1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 #include <linux/kvm_dirty_ring.h>
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
78 module_param(halt_poll_ns
, uint
, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns
);
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow
= 2;
83 module_param(halt_poll_ns_grow
, uint
, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start
= 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start
, uint
, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start
);
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink
;
93 module_param(halt_poll_ns_shrink
, uint
, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
99 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
102 DEFINE_MUTEX(kvm_lock
);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
106 static cpumask_var_t cpus_hardware_enabled
;
107 static int kvm_usage_count
;
108 static atomic_t hardware_enable_failed
;
110 static struct kmem_cache
*kvm_vcpu_cache
;
112 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
113 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_running_vcpu
);
115 struct dentry
*kvm_debugfs_dir
;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
118 static const struct file_operations stat_fops_per_vm
;
120 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file
*file
, unsigned int ioctl
,
135 unsigned long arg
) { return -EINVAL
; }
137 static int kvm_no_compat_open(struct inode
*inode
, struct file
*file
)
139 return is_compat_task() ? -ENODEV
: 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
149 __visible
bool kvm_rebooting
;
150 EXPORT_SYMBOL_GPL(kvm_rebooting
);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
155 static unsigned long long kvm_createvm_count
;
156 static unsigned long long kvm_active_vms
;
158 __weak
void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
159 unsigned long start
, unsigned long end
)
163 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn
)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (!pfn_valid(pfn
) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn
))))
174 return is_zone_device_page(pfn_to_page(pfn
));
177 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
180 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
181 * perspective they are "normal" pages, albeit with slightly different
185 return PageReserved(pfn_to_page(pfn
)) &&
187 !kvm_is_zone_device_pfn(pfn
);
192 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn
)
194 struct page
*page
= pfn_to_page(pfn
);
196 if (!PageTransCompoundMap(page
))
199 return is_transparent_hugepage(compound_head(page
));
203 * Switches to specified vcpu, until a matching vcpu_put()
205 void vcpu_load(struct kvm_vcpu
*vcpu
)
209 __this_cpu_write(kvm_running_vcpu
, vcpu
);
210 preempt_notifier_register(&vcpu
->preempt_notifier
);
211 kvm_arch_vcpu_load(vcpu
, cpu
);
214 EXPORT_SYMBOL_GPL(vcpu_load
);
216 void vcpu_put(struct kvm_vcpu
*vcpu
)
219 kvm_arch_vcpu_put(vcpu
);
220 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
221 __this_cpu_write(kvm_running_vcpu
, NULL
);
224 EXPORT_SYMBOL_GPL(vcpu_put
);
226 /* TODO: merge with kvm_arch_vcpu_should_kick */
227 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
229 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
232 * We need to wait for the VCPU to reenable interrupts and get out of
233 * READING_SHADOW_PAGE_TABLES mode.
235 if (req
& KVM_REQUEST_WAIT
)
236 return mode
!= OUTSIDE_GUEST_MODE
;
239 * Need to kick a running VCPU, but otherwise there is nothing to do.
241 return mode
== IN_GUEST_MODE
;
244 static void ack_flush(void *_completed
)
248 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
251 cpus
= cpu_online_mask
;
253 if (cpumask_empty(cpus
))
256 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
260 bool kvm_make_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
261 struct kvm_vcpu
*except
,
262 unsigned long *vcpu_bitmap
, cpumask_var_t tmp
)
265 struct kvm_vcpu
*vcpu
;
270 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
271 if ((vcpu_bitmap
&& !test_bit(i
, vcpu_bitmap
)) ||
275 kvm_make_request(req
, vcpu
);
278 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
281 if (tmp
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
282 kvm_request_needs_ipi(vcpu
, req
))
283 __cpumask_set_cpu(cpu
, tmp
);
286 called
= kvm_kick_many_cpus(tmp
, !!(req
& KVM_REQUEST_WAIT
));
292 bool kvm_make_all_cpus_request_except(struct kvm
*kvm
, unsigned int req
,
293 struct kvm_vcpu
*except
)
298 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
300 called
= kvm_make_vcpus_request_mask(kvm
, req
, except
, NULL
, cpus
);
302 free_cpumask_var(cpus
);
306 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
308 return kvm_make_all_cpus_request_except(kvm
, req
, NULL
);
310 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request
);
312 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
313 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
316 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
317 * kvm_make_all_cpus_request.
319 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
322 * We want to publish modifications to the page tables before reading
323 * mode. Pairs with a memory barrier in arch-specific code.
324 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
325 * and smp_mb in walk_shadow_page_lockless_begin/end.
326 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
328 * There is already an smp_mb__after_atomic() before
329 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
332 if (!kvm_arch_flush_remote_tlb(kvm
)
333 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
334 ++kvm
->stat
.generic
.remote_tlb_flush
;
335 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
337 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
340 void kvm_reload_remote_mmus(struct kvm
*kvm
)
342 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
345 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
346 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache
*mc
,
349 gfp_flags
|= mc
->gfp_zero
;
352 return kmem_cache_alloc(mc
->kmem_cache
, gfp_flags
);
354 return (void *)__get_free_page(gfp_flags
);
357 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int min
)
361 if (mc
->nobjs
>= min
)
363 while (mc
->nobjs
< ARRAY_SIZE(mc
->objects
)) {
364 obj
= mmu_memory_cache_alloc_obj(mc
, GFP_KERNEL_ACCOUNT
);
366 return mc
->nobjs
>= min
? 0 : -ENOMEM
;
367 mc
->objects
[mc
->nobjs
++] = obj
;
372 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache
*mc
)
377 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
381 kmem_cache_free(mc
->kmem_cache
, mc
->objects
[--mc
->nobjs
]);
383 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
387 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
)
391 if (WARN_ON(!mc
->nobjs
))
392 p
= mmu_memory_cache_alloc_obj(mc
, GFP_ATOMIC
| __GFP_ACCOUNT
);
394 p
= mc
->objects
[--mc
->nobjs
];
400 static void kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
402 mutex_init(&vcpu
->mutex
);
407 rcuwait_init(&vcpu
->wait
);
408 kvm_async_pf_vcpu_init(vcpu
);
411 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
413 kvm_vcpu_set_in_spin_loop(vcpu
, false);
414 kvm_vcpu_set_dy_eligible(vcpu
, false);
415 vcpu
->preempted
= false;
417 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
420 void kvm_vcpu_destroy(struct kvm_vcpu
*vcpu
)
422 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
423 kvm_arch_vcpu_destroy(vcpu
);
426 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
427 * the vcpu->pid pointer, and at destruction time all file descriptors
430 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
432 free_page((unsigned long)vcpu
->run
);
433 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
435 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy
);
437 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
438 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
440 return container_of(mn
, struct kvm
, mmu_notifier
);
443 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier
*mn
,
444 struct mm_struct
*mm
,
445 unsigned long start
, unsigned long end
)
447 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
450 idx
= srcu_read_lock(&kvm
->srcu
);
451 kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
);
452 srcu_read_unlock(&kvm
->srcu
, idx
);
455 typedef bool (*hva_handler_t
)(struct kvm
*kvm
, struct kvm_gfn_range
*range
);
457 typedef void (*on_lock_fn_t
)(struct kvm
*kvm
, unsigned long start
,
460 struct kvm_hva_range
{
464 hva_handler_t handler
;
465 on_lock_fn_t on_lock
;
471 * Use a dedicated stub instead of NULL to indicate that there is no callback
472 * function/handler. The compiler technically can't guarantee that a real
473 * function will have a non-zero address, and so it will generate code to
474 * check for !NULL, whereas comparing against a stub will be elided at compile
475 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
477 static void kvm_null_fn(void)
481 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
483 static __always_inline
int __kvm_handle_hva_range(struct kvm
*kvm
,
484 const struct kvm_hva_range
*range
)
486 bool ret
= false, locked
= false;
487 struct kvm_gfn_range gfn_range
;
488 struct kvm_memory_slot
*slot
;
489 struct kvm_memslots
*slots
;
492 /* A null handler is allowed if and only if on_lock() is provided. */
493 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range
->on_lock
) &&
494 IS_KVM_NULL_FN(range
->handler
)))
497 idx
= srcu_read_lock(&kvm
->srcu
);
499 /* The on_lock() path does not yet support lock elision. */
500 if (!IS_KVM_NULL_FN(range
->on_lock
)) {
504 range
->on_lock(kvm
, range
->start
, range
->end
);
506 if (IS_KVM_NULL_FN(range
->handler
))
510 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
511 slots
= __kvm_memslots(kvm
, i
);
512 kvm_for_each_memslot(slot
, slots
) {
513 unsigned long hva_start
, hva_end
;
515 hva_start
= max(range
->start
, slot
->userspace_addr
);
516 hva_end
= min(range
->end
, slot
->userspace_addr
+
517 (slot
->npages
<< PAGE_SHIFT
));
518 if (hva_start
>= hva_end
)
522 * To optimize for the likely case where the address
523 * range is covered by zero or one memslots, don't
524 * bother making these conditional (to avoid writes on
525 * the second or later invocation of the handler).
527 gfn_range
.pte
= range
->pte
;
528 gfn_range
.may_block
= range
->may_block
;
531 * {gfn(page) | page intersects with [hva_start, hva_end)} =
532 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
534 gfn_range
.start
= hva_to_gfn_memslot(hva_start
, slot
);
535 gfn_range
.end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, slot
);
536 gfn_range
.slot
= slot
;
542 ret
|= range
->handler(kvm
, &gfn_range
);
546 if (range
->flush_on_ret
&& (ret
|| kvm
->tlbs_dirty
))
547 kvm_flush_remote_tlbs(kvm
);
553 srcu_read_unlock(&kvm
->srcu
, idx
);
555 /* The notifiers are averse to booleans. :-( */
559 static __always_inline
int kvm_handle_hva_range(struct mmu_notifier
*mn
,
563 hva_handler_t handler
)
565 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
566 const struct kvm_hva_range range
= {
571 .on_lock
= (void *)kvm_null_fn
,
572 .flush_on_ret
= true,
576 return __kvm_handle_hva_range(kvm
, &range
);
579 static __always_inline
int kvm_handle_hva_range_no_flush(struct mmu_notifier
*mn
,
582 hva_handler_t handler
)
584 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
585 const struct kvm_hva_range range
= {
590 .on_lock
= (void *)kvm_null_fn
,
591 .flush_on_ret
= false,
595 return __kvm_handle_hva_range(kvm
, &range
);
597 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
598 struct mm_struct
*mm
,
599 unsigned long address
,
602 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
604 trace_kvm_set_spte_hva(address
);
607 * .change_pte() must be surrounded by .invalidate_range_{start,end}(),
608 * and so always runs with an elevated notifier count. This obviates
609 * the need to bump the sequence count.
611 WARN_ON_ONCE(!kvm
->mmu_notifier_count
);
613 kvm_handle_hva_range(mn
, address
, address
+ 1, pte
, kvm_set_spte_gfn
);
616 static void kvm_inc_notifier_count(struct kvm
*kvm
, unsigned long start
,
620 * The count increase must become visible at unlock time as no
621 * spte can be established without taking the mmu_lock and
622 * count is also read inside the mmu_lock critical section.
624 kvm
->mmu_notifier_count
++;
625 if (likely(kvm
->mmu_notifier_count
== 1)) {
626 kvm
->mmu_notifier_range_start
= start
;
627 kvm
->mmu_notifier_range_end
= end
;
630 * Fully tracking multiple concurrent ranges has dimishing
631 * returns. Keep things simple and just find the minimal range
632 * which includes the current and new ranges. As there won't be
633 * enough information to subtract a range after its invalidate
634 * completes, any ranges invalidated concurrently will
635 * accumulate and persist until all outstanding invalidates
638 kvm
->mmu_notifier_range_start
=
639 min(kvm
->mmu_notifier_range_start
, start
);
640 kvm
->mmu_notifier_range_end
=
641 max(kvm
->mmu_notifier_range_end
, end
);
645 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
646 const struct mmu_notifier_range
*range
)
648 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
649 const struct kvm_hva_range hva_range
= {
650 .start
= range
->start
,
653 .handler
= kvm_unmap_gfn_range
,
654 .on_lock
= kvm_inc_notifier_count
,
655 .flush_on_ret
= true,
656 .may_block
= mmu_notifier_range_blockable(range
),
659 trace_kvm_unmap_hva_range(range
->start
, range
->end
);
661 __kvm_handle_hva_range(kvm
, &hva_range
);
666 static void kvm_dec_notifier_count(struct kvm
*kvm
, unsigned long start
,
670 * This sequence increase will notify the kvm page fault that
671 * the page that is going to be mapped in the spte could have
674 kvm
->mmu_notifier_seq
++;
677 * The above sequence increase must be visible before the
678 * below count decrease, which is ensured by the smp_wmb above
679 * in conjunction with the smp_rmb in mmu_notifier_retry().
681 kvm
->mmu_notifier_count
--;
684 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
685 const struct mmu_notifier_range
*range
)
687 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
688 const struct kvm_hva_range hva_range
= {
689 .start
= range
->start
,
692 .handler
= (void *)kvm_null_fn
,
693 .on_lock
= kvm_dec_notifier_count
,
694 .flush_on_ret
= false,
695 .may_block
= mmu_notifier_range_blockable(range
),
698 __kvm_handle_hva_range(kvm
, &hva_range
);
700 BUG_ON(kvm
->mmu_notifier_count
< 0);
703 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
704 struct mm_struct
*mm
,
708 trace_kvm_age_hva(start
, end
);
710 return kvm_handle_hva_range(mn
, start
, end
, __pte(0), kvm_age_gfn
);
713 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
714 struct mm_struct
*mm
,
718 trace_kvm_age_hva(start
, end
);
721 * Even though we do not flush TLB, this will still adversely
722 * affect performance on pre-Haswell Intel EPT, where there is
723 * no EPT Access Bit to clear so that we have to tear down EPT
724 * tables instead. If we find this unacceptable, we can always
725 * add a parameter to kvm_age_hva so that it effectively doesn't
726 * do anything on clear_young.
