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 #include <trace/events/ipi.h>
67 #define CREATE_TRACE_POINTS
68 #include <trace/events/kvm.h>
70 #include <linux/kvm_dirty_ring.h>
73 /* Worst case buffer size needed for holding an integer. */
74 #define ITOA_MAX_LEN 12
76 MODULE_AUTHOR("Qumranet");
77 MODULE_LICENSE("GPL");
79 /* Architectures should define their poll value according to the halt latency */
80 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
81 module_param(halt_poll_ns
, uint
, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns
);
84 /* Default doubles per-vcpu halt_poll_ns. */
85 unsigned int halt_poll_ns_grow
= 2;
86 module_param(halt_poll_ns_grow
, uint
, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
89 /* The start value to grow halt_poll_ns from */
90 unsigned int halt_poll_ns_grow_start
= 10000; /* 10us */
91 module_param(halt_poll_ns_grow_start
, uint
, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start
);
94 /* Default resets per-vcpu halt_poll_ns . */
95 unsigned int halt_poll_ns_shrink
;
96 module_param(halt_poll_ns_shrink
, uint
, 0644);
97 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
102 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
105 DEFINE_MUTEX(kvm_lock
);
108 static struct kmem_cache
*kvm_vcpu_cache
;
110 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
111 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_running_vcpu
);
113 struct dentry
*kvm_debugfs_dir
;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
116 static const struct file_operations stat_fops_per_vm
;
118 static struct file_operations kvm_chardev_ops
;
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 #define KVM_EVENT_CREATE_VM 0
150 #define KVM_EVENT_DESTROY_VM 1
151 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
152 static unsigned long long kvm_createvm_count
;
153 static unsigned long long kvm_active_vms
;
155 static DEFINE_PER_CPU(cpumask_var_t
, cpu_kick_mask
);
157 __weak
void kvm_arch_guest_memory_reclaimed(struct kvm
*kvm
)
161 bool kvm_is_zone_device_page(struct page
*page
)
164 * The metadata used by is_zone_device_page() to determine whether or
165 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
166 * the device has been pinned, e.g. by get_user_pages(). WARN if the
167 * page_count() is zero to help detect bad usage of this helper.
169 if (WARN_ON_ONCE(!page_count(page
)))
172 return is_zone_device_page(page
);
176 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
177 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
178 * is likely incomplete, it has been compiled purely through people wanting to
179 * back guest with a certain type of memory and encountering issues.
181 struct page
*kvm_pfn_to_refcounted_page(kvm_pfn_t pfn
)
188 page
= pfn_to_page(pfn
);
189 if (!PageReserved(page
))
192 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
193 if (is_zero_pfn(pfn
))
197 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
198 * perspective they are "normal" pages, albeit with slightly different
201 if (kvm_is_zone_device_page(page
))
208 * Switches to specified vcpu, until a matching vcpu_put()
210 void vcpu_load(struct kvm_vcpu
*vcpu
)
214 __this_cpu_write(kvm_running_vcpu
, vcpu
);
215 preempt_notifier_register(&vcpu
->preempt_notifier
);
216 kvm_arch_vcpu_load(vcpu
, cpu
);
219 EXPORT_SYMBOL_GPL(vcpu_load
);
221 void vcpu_put(struct kvm_vcpu
*vcpu
)
224 kvm_arch_vcpu_put(vcpu
);
225 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
226 __this_cpu_write(kvm_running_vcpu
, NULL
);
229 EXPORT_SYMBOL_GPL(vcpu_put
);
231 /* TODO: merge with kvm_arch_vcpu_should_kick */
232 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
234 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
237 * We need to wait for the VCPU to reenable interrupts and get out of
238 * READING_SHADOW_PAGE_TABLES mode.
240 if (req
& KVM_REQUEST_WAIT
)
241 return mode
!= OUTSIDE_GUEST_MODE
;
244 * Need to kick a running VCPU, but otherwise there is nothing to do.
246 return mode
== IN_GUEST_MODE
;
249 static void ack_kick(void *_completed
)
253 static inline bool kvm_kick_many_cpus(struct cpumask
*cpus
, bool wait
)
255 if (cpumask_empty(cpus
))
258 smp_call_function_many(cpus
, ack_kick
, NULL
, wait
);
262 static void kvm_make_vcpu_request(struct kvm_vcpu
*vcpu
, unsigned int req
,
263 struct cpumask
*tmp
, int current_cpu
)
267 if (likely(!(req
& KVM_REQUEST_NO_ACTION
)))
268 __kvm_make_request(req
, vcpu
);
270 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
274 * Note, the vCPU could get migrated to a different pCPU at any point
275 * after kvm_request_needs_ipi(), which could result in sending an IPI
276 * to the previous pCPU. But, that's OK because the purpose of the IPI
277 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
278 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
279 * after this point is also OK, as the requirement is only that KVM wait
280 * for vCPUs that were reading SPTEs _before_ any changes were
281 * finalized. See kvm_vcpu_kick() for more details on handling requests.
283 if (kvm_request_needs_ipi(vcpu
, req
)) {
284 cpu
= READ_ONCE(vcpu
->cpu
);
285 if (cpu
!= -1 && cpu
!= current_cpu
)
286 __cpumask_set_cpu(cpu
, tmp
);
290 bool kvm_make_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
291 unsigned long *vcpu_bitmap
)
293 struct kvm_vcpu
*vcpu
;
294 struct cpumask
*cpus
;
300 cpus
= this_cpu_cpumask_var_ptr(cpu_kick_mask
);
303 for_each_set_bit(i
, vcpu_bitmap
, KVM_MAX_VCPUS
) {
304 vcpu
= kvm_get_vcpu(kvm
, i
);
307 kvm_make_vcpu_request(vcpu
, req
, cpus
, me
);
310 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
316 bool kvm_make_all_cpus_request_except(struct kvm
*kvm
, unsigned int req
,
317 struct kvm_vcpu
*except
)
319 struct kvm_vcpu
*vcpu
;
320 struct cpumask
*cpus
;
327 cpus
= this_cpu_cpumask_var_ptr(cpu_kick_mask
);
330 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
333 kvm_make_vcpu_request(vcpu
, req
, cpus
, me
);
336 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
342 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
344 return kvm_make_all_cpus_request_except(kvm
, req
, NULL
);
346 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request
);
348 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
350 ++kvm
->stat
.generic
.remote_tlb_flush_requests
;
353 * We want to publish modifications to the page tables before reading
354 * mode. Pairs with a memory barrier in arch-specific code.
355 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
356 * and smp_mb in walk_shadow_page_lockless_begin/end.
357 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
359 * There is already an smp_mb__after_atomic() before
360 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
363 if (!kvm_arch_flush_remote_tlbs(kvm
)
364 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
365 ++kvm
->stat
.generic
.remote_tlb_flush
;
367 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
369 void kvm_flush_remote_tlbs_range(struct kvm
*kvm
, gfn_t gfn
, u64 nr_pages
)
371 if (!kvm_arch_flush_remote_tlbs_range(kvm
, gfn
, nr_pages
))
375 * Fall back to a flushing entire TLBs if the architecture range-based
376 * TLB invalidation is unsupported or can't be performed for whatever
379 kvm_flush_remote_tlbs(kvm
);
382 void kvm_flush_remote_tlbs_memslot(struct kvm
*kvm
,
383 const struct kvm_memory_slot
*memslot
)
386 * All current use cases for flushing the TLBs for a specific memslot
387 * are related to dirty logging, and many do the TLB flush out of
388 * mmu_lock. The interaction between the various operations on memslot
389 * must be serialized by slots_locks to ensure the TLB flush from one
390 * operation is observed by any other operation on the same memslot.
392 lockdep_assert_held(&kvm
->slots_lock
);
393 kvm_flush_remote_tlbs_range(kvm
, memslot
->base_gfn
, memslot
->npages
);
396 static void kvm_flush_shadow_all(struct kvm
*kvm
)
398 kvm_arch_flush_shadow_all(kvm
);
399 kvm_arch_guest_memory_reclaimed(kvm
);
402 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
403 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache
*mc
,
406 gfp_flags
|= mc
->gfp_zero
;
409 return kmem_cache_alloc(mc
->kmem_cache
, gfp_flags
);
411 return (void *)__get_free_page(gfp_flags
);
414 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int capacity
, int min
)
416 gfp_t gfp
= mc
->gfp_custom
? mc
->gfp_custom
: GFP_KERNEL_ACCOUNT
;
419 if (mc
->nobjs
>= min
)
422 if (unlikely(!mc
->objects
)) {
423 if (WARN_ON_ONCE(!capacity
))
426 mc
->objects
= kvmalloc_array(sizeof(void *), capacity
, gfp
);
430 mc
->capacity
= capacity
;
433 /* It is illegal to request a different capacity across topups. */
434 if (WARN_ON_ONCE(mc
->capacity
!= capacity
))
437 while (mc
->nobjs
< mc
->capacity
) {
438 obj
= mmu_memory_cache_alloc_obj(mc
, gfp
);
440 return mc
->nobjs
>= min
? 0 : -ENOMEM
;
441 mc
->objects
[mc
->nobjs
++] = obj
;
446 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int min
)
448 return __kvm_mmu_topup_memory_cache(mc
, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
, min
);
451 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache
*mc
)
456 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
460 kmem_cache_free(mc
->kmem_cache
, mc
->objects
[--mc
->nobjs
]);
462 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
471 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
)
475 if (WARN_ON(!mc
->nobjs
))
476 p
= mmu_memory_cache_alloc_obj(mc
, GFP_ATOMIC
| __GFP_ACCOUNT
);
478 p
= mc
->objects
[--mc
->nobjs
];
484 static void kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
486 mutex_init(&vcpu
->mutex
);
491 #ifndef __KVM_HAVE_ARCH_WQP
492 rcuwait_init(&vcpu
->wait
);
494 kvm_async_pf_vcpu_init(vcpu
);
496 kvm_vcpu_set_in_spin_loop(vcpu
, false);
497 kvm_vcpu_set_dy_eligible(vcpu
, false);
498 vcpu
->preempted
= false;
500 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
501 vcpu
->last_used_slot
= NULL
;
503 /* Fill the stats id string for the vcpu */
504 snprintf(vcpu
->stats_id
, sizeof(vcpu
->stats_id
), "kvm-%d/vcpu-%d",
505 task_pid_nr(current
), id
);
508 static void kvm_vcpu_destroy(struct kvm_vcpu
*vcpu
)
510 kvm_arch_vcpu_destroy(vcpu
);
511 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
514 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
515 * the vcpu->pid pointer, and at destruction time all file descriptors
518 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
520 free_page((unsigned long)vcpu
->run
);
521 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
524 void kvm_destroy_vcpus(struct kvm
*kvm
)
527 struct kvm_vcpu
*vcpu
;
529 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
530 kvm_vcpu_destroy(vcpu
);
531 xa_erase(&kvm
->vcpu_array
, i
);
534 atomic_set(&kvm
->online_vcpus
, 0);
536 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus
);
538 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
539 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
541 return container_of(mn
, struct kvm
, mmu_notifier
);
544 typedef bool (*gfn_handler_t
)(struct kvm
*kvm
, struct kvm_gfn_range
*range
);
546 typedef void (*on_lock_fn_t
)(struct kvm
*kvm
);
547 typedef void (*on_unlock_fn_t
)(struct kvm
*kvm
);
549 struct kvm_mmu_notifier_range
{
551 * 64-bit addresses, as KVM notifiers can operate on host virtual
552 * addresses (unsigned long) and guest physical addresses (64-bit).
556 union kvm_mmu_notifier_arg arg
;
557 gfn_handler_t handler
;
558 on_lock_fn_t on_lock
;
559 on_unlock_fn_t on_unlock
;
565 * Use a dedicated stub instead of NULL to indicate that there is no callback
566 * function/handler. The compiler technically can't guarantee that a real
567 * function will have a non-zero address, and so it will generate code to
568 * check for !NULL, whereas comparing against a stub will be elided at compile
569 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
571 static void kvm_null_fn(void)
575 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
577 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG
;
579 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
580 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
581 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
583 node = interval_tree_iter_next(node, start, last)) \
585 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
586 const struct kvm_mmu_notifier_range
*range
)
588 bool ret
= false, locked
= false;
589 struct kvm_gfn_range gfn_range
;
590 struct kvm_memory_slot
*slot
;
591 struct kvm_memslots
*slots
;
594 if (WARN_ON_ONCE(range
->end
<= range
->start
))
597 /* A null handler is allowed if and only if on_lock() is provided. */
598 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range
->on_lock
) &&
599 IS_KVM_NULL_FN(range
->handler
)))
602 idx
= srcu_read_lock(&kvm
->srcu
);
604 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
605 struct interval_tree_node
*node
;
607 slots
= __kvm_memslots(kvm
, i
);
608 kvm_for_each_memslot_in_hva_range(node
, slots
,
609 range
->start
, range
->end
- 1) {
610 unsigned long hva_start
, hva_end
;
612 slot
= container_of(node
, struct kvm_memory_slot
, hva_node
[slots
->node_idx
]);
613 hva_start
= max_t(unsigned long, range
->start
, slot
->userspace_addr
);
614 hva_end
= min_t(unsigned long, range
->end
,
615 slot
->userspace_addr
+ (slot
->npages
<< PAGE_SHIFT
));
618 * To optimize for the likely case where the address
619 * range is covered by zero or one memslots, don't
620 * bother making these conditional (to avoid writes on
621 * the second or later invocation of the handler).
623 gfn_range
.arg
= range
->arg
;
624 gfn_range
.may_block
= range
->may_block
;
627 * {gfn(page) | page intersects with [hva_start, hva_end)} =
628 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
630 gfn_range
.start
= hva_to_gfn_memslot(hva_start
, slot
);
631 gfn_range
.end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, slot
);
632 gfn_range
.slot
= slot
;
637 if (!IS_KVM_NULL_FN(range
->on_lock
))
640 if (IS_KVM_NULL_FN(range
->handler
))
643 ret
|= range
->handler(kvm
, &gfn_range
);
647 if (range
->flush_on_ret
&& ret
)
648 kvm_flush_remote_tlbs(kvm
);
652 if (!IS_KVM_NULL_FN(range
->on_unlock
))
653 range
->on_unlock(kvm
);
656 srcu_read_unlock(&kvm
->srcu
, idx
);
658 /* The notifiers are averse to booleans. :-( */
662 static __always_inline
int kvm_handle_hva_range(struct mmu_notifier
*mn
,
665 union kvm_mmu_notifier_arg arg
,
666 gfn_handler_t handler
)
668 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
669 const struct kvm_mmu_notifier_range range
= {
674 .on_lock
= (void *)kvm_null_fn
,
675 .on_unlock
= (void *)kvm_null_fn
,
676 .flush_on_ret
= true,
680 return __kvm_handle_hva_range(kvm
, &range
);
683 static __always_inline
int kvm_handle_hva_range_no_flush(struct mmu_notifier
*mn
,
686 gfn_handler_t handler
)
688 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
689 const struct kvm_mmu_notifier_range range
= {
693 .on_lock
= (void *)kvm_null_fn
,
694 .on_unlock
= (void *)kvm_null_fn
,
695 .flush_on_ret
= false,
699 return __kvm_handle_hva_range(kvm
, &range
);
702 static bool kvm_change_spte_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
705 * Skipping invalid memslots is correct if and only change_pte() is
706 * surrounded by invalidate_range_{start,end}(), which is currently
707 * guaranteed by the primary MMU. If that ever changes, KVM needs to
708 * unmap the memslot instead of skipping the memslot to ensure that KVM
709 * doesn't hold references to the old PFN.
711 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
713 if (range
->slot
->flags
& KVM_MEMSLOT_INVALID
)
716 return kvm_set_spte_gfn(kvm
, range
);
719 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
720 struct mm_struct
*mm
,
721 unsigned long address
,
724 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
725 const union kvm_mmu_notifier_arg arg
= { .pte
= pte
};
727 trace_kvm_set_spte_hva(address
);
730 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
731 * If mmu_invalidate_in_progress is zero, then no in-progress
732 * invalidations, including this one, found a relevant memslot at
733 * start(); rechecking memslots here is unnecessary. Note, a false
734 * positive (count elevated by a different invalidation) is sub-optimal
735 * but functionally ok.
737 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
738 if (!READ_ONCE(kvm
->mmu_invalidate_in_progress
))
741 kvm_handle_hva_range(mn
, address
, address
+ 1, arg
, kvm_change_spte_gfn
);
744 void kvm_mmu_invalidate_begin(struct kvm
*kvm
)
746 lockdep_assert_held_write(&kvm
->mmu_lock
);
748 * The count increase must become visible at unlock time as no
749 * spte can be established without taking the mmu_lock and
750 * count is also read inside the mmu_lock critical section.
