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 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_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
, unsigned long start
,
549 typedef void (*on_unlock_fn_t
)(struct kvm
*kvm
);
551 struct kvm_mmu_notifier_range
{
553 * 64-bit addresses, as KVM notifiers can operate on host virtual
554 * addresses (unsigned long) and guest physical addresses (64-bit).
558 union kvm_mmu_notifier_arg arg
;
559 gfn_handler_t handler
;
560 on_lock_fn_t on_lock
;
561 on_unlock_fn_t on_unlock
;
567 * Use a dedicated stub instead of NULL to indicate that there is no callback
568 * function/handler. The compiler technically can't guarantee that a real
569 * function will have a non-zero address, and so it will generate code to
570 * check for !NULL, whereas comparing against a stub will be elided at compile
571 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
573 static void kvm_null_fn(void)
577 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
579 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG
;
581 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
582 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
583 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
585 node = interval_tree_iter_next(node, start, last)) \
587 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
588 const struct kvm_mmu_notifier_range
*range
)
590 bool ret
= false, locked
= false;
591 struct kvm_gfn_range gfn_range
;
592 struct kvm_memory_slot
*slot
;
593 struct kvm_memslots
*slots
;
596 if (WARN_ON_ONCE(range
->end
<= range
->start
))
599 /* A null handler is allowed if and only if on_lock() is provided. */
600 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range
->on_lock
) &&
601 IS_KVM_NULL_FN(range
->handler
)))
604 idx
= srcu_read_lock(&kvm
->srcu
);
606 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
607 struct interval_tree_node
*node
;
609 slots
= __kvm_memslots(kvm
, i
);
610 kvm_for_each_memslot_in_hva_range(node
, slots
,
611 range
->start
, range
->end
- 1) {
612 unsigned long hva_start
, hva_end
;
614 slot
= container_of(node
, struct kvm_memory_slot
, hva_node
[slots
->node_idx
]);
615 hva_start
= max_t(unsigned long, range
->start
, slot
->userspace_addr
);
616 hva_end
= min_t(unsigned long, range
->end
,
617 slot
->userspace_addr
+ (slot
->npages
<< PAGE_SHIFT
));
620 * To optimize for the likely case where the address
621 * range is covered by zero or one memslots, don't
622 * bother making these conditional (to avoid writes on
623 * the second or later invocation of the handler).
625 gfn_range
.arg
= range
->arg
;
626 gfn_range
.may_block
= range
->may_block
;
629 * {gfn(page) | page intersects with [hva_start, hva_end)} =
630 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
632 gfn_range
.start
= hva_to_gfn_memslot(hva_start
, slot
);
633 gfn_range
.end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, slot
);
634 gfn_range
.slot
= slot
;
639 if (!IS_KVM_NULL_FN(range
->on_lock
))
640 range
->on_lock(kvm
, range
->start
, range
->end
);
641 if (IS_KVM_NULL_FN(range
->handler
))
644 ret
|= range
->handler(kvm
, &gfn_range
);
648 if (range
->flush_on_ret
&& ret
)
649 kvm_flush_remote_tlbs(kvm
);
653 if (!IS_KVM_NULL_FN(range
->on_unlock
))
654 range
->on_unlock(kvm
);
657 srcu_read_unlock(&kvm
->srcu
, idx
);
659 /* The notifiers are averse to booleans. :-( */
663 static __always_inline
int kvm_handle_hva_range(struct mmu_notifier
*mn
,
666 union kvm_mmu_notifier_arg arg
,
667 gfn_handler_t handler
)
669 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
670 const struct kvm_mmu_notifier_range range
= {
675 .on_lock
= (void *)kvm_null_fn
,
676 .on_unlock
= (void *)kvm_null_fn
,
677 .flush_on_ret
= true,
681 return __kvm_handle_hva_range(kvm
, &range
);
684 static __always_inline
int kvm_handle_hva_range_no_flush(struct mmu_notifier
*mn
,
687 gfn_handler_t handler
)
689 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
690 const struct kvm_mmu_notifier_range range
= {
694 .on_lock
= (void *)kvm_null_fn
,
695 .on_unlock
= (void *)kvm_null_fn
,
696 .flush_on_ret
= false,
700 return __kvm_handle_hva_range(kvm
, &range
);
703 static bool kvm_change_spte_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
706 * Skipping invalid memslots is correct if and only change_pte() is
707 * surrounded by invalidate_range_{start,end}(), which is currently
708 * guaranteed by the primary MMU. If that ever changes, KVM needs to
709 * unmap the memslot instead of skipping the memslot to ensure that KVM
710 * doesn't hold references to the old PFN.
712 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
714 if (range
->slot
->flags
& KVM_MEMSLOT_INVALID
)
717 return kvm_set_spte_gfn(kvm
, range
);
720 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
721 struct mm_struct
*mm
,
722 unsigned long address
,
725 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
726 const union kvm_mmu_notifier_arg arg
= { .pte
= pte
};
728 trace_kvm_set_spte_hva(address
);
731 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
732 * If mmu_invalidate_in_progress is zero, then no in-progress
733 * invalidations, including this one, found a relevant memslot at
734 * start(); rechecking memslots here is unnecessary. Note, a false
735 * positive (count elevated by a different invalidation) is sub-optimal
736 * but functionally ok.
738 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
739 if (!READ_ONCE(kvm
->mmu_invalidate_in_progress
))
742 kvm_handle_hva_range(mn
, address
, address
+ 1, arg
, kvm_change_spte_gfn
);
745 void kvm_mmu_invalidate_begin(struct kvm
*kvm
, unsigned long start
,
749 * The count increase must become visible at unlock time as no
750 * spte can be established without taking the mmu_lock and
751 * count is also read inside the mmu_lock critical section.
753 kvm
->mmu_invalidate_in_progress
++;
754 if (likely(kvm
->mmu_invalidate_in_progress
== 1)) {
755 kvm
->mmu_invalidate_range_start
= start
;
756 kvm
->mmu_invalidate_range_end
= end
;
759 * Fully tracking multiple concurrent ranges has diminishing
760 * returns. Keep things simple and just find the minimal range
761 * which includes the current and new ranges. As there won't be
762 * enough information to subtract a range after its invalidate
763 * completes, any ranges invalidated concurrently will
764 * accumulate and persist until all outstanding invalidates
767 kvm
->mmu_invalidate_range_start
=
768 min(kvm
->mmu_invalidate_range_start
, start
);
769 kvm
->mmu_invalidate_range_end
=
770 max(kvm
->mmu_invalidate_range_end
, end
);
774 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
775 const struct mmu_notifier_range
*range
)
777 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
778 const struct kvm_mmu_notifier_range hva_range
= {
779 .start
= range
->start
,
781 .handler
= kvm_unmap_gfn_range
,
782 .on_lock
= kvm_mmu_invalidate_begin
,
783 .on_unlock
= kvm_arch_guest_memory_reclaimed
,
784 .flush_on_ret
= true,
785 .may_block
= mmu_notifier_range_blockable(range
),
788 trace_kvm_unmap_hva_range(range
->start
, range
->end
);
791 * Prevent memslot modification between range_start() and range_end()
792 * so that conditionally locking provides the same result in both
793 * functions. Without that guarantee, the mmu_invalidate_in_progress
794 * adjustments will be imbalanced.
796 * Pairs with the decrement in range_end().
798 spin_lock(&kvm
->mn_invalidate_lock
);
799 kvm
->mn_active_invalidate_count
++;
800 spin_unlock(&kvm
->mn_invalidate_lock
);
803 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
804 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
805 * each cache's lock. There are relatively few caches in existence at
806 * any given time, and the caches themselves can check for hva overlap,
807 * i.e. don't need to rely on memslot overlap checks for performance.
808 * Because this runs without holding mmu_lock, the pfn caches must use
809 * mn_active_invalidate_count (see above) instead of
810 * mmu_invalidate_in_progress.
812 gfn_to_pfn_cache_invalidate_start(kvm
, range
->start
, range
->end
,
813 hva_range
.may_block
);
815 __kvm_handle_hva_range(kvm
, &hva_range
);
820 void kvm_mmu_invalidate_end(struct kvm
*kvm
, unsigned long start
,
824 * This sequence increase will notify the kvm page fault that
825 * the page that is going to be mapped in the spte could have
828 kvm
->mmu_invalidate_seq
++;
831 * The above sequence increase must be visible before the
832 * below count decrease, which is ensured by the smp_wmb above
833 * in conjunction with the smp_rmb in mmu_invalidate_retry().
835 kvm
->mmu_invalidate_in_progress
--;
838 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
839 const struct mmu_notifier_range
*range
)
841 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
842 const struct kvm_mmu_notifier_range hva_range
= {
843 .start
= range
->start
,
845 .handler
= (void *)kvm_null_fn
,
846 .on_lock
= kvm_mmu_invalidate_end
,
847 .on_unlock
= (void *)kvm_null_fn
,
848 .flush_on_ret
= false,
849 .may_block
= mmu_notifier_range_blockable(range
),
853 __kvm_handle_hva_range(kvm
, &hva_range
);
855 /* Pairs with the increment in range_start(). */
856 spin_lock(&kvm
->mn_invalidate_lock
);
857 wake
= (--kvm
->mn_active_invalidate_count
== 0);
858 spin_unlock(&kvm
->mn_invalidate_lock
);
861 * There can only be one waiter, since the wait happens under
865 rcuwait_wake_up(&kvm
->mn_memslots_update_rcuwait
);
867 BUG_ON(kvm
->mmu_invalidate_in_progress
< 0);
870 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
871 struct mm_struct
*mm
,
875 trace_kvm_age_hva(start
, end
);
877 return kvm_handle_hva_range(mn
, start
, end
, KVM_MMU_NOTIFIER_NO_ARG
,
881 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
882 struct mm_struct
*mm
,
886 trace_kvm_age_hva(start
, end
);
889 * Even though we do not flush TLB, this will still adversely
890 * affect performance on pre-Haswell Intel EPT, where there is
891 * no EPT Access Bit to clear so that we have to tear down EPT
892 * tables instead. If we find this unacceptable, we can always
893 * add a parameter to kvm_age_hva so that it effectively doesn't
894 * do anything on clear_young.
896 * Also note that currently we never issue secondary TLB flushes
897 * from clear_young, leaving this job up to the regular system
898 * cadence. If we find this inaccurate, we might come up with a
899 * more sophisticated heuristic later.
901 return kvm_handle_hva_range_no_flush(mn
, start
, end
, kvm_age_gfn
);
904 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
905 struct mm_struct
*mm
,
906 unsigned long address
)
908 trace_kvm_test_age_hva(address
);
910 return kvm_handle_hva_range_no_flush(mn
, address
, address
+ 1,
914 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
915 struct mm_struct
*mm
)
917 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
920 idx
= srcu_read_lock(&kvm
->srcu
);
921 kvm_flush_shadow_all(kvm
);
922 srcu_read_unlock(&kvm
->srcu
, idx
);
925 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
926 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
927 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
928 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
929 .clear_young
= kvm_mmu_notifier_clear_young
,
930 .test_young
= kvm_mmu_notifier_test_young
,
931 .change_pte
= kvm_mmu_notifier_change_pte
,
932 .release
= kvm_mmu_notifier_release
,
935 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
937 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
938 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
941 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
943 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
948 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
950 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
951 static int kvm_pm_notifier_call(struct notifier_block
*bl
,
955 struct kvm
*kvm
= container_of(bl
, struct kvm
, pm_notifier
);
957 return kvm_arch_pm_notifier(kvm
, state
);
960 static void kvm_init_pm_notifier(struct kvm
*kvm
)
962 kvm
->pm_notifier
.notifier_call
= kvm_pm_notifier_call
;
963 /* Suspend KVM before we suspend ftrace, RCU, etc. */
964 kvm
->pm_notifier
.priority
= INT_MAX
;
965 register_pm_notifier(&kvm
->pm_notifier
);
968 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
970 unregister_pm_notifier(&kvm
->pm_notifier
);
972 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
973 static void kvm_init_pm_notifier(struct kvm
*kvm
)
977 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
980 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
982 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
984 if (!memslot
->dirty_bitmap
)
987 kvfree(memslot
->dirty_bitmap
);
988 memslot
->dirty_bitmap
= NULL
;
991 /* This does not remove the slot from struct kvm_memslots data structures */
992 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
994 kvm_destroy_dirty_bitmap(slot
);
996 kvm_arch_free_memslot(kvm
, slot
);
1001 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
1003 struct hlist_node
*idnode
;
1004 struct kvm_memory_slot
*memslot
;
1008 * The same memslot objects live in both active and inactive sets,
1009 * arbitrarily free using index '1' so the second invocation of this
1010 * function isn't operating over a structure with dangling pointers
1011 * (even though this function isn't actually touching them).
1013 if (!slots
->node_idx
)
1016 hash_for_each_safe(slots
->id_hash
, bkt
, idnode
, memslot
, id_node
[1])
1017 kvm_free_memslot(kvm
, memslot
);
1020 static umode_t
kvm_stats_debugfs_mode(const struct _kvm_stats_desc
*pdesc
)
1022 switch (pdesc
->desc
.flags
& KVM_STATS_TYPE_MASK
) {
1023 case KVM_STATS_TYPE_INSTANT
:
1025 case KVM_STATS_TYPE_CUMULATIVE
:
1026 case KVM_STATS_TYPE_PEAK
:
1033 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
1036 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1037 kvm_vcpu_stats_header
.num_desc
;
1039 if (IS_ERR(kvm
->debugfs_dentry
))
1042 debugfs_remove_recursive(kvm
->debugfs_dentry
);
1044 if (kvm
->debugfs_stat_data
) {
1045 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
1046 kfree(kvm
->debugfs_stat_data
[i
]);
1047 kfree(kvm
->debugfs_stat_data
);
1051 static int kvm_create_vm_debugfs(struct kvm
*kvm
, const char *fdname
)
1053 static DEFINE_MUTEX(kvm_debugfs_lock
);
1054 struct dentry
*dent
;
1055 char dir_name
[ITOA_MAX_LEN
* 2];
1056 struct kvm_stat_data
*stat_data
;
1057 const struct _kvm_stats_desc
*pdesc
;
1058 int i
, ret
= -ENOMEM
;
1059 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1060 kvm_vcpu_stats_header
.num_desc
;
1062 if (!debugfs_initialized())
1065 snprintf(dir_name
, sizeof(dir_name
), "%d-%s", task_pid_nr(current
), fdname
);
1066 mutex_lock(&kvm_debugfs_lock
);
1067 dent
= debugfs_lookup(dir_name
, kvm_debugfs_dir
);
1069 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name
);
1071 mutex_unlock(&kvm_debugfs_lock
);
1074 dent
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
1075 mutex_unlock(&kvm_debugfs_lock
);
1079 kvm
->debugfs_dentry
= dent
;
1080 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
1081 sizeof(*kvm
->debugfs_stat_data
),
1082 GFP_KERNEL_ACCOUNT
);
1083 if (!kvm
->debugfs_stat_data
)
1086 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
1087 pdesc
= &kvm_vm_stats_desc
[i
];
1088 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1092 stat_data
->kvm
= kvm
;
1093 stat_data
->desc
= pdesc
;
1094 stat_data
->kind
= KVM_STAT_VM
;
1095 kvm
->debugfs_stat_data
[i
] = stat_data
;
1096 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1097 kvm
->debugfs_dentry
, stat_data
,
1101 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
1102 pdesc
= &kvm_vcpu_stats_desc
[i
];
1103 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1107 stat_data
->kvm
= kvm
;
1108 stat_data
->desc
= pdesc
;
1109 stat_data
->kind
= KVM_STAT_VCPU
;
1110 kvm
->debugfs_stat_data
[i
+ kvm_vm_stats_header
.num_desc
] = stat_data
;
1111 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1112 kvm
->debugfs_dentry
, stat_data
,
1116 ret
= kvm_arch_create_vm_debugfs(kvm
);
1122 kvm_destroy_vm_debugfs(kvm
);
1127 * Called after the VM is otherwise initialized, but just before adding it to
1130 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
1136 * Called just after removing the VM from the vm_list, but before doing any
1137 * other destruction.