728 * Also note that currently we never issue secondary TLB flushes
729 * from clear_young, leaving this job up to the regular system
730 * cadence. If we find this inaccurate, we might come up with a
731 * more sophisticated heuristic later.
733 return kvm_handle_hva_range_no_flush(mn
, start
, end
, kvm_age_gfn
);
736 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
737 struct mm_struct
*mm
,
738 unsigned long address
)
740 trace_kvm_test_age_hva(address
);
742 return kvm_handle_hva_range_no_flush(mn
, address
, address
+ 1,
746 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
747 struct mm_struct
*mm
)
749 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
752 idx
= srcu_read_lock(&kvm
->srcu
);
753 kvm_arch_flush_shadow_all(kvm
);
754 srcu_read_unlock(&kvm
->srcu
, idx
);
757 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
758 .invalidate_range
= kvm_mmu_notifier_invalidate_range
,
759 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
760 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
761 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
762 .clear_young
= kvm_mmu_notifier_clear_young
,
763 .test_young
= kvm_mmu_notifier_test_young
,
764 .change_pte
= kvm_mmu_notifier_change_pte
,
765 .release
= kvm_mmu_notifier_release
,
768 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
770 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
771 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
774 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
776 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
781 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
783 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
784 static int kvm_pm_notifier_call(struct notifier_block
*bl
,
788 struct kvm
*kvm
= container_of(bl
, struct kvm
, pm_notifier
);
790 return kvm_arch_pm_notifier(kvm
, state
);
793 static void kvm_init_pm_notifier(struct kvm
*kvm
)
795 kvm
->pm_notifier
.notifier_call
= kvm_pm_notifier_call
;
796 /* Suspend KVM before we suspend ftrace, RCU, etc. */
797 kvm
->pm_notifier
.priority
= INT_MAX
;
798 register_pm_notifier(&kvm
->pm_notifier
);
801 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
803 unregister_pm_notifier(&kvm
->pm_notifier
);
805 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
806 static void kvm_init_pm_notifier(struct kvm
*kvm
)
810 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
813 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
815 static struct kvm_memslots
*kvm_alloc_memslots(void)
818 struct kvm_memslots
*slots
;
820 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL_ACCOUNT
);
824 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
825 slots
->id_to_index
[i
] = -1;
830 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
832 if (!memslot
->dirty_bitmap
)
835 kvfree(memslot
->dirty_bitmap
);
836 memslot
->dirty_bitmap
= NULL
;
839 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
841 kvm_destroy_dirty_bitmap(slot
);
843 kvm_arch_free_memslot(kvm
, slot
);
849 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
851 struct kvm_memory_slot
*memslot
;
856 kvm_for_each_memslot(memslot
, slots
)
857 kvm_free_memslot(kvm
, memslot
);
862 static umode_t
kvm_stats_debugfs_mode(const struct _kvm_stats_desc
*pdesc
)
864 switch (pdesc
->desc
.flags
& KVM_STATS_TYPE_MASK
) {
865 case KVM_STATS_TYPE_INSTANT
:
867 case KVM_STATS_TYPE_CUMULATIVE
:
868 case KVM_STATS_TYPE_PEAK
:
875 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
878 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
879 kvm_vcpu_stats_header
.num_desc
;
881 if (!kvm
->debugfs_dentry
)
884 debugfs_remove_recursive(kvm
->debugfs_dentry
);
886 if (kvm
->debugfs_stat_data
) {
887 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
888 kfree(kvm
->debugfs_stat_data
[i
]);
889 kfree(kvm
->debugfs_stat_data
);
893 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
895 char dir_name
[ITOA_MAX_LEN
* 2];
896 struct kvm_stat_data
*stat_data
;
897 const struct _kvm_stats_desc
*pdesc
;
899 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
900 kvm_vcpu_stats_header
.num_desc
;
902 if (!debugfs_initialized())
905 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
906 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
908 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
909 sizeof(*kvm
->debugfs_stat_data
),
911 if (!kvm
->debugfs_stat_data
)
914 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
915 pdesc
= &kvm_vm_stats_desc
[i
];
916 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
920 stat_data
->kvm
= kvm
;
921 stat_data
->desc
= pdesc
;
922 stat_data
->kind
= KVM_STAT_VM
;
923 kvm
->debugfs_stat_data
[i
] = stat_data
;
924 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
925 kvm
->debugfs_dentry
, stat_data
,
929 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
930 pdesc
= &kvm_vcpu_stats_desc
[i
];
931 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
935 stat_data
->kvm
= kvm
;
936 stat_data
->desc
= pdesc
;
937 stat_data
->kind
= KVM_STAT_VCPU
;
938 kvm
->debugfs_stat_data
[i
+ kvm_vm_stats_header
.num_desc
] = stat_data
;
939 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
940 kvm
->debugfs_dentry
, stat_data
,
947 * Called after the VM is otherwise initialized, but just before adding it to
950 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
956 * Called just after removing the VM from the vm_list, but before doing any
959 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
963 static struct kvm
*kvm_create_vm(unsigned long type
)
965 struct kvm
*kvm
= kvm_arch_alloc_vm();
970 return ERR_PTR(-ENOMEM
);
972 KVM_MMU_LOCK_INIT(kvm
);
974 kvm
->mm
= current
->mm
;
975 kvm_eventfd_init(kvm
);
976 mutex_init(&kvm
->lock
);
977 mutex_init(&kvm
->irq_lock
);
978 mutex_init(&kvm
->slots_lock
);
979 mutex_init(&kvm
->slots_arch_lock
);
980 INIT_LIST_HEAD(&kvm
->devices
);
982 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
984 if (init_srcu_struct(&kvm
->srcu
))
985 goto out_err_no_srcu
;
986 if (init_srcu_struct(&kvm
->irq_srcu
))
987 goto out_err_no_irq_srcu
;
989 refcount_set(&kvm
->users_count
, 1);
990 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
991 struct kvm_memslots
*slots
= kvm_alloc_memslots();
994 goto out_err_no_arch_destroy_vm
;
995 /* Generations must be different for each address space. */
996 slots
->generation
= i
;
997 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
1000 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1001 rcu_assign_pointer(kvm
->buses
[i
],
1002 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
1004 goto out_err_no_arch_destroy_vm
;
1007 kvm
->max_halt_poll_ns
= halt_poll_ns
;
1009 r
= kvm_arch_init_vm(kvm
, type
);
1011 goto out_err_no_arch_destroy_vm
;
1013 r
= hardware_enable_all();
1015 goto out_err_no_disable
;
1017 #ifdef CONFIG_HAVE_KVM_IRQFD
1018 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
1021 r
= kvm_init_mmu_notifier(kvm
);
1023 goto out_err_no_mmu_notifier
;
1025 r
= kvm_arch_post_init_vm(kvm
);
1029 mutex_lock(&kvm_lock
);
1030 list_add(&kvm
->vm_list
, &vm_list
);
1031 mutex_unlock(&kvm_lock
);
1033 preempt_notifier_inc();
1034 kvm_init_pm_notifier(kvm
);
1039 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1040 if (kvm
->mmu_notifier
.ops
)
1041 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
1043 out_err_no_mmu_notifier
:
1044 hardware_disable_all();
1046 kvm_arch_destroy_vm(kvm
);
1047 out_err_no_arch_destroy_vm
:
1048 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
1049 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
1050 kfree(kvm_get_bus(kvm
, i
));
1051 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
1052 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
1053 cleanup_srcu_struct(&kvm
->irq_srcu
);
1054 out_err_no_irq_srcu
:
1055 cleanup_srcu_struct(&kvm
->srcu
);
1057 kvm_arch_free_vm(kvm
);
1058 mmdrop(current
->mm
);
1062 static void kvm_destroy_devices(struct kvm
*kvm
)
1064 struct kvm_device
*dev
, *tmp
;
1067 * We do not need to take the kvm->lock here, because nobody else
1068 * has a reference to the struct kvm at this point and therefore
1069 * cannot access the devices list anyhow.
1071 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
1072 list_del(&dev
->vm_node
);
1073 dev
->ops
->destroy(dev
);
1077 static void kvm_destroy_vm(struct kvm
*kvm
)
1080 struct mm_struct
*mm
= kvm
->mm
;
1082 kvm_destroy_pm_notifier(kvm
);
1083 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
1084 kvm_destroy_vm_debugfs(kvm
);
1085 kvm_arch_sync_events(kvm
);
1086 mutex_lock(&kvm_lock
);
1087 list_del(&kvm
->vm_list
);
1088 mutex_unlock(&kvm_lock
);
1089 kvm_arch_pre_destroy_vm(kvm
);
1091 kvm_free_irq_routing(kvm
);
1092 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1093 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
1096 kvm_io_bus_destroy(bus
);
1097 kvm
->buses
[i
] = NULL
;
1099 kvm_coalesced_mmio_free(kvm
);
1100 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1101 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1103 kvm_arch_flush_shadow_all(kvm
);
1105 kvm_arch_destroy_vm(kvm
);
1106 kvm_destroy_devices(kvm
);
1107 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
1108 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
1109 cleanup_srcu_struct(&kvm
->irq_srcu
);
1110 cleanup_srcu_struct(&kvm
->srcu
);
1111 kvm_arch_free_vm(kvm
);
1112 preempt_notifier_dec();
1113 hardware_disable_all();
1117 void kvm_get_kvm(struct kvm
*kvm
)
1119 refcount_inc(&kvm
->users_count
);
1121 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1123 void kvm_put_kvm(struct kvm
*kvm
)
1125 if (refcount_dec_and_test(&kvm
->users_count
))
1126 kvm_destroy_vm(kvm
);
1128 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1131 * Used to put a reference that was taken on behalf of an object associated
1132 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1133 * of the new file descriptor fails and the reference cannot be transferred to
1134 * its final owner. In such cases, the caller is still actively using @kvm and
1135 * will fail miserably if the refcount unexpectedly hits zero.
1137 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
1139 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
1141 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
1143 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1145 struct kvm
*kvm
= filp
->private_data
;
1147 kvm_irqfd_release(kvm
);
1154 * Allocation size is twice as large as the actual dirty bitmap size.
1155 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1157 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1159 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
1161 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL_ACCOUNT
);
1162 if (!memslot
->dirty_bitmap
)
1169 * Delete a memslot by decrementing the number of used slots and shifting all
1170 * other entries in the array forward one spot.
1172 static inline void kvm_memslot_delete(struct kvm_memslots
*slots
,
1173 struct kvm_memory_slot
*memslot
)
1175 struct kvm_memory_slot
*mslots
= slots
->memslots
;
1178 if (WARN_ON(slots
->id_to_index
[memslot
->id
] == -1))
1181 slots
->used_slots
--;
1183 if (atomic_read(&slots
->lru_slot
) >= slots
->used_slots
)
1184 atomic_set(&slots
->lru_slot
, 0);
1186 for (i
= slots
->id_to_index
[memslot
->id
]; i
< slots
->used_slots
; i
++) {
1187 mslots
[i
] = mslots
[i
+ 1];
1188 slots
->id_to_index
[mslots
[i
].id
] = i
;
1190 mslots
[i
] = *memslot
;
1191 slots
->id_to_index
[memslot
->id
] = -1;
1195 * "Insert" a new memslot by incrementing the number of used slots. Returns
1196 * the new slot's initial index into the memslots array.
1198 static inline int kvm_memslot_insert_back(struct kvm_memslots
*slots
)
1200 return slots
->used_slots
++;
1204 * Move a changed memslot backwards in the array by shifting existing slots
1205 * with a higher GFN toward the front of the array. Note, the changed memslot
1206 * itself is not preserved in the array, i.e. not swapped at this time, only
1207 * its new index into the array is tracked. Returns the changed memslot's
1208 * current index into the memslots array.
1210 static inline int kvm_memslot_move_backward(struct kvm_memslots
*slots
,
1211 struct kvm_memory_slot
*memslot
)
1213 struct kvm_memory_slot
*mslots
= slots
->memslots
;
1216 if (WARN_ON_ONCE(slots
->id_to_index
[memslot
->id
] == -1) ||
1217 WARN_ON_ONCE(!slots
->used_slots
))
1221 * Move the target memslot backward in the array by shifting existing
1222 * memslots with a higher GFN (than the target memslot) towards the
1223 * front of the array.
1225 for (i
= slots
->id_to_index
[memslot
->id
]; i
< slots
->used_slots
- 1; i
++) {
1226 if (memslot
->base_gfn
> mslots
[i
+ 1].base_gfn
)
1229 WARN_ON_ONCE(memslot
->base_gfn
== mslots
[i
+ 1].base_gfn
);
1231 /* Shift the next memslot forward one and update its index. */
1232 mslots
[i
] = mslots
[i
+ 1];
1233 slots
->id_to_index
[mslots
[i
].id
] = i
;
1239 * Move a changed memslot forwards in the array by shifting existing slots with
1240 * a lower GFN toward the back of the array. Note, the changed memslot itself
1241 * is not preserved in the array, i.e. not swapped at this time, only its new
1242 * index into the array is tracked. Returns the changed memslot's final index
1243 * into the memslots array.
1245 static inline int kvm_memslot_move_forward(struct kvm_memslots
*slots
,
1246 struct kvm_memory_slot
*memslot
,
1249 struct kvm_memory_slot
*mslots
= slots
->memslots
;
1252 for (i
= start
; i
> 0; i
--) {
1253 if (memslot
->base_gfn
< mslots
[i
- 1].base_gfn
)
1256 WARN_ON_ONCE(memslot
->base_gfn
== mslots
[i
- 1].base_gfn
);
1258 /* Shift the next memslot back one and update its index. */
1259 mslots
[i
] = mslots
[i
- 1];
1260 slots
->id_to_index
[mslots
[i
].id
] = i
;
1266 * Re-sort memslots based on their GFN to account for an added, deleted, or
1267 * moved memslot. Sorting memslots by GFN allows using a binary search during
1270 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1271 * at memslots[0] has the highest GFN.
1273 * The sorting algorithm takes advantage of having initially sorted memslots
1274 * and knowing the position of the changed memslot. Sorting is also optimized
1275 * by not swapping the updated memslot and instead only shifting other memslots
1276 * and tracking the new index for the update memslot. Only once its final
1277 * index is known is the updated memslot copied into its position in the array.