752 kvm
->mmu_invalidate_in_progress
++;
754 if (likely(kvm
->mmu_invalidate_in_progress
== 1)) {
755 kvm
->mmu_invalidate_range_start
= INVALID_GPA
;
756 kvm
->mmu_invalidate_range_end
= INVALID_GPA
;
760 void kvm_mmu_invalidate_range_add(struct kvm
*kvm
, gfn_t start
, gfn_t end
)
762 lockdep_assert_held_write(&kvm
->mmu_lock
);
764 WARN_ON_ONCE(!kvm
->mmu_invalidate_in_progress
);
766 if (likely(kvm
->mmu_invalidate_range_start
== INVALID_GPA
)) {
767 kvm
->mmu_invalidate_range_start
= start
;
768 kvm
->mmu_invalidate_range_end
= end
;
771 * Fully tracking multiple concurrent ranges has diminishing
772 * returns. Keep things simple and just find the minimal range
773 * which includes the current and new ranges. As there won't be
774 * enough information to subtract a range after its invalidate
775 * completes, any ranges invalidated concurrently will
776 * accumulate and persist until all outstanding invalidates
779 kvm
->mmu_invalidate_range_start
=
780 min(kvm
->mmu_invalidate_range_start
, start
);
781 kvm
->mmu_invalidate_range_end
=
782 max(kvm
->mmu_invalidate_range_end
, end
);
786 static bool kvm_mmu_unmap_gfn_range(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
788 kvm_mmu_invalidate_range_add(kvm
, range
->start
, range
->end
);
789 return kvm_unmap_gfn_range(kvm
, range
);
792 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
793 const struct mmu_notifier_range
*range
)
795 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
796 const struct kvm_mmu_notifier_range hva_range
= {
797 .start
= range
->start
,
799 .handler
= kvm_mmu_unmap_gfn_range
,
800 .on_lock
= kvm_mmu_invalidate_begin
,
801 .on_unlock
= kvm_arch_guest_memory_reclaimed
,
802 .flush_on_ret
= true,
803 .may_block
= mmu_notifier_range_blockable(range
),
806 trace_kvm_unmap_hva_range(range
->start
, range
->end
);
809 * Prevent memslot modification between range_start() and range_end()
810 * so that conditionally locking provides the same result in both
811 * functions. Without that guarantee, the mmu_invalidate_in_progress
812 * adjustments will be imbalanced.
814 * Pairs with the decrement in range_end().
816 spin_lock(&kvm
->mn_invalidate_lock
);
817 kvm
->mn_active_invalidate_count
++;
818 spin_unlock(&kvm
->mn_invalidate_lock
);
821 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
822 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
823 * each cache's lock. There are relatively few caches in existence at
824 * any given time, and the caches themselves can check for hva overlap,
825 * i.e. don't need to rely on memslot overlap checks for performance.
826 * Because this runs without holding mmu_lock, the pfn caches must use
827 * mn_active_invalidate_count (see above) instead of
828 * mmu_invalidate_in_progress.
830 gfn_to_pfn_cache_invalidate_start(kvm
, range
->start
, range
->end
,
831 hva_range
.may_block
);
833 __kvm_handle_hva_range(kvm
, &hva_range
);
838 void kvm_mmu_invalidate_end(struct kvm
*kvm
)
840 lockdep_assert_held_write(&kvm
->mmu_lock
);
843 * This sequence increase will notify the kvm page fault that
844 * the page that is going to be mapped in the spte could have
847 kvm
->mmu_invalidate_seq
++;
850 * The above sequence increase must be visible before the
851 * below count decrease, which is ensured by the smp_wmb above
852 * in conjunction with the smp_rmb in mmu_invalidate_retry().
854 kvm
->mmu_invalidate_in_progress
--;
855 KVM_BUG_ON(kvm
->mmu_invalidate_in_progress
< 0, kvm
);
858 * Assert that at least one range was added between start() and end().
859 * Not adding a range isn't fatal, but it is a KVM bug.
861 WARN_ON_ONCE(kvm
->mmu_invalidate_range_start
== INVALID_GPA
);
864 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
865 const struct mmu_notifier_range
*range
)
867 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
868 const struct kvm_mmu_notifier_range hva_range
= {
869 .start
= range
->start
,
871 .handler
= (void *)kvm_null_fn
,
872 .on_lock
= kvm_mmu_invalidate_end
,
873 .on_unlock
= (void *)kvm_null_fn
,
874 .flush_on_ret
= false,
875 .may_block
= mmu_notifier_range_blockable(range
),
879 __kvm_handle_hva_range(kvm
, &hva_range
);
881 /* Pairs with the increment in range_start(). */
882 spin_lock(&kvm
->mn_invalidate_lock
);
883 wake
= (--kvm
->mn_active_invalidate_count
== 0);
884 spin_unlock(&kvm
->mn_invalidate_lock
);
887 * There can only be one waiter, since the wait happens under
891 rcuwait_wake_up(&kvm
->mn_memslots_update_rcuwait
);
894 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
895 struct mm_struct
*mm
,
899 trace_kvm_age_hva(start
, end
);
901 return kvm_handle_hva_range(mn
, start
, end
, KVM_MMU_NOTIFIER_NO_ARG
,
905 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
906 struct mm_struct
*mm
,
910 trace_kvm_age_hva(start
, end
);
913 * Even though we do not flush TLB, this will still adversely
914 * affect performance on pre-Haswell Intel EPT, where there is
915 * no EPT Access Bit to clear so that we have to tear down EPT
916 * tables instead. If we find this unacceptable, we can always
917 * add a parameter to kvm_age_hva so that it effectively doesn't
918 * do anything on clear_young.
920 * Also note that currently we never issue secondary TLB flushes
921 * from clear_young, leaving this job up to the regular system
922 * cadence. If we find this inaccurate, we might come up with a
923 * more sophisticated heuristic later.
925 return kvm_handle_hva_range_no_flush(mn
, start
, end
, kvm_age_gfn
);
928 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
929 struct mm_struct
*mm
,
930 unsigned long address
)
932 trace_kvm_test_age_hva(address
);
934 return kvm_handle_hva_range_no_flush(mn
, address
, address
+ 1,
938 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
939 struct mm_struct
*mm
)
941 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
944 idx
= srcu_read_lock(&kvm
->srcu
);
945 kvm_flush_shadow_all(kvm
);
946 srcu_read_unlock(&kvm
->srcu
, idx
);
949 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
950 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
951 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
952 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
953 .clear_young
= kvm_mmu_notifier_clear_young
,
954 .test_young
= kvm_mmu_notifier_test_young
,
955 .change_pte
= kvm_mmu_notifier_change_pte
,
956 .release
= kvm_mmu_notifier_release
,
959 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
961 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
962 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
965 #else /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */
967 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
972 #endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */
974 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
975 static int kvm_pm_notifier_call(struct notifier_block
*bl
,
979 struct kvm
*kvm
= container_of(bl
, struct kvm
, pm_notifier
);
981 return kvm_arch_pm_notifier(kvm
, state
);
984 static void kvm_init_pm_notifier(struct kvm
*kvm
)
986 kvm
->pm_notifier
.notifier_call
= kvm_pm_notifier_call
;
987 /* Suspend KVM before we suspend ftrace, RCU, etc. */
988 kvm
->pm_notifier
.priority
= INT_MAX
;
989 register_pm_notifier(&kvm
->pm_notifier
);
992 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
994 unregister_pm_notifier(&kvm
->pm_notifier
);
996 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
997 static void kvm_init_pm_notifier(struct kvm
*kvm
)
1001 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
1004 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
1006 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1008 if (!memslot
->dirty_bitmap
)
1011 kvfree(memslot
->dirty_bitmap
);
1012 memslot
->dirty_bitmap
= NULL
;
1015 /* This does not remove the slot from struct kvm_memslots data structures */
1016 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
1018 kvm_destroy_dirty_bitmap(slot
);
1020 kvm_arch_free_memslot(kvm
, slot
);
1025 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
1027 struct hlist_node
*idnode
;
1028 struct kvm_memory_slot
*memslot
;
1032 * The same memslot objects live in both active and inactive sets,
1033 * arbitrarily free using index '1' so the second invocation of this
1034 * function isn't operating over a structure with dangling pointers
1035 * (even though this function isn't actually touching them).
1037 if (!slots
->node_idx
)
1040 hash_for_each_safe(slots
->id_hash
, bkt
, idnode
, memslot
, id_node
[1])
1041 kvm_free_memslot(kvm
, memslot
);
1044 static umode_t
kvm_stats_debugfs_mode(const struct _kvm_stats_desc
*pdesc
)
1046 switch (pdesc
->desc
.flags
& KVM_STATS_TYPE_MASK
) {
1047 case KVM_STATS_TYPE_INSTANT
:
1049 case KVM_STATS_TYPE_CUMULATIVE
:
1050 case KVM_STATS_TYPE_PEAK
:
1057 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
1060 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1061 kvm_vcpu_stats_header
.num_desc
;
1063 if (IS_ERR(kvm
->debugfs_dentry
))
1066 debugfs_remove_recursive(kvm
->debugfs_dentry
);
1068 if (kvm
->debugfs_stat_data
) {
1069 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
1070 kfree(kvm
->debugfs_stat_data
[i
]);
1071 kfree(kvm
->debugfs_stat_data
);
1075 static int kvm_create_vm_debugfs(struct kvm
*kvm
, const char *fdname
)
1077 static DEFINE_MUTEX(kvm_debugfs_lock
);
1078 struct dentry
*dent
;
1079 char dir_name
[ITOA_MAX_LEN
* 2];
1080 struct kvm_stat_data
*stat_data
;
1081 const struct _kvm_stats_desc
*pdesc
;
1082 int i
, ret
= -ENOMEM
;
1083 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1084 kvm_vcpu_stats_header
.num_desc
;
1086 if (!debugfs_initialized())
1089 snprintf(dir_name
, sizeof(dir_name
), "%d-%s", task_pid_nr(current
), fdname
);
1090 mutex_lock(&kvm_debugfs_lock
);
1091 dent
= debugfs_lookup(dir_name
, kvm_debugfs_dir
);
1093 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name
);
1095 mutex_unlock(&kvm_debugfs_lock
);
1098 dent
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
1099 mutex_unlock(&kvm_debugfs_lock
);
1103 kvm
->debugfs_dentry
= dent
;
1104 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
1105 sizeof(*kvm
->debugfs_stat_data
),
1106 GFP_KERNEL_ACCOUNT
);
1107 if (!kvm
->debugfs_stat_data
)
1110 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
1111 pdesc
= &kvm_vm_stats_desc
[i
];
1112 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1116 stat_data
->kvm
= kvm
;
1117 stat_data
->desc
= pdesc
;
1118 stat_data
->kind
= KVM_STAT_VM
;
1119 kvm
->debugfs_stat_data
[i
] = stat_data
;
1120 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1121 kvm
->debugfs_dentry
, stat_data
,
1125 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
1126 pdesc
= &kvm_vcpu_stats_desc
[i
];
1127 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1131 stat_data
->kvm
= kvm
;
1132 stat_data
->desc
= pdesc
;
1133 stat_data
->kind
= KVM_STAT_VCPU
;
1134 kvm
->debugfs_stat_data
[i
+ kvm_vm_stats_header
.num_desc
] = stat_data
;
1135 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1136 kvm
->debugfs_dentry
, stat_data
,
1140 ret
= kvm_arch_create_vm_debugfs(kvm
);
1146 kvm_destroy_vm_debugfs(kvm
);
1151 * Called after the VM is otherwise initialized, but just before adding it to
1154 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
1160 * Called just after removing the VM from the vm_list, but before doing any
1161 * other destruction.
1163 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
1168 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1169 * be setup already, so we can create arch-specific debugfs entries under it.
1170 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1171 * a per-arch destroy interface is not needed.
1173 int __weak
kvm_arch_create_vm_debugfs(struct kvm
*kvm
)
1178 static struct kvm
*kvm_create_vm(unsigned long type
, const char *fdname
)
1180 struct kvm
*kvm
= kvm_arch_alloc_vm();
1181 struct kvm_memslots
*slots
;
1186 return ERR_PTR(-ENOMEM
);
1188 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1189 __module_get(kvm_chardev_ops
.owner
);
1191 KVM_MMU_LOCK_INIT(kvm
);
1192 mmgrab(current
->mm
);
1193 kvm
->mm
= current
->mm
;
1194 kvm_eventfd_init(kvm
);
1195 mutex_init(&kvm
->lock
);
1196 mutex_init(&kvm
->irq_lock
);
1197 mutex_init(&kvm
->slots_lock
);
1198 mutex_init(&kvm
->slots_arch_lock
);
1199 spin_lock_init(&kvm
->mn_invalidate_lock
);
1200 rcuwait_init(&kvm
->mn_memslots_update_rcuwait
);
1201 xa_init(&kvm
->vcpu_array
);
1203 INIT_LIST_HEAD(&kvm
->gpc_list
);
1204 spin_lock_init(&kvm
->gpc_lock
);
1206 INIT_LIST_HEAD(&kvm
->devices
);
1207 kvm
->max_vcpus
= KVM_MAX_VCPUS
;
1209 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
1212 * Force subsequent debugfs file creations to fail if the VM directory
1213 * is not created (by kvm_create_vm_debugfs()).
1215 kvm
->debugfs_dentry
= ERR_PTR(-ENOENT
);
1217 snprintf(kvm
->stats_id
, sizeof(kvm
->stats_id
), "kvm-%d",
1218 task_pid_nr(current
));
1220 if (init_srcu_struct(&kvm
->srcu
))
1221 goto out_err_no_srcu
;
1222 if (init_srcu_struct(&kvm
->irq_srcu
))
1223 goto out_err_no_irq_srcu
;
1225 refcount_set(&kvm
->users_count
, 1);
1226 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
1227 for (j
= 0; j
< 2; j
++) {
1228 slots
= &kvm
->__memslots
[i
][j
];
1230 atomic_long_set(&slots
->last_used_slot
, (unsigned long)NULL
);
1231 slots
->hva_tree
= RB_ROOT_CACHED
;
1232 slots
->gfn_tree
= RB_ROOT
;
1233 hash_init(slots
->id_hash
);
1234 slots
->node_idx
= j
;
1236 /* Generations must be different for each address space. */
1237 slots
->generation
= i
;
1240 rcu_assign_pointer(kvm
->memslots
[i
], &kvm
->__memslots
[i
][0]);
1243 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1244 rcu_assign_pointer(kvm
->buses
[i
],
1245 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
1247 goto out_err_no_arch_destroy_vm
;
1250 r
= kvm_arch_init_vm(kvm
, type
);
1252 goto out_err_no_arch_destroy_vm
;
1254 r
= hardware_enable_all();
1256 goto out_err_no_disable
;
1258 #ifdef CONFIG_HAVE_KVM_IRQFD
1259 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
1262 r
= kvm_init_mmu_notifier(kvm
);
1264 goto out_err_no_mmu_notifier
;
1266 r
= kvm_coalesced_mmio_init(kvm
);
1268 goto out_no_coalesced_mmio
;
1270 r
= kvm_create_vm_debugfs(kvm
, fdname
);
1272 goto out_err_no_debugfs
;
1274 r
= kvm_arch_post_init_vm(kvm
);
1278 mutex_lock(&kvm_lock
);
1279 list_add(&kvm
->vm_list
, &vm_list
);
1280 mutex_unlock(&kvm_lock
);
1282 preempt_notifier_inc();
1283 kvm_init_pm_notifier(kvm
);
1288 kvm_destroy_vm_debugfs(kvm
);
1290 kvm_coalesced_mmio_free(kvm
);
1291 out_no_coalesced_mmio
:
1292 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1293 if (kvm
->mmu_notifier
.ops
)
1294 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
1296 out_err_no_mmu_notifier
:
1297 hardware_disable_all();
1299 kvm_arch_destroy_vm(kvm
);
1300 out_err_no_arch_destroy_vm
:
1301 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
1302 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
1303 kfree(kvm_get_bus(kvm
, i
));
1304 cleanup_srcu_struct(&kvm
->irq_srcu
);
1305 out_err_no_irq_srcu
:
1306 cleanup_srcu_struct(&kvm
->srcu
);
1308 kvm_arch_free_vm(kvm
);
1309 mmdrop(current
->mm
);
1310 module_put(kvm_chardev_ops
.owner
);
1314 static void kvm_destroy_devices(struct kvm
*kvm
)
1316 struct kvm_device
*dev
, *tmp
;
1319 * We do not need to take the kvm->lock here, because nobody else
1320 * has a reference to the struct kvm at this point and therefore
1321 * cannot access the devices list anyhow.
1323 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
1324 list_del(&dev
->vm_node
);
1325 dev
->ops
->destroy(dev
);
1329 static void kvm_destroy_vm(struct kvm
*kvm
)
1332 struct mm_struct
*mm
= kvm
->mm
;
1334 kvm_destroy_pm_notifier(kvm
);
1335 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
1336 kvm_destroy_vm_debugfs(kvm
);
1337 kvm_arch_sync_events(kvm
);
1338 mutex_lock(&kvm_lock
);
1339 list_del(&kvm
->vm_list
);
1340 mutex_unlock(&kvm_lock
);
1341 kvm_arch_pre_destroy_vm(kvm
);
1343 kvm_free_irq_routing(kvm
);
1344 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1345 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
1348 kvm_io_bus_destroy(bus
);
1349 kvm
->buses
[i
] = NULL
;
1351 kvm_coalesced_mmio_free(kvm
);
1352 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1353 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1355 * At this point, pending calls to invalidate_range_start()
1356 * have completed but no more MMU notifiers will run, so
1357 * mn_active_invalidate_count may remain unbalanced.
1358 * No threads can be waiting in kvm_swap_active_memslots() as the
1359 * last reference on KVM has been dropped, but freeing
1360 * memslots would deadlock without this manual intervention.
1362 * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU
1363 * notifier between a start() and end(), then there shouldn't be any
1364 * in-progress invalidations.
