
1482 flags
|= FOLL_NOWAIT
;
1484 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
1488 /* map read fault as writable if possible */
1489 if (unlikely(!write_fault
) && writable
) {
1492 if (__get_user_pages_fast(addr
, 1, 1, &wpage
) == 1) {
1498 *pfn
= page_to_pfn(page
);
1502 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1504 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1507 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1513 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1514 unsigned long addr
, bool *async
,
1515 bool write_fault
, bool *writable
,
1521 r
= follow_pfn(vma
, addr
, &pfn
);
1524 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1525 * not call the fault handler, so do it here.
1527 bool unlocked
= false;
1528 r
= fixup_user_fault(current
, current
->mm
, addr
,
1529 (write_fault
? FAULT_FLAG_WRITE
: 0),
1536 r
= follow_pfn(vma
, addr
, &pfn
);
1546 * Get a reference here because callers of *hva_to_pfn* and
1547 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1548 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1549 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1550 * simply do nothing for reserved pfns.
1552 * Whoever called remap_pfn_range is also going to call e.g.
1553 * unmap_mapping_range before the underlying pages are freed,
1554 * causing a call to our MMU notifier.
1563 * Pin guest page in memory and return its pfn.
1564 * @addr: host virtual address which maps memory to the guest
1565 * @atomic: whether this function can sleep
1566 * @async: whether this function need to wait IO complete if the
1567 * host page is not in the memory
1568 * @write_fault: whether we should get a writable host page
1569 * @writable: whether it allows to map a writable host page for !@write_fault
1571 * The function will map a writable host page for these two cases:
1572 * 1): @write_fault = true
1573 * 2): @write_fault = false && @writable, @writable will tell the caller
1574 * whether the mapping is writable.
1576 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1577 bool write_fault
, bool *writable
)
1579 struct vm_area_struct
*vma
;
1583 /* we can do it either atomically or asynchronously, not both */
1584 BUG_ON(atomic
&& async
);
1586 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
1590 return KVM_PFN_ERR_FAULT
;
1592 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1596 down_read(¤t
->mm
->mmap_sem
);
1597 if (npages
== -EHWPOISON
||
1598 (!async
&& check_user_page_hwpoison(addr
))) {
1599 pfn
= KVM_PFN_ERR_HWPOISON
;
1604 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1607 pfn
= KVM_PFN_ERR_FAULT
;
1608 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1609 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1613 pfn
= KVM_PFN_ERR_FAULT
;
1615 if (async
&& vma_is_valid(vma
, write_fault
))
1617 pfn
= KVM_PFN_ERR_FAULT
;
1620 up_read(¤t
->mm
->mmap_sem
);
1624 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1625 bool atomic
, bool *async
, bool write_fault
,
1628 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1630 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1633 return KVM_PFN_ERR_RO_FAULT
;
1636 if (kvm_is_error_hva(addr
)) {
1639 return KVM_PFN_NOSLOT
;
1642 /* Do not map writable pfn in the readonly memslot. */
1643 if (writable
&& memslot_is_readonly(slot
)) {
1648 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1651 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1653 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1656 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1657 write_fault
, writable
);
1659 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1661 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1663 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1665 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1667 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1669 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1671 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1673 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1675 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1677 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1679 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1681 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1683 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1685 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1687 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1689 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1691 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1693 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1695 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1697 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1698 struct page
**pages
, int nr_pages
)
1703 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1704 if (kvm_is_error_hva(addr
))
1707 if (entry
< nr_pages
)
1710 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1712 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1714 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1716 if (is_error_noslot_pfn(pfn
))
1717 return KVM_ERR_PTR_BAD_PAGE
;
1719 if (kvm_is_reserved_pfn(pfn
)) {
1721 return KVM_ERR_PTR_BAD_PAGE
;
1724 return pfn_to_page(pfn
);
1727 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1731 pfn
= gfn_to_pfn(kvm
, gfn
);
1733 return kvm_pfn_to_page(pfn
);
1735 EXPORT_SYMBOL_GPL(gfn_to_page
);
1737 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1741 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1743 return kvm_pfn_to_page(pfn
);
1745 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1747 void kvm_release_page_clean(struct page
*page
)
1749 WARN_ON(is_error_page(page
));
1751 kvm_release_pfn_clean(page_to_pfn(page
));
1753 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1755 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1757 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1758 put_page(pfn_to_page(pfn
));
1760 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1762 void kvm_release_page_dirty(struct page
*page
)
1764 WARN_ON(is_error_page(page
));
1766 kvm_release_pfn_dirty(page_to_pfn(page
));
1768 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1770 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1772 kvm_set_pfn_dirty(pfn
);
1773 kvm_release_pfn_clean(pfn
);
1775 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1777 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1779 if (!kvm_is_reserved_pfn(pfn
)) {
1780 struct page
*page
= pfn_to_page(pfn
);
1782 if (!PageReserved(page
))
1786 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1788 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1790 if (!kvm_is_reserved_pfn(pfn
))
1791 mark_page_accessed(pfn_to_page(pfn
));
1793 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1795 void kvm_get_pfn(kvm_pfn_t pfn
)
1797 if (!kvm_is_reserved_pfn(pfn
))
1798 get_page(pfn_to_page(pfn
));
1800 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1802 static int next_segment(unsigned long len
, int offset
)
1804 if (len
> PAGE_SIZE
- offset
)
1805 return PAGE_SIZE
- offset
;
1810 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1811 void *data
, int offset
, int len
)
1816 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1817 if (kvm_is_error_hva(addr
))
1819 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1825 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1828 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1830 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1832 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1834 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1835 int offset
, int len
)
1837 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1839 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1841 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1843 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1845 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1847 int offset