1139 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
1144 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1145 * be setup already, so we can create arch-specific debugfs entries under it.
1146 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1147 * a per-arch destroy interface is not needed.
1149 int __weak
kvm_arch_create_vm_debugfs(struct kvm
*kvm
)
1154 static struct kvm
*kvm_create_vm(unsigned long type
, const char *fdname
)
1156 struct kvm
*kvm
= kvm_arch_alloc_vm();
1157 struct kvm_memslots
*slots
;
1162 return ERR_PTR(-ENOMEM
);
1164 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1165 __module_get(kvm_chardev_ops
.owner
);
1167 KVM_MMU_LOCK_INIT(kvm
);
1168 mmgrab(current
->mm
);
1169 kvm
->mm
= current
->mm
;
1170 kvm_eventfd_init(kvm
);
1171 mutex_init(&kvm
->lock
);
1172 mutex_init(&kvm
->irq_lock
);
1173 mutex_init(&kvm
->slots_lock
);
1174 mutex_init(&kvm
->slots_arch_lock
);
1175 spin_lock_init(&kvm
->mn_invalidate_lock
);
1176 rcuwait_init(&kvm
->mn_memslots_update_rcuwait
);
1177 xa_init(&kvm
->vcpu_array
);
1179 INIT_LIST_HEAD(&kvm
->gpc_list
);
1180 spin_lock_init(&kvm
->gpc_lock
);
1182 INIT_LIST_HEAD(&kvm
->devices
);
1183 kvm
->max_vcpus
= KVM_MAX_VCPUS
;
1185 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
1188 * Force subsequent debugfs file creations to fail if the VM directory
1189 * is not created (by kvm_create_vm_debugfs()).
1191 kvm
->debugfs_dentry
= ERR_PTR(-ENOENT
);
1193 snprintf(kvm
->stats_id
, sizeof(kvm
->stats_id
), "kvm-%d",
1194 task_pid_nr(current
));
1196 if (init_srcu_struct(&kvm
->srcu
))
1197 goto out_err_no_srcu
;
1198 if (init_srcu_struct(&kvm
->irq_srcu
))
1199 goto out_err_no_irq_srcu
;
1201 refcount_set(&kvm
->users_count
, 1);
1202 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
1203 for (j
= 0; j
< 2; j
++) {
1204 slots
= &kvm
->__memslots
[i
][j
];
1206 atomic_long_set(&slots
->last_used_slot
, (unsigned long)NULL
);
1207 slots
->hva_tree
= RB_ROOT_CACHED
;
1208 slots
->gfn_tree
= RB_ROOT
;
1209 hash_init(slots
->id_hash
);
1210 slots
->node_idx
= j
;
1212 /* Generations must be different for each address space. */
1213 slots
->generation
= i
;
1216 rcu_assign_pointer(kvm
->memslots
[i
], &kvm
->__memslots
[i
][0]);
1219 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1220 rcu_assign_pointer(kvm
->buses
[i
],
1221 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
1223 goto out_err_no_arch_destroy_vm
;
1226 r
= kvm_arch_init_vm(kvm
, type
);
1228 goto out_err_no_arch_destroy_vm
;
1230 r
= hardware_enable_all();
1232 goto out_err_no_disable
;
1234 #ifdef CONFIG_HAVE_KVM_IRQFD
1235 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
1238 r
= kvm_init_mmu_notifier(kvm
);
1240 goto out_err_no_mmu_notifier
;
1242 r
= kvm_coalesced_mmio_init(kvm
);
1244 goto out_no_coalesced_mmio
;
1246 r
= kvm_create_vm_debugfs(kvm
, fdname
);
1248 goto out_err_no_debugfs
;
1250 r
= kvm_arch_post_init_vm(kvm
);
1254 mutex_lock(&kvm_lock
);
1255 list_add(&kvm
->vm_list
, &vm_list
);
1256 mutex_unlock(&kvm_lock
);
1258 preempt_notifier_inc();
1259 kvm_init_pm_notifier(kvm
);
1264 kvm_destroy_vm_debugfs(kvm
);
1266 kvm_coalesced_mmio_free(kvm
);
1267 out_no_coalesced_mmio
:
1268 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1269 if (kvm
->mmu_notifier
.ops
)
1270 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
1272 out_err_no_mmu_notifier
:
1273 hardware_disable_all();
1275 kvm_arch_destroy_vm(kvm
);
1276 out_err_no_arch_destroy_vm
:
1277 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
1278 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
1279 kfree(kvm_get_bus(kvm
, i
));
1280 cleanup_srcu_struct(&kvm
->irq_srcu
);
1281 out_err_no_irq_srcu
:
1282 cleanup_srcu_struct(&kvm
->srcu
);
1284 kvm_arch_free_vm(kvm
);
1285 mmdrop(current
->mm
);
1286 module_put(kvm_chardev_ops
.owner
);
1290 static void kvm_destroy_devices(struct kvm
*kvm
)
1292 struct kvm_device
*dev
, *tmp
;
1295 * We do not need to take the kvm->lock here, because nobody else
1296 * has a reference to the struct kvm at this point and therefore
1297 * cannot access the devices list anyhow.
1299 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
1300 list_del(&dev
->vm_node
);
1301 dev
->ops
->destroy(dev
);
1305 static void kvm_destroy_vm(struct kvm
*kvm
)
1308 struct mm_struct
*mm
= kvm
->mm
;
1310 kvm_destroy_pm_notifier(kvm
);
1311 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
1312 kvm_destroy_vm_debugfs(kvm
);
1313 kvm_arch_sync_events(kvm
);
1314 mutex_lock(&kvm_lock
);
1315 list_del(&kvm
->vm_list
);
1316 mutex_unlock(&kvm_lock
);
1317 kvm_arch_pre_destroy_vm(kvm
);
1319 kvm_free_irq_routing(kvm
);
1320 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1321 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
1324 kvm_io_bus_destroy(bus
);
1325 kvm
->buses
[i
] = NULL
;
1327 kvm_coalesced_mmio_free(kvm
);
1328 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1329 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1331 * At this point, pending calls to invalidate_range_start()
1332 * have completed but no more MMU notifiers will run, so
1333 * mn_active_invalidate_count may remain unbalanced.
1334 * No threads can be waiting in kvm_swap_active_memslots() as the
1335 * last reference on KVM has been dropped, but freeing
1336 * memslots would deadlock without this manual intervention.
1338 WARN_ON(rcuwait_active(&kvm
->mn_memslots_update_rcuwait
));
1339 kvm
->mn_active_invalidate_count
= 0;
1341 kvm_flush_shadow_all(kvm
);
1343 kvm_arch_destroy_vm(kvm
);
1344 kvm_destroy_devices(kvm
);
1345 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
1346 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][0]);
1347 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][1]);
1349 cleanup_srcu_struct(&kvm
->irq_srcu
);
1350 cleanup_srcu_struct(&kvm
->srcu
);
1351 kvm_arch_free_vm(kvm
);
1352 preempt_notifier_dec();
1353 hardware_disable_all();
1355 module_put(kvm_chardev_ops
.owner
);
1358 void kvm_get_kvm(struct kvm
*kvm
)
1360 refcount_inc(&kvm
->users_count
);
1362 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1365 * Make sure the vm is not during destruction, which is a safe version of
1366 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1368 bool kvm_get_kvm_safe(struct kvm
*kvm
)
1370 return refcount_inc_not_zero(&kvm
->users_count
);
1372 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe
);
1374 void kvm_put_kvm(struct kvm
*kvm
)
1376 if (refcount_dec_and_test(&kvm
->users_count
))
1377 kvm_destroy_vm(kvm
);
1379 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1382 * Used to put a reference that was taken on behalf of an object associated
1383 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1384 * of the new file descriptor fails and the reference cannot be transferred to
1385 * its final owner. In such cases, the caller is still actively using @kvm and
1386 * will fail miserably if the refcount unexpectedly hits zero.
1388 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
1390 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
1392 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
1394 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1396 struct kvm
*kvm
= filp
->private_data
;
1398 kvm_irqfd_release(kvm
);
1405 * Allocation size is twice as large as the actual dirty bitmap size.
1406 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1408 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1410 unsigned long dirty_bytes
= kvm_dirty_bitmap_bytes(memslot
);
1412 memslot
->dirty_bitmap
= __vcalloc(2, dirty_bytes
, GFP_KERNEL_ACCOUNT
);
1413 if (!memslot
->dirty_bitmap
)
1419 static struct kvm_memslots
*kvm_get_inactive_memslots(struct kvm
*kvm
, int as_id
)
1421 struct kvm_memslots
*active
= __kvm_memslots(kvm
, as_id
);
1422 int node_idx_inactive
= active
->node_idx
^ 1;
1424 return &kvm
->__memslots
[as_id
][node_idx_inactive
];
1428 * Helper to get the address space ID when one of memslot pointers may be NULL.
1429 * This also serves as a sanity that at least one of the pointers is non-NULL,
1430 * and that their address space IDs don't diverge.
1432 static int kvm_memslots_get_as_id(struct kvm_memory_slot
*a
,
1433 struct kvm_memory_slot
*b
)
1435 if (WARN_ON_ONCE(!a
&& !b
))
1443 WARN_ON_ONCE(a
->as_id
!= b
->as_id
);
1447 static void kvm_insert_gfn_node(struct kvm_memslots
*slots
,
1448 struct kvm_memory_slot
*slot
)
1450 struct rb_root
*gfn_tree
= &slots
->gfn_tree
;
1451 struct rb_node
**node
, *parent
;
1452 int idx
= slots
->node_idx
;
1455 for (node
= &gfn_tree
->rb_node
; *node
; ) {
1456 struct kvm_memory_slot
*tmp
;
1458 tmp
= container_of(*node
, struct kvm_memory_slot
, gfn_node
[idx
]);
1460 if (slot
->base_gfn
< tmp
->base_gfn
)
1461 node
= &(*node
)->rb_left
;
1462 else if (slot
->base_gfn
> tmp
->base_gfn
)
1463 node
= &(*node
)->rb_right
;
1468 rb_link_node(&slot
->gfn_node
[idx
], parent
, node
);
1469 rb_insert_color(&slot
->gfn_node
[idx
], gfn_tree
);
1472 static void kvm_erase_gfn_node(struct kvm_memslots
*slots
,
1473 struct kvm_memory_slot
*slot
)
1475 rb_erase(&slot
->gfn_node
[slots
->node_idx
], &slots
->gfn_tree
);
1478 static void kvm_replace_gfn_node(struct kvm_memslots
*slots
,
1479 struct kvm_memory_slot
*old
,
1480 struct kvm_memory_slot
*new)
1482 int idx
= slots
->node_idx
;
1484 WARN_ON_ONCE(old
->base_gfn
!= new->base_gfn
);
1486 rb_replace_node(&old
->gfn_node
[idx
], &new->gfn_node
[idx
],
1491 * Replace @old with @new in the inactive memslots.
1493 * With NULL @old this simply adds @new.
1494 * With NULL @new this simply removes @old.
1496 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1499 static void kvm_replace_memslot(struct kvm
*kvm
,
1500 struct kvm_memory_slot
*old
,
1501 struct kvm_memory_slot
*new)
1503 int as_id
= kvm_memslots_get_as_id(old
, new);
1504 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1505 int idx
= slots
->node_idx
;
1508 hash_del(&old
->id_node
[idx
]);
1509 interval_tree_remove(&old
->hva_node
[idx
], &slots
->hva_tree
);
1511 if ((long)old
== atomic_long_read(&slots
->last_used_slot
))
1512 atomic_long_set(&slots
->last_used_slot
, (long)new);
1515 kvm_erase_gfn_node(slots
, old
);
1521 * Initialize @new's hva range. Do this even when replacing an @old
1522 * slot, kvm_copy_memslot() deliberately does not touch node data.
1524 new->hva_node
[idx
].start
= new->userspace_addr
;
1525 new->hva_node
[idx
].last
= new->userspace_addr
+
1526 (new->npages
<< PAGE_SHIFT
) - 1;
1529 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1530 * hva_node needs to be swapped with remove+insert even though hva can't
1531 * change when replacing an existing slot.
1533 hash_add(slots
->id_hash
, &new->id_node
[idx
], new->id
);
1534 interval_tree_insert(&new->hva_node
[idx
], &slots
->hva_tree
);
1537 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1538 * switch the node in the gfn tree instead of removing the old and
1539 * inserting the new as two separate operations. Replacement is a
1540 * single O(1) operation versus two O(log(n)) operations for
1543 if (old
&& old
->base_gfn
== new->base_gfn
) {
1544 kvm_replace_gfn_node(slots
, old
, new);
1547 kvm_erase_gfn_node(slots
, old
);
1548 kvm_insert_gfn_node(slots
, new);
1552 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
1554 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1556 #ifdef __KVM_HAVE_READONLY_MEM
1557 valid_flags
|= KVM_MEM_READONLY
;
1560 if (mem
->flags
& ~valid_flags
)
1566 static void kvm_swap_active_memslots(struct kvm
*kvm
, int as_id
)
1568 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1570 /* Grab the generation from the activate memslots. */
1571 u64 gen
= __kvm_memslots(kvm
, as_id
)->generation
;
1573 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1574 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1577 * Do not store the new memslots while there are invalidations in
1578 * progress, otherwise the locking in invalidate_range_start and
1579 * invalidate_range_end will be unbalanced.
1581 spin_lock(&kvm
->mn_invalidate_lock
);
1582 prepare_to_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1583 while (kvm
->mn_active_invalidate_count
) {
1584 set_current_state(TASK_UNINTERRUPTIBLE
);
1585 spin_unlock(&kvm
->mn_invalidate_lock
);
1587 spin_lock(&kvm
->mn_invalidate_lock
);
1589 finish_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1590 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1591 spin_unlock(&kvm
->mn_invalidate_lock
);
1594 * Acquired in kvm_set_memslot. Must be released before synchronize
1595 * SRCU below in order to avoid deadlock with another thread
1596 * acquiring the slots_arch_lock in an srcu critical section.
1598 mutex_unlock(&kvm
->slots_arch_lock
);
1600 synchronize_srcu_expedited(&kvm
->srcu
);
1603 * Increment the new memslot generation a second time, dropping the
1604 * update in-progress flag and incrementing the generation based on
1605 * the number of address spaces. This provides a unique and easily
1606 * identifiable generation number while the memslots are in flux.
1608 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1611 * Generations must be unique even across address spaces. We do not need
1612 * a global counter for that, instead the generation space is evenly split
1613 * across address spaces. For example, with two address spaces, address
1614 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1615 * use generations 1, 3, 5, ...
1617 gen
+= KVM_ADDRESS_SPACE_NUM
;
1619 kvm_arch_memslots_updated(kvm
, gen
);
1621 slots
->generation
= gen
;
1624 static int kvm_prepare_memory_region(struct kvm
*kvm
,
1625 const struct kvm_memory_slot
*old
,
1626 struct kvm_memory_slot
*new,
1627 enum kvm_mr_change change
)
1632 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1633 * will be freed on "commit". If logging is enabled in both old and
1634 * new, reuse the existing bitmap. If logging is enabled only in the
1635 * new and KVM isn't using a ring buffer, allocate and initialize a
1638 if (change
!= KVM_MR_DELETE
) {
1639 if (!(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1640 new->dirty_bitmap
= NULL
;
1641 else if (old
&& old
->dirty_bitmap
)
1642 new->dirty_bitmap
= old
->dirty_bitmap
;
1643 else if (kvm_use_dirty_bitmap(kvm
)) {
1644 r
= kvm_alloc_dirty_bitmap(new);
1648 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1649 bitmap_set(new->dirty_bitmap
, 0, new->npages
);
1653 r
= kvm_arch_prepare_memory_region(kvm
, old
, new, change
);
1655 /* Free the bitmap on failure if it was allocated above. */
1656 if (r
&& new && new->dirty_bitmap
&& (!old
|| !old
->dirty_bitmap
))
1657 kvm_destroy_dirty_bitmap(new);
1662 static void kvm_commit_memory_region(struct kvm
*kvm
,
1663 struct kvm_memory_slot
*old
,
1664 const struct kvm_memory_slot
*new,
1665 enum kvm_mr_change change
)
1667 int old_flags
= old
? old
->flags
: 0;
1668 int new_flags
= new ? new->flags
: 0;
1670 * Update the total number of memslot pages before calling the arch
1671 * hook so that architectures can consume the result directly.