1279 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1280 * the end of the array.
1282 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1283 * end of the array and then it forward to its correct location.
1285 * - When moving a memslot, the algorithm first moves the updated memslot
1286 * backward to handle the scenario where the memslot's GFN was changed to a
1287 * lower value. update_memslots() then falls through and runs the same flow
1288 * as creating a memslot to move the memslot forward to handle the scenario
1289 * where its GFN was changed to a higher value.
1291 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1292 * historical reasons. Originally, invalid memslots where denoted by having
1293 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1294 * to the end of the array. The current algorithm uses dedicated logic to
1295 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1297 * The other historical motiviation for highest->lowest was to improve the
1298 * performance of memslot lookup. KVM originally used a linear search starting
1299 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1300 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1301 * single memslot above the 4gb boundary. As the largest memslot is also the
1302 * most likely to be referenced, sorting it to the front of the array was
1303 * advantageous. The current binary search starts from the middle of the array
1304 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1306 static void update_memslots(struct kvm_memslots
*slots
,
1307 struct kvm_memory_slot
*memslot
,
1308 enum kvm_mr_change change
)
1312 if (change
== KVM_MR_DELETE
) {
1313 kvm_memslot_delete(slots
, memslot
);
1315 if (change
== KVM_MR_CREATE
)
1316 i
= kvm_memslot_insert_back(slots
);
1318 i
= kvm_memslot_move_backward(slots
, memslot
);
1319 i
= kvm_memslot_move_forward(slots
, memslot
, i
);
1322 * Copy the memslot to its new position in memslots and update
1323 * its index accordingly.
1325 slots
->memslots
[i
] = *memslot
;
1326 slots
->id_to_index
[memslot
->id
] = i
;
1330 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
1332 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1334 #ifdef __KVM_HAVE_READONLY_MEM
1335 valid_flags
|= KVM_MEM_READONLY
;
1338 if (mem
->flags
& ~valid_flags
)
1344 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
1345 int as_id
, struct kvm_memslots
*slots
)
1347 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
1348 u64 gen
= old_memslots
->generation
;
1350 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1351 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1353 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1356 * Acquired in kvm_set_memslot. Must be released before synchronize
1357 * SRCU below in order to avoid deadlock with another thread
1358 * acquiring the slots_arch_lock in an srcu critical section.
1360 mutex_unlock(&kvm
->slots_arch_lock
);
1362 synchronize_srcu_expedited(&kvm
->srcu
);
1365 * Increment the new memslot generation a second time, dropping the
1366 * update in-progress flag and incrementing the generation based on
1367 * the number of address spaces. This provides a unique and easily
1368 * identifiable generation number while the memslots are in flux.
1370 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1373 * Generations must be unique even across address spaces. We do not need
1374 * a global counter for that, instead the generation space is evenly split
1375 * across address spaces. For example, with two address spaces, address
1376 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1377 * use generations 1, 3, 5, ...
1379 gen
+= KVM_ADDRESS_SPACE_NUM
;
1381 kvm_arch_memslots_updated(kvm
, gen
);
1383 slots
->generation
= gen
;
1385 return old_memslots
;
1388 static size_t kvm_memslots_size(int slots
)
1390 return sizeof(struct kvm_memslots
) +
1391 (sizeof(struct kvm_memory_slot
) * slots
);
1394 static void kvm_copy_memslots(struct kvm_memslots
*to
,
1395 struct kvm_memslots
*from
)
1397 memcpy(to
, from
, kvm_memslots_size(from
->used_slots
));
1401 * Note, at a minimum, the current number of used slots must be allocated, even
1402 * when deleting a memslot, as we need a complete duplicate of the memslots for
1403 * use when invalidating a memslot prior to deleting/moving the memslot.
1405 static struct kvm_memslots
*kvm_dup_memslots(struct kvm_memslots
*old
,
1406 enum kvm_mr_change change
)
1408 struct kvm_memslots
*slots
;
1411 if (change
== KVM_MR_CREATE
)
1412 new_size
= kvm_memslots_size(old
->used_slots
+ 1);
1414 new_size
= kvm_memslots_size(old
->used_slots
);
1416 slots
= kvzalloc(new_size
, GFP_KERNEL_ACCOUNT
);
1418 kvm_copy_memslots(slots
, old
);
1423 static int kvm_set_memslot(struct kvm
*kvm
,
1424 const struct kvm_userspace_memory_region
*mem
,
1425 struct kvm_memory_slot
*old
,
1426 struct kvm_memory_slot
*new, int as_id
,
1427 enum kvm_mr_change change
)
1429 struct kvm_memory_slot
*slot
;
1430 struct kvm_memslots
*slots
;
1434 * Released in install_new_memslots.
1436 * Must be held from before the current memslots are copied until
1437 * after the new memslots are installed with rcu_assign_pointer,
1438 * then released before the synchronize srcu in install_new_memslots.
1440 * When modifying memslots outside of the slots_lock, must be held
1441 * before reading the pointer to the current memslots until after all
1442 * changes to those memslots are complete.
1444 * These rules ensure that installing new memslots does not lose
1445 * changes made to the previous memslots.
1447 mutex_lock(&kvm
->slots_arch_lock
);
1449 slots
= kvm_dup_memslots(__kvm_memslots(kvm
, as_id
), change
);
1451 mutex_unlock(&kvm
->slots_arch_lock
);
1455 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1457 * Note, the INVALID flag needs to be in the appropriate entry
1458 * in the freshly allocated memslots, not in @old or @new.
1460 slot
= id_to_memslot(slots
, old
->id
);
1461 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1464 * We can re-use the memory from the old memslots.
1465 * It will be overwritten with a copy of the new memslots
1466 * after reacquiring the slots_arch_lock below.
1468 slots
= install_new_memslots(kvm
, as_id
, slots
);
1470 /* From this point no new shadow pages pointing to a deleted,
1471 * or moved, memslot will be created.
1473 * validation of sp->gfn happens in:
1474 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1475 * - kvm_is_visible_gfn (mmu_check_root)
1477 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1479 /* Released in install_new_memslots. */
1480 mutex_lock(&kvm
->slots_arch_lock
);
1483 * The arch-specific fields of the memslots could have changed
1484 * between releasing the slots_arch_lock in
1485 * install_new_memslots and here, so get a fresh copy of the
1488 kvm_copy_memslots(slots
, __kvm_memslots(kvm
, as_id
));
1491 r
= kvm_arch_prepare_memory_region(kvm
, new, mem
, change
);
1495 update_memslots(slots
, new, change
);
1496 slots
= install_new_memslots(kvm
, as_id
, slots
);
1498 kvm_arch_commit_memory_region(kvm
, mem
, old
, new, change
);
1504 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1505 slot
= id_to_memslot(slots
, old
->id
);
1506 slot
->flags
&= ~KVM_MEMSLOT_INVALID
;
1507 slots
= install_new_memslots(kvm
, as_id
, slots
);
1509 mutex_unlock(&kvm
->slots_arch_lock
);
1515 static int kvm_delete_memslot(struct kvm
*kvm
,
1516 const struct kvm_userspace_memory_region
*mem
,
1517 struct kvm_memory_slot
*old
, int as_id
)
1519 struct kvm_memory_slot
new;
1525 memset(&new, 0, sizeof(new));
1528 * This is only for debugging purpose; it should never be referenced
1529 * for a removed memslot.
1533 r
= kvm_set_memslot(kvm
, mem
, old
, &new, as_id
, KVM_MR_DELETE
);
1537 kvm_free_memslot(kvm
, old
);
1542 * Allocate some memory and give it an address in the guest physical address
1545 * Discontiguous memory is allowed, mostly for framebuffers.
1547 * Must be called holding kvm->slots_lock for write.
1549 int __kvm_set_memory_region(struct kvm
*kvm
,
1550 const struct kvm_userspace_memory_region
*mem
)
1552 struct kvm_memory_slot old
, new;
1553 struct kvm_memory_slot
*tmp
;
1554 enum kvm_mr_change change
;
1558 r
= check_memory_region_flags(mem
);
1562 as_id
= mem
->slot
>> 16;
1563 id
= (u16
)mem
->slot
;
1565 /* General sanity checks */
1566 if (mem
->memory_size
& (PAGE_SIZE
- 1))
1568 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
1570 /* We can read the guest memory with __xxx_user() later on. */
1571 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
1572 (mem
->userspace_addr
!= untagged_addr(mem
->userspace_addr
)) ||
1573 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
1576 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
1578 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
1582 * Make a full copy of the old memslot, the pointer will become stale
1583 * when the memslots are re-sorted by update_memslots(), and the old
1584 * memslot needs to be referenced after calling update_memslots(), e.g.
1585 * to free its resources and for arch specific behavior.
1587 tmp
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
1592 memset(&old
, 0, sizeof(old
));
1596 if (!mem
->memory_size
)
1597 return kvm_delete_memslot(kvm
, mem
, &old
, as_id
);
1601 new.base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
1602 new.npages
= mem
->memory_size
>> PAGE_SHIFT
;
1603 new.flags
= mem
->flags
;
1604 new.userspace_addr
= mem
->userspace_addr
;
1606 if (new.npages
> KVM_MEM_MAX_NR_PAGES
)
1610 change
= KVM_MR_CREATE
;
1611 new.dirty_bitmap
= NULL
;
1612 memset(&new.arch
, 0, sizeof(new.arch
));
1613 } else { /* Modify an existing slot. */
1614 if ((new.userspace_addr
!= old
.userspace_addr
) ||
1615 (new.npages
!= old
.npages
) ||
1616 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
1619 if (new.base_gfn
!= old
.base_gfn
)
1620 change
= KVM_MR_MOVE
;
1621 else if (new.flags
!= old
.flags
)
1622 change
= KVM_MR_FLAGS_ONLY
;
1623 else /* Nothing to change. */
1626 /* Copy dirty_bitmap and arch from the current memslot. */
1627 new.dirty_bitmap
= old
.dirty_bitmap
;
1628 memcpy(&new.arch
, &old
.arch
, sizeof(new.arch
));
1631 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1632 /* Check for overlaps */
1633 kvm_for_each_memslot(tmp
, __kvm_memslots(kvm
, as_id
)) {
1636 if (!((new.base_gfn
+ new.npages
<= tmp
->base_gfn
) ||
1637 (new.base_gfn
>= tmp
->base_gfn
+ tmp
->npages
)))
1642 /* Allocate/free page dirty bitmap as needed */
1643 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1644 new.dirty_bitmap
= NULL
;
1645 else if (!new.dirty_bitmap
&& !kvm
->dirty_ring_size
) {
1646 r
= kvm_alloc_dirty_bitmap(&new);
1650 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1651 bitmap_set(new.dirty_bitmap
, 0, new.npages
);
1654 r
= kvm_set_memslot(kvm
, mem
, &old
, &new, as_id
, change
);
1658 if (old
.dirty_bitmap
&& !new.dirty_bitmap
)
1659 kvm_destroy_dirty_bitmap(&old
);
1663 if (new.dirty_bitmap
&& !old
.dirty_bitmap
)
1664 kvm_destroy_dirty_bitmap(&new);
1667 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1669 int kvm_set_memory_region(struct kvm
*kvm
,
1670 const struct kvm_userspace_memory_region
*mem
)
1674 mutex_lock(&kvm
->slots_lock
);
1675 r
= __kvm_set_memory_region(kvm
, mem
);
1676 mutex_unlock(&kvm
->slots_lock
);
1679 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1681 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1682 struct kvm_userspace_memory_region
*mem
)
1684 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1687 return kvm_set_memory_region(kvm
, mem
);
1690 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1692 * kvm_get_dirty_log - get a snapshot of dirty pages
1693 * @kvm: pointer to kvm instance
1694 * @log: slot id and address to which we copy the log
1695 * @is_dirty: set to '1' if any dirty pages were found
1696 * @memslot: set to the associated memslot, always valid on success
1698 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
1699 int *is_dirty
, struct kvm_memory_slot
**memslot
)
1701 struct kvm_memslots
*slots
;
1704 unsigned long any
= 0;
1706 /* Dirty ring tracking is exclusive to dirty log tracking */
1707 if (kvm
->dirty_ring_size
)
1713 as_id
= log
->slot
>> 16;
1714 id
= (u16
)log
->slot
;
1715 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1718 slots
= __kvm_memslots(kvm
, as_id
);
1719 *memslot
= id_to_memslot(slots
, id
);
1720 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
1723 kvm_arch_sync_dirty_log(kvm
, *memslot
);
1725 n
= kvm_dirty_bitmap_bytes(*memslot
);
1727 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1728 any
= (*memslot
)->dirty_bitmap
[i
];
1730 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
1737 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1739 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1741 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1742 * and reenable dirty page tracking for the corresponding pages.
1743 * @kvm: pointer to kvm instance
1744 * @log: slot id and address to which we copy the log
1746 * We need to keep it in mind that VCPU threads can write to the bitmap
1747 * concurrently. So, to avoid losing track of dirty pages we keep the
1750 * 1. Take a snapshot of the bit and clear it if needed.
1751 * 2. Write protect the corresponding page.
1752 * 3. Copy the snapshot to the userspace.
1753 * 4. Upon return caller flushes TLB's if needed.
1755 * Between 2 and 4, the guest may write to the page using the remaining TLB
1756 * entry. This is not a problem because the page is reported dirty using
1757 * the snapshot taken before and step 4 ensures that writes done after
1758 * exiting to userspace will be logged for the next call.