= offset_in_page(gpa
);
1850 while ((seg
= next_segment(len
, offset
)) != 0) {
1851 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1861 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1863 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1865 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1867 int offset
= offset_in_page(gpa
);
1870 while ((seg
= next_segment(len
, offset
)) != 0) {
1871 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1881 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1883 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1884 void *data
, int offset
, unsigned long len
)
1889 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1890 if (kvm_is_error_hva(addr
))
1892 pagefault_disable();
1893 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1900 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1903 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1904 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1905 int offset
= offset_in_page(gpa
);
1907 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1909 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1911 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1912 void *data
, unsigned long len
)
1914 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1915 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1916 int offset
= offset_in_page(gpa
);
1918 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1920 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1922 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1923 const void *data
, int offset
, int len
)
1928 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1929 if (kvm_is_error_hva(addr
))
1931 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1934 mark_page_dirty_in_slot(memslot
, gfn
);
1938 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1939 const void *data
, int offset
, int len
)
1941 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1943 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1945 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1947 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1948 const void *data
, int offset
, int len
)
1950 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1952 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1954 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1956 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1959 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1961 int offset
= offset_in_page(gpa
);
1964 while ((seg
= next_segment(len
, offset
)) != 0) {
1965 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1975 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1977 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1980 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1982 int offset
= offset_in_page(gpa
);
1985 while ((seg
= next_segment(len
, offset
)) != 0) {
1986 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1996 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1998 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1999 struct gfn_to_hva_cache
*ghc
,
2000 gpa_t gpa
, unsigned long len
)
2002 int offset
= offset_in_page(gpa
);
2003 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
2004 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
2005 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
2006 gfn_t nr_pages_avail
;
2007 int r
= start_gfn
<= end_gfn
? 0 : -EINVAL
;
2010 ghc
->generation
= slots
->generation
;
2012 ghc
->hva
= KVM_HVA_ERR_BAD
;
2015 * If the requested region crosses two memslots, we still
2016 * verify that the entire region is valid here.
2018 while (!r
&& start_gfn
<= end_gfn
) {
2019 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
2020 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
2022 if (kvm_is_error_hva(ghc
->hva
))
2024 start_gfn
+= nr_pages_avail
;
2027 /* Use the slow path for cross page reads and writes. */
2028 if (!r
&& nr_pages_needed
== 1)
2031 ghc
->memslot
= NULL
;
2036 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2037 gpa_t gpa
, unsigned long len
)
2039 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2040 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
2042 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
2044 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2045 void *data
, unsigned int offset
,
2048 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2050 gpa_t gpa
= ghc
->gpa
+ offset
;
2052 BUG_ON(len
+ offset
> ghc
->len
);
2054 if (slots
->generation
!= ghc
->generation
)
2055 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2057 if (unlikely(!ghc
->memslot
))
2058 return kvm_write_guest(kvm
, gpa
, data
, len
);
2060 if (kvm_is_error_hva(ghc
->hva
))
2063 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2066 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2070 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2072 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2073 void *data
, unsigned long len
)
2075 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2077 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2079 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2080 void *data
, unsigned long len
)
2082 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2085 BUG_ON(len
> ghc
->len
);
2087 if (slots
->generation
!= ghc
->generation
)
2088 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2090 if (unlikely(!ghc
->memslot
))
2091 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2093 if (kvm_is_error_hva(ghc
->hva
))
2096 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2102 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2104 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2106 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2108 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2110 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2112 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2114 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2116 int offset
= offset_in_page(gpa
);
2119 while ((seg
= next_segment(len
, offset
)) != 0) {
2120 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2129 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2131 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2134 if (memslot
&& memslot
->dirty_bitmap
) {
2135 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2137 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2141 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2143 struct kvm_memory_slot
*memslot
;
2145 memslot
= gfn_to_memslot(kvm
, gfn
);
2146 mark_page_dirty_in_slot(memslot
, gfn
);
2148 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2150 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2152 struct kvm_memory_slot
*memslot
;
2154 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2155 mark_page_dirty_in_slot(memslot
, gfn
);
2157 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2159 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2161 if (!vcpu
->sigset_active
)
2165 * This does a lockless modification of ->real_blocked, which is fine
2166 * because, only current can change ->real_blocked and all readers of
2167 * ->real_blocked don't care as long ->real_blocked is always a subset
2170 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2173 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2175 if (!vcpu
->sigset_active
)
2178 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2179 sigemptyset(¤t
->real_blocked
);
2182 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2184 unsigned int old
, val
, grow
;
2186 old
= val
= vcpu
->halt_poll_ns
;
2187 grow
= READ_ONCE(halt_poll_ns_grow
);
2189 if (val
== 0 && grow
)
2194 if (val
> halt_poll_ns
)
2197 vcpu