1673 if (change
== KVM_MR_DELETE
)
1674 kvm
->nr_memslot_pages
-= old
->npages
;
1675 else if (change
== KVM_MR_CREATE
)
1676 kvm
->nr_memslot_pages
+= new->npages
;
1678 if ((old_flags
^ new_flags
) & KVM_MEM_LOG_DIRTY_PAGES
) {
1679 int change
= (new_flags
& KVM_MEM_LOG_DIRTY_PAGES
) ? 1 : -1;
1680 atomic_set(&kvm
->nr_memslots_dirty_logging
,
1681 atomic_read(&kvm
->nr_memslots_dirty_logging
) + change
);
1684 kvm_arch_commit_memory_region(kvm
, old
, new, change
);
1688 /* Nothing more to do. */
1691 /* Free the old memslot and all its metadata. */
1692 kvm_free_memslot(kvm
, old
);
1695 case KVM_MR_FLAGS_ONLY
:
1697 * Free the dirty bitmap as needed; the below check encompasses
1698 * both the flags and whether a ring buffer is being used)
1700 if (old
->dirty_bitmap
&& !new->dirty_bitmap
)
1701 kvm_destroy_dirty_bitmap(old
);
1704 * The final quirk. Free the detached, old slot, but only its
1705 * memory, not any metadata. Metadata, including arch specific
1706 * data, may be reused by @new.
1716 * Activate @new, which must be installed in the inactive slots by the caller,
1717 * by swapping the active slots and then propagating @new to @old once @old is
1718 * unreachable and can be safely modified.
1720 * With NULL @old this simply adds @new to @active (while swapping the sets).
1721 * With NULL @new this simply removes @old from @active and frees it
1722 * (while also swapping the sets).
1724 static void kvm_activate_memslot(struct kvm
*kvm
,
1725 struct kvm_memory_slot
*old
,
1726 struct kvm_memory_slot
*new)
1728 int as_id
= kvm_memslots_get_as_id(old
, new);
1730 kvm_swap_active_memslots(kvm
, as_id
);
1732 /* Propagate the new memslot to the now inactive memslots. */
1733 kvm_replace_memslot(kvm
, old
, new);
1736 static void kvm_copy_memslot(struct kvm_memory_slot
*dest
,
1737 const struct kvm_memory_slot
*src
)
1739 dest
->base_gfn
= src
->base_gfn
;
1740 dest
->npages
= src
->npages
;
1741 dest
->dirty_bitmap
= src
->dirty_bitmap
;
1742 dest
->arch
= src
->arch
;
1743 dest
->userspace_addr
= src
->userspace_addr
;
1744 dest
->flags
= src
->flags
;
1746 dest
->as_id
= src
->as_id
;
1749 static void kvm_invalidate_memslot(struct kvm
*kvm
,
1750 struct kvm_memory_slot
*old
,
1751 struct kvm_memory_slot
*invalid_slot
)
1754 * Mark the current slot INVALID. As with all memslot modifications,
1755 * this must be done on an unreachable slot to avoid modifying the
1756 * current slot in the active tree.
1758 kvm_copy_memslot(invalid_slot
, old
);
1759 invalid_slot
->flags
|= KVM_MEMSLOT_INVALID
;
1760 kvm_replace_memslot(kvm
, old
, invalid_slot
);
1763 * Activate the slot that is now marked INVALID, but don't propagate
1764 * the slot to the now inactive slots. The slot is either going to be
1765 * deleted or recreated as a new slot.
1767 kvm_swap_active_memslots(kvm
, old
->as_id
);
1770 * From this point no new shadow pages pointing to a deleted, or moved,
1771 * memslot will be created. Validation of sp->gfn happens in:
1772 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1773 * - kvm_is_visible_gfn (mmu_check_root)
1775 kvm_arch_flush_shadow_memslot(kvm
, old
);
1776 kvm_arch_guest_memory_reclaimed(kvm
);
1778 /* Was released by kvm_swap_active_memslots(), reacquire. */
1779 mutex_lock(&kvm
->slots_arch_lock
);
1782 * Copy the arch-specific field of the newly-installed slot back to the
1783 * old slot as the arch data could have changed between releasing
1784 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1785 * above. Writers are required to retrieve memslots *after* acquiring
1786 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1788 old
->arch
= invalid_slot
->arch
;
1791 static void kvm_create_memslot(struct kvm
*kvm
,
1792 struct kvm_memory_slot
*new)
1794 /* Add the new memslot to the inactive set and activate. */
1795 kvm_replace_memslot(kvm
, NULL
, new);
1796 kvm_activate_memslot(kvm
, NULL
, new);
1799 static void kvm_delete_memslot(struct kvm
*kvm
,
1800 struct kvm_memory_slot
*old
,
1801 struct kvm_memory_slot
*invalid_slot
)
1804 * Remove the old memslot (in the inactive memslots) by passing NULL as
1805 * the "new" slot, and for the invalid version in the active slots.
1807 kvm_replace_memslot(kvm
, old
, NULL
);
1808 kvm_activate_memslot(kvm
, invalid_slot
, NULL
);
1811 static void kvm_move_memslot(struct kvm
*kvm
,
1812 struct kvm_memory_slot
*old
,
1813 struct kvm_memory_slot
*new,
1814 struct kvm_memory_slot
*invalid_slot
)
1817 * Replace the old memslot in the inactive slots, and then swap slots
1818 * and replace the current INVALID with the new as well.
1820 kvm_replace_memslot(kvm
, old
, new);
1821 kvm_activate_memslot(kvm
, invalid_slot
, new);
1824 static void kvm_update_flags_memslot(struct kvm
*kvm
,
1825 struct kvm_memory_slot
*old
,
1826 struct kvm_memory_slot
*new)
1829 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1830 * an intermediate step. Instead, the old memslot is simply replaced
1831 * with a new, updated copy in both memslot sets.
1833 kvm_replace_memslot(kvm
, old
, new);
1834 kvm_activate_memslot(kvm
, old
, new);
1837 static int kvm_set_memslot(struct kvm
*kvm
,
1838 struct kvm_memory_slot
*old
,
1839 struct kvm_memory_slot
*new,
1840 enum kvm_mr_change change
)
1842 struct kvm_memory_slot
*invalid_slot
;
1846 * Released in kvm_swap_active_memslots().
1848 * Must be held from before the current memslots are copied until after
1849 * the new memslots are installed with rcu_assign_pointer, then
1850 * released before the synchronize srcu in kvm_swap_active_memslots().
1852 * When modifying memslots outside of the slots_lock, must be held
1853 * before reading the pointer to the current memslots until after all
1854 * changes to those memslots are complete.
1856 * These rules ensure that installing new memslots does not lose
1857 * changes made to the previous memslots.
1859 mutex_lock(&kvm
->slots_arch_lock
);
1862 * Invalidate the old slot if it's being deleted or moved. This is
1863 * done prior to actually deleting/moving the memslot to allow vCPUs to
1864 * continue running by ensuring there are no mappings or shadow pages
1865 * for the memslot when it is deleted/moved. Without pre-invalidation
1866 * (and without a lock), a window would exist between effecting the
1867 * delete/move and committing the changes in arch code where KVM or a
1868 * guest could access a non-existent memslot.
1870 * Modifications are done on a temporary, unreachable slot. The old
1871 * slot needs to be preserved in case a later step fails and the
1872 * invalidation needs to be reverted.
1874 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1875 invalid_slot
= kzalloc(sizeof(*invalid_slot
), GFP_KERNEL_ACCOUNT
);
1876 if (!invalid_slot
) {
1877 mutex_unlock(&kvm
->slots_arch_lock
);
1880 kvm_invalidate_memslot(kvm
, old
, invalid_slot
);
1883 r
= kvm_prepare_memory_region(kvm
, old
, new, change
);
1886 * For DELETE/MOVE, revert the above INVALID change. No
1887 * modifications required since the original slot was preserved
1888 * in the inactive slots. Changing the active memslots also
1889 * release slots_arch_lock.
1891 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1892 kvm_activate_memslot(kvm
, invalid_slot
, old
);
1893 kfree(invalid_slot
);
1895 mutex_unlock(&kvm
->slots_arch_lock
);
1901 * For DELETE and MOVE, the working slot is now active as the INVALID
1902 * version of the old slot. MOVE is particularly special as it reuses
1903 * the old slot and returns a copy of the old slot (in working_slot).
1904 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1905 * old slot is detached but otherwise preserved.
1907 if (change
== KVM_MR_CREATE
)
1908 kvm_create_memslot(kvm
, new);
1909 else if (change
== KVM_MR_DELETE
)
1910 kvm_delete_memslot(kvm
, old
, invalid_slot
);
1911 else if (change
== KVM_MR_MOVE
)
1912 kvm_move_memslot(kvm
, old
, new, invalid_slot
);
1913 else if (change
== KVM_MR_FLAGS_ONLY
)
1914 kvm_update_flags_memslot(kvm
, old
, new);
1918 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1919 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1920 kfree(invalid_slot
);
1923 * No need to refresh new->arch, changes after dropping slots_arch_lock
1924 * will directly hit the final, active memslot. Architectures are
1925 * responsible for knowing that new->arch may be stale.
1927 kvm_commit_memory_region(kvm
, old
, new, change
);
1932 static bool kvm_check_memslot_overlap(struct kvm_memslots
*slots
, int id
,
1933 gfn_t start
, gfn_t end
)
1935 struct kvm_memslot_iter iter
;
1937 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, start
, end
) {
1938 if (iter
.slot
->id
!= id
)
1946 * Allocate some memory and give it an address in the guest physical address
1949 * Discontiguous memory is allowed, mostly for framebuffers.
1951 * Must be called holding kvm->slots_lock for write.
1953 int __kvm_set_memory_region(struct kvm
*kvm
,
1954 const struct kvm_userspace_memory_region
*mem
)
1956 struct kvm_memory_slot
*old
, *new;
1957 struct kvm_memslots
*slots
;
1958 enum kvm_mr_change change
;
1959 unsigned long npages
;
1964 r
= check_memory_region_flags(mem
);
1968 as_id
= mem
->slot
>> 16;
1969 id
= (u16
)mem
->slot
;
1971 /* General sanity checks */
1972 if ((mem
->memory_size
& (PAGE_SIZE
- 1)) ||
1973 (mem
->memory_size
!= (unsigned long)mem
->memory_size
))
1975 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
1977 /* We can read the guest memory with __xxx_user() later on. */
1978 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
1979 (mem
->userspace_addr
!= untagged_addr(mem
->userspace_addr
)) ||
1980 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
1983 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
1985 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
1987 if ((mem
->memory_size
>> PAGE_SHIFT
) > KVM_MEM_MAX_NR_PAGES
)
1990 slots
= __kvm_memslots(kvm
, as_id
);
1993 * Note, the old memslot (and the pointer itself!) may be invalidated
1994 * and/or destroyed by kvm_set_memslot().
1996 old
= id_to_memslot(slots
, id
);
1998 if (!mem
->memory_size
) {
1999 if (!old
|| !old
->npages
)
2002 if (WARN_ON_ONCE(kvm
->nr_memslot_pages
< old
->npages
))
2005 return kvm_set_memslot(kvm
, old
, NULL
, KVM_MR_DELETE
);
2008 base_gfn
= (mem
->guest_phys_addr
>> PAGE_SHIFT
);
2009 npages
= (mem
->memory_size
>> PAGE_SHIFT
);
2011 if (!old
|| !old
->npages
) {
2012 change
= KVM_MR_CREATE
;
2015 * To simplify KVM internals, the total number of pages across
2016 * all memslots must fit in an unsigned long.
2018 if ((kvm
->nr_memslot_pages
+ npages
) < kvm
->nr_memslot_pages
)
2020 } else { /* Modify an existing slot. */
2021 if ((mem
->userspace_addr
!= old
->userspace_addr
) ||
2022 (npages
!= old
->npages
) ||
2023 ((mem
->flags
^ old
->flags
) & KVM_MEM_READONLY
))
2026 if (base_gfn
!= old
->base_gfn
)
2027 change
= KVM_MR_MOVE
;
2028 else if (mem
->flags
!= old
->flags
)
2029 change
= KVM_MR_FLAGS_ONLY
;
2030 else /* Nothing to change. */
2034 if ((change
== KVM_MR_CREATE
|| change
== KVM_MR_MOVE
) &&
2035 kvm_check_memslot_overlap(slots
, id
, base_gfn
, base_gfn
+ npages
))
2038 /* Allocate a slot that will persist in the memslot. */
2039 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT
);
2045 new->base_gfn
= base_gfn
;
2046 new->npages
= npages
;
2047 new->flags
= mem
->flags
;
2048 new->userspace_addr
= mem
->userspace_addr
;
2050 r
= kvm_set_memslot(kvm
, old
, new, change
);
2055 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
2057 int kvm_set_memory_region(struct kvm
*kvm
,
2058 const struct kvm_userspace_memory_region
*mem
)
2062 mutex_lock(&kvm
->slots_lock
);
2063 r
= __kvm_set_memory_region(kvm
, mem
);
2064 mutex_unlock(&kvm
->slots_lock
);
2067 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
2069 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
2070 struct kvm_userspace_memory_region
*mem
)
2072 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
2075 return kvm_set_memory_region(kvm
, mem
);
2078 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2080 * kvm_get_dirty_log - get a snapshot of dirty pages
2081 * @kvm: pointer to kvm instance
2082 * @log: slot id and address to which we copy the log
2083 * @is_dirty: set to '1' if any dirty pages were found
2084 * @memslot: set to the associated memslot, always valid on success
2086 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
2087 int *is_dirty
, struct kvm_memory_slot
**memslot
)
2089 struct kvm_memslots
*slots
;
2092 unsigned long any
= 0;
2094 /* Dirty ring tracking may be exclusive to dirty log tracking */
2095 if (!kvm_use_dirty_bitmap(kvm
))
2101 as_id
= log
->slot
>> 16;
2102 id
= (u16
)log
->slot
;
2103 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2106 slots
= __kvm_memslots(kvm
, as_id
);
2107 *memslot
= id_to_memslot(slots
, id
);
2108 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
2111 kvm_arch_sync_dirty_log(kvm
, *memslot
);
2113 n
= kvm_dirty_bitmap_bytes(*memslot
);
2115 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
2116 any
= (*memslot
)->dirty_bitmap
[i
];
2118 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
2125 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
2127 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2129 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2130 * and reenable dirty page tracking for the corresponding pages.
2131 * @kvm: pointer to kvm instance
2132 * @log: slot id and address to which we copy the log
2134 * We need to keep it in mind that VCPU threads can write to the bitmap
2135 * concurrently. So, to avoid losing track of dirty pages we keep the
2138 * 1. Take a snapshot of the bit and clear it if needed.
2139 * 2. Write protect the corresponding page.
2140 * 3. Copy the snapshot to the userspace.
2141 * 4. Upon return caller flushes TLB's if needed.
2143 * Between 2 and 4, the guest may write to the page using the remaining TLB
2144 * entry. This is not a problem because the page is reported dirty using
2145 * the snapshot taken before and step 4 ensures that writes done after
2146 * exiting to userspace will be logged for the next call.