1761 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1763 struct kvm_memslots
*slots
;
1764 struct kvm_memory_slot
*memslot
;
1767 unsigned long *dirty_bitmap
;
1768 unsigned long *dirty_bitmap_buffer
;
1771 /* Dirty ring tracking is exclusive to dirty log tracking */
1772 if (kvm
->dirty_ring_size
)
1775 as_id
= log
->slot
>> 16;
1776 id
= (u16
)log
->slot
;
1777 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1780 slots
= __kvm_memslots(kvm
, as_id
);
1781 memslot
= id_to_memslot(slots
, id
);
1782 if (!memslot
|| !memslot
->dirty_bitmap
)
1785 dirty_bitmap
= memslot
->dirty_bitmap
;
1787 kvm_arch_sync_dirty_log(kvm
, memslot
);
1789 n
= kvm_dirty_bitmap_bytes(memslot
);
1791 if (kvm
->manual_dirty_log_protect
) {
1793 * Unlike kvm_get_dirty_log, we always return false in *flush,
1794 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1795 * is some code duplication between this function and
1796 * kvm_get_dirty_log, but hopefully all architecture
1797 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1798 * can be eliminated.
1800 dirty_bitmap_buffer
= dirty_bitmap
;
1802 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1803 memset(dirty_bitmap_buffer
, 0, n
);
1806 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1810 if (!dirty_bitmap
[i
])
1814 mask
= xchg(&dirty_bitmap
[i
], 0);
1815 dirty_bitmap_buffer
[i
] = mask
;
1817 offset
= i
* BITS_PER_LONG
;
1818 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1821 KVM_MMU_UNLOCK(kvm
);
1825 kvm_arch_flush_remote_tlbs_memslot(kvm
, memslot
);
1827 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1834 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1835 * @kvm: kvm instance
1836 * @log: slot id and address to which we copy the log
1838 * Steps 1-4 below provide general overview of dirty page logging. See
1839 * kvm_get_dirty_log_protect() function description for additional details.
1841 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1842 * always flush the TLB (step 4) even if previous step failed and the dirty
1843 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1844 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1845 * writes will be marked dirty for next log read.
1847 * 1. Take a snapshot of the bit and clear it if needed.
1848 * 2. Write protect the corresponding page.
1849 * 3. Copy the snapshot to the userspace.
1850 * 4. Flush TLB's if needed.
1852 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1853 struct kvm_dirty_log
*log
)
1857 mutex_lock(&kvm
->slots_lock
);
1859 r
= kvm_get_dirty_log_protect(kvm
, log
);
1861 mutex_unlock(&kvm
->slots_lock
);
1866 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1867 * and reenable dirty page tracking for the corresponding pages.
1868 * @kvm: pointer to kvm instance
1869 * @log: slot id and address from which to fetch the bitmap of dirty pages
1871 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
1872 struct kvm_clear_dirty_log
*log
)
1874 struct kvm_memslots
*slots
;
1875 struct kvm_memory_slot
*memslot
;
1879 unsigned long *dirty_bitmap
;
1880 unsigned long *dirty_bitmap_buffer
;
1883 /* Dirty ring tracking is exclusive to dirty log tracking */
1884 if (kvm
->dirty_ring_size
)
1887 as_id
= log
->slot
>> 16;
1888 id
= (u16
)log
->slot
;
1889 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1892 if (log
->first_page
& 63)
1895 slots
= __kvm_memslots(kvm
, as_id
);
1896 memslot
= id_to_memslot(slots
, id
);
1897 if (!memslot
|| !memslot
->dirty_bitmap
)
1900 dirty_bitmap
= memslot
->dirty_bitmap
;
1902 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
1904 if (log
->first_page
> memslot
->npages
||
1905 log
->num_pages
> memslot
->npages
- log
->first_page
||
1906 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
1909 kvm_arch_sync_dirty_log(kvm
, memslot
);
1912 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1913 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
1917 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
1918 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
1919 i
++, offset
+= BITS_PER_LONG
) {
1920 unsigned long mask
= *dirty_bitmap_buffer
++;
1921 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
1925 mask
&= atomic_long_fetch_andnot(mask
, p
);
1928 * mask contains the bits that really have been cleared. This
1929 * never includes any bits beyond the length of the memslot (if
1930 * the length is not aligned to 64 pages), therefore it is not
1931 * a problem if userspace sets them in log->dirty_bitmap.
1935 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1939 KVM_MMU_UNLOCK(kvm
);
1942 kvm_arch_flush_remote_tlbs_memslot(kvm
, memslot
);
1947 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
1948 struct kvm_clear_dirty_log
*log
)
1952 mutex_lock(&kvm
->slots_lock
);
1954 r
= kvm_clear_dirty_log_protect(kvm
, log
);
1956 mutex_unlock(&kvm
->slots_lock
);
1959 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1961 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1963 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1965 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1967 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1969 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1971 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot
);
1973 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1975 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1977 return kvm_is_visible_memslot(memslot
);
1979 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1981 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1983 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1985 return kvm_is_visible_memslot(memslot
);
1987 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
1989 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1991 struct vm_area_struct
*vma
;
1992 unsigned long addr
, size
;
1996 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
1997 if (kvm_is_error_hva(addr
))
2000 mmap_read_lock(current
->mm
);
2001 vma
= find_vma(current
->mm
, addr
);
2005 size
= vma_kernel_pagesize(vma
);
2008 mmap_read_unlock(current
->mm
);
2013 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
2015 return slot
->flags
& KVM_MEM_READONLY
;
2018 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2019 gfn_t
*nr_pages
, bool write
)
2021 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
2022 return KVM_HVA_ERR_BAD
;
2024 if (memslot_is_readonly(slot
) && write
)
2025 return KVM_HVA_ERR_RO_BAD
;
2028 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
2030 return __gfn_to_hva_memslot(slot
, gfn
);
2033 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2036 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
2039 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
2042 return gfn_to_hva_many(slot
, gfn
, NULL
);
2044 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
2046 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
2048 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
2050 EXPORT_SYMBOL_GPL(gfn_to_hva
);
2052 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2054 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
2056 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
2059 * Return the hva of a @gfn and the R/W attribute if possible.
2061 * @slot: the kvm_memory_slot which contains @gfn
2062 * @gfn: the gfn to be translated
2063 * @writable: used to return the read/write attribute of the @slot if the hva
2064 * is valid and @writable is not NULL
2066 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
2067 gfn_t gfn
, bool *writable
)
2069 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
2071 if (!kvm_is_error_hva(hva
) && writable
)
2072 *writable
= !memslot_is_readonly(slot
);
2077 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
2079 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2081 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2084 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
2086 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2088 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2091 static inline int check_user_page_hwpoison(unsigned long addr
)
2093 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
2095 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
2096 return rc
== -EHWPOISON
;
2100 * The fast path to get the writable pfn which will be stored in @pfn,
2101 * true indicates success, otherwise false is returned. It's also the
2102 * only part that runs if we can in atomic context.
2104 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
2105 bool *writable
, kvm_pfn_t
*pfn
)
2107 struct page
*page
[1];
2110 * Fast pin a writable pfn only if it is a write fault request
2111 * or the caller allows to map a writable pfn for a read fault
2114 if (!(write_fault
|| writable
))
2117 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
2118 *pfn
= page_to_pfn(page
[0]);
2129 * The slow path to get the pfn of the specified host virtual address,
2130 * 1 indicates success, -errno is returned if error is detected.
2132 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
2133 bool *writable
, kvm_pfn_t
*pfn
)
2135 unsigned int flags
= FOLL_HWPOISON
;
2142 *writable
= write_fault
;
2145 flags
|= FOLL_WRITE
;
2147 flags
|= FOLL_NOWAIT
;
2149 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
2153 /* map read fault as writable if possible */
2154 if (unlikely(!write_fault
) && writable
) {
2157 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
2163 *pfn
= page_to_pfn(page
);
2167 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
2169 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
2172 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
2178 static int kvm_try_get_pfn(kvm_pfn_t pfn
)
2180 if (kvm_is_reserved_pfn(pfn
))
2182 return get_page_unless_zero(pfn_to_page(pfn
));
2185 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
2186 unsigned long addr
, bool *async
,
2187 bool write_fault
, bool *writable
,
2195 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2198 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2199 * not call the fault handler, so do it here.
2201 bool unlocked
= false;
2202 r
= fixup_user_fault(current
->mm
, addr
,
2203 (write_fault
? FAULT_FLAG_WRITE
: 0),
2210 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2215 if (write_fault
&& !pte_write(*ptep
)) {
2216 pfn
= KVM_PFN_ERR_RO_FAULT
;
2221 *writable
= pte_write(*ptep
);
2222 pfn
= pte_pfn(*ptep
);
2225 * Get a reference here because callers of *hva_to_pfn* and
2226 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2227 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2228 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2229 * simply do nothing for reserved pfns.
2231 * Whoever called remap_pfn_range is also going to call e.g.
2232 * unmap_mapping_range before the underlying pages are freed,
2233 * causing a call to our MMU notifier.
2235 * Certain IO or PFNMAP mappings can be backed with valid
2236 * struct pages, but be allocated without refcounting e.g.,
2237 * tail pages of non-compound higher order allocations, which
2238 * would then underflow the refcount when the caller does the
2239 * required put_page. Don't allow those pages here.
2241 if (!kvm_try_get_pfn(pfn
))
2245 pte_unmap_unlock(ptep
, ptl
);
2252 * Pin guest page in memory and return its pfn.
2253 * @addr: host virtual address which maps memory to the guest
2254 * @atomic: whether this function can sleep
2255 * @async: whether this function need to wait IO complete if the
2256 * host page is not in the memory
2257 * @write_fault: whether we should get a writable host page
2258 * @writable: whether it allows to map a writable host page for !@write_fault
2260 * The function will map a writable host page for these two cases:
2261 * 1): @write_fault = true
2262 * 2): @write_fault = false && @writable, @writable will tell the caller
2263 * whether the mapping is writable.
2265 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
2266 bool write_fault
, bool *writable
)
2268 struct vm_area_struct
*vma
;
2272 /* we can do it either atomically or asynchronously, not both */
2273 BUG_ON(atomic
&& async
);
2275 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
2279 return KVM_PFN_ERR_FAULT
;
2281 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
2285 mmap_read_lock(current
->mm
);
2286 if (npages
== -EHWPOISON
||
2287 (!async
&& check_user_page_hwpoison(addr
))) {
2288 pfn
= KVM_PFN_ERR_HWPOISON
;
2293 vma
= vma_lookup(current
->mm
, addr
);
2296 pfn
= KVM_PFN_ERR_FAULT
;
2297 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
2298 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
2302 pfn
= KVM_PFN_ERR_FAULT
;
2304 if (async
&& vma_is_valid(vma
, write_fault
))
2306 pfn
= KVM_PFN_ERR_FAULT
;
2309 mmap_read_unlock(current
->mm
);
2313 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2314 bool atomic
, bool *async
, bool write_fault
,
2315 bool *writable
, hva_t
*hva
)
2317 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
2322 if (addr
== KVM_HVA_ERR_RO_BAD
) {
2325 return KVM_PFN_ERR_RO_FAULT
;
2328 if (kvm_is_error_hva(addr
)) {
2331 return KVM_PFN_NOSLOT
;
2334 /* Do not map writable pfn in the readonly memslot. */
2335 if (writable
&& memslot_is_readonly(slot
)) {
2340 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
2343 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
2345 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
2348 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
2349 write_fault
, writable
, NULL
);
2351 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
2353 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
2355 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
, NULL
);
2357 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
2359 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
2361 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
, NULL
);
2363 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
2365 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2367 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2369 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
2371 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
2373 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
2375 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
2377 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2379 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2381 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
2383 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2384 struct page
**pages
, int nr_pages
)
2389 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
2390 if (kvm_is_error_hva(addr
))
2393 if (entry
< nr_pages
)
2396 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
2398 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
2400 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
2402 if (is_error_noslot_pfn(pfn
))
2403 return KVM_ERR_PTR_BAD_PAGE
;
2405 if (kvm_is_reserved_pfn(pfn
)) {
2407 return KVM_ERR_PTR_BAD_PAGE
;
2410 return pfn_to_page(pfn
);
2413 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
2417 pfn
= gfn_to_pfn(kvm
, gfn
);
2419 return kvm_pfn_to_page(pfn
);
2421 EXPORT_SYMBOL_GPL(gfn_to_page
);
2423 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
, struct gfn_to_pfn_cache
*cache
)
2429 cache
->pfn
= cache
->gfn
= 0;
2432 kvm_release_pfn_dirty(pfn
);
2434 kvm_release_pfn_clean(pfn
);
2437 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2438 struct gfn_to_pfn_cache
*cache
, u64 gen
)
2440 kvm_release_pfn(cache
->pfn
, cache
->dirty
, cache
);
2442 cache
->pfn
= gfn_to_pfn_memslot(slot
, gfn
);
2444 cache
->dirty
= false;
2445 cache
->generation
= gen
;
2448 static int __kvm_map_gfn(struct kvm_memslots
*slots
, gfn_t gfn
,
2449 struct kvm_host_map
*map
,
2450 struct gfn_to_pfn_cache
*cache
,
2455 struct page
*page
= KVM_UNMAPPED_PAGE
;
2456 struct kvm_memory_slot
*slot
= __gfn_to_memslot(slots
, gfn
);
2457 u64 gen
= slots
->generation
;
2463 if (!cache
->pfn
|| cache
->gfn
!= gfn
||
2464 cache
->generation
!= gen
) {
2467 kvm_cache_gfn_to_pfn(slot
, gfn
, cache
, gen
);
2473 pfn
= gfn_to_pfn_memslot(slot
, gfn
);
2475 if (is_error_noslot_pfn(pfn
))
2478 if (pfn_valid(pfn
)) {
2479 page
= pfn_to_page(pfn
);
2481 hva
= kmap_atomic(page
);
2484 #ifdef CONFIG_HAS_IOMEM
2485 } else if (!atomic
) {
2486 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
2503 int kvm_map_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
,
2504 struct gfn_to_pfn_cache
*cache
, bool atomic
)
2506 return __kvm_map_gfn(kvm_memslots(vcpu
->kvm
), gfn
, map
,
2509 EXPORT_SYMBOL_GPL(kvm_map_gfn
);
2511 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
2513 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu
), gfn
, map
,
2516 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
2518 static void __kvm_unmap_gfn(struct kvm
*kvm
,
2519 struct kvm_memory_slot
*memslot
,
2520 struct kvm_host_map
*map
,
2521 struct gfn_to_pfn_cache
*cache
,
2522 bool dirty
, bool atomic
)
2530 if (map
->page
!= KVM_UNMAPPED_PAGE
) {
2532 kunmap_atomic(map
->hva
);
2536 #ifdef CONFIG_HAS_IOMEM
2540 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2544 mark_page_dirty_in_slot(kvm
, memslot
, map
->gfn
);
2547 cache
->dirty
|= dirty
;
2549 kvm_release_pfn(map
->pfn
, dirty
, NULL
);
2555 int kvm_unmap_gfn(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