1366 WARN_ON(rcuwait_active(&kvm
->mn_memslots_update_rcuwait
));
1367 if (kvm
->mn_active_invalidate_count
)
1368 kvm
->mn_active_invalidate_count
= 0;
1370 WARN_ON(kvm
->mmu_invalidate_in_progress
);
1372 kvm_flush_shadow_all(kvm
);
1374 kvm_arch_destroy_vm(kvm
);
1375 kvm_destroy_devices(kvm
);
1376 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
1377 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][0]);
1378 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][1]);
1380 cleanup_srcu_struct(&kvm
->irq_srcu
);
1381 cleanup_srcu_struct(&kvm
->srcu
);
1382 kvm_arch_free_vm(kvm
);
1383 preempt_notifier_dec();
1384 hardware_disable_all();
1386 module_put(kvm_chardev_ops
.owner
);
1389 void kvm_get_kvm(struct kvm
*kvm
)
1391 refcount_inc(&kvm
->users_count
);
1393 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1396 * Make sure the vm is not during destruction, which is a safe version of
1397 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1399 bool kvm_get_kvm_safe(struct kvm
*kvm
)
1401 return refcount_inc_not_zero(&kvm
->users_count
);
1403 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe
);
1405 void kvm_put_kvm(struct kvm
*kvm
)
1407 if (refcount_dec_and_test(&kvm
->users_count
))
1408 kvm_destroy_vm(kvm
);
1410 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1413 * Used to put a reference that was taken on behalf of an object associated
1414 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1415 * of the new file descriptor fails and the reference cannot be transferred to
1416 * its final owner. In such cases, the caller is still actively using @kvm and
1417 * will fail miserably if the refcount unexpectedly hits zero.
1419 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
1421 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
1423 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
1425 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1427 struct kvm
*kvm
= filp
->private_data
;
1429 kvm_irqfd_release(kvm
);
1436 * Allocation size is twice as large as the actual dirty bitmap size.
1437 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1439 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1441 unsigned long dirty_bytes
= kvm_dirty_bitmap_bytes(memslot
);
1443 memslot
->dirty_bitmap
= __vcalloc(2, dirty_bytes
, GFP_KERNEL_ACCOUNT
);
1444 if (!memslot
->dirty_bitmap
)
1450 static struct kvm_memslots
*kvm_get_inactive_memslots(struct kvm
*kvm
, int as_id
)
1452 struct kvm_memslots
*active
= __kvm_memslots(kvm
, as_id
);
1453 int node_idx_inactive
= active
->node_idx
^ 1;
1455 return &kvm
->__memslots
[as_id
][node_idx_inactive
];
1459 * Helper to get the address space ID when one of memslot pointers may be NULL.
1460 * This also serves as a sanity that at least one of the pointers is non-NULL,
1461 * and that their address space IDs don't diverge.
1463 static int kvm_memslots_get_as_id(struct kvm_memory_slot
*a
,
1464 struct kvm_memory_slot
*b
)
1466 if (WARN_ON_ONCE(!a
&& !b
))
1474 WARN_ON_ONCE(a
->as_id
!= b
->as_id
);
1478 static void kvm_insert_gfn_node(struct kvm_memslots
*slots
,
1479 struct kvm_memory_slot
*slot
)
1481 struct rb_root
*gfn_tree
= &slots
->gfn_tree
;
1482 struct rb_node
**node
, *parent
;
1483 int idx
= slots
->node_idx
;
1486 for (node
= &gfn_tree
->rb_node
; *node
; ) {
1487 struct kvm_memory_slot
*tmp
;
1489 tmp
= container_of(*node
, struct kvm_memory_slot
, gfn_node
[idx
]);
1491 if (slot
->base_gfn
< tmp
->base_gfn
)
1492 node
= &(*node
)->rb_left
;
1493 else if (slot
->base_gfn
> tmp
->base_gfn
)
1494 node
= &(*node
)->rb_right
;
1499 rb_link_node(&slot
->gfn_node
[idx
], parent
, node
);
1500 rb_insert_color(&slot
->gfn_node
[idx
], gfn_tree
);
1503 static void kvm_erase_gfn_node(struct kvm_memslots
*slots
,
1504 struct kvm_memory_slot
*slot
)
1506 rb_erase(&slot
->gfn_node
[slots
->node_idx
], &slots
->gfn_tree
);
1509 static void kvm_replace_gfn_node(struct kvm_memslots
*slots
,
1510 struct kvm_memory_slot
*old
,
1511 struct kvm_memory_slot
*new)
1513 int idx
= slots
->node_idx
;
1515 WARN_ON_ONCE(old
->base_gfn
!= new->base_gfn
);
1517 rb_replace_node(&old
->gfn_node
[idx
], &new->gfn_node
[idx
],
1522 * Replace @old with @new in the inactive memslots.
1524 * With NULL @old this simply adds @new.
1525 * With NULL @new this simply removes @old.
1527 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1530 static void kvm_replace_memslot(struct kvm
*kvm
,
1531 struct kvm_memory_slot
*old
,
1532 struct kvm_memory_slot
*new)
1534 int as_id
= kvm_memslots_get_as_id(old
, new);
1535 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1536 int idx
= slots
->node_idx
;
1539 hash_del(&old
->id_node
[idx
]);
1540 interval_tree_remove(&old
->hva_node
[idx
], &slots
->hva_tree
);
1542 if ((long)old
== atomic_long_read(&slots
->last_used_slot
))
1543 atomic_long_set(&slots
->last_used_slot
, (long)new);
1546 kvm_erase_gfn_node(slots
, old
);
1552 * Initialize @new's hva range. Do this even when replacing an @old
1553 * slot, kvm_copy_memslot() deliberately does not touch node data.
1555 new->hva_node
[idx
].start
= new->userspace_addr
;
1556 new->hva_node
[idx
].last
= new->userspace_addr
+
1557 (new->npages
<< PAGE_SHIFT
) - 1;
1560 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1561 * hva_node needs to be swapped with remove+insert even though hva can't
1562 * change when replacing an existing slot.
1564 hash_add(slots
->id_hash
, &new->id_node
[idx
], new->id
);
1565 interval_tree_insert(&new->hva_node
[idx
], &slots
->hva_tree
);
1568 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1569 * switch the node in the gfn tree instead of removing the old and
1570 * inserting the new as two separate operations. Replacement is a
1571 * single O(1) operation versus two O(log(n)) operations for
1574 if (old
&& old
->base_gfn
== new->base_gfn
) {
1575 kvm_replace_gfn_node(slots
, old
, new);
1578 kvm_erase_gfn_node(slots
, old
);
1579 kvm_insert_gfn_node(slots
, new);
1584 * Flags that do not access any of the extra space of struct
1585 * kvm_userspace_memory_region2. KVM_SET_USER_MEMORY_REGION_V1_FLAGS
1586 * only allows these.
1588 #define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \
1589 (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY)
1591 static int check_memory_region_flags(const struct kvm_userspace_memory_region2
*mem
)
1593 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1595 #ifdef __KVM_HAVE_READONLY_MEM
1596 valid_flags
|= KVM_MEM_READONLY
;
1599 if (mem
->flags
& ~valid_flags
)
1605 static void kvm_swap_active_memslots(struct kvm
*kvm
, int as_id
)
1607 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1609 /* Grab the generation from the activate memslots. */
1610 u64 gen
= __kvm_memslots(kvm
, as_id
)->generation
;
1612 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1613 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1616 * Do not store the new memslots while there are invalidations in
1617 * progress, otherwise the locking in invalidate_range_start and
1618 * invalidate_range_end will be unbalanced.
1620 spin_lock(&kvm
->mn_invalidate_lock
);
1621 prepare_to_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1622 while (kvm
->mn_active_invalidate_count
) {
1623 set_current_state(TASK_UNINTERRUPTIBLE
);
1624 spin_unlock(&kvm
->mn_invalidate_lock
);
1626 spin_lock(&kvm
->mn_invalidate_lock
);
1628 finish_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1629 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1630 spin_unlock(&kvm
->mn_invalidate_lock
);
1633 * Acquired in kvm_set_memslot. Must be released before synchronize
1634 * SRCU below in order to avoid deadlock with another thread
1635 * acquiring the slots_arch_lock in an srcu critical section.
1637 mutex_unlock(&kvm
->slots_arch_lock
);
1639 synchronize_srcu_expedited(&kvm
->srcu
);
1642 * Increment the new memslot generation a second time, dropping the
1643 * update in-progress flag and incrementing the generation based on
1644 * the number of address spaces. This provides a unique and easily
1645 * identifiable generation number while the memslots are in flux.
1647 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1650 * Generations must be unique even across address spaces. We do not need
1651 * a global counter for that, instead the generation space is evenly split
1652 * across address spaces. For example, with two address spaces, address
1653 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1654 * use generations 1, 3, 5, ...
1656 gen
+= KVM_ADDRESS_SPACE_NUM
;
1658 kvm_arch_memslots_updated(kvm
, gen
);
1660 slots
->generation
= gen
;
1663 static int kvm_prepare_memory_region(struct kvm
*kvm
,
1664 const struct kvm_memory_slot
*old
,
1665 struct kvm_memory_slot
*new,
1666 enum kvm_mr_change change
)
1671 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1672 * will be freed on "commit". If logging is enabled in both old and
1673 * new, reuse the existing bitmap. If logging is enabled only in the
1674 * new and KVM isn't using a ring buffer, allocate and initialize a
1677 if (change
!= KVM_MR_DELETE
) {
1678 if (!(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1679 new->dirty_bitmap
= NULL
;
1680 else if (old
&& old
->dirty_bitmap
)
1681 new->dirty_bitmap
= old
->dirty_bitmap
;
1682 else if (kvm_use_dirty_bitmap(kvm
)) {
1683 r
= kvm_alloc_dirty_bitmap(new);
1687 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1688 bitmap_set(new->dirty_bitmap
, 0, new->npages
);
1692 r
= kvm_arch_prepare_memory_region(kvm
, old
, new, change
);
1694 /* Free the bitmap on failure if it was allocated above. */
1695 if (r
&& new && new->dirty_bitmap
&& (!old
|| !old
->dirty_bitmap
))
1696 kvm_destroy_dirty_bitmap(new);
1701 static void kvm_commit_memory_region(struct kvm
*kvm
,
1702 struct kvm_memory_slot
*old
,
1703 const struct kvm_memory_slot
*new,
1704 enum kvm_mr_change change
)
1706 int old_flags
= old
? old
->flags
: 0;
1707 int new_flags
= new ? new->flags
: 0;
1709 * Update the total number of memslot pages before calling the arch
1710 * hook so that architectures can consume the result directly.
1712 if (change
== KVM_MR_DELETE
)
1713 kvm
->nr_memslot_pages
-= old
->npages
;
1714 else if (change
== KVM_MR_CREATE
)
1715 kvm
->nr_memslot_pages
+= new->npages
;
1717 if ((old_flags
^ new_flags
) & KVM_MEM_LOG_DIRTY_PAGES
) {
1718 int change
= (new_flags
& KVM_MEM_LOG_DIRTY_PAGES
) ? 1 : -1;
1719 atomic_set(&kvm
->nr_memslots_dirty_logging
,
1720 atomic_read(&kvm
->nr_memslots_dirty_logging
) + change
);
1723 kvm_arch_commit_memory_region(kvm
, old
, new, change
);
1727 /* Nothing more to do. */
1730 /* Free the old memslot and all its metadata. */
1731 kvm_free_memslot(kvm
, old
);
1734 case KVM_MR_FLAGS_ONLY
:
1736 * Free the dirty bitmap as needed; the below check encompasses
1737 * both the flags and whether a ring buffer is being used)
1739 if (old
->dirty_bitmap
&& !new->dirty_bitmap
)
1740 kvm_destroy_dirty_bitmap(old
);
1743 * The final quirk. Free the detached, old slot, but only its
1744 * memory, not any metadata. Metadata, including arch specific
1745 * data, may be reused by @new.
1755 * Activate @new, which must be installed in the inactive slots by the caller,
1756 * by swapping the active slots and then propagating @new to @old once @old is
1757 * unreachable and can be safely modified.
1759 * With NULL @old this simply adds @new to @active (while swapping the sets).
1760 * With NULL @new this simply removes @old from @active and frees it
1761 * (while also swapping the sets).
1763 static void kvm_activate_memslot(struct kvm
*kvm
,
1764 struct kvm_memory_slot
*old
,
1765 struct kvm_memory_slot
*new)
1767 int as_id
= kvm_memslots_get_as_id(old
, new);
1769 kvm_swap_active_memslots(kvm
, as_id
);
1771 /* Propagate the new memslot to the now inactive memslots. */
1772 kvm_replace_memslot(kvm
, old
, new);
1775 static void kvm_copy_memslot(struct kvm_memory_slot
*dest
,
1776 const struct kvm_memory_slot
*src
)
1778 dest
->base_gfn
= src
->base_gfn
;
1779 dest
->npages
= src
->npages
;
1780 dest
->dirty_bitmap
= src
->dirty_bitmap
;
1781 dest
->arch
= src
->arch
;
1782 dest
->userspace_addr
= src
->userspace_addr
;
1783 dest
->flags
= src
->flags
;
1785 dest
->as_id
= src
->as_id
;
1788 static void kvm_invalidate_memslot(struct kvm
*kvm
,
1789 struct kvm_memory_slot
*old
,
1790 struct kvm_memory_slot
*invalid_slot
)
1793 * Mark the current slot INVALID. As with all memslot modifications,
1794 * this must be done on an unreachable slot to avoid modifying the
1795 * current slot in the active tree.
1797 kvm_copy_memslot(invalid_slot
, old
);
1798 invalid_slot
->flags
|= KVM_MEMSLOT_INVALID
;
1799 kvm_replace_memslot(kvm
, old
, invalid_slot
);
1802 * Activate the slot that is now marked INVALID, but don't propagate
1803 * the slot to the now inactive slots. The slot is either going to be
1804 * deleted or recreated as a new slot.
1806 kvm_swap_active_memslots(kvm
, old
->as_id
);
1809 * From this point no new shadow pages pointing to a deleted, or moved,
1810 * memslot will be created. Validation of sp->gfn happens in:
1811 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1812 * - kvm_is_visible_gfn (mmu_check_root)
1814 kvm_arch_flush_shadow_memslot(kvm
, old
);
1815 kvm_arch_guest_memory_reclaimed(kvm
);
1817 /* Was released by kvm_swap_active_memslots(), reacquire. */
1818 mutex_lock(&kvm
->slots_arch_lock
);
1821 * Copy the arch-specific field of the newly-installed slot back to the
1822 * old slot as the arch data could have changed between releasing
1823 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1824 * above. Writers are required to retrieve memslots *after* acquiring
1825 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1827 old
->arch
= invalid_slot
->arch
;
1830 static void kvm_create_memslot(struct kvm
*kvm
,
1831 struct kvm_memory_slot
*new)
1833 /* Add the new memslot to the inactive set and activate. */
1834 kvm_replace_memslot(kvm
, NULL
, new);
1835 kvm_activate_memslot(kvm
, NULL
, new);
1838 static void kvm_delete_memslot(struct kvm
*kvm
,
1839 struct kvm_memory_slot
*old
,
1840 struct kvm_memory_slot
*invalid_slot
)
1843 * Remove the old memslot (in the inactive memslots) by passing NULL as
1844 * the "new" slot, and for the invalid version in the active slots.
1846 kvm_replace_memslot(kvm
, old
, NULL
);
1847 kvm_activate_memslot(kvm
, invalid_slot
, NULL
);
1850 static void kvm_move_memslot(struct kvm
*kvm
,
1851 struct kvm_memory_slot
*old
,
1852 struct kvm_memory_slot
*new,
1853 struct kvm_memory_slot
*invalid_slot
)
1856 * Replace the old memslot in the inactive slots, and then swap slots
1857 * and replace the current INVALID with the new as well.
1859 kvm_replace_memslot(kvm
, old
, new);
1860 kvm_activate_memslot(kvm
, invalid_slot
, new);
1863 static void kvm_update_flags_memslot(struct kvm
*kvm
,
1864 struct kvm_memory_slot
*old
,
1865 struct kvm_memory_slot
*new)
1868 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1869 * an intermediate step. Instead, the old memslot is simply replaced
1870 * with a new, updated copy in both memslot sets.
1872 kvm_replace_memslot(kvm
, old
, new);
1873 kvm_activate_memslot(kvm
, old
, new);
1876 static int kvm_set_memslot(struct kvm
*kvm
,
1877 struct kvm_memory_slot
*old
,
1878 struct kvm_memory_slot
*new,
1879 enum kvm_mr_change change
)
1881 struct kvm_memory_slot
*invalid_slot
;
1885 * Released in kvm_swap_active_memslots().
1887 * Must be held from before the current memslots are copied until after
1888 * the new memslots are installed with rcu_assign_pointer, then
1889 * released before the synchronize srcu in kvm_swap_active_memslots().
1891 * When modifying memslots outside of the slots_lock, must be held
1892 * before reading the pointer to the current memslots until after all
1893 * changes to those memslots are complete.
1895 * These rules ensure that installing new memslots does not lose
1896 * changes made to the previous memslots.
1898 mutex_lock(&kvm
->slots_arch_lock
);
1901 * Invalidate the old slot if it's being deleted or moved. This is
1902 * done prior to actually deleting/moving the memslot to allow vCPUs to
1903 * continue running by ensuring there are no mappings or shadow pages
1904 * for the memslot when it is deleted/moved. Without pre-invalidation
1905 * (and without a lock), a window would exist between effecting the
1906 * delete/move and committing the changes in arch code where KVM or a
1907 * guest could access a non-existent memslot.
1909 * Modifications are done on a temporary, unreachable slot. The old
1910 * slot needs to be preserved in case a later step fails and the
1911 * invalidation needs to be reverted.
1913 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1914 invalid_slot
= kzalloc(sizeof(*invalid_slot
), GFP_KERNEL_ACCOUNT
);
1915 if (!invalid_slot
) {
1916 mutex_unlock(&kvm
->slots_arch_lock
);
1919 kvm_invalidate_memslot(kvm
, old
, invalid_slot
);
1922 r
= kvm_prepare_memory_region(kvm
, old
, new, change
);
1925 * For DELETE/MOVE, revert the above INVALID change. No
1926 * modifications required since the original slot was preserved
1927 * in the inactive slots. Changing the active memslots also
1928 * release slots_arch_lock.