2149 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
2151 struct kvm_memslots
*slots
;
2152 struct kvm_memory_slot
*memslot
;
2155 unsigned long *dirty_bitmap
;
2156 unsigned long *dirty_bitmap_buffer
;
2159 /* Dirty ring tracking may be exclusive to dirty log tracking */
2160 if (!kvm_use_dirty_bitmap(kvm
))
2163 as_id
= log
->slot
>> 16;
2164 id
= (u16
)log
->slot
;
2165 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2168 slots
= __kvm_memslots(kvm
, as_id
);
2169 memslot
= id_to_memslot(slots
, id
);
2170 if (!memslot
|| !memslot
->dirty_bitmap
)
2173 dirty_bitmap
= memslot
->dirty_bitmap
;
2175 kvm_arch_sync_dirty_log(kvm
, memslot
);
2177 n
= kvm_dirty_bitmap_bytes(memslot
);
2179 if (kvm
->manual_dirty_log_protect
) {
2181 * Unlike kvm_get_dirty_log, we always return false in *flush,
2182 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2183 * is some code duplication between this function and
2184 * kvm_get_dirty_log, but hopefully all architecture
2185 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2186 * can be eliminated.
2188 dirty_bitmap_buffer
= dirty_bitmap
;
2190 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2191 memset(dirty_bitmap_buffer
, 0, n
);
2194 for (i
= 0; i
< n
/ sizeof(long); i
++) {
2198 if (!dirty_bitmap
[i
])
2202 mask
= xchg(&dirty_bitmap
[i
], 0);
2203 dirty_bitmap_buffer
[i
] = mask
;
2205 offset
= i
* BITS_PER_LONG
;
2206 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2209 KVM_MMU_UNLOCK(kvm
);
2213 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2215 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
2222 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2223 * @kvm: kvm instance
2224 * @log: slot id and address to which we copy the log
2226 * Steps 1-4 below provide general overview of dirty page logging. See
2227 * kvm_get_dirty_log_protect() function description for additional details.
2229 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2230 * always flush the TLB (step 4) even if previous step failed and the dirty
2231 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2232 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2233 * writes will be marked dirty for next log read.
2235 * 1. Take a snapshot of the bit and clear it if needed.
2236 * 2. Write protect the corresponding page.
2237 * 3. Copy the snapshot to the userspace.
2238 * 4. Flush TLB's if needed.
2240 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2241 struct kvm_dirty_log
*log
)
2245 mutex_lock(&kvm
->slots_lock
);
2247 r
= kvm_get_dirty_log_protect(kvm
, log
);
2249 mutex_unlock(&kvm
->slots_lock
);
2254 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2255 * and reenable dirty page tracking for the corresponding pages.
2256 * @kvm: pointer to kvm instance
2257 * @log: slot id and address from which to fetch the bitmap of dirty pages
2259 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
2260 struct kvm_clear_dirty_log
*log
)
2262 struct kvm_memslots
*slots
;
2263 struct kvm_memory_slot
*memslot
;
2267 unsigned long *dirty_bitmap
;
2268 unsigned long *dirty_bitmap_buffer
;
2271 /* Dirty ring tracking may be exclusive to dirty log tracking */
2272 if (!kvm_use_dirty_bitmap(kvm
))
2275 as_id
= log
->slot
>> 16;
2276 id
= (u16
)log
->slot
;
2277 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2280 if (log
->first_page
& 63)
2283 slots
= __kvm_memslots(kvm
, as_id
);
2284 memslot
= id_to_memslot(slots
, id
);
2285 if (!memslot
|| !memslot
->dirty_bitmap
)
2288 dirty_bitmap
= memslot
->dirty_bitmap
;
2290 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
2292 if (log
->first_page
> memslot
->npages
||
2293 log
->num_pages
> memslot
->npages
- log
->first_page
||
2294 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
2297 kvm_arch_sync_dirty_log(kvm
, memslot
);
2300 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2301 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
2305 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
2306 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
2307 i
++, offset
+= BITS_PER_LONG
) {
2308 unsigned long mask
= *dirty_bitmap_buffer
++;
2309 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
2313 mask
&= atomic_long_fetch_andnot(mask
, p
);
2316 * mask contains the bits that really have been cleared. This
2317 * never includes any bits beyond the length of the memslot (if
2318 * the length is not aligned to 64 pages), therefore it is not
2319 * a problem if userspace sets them in log->dirty_bitmap.
2323 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2327 KVM_MMU_UNLOCK(kvm
);
2330 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2335 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
2336 struct kvm_clear_dirty_log
*log
)
2340 mutex_lock(&kvm
->slots_lock
);
2342 r
= kvm_clear_dirty_log_protect(kvm
, log
);
2344 mutex_unlock(&kvm
->slots_lock
);
2347 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2349 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
2351 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
2353 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
2355 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2357 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2358 u64 gen
= slots
->generation
;
2359 struct kvm_memory_slot
*slot
;
2362 * This also protects against using a memslot from a different address space,
2363 * since different address spaces have different generation numbers.
2365 if (unlikely(gen
!= vcpu
->last_used_slot_gen
)) {
2366 vcpu
->last_used_slot
= NULL
;
2367 vcpu
->last_used_slot_gen
= gen
;
2370 slot
= try_get_memslot(vcpu
->last_used_slot
, gfn
);
2375 * Fall back to searching all memslots. We purposely use
2376 * search_memslots() instead of __gfn_to_memslot() to avoid
2377 * thrashing the VM-wide last_used_slot in kvm_memslots.
2379 slot
= search_memslots(slots
, gfn
, false);
2381 vcpu
->last_used_slot
= slot
;
2388 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
2390 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
2392 return kvm_is_visible_memslot(memslot
);
2394 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
2396 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2398 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2400 return kvm_is_visible_memslot(memslot
);
2402 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
2404 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2406 struct vm_area_struct
*vma
;
2407 unsigned long addr
, size
;
2411 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
2412 if (kvm_is_error_hva(addr
))
2415 mmap_read_lock(current
->mm
);
2416 vma
= find_vma(current
->mm
, addr
);
2420 size
= vma_kernel_pagesize(vma
);
2423 mmap_read_unlock(current
->mm
);
2428 static bool memslot_is_readonly(const struct kvm_memory_slot
*slot
)
2430 return slot
->flags
& KVM_MEM_READONLY
;
2433 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2434 gfn_t
*nr_pages
, bool write
)
2436 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
2437 return KVM_HVA_ERR_BAD
;
2439 if (memslot_is_readonly(slot
) && write
)
2440 return KVM_HVA_ERR_RO_BAD
;
2443 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
2445 return __gfn_to_hva_memslot(slot
, gfn
);
2448 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2451 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
2454 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
2457 return gfn_to_hva_many(slot
, gfn
, NULL
);
2459 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
2461 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
2463 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
2465 EXPORT_SYMBOL_GPL(gfn_to_hva
);
2467 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2469 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
2471 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
2474 * Return the hva of a @gfn and the R/W attribute if possible.
2476 * @slot: the kvm_memory_slot which contains @gfn
2477 * @gfn: the gfn to be translated
2478 * @writable: used to return the read/write attribute of the @slot if the hva
2479 * is valid and @writable is not NULL
2481 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
2482 gfn_t gfn
, bool *writable
)
2484 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
2486 if (!kvm_is_error_hva(hva
) && writable
)
2487 *writable
= !memslot_is_readonly(slot
);
2492 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
2494 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2496 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2499 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
2501 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2503 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2506 static inline int check_user_page_hwpoison(unsigned long addr
)
2508 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
2510 rc
= get_user_pages(addr
, 1, flags
, NULL
);
2511 return rc
== -EHWPOISON
;
2515 * The fast path to get the writable pfn which will be stored in @pfn,
2516 * true indicates success, otherwise false is returned. It's also the
2517 * only part that runs if we can in atomic context.
2519 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
2520 bool *writable
, kvm_pfn_t
*pfn
)
2522 struct page
*page
[1];
2525 * Fast pin a writable pfn only if it is a write fault request
2526 * or the caller allows to map a writable pfn for a read fault
2529 if (!(write_fault
|| writable
))
2532 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
2533 *pfn
= page_to_pfn(page
[0]);
2544 * The slow path to get the pfn of the specified host virtual address,
2545 * 1 indicates success, -errno is returned if error is detected.
2547 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
2548 bool interruptible
, bool *writable
, kvm_pfn_t
*pfn
)
2551 * When a VCPU accesses a page that is not mapped into the secondary
2552 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2553 * make progress. We always want to honor NUMA hinting faults in that
2554 * case, because GUP usage corresponds to memory accesses from the VCPU.
2555 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2556 * mapped into the secondary MMU and gets accessed by a VCPU.
2558 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2559 * implicitly honor NUMA hinting faults and don't need this flag.
2561 unsigned int flags
= FOLL_HWPOISON
| FOLL_HONOR_NUMA_FAULT
;
2568 *writable
= write_fault
;
2571 flags
|= FOLL_WRITE
;
2573 flags
|= FOLL_NOWAIT
;
2575 flags
|= FOLL_INTERRUPTIBLE
;
2577 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
2581 /* map read fault as writable if possible */
2582 if (unlikely(!write_fault
) && writable
) {
2585 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
2591 *pfn
= page_to_pfn(page
);
2595 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
2597 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
2600 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
2606 static int kvm_try_get_pfn(kvm_pfn_t pfn
)
2608 struct page
*page
= kvm_pfn_to_refcounted_page(pfn
);
2613 return get_page_unless_zero(page
);
2616 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
2617 unsigned long addr
, bool write_fault
,
2618 bool *writable
, kvm_pfn_t
*p_pfn
)
2626 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2629 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2630 * not call the fault handler, so do it here.
2632 bool unlocked
= false;
2633 r
= fixup_user_fault(current
->mm
, addr
,
2634 (write_fault
? FAULT_FLAG_WRITE
: 0),
2641 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2646 pte
= ptep_get(ptep
);
2648 if (write_fault
&& !pte_write(pte
)) {
2649 pfn
= KVM_PFN_ERR_RO_FAULT
;
2654 *writable
= pte_write(pte
);
2658 * Get a reference here because callers of *hva_to_pfn* and
2659 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2660 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2661 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2662 * simply do nothing for reserved pfns.
2664 * Whoever called remap_pfn_range is also going to call e.g.
2665 * unmap_mapping_range before the underlying pages are freed,
2666 * causing a call to our MMU notifier.
2668 * Certain IO or PFNMAP mappings can be backed with valid
2669 * struct pages, but be allocated without refcounting e.g.,
2670 * tail pages of non-compound higher order allocations, which
2671 * would then underflow the refcount when the caller does the
2672 * required put_page. Don't allow those pages here.
2674 if (!kvm_try_get_pfn(pfn
))
2678 pte_unmap_unlock(ptep
, ptl
);
2685 * Pin guest page in memory and return its pfn.
2686 * @addr: host virtual address which maps memory to the guest
2687 * @atomic: whether this function can sleep
2688 * @interruptible: whether the process can be interrupted by non-fatal signals
2689 * @async: whether this function need to wait IO complete if the
2690 * host page is not in the memory
2691 * @write_fault: whether we should get a writable host page
2692 * @writable: whether it allows to map a writable host page for !@write_fault
2694 * The function will map a writable host page for these two cases:
2695 * 1): @write_fault = true
2696 * 2): @write_fault = false && @writable, @writable will tell the caller
2697 * whether the mapping is writable.
2699 kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool interruptible
,
2700 bool *async
, bool write_fault
, bool *writable
)
2702 struct vm_area_struct
*vma
;
2706 /* we can do it either atomically or asynchronously, not both */
2707 BUG_ON(atomic
&& async
);
2709 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
2713 return KVM_PFN_ERR_FAULT
;
2715 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, interruptible
,
2719 if (npages
== -EINTR
)
2720 return KVM_PFN_ERR_SIGPENDING
;
2722 mmap_read_lock(current
->mm
);
2723 if (npages
== -EHWPOISON
||
2724 (!async
&& check_user_page_hwpoison(addr
))) {
2725 pfn
= KVM_PFN_ERR_HWPOISON
;
2730 vma
= vma_lookup(current
->mm
, addr
);
2733 pfn
= KVM_PFN_ERR_FAULT
;
2734 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
2735 r
= hva_to_pfn_remapped(vma
, addr
, write_fault
, writable
, &pfn
);
2739 pfn
= KVM_PFN_ERR_FAULT
;
2741 if (async
&& vma_is_valid(vma
, write_fault
))
2743 pfn
= KVM_PFN_ERR_FAULT
;
2746 mmap_read_unlock(current
->mm
);
2750 kvm_pfn_t
__gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2751 bool atomic
, bool interruptible
, bool *async
,
2752 bool write_fault
, bool *writable
, hva_t
*hva
)
2754 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
2759 if (addr
== KVM_HVA_ERR_RO_BAD
) {
2762 return KVM_PFN_ERR_RO_FAULT
;
2765 if (kvm_is_error_hva(addr
)) {
2768 return KVM_PFN_NOSLOT
;
2771 /* Do not map writable pfn in the readonly memslot. */
2772 if (writable
&& memslot_is_readonly(slot
)) {
2777 return hva_to_pfn(addr
, atomic
, interruptible
, async
, write_fault
,
2780 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
2782 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
2785 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, false,
2786 NULL
, write_fault
, writable
, NULL
);
2788 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
2790 kvm_pfn_t
gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
2792 return __gfn_to_pfn_memslot(slot
, gfn
, false, false, NULL
, true,
2795 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
2797 kvm_pfn_t
gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
2799 return __gfn_to_pfn_memslot(slot
, gfn
, true, false, NULL
, true,
2802 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
2804 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2806 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2808 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
2810 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
2812 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
2814 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
2816 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2818 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2820 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
2822 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2823 struct page
**pages
, int nr_pages
)
2828 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
2829 if (kvm_is_error_hva(addr
))
2832 if (entry
< nr_pages
)
2835 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
2837 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
2840 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2841 * backed by 'struct page'. A valid example is if the backing memslot is
2842 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2843 * been elevated by gfn_to_pfn().
2845 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
2850 pfn
= gfn_to_pfn(kvm
, gfn
);
2852 if (is_error_noslot_pfn(pfn
))
2853 return KVM_ERR_PTR_BAD_PAGE
;
2855 page
= kvm_pfn_to_refcounted_page(pfn
);
2857 return KVM_ERR_PTR_BAD_PAGE
;
2861 EXPORT_SYMBOL_GPL(gfn_to_page
);
2863 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
)
2866 kvm_release_pfn_dirty(pfn
);
2868 kvm_release_pfn_clean(pfn
);
2871 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
2875 struct page
*page
= KVM_UNMAPPED_PAGE
;
2880 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2881 if (is_error_noslot_pfn(pfn
))
2884 if (pfn_valid(pfn
)) {
2885 page
= pfn_to_page(pfn
);
2887 #ifdef CONFIG_HAS_IOMEM
2889 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
2903 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
2905 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
2913 if (map
->page
!= KVM_UNMAPPED_PAGE
)
2915 #ifdef CONFIG_HAS_IOMEM
2921 kvm_vcpu_mark_page_dirty(vcpu
, map
->gfn
);
2923 kvm_release_pfn(map
->pfn
, dirty
);
2928 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
2930 static bool kvm_is_ad_tracked_page(struct page
*page
)
2933 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2934 * touched (e.g. set dirty) except by its owner".
2936 return !PageReserved(page
);
2939 static void kvm_set_page_dirty(struct page
*page
)
2941 if (kvm_is_ad_tracked_page(page
))
2945 static void kvm_set_page_accessed(struct page
*page
)
2947 if (kvm_is_ad_tracked_page(page
))
2948 mark_page_accessed(page
);
2951 void kvm_release_page_clean(struct page
*page
)
2953 WARN_ON(is_error_page(page
));
2955 kvm_set_page_accessed(page
);
2958 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
2960 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
2964 if (is_error_noslot_pfn(pfn
))
2967 page
= kvm_pfn_to_refcounted_page(pfn
);
2971 kvm_release_page_clean(page
);
2973 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
2975 void kvm_release_page_dirty(struct page
*page
)
2977 WARN_ON(is_error_page(page
));
2979 kvm_set_page_dirty(page
);
2980 kvm_release_page_clean(page
);
2982 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
2984 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
2988 if (is_error_noslot_pfn(pfn
))
2991 page
= kvm_pfn_to_refcounted_page(pfn
);
2995 kvm_release_page_dirty(page
);
2997 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
3000 * Note, checking for an error/noslot pfn is the caller's responsibility when
3001 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
3002 * "set" helpers are not to be used when the pfn might point at garbage.