,
2556 struct gfn_to_pfn_cache
*cache
, bool dirty
, bool atomic
)
2558 __kvm_unmap_gfn(vcpu
->kvm
, gfn_to_memslot(vcpu
->kvm
, map
->gfn
), map
,
2559 cache
, dirty
, atomic
);
2562 EXPORT_SYMBOL_GPL(kvm_unmap_gfn
);
2564 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
2566 __kvm_unmap_gfn(vcpu
->kvm
, kvm_vcpu_gfn_to_memslot(vcpu
, map
->gfn
),
2567 map
, NULL
, dirty
, false);
2569 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
2571 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2575 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
2577 return kvm_pfn_to_page(pfn
);
2579 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
2581 void kvm_release_page_clean(struct page
*page
)
2583 WARN_ON(is_error_page(page
));
2585 kvm_release_pfn_clean(page_to_pfn(page
));
2587 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
2589 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
2591 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
2592 put_page(pfn_to_page(pfn
));
2594 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
2596 void kvm_release_page_dirty(struct page
*page
)
2598 WARN_ON(is_error_page(page
));
2600 kvm_release_pfn_dirty(page_to_pfn(page
));
2602 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
2604 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
2606 kvm_set_pfn_dirty(pfn
);
2607 kvm_release_pfn_clean(pfn
);
2609 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
2611 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
2613 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
2614 SetPageDirty(pfn_to_page(pfn
));
2616 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
2618 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
2620 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
2621 mark_page_accessed(pfn_to_page(pfn
));
2623 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
2625 void kvm_get_pfn(kvm_pfn_t pfn
)
2627 if (!kvm_is_reserved_pfn(pfn
))
2628 get_page(pfn_to_page(pfn
));
2630 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
2632 static int next_segment(unsigned long len
, int offset
)
2634 if (len
> PAGE_SIZE
- offset
)
2635 return PAGE_SIZE
- offset
;
2640 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2641 void *data
, int offset
, int len
)
2646 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
2647 if (kvm_is_error_hva(addr
))
2649 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
2655 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
2658 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2660 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
2662 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
2664 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
2665 int offset
, int len
)
2667 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2669 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
2671 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
2673 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
2675 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2677 int offset
= offset_in_page(gpa
);
2680 while ((seg
= next_segment(len
, offset
)) != 0) {
2681 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
2691 EXPORT_SYMBOL_GPL(kvm_read_guest
);
2693 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
2695 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2697 int offset
= offset_in_page(gpa
);
2700 while ((seg
= next_segment(len
, offset
)) != 0) {
2701 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2711 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
2713 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2714 void *data
, int offset
, unsigned long len
)
2719 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
2720 if (kvm_is_error_hva(addr
))
2722 pagefault_disable();
2723 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
2730 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2731 void *data
, unsigned long len
)
2733 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2734 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2735 int offset
= offset_in_page(gpa
);
2737 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
2739 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
2741 static int __kvm_write_guest_page(struct kvm
*kvm
,
2742 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
2743 const void *data
, int offset
, int len
)
2748 addr
= gfn_to_hva_memslot(memslot
, gfn
);
2749 if (kvm_is_error_hva(addr
))
2751 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
2754 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
2758 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
2759 const void *data
, int offset
, int len
)
2761 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2763 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
2765 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
2767 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
2768 const void *data
, int offset
, int len
)
2770 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2772 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
2774 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
2776 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
2779 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2781 int offset
= offset_in_page(gpa
);
2784 while ((seg
= next_segment(len
, offset
)) != 0) {
2785 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
2795 EXPORT_SYMBOL_GPL(kvm_write_guest
);
2797 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
2800 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2802 int offset
= offset_in_page(gpa
);
2805 while ((seg
= next_segment(len
, offset
)) != 0) {
2806 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2816 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
2818 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
2819 struct gfn_to_hva_cache
*ghc
,
2820 gpa_t gpa
, unsigned long len
)
2822 int offset
= offset_in_page(gpa
);
2823 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
2824 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
2825 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
2826 gfn_t nr_pages_avail
;
2828 /* Update ghc->generation before performing any error checks. */
2829 ghc
->generation
= slots
->generation
;
2831 if (start_gfn
> end_gfn
) {
2832 ghc
->hva
= KVM_HVA_ERR_BAD
;
2837 * If the requested region crosses two memslots, we still
2838 * verify that the entire region is valid here.
2840 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
2841 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
2842 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
2844 if (kvm_is_error_hva(ghc
->hva
))
2848 /* Use the slow path for cross page reads and writes. */
2849 if (nr_pages_needed
== 1)
2852 ghc
->memslot
= NULL
;
2859 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2860 gpa_t gpa
, unsigned long len
)
2862 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2863 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
2865 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
2867 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2868 void *data
, unsigned int offset
,
2871 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2873 gpa_t gpa
= ghc
->gpa
+ offset
;
2875 BUG_ON(len
+ offset
> ghc
->len
);
2877 if (slots
->generation
!= ghc
->generation
) {
2878 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
2882 if (kvm_is_error_hva(ghc
->hva
))
2885 if (unlikely(!ghc
->memslot
))
2886 return kvm_write_guest(kvm
, gpa
, data
, len
);
2888 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2891 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2895 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2897 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2898 void *data
, unsigned long len
)
2900 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2902 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2904 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2905 void *data
, unsigned int offset
,
2908 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2910 gpa_t gpa
= ghc
->gpa
+ offset
;
2912 BUG_ON(len
+ offset
> ghc
->len
);
2914 if (slots
->generation
!= ghc
->generation
) {
2915 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
2919 if (kvm_is_error_hva(ghc
->hva
))
2922 if (unlikely(!ghc
->memslot
))
2923 return kvm_read_guest(kvm
, gpa
, data
, len
);
2925 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
2931 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
2933 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2934 void *data
, unsigned long len
)
2936 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2938 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2940 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2942 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2943 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2945 int offset
= offset_in_page(gpa
);
2948 while ((seg
= next_segment(len
, offset
)) != 0) {
2949 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2958 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2960 void mark_page_dirty_in_slot(struct kvm
*kvm
,
2961 struct kvm_memory_slot
*memslot
,
2964 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
2965 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2966 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
2968 if (kvm
->dirty_ring_size
)
2969 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm
),
2972 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2975 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
2977 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2979 struct kvm_memory_slot
*memslot
;
2981 memslot
= gfn_to_memslot(kvm
, gfn
);
2982 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
2984 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2986 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2988 struct kvm_memory_slot
*memslot
;
2990 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2991 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
2993 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2995 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2997 if (!vcpu
->sigset_active
)
3001 * This does a lockless modification of ->real_blocked, which is fine
3002 * because, only current can change ->real_blocked and all readers of
3003 * ->real_blocked don't care as long ->real_blocked is always a subset
3006 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
3009 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
3011 if (!vcpu
->sigset_active
)
3014 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
3015 sigemptyset(¤t
->real_blocked
);
3018 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3020 unsigned int old
, val
, grow
, grow_start
;
3022 old
= val
= vcpu
->halt_poll_ns
;
3023 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3024 grow
= READ_ONCE(halt_poll_ns_grow
);
3029 if (val
< grow_start
)
3032 if (val
> vcpu
->kvm
->max_halt_poll_ns
)
3033 val
= vcpu
->kvm
->max_halt_poll_ns
;
3035 vcpu
->halt_poll_ns
= val
;
3037 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
3040 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3042 unsigned int old
, val
, shrink
;
3044 old
= val
= vcpu
->halt_poll_ns
;
3045 shrink
= READ_ONCE(halt_poll_ns_shrink
);
3051 vcpu
->halt_poll_ns
= val
;
3052 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
3055 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
3058 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3060 if (kvm_arch_vcpu_runnable(vcpu
)) {
3061 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
3064 if (kvm_cpu_has_pending_timer(vcpu
))
3066 if (signal_pending(current
))
3068 if (kvm_check_request(KVM_REQ_UNBLOCK
, vcpu
))
3073 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3078 update_halt_poll_stats(struct kvm_vcpu
*vcpu
, u64 poll_ns
, bool waited
)
3081 vcpu
->stat
.generic
.halt_poll_fail_ns
+= poll_ns
;
3083 vcpu
->stat
.generic
.halt_poll_success_ns
+= poll_ns
;
3087 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3089 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
3091 ktime_t start
, cur
, poll_end
;
3092 bool waited
= false;
3095 kvm_arch_vcpu_blocking(vcpu
);
3097 start
= cur
= poll_end
= ktime_get();
3098 if (vcpu
->halt_poll_ns
&& !kvm_arch_no_poll(vcpu
)) {
3099 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
3101 ++vcpu
->stat
.generic
.halt_attempted_poll
;
3104 * This sets KVM_REQ_UNHALT if an interrupt
3107 if (kvm_vcpu_check_block(vcpu
) < 0) {
3108 ++vcpu
->stat
.generic
.halt_successful_poll
;
3109 if (!vcpu_valid_wakeup(vcpu
))
3110 ++vcpu
->stat
.generic
.halt_poll_invalid
;
3114 poll_end
= cur
= ktime_get();
3115 } while (kvm_vcpu_can_poll(cur
, stop
));
3118 prepare_to_rcuwait(&vcpu
->wait
);
3120 set_current_state(TASK_INTERRUPTIBLE
);
3122 if (kvm_vcpu_check_block(vcpu
) < 0)
3128 finish_rcuwait(&vcpu
->wait
);
3131 kvm_arch_vcpu_unblocking(vcpu
);
3132 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
3134 update_halt_poll_stats(
3135 vcpu
, ktime_to_ns(ktime_sub(poll_end
, start
)), waited
);
3137 if (!kvm_arch_no_poll(vcpu
)) {
3138 if (!vcpu_valid_wakeup(vcpu
)) {
3139 shrink_halt_poll_ns(vcpu
);
3140 } else if (vcpu
->kvm
->max_halt_poll_ns
) {
3141 if (block_ns
<= vcpu
->halt_poll_ns
)
3143 /* we had a long block, shrink polling */
3144 else if (vcpu
->halt_poll_ns
&&
3145 block_ns
> vcpu
->kvm
->max_halt_poll_ns
)
3146 shrink_halt_poll_ns(vcpu
);
3147 /* we had a short halt and our poll time is too small */
3148 else if (vcpu
->halt_poll_ns
< vcpu
->kvm
->max_halt_poll_ns
&&
3149 block_ns
< vcpu
->kvm
->max_halt_poll_ns
)
3150 grow_halt_poll_ns(vcpu
);
3152 vcpu
->halt_poll_ns
= 0;
3156 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
3157 kvm_arch_vcpu_block_finish(vcpu
);
3159 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
3161 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
3163 struct rcuwait
*waitp
;
3165 waitp
= kvm_arch_vcpu_get_wait(vcpu
);
3166 if (rcuwait_wake_up(waitp
)) {
3167 WRITE_ONCE(vcpu
->ready
, true);
3168 ++vcpu
->stat
.generic
.halt_wakeup
;
3174 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
3178 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3180 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3183 int cpu
= vcpu
->cpu
;
3185 if (kvm_vcpu_wake_up(vcpu
))
3189 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
3190 if (kvm_arch_vcpu_should_kick(vcpu
))
3191 smp_send_reschedule(cpu
);
3194 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
3195 #endif /* !CONFIG_S390 */
3197 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
3200 struct task_struct
*task
= NULL
;
3204 pid
= rcu_dereference(target
->pid
);
3206 task
= get_pid_task(pid
, PIDTYPE_PID
);
3210 ret
= yield_to(task
, 1);
3211 put_task_struct(task
);
3215 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
3218 * Helper that checks whether a VCPU is eligible for directed yield.
3219 * Most eligible candidate to yield is decided by following heuristics:
3221 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3222 * (preempted lock holder), indicated by @in_spin_loop.
3223 * Set at the beginning and cleared at the end of interception/PLE handler.
3225 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3226 * chance last time (mostly it has become eligible now since we have probably
3227 * yielded to lockholder in last iteration. This is done by toggling
3228 * @dy_eligible each time a VCPU checked for eligibility.)
3230 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3231 * to preempted lock-holder could result in wrong VCPU selection and CPU
3232 * burning. Giving priority for a potential lock-holder increases lock
3235 * Since algorithm is based on heuristics, accessing another VCPU data without
3236 * locking does not harm. It may result in trying to yield to same VCPU, fail
3237 * and continue with next VCPU and so on.
3239 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
3241 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3244 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
3245 vcpu
->spin_loop
.dy_eligible
;
3247 if (vcpu
->spin_loop
.in_spin_loop
)
3248 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
3257 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3258 * a vcpu_load/vcpu_put pair. However, for most architectures
3259 * kvm_arch_vcpu_runnable does not require vcpu_load.
3261 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
3263 return kvm_arch_vcpu_runnable(vcpu
);
3266 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
3268 if (kvm_arch_dy_runnable(vcpu
))
3271 #ifdef CONFIG_KVM_ASYNC_PF
3272 if (!list_empty_careful(&vcpu
->async_pf
.done
))
3279 bool __weak
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu
*vcpu
)
3284 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
3286 struct kvm
*kvm
= me
->kvm
;
3287 struct kvm_vcpu
*vcpu
;
3288 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
3294 kvm_vcpu_set_in_spin_loop(me
, true);
3296 * We boost the priority of a VCPU that is runnable but not
3297 * currently running, because it got preempted by something
3298 * else and called schedule in __vcpu_run. Hopefully that
3299 * VCPU is holding the lock that we need and will release it.
3300 * We approximate round-robin by starting at the last boosted VCPU.