1930 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1931 kvm_activate_memslot(kvm
, invalid_slot
, old
);
1932 kfree(invalid_slot
);
1934 mutex_unlock(&kvm
->slots_arch_lock
);
1940 * For DELETE and MOVE, the working slot is now active as the INVALID
1941 * version of the old slot. MOVE is particularly special as it reuses
1942 * the old slot and returns a copy of the old slot (in working_slot).
1943 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1944 * old slot is detached but otherwise preserved.
1946 if (change
== KVM_MR_CREATE
)
1947 kvm_create_memslot(kvm
, new);
1948 else if (change
== KVM_MR_DELETE
)
1949 kvm_delete_memslot(kvm
, old
, invalid_slot
);
1950 else if (change
== KVM_MR_MOVE
)
1951 kvm_move_memslot(kvm
, old
, new, invalid_slot
);
1952 else if (change
== KVM_MR_FLAGS_ONLY
)
1953 kvm_update_flags_memslot(kvm
, old
, new);
1957 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1958 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1959 kfree(invalid_slot
);
1962 * No need to refresh new->arch, changes after dropping slots_arch_lock
1963 * will directly hit the final, active memslot. Architectures are
1964 * responsible for knowing that new->arch may be stale.
1966 kvm_commit_memory_region(kvm
, old
, new, change
);
1971 static bool kvm_check_memslot_overlap(struct kvm_memslots
*slots
, int id
,
1972 gfn_t start
, gfn_t end
)
1974 struct kvm_memslot_iter iter
;
1976 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, start
, end
) {
1977 if (iter
.slot
->id
!= id
)
1985 * Allocate some memory and give it an address in the guest physical address
1988 * Discontiguous memory is allowed, mostly for framebuffers.
1990 * Must be called holding kvm->slots_lock for write.
1992 int __kvm_set_memory_region(struct kvm
*kvm
,
1993 const struct kvm_userspace_memory_region2
*mem
)
1995 struct kvm_memory_slot
*old
, *new;
1996 struct kvm_memslots
*slots
;
1997 enum kvm_mr_change change
;
1998 unsigned long npages
;
2003 r
= check_memory_region_flags(mem
);
2007 as_id
= mem
->slot
>> 16;
2008 id
= (u16
)mem
->slot
;
2010 /* General sanity checks */
2011 if ((mem
->memory_size
& (PAGE_SIZE
- 1)) ||
2012 (mem
->memory_size
!= (unsigned long)mem
->memory_size
))
2014 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
2016 /* We can read the guest memory with __xxx_user() later on. */
2017 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
2018 (mem
->userspace_addr
!= untagged_addr(mem
->userspace_addr
)) ||
2019 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
2022 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
2024 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
2026 if ((mem
->memory_size
>> PAGE_SHIFT
) > KVM_MEM_MAX_NR_PAGES
)
2029 slots
= __kvm_memslots(kvm
, as_id
);
2032 * Note, the old memslot (and the pointer itself!) may be invalidated
2033 * and/or destroyed by kvm_set_memslot().
2035 old
= id_to_memslot(slots
, id
);
2037 if (!mem
->memory_size
) {
2038 if (!old
|| !old
->npages
)
2041 if (WARN_ON_ONCE(kvm
->nr_memslot_pages
< old
->npages
))
2044 return kvm_set_memslot(kvm
, old
, NULL
, KVM_MR_DELETE
);
2047 base_gfn
= (mem
->guest_phys_addr
>> PAGE_SHIFT
);
2048 npages
= (mem
->memory_size
>> PAGE_SHIFT
);
2050 if (!old
|| !old
->npages
) {
2051 change
= KVM_MR_CREATE
;
2054 * To simplify KVM internals, the total number of pages across
2055 * all memslots must fit in an unsigned long.
2057 if ((kvm
->nr_memslot_pages
+ npages
) < kvm
->nr_memslot_pages
)
2059 } else { /* Modify an existing slot. */
2060 if ((mem
->userspace_addr
!= old
->userspace_addr
) ||
2061 (npages
!= old
->npages
) ||
2062 ((mem
->flags
^ old
->flags
) & KVM_MEM_READONLY
))
2065 if (base_gfn
!= old
->base_gfn
)
2066 change
= KVM_MR_MOVE
;
2067 else if (mem
->flags
!= old
->flags
)
2068 change
= KVM_MR_FLAGS_ONLY
;
2069 else /* Nothing to change. */
2073 if ((change
== KVM_MR_CREATE
|| change
== KVM_MR_MOVE
) &&
2074 kvm_check_memslot_overlap(slots
, id
, base_gfn
, base_gfn
+ npages
))
2077 /* Allocate a slot that will persist in the memslot. */
2078 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT
);
2084 new->base_gfn
= base_gfn
;
2085 new->npages
= npages
;
2086 new->flags
= mem
->flags
;
2087 new->userspace_addr
= mem
->userspace_addr
;
2089 r
= kvm_set_memslot(kvm
, old
, new, change
);
2094 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
2096 int kvm_set_memory_region(struct kvm
*kvm
,
2097 const struct kvm_userspace_memory_region2
*mem
)
2101 mutex_lock(&kvm
->slots_lock
);
2102 r
= __kvm_set_memory_region(kvm
, mem
);
2103 mutex_unlock(&kvm
->slots_lock
);
2106 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
2108 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
2109 struct kvm_userspace_memory_region2
*mem
)
2111 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
2114 return kvm_set_memory_region(kvm
, mem
);
2117 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2119 * kvm_get_dirty_log - get a snapshot of dirty pages
2120 * @kvm: pointer to kvm instance
2121 * @log: slot id and address to which we copy the log
2122 * @is_dirty: set to '1' if any dirty pages were found
2123 * @memslot: set to the associated memslot, always valid on success
2125 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
2126 int *is_dirty
, struct kvm_memory_slot
**memslot
)
2128 struct kvm_memslots
*slots
;
2131 unsigned long any
= 0;
2133 /* Dirty ring tracking may be exclusive to dirty log tracking */
2134 if (!kvm_use_dirty_bitmap(kvm
))
2140 as_id
= log
->slot
>> 16;
2141 id
= (u16
)log
->slot
;
2142 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2145 slots
= __kvm_memslots(kvm
, as_id
);
2146 *memslot
= id_to_memslot(slots
, id
);
2147 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
2150 kvm_arch_sync_dirty_log(kvm
, *memslot
);
2152 n
= kvm_dirty_bitmap_bytes(*memslot
);
2154 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
2155 any
= (*memslot
)->dirty_bitmap
[i
];
2157 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
2164 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
2166 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2168 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2169 * and reenable dirty page tracking for the corresponding pages.
2170 * @kvm: pointer to kvm instance
2171 * @log: slot id and address to which we copy the log
2173 * We need to keep it in mind that VCPU threads can write to the bitmap
2174 * concurrently. So, to avoid losing track of dirty pages we keep the
2177 * 1. Take a snapshot of the bit and clear it if needed.
2178 * 2. Write protect the corresponding page.
2179 * 3. Copy the snapshot to the userspace.
2180 * 4. Upon return caller flushes TLB's if needed.
2182 * Between 2 and 4, the guest may write to the page using the remaining TLB
2183 * entry. This is not a problem because the page is reported dirty using
2184 * the snapshot taken before and step 4 ensures that writes done after
2185 * exiting to userspace will be logged for the next call.
2188 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
2190 struct kvm_memslots
*slots
;
2191 struct kvm_memory_slot
*memslot
;
2194 unsigned long *dirty_bitmap
;
2195 unsigned long *dirty_bitmap_buffer
;
2198 /* Dirty ring tracking may be exclusive to dirty log tracking */
2199 if (!kvm_use_dirty_bitmap(kvm
))
2202 as_id
= log
->slot
>> 16;
2203 id
= (u16
)log
->slot
;
2204 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2207 slots
= __kvm_memslots(kvm
, as_id
);
2208 memslot
= id_to_memslot(slots
, id
);
2209 if (!memslot
|| !memslot
->dirty_bitmap
)
2212 dirty_bitmap
= memslot
->dirty_bitmap
;
2214 kvm_arch_sync_dirty_log(kvm
, memslot
);
2216 n
= kvm_dirty_bitmap_bytes(memslot
);
2218 if (kvm
->manual_dirty_log_protect
) {
2220 * Unlike kvm_get_dirty_log, we always return false in *flush,
2221 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2222 * is some code duplication between this function and
2223 * kvm_get_dirty_log, but hopefully all architecture
2224 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2225 * can be eliminated.
2227 dirty_bitmap_buffer
= dirty_bitmap
;
2229 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2230 memset(dirty_bitmap_buffer
, 0, n
);
2233 for (i
= 0; i
< n
/ sizeof(long); i
++) {
2237 if (!dirty_bitmap
[i
])
2241 mask
= xchg(&dirty_bitmap
[i
], 0);
2242 dirty_bitmap_buffer
[i
] = mask
;
2244 offset
= i
* BITS_PER_LONG
;
2245 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2248 KVM_MMU_UNLOCK(kvm
);
2252 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2254 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
2261 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2262 * @kvm: kvm instance
2263 * @log: slot id and address to which we copy the log
2265 * Steps 1-4 below provide general overview of dirty page logging. See
2266 * kvm_get_dirty_log_protect() function description for additional details.
2268 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2269 * always flush the TLB (step 4) even if previous step failed and the dirty
2270 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2271 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2272 * writes will be marked dirty for next log read.
2274 * 1. Take a snapshot of the bit and clear it if needed.
2275 * 2. Write protect the corresponding page.
2276 * 3. Copy the snapshot to the userspace.
2277 * 4. Flush TLB's if needed.
2279 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2280 struct kvm_dirty_log
*log
)
2284 mutex_lock(&kvm
->slots_lock
);
2286 r
= kvm_get_dirty_log_protect(kvm
, log
);
2288 mutex_unlock(&kvm
->slots_lock
);
2293 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2294 * and reenable dirty page tracking for the corresponding pages.
2295 * @kvm: pointer to kvm instance
2296 * @log: slot id and address from which to fetch the bitmap of dirty pages
2298 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
2299 struct kvm_clear_dirty_log
*log
)
2301 struct kvm_memslots
*slots
;
2302 struct kvm_memory_slot
*memslot
;
2306 unsigned long *dirty_bitmap
;
2307 unsigned long *dirty_bitmap_buffer
;
2310 /* Dirty ring tracking may be exclusive to dirty log tracking */
2311 if (!kvm_use_dirty_bitmap(kvm
))
2314 as_id
= log
->slot
>> 16;
2315 id
= (u16
)log
->slot
;
2316 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2319 if (log
->first_page
& 63)
2322 slots
= __kvm_memslots(kvm
, as_id
);
2323 memslot
= id_to_memslot(slots
, id
);
2324 if (!memslot
|| !memslot
->dirty_bitmap
)
2327 dirty_bitmap
= memslot
->dirty_bitmap
;
2329 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
2331 if (log
->first_page
> memslot
->npages
||
2332 log
->num_pages
> memslot
->npages
- log
->first_page
||
2333 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
2336 kvm_arch_sync_dirty_log(kvm
, memslot
);
2339 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2340 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
2344 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
2345 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
2346 i
++, offset
+= BITS_PER_LONG
) {
2347 unsigned long mask
= *dirty_bitmap_buffer
++;
2348 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
2352 mask
&= atomic_long_fetch_andnot(mask
, p
);
2355 * mask contains the bits that really have been cleared. This
2356 * never includes any bits beyond the length of the memslot (if
2357 * the length is not aligned to 64 pages), therefore it is not
2358 * a problem if userspace sets them in log->dirty_bitmap.
2362 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2366 KVM_MMU_UNLOCK(kvm
);
2369 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2374 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
2375 struct kvm_clear_dirty_log
*log
)
2379 mutex_lock(&kvm
->slots_lock
);
2381 r
= kvm_clear_dirty_log_protect(kvm
, log
);
2383 mutex_unlock(&kvm
->slots_lock
);
2386 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2388 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
2390 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
2392 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
2394 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2396 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2397 u64 gen
= slots
->generation
;
2398 struct kvm_memory_slot
*slot
;
2401 * This also protects against using a memslot from a different address space,
2402 * since different address spaces have different generation numbers.
2404 if (unlikely(gen
!= vcpu
->last_used_slot_gen
)) {
2405 vcpu
->last_used_slot
= NULL
;
2406 vcpu
->last_used_slot_gen
= gen
;
2409 slot
= try_get_memslot(vcpu
->last_used_slot
, gfn
);
2414 * Fall back to searching all memslots. We purposely use
2415 * search_memslots() instead of __gfn_to_memslot() to avoid
2416 * thrashing the VM-wide last_used_slot in kvm_memslots.
2418 slot
= search_memslots(slots
, gfn
, false);
2420 vcpu
->last_used_slot
= slot
;
2427 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
2429 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
2431 return kvm_is_visible_memslot(memslot
);
2433 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
2435 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2437 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2439 return kvm_is_visible_memslot(memslot
);
2441 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
2443 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2445 struct vm_area_struct
*vma
;
2446 unsigned long addr
, size
;
2450 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
2451 if (kvm_is_error_hva(addr
))
2454 mmap_read_lock(current
->mm
);
2455 vma
= find_vma(current
->mm
, addr
);
2459 size
= vma_kernel_pagesize(vma
);
2462 mmap_read_unlock(current
->mm
);
2467 static bool memslot_is_readonly(const struct kvm_memory_slot
*slot
)
2469 return slot
->flags
& KVM_MEM_READONLY
;
2472 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2473 gfn_t
*nr_pages
, bool write
)
2475 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
2476 return KVM_HVA_ERR_BAD
;
2478 if (memslot_is_readonly(slot
) && write
)
2479 return KVM_HVA_ERR_RO_BAD
;
2482 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
2484 return __gfn_to_hva_memslot(slot
, gfn
);
2487 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2490 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
2493 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
2496 return gfn_to_hva_many(slot
, gfn
, NULL
);
2498 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
2500 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
2502 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
2504 EXPORT_SYMBOL_GPL(gfn_to_hva
);
2506 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2508 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
2510 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
2513 * Return the hva of a @gfn and the R/W attribute if possible.
2515 * @slot: the kvm_memory_slot which contains @gfn
2516 * @gfn: the gfn to be translated
2517 * @writable: used to return the read/write attribute of the @slot if the hva
2518 * is valid and @writable is not NULL
2520 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
2521 gfn_t gfn
, bool *writable
)
2523 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
2525 if (!kvm_is_error_hva(hva
) && writable
)
2526 *writable
= !memslot_is_readonly(slot
);
2531 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
2533 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2535 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2538 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
2540 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2542 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2545 static inline int check_user_page_hwpoison(unsigned long addr
)
2547 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
2549 rc
= get_user_pages(addr
, 1, flags
, NULL
);
2550 return rc
== -EHWPOISON
;
2554 * The fast path to get the writable pfn which will be stored in @pfn,
2555 * true indicates success, otherwise false is returned. It's also the
2556 * only part that runs if we can in atomic context.
2558 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
2559 bool *writable
, kvm_pfn_t
*pfn
)
2561 struct page
*page
[1];
2564 * Fast pin a writable pfn only if it is a write fault request
2565 * or the caller allows to map a writable pfn for a read fault
2568 if (!(write_fault
|| writable
))
2571 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
2572 *pfn
= page_to_pfn(page
[0]);
2583 * The slow path to get the pfn of the specified host virtual address,
2584 * 1 indicates success, -errno is returned if error is detected.
2586 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
2587 bool interruptible
, bool *writable
, kvm_pfn_t
*pfn
)
2590 * When a VCPU accesses a page that is not mapped into the secondary
2591 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2592 * make progress. We always want to honor NUMA hinting faults in that
2593 * case, because GUP usage corresponds to memory accesses from the VCPU.
2594 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2595 * mapped into the secondary MMU and gets accessed by a VCPU.
2597 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2598 * implicitly honor NUMA hinting faults and don't need this flag.
2600 unsigned int flags
= FOLL_HWPOISON
| FOLL_HONOR_NUMA_FAULT
;
2607 *writable
= write_fault
;
2610 flags
|= FOLL_WRITE
;
2612 flags
|= FOLL_NOWAIT
;
2614 flags
|= FOLL_INTERRUPTIBLE
;
2616 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
2620 /* map read fault as writable if possible */
2621 if (unlikely(!write_fault
) && writable
) {
2624 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
2630 *pfn
= page_to_pfn(page
);
2634 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
2636 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
2639 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
2645 static int kvm_try_get_pfn(kvm_pfn_t pfn
)
2647 struct page
*page
= kvm_pfn_to_refcounted_page(pfn
);
2652 return get_page_unless_zero(page
);
2655 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
2656 unsigned long addr
, bool write_fault
,
2657 bool *writable
, kvm_pfn_t
*p_pfn
)
2665 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2668 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2669 * not call the fault handler, so do it here.
2671 bool unlocked
= false;
2672 r
= fixup_user_fault(current
->mm
, addr
,
2673 (write_fault
? FAULT_FLAG_WRITE
: 0),
2680 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2685 pte
= ptep_get(ptep
);
2687 if (write_fault
&& !pte_write(pte
)) {
2688 pfn
= KVM_PFN_ERR_RO_FAULT
;
2693 *writable
= pte_write(pte
);
2697 * Get a reference here because callers of *hva_to_pfn* and
2698 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2699 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2700 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2701 * simply do nothing for reserved pfns.
2703 * Whoever called remap_pfn_range is also going to call e.g.
2704 * unmap_mapping_range before the underlying pages are freed,
2705 * causing a call to our MMU notifier.