3004 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
3006 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3010 kvm_set_page_dirty(pfn_to_page(pfn
));
3012 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
3014 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
3016 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3020 kvm_set_page_accessed(pfn_to_page(pfn
));
3022 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
3024 static int next_segment(unsigned long len
, int offset
)
3026 if (len
> PAGE_SIZE
- offset
)
3027 return PAGE_SIZE
- offset
;
3032 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3033 void *data
, int offset
, int len
)
3038 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3039 if (kvm_is_error_hva(addr
))
3041 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
3047 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
3050 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3052 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3054 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
3056 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
3057 int offset
, int len
)
3059 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3061 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3063 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
3065 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
3067 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3069 int offset
= offset_in_page(gpa
);
3072 while ((seg
= next_segment(len
, offset
)) != 0) {
3073 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
3083 EXPORT_SYMBOL_GPL(kvm_read_guest
);
3085 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
3087 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3089 int offset
= offset_in_page(gpa
);
3092 while ((seg
= next_segment(len
, offset
)) != 0) {
3093 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3103 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
3105 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3106 void *data
, int offset
, unsigned long len
)
3111 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3112 if (kvm_is_error_hva(addr
))
3114 pagefault_disable();
3115 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
3122 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3123 void *data
, unsigned long len
)
3125 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3126 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3127 int offset
= offset_in_page(gpa
);
3129 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
3131 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
3133 static int __kvm_write_guest_page(struct kvm
*kvm
,
3134 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
3135 const void *data
, int offset
, int len
)
3140 addr
= gfn_to_hva_memslot(memslot
, gfn
);
3141 if (kvm_is_error_hva(addr
))
3143 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
3146 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3150 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
3151 const void *data
, int offset
, int len
)
3153 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3155 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
3157 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
3159 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
3160 const void *data
, int offset
, int len
)
3162 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3164 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
3166 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
3168 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
3171 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3173 int offset
= offset_in_page(gpa
);
3176 while ((seg
= next_segment(len
, offset
)) != 0) {
3177 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
3187 EXPORT_SYMBOL_GPL(kvm_write_guest
);
3189 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
3192 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3194 int offset
= offset_in_page(gpa
);
3197 while ((seg
= next_segment(len
, offset
)) != 0) {
3198 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3208 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
3210 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
3211 struct gfn_to_hva_cache
*ghc
,
3212 gpa_t gpa
, unsigned long len
)
3214 int offset
= offset_in_page(gpa
);
3215 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
3216 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
3217 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
3218 gfn_t nr_pages_avail
;
3220 /* Update ghc->generation before performing any error checks. */
3221 ghc
->generation
= slots
->generation
;
3223 if (start_gfn
> end_gfn
) {
3224 ghc
->hva
= KVM_HVA_ERR_BAD
;
3229 * If the requested region crosses two memslots, we still
3230 * verify that the entire region is valid here.
3232 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
3233 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
3234 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
3236 if (kvm_is_error_hva(ghc
->hva
))
3240 /* Use the slow path for cross page reads and writes. */
3241 if (nr_pages_needed
== 1)
3244 ghc
->memslot
= NULL
;
3251 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3252 gpa_t gpa
, unsigned long len
)
3254 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3255 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
3257 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
3259 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3260 void *data
, unsigned int offset
,
3263 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3265 gpa_t gpa
= ghc
->gpa
+ offset
;
3267 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3270 if (slots
->generation
!= ghc
->generation
) {
3271 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3275 if (kvm_is_error_hva(ghc
->hva
))
3278 if (unlikely(!ghc
->memslot
))
3279 return kvm_write_guest(kvm
, gpa
, data
, len
);
3281 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
3284 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
3288 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
3290 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3291 void *data
, unsigned long len
)
3293 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3295 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
3297 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3298 void *data
, unsigned int offset
,
3301 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3303 gpa_t gpa
= ghc
->gpa
+ offset
;
3305 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3308 if (slots
->generation
!= ghc
->generation
) {
3309 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3313 if (kvm_is_error_hva(ghc
->hva
))
3316 if (unlikely(!ghc
->memslot
))
3317 return kvm_read_guest(kvm
, gpa
, data
, len
);
3319 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
3325 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
3327 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3328 void *data
, unsigned long len
)
3330 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3332 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
3334 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
3336 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3337 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3339 int offset
= offset_in_page(gpa
);
3342 while ((seg
= next_segment(len
, offset
)) != 0) {
3343 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
3352 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
3354 void mark_page_dirty_in_slot(struct kvm
*kvm
,
3355 const struct kvm_memory_slot
*memslot
,
3358 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
3360 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3361 if (WARN_ON_ONCE(vcpu
&& vcpu
->kvm
!= kvm
))
3364 WARN_ON_ONCE(!vcpu
&& !kvm_arch_allow_write_without_running_vcpu(kvm
));
3367 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
3368 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
3369 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
3371 if (kvm
->dirty_ring_size
&& vcpu
)
3372 kvm_dirty_ring_push(vcpu
, slot
, rel_gfn
);
3373 else if (memslot
->dirty_bitmap
)
3374 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
3377 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
3379 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
3381 struct kvm_memory_slot
*memslot
;
3383 memslot
= gfn_to_memslot(kvm
, gfn
);
3384 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3386 EXPORT_SYMBOL_GPL(mark_page_dirty
);
3388 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3390 struct kvm_memory_slot
*memslot
;
3392 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3393 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
3395 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
3397 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
3399 if (!vcpu
->sigset_active
)
3403 * This does a lockless modification of ->real_blocked, which is fine
3404 * because, only current can change ->real_blocked and all readers of
3405 * ->real_blocked don't care as long ->real_blocked is always a subset
3408 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
3411 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
3413 if (!vcpu
->sigset_active
)
3416 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
3417 sigemptyset(¤t
->real_blocked
);
3420 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3422 unsigned int old
, val
, grow
, grow_start
;
3424 old
= val
= vcpu
->halt_poll_ns
;
3425 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3426 grow
= READ_ONCE(halt_poll_ns_grow
);
3431 if (val
< grow_start
)
3434 vcpu
->halt_poll_ns
= val
;
3436 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
3439 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3441 unsigned int old
, val
, shrink
, grow_start
;
3443 old
= val
= vcpu
->halt_poll_ns
;
3444 shrink
= READ_ONCE(halt_poll_ns_shrink
);
3445 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3451 if (val
< grow_start
)
3454 vcpu
->halt_poll_ns
= val
;
3455 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
3458 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
3461 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3463 if (kvm_arch_vcpu_runnable(vcpu
))
3465 if (kvm_cpu_has_pending_timer(vcpu
))
3467 if (signal_pending(current
))
3469 if (kvm_check_request(KVM_REQ_UNBLOCK
, vcpu
))
3474 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3479 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3480 * pending. This is mostly used when halting a vCPU, but may also be used
3481 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3483 bool kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
3485 struct rcuwait
*wait
= kvm_arch_vcpu_get_wait(vcpu
);
3486 bool waited
= false;
3488 vcpu
->stat
.generic
.blocking
= 1;
3491 kvm_arch_vcpu_blocking(vcpu
);
3492 prepare_to_rcuwait(wait
);
3496 set_current_state(TASK_INTERRUPTIBLE
);
3498 if (kvm_vcpu_check_block(vcpu
) < 0)
3506 finish_rcuwait(wait
);
3507 kvm_arch_vcpu_unblocking(vcpu
);
3510 vcpu
->stat
.generic
.blocking
= 0;
3515 static inline void update_halt_poll_stats(struct kvm_vcpu
*vcpu
, ktime_t start
,
3516 ktime_t end
, bool success
)
3518 struct kvm_vcpu_stat_generic
*stats
= &vcpu
->stat
.generic
;
3519 u64 poll_ns
= ktime_to_ns(ktime_sub(end
, start
));
3521 ++vcpu
->stat
.generic
.halt_attempted_poll
;
3524 ++vcpu
->stat
.generic
.halt_successful_poll
;
3526 if (!vcpu_valid_wakeup(vcpu
))
3527 ++vcpu
->stat
.generic
.halt_poll_invalid
;
3529 stats
->halt_poll_success_ns
+= poll_ns
;
3530 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_success_hist
, poll_ns
);
3532 stats
->halt_poll_fail_ns
+= poll_ns
;
3533 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_fail_hist
, poll_ns
);
3537 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3539 struct kvm
*kvm
= vcpu
->kvm
;
3541 if (kvm
->override_halt_poll_ns
) {
3543 * Ensure kvm->max_halt_poll_ns is not read before
3544 * kvm->override_halt_poll_ns.
3546 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3549 return READ_ONCE(kvm
->max_halt_poll_ns
);
3552 return READ_ONCE(halt_poll_ns
);
3556 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3557 * polling is enabled, busy wait for a short time before blocking to avoid the
3558 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3561 void kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
3563 unsigned int max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3564 bool halt_poll_allowed
= !kvm_arch_no_poll(vcpu
);
3565 ktime_t start
, cur
, poll_end
;
3566 bool waited
= false;
3570 if (vcpu
->halt_poll_ns
> max_halt_poll_ns
)
3571 vcpu
->halt_poll_ns
= max_halt_poll_ns
;
3573 do_halt_poll
= halt_poll_allowed
&& vcpu
->halt_poll_ns
;
3575 start
= cur
= poll_end
= ktime_get();
3577 ktime_t stop
= ktime_add_ns(start
, vcpu
->halt_poll_ns
);
3580 if (kvm_vcpu_check_block(vcpu
) < 0)
3583 poll_end
= cur
= ktime_get();
3584 } while (kvm_vcpu_can_poll(cur
, stop
));
3587 waited
= kvm_vcpu_block(vcpu
);
3591 vcpu
->stat
.generic
.halt_wait_ns
+=
3592 ktime_to_ns(cur
) - ktime_to_ns(poll_end
);
3593 KVM_STATS_LOG_HIST_UPDATE(vcpu
->stat
.generic
.halt_wait_hist
,
3594 ktime_to_ns(cur
) - ktime_to_ns(poll_end
));
3597 /* The total time the vCPU was "halted", including polling time. */
3598 halt_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
3601 * Note, halt-polling is considered successful so long as the vCPU was
3602 * never actually scheduled out, i.e. even if the wake event arrived
3603 * after of the halt-polling loop itself, but before the full wait.
3606 update_halt_poll_stats(vcpu
, start
, poll_end
, !waited
);
3608 if (halt_poll_allowed
) {
3609 /* Recompute the max halt poll time in case it changed. */
3610 max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3612 if (!vcpu_valid_wakeup(vcpu
)) {
3613 shrink_halt_poll_ns(vcpu
);
3614 } else if (max_halt_poll_ns
) {
3615 if (halt_ns
<= vcpu
->halt_poll_ns
)
3617 /* we had a long block, shrink polling */
3618 else if (vcpu
->halt_poll_ns
&&
3619 halt_ns
> max_halt_poll_ns
)
3620 shrink_halt_poll_ns(vcpu
);
3621 /* we had a short halt and our poll time is too small */
3622 else if (vcpu
->halt_poll_ns
< max_halt_poll_ns
&&
3623 halt_ns
< max_halt_poll_ns
)
3624 grow_halt_poll_ns(vcpu
);
3626 vcpu
->halt_poll_ns
= 0;
3630 trace_kvm_vcpu_wakeup(halt_ns
, waited
, vcpu_valid_wakeup(vcpu
));
3632 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
3634 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
3636 if (__kvm_vcpu_wake_up(vcpu
)) {
3637 WRITE_ONCE(vcpu
->ready
, true);
3638 ++vcpu
->stat
.generic
.halt_wakeup
;
3644 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
3648 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3650 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3654 if (kvm_vcpu_wake_up(vcpu
))
3659 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3660 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3661 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3662 * within the vCPU thread itself.
3664 if (vcpu
== __this_cpu_read(kvm_running_vcpu
)) {
3665 if (vcpu
->mode
== IN_GUEST_MODE
)
3666 WRITE_ONCE(vcpu
->mode
, EXITING_GUEST_MODE
);
3671 * Note, the vCPU could get migrated to a different pCPU at any point
3672 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3673 * IPI to the previous pCPU. But, that's ok because the purpose of the
3674 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3675 * vCPU also requires it to leave IN_GUEST_MODE.
3677 if (kvm_arch_vcpu_should_kick(vcpu
)) {
3678 cpu
= READ_ONCE(vcpu
->cpu
);
3679 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
3680 smp_send_reschedule(cpu
);
3685 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
3686 #endif /* !CONFIG_S390 */
3688 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
3691 struct task_struct
*task
= NULL
;
3695 pid
= rcu_dereference(target
->pid
);
3697 task
= get_pid_task(pid
, PIDTYPE_PID
);
3701 ret
= yield_to(task
, 1);
3702 put_task_struct(task
);
3706 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
3709 * Helper that checks whether a VCPU is eligible for directed yield.
3710 * Most eligible candidate to yield is decided by following heuristics:
3712 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3713 * (preempted lock holder), indicated by @in_spin_loop.
3714 * Set at the beginning and cleared at the end of interception/PLE handler.
3716 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3717 * chance last time (mostly it has become eligible now since we have probably
3718 * yielded to lockholder in last iteration. This is done by toggling
3719 * @dy_eligible each time a VCPU checked for eligibility.)
3721 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3722 * to preempted lock-holder could result in wrong VCPU selection and CPU
3723 * burning. Giving priority for a potential lock-holder increases lock
3726 * Since algorithm is based on heuristics, accessing another VCPU data without
3727 * locking does not harm. It may result in trying to yield to same VCPU, fail
3728 * and continue with next VCPU and so on.
3730 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
3732 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3735 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
3736 vcpu
->spin_loop
.dy_eligible
;
3738 if (vcpu
->spin_loop
.in_spin_loop
)
3739 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
3748 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3749 * a vcpu_load/vcpu_put pair. However, for most architectures
3750 * kvm_arch_vcpu_runnable does not require vcpu_load.
3752 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
3754 return kvm_arch_vcpu_runnable(vcpu
);
3757 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
3759 if (kvm_arch_dy_runnable(vcpu
))
3762 #ifdef CONFIG_KVM_ASYNC_PF
3763 if (!list_empty_careful(&vcpu
->async_pf
.done
))
3770 bool __weak
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu
*vcpu
)
3775 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
3777 struct kvm
*kvm
= me
->kvm
;
3778 struct kvm_vcpu
*vcpu
;
3779 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
3785 kvm_vcpu_set_in_spin_loop(me
, true);
3787 * We boost the priority of a VCPU that is runnable but not
3788 * currently running, because it got preempted by something
3789 * else and called schedule in __vcpu_run. Hopefully that
3790 * VCPU is holding the lock that we need and will release it.
3791 * We approximate round-robin by starting at the last boosted VCPU.