->halt_poll_ns
= val
;
2198 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2201 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2203 unsigned int old
, val
, shrink
;
2205 old
= val
= vcpu
->halt_poll_ns
;
2206 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2212 vcpu
->halt_poll_ns
= val
;
2213 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2216 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2219 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2221 if (kvm_arch_vcpu_runnable(vcpu
)) {
2222 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2225 if (kvm_cpu_has_pending_timer(vcpu
))
2227 if (signal_pending(current
))
2232 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2237 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2239 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2242 DECLARE_SWAITQUEUE(wait
);
2243 bool waited
= false;
2246 start
= cur
= ktime_get();
2247 if (vcpu
->halt_poll_ns
) {
2248 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2250 ++vcpu
->stat
.halt_attempted_poll
;
2253 * This sets KVM_REQ_UNHALT if an interrupt
2256 if (kvm_vcpu_check_block(vcpu
) < 0) {
2257 ++vcpu
->stat
.halt_successful_poll
;
2258 if (!vcpu_valid_wakeup(vcpu
))
2259 ++vcpu
->stat
.halt_poll_invalid
;
2263 } while (single_task_running() && ktime_before(cur
, stop
));
2266 kvm_arch_vcpu_blocking(vcpu
);
2269 prepare_to_swait_exclusive(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2271 if (kvm_vcpu_check_block(vcpu
) < 0)
2278 finish_swait(&vcpu
->wq
, &wait
);
2281 kvm_arch_vcpu_unblocking(vcpu
);
2283 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2285 if (!vcpu_valid_wakeup(vcpu
))
2286 shrink_halt_poll_ns(vcpu
);
2287 else if (halt_poll_ns
) {
2288 if (block_ns
<= vcpu
->halt_poll_ns
)
2290 /* we had a long block, shrink polling */
2291 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2292 shrink_halt_poll_ns(vcpu
);
2293 /* we had a short halt and our poll time is too small */
2294 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2295 block_ns
< halt_poll_ns
)
2296 grow_halt_poll_ns(vcpu
);
2298 vcpu
->halt_poll_ns
= 0;
2300 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2301 kvm_arch_vcpu_block_finish(vcpu
);
2303 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2305 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2307 struct swait_queue_head
*wqp
;
2309 wqp
= kvm_arch_vcpu_wq(vcpu
);
2310 if (swq_has_sleeper(wqp
)) {
2312 ++vcpu
->stat
.halt_wakeup
;
2318 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2322 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2324 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2327 int cpu
= vcpu
->cpu
;
2329 if (kvm_vcpu_wake_up(vcpu
))
2333 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2334 if (kvm_arch_vcpu_should_kick(vcpu
))
2335 smp_send_reschedule(cpu
);
2338 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2339 #endif /* !CONFIG_S390 */
2341 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2344 struct task_struct
*task
= NULL
;
2348 pid
= rcu_dereference(target
->pid
);
2350 task
= get_pid_task(pid
, PIDTYPE_PID
);
2354 ret
= yield_to(task
, 1);
2355 put_task_struct(task
);
2359 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2362 * Helper that checks whether a VCPU is eligible for directed yield.
2363 * Most eligible candidate to yield is decided by following heuristics:
2365 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2366 * (preempted lock holder), indicated by @in_spin_loop.
2367 * Set at the beiginning and cleared at the end of interception/PLE handler.
2369 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2370 * chance last time (mostly it has become eligible now since we have probably
2371 * yielded to lockholder in last iteration. This is done by toggling
2372 * @dy_eligible each time a VCPU checked for eligibility.)
2374 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2375 * to preempted lock-holder could result in wrong VCPU selection and CPU
2376 * burning. Giving priority for a potential lock-holder increases lock
2379 * Since algorithm is based on heuristics, accessing another VCPU data without
2380 * locking does not harm. It may result in trying to yield to same VCPU, fail
2381 * and continue with next VCPU and so on.
2383 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2385 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2388 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2389 vcpu
->spin_loop
.dy_eligible
;
2391 if (vcpu
->spin_loop
.in_spin_loop
)
2392 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2400 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2402 struct kvm
*kvm
= me
->kvm
;
2403 struct kvm_vcpu
*vcpu
;
2404 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2410 kvm_vcpu_set_in_spin_loop(me
, true);
2412 * We boost the priority of a VCPU that is runnable but not
2413 * currently running, because it got preempted by something
2414 * else and called schedule in __vcpu_run. Hopefully that
2415 * VCPU is holding the lock that we need and will release it.
2416 * We approximate round-robin by starting at the last boosted VCPU.
2418 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2419 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2420 if (!pass
&& i
<= last_boosted_vcpu
) {
2421 i
= last_boosted_vcpu
;
2423 } else if (pass
&& i
> last_boosted_vcpu
)
2425 if (!READ_ONCE(vcpu
->preempted
))
2429 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2431 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2433 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2436 yielded
= kvm_vcpu_yield_to(vcpu
);
2438 kvm
->last_boosted_vcpu
= i
;
2440 } else if (yielded
< 0) {
2447 kvm_vcpu_set_in_spin_loop(me
, false);
2449 /* Ensure vcpu is not eligible during next spinloop */
2450 kvm_vcpu_set_dy_eligible(me
, false);
2452 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2454 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
2456 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2459 if (vmf
->pgoff
== 0)
2460 page
= virt_to_page(vcpu
->run
);
2462 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2463 page
= virt_to_page(vcpu
->arch
.pio_data
);
2465 #ifdef CONFIG_KVM_MMIO
2466 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2467 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2470 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2476 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2477 .fault
= kvm_vcpu_fault
,
2480 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2482 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2486 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2488 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2490 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2491 kvm_put_kvm(vcpu
->kvm
);
2495 static struct file_operations kvm_vcpu_fops
= {
2496 .release
= kvm_vcpu_release
,
2497 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2498 .mmap
= kvm_vcpu_mmap
,
2499 .llseek
= noop_llseek
,
2500 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
2504 * Allocates an inode for the vcpu.
2506 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2508 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
2510 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
2511 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2514 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2516 char dir_name
[ITOA_MAX_LEN
* 2];
2519 if (!kvm_arch_has_vcpu_debugfs())
2522 if (!debugfs_initialized())
2525 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2526 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2527 vcpu
->kvm
->debugfs_dentry
);
2528 if (!vcpu
->debugfs_dentry
)
2531 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2533 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2541 * Creates some virtual cpus. Good luck creating more than one.