2707 * Certain IO or PFNMAP mappings can be backed with valid
2708 * struct pages, but be allocated without refcounting e.g.,
2709 * tail pages of non-compound higher order allocations, which
2710 * would then underflow the refcount when the caller does the
2711 * required put_page. Don't allow those pages here.
2713 if (!kvm_try_get_pfn(pfn
))
2717 pte_unmap_unlock(ptep
, ptl
);
2724 * Pin guest page in memory and return its pfn.
2725 * @addr: host virtual address which maps memory to the guest
2726 * @atomic: whether this function can sleep
2727 * @interruptible: whether the process can be interrupted by non-fatal signals
2728 * @async: whether this function need to wait IO complete if the
2729 * host page is not in the memory
2730 * @write_fault: whether we should get a writable host page
2731 * @writable: whether it allows to map a writable host page for !@write_fault
2733 * The function will map a writable host page for these two cases:
2734 * 1): @write_fault = true
2735 * 2): @write_fault = false && @writable, @writable will tell the caller
2736 * whether the mapping is writable.
2738 kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool interruptible
,
2739 bool *async
, bool write_fault
, bool *writable
)
2741 struct vm_area_struct
*vma
;
2745 /* we can do it either atomically or asynchronously, not both */
2746 BUG_ON(atomic
&& async
);
2748 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
2752 return KVM_PFN_ERR_FAULT
;
2754 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, interruptible
,
2758 if (npages
== -EINTR
)
2759 return KVM_PFN_ERR_SIGPENDING
;
2761 mmap_read_lock(current
->mm
);
2762 if (npages
== -EHWPOISON
||
2763 (!async
&& check_user_page_hwpoison(addr
))) {
2764 pfn
= KVM_PFN_ERR_HWPOISON
;
2769 vma
= vma_lookup(current
->mm
, addr
);
2772 pfn
= KVM_PFN_ERR_FAULT
;
2773 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
2774 r
= hva_to_pfn_remapped(vma
, addr
, write_fault
, writable
, &pfn
);
2778 pfn
= KVM_PFN_ERR_FAULT
;
2780 if (async
&& vma_is_valid(vma
, write_fault
))
2782 pfn
= KVM_PFN_ERR_FAULT
;
2785 mmap_read_unlock(current
->mm
);
2789 kvm_pfn_t
__gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2790 bool atomic
, bool interruptible
, bool *async
,
2791 bool write_fault
, bool *writable
, hva_t
*hva
)
2793 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
2798 if (addr
== KVM_HVA_ERR_RO_BAD
) {
2801 return KVM_PFN_ERR_RO_FAULT
;
2804 if (kvm_is_error_hva(addr
)) {
2807 return KVM_PFN_NOSLOT
;
2810 /* Do not map writable pfn in the readonly memslot. */
2811 if (writable
&& memslot_is_readonly(slot
)) {
2816 return hva_to_pfn(addr
, atomic
, interruptible
, async
, write_fault
,
2819 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
2821 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
2824 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, false,
2825 NULL
, write_fault
, writable
, NULL
);
2827 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
2829 kvm_pfn_t
gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
2831 return __gfn_to_pfn_memslot(slot
, gfn
, false, false, NULL
, true,
2834 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
2836 kvm_pfn_t
gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
2838 return __gfn_to_pfn_memslot(slot
, gfn
, true, false, NULL
, true,
2841 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
2843 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2845 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2847 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
2849 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
2851 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
2853 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
2855 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2857 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2859 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
2861 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2862 struct page
**pages
, int nr_pages
)
2867 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
2868 if (kvm_is_error_hva(addr
))
2871 if (entry
< nr_pages
)
2874 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
2876 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
2879 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2880 * backed by 'struct page'. A valid example is if the backing memslot is
2881 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2882 * been elevated by gfn_to_pfn().
2884 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
2889 pfn
= gfn_to_pfn(kvm
, gfn
);
2891 if (is_error_noslot_pfn(pfn
))
2892 return KVM_ERR_PTR_BAD_PAGE
;
2894 page
= kvm_pfn_to_refcounted_page(pfn
);
2896 return KVM_ERR_PTR_BAD_PAGE
;
2900 EXPORT_SYMBOL_GPL(gfn_to_page
);
2902 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
)
2905 kvm_release_pfn_dirty(pfn
);
2907 kvm_release_pfn_clean(pfn
);
2910 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
2914 struct page
*page
= KVM_UNMAPPED_PAGE
;
2919 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2920 if (is_error_noslot_pfn(pfn
))
2923 if (pfn_valid(pfn
)) {
2924 page
= pfn_to_page(pfn
);
2926 #ifdef CONFIG_HAS_IOMEM
2928 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
2942 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
2944 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
2952 if (map
->page
!= KVM_UNMAPPED_PAGE
)
2954 #ifdef CONFIG_HAS_IOMEM
2960 kvm_vcpu_mark_page_dirty(vcpu
, map
->gfn
);
2962 kvm_release_pfn(map
->pfn
, dirty
);
2967 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
2969 static bool kvm_is_ad_tracked_page(struct page
*page
)
2972 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2973 * touched (e.g. set dirty) except by its owner".
2975 return !PageReserved(page
);
2978 static void kvm_set_page_dirty(struct page
*page
)
2980 if (kvm_is_ad_tracked_page(page
))
2984 static void kvm_set_page_accessed(struct page
*page
)
2986 if (kvm_is_ad_tracked_page(page
))
2987 mark_page_accessed(page
);
2990 void kvm_release_page_clean(struct page
*page
)
2992 WARN_ON(is_error_page(page
));
2994 kvm_set_page_accessed(page
);
2997 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
2999 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
3003 if (is_error_noslot_pfn(pfn
))
3006 page
= kvm_pfn_to_refcounted_page(pfn
);
3010 kvm_release_page_clean(page
);
3012 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
3014 void kvm_release_page_dirty(struct page
*page
)
3016 WARN_ON(is_error_page(page
));
3018 kvm_set_page_dirty(page
);
3019 kvm_release_page_clean(page
);
3021 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
3023 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
3027 if (is_error_noslot_pfn(pfn
))
3030 page
= kvm_pfn_to_refcounted_page(pfn
);
3034 kvm_release_page_dirty(page
);
3036 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
3039 * Note, checking for an error/noslot pfn is the caller's responsibility when
3040 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
3041 * "set" helpers are not to be used when the pfn might point at garbage.
3043 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
3045 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3049 kvm_set_page_dirty(pfn_to_page(pfn
));
3051 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
3053 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
3055 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3059 kvm_set_page_accessed(pfn_to_page(pfn
));
3061 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
3063 static int next_segment(unsigned long len
, int offset
)
3065 if (len
> PAGE_SIZE
- offset
)
3066 return PAGE_SIZE
- offset
;
3071 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3072 void *data
, int offset
, int len
)
3077 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3078 if (kvm_is_error_hva(addr
))
3080 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
3086 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
3089 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3091 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3093 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
3095 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
3096 int offset
, int len
)
3098 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3100 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3102 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
3104 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
3106 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3108 int offset
= offset_in_page(gpa
);
3111 while ((seg
= next_segment(len
, offset
)) != 0) {
3112 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
3122 EXPORT_SYMBOL_GPL(kvm_read_guest
);
3124 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
3126 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3128 int offset
= offset_in_page(gpa
);
3131 while ((seg
= next_segment(len
, offset
)) != 0) {
3132 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3142 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
3144 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3145 void *data
, int offset
, unsigned long len
)
3150 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3151 if (kvm_is_error_hva(addr
))
3153 pagefault_disable();
3154 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
3161 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3162 void *data
, unsigned long len
)
3164 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3165 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3166 int offset
= offset_in_page(gpa
);
3168 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
3170 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
3172 static int __kvm_write_guest_page(struct kvm
*kvm
,
3173 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
3174 const void *data
, int offset
, int len
)
3179 addr
= gfn_to_hva_memslot(memslot
, gfn
);
3180 if (kvm_is_error_hva(addr
))
3182 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
3185 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3189 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
3190 const void *data
, int offset
, int len
)
3192 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3194 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
3196 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
3198 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
3199 const void *data
, int offset
, int len
)
3201 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3203 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
3205 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
3207 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
3210 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3212 int offset
= offset_in_page(gpa
);
3215 while ((seg
= next_segment(len
, offset
)) != 0) {
3216 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
3226 EXPORT_SYMBOL_GPL(kvm_write_guest
);
3228 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
3231 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3233 int offset
= offset_in_page(gpa
);
3236 while ((seg
= next_segment(len
, offset
)) != 0) {
3237 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3247 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
3249 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
3250 struct gfn_to_hva_cache
*ghc
,
3251 gpa_t gpa
, unsigned long len
)
3253 int offset
= offset_in_page(gpa
);
3254 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
3255 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
3256 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
3257 gfn_t nr_pages_avail
;
3259 /* Update ghc->generation before performing any error checks. */
3260 ghc
->generation
= slots
->generation
;
3262 if (start_gfn
> end_gfn
) {
3263 ghc
->hva
= KVM_HVA_ERR_BAD
;
3268 * If the requested region crosses two memslots, we still
3269 * verify that the entire region is valid here.
3271 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
3272 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
3273 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
3275 if (kvm_is_error_hva(ghc
->hva
))
3279 /* Use the slow path for cross page reads and writes. */
3280 if (nr_pages_needed
== 1)
3283 ghc
->memslot
= NULL
;
3290 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3291 gpa_t gpa
, unsigned long len
)
3293 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3294 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
3296 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
3298 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3299 void *data
, unsigned int offset
,
3302 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3304 gpa_t gpa
= ghc
->gpa
+ offset
;
3306 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3309 if (slots
->generation
!= ghc
->generation
) {
3310 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3314 if (kvm_is_error_hva(ghc
->hva
))
3317 if (unlikely(!ghc
->memslot
))
3318 return kvm_write_guest(kvm
, gpa
, data
, len
);
3320 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
3323 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
3327 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
3329 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3330 void *data
, unsigned long len
)
3332 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3334 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
3336 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3337 void *data
, unsigned int offset
,
3340 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3342 gpa_t gpa
= ghc
->gpa
+ offset
;
3344 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3347 if (slots
->generation
!= ghc
->generation
) {
3348 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3352 if (kvm_is_error_hva(ghc
->hva
))
3355 if (unlikely(!ghc
->memslot
))
3356 return kvm_read_guest(kvm
, gpa
, data
, len
);
3358 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
3364 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
3366 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3367 void *data
, unsigned long len
)
3369 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3371 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
3373 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
3375 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3376 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3378 int offset
= offset_in_page(gpa
);
3381 while ((seg
= next_segment(len
, offset
)) != 0) {
3382 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
3391 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
3393 void mark_page_dirty_in_slot(struct kvm
*kvm
,
3394 const struct kvm_memory_slot
*memslot
,
3397 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
3399 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3400 if (WARN_ON_ONCE(vcpu
&& vcpu
->kvm
!= kvm
))
3403 WARN_ON_ONCE(!vcpu
&& !kvm_arch_allow_write_without_running_vcpu(kvm
));
3406 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
3407 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
3408 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
3410 if (kvm
->dirty_ring_size
&& vcpu
)
3411 kvm_dirty_ring_push(vcpu
, slot
, rel_gfn
);
3412 else if (memslot
->dirty_bitmap
)
3413 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
3416 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
3418 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
3420 struct kvm_memory_slot
*memslot
;
3422 memslot
= gfn_to_memslot(kvm
, gfn
);
3423 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3425 EXPORT_SYMBOL_GPL(mark_page_dirty
);
3427 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3429 struct kvm_memory_slot
*memslot
;
3431 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3432 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
3434 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
3436 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
3438 if (!vcpu
->sigset_active
)
3442 * This does a lockless modification of ->real_blocked, which is fine
3443 * because, only current can change ->real_blocked and all readers of
3444 * ->real_blocked don't care as long ->real_blocked is always a subset
3447 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
3450 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
3452 if (!vcpu
->sigset_active
)
3455 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
3456 sigemptyset(¤t
->real_blocked
);
3459 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3461 unsigned int old
, val
, grow
, grow_start
;
3463 old
= val
= vcpu
->halt_poll_ns
;
3464 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3465 grow
= READ_ONCE(halt_poll_ns_grow
);
3470 if (val
< grow_start
)
3473 vcpu
->halt_poll_ns
= val
;
3475 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
3478 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3480 unsigned int old
, val
, shrink
, grow_start
;
3482 old
= val
= vcpu
->halt_poll_ns
;
3483 shrink
= READ_ONCE(halt_poll_ns_shrink
);
3484 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3490 if (val
< grow_start
)
3493 vcpu
->halt_poll_ns
= val
;
3494 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
3497 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
3500 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3502 if (kvm_arch_vcpu_runnable(vcpu
))
3504 if (kvm_cpu_has_pending_timer(vcpu
))
3506 if (signal_pending(current
))
3508 if (kvm_check_request(KVM_REQ_UNBLOCK
, vcpu
))
3513 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3518 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3519 * pending. This is mostly used when halting a vCPU, but may also be used
3520 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3522 bool kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
3524 struct rcuwait
*wait
= kvm_arch_vcpu_get_wait(vcpu
);
3525 bool waited
= false;
3527 vcpu
->stat
.generic
.blocking
= 1;
3530 kvm_arch_vcpu_blocking(vcpu
);
3531 prepare_to_rcuwait(wait
);
3535 set_current_state(TASK_INTERRUPTIBLE
);
3537 if (kvm_vcpu_check_block(vcpu
) < 0)
3545 finish_rcuwait(wait
);
3546 kvm_arch_vcpu_unblocking(vcpu
);
3549 vcpu
->stat
.generic
.blocking
= 0;
3554 static inline void update_halt_poll_stats(struct kvm_vcpu
*vcpu
, ktime_t start
,
3555 ktime_t end
, bool success
)
3557 struct kvm_vcpu_stat_generic
*stats
= &vcpu
->stat
.generic
;
3558 u64 poll_ns
= ktime_to_ns(ktime_sub(end
, start
));
3560 ++vcpu
->stat
.generic
.halt_attempted_poll
;
3563 ++vcpu
->stat
.generic
.halt_successful_poll
;
3565 if (!vcpu_valid_wakeup(vcpu
))
3566 ++vcpu
->stat
.generic
.halt_poll_invalid
;
3568 stats
->halt_poll_success_ns
+= poll_ns
;
3569 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_success_hist
, poll_ns
);
3571 stats
->halt_poll_fail_ns
+= poll_ns
;
3572 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_fail_hist
, poll_ns
);
3576 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3578 struct kvm
*kvm
= vcpu
->kvm
;
3580 if (kvm
->override_halt_poll_ns
) {
3582 * Ensure kvm->max_halt_poll_ns is not read before
3583 * kvm->override_halt_poll_ns.
3585 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3588 return READ_ONCE(kvm
->max_halt_poll_ns
);
3591 return READ_ONCE(halt_poll_ns
);
3595 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3596 * polling is enabled, busy wait for a short time before blocking to avoid the
3597 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3600 void kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
3602 unsigned int max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3603 bool halt_poll_allowed
= !kvm_arch_no_poll(vcpu
);
3604 ktime_t start
, cur
, poll_end
;
3605 bool waited
= false;
3609 if (vcpu
->halt_poll_ns
> max_halt_poll_ns
)
3610 vcpu
->halt_poll_ns
= max_halt_poll_ns
;
3612 do_halt_poll
= halt_poll_allowed
&& vcpu
->halt_poll_ns
;
3614 start
= cur
= poll_end
= ktime_get();
3616 ktime_t stop
= ktime_add_ns(start
, vcpu
->halt_poll_ns
);
3619 if (kvm_vcpu_check_block(vcpu
) < 0)
3622 poll_end
= cur
= ktime_get();
3623 } while (kvm_vcpu_can_poll(cur
, stop
));
3626 waited
= kvm_vcpu_block(vcpu
);
3630 vcpu
->stat
.generic
.halt_wait_ns
+=
3631 ktime_to_ns(cur
) - ktime_to_ns(poll_end
);
3632 KVM_STATS_LOG_HIST_UPDATE(vcpu
->stat
.generic
.halt_wait_hist
,
3633 ktime_to_ns(cur
) - ktime_to_ns(poll_end
));
3636 /* The total time the vCPU was "halted", including polling time. */
3637 halt_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
3640 * Note, halt-polling is considered successful so long as the vCPU was
3641 * never actually scheduled out, i.e. even if the wake event arrived
3642 * after of the halt-polling loop itself, but before the full wait.
3645 update_halt_poll_stats(vcpu
, start
, poll_end
, !waited
);
3647 if (halt_poll_allowed
) {
3648 /* Recompute the max halt poll time in case it changed. */
3649 max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3651 if (!vcpu_valid_wakeup(vcpu
)) {
3652 shrink_halt_poll_ns(vcpu
);
3653 } else if (max_halt_poll_ns
) {
3654 if (halt_ns
<= vcpu
->halt_poll_ns
)
3656 /* we had a long block, shrink polling */
3657 else if (vcpu
->halt_poll_ns
&&
3658 halt_ns
> max_halt_poll_ns
)
3659 shrink_halt_poll_ns(vcpu
);
3660 /* we had a short halt and our poll time is too small */
3661 else if (vcpu
->halt_poll_ns
< max_halt_poll_ns
&&
3662 halt_ns
< max_halt_poll_ns
)
3663 grow_halt_poll_ns(vcpu
);
3665 vcpu
->halt_poll_ns
= 0;
3669 trace_kvm_vcpu_wakeup(halt_ns
, waited
, vcpu_valid_wakeup(vcpu
));
3671 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
3673 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
3675 if (__kvm_vcpu_wake_up(vcpu
)) {
3676 WRITE_ONCE(vcpu
->ready
, true);
3677 ++vcpu
->stat
.generic
.halt_wakeup
;
3683 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
3687 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3689 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3693 if (kvm_vcpu_wake_up(vcpu
))
3698 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3699 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3700 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3701 * within the vCPU thread itself.