3793 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
3794 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3795 if (!pass
&& i
<= last_boosted_vcpu
) {
3796 i
= last_boosted_vcpu
;
3798 } else if (pass
&& i
> last_boosted_vcpu
)
3800 if (!READ_ONCE(vcpu
->ready
))
3804 if (kvm_vcpu_is_blocking(vcpu
) && !vcpu_dy_runnable(vcpu
))
3806 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
3807 !kvm_arch_dy_has_pending_interrupt(vcpu
) &&
3808 !kvm_arch_vcpu_in_kernel(vcpu
))
3810 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
3813 yielded
= kvm_vcpu_yield_to(vcpu
);
3815 kvm
->last_boosted_vcpu
= i
;
3817 } else if (yielded
< 0) {
3824 kvm_vcpu_set_in_spin_loop(me
, false);
3826 /* Ensure vcpu is not eligible during next spinloop */
3827 kvm_vcpu_set_dy_eligible(me
, false);
3829 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
3831 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
3833 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3834 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
3835 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
3836 kvm
->dirty_ring_size
/ PAGE_SIZE
);
3842 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
3844 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
3847 if (vmf
->pgoff
== 0)
3848 page
= virt_to_page(vcpu
->run
);
3850 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
3851 page
= virt_to_page(vcpu
->arch
.pio_data
);
3853 #ifdef CONFIG_KVM_MMIO
3854 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
3855 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
3857 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
3858 page
= kvm_dirty_ring_get_page(
3860 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
3862 return kvm_arch_vcpu_fault(vcpu
, vmf
);
3868 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
3869 .fault
= kvm_vcpu_fault
,
3872 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3874 struct kvm_vcpu
*vcpu
= file
->private_data
;
3875 unsigned long pages
= vma_pages(vma
);
3877 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
3878 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
3879 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
3882 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
3886 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
3888 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3890 kvm_put_kvm(vcpu
->kvm
);
3894 static const struct file_operations kvm_vcpu_fops
= {
3895 .release
= kvm_vcpu_release
,
3896 .unlocked_ioctl
= kvm_vcpu_ioctl
,
3897 .mmap
= kvm_vcpu_mmap
,
3898 .llseek
= noop_llseek
,
3899 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
3903 * Allocates an inode for the vcpu.
3905 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
3907 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
3909 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
3910 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
3913 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3914 static int vcpu_get_pid(void *data
, u64
*val
)
3916 struct kvm_vcpu
*vcpu
= data
;
3919 *val
= pid_nr(rcu_dereference(vcpu
->pid
));
3924 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops
, vcpu_get_pid
, NULL
, "%llu\n");
3926 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
3928 struct dentry
*debugfs_dentry
;
3929 char dir_name
[ITOA_MAX_LEN
* 2];
3931 if (!debugfs_initialized())
3934 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
3935 debugfs_dentry
= debugfs_create_dir(dir_name
,
3936 vcpu
->kvm
->debugfs_dentry
);
3937 debugfs_create_file("pid", 0444, debugfs_dentry
, vcpu
,
3938 &vcpu_get_pid_fops
);
3940 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
3945 * Creates some virtual cpus. Good luck creating more than one.
3947 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
3950 struct kvm_vcpu
*vcpu
;
3953 if (id
>= KVM_MAX_VCPU_IDS
)
3956 mutex_lock(&kvm
->lock
);
3957 if (kvm
->created_vcpus
>= kvm
->max_vcpus
) {
3958 mutex_unlock(&kvm
->lock
);
3962 r
= kvm_arch_vcpu_precreate(kvm
, id
);
3964 mutex_unlock(&kvm
->lock
);
3968 kvm
->created_vcpus
++;
3969 mutex_unlock(&kvm
->lock
);
3971 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL_ACCOUNT
);
3974 goto vcpu_decrement
;
3977 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
3978 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
3983 vcpu
->run
= page_address(page
);
3985 kvm_vcpu_init(vcpu
, kvm
, id
);
3987 r
= kvm_arch_vcpu_create(vcpu
);
3989 goto vcpu_free_run_page
;
3991 if (kvm
->dirty_ring_size
) {
3992 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
3993 id
, kvm
->dirty_ring_size
);
3995 goto arch_vcpu_destroy
;
3998 mutex_lock(&kvm
->lock
);
4000 #ifdef CONFIG_LOCKDEP
4001 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
4002 mutex_lock(&vcpu
->mutex
);
4003 mutex_unlock(&vcpu
->mutex
);
4006 if (kvm_get_vcpu_by_id(kvm
, id
)) {
4008 goto unlock_vcpu_destroy
;
4011 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
4012 r
= xa_reserve(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, GFP_KERNEL_ACCOUNT
);
4014 goto unlock_vcpu_destroy
;
4016 /* Now it's all set up, let userspace reach it */
4018 r
= create_vcpu_fd(vcpu
);
4020 goto kvm_put_xa_release
;
4022 if (KVM_BUG_ON(xa_store(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, vcpu
, 0), kvm
)) {
4024 goto kvm_put_xa_release
;
4028 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4029 * pointer before kvm->online_vcpu's incremented value.
4032 atomic_inc(&kvm
->online_vcpus
);
4034 mutex_unlock(&kvm
->lock
);
4035 kvm_arch_vcpu_postcreate(vcpu
);
4036 kvm_create_vcpu_debugfs(vcpu
);
4040 kvm_put_kvm_no_destroy(kvm
);
4041 xa_release(&kvm
->vcpu_array
, vcpu
->vcpu_idx
);
4042 unlock_vcpu_destroy
:
4043 mutex_unlock(&kvm
->lock
);
4044 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
4046 kvm_arch_vcpu_destroy(vcpu
);
4048 free_page((unsigned long)vcpu
->run
);
4050 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
4052 mutex_lock(&kvm
->lock
);
4053 kvm
->created_vcpus
--;
4054 mutex_unlock(&kvm
->lock
);
4058 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
4061 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
4062 vcpu
->sigset_active
= 1;
4063 vcpu
->sigset
= *sigset
;
4065 vcpu
->sigset_active
= 0;
4069 static ssize_t
kvm_vcpu_stats_read(struct file
*file
, char __user
*user_buffer
,
4070 size_t size
, loff_t
*offset
)
4072 struct kvm_vcpu
*vcpu
= file
->private_data
;
4074 return kvm_stats_read(vcpu
->stats_id
, &kvm_vcpu_stats_header
,
4075 &kvm_vcpu_stats_desc
[0], &vcpu
->stat
,
4076 sizeof(vcpu
->stat
), user_buffer
, size
, offset
);
4079 static int kvm_vcpu_stats_release(struct inode
*inode
, struct file
*file
)
4081 struct kvm_vcpu
*vcpu
= file
->private_data
;
4083 kvm_put_kvm(vcpu
->kvm
);
4087 static const struct file_operations kvm_vcpu_stats_fops
= {
4088 .read
= kvm_vcpu_stats_read
,
4089 .release
= kvm_vcpu_stats_release
,
4090 .llseek
= noop_llseek
,
4093 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu
*vcpu
)
4097 char name
[15 + ITOA_MAX_LEN
+ 1];
4099 snprintf(name
, sizeof(name
), "kvm-vcpu-stats:%d", vcpu
->vcpu_id
);
4101 fd
= get_unused_fd_flags(O_CLOEXEC
);
4105 file
= anon_inode_getfile(name
, &kvm_vcpu_stats_fops
, vcpu
, O_RDONLY
);
4108 return PTR_ERR(file
);
4111 kvm_get_kvm(vcpu
->kvm
);
4113 file
->f_mode
|= FMODE_PREAD
;
4114 fd_install(fd
, file
);
4119 static long kvm_vcpu_ioctl(struct file
*filp
,
4120 unsigned int ioctl
, unsigned long arg
)
4122 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4123 void __user
*argp
= (void __user
*)arg
;
4125 struct kvm_fpu
*fpu
= NULL
;
4126 struct kvm_sregs
*kvm_sregs
= NULL
;
4128 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4131 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
4135 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4136 * execution; mutex_lock() would break them.
4138 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
4139 if (r
!= -ENOIOCTLCMD
)
4142 if (mutex_lock_killable(&vcpu
->mutex
))
4150 oldpid
= rcu_access_pointer(vcpu
->pid
);
4151 if (unlikely(oldpid
!= task_pid(current
))) {
4152 /* The thread running this VCPU changed. */
4155 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
4159 newpid
= get_task_pid(current
, PIDTYPE_PID
);
4160 rcu_assign_pointer(vcpu
->pid
, newpid
);
4165 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
4166 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
4169 case KVM_GET_REGS
: {
4170 struct kvm_regs
*kvm_regs
;
4173 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
4176 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
4180 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
4187 case KVM_SET_REGS
: {
4188 struct kvm_regs
*kvm_regs
;
4190 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
4191 if (IS_ERR(kvm_regs
)) {
4192 r
= PTR_ERR(kvm_regs
);
4195 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
4199 case KVM_GET_SREGS
: {
4200 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
4201 GFP_KERNEL_ACCOUNT
);
4205 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
4209 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
4214 case KVM_SET_SREGS
: {
4215 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
4216 if (IS_ERR(kvm_sregs
)) {
4217 r
= PTR_ERR(kvm_sregs
);
4221 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
4224 case KVM_GET_MP_STATE
: {
4225 struct kvm_mp_state mp_state
;
4227 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
4231 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
4236 case KVM_SET_MP_STATE
: {
4237 struct kvm_mp_state mp_state
;
4240 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
4242 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
4245 case KVM_TRANSLATE
: {
4246 struct kvm_translation tr
;
4249 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
4251 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
4255 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
4260 case KVM_SET_GUEST_DEBUG
: {
4261 struct kvm_guest_debug dbg
;
4264 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
4266 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
4269 case KVM_SET_SIGNAL_MASK
: {
4270 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4271 struct kvm_signal_mask kvm_sigmask
;
4272 sigset_t sigset
, *p
;
4277 if (copy_from_user(&kvm_sigmask
, argp
,
4278 sizeof(kvm_sigmask
)))
4281 if (kvm_sigmask
.len
!= sizeof(sigset
))
4284 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
4289 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
4293 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
4297 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
4301 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
4307 fpu
= memdup_user(argp
, sizeof(*fpu
));
4313 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
4316 case KVM_GET_STATS_FD
: {
4317 r
= kvm_vcpu_ioctl_get_stats_fd(vcpu
);
4321 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
4324 mutex_unlock(&vcpu
->mutex
);
4330 #ifdef CONFIG_KVM_COMPAT
4331 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
4332 unsigned int ioctl
, unsigned long arg
)
4334 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4335 void __user
*argp
= compat_ptr(arg
);
4338 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4342 case KVM_SET_SIGNAL_MASK
: {
4343 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4344 struct kvm_signal_mask kvm_sigmask
;
4349 if (copy_from_user(&kvm_sigmask
, argp
,
4350 sizeof(kvm_sigmask
)))
4353 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
4356 if (get_compat_sigset(&sigset
,
4357 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
4359 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
4361 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
4365 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
4373 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
4375 struct kvm_device
*dev
= filp
->private_data
;
4378 return dev
->ops
->mmap(dev
, vma
);
4383 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
4384 int (*accessor
)(struct kvm_device
*dev
,
4385 struct kvm_device_attr
*attr
),
4388 struct kvm_device_attr attr
;
4393 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
4396 return accessor(dev
, &attr
);
4399 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
4402 struct kvm_device
*dev
= filp
->private_data
;
4404 if (dev
->kvm
->mm
!= current
->mm
|| dev
->kvm
->vm_dead
)
4408 case KVM_SET_DEVICE_ATTR
:
4409 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
4410 case KVM_GET_DEVICE_ATTR
:
4411 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
4412 case KVM_HAS_DEVICE_ATTR
:
4413 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
4415 if (dev
->ops
->ioctl
)
4416 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
4422 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
4424 struct kvm_device
*dev
= filp
->private_data
;
4425 struct kvm
*kvm
= dev
->kvm
;
4427 if (dev
->ops
->release
) {
4428 mutex_lock(&kvm
->lock
);
4429 list_del(&dev
->vm_node
);
4430 dev
->ops
->release(dev
);
4431 mutex_unlock(&kvm
->lock
);
4438 static const struct file_operations kvm_device_fops
= {
4439 .unlocked_ioctl
= kvm_device_ioctl
,
4440 .release
= kvm_device_release
,
4441 KVM_COMPAT(kvm_device_ioctl
),
4442 .mmap
= kvm_device_mmap
,
4445 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
4447 if (filp
->f_op
!= &kvm_device_fops
)
4450 return filp
->private_data
;
4453 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
4454 #ifdef CONFIG_KVM_MPIC
4455 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
4456 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
4460 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
4462 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
4465 if (kvm_device_ops_table
[type
] != NULL
)
4468 kvm_device_ops_table
[type
] = ops
;
4472 void kvm_unregister_device_ops(u32 type
)
4474 if (kvm_device_ops_table
[type
] != NULL
)
4475 kvm_device_ops_table
[type
] = NULL
;
4478 static int kvm_ioctl_create_device(struct kvm
*kvm
,
4479 struct kvm_create_device
*cd
)
4481 const struct kvm_device_ops
*ops
;
4482 struct kvm_device
*dev
;
4483 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
4487 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
4490 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
4491 ops
= kvm_device_ops_table
[type
];
4498 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
4505 mutex_lock(&kvm
->lock
);
4506 ret
= ops
->create(dev
, type
);
4508 mutex_unlock(&kvm
->lock
);
4512 list_add(&dev
->vm_node
, &kvm
->devices
);
4513 mutex_unlock(&kvm
->lock
);
4519 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
4521 kvm_put_kvm_no_destroy(kvm
);
4522 mutex_lock(&kvm
->lock
);
4523 list_del(&dev
->vm_node
);
4526 mutex_unlock(&kvm
->lock
);
4536 static int kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
4539 case KVM_CAP_USER_MEMORY
:
4540 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
4541 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
4542 case KVM_CAP_INTERNAL_ERROR_DATA
:
4543 #ifdef CONFIG_HAVE_KVM_MSI
4544 case KVM_CAP_SIGNAL_MSI
:
4546 #ifdef CONFIG_HAVE_KVM_IRQFD
4549 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
4550 case KVM_CAP_CHECK_EXTENSION_VM
:
4551 case KVM_CAP_ENABLE_CAP_VM
:
4552 case KVM_CAP_HALT_POLL
:
4554 #ifdef CONFIG_KVM_MMIO
4555 case KVM_CAP_COALESCED_MMIO
:
4556 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
4557 case KVM_CAP_COALESCED_PIO
:
4560 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4561 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
4562 return KVM_DIRTY_LOG_MANUAL_CAPS
;
4564 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4565 case KVM_CAP_IRQ_ROUTING
:
4566 return KVM_MAX_IRQ_ROUTES
;
4568 #if KVM_ADDRESS_SPACE_NUM > 1
4569 case KVM_CAP_MULTI_ADDRESS_SPACE
:
4570 return KVM_ADDRESS_SPACE_NUM
;
4572 case KVM_CAP_NR_MEMSLOTS
:
4573 return KVM_USER_MEM_SLOTS
;
4574 case KVM_CAP_DIRTY_LOG_RING
:
4575 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4576 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4580 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4581 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4582 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4586 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4587 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
:
4589 case KVM_CAP_BINARY_STATS_FD
:
4590 case KVM_CAP_SYSTEM_EVENT_DATA
:
4595 return kvm_vm_ioctl_check_extension(kvm
, arg
);
4598 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
4602 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
4605 /* the size should be power of 2 */
4606 if (!size
|| (size
& (size
- 1)))
4609 /* Should be bigger to keep the reserved entries, or a page */
4610 if (size
< kvm_dirty_ring_get_rsvd_entries() *
4611 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
4614 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
4615 sizeof(struct kvm_dirty_gfn
))
4618 /* We only allow it to set once */
4619 if (kvm
->dirty_ring_size
)
4622 mutex_lock(&kvm
->lock
);
4624 if (kvm
->created_vcpus
) {
4625 /* We don't allow to change this value after vcpu created */
4628 kvm
->dirty_ring_size
= size
;
4632 mutex_unlock(&kvm
->lock
);
4636 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
4639 struct kvm_vcpu
*vcpu
;
4642 if (!kvm
->dirty_ring_size
)
4645 mutex_lock(&kvm
->slots_lock
);
4647 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4648 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
4650 mutex_unlock(&kvm
->slots_lock
);
4653 kvm_flush_remote_tlbs(kvm
);
4658 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4659 struct kvm_enable_cap
*cap
)
4664 bool kvm_are_all_memslots_empty(struct kvm
*kvm
)
4668 lockdep_assert_held(&kvm
->slots_lock
);
4670 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
4671 if (!kvm_memslots_empty(__kvm_memslots(kvm
, i
)))
4677 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty
);
4679 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
4680 struct kvm_enable_cap
*cap
)
4683 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4684 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
4685 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
4687 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
4688 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
4690 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
4692 kvm
->manual_dirty_log_protect
= cap
->args
[0];
4696 case KVM_CAP_HALT_POLL
: {
4697 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
4700 kvm
->max_halt_poll_ns
= cap
->args
[0];
4703 * Ensure kvm->override_halt_poll_ns does not become visible
4704 * before kvm->max_halt_poll_ns.