3302 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
3303 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3304 if (!pass
&& i
<= last_boosted_vcpu
) {
3305 i
= last_boosted_vcpu
;
3307 } else if (pass
&& i
> last_boosted_vcpu
)
3309 if (!READ_ONCE(vcpu
->ready
))
3313 if (rcuwait_active(&vcpu
->wait
) &&
3314 !vcpu_dy_runnable(vcpu
))
3316 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
3317 !kvm_arch_dy_has_pending_interrupt(vcpu
) &&
3318 !kvm_arch_vcpu_in_kernel(vcpu
))
3320 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
3323 yielded
= kvm_vcpu_yield_to(vcpu
);
3325 kvm
->last_boosted_vcpu
= i
;
3327 } else if (yielded
< 0) {
3334 kvm_vcpu_set_in_spin_loop(me
, false);
3336 /* Ensure vcpu is not eligible during next spinloop */
3337 kvm_vcpu_set_dy_eligible(me
, false);
3339 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
3341 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
3343 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3344 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
3345 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
3346 kvm
->dirty_ring_size
/ PAGE_SIZE
);
3352 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
3354 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
3357 if (vmf
->pgoff
== 0)
3358 page
= virt_to_page(vcpu
->run
);
3360 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
3361 page
= virt_to_page(vcpu
->arch
.pio_data
);
3363 #ifdef CONFIG_KVM_MMIO
3364 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
3365 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
3367 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
3368 page
= kvm_dirty_ring_get_page(
3370 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
3372 return kvm_arch_vcpu_fault(vcpu
, vmf
);
3378 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
3379 .fault
= kvm_vcpu_fault
,
3382 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3384 struct kvm_vcpu
*vcpu
= file
->private_data
;
3385 unsigned long pages
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
3387 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
3388 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
3389 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
3392 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
3396 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
3398 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3400 kvm_put_kvm(vcpu
->kvm
);
3404 static struct file_operations kvm_vcpu_fops
= {
3405 .release
= kvm_vcpu_release
,
3406 .unlocked_ioctl
= kvm_vcpu_ioctl
,
3407 .mmap
= kvm_vcpu_mmap
,
3408 .llseek
= noop_llseek
,
3409 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
3413 * Allocates an inode for the vcpu.
3415 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
3417 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
3419 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
3420 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
3423 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
3425 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3426 struct dentry
*debugfs_dentry
;
3427 char dir_name
[ITOA_MAX_LEN
* 2];
3429 if (!debugfs_initialized())
3432 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
3433 debugfs_dentry
= debugfs_create_dir(dir_name
,
3434 vcpu
->kvm
->debugfs_dentry
);
3436 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
3441 * Creates some virtual cpus. Good luck creating more than one.
3443 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
3446 struct kvm_vcpu
*vcpu
;
3449 if (id
>= KVM_MAX_VCPU_ID
)
3452 mutex_lock(&kvm
->lock
);
3453 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
3454 mutex_unlock(&kvm
->lock
);
3458 kvm
->created_vcpus
++;
3459 mutex_unlock(&kvm
->lock
);
3461 r
= kvm_arch_vcpu_precreate(kvm
, id
);
3463 goto vcpu_decrement
;
3465 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL_ACCOUNT
);
3468 goto vcpu_decrement
;
3471 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
3472 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
3477 vcpu
->run
= page_address(page
);
3479 kvm_vcpu_init(vcpu
, kvm
, id
);
3481 r
= kvm_arch_vcpu_create(vcpu
);
3483 goto vcpu_free_run_page
;
3485 if (kvm
->dirty_ring_size
) {
3486 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
3487 id
, kvm
->dirty_ring_size
);
3489 goto arch_vcpu_destroy
;
3492 mutex_lock(&kvm
->lock
);
3493 if (kvm_get_vcpu_by_id(kvm
, id
)) {
3495 goto unlock_vcpu_destroy
;
3498 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
3499 BUG_ON(kvm
->vcpus
[vcpu
->vcpu_idx
]);
3501 /* Fill the stats id string for the vcpu */
3502 snprintf(vcpu
->stats_id
, sizeof(vcpu
->stats_id
), "kvm-%d/vcpu-%d",
3503 task_pid_nr(current
), id
);
3505 /* Now it's all set up, let userspace reach it */
3507 r
= create_vcpu_fd(vcpu
);
3509 kvm_put_kvm_no_destroy(kvm
);
3510 goto unlock_vcpu_destroy
;
3513 kvm
->vcpus
[vcpu
->vcpu_idx
] = vcpu
;
3516 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3517 * before kvm->online_vcpu's incremented value.
3520 atomic_inc(&kvm
->online_vcpus
);
3522 mutex_unlock(&kvm
->lock
);
3523 kvm_arch_vcpu_postcreate(vcpu
);
3524 kvm_create_vcpu_debugfs(vcpu
);
3527 unlock_vcpu_destroy
:
3528 mutex_unlock(&kvm
->lock
);
3529 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
3531 kvm_arch_vcpu_destroy(vcpu
);
3533 free_page((unsigned long)vcpu
->run
);
3535 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
3537 mutex_lock(&kvm
->lock
);
3538 kvm
->created_vcpus
--;
3539 mutex_unlock(&kvm
->lock
);
3543 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
3546 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
3547 vcpu
->sigset_active
= 1;
3548 vcpu
->sigset
= *sigset
;
3550 vcpu
->sigset_active
= 0;
3554 static ssize_t
kvm_vcpu_stats_read(struct file
*file
, char __user
*user_buffer
,
3555 size_t size
, loff_t
*offset
)
3557 struct kvm_vcpu
*vcpu
= file
->private_data
;
3559 return kvm_stats_read(vcpu
->stats_id
, &kvm_vcpu_stats_header
,
3560 &kvm_vcpu_stats_desc
[0], &vcpu
->stat
,
3561 sizeof(vcpu
->stat
), user_buffer
, size
, offset
);
3564 static const struct file_operations kvm_vcpu_stats_fops
= {
3565 .read
= kvm_vcpu_stats_read
,
3566 .llseek
= noop_llseek
,
3569 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu
*vcpu
)
3573 char name
[15 + ITOA_MAX_LEN
+ 1];
3575 snprintf(name
, sizeof(name
), "kvm-vcpu-stats:%d", vcpu
->vcpu_id
);
3577 fd
= get_unused_fd_flags(O_CLOEXEC
);
3581 file
= anon_inode_getfile(name
, &kvm_vcpu_stats_fops
, vcpu
, O_RDONLY
);
3584 return PTR_ERR(file
);
3586 file
->f_mode
|= FMODE_PREAD
;
3587 fd_install(fd
, file
);
3592 static long kvm_vcpu_ioctl(struct file
*filp
,
3593 unsigned int ioctl
, unsigned long arg
)
3595 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3596 void __user
*argp
= (void __user
*)arg
;
3598 struct kvm_fpu
*fpu
= NULL
;
3599 struct kvm_sregs
*kvm_sregs
= NULL
;
3601 if (vcpu
->kvm
->mm
!= current
->mm
)
3604 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
3608 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3609 * execution; mutex_lock() would break them.
3611 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
3612 if (r
!= -ENOIOCTLCMD
)
3615 if (mutex_lock_killable(&vcpu
->mutex
))
3623 oldpid
= rcu_access_pointer(vcpu
->pid
);
3624 if (unlikely(oldpid
!= task_pid(current
))) {
3625 /* The thread running this VCPU changed. */
3628 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
3632 newpid
= get_task_pid(current
, PIDTYPE_PID
);
3633 rcu_assign_pointer(vcpu
->pid
, newpid
);
3638 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
3639 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
3642 case KVM_GET_REGS
: {
3643 struct kvm_regs
*kvm_regs
;
3646 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
3649 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
3653 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
3660 case KVM_SET_REGS
: {
3661 struct kvm_regs
*kvm_regs
;
3663 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
3664 if (IS_ERR(kvm_regs
)) {
3665 r
= PTR_ERR(kvm_regs
);
3668 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
3672 case KVM_GET_SREGS
: {
3673 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
3674 GFP_KERNEL_ACCOUNT
);
3678 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
3682 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
3687 case KVM_SET_SREGS
: {
3688 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
3689 if (IS_ERR(kvm_sregs
)) {
3690 r
= PTR_ERR(kvm_sregs
);
3694 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
3697 case KVM_GET_MP_STATE
: {
3698 struct kvm_mp_state mp_state
;
3700 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
3704 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
3709 case KVM_SET_MP_STATE
: {
3710 struct kvm_mp_state mp_state
;
3713 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
3715 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
3718 case KVM_TRANSLATE
: {
3719 struct kvm_translation tr
;
3722 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
3724 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
3728 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
3733 case KVM_SET_GUEST_DEBUG
: {
3734 struct kvm_guest_debug dbg
;
3737 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
3739 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
3742 case KVM_SET_SIGNAL_MASK
: {
3743 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
3744 struct kvm_signal_mask kvm_sigmask
;
3745 sigset_t sigset
, *p
;
3750 if (copy_from_user(&kvm_sigmask
, argp
,
3751 sizeof(kvm_sigmask
)))
3754 if (kvm_sigmask
.len
!= sizeof(sigset
))
3757 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
3762 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
3766 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
3770 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
3774 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
3780 fpu
= memdup_user(argp
, sizeof(*fpu
));
3786 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
3789 case KVM_GET_STATS_FD
: {
3790 r
= kvm_vcpu_ioctl_get_stats_fd(vcpu
);
3794 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
3797 mutex_unlock(&vcpu
->mutex
);
3803 #ifdef CONFIG_KVM_COMPAT
3804 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
3805 unsigned int ioctl
, unsigned long arg
)
3807 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3808 void __user
*argp
= compat_ptr(arg
);
3811 if (vcpu
->kvm
->mm
!= current
->mm
)
3815 case KVM_SET_SIGNAL_MASK
: {
3816 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
3817 struct kvm_signal_mask kvm_sigmask
;
3822 if (copy_from_user(&kvm_sigmask
, argp
,
3823 sizeof(kvm_sigmask
)))
3826 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
3829 if (get_compat_sigset(&sigset
,
3830 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
3832 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
3834 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
3838 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
3846 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
3848 struct kvm_device
*dev
= filp
->private_data
;
3851 return dev
->ops
->mmap(dev
, vma
);
3856 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
3857 int (*accessor
)(struct kvm_device
*dev
,
3858 struct kvm_device_attr
*attr
),
3861 struct kvm_device_attr attr
;
3866 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
3869 return accessor(dev
, &attr
);
3872 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
3875 struct kvm_device
*dev
= filp
->private_data
;
3877 if (dev
->kvm
->mm
!= current
->mm
)
3881 case KVM_SET_DEVICE_ATTR
:
3882 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
3883 case KVM_GET_DEVICE_ATTR
:
3884 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
3885 case KVM_HAS_DEVICE_ATTR
:
3886 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
3888 if (dev
->ops
->ioctl
)
3889 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
3895 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
3897 struct kvm_device
*dev
= filp
->private_data
;
3898 struct kvm
*kvm
= dev
->kvm
;
3900 if (dev
->ops
->release
) {
3901 mutex_lock(&kvm
->lock
);
3902 list_del(&dev
->vm_node
);
3903 dev
->ops
->release(dev
);
3904 mutex_unlock(&kvm
->lock
);
3911 static const struct file_operations kvm_device_fops
= {
3912 .unlocked_ioctl
= kvm_device_ioctl
,
3913 .release
= kvm_device_release
,
3914 KVM_COMPAT(kvm_device_ioctl
),
3915 .mmap
= kvm_device_mmap
,
3918 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
3920 if (filp
->f_op
!= &kvm_device_fops
)
3923 return filp
->private_data
;
3926 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
3927 #ifdef CONFIG_KVM_MPIC
3928 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
3929 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
3933 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
3935 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
3938 if (kvm_device_ops_table
[type
] != NULL
)
3941 kvm_device_ops_table
[type
] = ops
;
3945 void kvm_unregister_device_ops(u32 type
)
3947 if (kvm_device_ops_table
[type
] != NULL
)
3948 kvm_device_ops_table
[type
] = NULL
;
3951 static int kvm_ioctl_create_device(struct kvm
*kvm
,
3952 struct kvm_create_device
*cd
)
3954 const struct kvm_device_ops
*ops
= NULL
;
3955 struct kvm_device
*dev
;
3956 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
3960 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
3963 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
3964 ops
= kvm_device_ops_table
[type
];
3971 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
3978 mutex_lock(&kvm
->lock
);
3979 ret
= ops
->create(dev
, type
);
3981 mutex_unlock(&kvm
->lock
);
3985 list_add(&dev
->vm_node
, &kvm
->devices
);
3986 mutex_unlock(&kvm
->lock
);
3992 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
3994 kvm_put_kvm_no_destroy(kvm
);
3995 mutex_lock(&kvm
->lock
);
3996 list_del(&dev
->vm_node
);
3997 mutex_unlock(&kvm
->lock
);
4006 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
4009 case KVM_CAP_USER_MEMORY
:
4010 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
4011 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
4012 case KVM_CAP_INTERNAL_ERROR_DATA
:
4013 #ifdef CONFIG_HAVE_KVM_MSI
4014 case KVM_CAP_SIGNAL_MSI
:
4016 #ifdef CONFIG_HAVE_KVM_IRQFD
4018 case KVM_CAP_IRQFD_RESAMPLE
:
4020 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
4021 case KVM_CAP_CHECK_EXTENSION_VM
:
4022 case KVM_CAP_ENABLE_CAP_VM
:
4023 case KVM_CAP_HALT_POLL