3703 if (vcpu
== __this_cpu_read(kvm_running_vcpu
)) {
3704 if (vcpu
->mode
== IN_GUEST_MODE
)
3705 WRITE_ONCE(vcpu
->mode
, EXITING_GUEST_MODE
);
3710 * Note, the vCPU could get migrated to a different pCPU at any point
3711 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3712 * IPI to the previous pCPU. But, that's ok because the purpose of the
3713 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3714 * vCPU also requires it to leave IN_GUEST_MODE.
3716 if (kvm_arch_vcpu_should_kick(vcpu
)) {
3717 cpu
= READ_ONCE(vcpu
->cpu
);
3718 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
3719 smp_send_reschedule(cpu
);
3724 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
3725 #endif /* !CONFIG_S390 */
3727 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
3730 struct task_struct
*task
= NULL
;
3734 pid
= rcu_dereference(target
->pid
);
3736 task
= get_pid_task(pid
, PIDTYPE_PID
);
3740 ret
= yield_to(task
, 1);
3741 put_task_struct(task
);
3745 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
3748 * Helper that checks whether a VCPU is eligible for directed yield.
3749 * Most eligible candidate to yield is decided by following heuristics:
3751 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3752 * (preempted lock holder), indicated by @in_spin_loop.
3753 * Set at the beginning and cleared at the end of interception/PLE handler.
3755 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3756 * chance last time (mostly it has become eligible now since we have probably
3757 * yielded to lockholder in last iteration. This is done by toggling
3758 * @dy_eligible each time a VCPU checked for eligibility.)
3760 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3761 * to preempted lock-holder could result in wrong VCPU selection and CPU
3762 * burning. Giving priority for a potential lock-holder increases lock
3765 * Since algorithm is based on heuristics, accessing another VCPU data without
3766 * locking does not harm. It may result in trying to yield to same VCPU, fail
3767 * and continue with next VCPU and so on.
3769 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
3771 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3774 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
3775 vcpu
->spin_loop
.dy_eligible
;
3777 if (vcpu
->spin_loop
.in_spin_loop
)
3778 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
3787 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3788 * a vcpu_load/vcpu_put pair. However, for most architectures
3789 * kvm_arch_vcpu_runnable does not require vcpu_load.
3791 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
3793 return kvm_arch_vcpu_runnable(vcpu
);
3796 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
3798 if (kvm_arch_dy_runnable(vcpu
))
3801 #ifdef CONFIG_KVM_ASYNC_PF
3802 if (!list_empty_careful(&vcpu
->async_pf
.done
))
3809 bool __weak
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu
*vcpu
)
3814 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
3816 struct kvm
*kvm
= me
->kvm
;
3817 struct kvm_vcpu
*vcpu
;
3818 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
3824 kvm_vcpu_set_in_spin_loop(me
, true);
3826 * We boost the priority of a VCPU that is runnable but not
3827 * currently running, because it got preempted by something
3828 * else and called schedule in __vcpu_run. Hopefully that
3829 * VCPU is holding the lock that we need and will release it.
3830 * We approximate round-robin by starting at the last boosted VCPU.
3832 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
3833 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3834 if (!pass
&& i
<= last_boosted_vcpu
) {
3835 i
= last_boosted_vcpu
;
3837 } else if (pass
&& i
> last_boosted_vcpu
)
3839 if (!READ_ONCE(vcpu
->ready
))
3843 if (kvm_vcpu_is_blocking(vcpu
) && !vcpu_dy_runnable(vcpu
))
3845 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
3846 !kvm_arch_dy_has_pending_interrupt(vcpu
) &&
3847 !kvm_arch_vcpu_in_kernel(vcpu
))
3849 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
3852 yielded
= kvm_vcpu_yield_to(vcpu
);
3854 kvm
->last_boosted_vcpu
= i
;
3856 } else if (yielded
< 0) {
3863 kvm_vcpu_set_in_spin_loop(me
, false);
3865 /* Ensure vcpu is not eligible during next spinloop */
3866 kvm_vcpu_set_dy_eligible(me
, false);
3868 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
3870 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
3872 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3873 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
3874 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
3875 kvm
->dirty_ring_size
/ PAGE_SIZE
);
3881 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
3883 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
3886 if (vmf
->pgoff
== 0)
3887 page
= virt_to_page(vcpu
->run
);
3889 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
3890 page
= virt_to_page(vcpu
->arch
.pio_data
);
3892 #ifdef CONFIG_KVM_MMIO
3893 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
3894 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
3896 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
3897 page
= kvm_dirty_ring_get_page(
3899 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
3901 return kvm_arch_vcpu_fault(vcpu
, vmf
);
3907 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
3908 .fault
= kvm_vcpu_fault
,
3911 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3913 struct kvm_vcpu
*vcpu
= file
->private_data
;
3914 unsigned long pages
= vma_pages(vma
);
3916 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
3917 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
3918 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
3921 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
3925 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
3927 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3929 kvm_put_kvm(vcpu
->kvm
);
3933 static const struct file_operations kvm_vcpu_fops
= {
3934 .release
= kvm_vcpu_release
,
3935 .unlocked_ioctl
= kvm_vcpu_ioctl
,
3936 .mmap
= kvm_vcpu_mmap
,
3937 .llseek
= noop_llseek
,
3938 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
3942 * Allocates an inode for the vcpu.
3944 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
3946 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
3948 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
3949 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
3952 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3953 static int vcpu_get_pid(void *data
, u64
*val
)
3955 struct kvm_vcpu
*vcpu
= data
;
3958 *val
= pid_nr(rcu_dereference(vcpu
->pid
));
3963 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops
, vcpu_get_pid
, NULL
, "%llu\n");
3965 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
3967 struct dentry
*debugfs_dentry
;
3968 char dir_name
[ITOA_MAX_LEN
* 2];
3970 if (!debugfs_initialized())
3973 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
3974 debugfs_dentry
= debugfs_create_dir(dir_name
,
3975 vcpu
->kvm
->debugfs_dentry
);
3976 debugfs_create_file("pid", 0444, debugfs_dentry
, vcpu
,
3977 &vcpu_get_pid_fops
);
3979 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
3984 * Creates some virtual cpus. Good luck creating more than one.
3986 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
3989 struct kvm_vcpu
*vcpu
;
3992 if (id
>= KVM_MAX_VCPU_IDS
)
3995 mutex_lock(&kvm
->lock
);
3996 if (kvm
->created_vcpus
>= kvm
->max_vcpus
) {
3997 mutex_unlock(&kvm
->lock
);
4001 r
= kvm_arch_vcpu_precreate(kvm
, id
);
4003 mutex_unlock(&kvm
->lock
);
4007 kvm
->created_vcpus
++;
4008 mutex_unlock(&kvm
->lock
);
4010 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL_ACCOUNT
);
4013 goto vcpu_decrement
;
4016 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
4017 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
4022 vcpu
->run
= page_address(page
);
4024 kvm_vcpu_init(vcpu
, kvm
, id
);
4026 r
= kvm_arch_vcpu_create(vcpu
);
4028 goto vcpu_free_run_page
;
4030 if (kvm
->dirty_ring_size
) {
4031 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
4032 id
, kvm
->dirty_ring_size
);
4034 goto arch_vcpu_destroy
;
4037 mutex_lock(&kvm
->lock
);
4039 #ifdef CONFIG_LOCKDEP
4040 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
4041 mutex_lock(&vcpu
->mutex
);
4042 mutex_unlock(&vcpu
->mutex
);
4045 if (kvm_get_vcpu_by_id(kvm
, id
)) {
4047 goto unlock_vcpu_destroy
;
4050 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
4051 r
= xa_reserve(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, GFP_KERNEL_ACCOUNT
);
4053 goto unlock_vcpu_destroy
;
4055 /* Now it's all set up, let userspace reach it */
4057 r
= create_vcpu_fd(vcpu
);
4059 goto kvm_put_xa_release
;
4061 if (KVM_BUG_ON(xa_store(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, vcpu
, 0), kvm
)) {
4063 goto kvm_put_xa_release
;
4067 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4068 * pointer before kvm->online_vcpu's incremented value.
4071 atomic_inc(&kvm
->online_vcpus
);
4073 mutex_unlock(&kvm
->lock
);
4074 kvm_arch_vcpu_postcreate(vcpu
);
4075 kvm_create_vcpu_debugfs(vcpu
);
4079 kvm_put_kvm_no_destroy(kvm
);
4080 xa_release(&kvm
->vcpu_array
, vcpu
->vcpu_idx
);
4081 unlock_vcpu_destroy
:
4082 mutex_unlock(&kvm
->lock
);
4083 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
4085 kvm_arch_vcpu_destroy(vcpu
);
4087 free_page((unsigned long)vcpu
->run
);
4089 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
4091 mutex_lock(&kvm
->lock
);
4092 kvm
->created_vcpus
--;
4093 mutex_unlock(&kvm
->lock
);
4097 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
4100 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
4101 vcpu
->sigset_active
= 1;
4102 vcpu
->sigset
= *sigset
;
4104 vcpu
->sigset_active
= 0;
4108 static ssize_t
kvm_vcpu_stats_read(struct file
*file
, char __user
*user_buffer
,
4109 size_t size
, loff_t
*offset
)
4111 struct kvm_vcpu
*vcpu
= file
->private_data
;
4113 return kvm_stats_read(vcpu
->stats_id
, &kvm_vcpu_stats_header
,
4114 &kvm_vcpu_stats_desc
[0], &vcpu
->stat
,
4115 sizeof(vcpu
->stat
), user_buffer
, size
, offset
);
4118 static int kvm_vcpu_stats_release(struct inode
*inode
, struct file
*file
)
4120 struct kvm_vcpu
*vcpu
= file
->private_data
;
4122 kvm_put_kvm(vcpu
->kvm
);
4126 static const struct file_operations kvm_vcpu_stats_fops
= {
4127 .read
= kvm_vcpu_stats_read
,
4128 .release
= kvm_vcpu_stats_release
,
4129 .llseek
= noop_llseek
,
4132 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu
*vcpu
)
4136 char name
[15 + ITOA_MAX_LEN
+ 1];
4138 snprintf(name
, sizeof(name
), "kvm-vcpu-stats:%d", vcpu
->vcpu_id
);
4140 fd
= get_unused_fd_flags(O_CLOEXEC
);
4144 file
= anon_inode_getfile(name
, &kvm_vcpu_stats_fops
, vcpu
, O_RDONLY
);
4147 return PTR_ERR(file
);
4150 kvm_get_kvm(vcpu
->kvm
);
4152 file
->f_mode
|= FMODE_PREAD
;
4153 fd_install(fd
, file
);
4158 static long kvm_vcpu_ioctl(struct file
*filp
,
4159 unsigned int ioctl
, unsigned long arg
)
4161 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4162 void __user
*argp
= (void __user
*)arg
;
4164 struct kvm_fpu
*fpu
= NULL
;
4165 struct kvm_sregs
*kvm_sregs
= NULL
;
4167 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4170 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
4174 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4175 * execution; mutex_lock() would break them.
4177 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
4178 if (r
!= -ENOIOCTLCMD
)
4181 if (mutex_lock_killable(&vcpu
->mutex
))
4189 oldpid
= rcu_access_pointer(vcpu
->pid
);
4190 if (unlikely(oldpid
!= task_pid(current
))) {
4191 /* The thread running this VCPU changed. */
4194 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
4198 newpid
= get_task_pid(current
, PIDTYPE_PID
);
4199 rcu_assign_pointer(vcpu
->pid
, newpid
);
4204 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
4205 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
4208 case KVM_GET_REGS
: {
4209 struct kvm_regs
*kvm_regs
;
4212 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
4215 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
4219 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
4226 case KVM_SET_REGS
: {
4227 struct kvm_regs
*kvm_regs
;
4229 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
4230 if (IS_ERR(kvm_regs
)) {
4231 r
= PTR_ERR(kvm_regs
);
4234 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
4238 case KVM_GET_SREGS
: {
4239 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
4240 GFP_KERNEL_ACCOUNT
);
4244 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
4248 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
4253 case KVM_SET_SREGS
: {
4254 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
4255 if (IS_ERR(kvm_sregs
)) {
4256 r
= PTR_ERR(kvm_sregs
);
4260 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
4263 case KVM_GET_MP_STATE
: {
4264 struct kvm_mp_state mp_state
;
4266 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
4270 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
4275 case KVM_SET_MP_STATE
: {
4276 struct kvm_mp_state mp_state
;
4279 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
4281 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
4284 case KVM_TRANSLATE
: {
4285 struct kvm_translation tr
;
4288 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
4290 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
4294 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
4299 case KVM_SET_GUEST_DEBUG
: {
4300 struct kvm_guest_debug dbg
;
4303 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
4305 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
4308 case KVM_SET_SIGNAL_MASK
: {
4309 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4310 struct kvm_signal_mask kvm_sigmask
;
4311 sigset_t sigset
, *p
;
4316 if (copy_from_user(&kvm_sigmask
, argp
,
4317 sizeof(kvm_sigmask
)))
4320 if (kvm_sigmask
.len
!= sizeof(sigset
))
4323 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
4328 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
4332 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
4336 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
4340 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
4346 fpu
= memdup_user(argp
, sizeof(*fpu
));
4352 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
4355 case KVM_GET_STATS_FD
: {
4356 r
= kvm_vcpu_ioctl_get_stats_fd(vcpu
);
4360 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
4363 mutex_unlock(&vcpu
->mutex
);
4369 #ifdef CONFIG_KVM_COMPAT
4370 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
4371 unsigned int ioctl
, unsigned long arg
)
4373 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4374 void __user
*argp
= compat_ptr(arg
);
4377 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4381 case KVM_SET_SIGNAL_MASK
: {
4382 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4383 struct kvm_signal_mask kvm_sigmask
;
4388 if (copy_from_user(&kvm_sigmask
, argp
,
4389 sizeof(kvm_sigmask
)))
4392 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
4395 if (get_compat_sigset(&sigset
,
4396 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
4398 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
4400 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
4404 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
4412 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
4414 struct kvm_device
*dev
= filp
->private_data
;
4417 return dev
->ops
->mmap(dev
, vma
);
4422 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
4423 int (*accessor
)(struct kvm_device
*dev
,
4424 struct kvm_device_attr
*attr
),
4427 struct kvm_device_attr attr
;
4432 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
4435 return accessor(dev
, &attr
);
4438 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
4441 struct kvm_device
*dev
= filp
->private_data
;
4443 if (dev
->kvm
->mm
!= current
->mm
|| dev
->kvm
->vm_dead
)
4447 case KVM_SET_DEVICE_ATTR
:
4448 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
4449 case KVM_GET_DEVICE_ATTR
:
4450 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
4451 case KVM_HAS_DEVICE_ATTR
:
4452 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
4454 if (dev
->ops
->ioctl
)
4455 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
4461 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
4463 struct kvm_device
*dev
= filp
->private_data
;
4464 struct kvm
*kvm
= dev
->kvm
;
4466 if (dev
->ops
->release
) {
4467 mutex_lock(&kvm
->lock
);
4468 list_del(&dev
->vm_node
);
4469 dev
->ops
->release(dev
);
4470 mutex_unlock(&kvm
->lock
);
4477 static const struct file_operations kvm_device_fops
= {
4478 .unlocked_ioctl
= kvm_device_ioctl
,
4479 .release
= kvm_device_release
,
4480 KVM_COMPAT(kvm_device_ioctl
),
4481 .mmap
= kvm_device_mmap
,
4484 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
4486 if (filp
->f_op
!= &kvm_device_fops
)
4489 return filp
->private_data
;
4492 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
4493 #ifdef CONFIG_KVM_MPIC
4494 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
4495 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
4499 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
4501 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
4504 if (kvm_device_ops_table
[type
] != NULL
)
4507 kvm_device_ops_table
[type
] = ops
;
4511 void kvm_unregister_device_ops(u32 type
)
4513 if (kvm_device_ops_table
[type
] != NULL
)
4514 kvm_device_ops_table
[type
] = NULL
;
4517 static int kvm_ioctl_create_device(struct kvm
*kvm
,
4518 struct kvm_create_device
*cd
)
4520 const struct kvm_device_ops
*ops
;
4521 struct kvm_device
*dev
;
4522 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
4526 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
4529 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
4530 ops
= kvm_device_ops_table
[type
];
4537 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
4544 mutex_lock(&kvm
->lock
);
4545 ret
= ops
->create(dev
, type
);
4547 mutex_unlock(&kvm
->lock
);
4551 list_add(&dev
->vm_node
, &kvm
->devices
);
4552 mutex_unlock(&kvm
->lock
);
4558 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
4560 kvm_put_kvm_no_destroy(kvm
);
4561 mutex_lock(&kvm
->lock
);
4562 list_del(&dev
->vm_node
);
4565 mutex_unlock(&kvm
->lock
);
4575 static int kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
4578 case KVM_CAP_USER_MEMORY
:
4579 case KVM_CAP_USER_MEMORY2
:
4580 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
4581 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
4582 case KVM_CAP_INTERNAL_ERROR_DATA
:
4583 #ifdef CONFIG_HAVE_KVM_MSI
4584 case KVM_CAP_SIGNAL_MSI
:
4586 #ifdef CONFIG_HAVE_KVM_IRQFD
4589 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
4590 case KVM_CAP_CHECK_EXTENSION_VM
:
4591 case KVM_CAP_ENABLE_CAP_VM
:
4592 case KVM_CAP_HALT_POLL
:
4594 #ifdef CONFIG_KVM_MMIO
4595 case KVM_CAP_COALESCED_MMIO
:
4596 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
4597 case KVM_CAP_COALESCED_PIO
:
4600 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4601 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
4602 return KVM_DIRTY_LOG_MANUAL_CAPS
;
4604 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4605 case KVM_CAP_IRQ_ROUTING
:
4606 return KVM_MAX_IRQ_ROUTES
;
4608 #if KVM_ADDRESS_SPACE_NUM > 1
4609 case KVM_CAP_MULTI_ADDRESS_SPACE
:
4610 return KVM_ADDRESS_SPACE_NUM
;
4612 case KVM_CAP_NR_MEMSLOTS
:
4613 return KVM_USER_MEM_SLOTS
;
4614 case KVM_CAP_DIRTY_LOG_RING
:
4615 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4616 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4620 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4621 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4622 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4626 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4627 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
:
4629 case KVM_CAP_BINARY_STATS_FD
:
4630 case KVM_CAP_SYSTEM_EVENT_DATA
:
4635 return kvm_vm_ioctl_check_extension(kvm
, arg
);
4638 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
4642 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
4645 /* the size should be power of 2 */
4646 if (!size
|| (size
& (size
- 1)))
4649 /* Should be bigger to keep the reserved entries, or a page */
4650 if (size
< kvm_dirty_ring_get_rsvd_entries() *
4651 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
4654 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
4655 sizeof(struct kvm_dirty_gfn
))
4658 /* We only allow it to set once */
4659 if (kvm
->dirty_ring_size
)
4662 mutex_lock(&kvm
->lock
);
4664 if (kvm
->created_vcpus
) {
4665 /* We don't allow to change this value after vcpu created */
4668 kvm
->dirty_ring_size
= size
;
4672 mutex_unlock(&kvm
->lock
);
4676 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
4679 struct kvm_vcpu
*vcpu
;
4682 if (!kvm
->dirty_ring_size
)
4685 mutex_lock(&kvm
->slots_lock
);
4687 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4688 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
4690 mutex_unlock(&kvm
->slots_lock
);
4693 kvm_flush_remote_tlbs(kvm
);
4698 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4699 struct kvm_enable_cap
*cap
)
4704 bool kvm_are_all_memslots_empty(struct kvm
*kvm
)
4708 lockdep_assert_held(&kvm
->slots_lock
);
4710 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
4711 if (!kvm_memslots_empty(__kvm_memslots(kvm
, i
)))
4717 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty
);
4719 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
4720 struct kvm_enable_cap
*cap
)
4723 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4724 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
4725 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
4727 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
4728 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
4730 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
4732 kvm
->manual_dirty_log_protect
= cap
->args
[0];
4736 case KVM_CAP_HALT_POLL
: {
4737 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
4740 kvm
->max_halt_poll_ns
= cap
->args
[0];
4743 * Ensure kvm->override_halt_poll_ns does not become visible
4744 * before kvm->max_halt_poll_ns.