4706 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4709 kvm
->override_halt_poll_ns
= true;
4713 case KVM_CAP_DIRTY_LOG_RING
:
4714 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4715 if (!kvm_vm_ioctl_check_extension_generic(kvm
, cap
->cap
))
4718 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
4719 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
: {
4722 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
) ||
4723 !kvm
->dirty_ring_size
|| cap
->flags
)
4726 mutex_lock(&kvm
->slots_lock
);
4729 * For simplicity, allow enabling ring+bitmap if and only if
4730 * there are no memslots, e.g. to ensure all memslots allocate
4731 * a bitmap after the capability is enabled.
4733 if (kvm_are_all_memslots_empty(kvm
)) {
4734 kvm
->dirty_ring_with_bitmap
= true;
4738 mutex_unlock(&kvm
->slots_lock
);
4743 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
4747 static ssize_t
kvm_vm_stats_read(struct file
*file
, char __user
*user_buffer
,
4748 size_t size
, loff_t
*offset
)
4750 struct kvm
*kvm
= file
->private_data
;
4752 return kvm_stats_read(kvm
->stats_id
, &kvm_vm_stats_header
,
4753 &kvm_vm_stats_desc
[0], &kvm
->stat
,
4754 sizeof(kvm
->stat
), user_buffer
, size
, offset
);
4757 static int kvm_vm_stats_release(struct inode
*inode
, struct file
*file
)
4759 struct kvm
*kvm
= file
->private_data
;
4765 static const struct file_operations kvm_vm_stats_fops
= {
4766 .read
= kvm_vm_stats_read
,
4767 .release
= kvm_vm_stats_release
,
4768 .llseek
= noop_llseek
,
4771 static int kvm_vm_ioctl_get_stats_fd(struct kvm
*kvm
)
4776 fd
= get_unused_fd_flags(O_CLOEXEC
);
4780 file
= anon_inode_getfile("kvm-vm-stats",
4781 &kvm_vm_stats_fops
, kvm
, O_RDONLY
);
4784 return PTR_ERR(file
);
4789 file
->f_mode
|= FMODE_PREAD
;
4790 fd_install(fd
, file
);
4795 static long kvm_vm_ioctl(struct file
*filp
,
4796 unsigned int ioctl
, unsigned long arg
)
4798 struct kvm
*kvm
= filp
->private_data
;
4799 void __user
*argp
= (void __user
*)arg
;
4802 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
4805 case KVM_CREATE_VCPU
:
4806 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
4808 case KVM_ENABLE_CAP
: {
4809 struct kvm_enable_cap cap
;
4812 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4814 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
4817 case KVM_SET_USER_MEMORY_REGION
: {
4818 struct kvm_userspace_memory_region kvm_userspace_mem
;
4821 if (copy_from_user(&kvm_userspace_mem
, argp
,
4822 sizeof(kvm_userspace_mem
)))
4825 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
4828 case KVM_GET_DIRTY_LOG
: {
4829 struct kvm_dirty_log log
;
4832 if (copy_from_user(&log
, argp
, sizeof(log
)))
4834 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
4837 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4838 case KVM_CLEAR_DIRTY_LOG
: {
4839 struct kvm_clear_dirty_log log
;
4842 if (copy_from_user(&log
, argp
, sizeof(log
)))
4844 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
4848 #ifdef CONFIG_KVM_MMIO
4849 case KVM_REGISTER_COALESCED_MMIO
: {
4850 struct kvm_coalesced_mmio_zone zone
;
4853 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4855 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
4858 case KVM_UNREGISTER_COALESCED_MMIO
: {
4859 struct kvm_coalesced_mmio_zone zone
;
4862 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4864 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
4869 struct kvm_irqfd data
;
4872 if (copy_from_user(&data
, argp
, sizeof(data
)))
4874 r
= kvm_irqfd(kvm
, &data
);
4877 case KVM_IOEVENTFD
: {
4878 struct kvm_ioeventfd data
;
4881 if (copy_from_user(&data
, argp
, sizeof(data
)))
4883 r
= kvm_ioeventfd(kvm
, &data
);
4886 #ifdef CONFIG_HAVE_KVM_MSI
4887 case KVM_SIGNAL_MSI
: {
4891 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
4893 r
= kvm_send_userspace_msi(kvm
, &msi
);
4897 #ifdef __KVM_HAVE_IRQ_LINE
4898 case KVM_IRQ_LINE_STATUS
:
4899 case KVM_IRQ_LINE
: {
4900 struct kvm_irq_level irq_event
;
4903 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
4906 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
4907 ioctl
== KVM_IRQ_LINE_STATUS
);
4912 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
4913 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
4921 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4922 case KVM_SET_GSI_ROUTING
: {
4923 struct kvm_irq_routing routing
;
4924 struct kvm_irq_routing __user
*urouting
;
4925 struct kvm_irq_routing_entry
*entries
= NULL
;
4928 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
4931 if (!kvm_arch_can_set_irq_routing(kvm
))
4933 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
4939 entries
= vmemdup_user(urouting
->entries
,
4940 array_size(sizeof(*entries
),
4942 if (IS_ERR(entries
)) {
4943 r
= PTR_ERR(entries
);
4947 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
4952 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4953 case KVM_CREATE_DEVICE
: {
4954 struct kvm_create_device cd
;
4957 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
4960 r
= kvm_ioctl_create_device(kvm
, &cd
);
4965 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
4971 case KVM_CHECK_EXTENSION
:
4972 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
4974 case KVM_RESET_DIRTY_RINGS
:
4975 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
4977 case KVM_GET_STATS_FD
:
4978 r
= kvm_vm_ioctl_get_stats_fd(kvm
);
4981 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
4987 #ifdef CONFIG_KVM_COMPAT
4988 struct compat_kvm_dirty_log
{
4992 compat_uptr_t dirty_bitmap
; /* one bit per page */
4997 struct compat_kvm_clear_dirty_log
{
5002 compat_uptr_t dirty_bitmap
; /* one bit per page */
5007 long __weak
kvm_arch_vm_compat_ioctl(struct file
*filp
, unsigned int ioctl
,
5013 static long kvm_vm_compat_ioctl(struct file
*filp
,
5014 unsigned int ioctl
, unsigned long arg
)
5016 struct kvm
*kvm
= filp
->private_data
;
5019 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5022 r
= kvm_arch_vm_compat_ioctl(filp
, ioctl
, arg
);
5027 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5028 case KVM_CLEAR_DIRTY_LOG
: {
5029 struct compat_kvm_clear_dirty_log compat_log
;
5030 struct kvm_clear_dirty_log log
;
5032 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5033 sizeof(compat_log
)))
5035 log
.slot
= compat_log
.slot
;
5036 log
.num_pages
= compat_log
.num_pages
;
5037 log
.first_page
= compat_log
.first_page
;
5038 log
.padding2
= compat_log
.padding2
;
5039 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5041 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5045 case KVM_GET_DIRTY_LOG
: {
5046 struct compat_kvm_dirty_log compat_log
;
5047 struct kvm_dirty_log log
;
5049 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5050 sizeof(compat_log
)))
5052 log
.slot
= compat_log
.slot
;
5053 log
.padding1
= compat_log
.padding1
;
5054 log
.padding2
= compat_log
.padding2
;
5055 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5057 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5061 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
5067 static const struct file_operations kvm_vm_fops
= {
5068 .release
= kvm_vm_release
,
5069 .unlocked_ioctl
= kvm_vm_ioctl
,
5070 .llseek
= noop_llseek
,
5071 KVM_COMPAT(kvm_vm_compat_ioctl
),
5074 bool file_is_kvm(struct file
*file
)
5076 return file
&& file
->f_op
== &kvm_vm_fops
;
5078 EXPORT_SYMBOL_GPL(file_is_kvm
);
5080 static int kvm_dev_ioctl_create_vm(unsigned long type
)
5082 char fdname
[ITOA_MAX_LEN
+ 1];
5087 fd
= get_unused_fd_flags(O_CLOEXEC
);
5091 snprintf(fdname
, sizeof(fdname
), "%d", fd
);
5093 kvm
= kvm_create_vm(type
, fdname
);
5099 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
5106 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5107 * already set, with ->release() being kvm_vm_release(). In error
5108 * cases it will be called by the final fput(file) and will take
5109 * care of doing kvm_put_kvm(kvm).
5111 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
5113 fd_install(fd
, file
);
5123 static long kvm_dev_ioctl(struct file
*filp
,
5124 unsigned int ioctl
, unsigned long arg
)
5129 case KVM_GET_API_VERSION
:
5132 r
= KVM_API_VERSION
;
5135 r
= kvm_dev_ioctl_create_vm(arg
);
5137 case KVM_CHECK_EXTENSION
:
5138 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
5140 case KVM_GET_VCPU_MMAP_SIZE
:
5143 r
= PAGE_SIZE
; /* struct kvm_run */
5145 r
+= PAGE_SIZE
; /* pio data page */
5147 #ifdef CONFIG_KVM_MMIO
5148 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
5151 case KVM_TRACE_ENABLE
:
5152 case KVM_TRACE_PAUSE
:
5153 case KVM_TRACE_DISABLE
:
5157 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
5163 static struct file_operations kvm_chardev_ops
= {
5164 .unlocked_ioctl
= kvm_dev_ioctl
,
5165 .llseek
= noop_llseek
,
5166 KVM_COMPAT(kvm_dev_ioctl
),
5169 static struct miscdevice kvm_dev
= {
5175 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5176 __visible
bool kvm_rebooting
;
5177 EXPORT_SYMBOL_GPL(kvm_rebooting
);
5179 static DEFINE_PER_CPU(bool, hardware_enabled
);
5180 static int kvm_usage_count
;
5182 static int __hardware_enable_nolock(void)
5184 if (__this_cpu_read(hardware_enabled
))
5187 if (kvm_arch_hardware_enable()) {
5188 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5189 raw_smp_processor_id());
5193 __this_cpu_write(hardware_enabled
, true);
5197 static void hardware_enable_nolock(void *failed
)
5199 if (__hardware_enable_nolock())
5203 static int kvm_online_cpu(unsigned int cpu
)
5208 * Abort the CPU online process if hardware virtualization cannot
5209 * be enabled. Otherwise running VMs would encounter unrecoverable
5210 * errors when scheduled to this CPU.
5212 mutex_lock(&kvm_lock
);
5213 if (kvm_usage_count
)
5214 ret
= __hardware_enable_nolock();
5215 mutex_unlock(&kvm_lock
);
5219 static void hardware_disable_nolock(void *junk
)
5222 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5223 * hardware, not just CPUs that successfully enabled hardware!
5225 if (!__this_cpu_read(hardware_enabled
))
5228 kvm_arch_hardware_disable();
5230 __this_cpu_write(hardware_enabled
, false);
5233 static int kvm_offline_cpu(unsigned int cpu
)
5235 mutex_lock(&kvm_lock
);
5236 if (kvm_usage_count
)
5237 hardware_disable_nolock(NULL
);
5238 mutex_unlock(&kvm_lock
);
5242 static void hardware_disable_all_nolock(void)
5244 BUG_ON(!kvm_usage_count
);
5247 if (!kvm_usage_count
)
5248 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5251 static void hardware_disable_all(void)
5254 mutex_lock(&kvm_lock
);
5255 hardware_disable_all_nolock();
5256 mutex_unlock(&kvm_lock
);
5260 static int hardware_enable_all(void)
5262 atomic_t failed
= ATOMIC_INIT(0);
5266 * Do not enable hardware virtualization if the system is going down.
5267 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5268 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5269 * after kvm_reboot() is called. Note, this relies on system_state
5270 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5271 * hook instead of registering a dedicated reboot notifier (the latter
5272 * runs before system_state is updated).
5274 if (system_state
== SYSTEM_HALT
|| system_state
== SYSTEM_POWER_OFF
||
5275 system_state
== SYSTEM_RESTART
)
5279 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5280 * is called, and so on_each_cpu() between them includes the CPU that
5281 * is being onlined. As a result, hardware_enable_nolock() may get
5282 * invoked before kvm_online_cpu(), which also enables hardware if the
5283 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5284 * enable hardware multiple times.
5287 mutex_lock(&kvm_lock
);
5292 if (kvm_usage_count
== 1) {
5293 on_each_cpu(hardware_enable_nolock
, &failed
, 1);
5295 if (atomic_read(&failed
)) {
5296 hardware_disable_all_nolock();
5301 mutex_unlock(&kvm_lock
);
5307 static void kvm_shutdown(void)
5310 * Disable hardware virtualization and set kvm_rebooting to indicate
5311 * that KVM has asynchronously disabled hardware virtualization, i.e.
5312 * that relevant errors and exceptions aren't entirely unexpected.
5313 * Some flavors of hardware virtualization need to be disabled before
5314 * transferring control to firmware (to perform shutdown/reboot), e.g.
5315 * on x86, virtualization can block INIT interrupts, which are used by
5316 * firmware to pull APs back under firmware control. Note, this path
5317 * is used for both shutdown and reboot scenarios, i.e. neither name is
5318 * 100% comprehensive.
5320 pr_info("kvm: exiting hardware virtualization\n");
5321 kvm_rebooting
= true;
5322 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5325 static int kvm_suspend(void)
5328 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5329 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5330 * is stable. Assert that kvm_lock is not held to ensure the system
5331 * isn't suspended while KVM is enabling hardware. Hardware enabling
5332 * can be preempted, but the task cannot be frozen until it has dropped
5333 * all locks (userspace tasks are frozen via a fake signal).
5335 lockdep_assert_not_held(&kvm_lock
);
5336 lockdep_assert_irqs_disabled();
5338 if (kvm_usage_count
)
5339 hardware_disable_nolock(NULL
);
5343 static void kvm_resume(void)
5345 lockdep_assert_not_held(&kvm_lock
);
5346 lockdep_assert_irqs_disabled();
5348 if (kvm_usage_count
)
5349 WARN_ON_ONCE(__hardware_enable_nolock());
5352 static struct syscore_ops kvm_syscore_ops
= {
5353 .suspend
= kvm_suspend
,
5354 .resume
= kvm_resume
,
5355 .shutdown
= kvm_shutdown
,
5357 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5358 static int hardware_enable_all(void)
5363 static void hardware_disable_all(void)
5367 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5369 static void kvm_iodevice_destructor(struct kvm_io_device
*dev
)
5371 if (dev
->ops
->destructor
)
5372 dev
->ops
->destructor(dev
);
5375 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
5379 for (i
= 0; i
< bus
->dev_count
; i
++) {
5380 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
5382 kvm_iodevice_destructor(pos
);
5387 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
5388 const struct kvm_io_range
*r2
)
5390 gpa_t addr1
= r1
->addr
;
5391 gpa_t addr2
= r2
->addr
;
5396 /* If r2->len == 0, match the exact address. If r2->len != 0,
5397 * accept any overlapping write. Any order is acceptable for
5398 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5399 * we process all of them.