2543 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2546 struct kvm_vcpu
*vcpu
;
2548 if (id
>= KVM_MAX_VCPU_ID
)
2551 mutex_lock(&kvm
->lock
);
2552 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2553 mutex_unlock(&kvm
->lock
);
2557 kvm
->created_vcpus
++;
2558 mutex_unlock(&kvm
->lock
);
2560 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2563 goto vcpu_decrement
;
2566 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2568 r
= kvm_arch_vcpu_setup(vcpu
);
2572 r
= kvm_create_vcpu_debugfs(vcpu
);
2576 mutex_lock(&kvm
->lock
);
2577 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2579 goto unlock_vcpu_destroy
;
2582 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2584 /* Now it's all set up, let userspace reach it */
2586 r
= create_vcpu_fd(vcpu
);
2589 goto unlock_vcpu_destroy
;
2592 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2595 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2596 * before kvm->online_vcpu's incremented value.
2599 atomic_inc(&kvm
->online_vcpus
);
2601 mutex_unlock(&kvm
->lock
);
2602 kvm_arch_vcpu_postcreate(vcpu
);
2605 unlock_vcpu_destroy
:
2606 mutex_unlock(&kvm
->lock
);
2607 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2609 kvm_arch_vcpu_destroy(vcpu
);
2611 mutex_lock(&kvm
->lock
);
2612 kvm
->created_vcpus
--;
2613 mutex_unlock(&kvm
->lock
);
2617 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2620 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2621 vcpu
->sigset_active
= 1;
2622 vcpu
->sigset
= *sigset
;
2624 vcpu
->sigset_active
= 0;
2628 static long kvm_vcpu_ioctl(struct file
*filp
,
2629 unsigned int ioctl
, unsigned long arg
)
2631 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2632 void __user
*argp
= (void __user
*)arg
;
2634 struct kvm_fpu
*fpu
= NULL
;
2635 struct kvm_sregs
*kvm_sregs
= NULL
;
2637 if (vcpu
->kvm
->mm
!= current
->mm
)
2640 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2644 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2645 * execution; mutex_lock() would break them.
2647 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
2648 if (r
!= -ENOIOCTLCMD
)
2651 if (mutex_lock_killable(&vcpu
->mutex
))
2659 oldpid
= rcu_access_pointer(vcpu
->pid
);
2660 if (unlikely(oldpid
!= task_pid(current
))) {
2661 /* The thread running this VCPU changed. */
2664 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
2668 newpid
= get_task_pid(current
, PIDTYPE_PID
);
2669 rcu_assign_pointer(vcpu
->pid
, newpid
);
2674 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2675 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2678 case KVM_GET_REGS
: {
2679 struct kvm_regs
*kvm_regs
;
2682 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2685 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2689 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2696 case KVM_SET_REGS
: {
2697 struct kvm_regs
*kvm_regs
;
2700 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2701 if (IS_ERR(kvm_regs
)) {
2702 r
= PTR_ERR(kvm_regs
);
2705 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2709 case KVM_GET_SREGS
: {
2710 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2714 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2718 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2723 case KVM_SET_SREGS
: {
2724 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2725 if (IS_ERR(kvm_sregs
)) {
2726 r
= PTR_ERR(kvm_sregs
);
2730 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2733 case KVM_GET_MP_STATE
: {
2734 struct kvm_mp_state mp_state
;
2736 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2740 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2745 case KVM_SET_MP_STATE
: {
2746 struct kvm_mp_state mp_state
;
2749 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2751 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2754 case KVM_TRANSLATE
: {
2755 struct kvm_translation tr
;
2758 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2760 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2764 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2769 case KVM_SET_GUEST_DEBUG
: {
2770 struct kvm_guest_debug dbg
;
2773 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2775 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2778 case KVM_SET_SIGNAL_MASK
: {
2779 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2780 struct kvm_signal_mask kvm_sigmask
;
2781 sigset_t sigset
, *p
;
2786 if (copy_from_user(&kvm_sigmask
, argp
,
2787 sizeof(kvm_sigmask
)))
2790 if (kvm_sigmask
.len
!= sizeof(sigset
))
2793 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2798 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2802 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2806 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2810 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2816 fpu
= memdup_user(argp
, sizeof(*fpu
));
2822 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2826 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2829 mutex_unlock(&vcpu
->mutex
);
2835 #ifdef CONFIG_KVM_COMPAT
2836 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2837 unsigned int ioctl
, unsigned long arg
)
2839 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2840 void __user
*argp
= compat_ptr(arg
);
2843 if (vcpu
->kvm
->mm
!= current
->mm
)
2847 case KVM_SET_SIGNAL_MASK
: {
2848 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2849 struct kvm_signal_mask kvm_sigmask
;
2854 if (copy_from_user(&kvm_sigmask
, argp
,
2855 sizeof(kvm_sigmask
)))
2858 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2861 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2863 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2865 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2869 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2877 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2878 int (*accessor
)(struct kvm_device
*dev
,
2879 struct kvm_device_attr
*attr
),
2882 struct kvm_device_attr attr
;
2887 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2890 return accessor(dev
, &attr
);
2893 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2896 struct kvm_device
*dev
= filp
->private_data
;