4746 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4749 kvm
->override_halt_poll_ns
= true;
4753 case KVM_CAP_DIRTY_LOG_RING
:
4754 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4755 if (!kvm_vm_ioctl_check_extension_generic(kvm
, cap
->cap
))
4758 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
4759 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
: {
4762 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
) ||
4763 !kvm
->dirty_ring_size
|| cap
->flags
)
4766 mutex_lock(&kvm
->slots_lock
);
4769 * For simplicity, allow enabling ring+bitmap if and only if
4770 * there are no memslots, e.g. to ensure all memslots allocate
4771 * a bitmap after the capability is enabled.
4773 if (kvm_are_all_memslots_empty(kvm
)) {
4774 kvm
->dirty_ring_with_bitmap
= true;
4778 mutex_unlock(&kvm
->slots_lock
);
4783 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
4787 static ssize_t
kvm_vm_stats_read(struct file
*file
, char __user
*user_buffer
,
4788 size_t size
, loff_t
*offset
)
4790 struct kvm
*kvm
= file
->private_data
;
4792 return kvm_stats_read(kvm
->stats_id
, &kvm_vm_stats_header
,
4793 &kvm_vm_stats_desc
[0], &kvm
->stat
,
4794 sizeof(kvm
->stat
), user_buffer
, size
, offset
);
4797 static int kvm_vm_stats_release(struct inode
*inode
, struct file
*file
)
4799 struct kvm
*kvm
= file
->private_data
;
4805 static const struct file_operations kvm_vm_stats_fops
= {
4806 .read
= kvm_vm_stats_read
,
4807 .release
= kvm_vm_stats_release
,
4808 .llseek
= noop_llseek
,
4811 static int kvm_vm_ioctl_get_stats_fd(struct kvm
*kvm
)
4816 fd
= get_unused_fd_flags(O_CLOEXEC
);
4820 file
= anon_inode_getfile("kvm-vm-stats",
4821 &kvm_vm_stats_fops
, kvm
, O_RDONLY
);
4824 return PTR_ERR(file
);
4829 file
->f_mode
|= FMODE_PREAD
;
4830 fd_install(fd
, file
);
4835 #define SANITY_CHECK_MEM_REGION_FIELD(field) \
4837 BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) != \
4838 offsetof(struct kvm_userspace_memory_region2, field)); \
4839 BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) != \
4840 sizeof_field(struct kvm_userspace_memory_region2, field)); \
4843 static long kvm_vm_ioctl(struct file
*filp
,
4844 unsigned int ioctl
, unsigned long arg
)
4846 struct kvm
*kvm
= filp
->private_data
;
4847 void __user
*argp
= (void __user
*)arg
;
4850 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
4853 case KVM_CREATE_VCPU
:
4854 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
4856 case KVM_ENABLE_CAP
: {
4857 struct kvm_enable_cap cap
;
4860 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4862 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
4865 case KVM_SET_USER_MEMORY_REGION2
:
4866 case KVM_SET_USER_MEMORY_REGION
: {
4867 struct kvm_userspace_memory_region2 mem
;
4870 if (ioctl
== KVM_SET_USER_MEMORY_REGION
) {
4872 * Fields beyond struct kvm_userspace_memory_region shouldn't be
4873 * accessed, but avoid leaking kernel memory in case of a bug.
4875 memset(&mem
, 0, sizeof(mem
));
4876 size
= sizeof(struct kvm_userspace_memory_region
);
4878 size
= sizeof(struct kvm_userspace_memory_region2
);
4881 /* Ensure the common parts of the two structs are identical. */
4882 SANITY_CHECK_MEM_REGION_FIELD(slot
);
4883 SANITY_CHECK_MEM_REGION_FIELD(flags
);
4884 SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr
);
4885 SANITY_CHECK_MEM_REGION_FIELD(memory_size
);
4886 SANITY_CHECK_MEM_REGION_FIELD(userspace_addr
);
4889 if (copy_from_user(&mem
, argp
, size
))
4893 if (ioctl
== KVM_SET_USER_MEMORY_REGION
&&
4894 (mem
.flags
& ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS
))
4897 r
= kvm_vm_ioctl_set_memory_region(kvm
, &mem
);
4900 case KVM_GET_DIRTY_LOG
: {
4901 struct kvm_dirty_log log
;
4904 if (copy_from_user(&log
, argp
, sizeof(log
)))
4906 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
4909 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4910 case KVM_CLEAR_DIRTY_LOG
: {
4911 struct kvm_clear_dirty_log log
;
4914 if (copy_from_user(&log
, argp
, sizeof(log
)))
4916 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
4920 #ifdef CONFIG_KVM_MMIO
4921 case KVM_REGISTER_COALESCED_MMIO
: {
4922 struct kvm_coalesced_mmio_zone zone
;
4925 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4927 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
4930 case KVM_UNREGISTER_COALESCED_MMIO
: {
4931 struct kvm_coalesced_mmio_zone zone
;
4934 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4936 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
4941 struct kvm_irqfd data
;
4944 if (copy_from_user(&data
, argp
, sizeof(data
)))
4946 r
= kvm_irqfd(kvm
, &data
);
4949 case KVM_IOEVENTFD
: {
4950 struct kvm_ioeventfd data
;
4953 if (copy_from_user(&data
, argp
, sizeof(data
)))
4955 r
= kvm_ioeventfd(kvm
, &data
);
4958 #ifdef CONFIG_HAVE_KVM_MSI
4959 case KVM_SIGNAL_MSI
: {
4963 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
4965 r
= kvm_send_userspace_msi(kvm
, &msi
);
4969 #ifdef __KVM_HAVE_IRQ_LINE
4970 case KVM_IRQ_LINE_STATUS
:
4971 case KVM_IRQ_LINE
: {
4972 struct kvm_irq_level irq_event
;
4975 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
4978 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
4979 ioctl
== KVM_IRQ_LINE_STATUS
);
4984 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
4985 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
4993 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4994 case KVM_SET_GSI_ROUTING
: {
4995 struct kvm_irq_routing routing
;
4996 struct kvm_irq_routing __user
*urouting
;
4997 struct kvm_irq_routing_entry
*entries
= NULL
;
5000 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
5003 if (!kvm_arch_can_set_irq_routing(kvm
))
5005 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
5011 entries
= vmemdup_user(urouting
->entries
,
5012 array_size(sizeof(*entries
),
5014 if (IS_ERR(entries
)) {
5015 r
= PTR_ERR(entries
);
5019 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
5024 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
5025 case KVM_CREATE_DEVICE
: {
5026 struct kvm_create_device cd
;
5029 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
5032 r
= kvm_ioctl_create_device(kvm
, &cd
);
5037 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
5043 case KVM_CHECK_EXTENSION
:
5044 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
5046 case KVM_RESET_DIRTY_RINGS
:
5047 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
5049 case KVM_GET_STATS_FD
:
5050 r
= kvm_vm_ioctl_get_stats_fd(kvm
);
5053 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
5059 #ifdef CONFIG_KVM_COMPAT
5060 struct compat_kvm_dirty_log
{
5064 compat_uptr_t dirty_bitmap
; /* one bit per page */
5069 struct compat_kvm_clear_dirty_log
{
5074 compat_uptr_t dirty_bitmap
; /* one bit per page */
5079 long __weak
kvm_arch_vm_compat_ioctl(struct file
*filp
, unsigned int ioctl
,
5085 static long kvm_vm_compat_ioctl(struct file
*filp
,
5086 unsigned int ioctl
, unsigned long arg
)
5088 struct kvm
*kvm
= filp
->private_data
;
5091 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5094 r
= kvm_arch_vm_compat_ioctl(filp
, ioctl
, arg
);
5099 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5100 case KVM_CLEAR_DIRTY_LOG
: {
5101 struct compat_kvm_clear_dirty_log compat_log
;
5102 struct kvm_clear_dirty_log log
;
5104 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5105 sizeof(compat_log
)))
5107 log
.slot
= compat_log
.slot
;
5108 log
.num_pages
= compat_log
.num_pages
;
5109 log
.first_page
= compat_log
.first_page
;
5110 log
.padding2
= compat_log
.padding2
;
5111 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5113 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5117 case KVM_GET_DIRTY_LOG
: {
5118 struct compat_kvm_dirty_log compat_log
;
5119 struct kvm_dirty_log log
;
5121 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5122 sizeof(compat_log
)))
5124 log
.slot
= compat_log
.slot
;
5125 log
.padding1
= compat_log
.padding1
;
5126 log
.padding2
= compat_log
.padding2
;
5127 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5129 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5133 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
5139 static const struct file_operations kvm_vm_fops
= {
5140 .release
= kvm_vm_release
,
5141 .unlocked_ioctl
= kvm_vm_ioctl
,
5142 .llseek
= noop_llseek
,
5143 KVM_COMPAT(kvm_vm_compat_ioctl
),
5146 bool file_is_kvm(struct file
*file
)
5148 return file
&& file
->f_op
== &kvm_vm_fops
;
5150 EXPORT_SYMBOL_GPL(file_is_kvm
);
5152 static int kvm_dev_ioctl_create_vm(unsigned long type
)
5154 char fdname
[ITOA_MAX_LEN
+ 1];
5159 fd
= get_unused_fd_flags(O_CLOEXEC
);
5163 snprintf(fdname
, sizeof(fdname
), "%d", fd
);
5165 kvm
= kvm_create_vm(type
, fdname
);
5171 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
5178 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5179 * already set, with ->release() being kvm_vm_release(). In error
5180 * cases it will be called by the final fput(file) and will take
5181 * care of doing kvm_put_kvm(kvm).
5183 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
5185 fd_install(fd
, file
);
5195 static long kvm_dev_ioctl(struct file
*filp
,
5196 unsigned int ioctl
, unsigned long arg
)
5201 case KVM_GET_API_VERSION
:
5204 r
= KVM_API_VERSION
;
5207 r
= kvm_dev_ioctl_create_vm(arg
);
5209 case KVM_CHECK_EXTENSION
:
5210 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
5212 case KVM_GET_VCPU_MMAP_SIZE
:
5215 r
= PAGE_SIZE
; /* struct kvm_run */
5217 r
+= PAGE_SIZE
; /* pio data page */
5219 #ifdef CONFIG_KVM_MMIO
5220 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
5223 case KVM_TRACE_ENABLE
:
5224 case KVM_TRACE_PAUSE
:
5225 case KVM_TRACE_DISABLE
:
5229 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
5235 static struct file_operations kvm_chardev_ops
= {
5236 .unlocked_ioctl
= kvm_dev_ioctl
,
5237 .llseek
= noop_llseek
,
5238 KVM_COMPAT(kvm_dev_ioctl
),
5241 static struct miscdevice kvm_dev
= {
5247 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5248 __visible
bool kvm_rebooting
;
5249 EXPORT_SYMBOL_GPL(kvm_rebooting
);
5251 static DEFINE_PER_CPU(bool, hardware_enabled
);
5252 static int kvm_usage_count
;
5254 static int __hardware_enable_nolock(void)
5256 if (__this_cpu_read(hardware_enabled
))
5259 if (kvm_arch_hardware_enable()) {
5260 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5261 raw_smp_processor_id());
5265 __this_cpu_write(hardware_enabled
, true);
5269 static void hardware_enable_nolock(void *failed
)
5271 if (__hardware_enable_nolock())
5275 static int kvm_online_cpu(unsigned int cpu
)
5280 * Abort the CPU online process if hardware virtualization cannot
5281 * be enabled. Otherwise running VMs would encounter unrecoverable
5282 * errors when scheduled to this CPU.
5284 mutex_lock(&kvm_lock
);
5285 if (kvm_usage_count
)
5286 ret
= __hardware_enable_nolock();
5287 mutex_unlock(&kvm_lock
);
5291 static void hardware_disable_nolock(void *junk
)
5294 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5295 * hardware, not just CPUs that successfully enabled hardware!
5297 if (!__this_cpu_read(hardware_enabled
))
5300 kvm_arch_hardware_disable();
5302 __this_cpu_write(hardware_enabled
, false);
5305 static int kvm_offline_cpu(unsigned int cpu
)
5307 mutex_lock(&kvm_lock
);
5308 if (kvm_usage_count
)
5309 hardware_disable_nolock(NULL
);
5310 mutex_unlock(&kvm_lock
);
5314 static void hardware_disable_all_nolock(void)
5316 BUG_ON(!kvm_usage_count
);
5319 if (!kvm_usage_count
)
5320 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5323 static void hardware_disable_all(void)
5326 mutex_lock(&kvm_lock
);
5327 hardware_disable_all_nolock();
5328 mutex_unlock(&kvm_lock
);
5332 static int hardware_enable_all(void)
5334 atomic_t failed
= ATOMIC_INIT(0);
5338 * Do not enable hardware virtualization if the system is going down.
5339 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5340 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5341 * after kvm_reboot() is called. Note, this relies on system_state
5342 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5343 * hook instead of registering a dedicated reboot notifier (the latter
5344 * runs before system_state is updated).
5346 if (system_state
== SYSTEM_HALT
|| system_state
== SYSTEM_POWER_OFF
||
5347 system_state
== SYSTEM_RESTART
)
5351 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5352 * is called, and so on_each_cpu() between them includes the CPU that
5353 * is being onlined. As a result, hardware_enable_nolock() may get
5354 * invoked before kvm_online_cpu(), which also enables hardware if the
5355 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5356 * enable hardware multiple times.
5359 mutex_lock(&kvm_lock
);
5364 if (kvm_usage_count
== 1) {
5365 on_each_cpu(hardware_enable_nolock
, &failed
, 1);
5367 if (atomic_read(&failed
)) {
5368 hardware_disable_all_nolock();
5373 mutex_unlock(&kvm_lock
);
5379 static void kvm_shutdown(void)
5382 * Disable hardware virtualization and set kvm_rebooting to indicate
5383 * that KVM has asynchronously disabled hardware virtualization, i.e.
5384 * that relevant errors and exceptions aren't entirely unexpected.
5385 * Some flavors of hardware virtualization need to be disabled before
5386 * transferring control to firmware (to perform shutdown/reboot), e.g.
5387 * on x86, virtualization can block INIT interrupts, which are used by
5388 * firmware to pull APs back under firmware control. Note, this path
5389 * is used for both shutdown and reboot scenarios, i.e. neither name is
5390 * 100% comprehensive.
5392 pr_info("kvm: exiting hardware virtualization\n");
5393 kvm_rebooting
= true;
5394 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5397 static int kvm_suspend(void)
5400 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5401 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5402 * is stable. Assert that kvm_lock is not held to ensure the system
5403 * isn't suspended while KVM is enabling hardware. Hardware enabling
5404 * can be preempted, but the task cannot be frozen until it has dropped
5405 * all locks (userspace tasks are frozen via a fake signal).