:
4025 #ifdef CONFIG_KVM_MMIO
4026 case KVM_CAP_COALESCED_MMIO
:
4027 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
4028 case KVM_CAP_COALESCED_PIO
:
4031 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4032 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
4033 return KVM_DIRTY_LOG_MANUAL_CAPS
;
4035 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4036 case KVM_CAP_IRQ_ROUTING
:
4037 return KVM_MAX_IRQ_ROUTES
;
4039 #if KVM_ADDRESS_SPACE_NUM > 1
4040 case KVM_CAP_MULTI_ADDRESS_SPACE
:
4041 return KVM_ADDRESS_SPACE_NUM
;
4043 case KVM_CAP_NR_MEMSLOTS
:
4044 return KVM_USER_MEM_SLOTS
;
4045 case KVM_CAP_DIRTY_LOG_RING
:
4046 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4047 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4051 case KVM_CAP_BINARY_STATS_FD
:
4056 return kvm_vm_ioctl_check_extension(kvm
, arg
);
4059 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
4063 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
4066 /* the size should be power of 2 */
4067 if (!size
|| (size
& (size
- 1)))
4070 /* Should be bigger to keep the reserved entries, or a page */
4071 if (size
< kvm_dirty_ring_get_rsvd_entries() *
4072 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
4075 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
4076 sizeof(struct kvm_dirty_gfn
))
4079 /* We only allow it to set once */
4080 if (kvm
->dirty_ring_size
)
4083 mutex_lock(&kvm
->lock
);
4085 if (kvm
->created_vcpus
) {
4086 /* We don't allow to change this value after vcpu created */
4089 kvm
->dirty_ring_size
= size
;
4093 mutex_unlock(&kvm
->lock
);
4097 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
4100 struct kvm_vcpu
*vcpu
;
4103 if (!kvm
->dirty_ring_size
)
4106 mutex_lock(&kvm
->slots_lock
);
4108 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4109 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
4111 mutex_unlock(&kvm
->slots_lock
);
4114 kvm_flush_remote_tlbs(kvm
);
4119 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4120 struct kvm_enable_cap
*cap
)
4125 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
4126 struct kvm_enable_cap
*cap
)
4129 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4130 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
4131 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
4133 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
4134 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
4136 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
4138 kvm
->manual_dirty_log_protect
= cap
->args
[0];
4142 case KVM_CAP_HALT_POLL
: {
4143 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
4146 kvm
->max_halt_poll_ns
= cap
->args
[0];
4149 case KVM_CAP_DIRTY_LOG_RING
:
4150 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
4152 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
4156 static ssize_t
kvm_vm_stats_read(struct file
*file
, char __user
*user_buffer
,
4157 size_t size
, loff_t
*offset
)
4159 struct kvm
*kvm
= file
->private_data
;
4161 return kvm_stats_read(kvm
->stats_id
, &kvm_vm_stats_header
,
4162 &kvm_vm_stats_desc
[0], &kvm
->stat
,
4163 sizeof(kvm
->stat
), user_buffer
, size
, offset
);
4166 static const struct file_operations kvm_vm_stats_fops
= {
4167 .read
= kvm_vm_stats_read
,
4168 .llseek
= noop_llseek
,
4171 static int kvm_vm_ioctl_get_stats_fd(struct kvm
*kvm
)
4176 fd
= get_unused_fd_flags(O_CLOEXEC
);
4180 file
= anon_inode_getfile("kvm-vm-stats",
4181 &kvm_vm_stats_fops
, kvm
, O_RDONLY
);
4184 return PTR_ERR(file
);
4186 file
->f_mode
|= FMODE_PREAD
;
4187 fd_install(fd
, file
);
4192 static long kvm_vm_ioctl(struct file
*filp
,
4193 unsigned int ioctl
, unsigned long arg
)
4195 struct kvm
*kvm
= filp
->private_data
;
4196 void __user
*argp
= (void __user
*)arg
;
4199 if (kvm
->mm
!= current
->mm
)
4202 case KVM_CREATE_VCPU
:
4203 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
4205 case KVM_ENABLE_CAP
: {
4206 struct kvm_enable_cap cap
;
4209 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4211 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
4214 case KVM_SET_USER_MEMORY_REGION
: {
4215 struct kvm_userspace_memory_region kvm_userspace_mem
;
4218 if (copy_from_user(&kvm_userspace_mem
, argp
,
4219 sizeof(kvm_userspace_mem
)))
4222 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
4225 case KVM_GET_DIRTY_LOG
: {
4226 struct kvm_dirty_log log
;
4229 if (copy_from_user(&log
, argp
, sizeof(log
)))
4231 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
4234 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4235 case KVM_CLEAR_DIRTY_LOG
: {
4236 struct kvm_clear_dirty_log log
;
4239 if (copy_from_user(&log
, argp
, sizeof(log
)))
4241 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
4245 #ifdef CONFIG_KVM_MMIO
4246 case KVM_REGISTER_COALESCED_MMIO
: {
4247 struct kvm_coalesced_mmio_zone zone
;
4250 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4252 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
4255 case KVM_UNREGISTER_COALESCED_MMIO
: {
4256 struct kvm_coalesced_mmio_zone zone
;
4259 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4261 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
4266 struct kvm_irqfd data
;
4269 if (copy_from_user(&data
, argp
, sizeof(data
)))
4271 r
= kvm_irqfd(kvm
, &data
);
4274 case KVM_IOEVENTFD
: {
4275 struct kvm_ioeventfd data
;
4278 if (copy_from_user(&data
, argp
, sizeof(data
)))
4280 r
= kvm_ioeventfd(kvm
, &data
);
4283 #ifdef CONFIG_HAVE_KVM_MSI
4284 case KVM_SIGNAL_MSI
: {
4288 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
4290 r
= kvm_send_userspace_msi(kvm
, &msi
);
4294 #ifdef __KVM_HAVE_IRQ_LINE
4295 case KVM_IRQ_LINE_STATUS
:
4296 case KVM_IRQ_LINE
: {
4297 struct kvm_irq_level irq_event
;
4300 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
4303 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
4304 ioctl
== KVM_IRQ_LINE_STATUS
);
4309 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
4310 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
4318 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4319 case KVM_SET_GSI_ROUTING
: {
4320 struct kvm_irq_routing routing
;
4321 struct kvm_irq_routing __user
*urouting
;
4322 struct kvm_irq_routing_entry
*entries
= NULL
;
4325 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
4328 if (!kvm_arch_can_set_irq_routing(kvm
))
4330 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
4336 entries
= vmemdup_user(urouting
->entries
,
4337 array_size(sizeof(*entries
),
4339 if (IS_ERR(entries
)) {
4340 r
= PTR_ERR(entries
);
4344 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
4349 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4350 case KVM_CREATE_DEVICE
: {
4351 struct kvm_create_device cd
;
4354 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
4357 r
= kvm_ioctl_create_device(kvm
, &cd
);
4362 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
4368 case KVM_CHECK_EXTENSION
:
4369 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
4371 case KVM_RESET_DIRTY_RINGS
:
4372 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
4374 case KVM_GET_STATS_FD
:
4375 r
= kvm_vm_ioctl_get_stats_fd(kvm
);
4378 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
4384 #ifdef CONFIG_KVM_COMPAT
4385 struct compat_kvm_dirty_log
{
4389 compat_uptr_t dirty_bitmap
; /* one bit per page */
4394 static long kvm_vm_compat_ioctl(struct file
*filp
,
4395 unsigned int ioctl
, unsigned long arg
)
4397 struct kvm
*kvm
= filp
->private_data
;
4400 if (kvm
->mm
!= current
->mm
)
4403 case KVM_GET_DIRTY_LOG
: {
4404 struct compat_kvm_dirty_log compat_log
;
4405 struct kvm_dirty_log log
;
4407 if (copy_from_user(&compat_log
, (void __user
*)arg
,
4408 sizeof(compat_log
)))
4410 log
.slot
= compat_log
.slot
;
4411 log
.padding1
= compat_log
.padding1
;
4412 log
.padding2
= compat_log
.padding2
;
4413 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
4415 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
4419 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
4425 static struct file_operations kvm_vm_fops
= {
4426 .release
= kvm_vm_release
,
4427 .unlocked_ioctl
= kvm_vm_ioctl
,
4428 .llseek
= noop_llseek
,
4429 KVM_COMPAT(kvm_vm_compat_ioctl
),
4432 bool file_is_kvm(struct file
*file
)
4434 return file
&& file
->f_op
== &kvm_vm_fops
;
4436 EXPORT_SYMBOL_GPL(file_is_kvm
);
4438 static int kvm_dev_ioctl_create_vm(unsigned long type
)
4444 kvm
= kvm_create_vm(type
);
4446 return PTR_ERR(kvm
);
4447 #ifdef CONFIG_KVM_MMIO
4448 r
= kvm_coalesced_mmio_init(kvm
);
4452 r
= get_unused_fd_flags(O_CLOEXEC
);
4456 snprintf(kvm
->stats_id
, sizeof(kvm
->stats_id
),
4457 "kvm-%d", task_pid_nr(current
));
4459 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
4467 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4468 * already set, with ->release() being kvm_vm_release(). In error
4469 * cases it will be called by the final fput(file) and will take
4470 * care of doing kvm_put_kvm(kvm).
4472 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
4477 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
4479 fd_install(r
, file
);
4487 static long kvm_dev_ioctl(struct file
*filp
,
4488 unsigned int ioctl
, unsigned long arg
)
4493 case KVM_GET_API_VERSION
:
4496 r
= KVM_API_VERSION
;
4499 r
= kvm_dev_ioctl_create_vm(arg
);
4501 case KVM_CHECK_EXTENSION
:
4502 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
4504 case KVM_GET_VCPU_MMAP_SIZE
:
4507 r
= PAGE_SIZE
; /* struct kvm_run */
4509 r
+= PAGE_SIZE
; /* pio data page */
4511 #ifdef CONFIG_KVM_MMIO
4512 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
4515 case KVM_TRACE_ENABLE
:
4516 case KVM_TRACE_PAUSE
:
4517 case KVM_TRACE_DISABLE
:
4521 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
4527 static struct file_operations kvm_chardev_ops
= {
4528 .unlocked_ioctl
= kvm_dev_ioctl
,
4529 .llseek
= noop_llseek
,
4530 KVM_COMPAT(kvm_dev_ioctl
),
4533 static struct miscdevice kvm_dev
= {
4539 static void hardware_enable_nolock(void *junk
)
4541 int cpu
= raw_smp_processor_id();
4544 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
4547 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
4549 r
= kvm_arch_hardware_enable();
4552 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
4553 atomic_inc(&hardware_enable_failed
);
4554 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
4558 static int kvm_starting_cpu(unsigned int cpu
)
4560 raw_spin_lock(&kvm_count_lock
);
4561 if (kvm_usage_count
)
4562 hardware_enable_nolock(NULL
);
4563 raw_spin_unlock(&kvm_count_lock
);
4567 static void hardware_disable_nolock(void *junk
)
4569 int cpu
= raw_smp_processor_id();
4571 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
4573 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
4574 kvm_arch_hardware_disable();
4577 static int kvm_dying_cpu(unsigned int cpu
)
4579 raw_spin_lock(&kvm_count_lock
);
4580 if (kvm_usage_count
)
4581 hardware_disable_nolock(NULL
);
4582 raw_spin_unlock(&kvm_count_lock
);
4586 static void hardware_disable_all_nolock(void)
4588 BUG_ON(!kvm_usage_count
);
4591 if (!kvm_usage_count
)
4592 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4595 static void hardware_disable_all(void)
4597 raw_spin_lock(&kvm_count_lock
);
4598 hardware_disable_all_nolock();
4599 raw_spin_unlock(&kvm_count_lock
);
4602 static int hardware_enable_all(void)
4606 raw_spin_lock(&kvm_count_lock
);
4609 if (kvm_usage_count
== 1) {
4610 atomic_set(&hardware_enable_failed
, 0);
4611 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
4613 if (atomic_read(&hardware_enable_failed
)) {
4614 hardware_disable_all_nolock();
4619 raw_spin_unlock(&kvm_count_lock
);
4624 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
4628 * Some (well, at least mine) BIOSes hang on reboot if
4631 * And Intel TXT required VMX off for all cpu when system shutdown.
4633 pr_info("kvm: exiting hardware virtualization\n");
4634 kvm_rebooting
= true;
4635 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4639 static struct notifier_block kvm_reboot_notifier
= {
4640 .notifier_call
= kvm_reboot
,
4644 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
4648 for (i
= 0; i
< bus
->dev_count
; i
++) {
4649 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
4651 kvm_iodevice_destructor(pos
);
4656 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
4657 const struct kvm_io_range
*r2
)
4659 gpa_t addr1
= r1
->addr
;
4660 gpa_t addr2
= r2
->addr
;
4665 /* If r2->len == 0, match the exact address. If r2->len != 0,
4666 * accept any overlapping write. Any order is acceptable for
4667 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4668 * we process all of them.
4681 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
4683 return kvm_io_bus_cmp(p1
, p2
);
4686 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
4687 gpa_t addr
, int len
)
4689 struct kvm_io_range
*range
, key
;
4692 key
= (struct kvm_io_range
) {
4697 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
4698 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
4702 off
= range
- bus
->range
;
4704 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
4710 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
4711 struct kvm_io_range
*range
, const void *val
)
4715 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
4719 while (idx
< bus
->dev_count
&&
4720 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
4721 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
4730 /* kvm_io_bus_write - called under kvm->slots_lock */
4731 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
4732 int len
, const void *val
)
4734 struct kvm_io_bus
*bus
;
4735 struct kvm_io_range range
;
4738 range
= (struct kvm_io_range
) {
4743 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4746 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
4747 return r
< 0 ? r
: 0;
4749 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
4751 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4752 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
4753 gpa_t addr
, int len
, const void *val
, long cookie
)
4755 struct kvm_io_bus
*bus
;
4756 struct kvm_io_range range
;
4758 range
= (struct kvm_io_range
) {
4763 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4767 /* First try the device referenced by cookie. */
4768 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
4769 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
4770 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
4775 * cookie contained garbage; fall back to search and return the
4776 * correct cookie value.