2899 case KVM_SET_DEVICE_ATTR
:
2900 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2901 case KVM_GET_DEVICE_ATTR
:
2902 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2903 case KVM_HAS_DEVICE_ATTR
:
2904 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2906 if (dev
->ops
->ioctl
)
2907 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2913 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2915 struct kvm_device
*dev
= filp
->private_data
;
2916 struct kvm
*kvm
= dev
->kvm
;
2922 static const struct file_operations kvm_device_fops
= {
2923 .unlocked_ioctl
= kvm_device_ioctl
,
2924 .release
= kvm_device_release
,
2925 KVM_COMPAT(kvm_device_ioctl
),
2928 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2930 if (filp
->f_op
!= &kvm_device_fops
)
2933 return filp
->private_data
;
2936 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2937 #ifdef CONFIG_KVM_MPIC
2938 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2939 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2943 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2945 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2948 if (kvm_device_ops_table
[type
] != NULL
)
2951 kvm_device_ops_table
[type
] = ops
;
2955 void kvm_unregister_device_ops(u32 type
)
2957 if (kvm_device_ops_table
[type
] != NULL
)
2958 kvm_device_ops_table
[type
] = NULL
;
2961 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2962 struct kvm_create_device
*cd
)
2964 struct kvm_device_ops
*ops
= NULL
;
2965 struct kvm_device
*dev
;
2966 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2969 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2972 ops
= kvm_device_ops_table
[cd
->type
];
2979 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2986 mutex_lock(&kvm
->lock
);
2987 ret
= ops
->create(dev
, cd
->type
);
2989 mutex_unlock(&kvm
->lock
);
2993 list_add(&dev
->vm_node
, &kvm
->devices
);
2994 mutex_unlock(&kvm
->lock
);
2999 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
3001 mutex_lock(&kvm
->lock
);
3002 list_del(&dev
->vm_node
);
3003 mutex_unlock(&kvm
->lock
);
3013 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
3016 case KVM_CAP_USER_MEMORY
:
3017 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
3018 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
3019 case KVM_CAP_INTERNAL_ERROR_DATA
:
3020 #ifdef CONFIG_HAVE_KVM_MSI
3021 case KVM_CAP_SIGNAL_MSI
:
3023 #ifdef CONFIG_HAVE_KVM_IRQFD
3025 case KVM_CAP_IRQFD_RESAMPLE
:
3027 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
3028 case KVM_CAP_CHECK_EXTENSION_VM
:
3029 case KVM_CAP_ENABLE_CAP_VM
:
3030 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3031 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
:
3034 #ifdef CONFIG_KVM_MMIO
3035 case KVM_CAP_COALESCED_MMIO
:
3036 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
3037 case KVM_CAP_COALESCED_PIO
:
3040 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3041 case KVM_CAP_IRQ_ROUTING
:
3042 return KVM_MAX_IRQ_ROUTES
;
3044 #if KVM_ADDRESS_SPACE_NUM > 1
3045 case KVM_CAP_MULTI_ADDRESS_SPACE
:
3046 return KVM_ADDRESS_SPACE_NUM
;
3048 case KVM_CAP_MAX_VCPU_ID
:
3049 return KVM_MAX_VCPU_ID
;
3053 return kvm_vm_ioctl_check_extension(kvm
, arg
);
3056 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
3057 struct kvm_enable_cap
*cap
)
3062 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
3063 struct kvm_enable_cap
*cap
)
3066 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3067 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
:
3068 if (cap
->flags
|| (cap
->args
[0] & ~1))
3070 kvm
->manual_dirty_log_protect
= cap
->args
[0];
3074 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
3078 static long kvm_vm_ioctl(struct file
*filp
,
3079 unsigned int ioctl
, unsigned long arg
)
3081 struct kvm
*kvm
= filp
->private_data
;
3082 void __user
*argp
= (void __user
*)arg
;
3085 if (kvm
->mm
!= current
->mm
)
3088 case KVM_CREATE_VCPU
:
3089 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3091 case KVM_ENABLE_CAP
: {
3092 struct kvm_enable_cap cap
;
3095 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3097 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
3100 case KVM_SET_USER_MEMORY_REGION
: {
3101 struct kvm_userspace_memory_region kvm_userspace_mem
;
3104 if (copy_from_user(&kvm_userspace_mem
, argp
,
3105 sizeof(kvm_userspace_mem
)))
3108 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3111 case KVM_GET_DIRTY_LOG
: {
3112 struct kvm_dirty_log log
;
3115 if (copy_from_user(&log
, argp
, sizeof(log
)))
3117 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3120 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3121 case KVM_CLEAR_DIRTY_LOG
: {
3122 struct kvm_clear_dirty_log log
;
3125 if (copy_from_user(&log
, argp
, sizeof(log
)))
3127 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
3131 #ifdef CONFIG_KVM_MMIO
3132 case KVM_REGISTER_COALESCED_MMIO
: {
3133 struct kvm_coalesced_mmio_zone zone
;
3136 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3138 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3141 case KVM_UNREGISTER_COALESCED_MMIO
: {
3142 struct kvm_coalesced_mmio_zone zone
;
3145 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3147 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3152 struct kvm_irqfd data
;
3155 if (copy_from_user(&data
, argp
, sizeof(data
)))
3157 r
= kvm_irqfd(kvm
, &data
);
3160 case KVM_IOEVENTFD
: {
3161 struct kvm_ioeventfd data
;
3164 if (copy_from_user(&data
, argp
, sizeof(data
)))
3166 r
= kvm_ioeventfd(kvm
, &data
);
3169 #ifdef CONFIG_HAVE_KVM_MSI
3170 case KVM_SIGNAL_MSI
: {
3174 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3176 r
= kvm_send_userspace_msi(kvm
, &msi
);
3180 #ifdef __KVM_HAVE_IRQ_LINE
3181 case KVM_IRQ_LINE_STATUS
:
3182 case KVM_IRQ_LINE
: {
3183 struct kvm_irq_level irq_event
;
3186 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3189 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3190 ioctl
== KVM_IRQ_LINE_STATUS
);
3195 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3196 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3204 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3205 case KVM_SET_GSI_ROUTING
: {