5407 lockdep_assert_not_held(&kvm_lock
);
5408 lockdep_assert_irqs_disabled();
5410 if (kvm_usage_count
)
5411 hardware_disable_nolock(NULL
);
5415 static void kvm_resume(void)
5417 lockdep_assert_not_held(&kvm_lock
);
5418 lockdep_assert_irqs_disabled();
5420 if (kvm_usage_count
)
5421 WARN_ON_ONCE(__hardware_enable_nolock());
5424 static struct syscore_ops kvm_syscore_ops
= {
5425 .suspend
= kvm_suspend
,
5426 .resume
= kvm_resume
,
5427 .shutdown
= kvm_shutdown
,
5429 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5430 static int hardware_enable_all(void)
5435 static void hardware_disable_all(void)
5439 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5441 static void kvm_iodevice_destructor(struct kvm_io_device
*dev
)
5443 if (dev
->ops
->destructor
)
5444 dev
->ops
->destructor(dev
);
5447 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
5451 for (i
= 0; i
< bus
->dev_count
; i
++) {
5452 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
5454 kvm_iodevice_destructor(pos
);
5459 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
5460 const struct kvm_io_range
*r2
)
5462 gpa_t addr1
= r1
->addr
;
5463 gpa_t addr2
= r2
->addr
;
5468 /* If r2->len == 0, match the exact address. If r2->len != 0,
5469 * accept any overlapping write. Any order is acceptable for
5470 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5471 * we process all of them.
5484 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
5486 return kvm_io_bus_cmp(p1
, p2
);
5489 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
5490 gpa_t addr
, int len
)
5492 struct kvm_io_range
*range
, key
;
5495 key
= (struct kvm_io_range
) {
5500 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
5501 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
5505 off
= range
- bus
->range
;
5507 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
5513 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5514 struct kvm_io_range
*range
, const void *val
)
5518 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5522 while (idx
< bus
->dev_count
&&
5523 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5524 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5533 /* kvm_io_bus_write - called under kvm->slots_lock */
5534 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5535 int len
, const void *val
)
5537 struct kvm_io_bus
*bus
;
5538 struct kvm_io_range range
;
5541 range
= (struct kvm_io_range
) {
5546 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5549 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5550 return r
< 0 ? r
: 0;
5552 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
5554 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5555 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
5556 gpa_t addr
, int len
, const void *val
, long cookie
)
5558 struct kvm_io_bus
*bus
;
5559 struct kvm_io_range range
;
5561 range
= (struct kvm_io_range
) {
5566 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5570 /* First try the device referenced by cookie. */
5571 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
5572 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
5573 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
5578 * cookie contained garbage; fall back to search and return the
5579 * correct cookie value.
5581 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5584 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5585 struct kvm_io_range
*range
, void *val
)
5589 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5593 while (idx
< bus
->dev_count
&&
5594 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5595 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5604 /* kvm_io_bus_read - called under kvm->slots_lock */
5605 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5608 struct kvm_io_bus
*bus
;
5609 struct kvm_io_range range
;
5612 range
= (struct kvm_io_range
) {
5617 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5620 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
5621 return r
< 0 ? r
: 0;
5624 /* Caller must hold slots_lock. */
5625 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
5626 int len
, struct kvm_io_device
*dev
)
5629 struct kvm_io_bus
*new_bus
, *bus
;
5630 struct kvm_io_range range
;
5632 bus
= kvm_get_bus(kvm
, bus_idx
);
5636 /* exclude ioeventfd which is limited by maximum fd */
5637 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
5640 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
5641 GFP_KERNEL_ACCOUNT
);
5645 range
= (struct kvm_io_range
) {
5651 for (i
= 0; i
< bus
->dev_count
; i
++)
5652 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
5655 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
5656 new_bus
->dev_count
++;
5657 new_bus
->range
[i
] = range
;
5658 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
5659 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
5660 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5661 synchronize_srcu_expedited(&kvm
->srcu
);
5667 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5668 struct kvm_io_device
*dev
)
5671 struct kvm_io_bus
*new_bus
, *bus
;
5673 lockdep_assert_held(&kvm
->slots_lock
);
5675 bus
= kvm_get_bus(kvm
, bus_idx
);
5679 for (i
= 0; i
< bus
->dev_count
; i
++) {
5680 if (bus
->range
[i
].dev
== dev
) {
5685 if (i
== bus
->dev_count
)
5688 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
5689 GFP_KERNEL_ACCOUNT
);
5691 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
5692 new_bus
->dev_count
--;
5693 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
5694 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
5697 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5698 synchronize_srcu_expedited(&kvm
->srcu
);
5701 * If NULL bus is installed, destroy the old bus, including all the
5702 * attached devices. Otherwise, destroy the caller's device only.
5705 pr_err("kvm: failed to shrink bus, removing it completely\n");
5706 kvm_io_bus_destroy(bus
);
5710 kvm_iodevice_destructor(dev
);
5715 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5718 struct kvm_io_bus
*bus
;
5719 int dev_idx
, srcu_idx
;
5720 struct kvm_io_device
*iodev
= NULL
;
5722 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5724 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
5728 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
5732 iodev
= bus
->range
[dev_idx
].dev
;
5735 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
5739 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
5741 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
5742 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
5746 struct kvm_stat_data
*stat_data
= inode
->i_private
;
5749 * The debugfs files are a reference to the kvm struct which
5750 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5751 * avoids the race between open and the removal of the debugfs directory.
5753 if (!kvm_get_kvm_safe(stat_data
->kvm
))
5756 ret
= simple_attr_open(inode
, file
, get
,
5757 kvm_stats_debugfs_mode(stat_data
->desc
) & 0222
5760 kvm_put_kvm(stat_data
->kvm
);
5765 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
5767 struct kvm_stat_data
*stat_data
= inode
->i_private
;
5769 simple_attr_release(inode
, file
);
5770 kvm_put_kvm(stat_data
->kvm
);
5775 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
5777 *val
= *(u64
*)((void *)(&kvm
->stat
) + offset
);
5782 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
5784 *(u64
*)((void *)(&kvm
->stat
) + offset
) = 0;
5789 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
5792 struct kvm_vcpu
*vcpu
;
5796 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5797 *val
+= *(u64
*)((void *)(&vcpu
->stat
) + offset
);
5802 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
5805 struct kvm_vcpu
*vcpu
;
5807 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5808 *(u64
*)((void *)(&vcpu
->stat
) + offset
) = 0;
5813 static int kvm_stat_data_get(void *data
, u64
*val
)
5816 struct kvm_stat_data
*stat_data
= data
;
5818 switch (stat_data
->kind
) {
5820 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
5821 stat_data
->desc
->desc
.offset
, val
);
5824 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
5825 stat_data
->desc
->desc
.offset
, val
);
5832 static int kvm_stat_data_clear(void *data
, u64 val
)
5835 struct kvm_stat_data
*stat_data
= data
;
5840 switch (stat_data
->kind
) {
5842 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
5843 stat_data
->desc
->desc
.offset
);
5846 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
5847 stat_data
->desc
->desc
.offset
);
5854 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
5856 __simple_attr_check_format("%llu\n", 0ull);
5857 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
5858 kvm_stat_data_clear
, "%llu\n");
5861 static const struct file_operations stat_fops_per_vm
= {
5862 .owner
= THIS_MODULE
,
5863 .open
= kvm_stat_data_open
,
5864 .release
= kvm_debugfs_release
,
5865 .read
= simple_attr_read
,
5866 .write
= simple_attr_write
,
5867 .llseek
= no_llseek
,
5870 static int vm_stat_get(void *_offset
, u64
*val
)
5872 unsigned offset
= (long)_offset
;
5877 mutex_lock(&kvm_lock
);
5878 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5879 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
5882 mutex_unlock(&kvm_lock
);
5886 static int vm_stat_clear(void *_offset
, u64 val
)
5888 unsigned offset
= (long)_offset
;
5894 mutex_lock(&kvm_lock
);
5895 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5896 kvm_clear_stat_per_vm(kvm
, offset
);
5898 mutex_unlock(&kvm_lock
);
5903 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
5904 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops
, vm_stat_get
, NULL
, "%llu\n");
5906 static int vcpu_stat_get(void *_offset
, u64
*val
)
5908 unsigned offset
= (long)_offset
;
5913 mutex_lock(&kvm_lock
);
5914 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5915 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
5918 mutex_unlock(&kvm_lock
);
5922 static int vcpu_stat_clear(void *_offset
, u64 val
)
5924 unsigned offset
= (long)_offset
;
5930 mutex_lock(&kvm_lock
);
5931 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5932 kvm_clear_stat_per_vcpu(kvm
, offset
);
5934 mutex_unlock(&kvm_lock
);
5939 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
5941 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops
, vcpu_stat_get
, NULL
, "%llu\n");
5943 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
5945 struct kobj_uevent_env
*env
;
5946 unsigned long long created
, active
;
5948 if (!kvm_dev
.this_device
|| !kvm
)
5951 mutex_lock(&kvm_lock
);
5952 if (type
== KVM_EVENT_CREATE_VM
) {
5953 kvm_createvm_count
++;
5955 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5958 created
= kvm_createvm_count
;
5959 active
= kvm_active_vms
;
5960 mutex_unlock(&kvm_lock
);
5962 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
5966 add_uevent_var(env
, "CREATED=%llu", created
);
5967 add_uevent_var(env
, "COUNT=%llu", active
);
5969 if (type
== KVM_EVENT_CREATE_VM
) {
5970 add_uevent_var(env
, "EVENT=create");
5971 kvm
->userspace_pid
= task_pid_nr(current
);
5972 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5973 add_uevent_var(env
, "EVENT=destroy");
5975 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
5977 if (!IS_ERR(kvm
->debugfs_dentry
)) {
5978 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
5981 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
5983 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
5987 /* no need for checks, since we are adding at most only 5 keys */
5988 env
->envp
[env
->envp_idx
++] = NULL
;
5989 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
5993 static void kvm_init_debug(void)
5995 const struct file_operations
*fops
;
5996 const struct _kvm_stats_desc
*pdesc
;
5999 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
6001 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
6002 pdesc
= &kvm_vm_stats_desc
[i
];
6003 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
6004 fops
= &vm_stat_fops
;
6006 fops
= &vm_stat_readonly_fops
;
6007 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
6009 (void *)(long)pdesc
->desc
.offset
, fops
);
6012 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
6013 pdesc
= &kvm_vcpu_stats_desc
[i
];
6014 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
6015 fops
= &vcpu_stat_fops
;
6017 fops
= &vcpu_stat_readonly_fops
;
6018 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
6020 (void *)(long)pdesc
->desc
.offset
, fops
);
6025 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
6027 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
6030 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
6032 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
6034 WRITE_ONCE(vcpu
->preempted
, false);
6035 WRITE_ONCE(vcpu
->ready
, false);
6037 __this_cpu_write(kvm_running_vcpu
, vcpu
);
6038 kvm_arch_sched_in(vcpu
, cpu
);
6039 kvm_arch_vcpu_load(vcpu
, cpu
);
6042 static void kvm_sched_out(struct preempt_notifier
*pn
,
6043 struct task_struct
*next
)
6045 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
6047 if (current
->on_rq
) {
6048 WRITE_ONCE(vcpu
->preempted
, true);
6049 WRITE_ONCE(vcpu
->ready
, true);
6051 kvm_arch_vcpu_put(vcpu
);
6052 __this_cpu_write(kvm_running_vcpu
, NULL
);
6056 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
6058 * We can disable preemption locally around accessing the per-CPU variable,
6059 * and use the resolved vcpu pointer after enabling preemption again,
6060 * because even if the current thread is migrated to another CPU, reading
6061 * the per-CPU value later will give us the same value as we update the
6062 * per-CPU variable in the preempt notifier handlers.
6064 struct kvm_vcpu
*kvm_get_running_vcpu(void)
6066 struct kvm_vcpu
*vcpu
;
6069 vcpu
= __this_cpu_read(kvm_running_vcpu
);
6074 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
6077 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
6079 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
6081 return &kvm_running_vcpu
;
6084 #ifdef CONFIG_GUEST_PERF_EVENTS
6085 static unsigned int kvm_guest_state(void)
6087 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6090 if (!kvm_arch_pmi_in_guest(vcpu
))
6093 state
= PERF_GUEST_ACTIVE
;
6094 if (!kvm_arch_vcpu_in_kernel(vcpu
))
6095 state
|= PERF_GUEST_USER
;
6100 static unsigned long kvm_guest_get_ip(void)
6102 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6104 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6105 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu
)))
6108 return kvm_arch_vcpu_get_ip(vcpu
);
6111 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6112 .state
= kvm_guest_state
,
6113 .get_ip
= kvm_guest_get_ip
,
6114 .handle_intel_pt_intr
= NULL
,
6117 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler
)(void))
6119 kvm_guest_cbs
.handle_intel_pt_intr
= pt_intr_handler
;
6120 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6122 void kvm_unregister_perf_callbacks(void)
6124 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6128 int kvm_init(unsigned vcpu_size
, unsigned vcpu_align
, struct module
*module
)
6133 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6134 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE
, "kvm/cpu:online",
6135 kvm_online_cpu
, kvm_offline_cpu
);
6139 register_syscore_ops(&kvm_syscore_ops
);
6142 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6144 vcpu_align
= __alignof__(struct kvm_vcpu
);
6146 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
6148 offsetof(struct kvm_vcpu
, arch
),
6149 offsetofend(struct kvm_vcpu
, stats_id
)
6150 - offsetof(struct kvm_vcpu
, arch
),
6152 if (!kvm_vcpu_cache
) {
6154 goto err_vcpu_cache
;
6157 for_each_possible_cpu(cpu
) {
6158 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask
, cpu
),
6159 GFP_KERNEL
, cpu_to_node(cpu
))) {
6161 goto err_cpu_kick_mask
;
6165 r
= kvm_irqfd_init();
6169 r
= kvm_async_pf_init();
6173 kvm_chardev_ops
.owner
= module
;
6175 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
6176 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
6180 r
= kvm_vfio_ops_init();
6181 if (WARN_ON_ONCE(r
))
6185 * Registration _must_ be the very last thing done, as this exposes
6186 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6188 r
= misc_register(&kvm_dev
);
6190 pr_err("kvm: misc device register failed\n");
6197 kvm_vfio_ops_exit();
6199 kvm_async_pf_deinit();
6204 for_each_possible_cpu(cpu
)
6205 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6206 kmem_cache_destroy(kvm_vcpu_cache
);
6208 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6209 unregister_syscore_ops(&kvm_syscore_ops
);
6210 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6214 EXPORT_SYMBOL_GPL(kvm_init
);
6221 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6222 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6223 * to KVM while the module is being stopped.
6225 misc_deregister(&kvm_dev
);
6227 debugfs_remove_recursive(kvm_debugfs_dir
);
6228 for_each_possible_cpu(cpu
)
6229 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6230 kmem_cache_destroy(kvm_vcpu_cache
);
6231 kvm_vfio_ops_exit();
6232 kvm_async_pf_deinit();
6233 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6234 unregister_syscore_ops(&kvm_syscore_ops
);
6235 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6239 EXPORT_SYMBOL_GPL(kvm_exit
);
6241 struct kvm_vm_worker_thread_context
{
6243 struct task_struct
*parent
;
6244 struct completion init_done
;
6245 kvm_vm_thread_fn_t thread_fn
;
6250 static int kvm_vm_worker_thread(void *context
)
6253 * The init_context is allocated on the stack of the parent thread, so
6254 * we have to locally copy anything that is needed beyond initialization
6256 struct kvm_vm_worker_thread_context
*init_context
= context
;
6257 struct task_struct
*parent
;
6258 struct kvm
*kvm
= init_context
->kvm
;
6259 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
6260 uintptr_t data
= init_context
->data
;
6263 err
= kthread_park(current
);
6264 /* kthread_park(current) is never supposed to return an error */
6269 err
= cgroup_attach_task_all(init_context
->parent
, current
);
6271 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6276 set_user_nice(current
, task_nice(init_context
->parent
));
6279 init_context
->err
= err
;
6280 complete(&init_context
->init_done
);
6281 init_context
= NULL
;
6286 /* Wait to be woken up by the spawner before proceeding. */
6289 if (!kthread_should_stop())
6290 err
= thread_fn(kvm
, data
);
6294 * Move kthread back to its original cgroup to prevent it lingering in
6295 * the cgroup of the VM process, after the latter finishes its
6298 * kthread_stop() waits on the 'exited' completion condition which is
6299 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6300 * kthread is removed from the cgroup in the cgroup_exit() which is
6301 * called after the exit_mm(). This causes the kthread_stop() to return
6302 * before the kthread actually quits the cgroup.
6305 parent
= rcu_dereference(current
->real_parent
);
6306 get_task_struct(parent
);
6308 cgroup_attach_task_all(parent
, current
);
6309 put_task_struct(parent
);
6314 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
6315 uintptr_t data
, const char *name
,
6316 struct task_struct
**thread_ptr
)
6318 struct kvm_vm_worker_thread_context init_context
= {};
6319 struct task_struct
*thread
;
6322 init_context
.kvm
= kvm
;
6323 init_context
.parent
= current
;
6324 init_context
.thread_fn
= thread_fn
;
6325 init_context
.data
= data
;
6326 init_completion(&init_context
.init_done
);
6328 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
6329 "%s-%d", name
, task_pid_nr(current
));
6331 return PTR_ERR(thread
);
6333 /* kthread_run is never supposed to return NULL */
6334 WARN_ON(thread
== NULL
);
6336 wait_for_completion(&init_context
.init_done
);
6338 if (!init_context
.err
)
6339 *thread_ptr
= thread
;
6341 return init_context
.err
;