4778 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
4781 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
4782 struct kvm_io_range
*range
, void *val
)
4786 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
4790 while (idx
< bus
->dev_count
&&
4791 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
4792 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
4801 /* kvm_io_bus_read - called under kvm->slots_lock */
4802 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
4805 struct kvm_io_bus
*bus
;
4806 struct kvm_io_range range
;
4809 range
= (struct kvm_io_range
) {
4814 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4817 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
4818 return r
< 0 ? r
: 0;
4821 /* Caller must hold slots_lock. */
4822 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
4823 int len
, struct kvm_io_device
*dev
)
4826 struct kvm_io_bus
*new_bus
, *bus
;
4827 struct kvm_io_range range
;
4829 bus
= kvm_get_bus(kvm
, bus_idx
);
4833 /* exclude ioeventfd which is limited by maximum fd */
4834 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
4837 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
4838 GFP_KERNEL_ACCOUNT
);
4842 range
= (struct kvm_io_range
) {
4848 for (i
= 0; i
< bus
->dev_count
; i
++)
4849 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
4852 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
4853 new_bus
->dev_count
++;
4854 new_bus
->range
[i
] = range
;
4855 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
4856 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
4857 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
4858 synchronize_srcu_expedited(&kvm
->srcu
);
4864 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
4865 struct kvm_io_device
*dev
)
4868 struct kvm_io_bus
*new_bus
, *bus
;
4870 lockdep_assert_held(&kvm
->slots_lock
);
4872 bus
= kvm_get_bus(kvm
, bus_idx
);
4876 for (i
= 0; i
< bus
->dev_count
; i
++) {
4877 if (bus
->range
[i
].dev
== dev
) {
4882 if (i
== bus
->dev_count
)
4885 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
4886 GFP_KERNEL_ACCOUNT
);
4888 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
4889 new_bus
->dev_count
--;
4890 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
4891 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
4894 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
4895 synchronize_srcu_expedited(&kvm
->srcu
);
4897 /* Destroy the old bus _after_ installing the (null) bus. */
4899 pr_err("kvm: failed to shrink bus, removing it completely\n");
4900 for (j
= 0; j
< bus
->dev_count
; j
++) {
4903 kvm_iodevice_destructor(bus
->range
[j
].dev
);
4908 return new_bus
? 0 : -ENOMEM
;
4911 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
4914 struct kvm_io_bus
*bus
;
4915 int dev_idx
, srcu_idx
;
4916 struct kvm_io_device
*iodev
= NULL
;
4918 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
4920 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
4924 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
4928 iodev
= bus
->range
[dev_idx
].dev
;
4931 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
4935 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
4937 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
4938 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
4941 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
4944 /* The debugfs files are a reference to the kvm struct which
4945 * is still valid when kvm_destroy_vm is called.
4946 * To avoid the race between open and the removal of the debugfs
4947 * directory we test against the users count.
4949 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
4952 if (simple_attr_open(inode
, file
, get
,
4953 kvm_stats_debugfs_mode(stat_data
->desc
) & 0222
4956 kvm_put_kvm(stat_data
->kvm
);
4963 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
4965 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
4968 simple_attr_release(inode
, file
);
4969 kvm_put_kvm(stat_data
->kvm
);
4974 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
4976 *val
= *(u64
*)((void *)(&kvm
->stat
) + offset
);
4981 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
4983 *(u64
*)((void *)(&kvm
->stat
) + offset
) = 0;
4988 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
4991 struct kvm_vcpu
*vcpu
;
4995 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4996 *val
+= *(u64
*)((void *)(&vcpu
->stat
) + offset
);
5001 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
5004 struct kvm_vcpu
*vcpu
;
5006 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5007 *(u64
*)((void *)(&vcpu
->stat
) + offset
) = 0;
5012 static int kvm_stat_data_get(void *data
, u64
*val
)
5015 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
5017 switch (stat_data
->kind
) {
5019 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
5020 stat_data
->desc
->desc
.offset
, val
);
5023 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
5024 stat_data
->desc
->desc
.offset
, val
);
5031 static int kvm_stat_data_clear(void *data
, u64 val
)
5034 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
5039 switch (stat_data
->kind
) {
5041 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
5042 stat_data
->desc
->desc
.offset
);
5045 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
5046 stat_data
->desc
->desc
.offset
);
5053 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
5055 __simple_attr_check_format("%llu\n", 0ull);
5056 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
5057 kvm_stat_data_clear
, "%llu\n");
5060 static const struct file_operations stat_fops_per_vm
= {
5061 .owner
= THIS_MODULE
,
5062 .open
= kvm_stat_data_open
,
5063 .release
= kvm_debugfs_release
,
5064 .read
= simple_attr_read
,
5065 .write
= simple_attr_write
,
5066 .llseek
= no_llseek
,
5069 static int vm_stat_get(void *_offset
, u64
*val
)
5071 unsigned offset
= (long)_offset
;
5076 mutex_lock(&kvm_lock
);
5077 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5078 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
5081 mutex_unlock(&kvm_lock
);
5085 static int vm_stat_clear(void *_offset
, u64 val
)
5087 unsigned offset
= (long)_offset
;
5093 mutex_lock(&kvm_lock
);
5094 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5095 kvm_clear_stat_per_vm(kvm
, offset
);
5097 mutex_unlock(&kvm_lock
);
5102 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
5103 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops
, vm_stat_get
, NULL
, "%llu\n");
5105 static int vcpu_stat_get(void *_offset
, u64
*val
)
5107 unsigned offset
= (long)_offset
;
5112 mutex_lock(&kvm_lock
);
5113 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5114 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
5117 mutex_unlock(&kvm_lock
);
5121 static int vcpu_stat_clear(void *_offset
, u64 val
)
5123 unsigned offset
= (long)_offset
;
5129 mutex_lock(&kvm_lock
);
5130 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5131 kvm_clear_stat_per_vcpu(kvm
, offset
);
5133 mutex_unlock(&kvm_lock
);
5138 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
5140 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops
, vcpu_stat_get
, NULL
, "%llu\n");
5142 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
5144 struct kobj_uevent_env
*env
;
5145 unsigned long long created
, active
;
5147 if (!kvm_dev
.this_device
|| !kvm
)
5150 mutex_lock(&kvm_lock
);
5151 if (type
== KVM_EVENT_CREATE_VM
) {
5152 kvm_createvm_count
++;
5154 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5157 created
= kvm_createvm_count
;
5158 active
= kvm_active_vms
;
5159 mutex_unlock(&kvm_lock
);
5161 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
5165 add_uevent_var(env
, "CREATED=%llu", created
);
5166 add_uevent_var(env
, "COUNT=%llu", active
);
5168 if (type
== KVM_EVENT_CREATE_VM
) {
5169 add_uevent_var(env
, "EVENT=create");
5170 kvm
->userspace_pid
= task_pid_nr(current
);
5171 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5172 add_uevent_var(env
, "EVENT=destroy");
5174 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
5176 if (!IS_ERR_OR_NULL(kvm
->debugfs_dentry
)) {
5177 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
5180 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
5182 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
5186 /* no need for checks, since we are adding at most only 5 keys */
5187 env
->envp
[env
->envp_idx
++] = NULL
;
5188 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
5192 static void kvm_init_debug(void)
5194 const struct file_operations
*fops
;
5195 const struct _kvm_stats_desc
*pdesc
;
5198 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
5200 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
5201 pdesc
= &kvm_vm_stats_desc
[i
];
5202 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
5203 fops
= &vm_stat_fops
;
5205 fops
= &vm_stat_readonly_fops
;
5206 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
5208 (void *)(long)pdesc
->desc
.offset
, fops
);
5211 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
5212 pdesc
= &kvm_vcpu_stats_desc
[i
];
5213 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
5214 fops
= &vcpu_stat_fops
;
5216 fops
= &vcpu_stat_readonly_fops
;
5217 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
5219 (void *)(long)pdesc
->desc
.offset
, fops
);
5223 static int kvm_suspend(void)
5225 if (kvm_usage_count
)
5226 hardware_disable_nolock(NULL
);
5230 static void kvm_resume(void)
5232 if (kvm_usage_count
) {
5233 #ifdef CONFIG_LOCKDEP
5234 WARN_ON(lockdep_is_held(&kvm_count_lock
));
5236 hardware_enable_nolock(NULL
);
5240 static struct syscore_ops kvm_syscore_ops
= {
5241 .suspend
= kvm_suspend
,
5242 .resume
= kvm_resume
,
5246 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
5248 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
5251 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
5253 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
5255 WRITE_ONCE(vcpu
->preempted
, false);
5256 WRITE_ONCE(vcpu
->ready
, false);
5258 __this_cpu_write(kvm_running_vcpu
, vcpu
);
5259 kvm_arch_sched_in(vcpu
, cpu
);
5260 kvm_arch_vcpu_load(vcpu
, cpu
);
5263 static void kvm_sched_out(struct preempt_notifier
*pn
,
5264 struct task_struct
*next
)
5266 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
5268 if (current
->on_rq
) {
5269 WRITE_ONCE(vcpu
->preempted
, true);
5270 WRITE_ONCE(vcpu
->ready
, true);
5272 kvm_arch_vcpu_put(vcpu
);
5273 __this_cpu_write(kvm_running_vcpu
, NULL
);
5277 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5279 * We can disable preemption locally around accessing the per-CPU variable,
5280 * and use the resolved vcpu pointer after enabling preemption again,
5281 * because even if the current thread is migrated to another CPU, reading
5282 * the per-CPU value later will give us the same value as we update the
5283 * per-CPU variable in the preempt notifier handlers.
5285 struct kvm_vcpu
*kvm_get_running_vcpu(void)
5287 struct kvm_vcpu
*vcpu
;
5290 vcpu
= __this_cpu_read(kvm_running_vcpu
);
5295 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
5298 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5300 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
5302 return &kvm_running_vcpu
;
5305 struct kvm_cpu_compat_check
{
5310 static void check_processor_compat(void *data
)
5312 struct kvm_cpu_compat_check
*c
= data
;
5314 *c
->ret
= kvm_arch_check_processor_compat(c
->opaque
);
5317 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
5318 struct module
*module
)
5320 struct kvm_cpu_compat_check c
;
5324 r
= kvm_arch_init(opaque
);
5329 * kvm_arch_init makes sure there's at most one caller
5330 * for architectures that support multiple implementations,
5331 * like intel and amd on x86.
5332 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5333 * conflicts in case kvm is already setup for another implementation.
5335 r
= kvm_irqfd_init();
5339 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
5344 r
= kvm_arch_hardware_setup(opaque
);
5350 for_each_online_cpu(cpu
) {
5351 smp_call_function_single(cpu
, check_processor_compat
, &c
, 1);
5356 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
5357 kvm_starting_cpu
, kvm_dying_cpu
);
5360 register_reboot_notifier(&kvm_reboot_notifier
);
5362 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5364 vcpu_align
= __alignof__(struct kvm_vcpu
);
5366 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
5368 offsetof(struct kvm_vcpu
, arch
),
5369 offsetofend(struct kvm_vcpu
, stats_id
)
5370 - offsetof(struct kvm_vcpu
, arch
),
5372 if (!kvm_vcpu_cache
) {
5377 r
= kvm_async_pf_init();
5381 kvm_chardev_ops
.owner
= module
;
5382 kvm_vm_fops
.owner
= module
;
5383 kvm_vcpu_fops
.owner
= module
;
5385 r
= misc_register(&kvm_dev
);
5387 pr_err("kvm: misc device register failed\n");
5391 register_syscore_ops(&kvm_syscore_ops
);
5393 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
5394 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
5398 r
= kvm_vfio_ops_init();
5404 kvm_async_pf_deinit();
5406 kmem_cache_destroy(kvm_vcpu_cache
);
5408 unregister_reboot_notifier(&kvm_reboot_notifier
);
5409 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
5411 kvm_arch_hardware_unsetup();
5413 free_cpumask_var(cpus_hardware_enabled
);
5421 EXPORT_SYMBOL_GPL(kvm_init
);
5425 debugfs_remove_recursive(kvm_debugfs_dir
);
5426 misc_deregister(&kvm_dev
);
5427 kmem_cache_destroy(kvm_vcpu_cache
);
5428 kvm_async_pf_deinit();
5429 unregister_syscore_ops(&kvm_syscore_ops
);
5430 unregister_reboot_notifier(&kvm_reboot_notifier
);
5431 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
5432 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5433 kvm_arch_hardware_unsetup();
5436 free_cpumask_var(cpus_hardware_enabled
);
5437 kvm_vfio_ops_exit();
5439 EXPORT_SYMBOL_GPL(kvm_exit
);
5441 struct kvm_vm_worker_thread_context
{
5443 struct task_struct
*parent
;
5444 struct completion init_done
;
5445 kvm_vm_thread_fn_t thread_fn
;
5450 static int kvm_vm_worker_thread(void *context
)
5453 * The init_context is allocated on the stack of the parent thread, so
5454 * we have to locally copy anything that is needed beyond initialization
5456 struct kvm_vm_worker_thread_context
*init_context
= context
;
5457 struct kvm
*kvm
= init_context
->kvm
;
5458 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
5459 uintptr_t data
= init_context
->data
;
5462 err
= kthread_park(current
);
5463 /* kthread_park(current) is never supposed to return an error */
5468 err
= cgroup_attach_task_all(init_context
->parent
, current
);
5470 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5475 set_user_nice(current
, task_nice(init_context
->parent
));
5478 init_context
->err
= err
;
5479 complete(&init_context
->init_done
);
5480 init_context
= NULL
;
5485 /* Wait to be woken up by the spawner before proceeding. */
5488 if (!kthread_should_stop())
5489 err
= thread_fn(kvm
, data
);
5494 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
5495 uintptr_t data
, const char *name
,
5496 struct task_struct
**thread_ptr
)
5498 struct kvm_vm_worker_thread_context init_context
= {};
5499 struct task_struct
*thread
;
5502 init_context
.kvm
= kvm
;
5503 init_context
.parent
= current
;
5504 init_context
.thread_fn
= thread_fn
;
5505 init_context
.data
= data
;
5506 init_completion(&init_context
.init_done
);
5508 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
5509 "%s-%d", name
, task_pid_nr(current
));
5511 return PTR_ERR(thread
);
5513 /* kthread_run is never supposed to return NULL */
5514 WARN_ON(thread
== NULL
);
5516 wait_for_completion(&init_context
.init_done
);
5518 if (!init_context
.err
)
5519 *thread_ptr
= thread
;
5521 return init_context
.err
;