3206 struct kvm_irq_routing routing
;
3207 struct kvm_irq_routing __user
*urouting
;
3208 struct kvm_irq_routing_entry
*entries
= NULL
;
3211 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3214 if (!kvm_arch_can_set_irq_routing(kvm
))
3216 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3222 entries
= vmalloc(array_size(sizeof(*entries
),
3228 if (copy_from_user(entries
, urouting
->entries
,
3229 routing
.nr
* sizeof(*entries
)))
3230 goto out_free_irq_routing
;
3232 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3234 out_free_irq_routing
:
3238 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3239 case KVM_CREATE_DEVICE
: {
3240 struct kvm_create_device cd
;
3243 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3246 r
= kvm_ioctl_create_device(kvm
, &cd
);
3251 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3257 case KVM_CHECK_EXTENSION
:
3258 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3261 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3267 #ifdef CONFIG_KVM_COMPAT
3268 struct compat_kvm_dirty_log
{
3272 compat_uptr_t dirty_bitmap
; /* one bit per page */
3277 static long kvm_vm_compat_ioctl(struct file
*filp
,
3278 unsigned int ioctl
, unsigned long arg
)
3280 struct kvm
*kvm
= filp
->private_data
;
3283 if (kvm
->mm
!= current
->mm
)
3286 case KVM_GET_DIRTY_LOG
: {
3287 struct compat_kvm_dirty_log compat_log
;
3288 struct kvm_dirty_log log
;
3290 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3291 sizeof(compat_log
)))
3293 log
.slot
= compat_log
.slot
;
3294 log
.padding1
= compat_log
.padding1
;
3295 log
.padding2
= compat_log
.padding2
;
3296 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3298 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3302 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3308 static struct file_operations kvm_vm_fops
= {
3309 .release
= kvm_vm_release
,
3310 .unlocked_ioctl
= kvm_vm_ioctl
,
3311 .llseek
= noop_llseek
,
3312 KVM_COMPAT(kvm_vm_compat_ioctl
),
3315 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3321 kvm
= kvm_create_vm(type
);
3323 return PTR_ERR(kvm
);
3324 #ifdef CONFIG_KVM_MMIO
3325 r
= kvm_coalesced_mmio_init(kvm
);
3329 r
= get_unused_fd_flags(O_CLOEXEC
);
3333 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3341 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3342 * already set, with ->release() being kvm_vm_release(). In error
3343 * cases it will be called by the final fput(file) and will take
3344 * care of doing kvm_put_kvm(kvm).
3346 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3351 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3353 fd_install(r
, file
);
3361 static long kvm_dev_ioctl(struct file
*filp
,
3362 unsigned int ioctl
, unsigned long arg
)
3367 case KVM_GET_API_VERSION
:
3370 r
= KVM_API_VERSION
;
3373 r
= kvm_dev_ioctl_create_vm(arg
);
3375 case KVM_CHECK_EXTENSION
:
3376 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3378 case KVM_GET_VCPU_MMAP_SIZE
:
3381 r
= PAGE_SIZE
; /* struct kvm_run */
3383 r
+= PAGE_SIZE
; /* pio data page */
3385 #ifdef CONFIG_KVM_MMIO
3386 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3389 case KVM_TRACE_ENABLE
:
3390 case KVM_TRACE_PAUSE
:
3391 case KVM_TRACE_DISABLE
:
3395 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3401 static struct file_operations kvm_chardev_ops
= {
3402 .unlocked_ioctl
= kvm_dev_ioctl
,
3403 .llseek
= noop_llseek
,
3404 KVM_COMPAT(kvm_dev_ioctl
),
3407 static struct miscdevice kvm_dev
= {
3413 static void hardware_enable_nolock(void *junk
)
3415 int cpu
= raw_smp_processor_id();
3418 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3421 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3423 r
= kvm_arch_hardware_enable();
3426 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3427 atomic_inc(&hardware_enable_failed
);
3428 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3432 static int kvm_starting_cpu(unsigned int cpu
)
3434 raw_spin_lock(&kvm_count_lock
);
3435 if (kvm_usage_count
)
3436 hardware_enable_nolock(NULL
);
3437 raw_spin_unlock(&kvm_count_lock
);
3441 static void hardware_disable_nolock(void *junk
)
3443 int cpu
= raw_smp_processor_id();
3445 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3447 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3448 kvm_arch_hardware_disable();
3451 static int kvm_dying_cpu(unsigned int cpu
)
3453 raw_spin_lock(&kvm_count_lock
);
3454 if (kvm_usage_count
)
3455 hardware_disable_nolock(NULL
);
3456 raw_spin_unlock(&kvm_count_lock
);
3460 static void hardware_disable_all_nolock(void)
3462 BUG_ON(!kvm_usage_count
);
3465 if (!kvm_usage_count
)
3466 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3469 static void hardware_disable_all(void)
3471 raw_spin_lock(&kvm_count_lock
);
3472 hardware_disable_all_nolock();
3473 raw_spin_unlock(&kvm_count_lock
);
3476 static int hardware_enable_all(void)
3480 raw_spin_lock(&kvm_count_lock
);
3483 if (kvm_usage_count
== 1) {
3484 atomic_set(&hardware_enable_failed
, 0);
3485 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3487 if (atomic_read(&hardware_enable_failed
)) {
3488 hardware_disable_all_nolock();
3493 raw_spin_unlock(&kvm_count_lock
);
3498 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3502 * Some (well, at least mine) BIOSes hang on reboot if
3505 * And Intel TXT required VMX off for all cpu when system shutdown.
3507 pr_info("kvm: exiting hardware virtualization\n");
3508 kvm_rebooting
= true;
3509 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3513 static struct notifier_block kvm_reboot_notifier
= {
3514 .notifier_call
= kvm_reboot
,
3518 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3522 for (i
= 0; i
< bus
->dev_count
; i
++) {
3523 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3525 kvm_iodevice_destructor(pos
);
3530 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3531 const struct kvm_io_range
*r2
)
3533 gpa_t addr1
= r1
->addr
;
3534 gpa_t addr2
= r2
->addr
;
3539 /* If r2->len == 0, match the exact address. If r2->len != 0,
3540 * accept any overlapping write. Any order is acceptable for
3541 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3542 * we process all of them.