5412 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
5414 return kvm_io_bus_cmp(p1
, p2
);
5417 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
5418 gpa_t addr
, int len
)
5420 struct kvm_io_range
*range
, key
;
5423 key
= (struct kvm_io_range
) {
5428 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
5429 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
5433 off
= range
- bus
->range
;
5435 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
5441 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5442 struct kvm_io_range
*range
, const void *val
)
5446 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5450 while (idx
< bus
->dev_count
&&
5451 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5452 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5461 /* kvm_io_bus_write - called under kvm->slots_lock */
5462 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5463 int len
, const void *val
)
5465 struct kvm_io_bus
*bus
;
5466 struct kvm_io_range range
;
5469 range
= (struct kvm_io_range
) {
5474 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5477 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5478 return r
< 0 ? r
: 0;
5480 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
5482 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5483 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
5484 gpa_t addr
, int len
, const void *val
, long cookie
)
5486 struct kvm_io_bus
*bus
;
5487 struct kvm_io_range range
;
5489 range
= (struct kvm_io_range
) {
5494 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5498 /* First try the device referenced by cookie. */
5499 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
5500 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
5501 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
5506 * cookie contained garbage; fall back to search and return the
5507 * correct cookie value.
5509 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5512 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5513 struct kvm_io_range
*range
, void *val
)
5517 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5521 while (idx
< bus
->dev_count
&&
5522 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5523 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5532 /* kvm_io_bus_read - called under kvm->slots_lock */
5533 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5536 struct kvm_io_bus
*bus
;
5537 struct kvm_io_range range
;
5540 range
= (struct kvm_io_range
) {
5545 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5548 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
5549 return r
< 0 ? r
: 0;
5552 /* Caller must hold slots_lock. */
5553 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
5554 int len
, struct kvm_io_device
*dev
)
5557 struct kvm_io_bus
*new_bus
, *bus
;
5558 struct kvm_io_range range
;
5560 bus
= kvm_get_bus(kvm
, bus_idx
);
5564 /* exclude ioeventfd which is limited by maximum fd */
5565 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
5568 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
5569 GFP_KERNEL_ACCOUNT
);
5573 range
= (struct kvm_io_range
) {
5579 for (i
= 0; i
< bus
->dev_count
; i
++)
5580 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
5583 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
5584 new_bus
->dev_count
++;
5585 new_bus
->range
[i
] = range
;
5586 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
5587 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
5588 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5589 synchronize_srcu_expedited(&kvm
->srcu
);
5595 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5596 struct kvm_io_device
*dev
)
5599 struct kvm_io_bus
*new_bus
, *bus
;
5601 lockdep_assert_held(&kvm
->slots_lock
);
5603 bus
= kvm_get_bus(kvm
, bus_idx
);
5607 for (i
= 0; i
< bus
->dev_count
; i
++) {
5608 if (bus
->range
[i
].dev
== dev
) {
5613 if (i
== bus
->dev_count
)
5616 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
5617 GFP_KERNEL_ACCOUNT
);
5619 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
5620 new_bus
->dev_count
--;
5621 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
5622 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
5625 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5626 synchronize_srcu_expedited(&kvm
->srcu
);
5629 * If NULL bus is installed, destroy the old bus, including all the
5630 * attached devices. Otherwise, destroy the caller's device only.
5633 pr_err("kvm: failed to shrink bus, removing it completely\n");
5634 kvm_io_bus_destroy(bus
);
5638 kvm_iodevice_destructor(dev
);
5643 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5646 struct kvm_io_bus
*bus
;
5647 int dev_idx
, srcu_idx
;
5648 struct kvm_io_device
*iodev
= NULL
;
5650 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5652 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
5656 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
5660 iodev
= bus
->range
[dev_idx
].dev
;
5663 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
5667 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
5669 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
5670 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
5674 struct kvm_stat_data
*stat_data
= inode
->i_private
;
5677 * The debugfs files are a reference to the kvm struct which
5678 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5679 * avoids the race between open and the removal of the debugfs directory.
5681 if (!kvm_get_kvm_safe(stat_data
->kvm
))
5684 ret
= simple_attr_open(inode
, file
, get
,
5685 kvm_stats_debugfs_mode(stat_data
->desc
) & 0222
5688 kvm_put_kvm(stat_data
->kvm
);
5693 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
5695 struct kvm_stat_data
*stat_data
= inode
->i_private
;
5697 simple_attr_release(inode
, file
);
5698 kvm_put_kvm(stat_data
->kvm
);
5703 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
5705 *val
= *(u64
*)((void *)(&kvm
->stat
) + offset
);
5710 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
5712 *(u64
*)((void *)(&kvm
->stat
) + offset
) = 0;
5717 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
5720 struct kvm_vcpu
*vcpu
;
5724 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5725 *val
+= *(u64
*)((void *)(&vcpu
->stat
) + offset
);
5730 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
5733 struct kvm_vcpu
*vcpu
;
5735 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5736 *(u64
*)((void *)(&vcpu
->stat
) + offset
) = 0;
5741 static int kvm_stat_data_get(void *data
, u64
*val
)
5744 struct kvm_stat_data
*stat_data
= data
;
5746 switch (stat_data
->kind
) {
5748 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
5749 stat_data
->desc
->desc
.offset
, val
);
5752 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
5753 stat_data
->desc
->desc
.offset
, val
);
5760 static int kvm_stat_data_clear(void *data
, u64 val
)
5763 struct kvm_stat_data
*stat_data
= data
;
5768 switch (stat_data
->kind
) {
5770 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
5771 stat_data
->desc
->desc
.offset
);
5774 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
5775 stat_data
->desc
->desc
.offset
);
5782 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
5784 __simple_attr_check_format("%llu\n", 0ull);
5785 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
5786 kvm_stat_data_clear
, "%llu\n");
5789 static const struct file_operations stat_fops_per_vm
= {
5790 .owner
= THIS_MODULE
,
5791 .open
= kvm_stat_data_open
,
5792 .release
= kvm_debugfs_release
,
5793 .read
= simple_attr_read
,
5794 .write
= simple_attr_write
,
5795 .llseek
= no_llseek
,
5798 static int vm_stat_get(void *_offset
, u64
*val
)
5800 unsigned offset
= (long)_offset
;
5805 mutex_lock(&kvm_lock
);
5806 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5807 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
5810 mutex_unlock(&kvm_lock
);
5814 static int vm_stat_clear(void *_offset
, u64 val
)
5816 unsigned offset
= (long)_offset
;
5822 mutex_lock(&kvm_lock
);
5823 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5824 kvm_clear_stat_per_vm(kvm
, offset
);
5826 mutex_unlock(&kvm_lock
);
5831 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
5832 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops
, vm_stat_get
, NULL
, "%llu\n");
5834 static int vcpu_stat_get(void *_offset
, u64
*val
)
5836 unsigned offset
= (long)_offset
;
5841 mutex_lock(&kvm_lock
);
5842 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5843 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
5846 mutex_unlock(&kvm_lock
);
5850 static int vcpu_stat_clear(void *_offset
, u64 val
)
5852 unsigned offset
= (long)_offset
;
5858 mutex_lock(&kvm_lock
);
5859 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5860 kvm_clear_stat_per_vcpu(kvm
, offset
);
5862 mutex_unlock(&kvm_lock
);
5867 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
5869 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops
, vcpu_stat_get
, NULL
, "%llu\n");
5871 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
5873 struct kobj_uevent_env
*env
;
5874 unsigned long long created
, active
;
5876 if (!kvm_dev
.this_device
|| !kvm
)
5879 mutex_lock(&kvm_lock
);
5880 if (type
== KVM_EVENT_CREATE_VM
) {
5881 kvm_createvm_count
++;
5883 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5886 created
= kvm_createvm_count
;
5887 active
= kvm_active_vms
;
5888 mutex_unlock(&kvm_lock
);
5890 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
5894 add_uevent_var(env
, "CREATED=%llu", created
);
5895 add_uevent_var(env
, "COUNT=%llu", active
);
5897 if (type
== KVM_EVENT_CREATE_VM
) {
5898 add_uevent_var(env
, "EVENT=create");
5899 kvm
->userspace_pid
= task_pid_nr(current
);
5900 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5901 add_uevent_var(env
, "EVENT=destroy");
5903 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
5905 if (!IS_ERR(kvm
->debugfs_dentry
)) {
5906 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
5909 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
5911 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
5915 /* no need for checks, since we are adding at most only 5 keys */
5916 env
->envp
[env
->envp_idx
++] = NULL
;
5917 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
5921 static void kvm_init_debug(void)
5923 const struct file_operations
*fops
;
5924 const struct _kvm_stats_desc
*pdesc
;
5927 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
5929 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
5930 pdesc
= &kvm_vm_stats_desc
[i
];
5931 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
5932 fops
= &vm_stat_fops
;
5934 fops
= &vm_stat_readonly_fops
;
5935 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
5937 (void *)(long)pdesc
->desc
.offset
, fops
);
5940 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
5941 pdesc
= &kvm_vcpu_stats_desc
[i
];
5942 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
5943 fops
= &vcpu_stat_fops
;
5945 fops
= &vcpu_stat_readonly_fops
;
5946 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
5948 (void *)(long)pdesc
->desc
.offset
, fops
);
5953 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
5955 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
5958 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
5960 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
5962 WRITE_ONCE(vcpu
->preempted
, false);
5963 WRITE_ONCE(vcpu
->ready
, false);
5965 __this_cpu_write(kvm_running_vcpu
, vcpu
);
5966 kvm_arch_sched_in(vcpu
, cpu
);
5967 kvm_arch_vcpu_load(vcpu
, cpu
);
5970 static void kvm_sched_out(struct preempt_notifier
*pn
,
5971 struct task_struct
*next
)
5973 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
5975 if (current
->on_rq
) {
5976 WRITE_ONCE(vcpu
->preempted
, true);
5977 WRITE_ONCE(vcpu
->ready
, true);
5979 kvm_arch_vcpu_put(vcpu
);
5980 __this_cpu_write(kvm_running_vcpu
, NULL
);
5984 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5986 * We can disable preemption locally around accessing the per-CPU variable,
5987 * and use the resolved vcpu pointer after enabling preemption again,
5988 * because even if the current thread is migrated to another CPU, reading
5989 * the per-CPU value later will give us the same value as we update the
5990 * per-CPU variable in the preempt notifier handlers.
5992 struct kvm_vcpu
*kvm_get_running_vcpu(void)
5994 struct kvm_vcpu
*vcpu
;
5997 vcpu
= __this_cpu_read(kvm_running_vcpu
);
6002 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
6005 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
6007 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
6009 return &kvm_running_vcpu
;
6012 #ifdef CONFIG_GUEST_PERF_EVENTS
6013 static unsigned int kvm_guest_state(void)
6015 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6018 if (!kvm_arch_pmi_in_guest(vcpu
))
6021 state
= PERF_GUEST_ACTIVE
;
6022 if (!kvm_arch_vcpu_in_kernel(vcpu
))
6023 state
|= PERF_GUEST_USER
;
6028 static unsigned long kvm_guest_get_ip(void)
6030 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6032 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6033 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu
)))
6036 return kvm_arch_vcpu_get_ip(vcpu
);
6039 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6040 .state
= kvm_guest_state
,
6041 .get_ip
= kvm_guest_get_ip
,
6042 .handle_intel_pt_intr
= NULL
,
6045 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler
)(void))
6047 kvm_guest_cbs
.handle_intel_pt_intr
= pt_intr_handler
;
6048 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6050 void kvm_unregister_perf_callbacks(void)
6052 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6056 int kvm_init(unsigned vcpu_size
, unsigned vcpu_align
, struct module
*module
)
6061 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6062 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE
, "kvm/cpu:online",
6063 kvm_online_cpu
, kvm_offline_cpu
);
6067 register_syscore_ops(&kvm_syscore_ops
);
6070 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6072 vcpu_align
= __alignof__(struct kvm_vcpu
);
6074 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
6076 offsetof(struct kvm_vcpu
, arch
),
6077 offsetofend(struct kvm_vcpu
, stats_id
)
6078 - offsetof(struct kvm_vcpu
, arch
),
6080 if (!kvm_vcpu_cache
) {
6082 goto err_vcpu_cache
;
6085 for_each_possible_cpu(cpu
) {
6086 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask
, cpu
),
6087 GFP_KERNEL
, cpu_to_node(cpu
))) {
6089 goto err_cpu_kick_mask
;
6093 r
= kvm_irqfd_init();
6097 r
= kvm_async_pf_init();
6101 kvm_chardev_ops
.owner
= module
;
6103 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
6104 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
6108 r
= kvm_vfio_ops_init();
6109 if (WARN_ON_ONCE(r
))
6113 * Registration _must_ be the very last thing done, as this exposes
6114 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6116 r
= misc_register(&kvm_dev
);
6118 pr_err("kvm: misc device register failed\n");
6125 kvm_vfio_ops_exit();
6127 kvm_async_pf_deinit();
6132 for_each_possible_cpu(cpu
)
6133 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6134 kmem_cache_destroy(kvm_vcpu_cache
);
6136 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6137 unregister_syscore_ops(&kvm_syscore_ops
);
6138 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6142 EXPORT_SYMBOL_GPL(kvm_init
);
6149 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6150 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6151 * to KVM while the module is being stopped.
6153 misc_deregister(&kvm_dev
);
6155 debugfs_remove_recursive(kvm_debugfs_dir
);
6156 for_each_possible_cpu(cpu
)
6157 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6158 kmem_cache_destroy(kvm_vcpu_cache
);
6159 kvm_vfio_ops_exit();
6160 kvm_async_pf_deinit();
6161 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6162 unregister_syscore_ops(&kvm_syscore_ops
);
6163 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6167 EXPORT_SYMBOL_GPL(kvm_exit
);
6169 struct kvm_vm_worker_thread_context
{
6171 struct task_struct
*parent
;
6172 struct completion init_done
;
6173 kvm_vm_thread_fn_t thread_fn
;
6178 static int kvm_vm_worker_thread(void *context
)
6181 * The init_context is allocated on the stack of the parent thread, so
6182 * we have to locally copy anything that is needed beyond initialization
6184 struct kvm_vm_worker_thread_context
*init_context
= context
;
6185 struct task_struct
*parent
;
6186 struct kvm
*kvm
= init_context
->kvm
;
6187 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
6188 uintptr_t data
= init_context
->data
;
6191 err
= kthread_park(current
);
6192 /* kthread_park(current) is never supposed to return an error */
6197 err
= cgroup_attach_task_all(init_context
->parent
, current
);
6199 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6204 set_user_nice(current
, task_nice(init_context
->parent
));
6207 init_context
->err
= err
;
6208 complete(&init_context
->init_done
);
6209 init_context
= NULL
;
6214 /* Wait to be woken up by the spawner before proceeding. */
6217 if (!kthread_should_stop())
6218 err
= thread_fn(kvm
, data
);
6222 * Move kthread back to its original cgroup to prevent it lingering in
6223 * the cgroup of the VM process, after the latter finishes its
6226 * kthread_stop() waits on the 'exited' completion condition which is
6227 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6228 * kthread is removed from the cgroup in the cgroup_exit() which is
6229 * called after the exit_mm(). This causes the kthread_stop() to return
6230 * before the kthread actually quits the cgroup.
6233 parent
= rcu_dereference(current
->real_parent
);
6234 get_task_struct(parent
);
6236 cgroup_attach_task_all(parent
, current
);
6237 put_task_struct(parent
);
6242 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
6243 uintptr_t data
, const char *name
,
6244 struct task_struct
**thread_ptr
)
6246 struct kvm_vm_worker_thread_context init_context
= {};
6247 struct task_struct
*thread
;
6250 init_context
.kvm
= kvm
;
6251 init_context
.parent
= current
;
6252 init_context
.thread_fn
= thread_fn
;
6253 init_context
.data
= data
;
6254 init_completion(&init_context
.init_done
);
6256 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
6257 "%s-%d", name
, task_pid_nr(current
));
6259 return PTR_ERR(thread
);
6261 /* kthread_run is never supposed to return NULL */
6262 WARN_ON(thread
== NULL
);
6264 wait_for_completion(&init_context
.init_done
);
6266 if (!init_context
.err
)
6267 *thread_ptr
= thread
;
6269 return init_context
.err
;