3555 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3557 return kvm_io_bus_cmp(p1
, p2
);
3560 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3561 gpa_t addr
, int len
)
3563 struct kvm_io_range
*range
, key
;
3566 key
= (struct kvm_io_range
) {
3571 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3572 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3576 off
= range
- bus
->range
;
3578 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3584 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3585 struct kvm_io_range
*range
, const void *val
)
3589 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3593 while (idx
< bus
->dev_count
&&
3594 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3595 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3604 /* kvm_io_bus_write - called under kvm->slots_lock */
3605 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3606 int len
, const void *val
)
3608 struct kvm_io_bus
*bus
;
3609 struct kvm_io_range range
;
3612 range
= (struct kvm_io_range
) {
3617 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3620 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3621 return r
< 0 ? r
: 0;
3624 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3625 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3626 gpa_t addr
, int len
, const void *val
, long cookie
)
3628 struct kvm_io_bus
*bus
;
3629 struct kvm_io_range range
;
3631 range
= (struct kvm_io_range
) {
3636 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3640 /* First try the device referenced by cookie. */
3641 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3642 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3643 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3648 * cookie contained garbage; fall back to search and return the
3649 * correct cookie value.
3651 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3654 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3655 struct kvm_io_range
*range
, void *val
)
3659 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3663 while (idx
< bus
->dev_count
&&
3664 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3665 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3673 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3675 /* kvm_io_bus_read - called under kvm->slots_lock */
3676 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3679 struct kvm_io_bus
*bus
;
3680 struct kvm_io_range range
;
3683 range
= (struct kvm_io_range
) {
3688 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3691 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3692 return r
< 0 ? r
: 0;
3696 /* Caller must hold slots_lock. */
3697 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3698 int len
, struct kvm_io_device
*dev
)
3701 struct kvm_io_bus
*new_bus
, *bus
;
3702 struct kvm_io_range range
;
3704 bus
= kvm_get_bus(kvm
, bus_idx
);
3708 /* exclude ioeventfd which is limited by maximum fd */
3709 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3712 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3713 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3717 range
= (struct kvm_io_range
) {
3723 for (i
= 0; i
< bus
->dev_count
; i
++)
3724 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
3727 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3728 new_bus
->dev_count
++;
3729 new_bus
->range
[i
] = range
;
3730 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
3731 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3732 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3733 synchronize_srcu_expedited(&kvm
->srcu
);
3739 /* Caller must hold slots_lock. */
3740 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3741 struct kvm_io_device
*dev
)
3744 struct kvm_io_bus
*new_bus
, *bus
;
3746 bus
= kvm_get_bus(kvm
, bus_idx
);
3750 for (i
= 0; i
< bus
->dev_count
; i
++)
3751 if (bus
->range
[i
].dev
== dev
) {
3755 if (i
== bus
->dev_count
)
3758 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3759 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3761 pr_err("kvm: failed to shrink bus, removing it completely\n");
3765 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3766 new_bus
->dev_count
--;
3767 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3768 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3771 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3772 synchronize_srcu_expedited(&kvm
->srcu
);
3777 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3780 struct kvm_io_bus
*bus
;
3781 int dev_idx
, srcu_idx
;
3782 struct kvm_io_device
*iodev
= NULL
;
3784 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3786 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3790 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3794 iodev
= bus
->range
[dev_idx
].dev
;
3797 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3801 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3803 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3804 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3807 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3810 /* The debugfs files are a reference to the kvm struct which
3811 * is still valid when kvm_destroy_vm is called.
3812 * To avoid the race between open and the removal of the debugfs
3813 * directory we test against the users count.
3815 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3818 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3819 kvm_put_kvm(stat_data
->kvm
);
3826 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3828 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3831 simple_attr_release(inode
, file
);
3832 kvm_put_kvm(stat_data
->kvm
);
3837 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3839 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3841 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3846 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3848 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3853 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3858 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3860 __simple_attr_check_format("%llu\n", 0ull);
3861 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3862 vm_stat_clear_per_vm
, "%llu\n");
3865 static const struct file_operations vm_stat_get_per_vm_fops
= {
3866 .owner
= THIS_MODULE
,
3867 .open
= vm_stat_get_per_vm_open
,
3868 .release
= kvm_debugfs_release
,
3869 .read
= simple_attr_read
,
3870 .write
= simple_attr_write
,
3871 .llseek
= no_llseek
,
3874 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3877 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3878 struct kvm_vcpu
*vcpu
;
3882 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3883 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3888 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3891 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3892 struct kvm_vcpu
*vcpu
;
3897 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3898 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3903 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3905 __simple_attr_check_format("%llu\n", 0ull);
3906 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3907 vcpu_stat_clear_per_vm
, "%llu\n");
3910 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3911 .owner
= THIS_MODULE
,
3912 .open
= vcpu_stat_get_per_vm_open
,
3913 .release
= kvm_debugfs_release
,
3914 .read
= simple_attr_read
,
3915 .write
= simple_attr_write
,
3916 .llseek
= no_llseek
,
3919 static const struct file_operations
*stat_fops_per_vm
[] = {
3920 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3921 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3924 static int vm_stat_get(void *_offset
, u64
*val
)
3926 unsigned offset
= (long)_offset
;
3928 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3932 spin_lock(&kvm_lock
);
3933 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3935 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3938 spin_unlock(&kvm_lock
);
3942 static int vm_stat_clear(void *_offset
, u64 val
)
3944 unsigned offset
= (long)_offset
;
3946 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3951 spin_lock(&kvm_lock
);
3952 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3954 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3956 spin_unlock(&kvm_lock
);
3961 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3963 static int vcpu_stat_get(void *_offset
, u64
*val
)
3965 unsigned offset
= (long)_offset
;
3967 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3971 spin_lock(&kvm_lock
);
3972 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3974 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3977 spin_unlock(&kvm_lock
);
3981 static int vcpu_stat_clear(void *_offset
, u64 val
)
3983 unsigned offset
= (long)_offset
;
3985 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3990 spin_lock(&kvm_lock
);
3991 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3993 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3995 spin_unlock(&kvm_lock
);
4000 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
4003 static const struct file_operations
*stat_fops
[] = {
4004 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
4005 [KVM_STAT_VM
] = &vm_stat_fops
,
4008 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
4010 struct kobj_uevent_env
*env
;
4011 unsigned long long created
, active
;
4013 if (!kvm_dev
.this_device
|| !kvm
)
4016 spin_lock(&kvm_lock
);
4017 if (type
== KVM_EVENT_CREATE_VM
) {
4018 kvm_createvm_count
++;
4020 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4023 created
= kvm_createvm_count
;
4024 active
= kvm_active_vms
;
4025 spin_unlock(&kvm_lock
);
4027 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
4031 add_uevent_var(env
, "CREATED=%llu", created
);
4032 add_uevent_var(env
, "COUNT=%llu", active
);
4034 if (type
== KVM_EVENT_CREATE_VM
) {
4035 add_uevent_var(env
, "EVENT=create");
4036 kvm
->userspace_pid
= task_pid_nr(current
);
4037 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4038 add_uevent_var(env
, "EVENT=destroy");
4040 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
4042 if (kvm
->debugfs_dentry
) {
4043 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
4046 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
4048 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
4052 /* no need for checks, since we are adding at most only 5 keys */
4053 env
->envp
[env
->envp_idx
++] = NULL
;
4054 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
4058 static void kvm_init_debug(void)
4060 struct kvm_stats_debugfs_item
*p
;
4062 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
4064 kvm_debugfs_num_entries
= 0;
4065 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
4066 debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
4067 (void *)(long)p
->offset
,
4068 stat_fops
[p
->kind
]);
4072 static int kvm_suspend(void)
4074 if (kvm_usage_count
)
4075 hardware_disable_nolock(NULL
);
4079 static void kvm_resume(void)
4081 if (kvm_usage_count
) {
4082 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
4083 hardware_enable_nolock(NULL
);
4087 static struct syscore_ops kvm_syscore_ops
= {
4088 .suspend
= kvm_suspend
,
4089 .resume
= kvm_resume
,
4093 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4095 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4098 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4100 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4102 if (vcpu
->preempted
)
4103 vcpu
->preempted
= false;
4105 kvm_arch_sched_in(vcpu
, cpu
);
4107 kvm_arch_vcpu_load(vcpu
, cpu
);
4110 static void kvm_sched_out(struct preempt_notifier
*pn
,
4111 struct task_struct
*next
)
4113 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4115 if (current
->state
== TASK_RUNNING
)
4116 vcpu
->preempted
= true;
4117 kvm_arch_vcpu_put(vcpu
);
4120 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4121 struct module
*module
)
4126 r
= kvm_arch_init(opaque
);
4131 * kvm_arch_init makes sure there's at most one caller
4132 * for architectures that support multiple implementations,
4133 * like intel and amd on x86.
4134 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4135 * conflicts in case kvm is already setup for another implementation.
4137 r
= kvm_irqfd_init();
4141 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4146 r
= kvm_arch_hardware_setup();
4150 for_each_online_cpu(cpu
) {
4151 smp_call_function_single(cpu
,
4152 kvm_arch_check_processor_compat
,
4158 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4159 kvm_starting_cpu
, kvm_dying_cpu
);
4162 register_reboot_notifier(&kvm_reboot_notifier
);
4164 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4166 vcpu_align
= __alignof__(struct kvm_vcpu
);
4168 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
4170 offsetof(struct kvm_vcpu
, arch
),
4171 sizeof_field(struct kvm_vcpu
, arch
),
4173 if (!kvm_vcpu_cache
) {
4178 r
= kvm_async_pf_init();
4182 kvm_chardev_ops
.owner
= module
;
4183 kvm_vm_fops
.owner
= module
;
4184 kvm_vcpu_fops
.owner
= module
;
4186 r
= misc_register(&kvm_dev
);
4188 pr_err("kvm: misc device register failed\n");
4192 register_syscore_ops(&kvm_syscore_ops
);
4194 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4195 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4199 r
= kvm_vfio_ops_init();
4205 kvm_async_pf_deinit();
4207 kmem_cache_destroy(kvm_vcpu_cache
);
4209 unregister_reboot_notifier(&kvm_reboot_notifier
);
4210 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4213 kvm_arch_hardware_unsetup();
4215 free_cpumask_var(cpus_hardware_enabled
);
4223 EXPORT_SYMBOL_GPL(kvm_init
);
4227 debugfs_remove_recursive(kvm_debugfs_dir
);
4228 misc_deregister(&kvm_dev
);
4229 kmem_cache_destroy(kvm_vcpu_cache
);
4230 kvm_async_pf_deinit();
4231 unregister_syscore_ops(&kvm_syscore_ops
);
4232 unregister_reboot_notifier(&kvm_reboot_notifier
);
4233 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4234 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4235 kvm_arch_hardware_unsetup();
4238 free_cpumask_var(cpus_hardware_enabled
);
4239 kvm_vfio_ops_exit();
4241 EXPORT_SYMBOL_GPL(kvm_exit
);