1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #include <trace/events/ipi.h>
67 #define CREATE_TRACE_POINTS
68 #include <trace/events/kvm.h>
70 #include <linux/kvm_dirty_ring.h>
73 /* Worst case buffer size needed for holding an integer. */
74 #define ITOA_MAX_LEN 12
76 MODULE_AUTHOR("Qumranet");
77 MODULE_LICENSE("GPL");
79 /* Architectures should define their poll value according to the halt latency */
80 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
81 module_param(halt_poll_ns
, uint
, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns
);
84 /* Default doubles per-vcpu halt_poll_ns. */
85 unsigned int halt_poll_ns_grow
= 2;
86 module_param(halt_poll_ns_grow
, uint
, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
89 /* The start value to grow halt_poll_ns from */
90 unsigned int halt_poll_ns_grow_start
= 10000; /* 10us */
91 module_param(halt_poll_ns_grow_start
, uint
, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start
);
94 /* Default resets per-vcpu halt_poll_ns . */
95 unsigned int halt_poll_ns_shrink
;
96 module_param(halt_poll_ns_shrink
, uint
, 0644);
97 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
102 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
105 DEFINE_MUTEX(kvm_lock
);
108 static struct kmem_cache
*kvm_vcpu_cache
;
110 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
111 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_running_vcpu
);
113 struct dentry
*kvm_debugfs_dir
;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
116 static const struct file_operations stat_fops_per_vm
;
118 static struct file_operations kvm_chardev_ops
;
120 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file
*file
, unsigned int ioctl
,
135 unsigned long arg
) { return -EINVAL
; }
137 static int kvm_no_compat_open(struct inode
*inode
, struct file
*file
)
139 return is_compat_task() ? -ENODEV
: 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
149 #define KVM_EVENT_CREATE_VM 0
150 #define KVM_EVENT_DESTROY_VM 1
151 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
152 static unsigned long long kvm_createvm_count
;
153 static unsigned long long kvm_active_vms
;
155 static DEFINE_PER_CPU(cpumask_var_t
, cpu_kick_mask
);
157 __weak
void kvm_arch_guest_memory_reclaimed(struct kvm
*kvm
)
161 bool kvm_is_zone_device_page(struct page
*page
)
164 * The metadata used by is_zone_device_page() to determine whether or
165 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
166 * the device has been pinned, e.g. by get_user_pages(). WARN if the
167 * page_count() is zero to help detect bad usage of this helper.
169 if (WARN_ON_ONCE(!page_count(page
)))
172 return is_zone_device_page(page
);
176 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
177 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
178 * is likely incomplete, it has been compiled purely through people wanting to
179 * back guest with a certain type of memory and encountering issues.
181 struct page
*kvm_pfn_to_refcounted_page(kvm_pfn_t pfn
)
188 page
= pfn_to_page(pfn
);
189 if (!PageReserved(page
))
192 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
193 if (is_zero_pfn(pfn
))
197 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
198 * perspective they are "normal" pages, albeit with slightly different
201 if (kvm_is_zone_device_page(page
))
208 * Switches to specified vcpu, until a matching vcpu_put()
210 void vcpu_load(struct kvm_vcpu
*vcpu
)
214 __this_cpu_write(kvm_running_vcpu
, vcpu
);
215 preempt_notifier_register(&vcpu
->preempt_notifier
);
216 kvm_arch_vcpu_load(vcpu
, cpu
);
219 EXPORT_SYMBOL_GPL(vcpu_load
);
221 void vcpu_put(struct kvm_vcpu
*vcpu
)
224 kvm_arch_vcpu_put(vcpu
);
225 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
226 __this_cpu_write(kvm_running_vcpu
, NULL
);
229 EXPORT_SYMBOL_GPL(vcpu_put
);
231 /* TODO: merge with kvm_arch_vcpu_should_kick */
232 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
234 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
237 * We need to wait for the VCPU to reenable interrupts and get out of
238 * READING_SHADOW_PAGE_TABLES mode.
240 if (req
& KVM_REQUEST_WAIT
)
241 return mode
!= OUTSIDE_GUEST_MODE
;
244 * Need to kick a running VCPU, but otherwise there is nothing to do.
246 return mode
== IN_GUEST_MODE
;
249 static void ack_kick(void *_completed
)
253 static inline bool kvm_kick_many_cpus(struct cpumask
*cpus
, bool wait
)
255 if (cpumask_empty(cpus
))
258 smp_call_function_many(cpus
, ack_kick
, NULL
, wait
);
262 static void kvm_make_vcpu_request(struct kvm_vcpu
*vcpu
, unsigned int req
,
263 struct cpumask
*tmp
, int current_cpu
)
267 if (likely(!(req
& KVM_REQUEST_NO_ACTION
)))
268 __kvm_make_request(req
, vcpu
);
270 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
274 * Note, the vCPU could get migrated to a different pCPU at any point
275 * after kvm_request_needs_ipi(), which could result in sending an IPI
276 * to the previous pCPU. But, that's OK because the purpose of the IPI
277 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
278 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
279 * after this point is also OK, as the requirement is only that KVM wait
280 * for vCPUs that were reading SPTEs _before_ any changes were
281 * finalized. See kvm_vcpu_kick() for more details on handling requests.
283 if (kvm_request_needs_ipi(vcpu
, req
)) {
284 cpu
= READ_ONCE(vcpu
->cpu
);
285 if (cpu
!= -1 && cpu
!= current_cpu
)
286 __cpumask_set_cpu(cpu
, tmp
);
290 bool kvm_make_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
291 unsigned long *vcpu_bitmap
)
293 struct kvm_vcpu
*vcpu
;
294 struct cpumask
*cpus
;
300 cpus
= this_cpu_cpumask_var_ptr(cpu_kick_mask
);
303 for_each_set_bit(i
, vcpu_bitmap
, KVM_MAX_VCPUS
) {
304 vcpu
= kvm_get_vcpu(kvm
, i
);
307 kvm_make_vcpu_request(vcpu
, req
, cpus
, me
);
310 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
316 bool kvm_make_all_cpus_request_except(struct kvm
*kvm
, unsigned int req
,
317 struct kvm_vcpu
*except
)
319 struct kvm_vcpu
*vcpu
;
320 struct cpumask
*cpus
;
327 cpus
= this_cpu_cpumask_var_ptr(cpu_kick_mask
);
330 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
333 kvm_make_vcpu_request(vcpu
, req
, cpus
, me
);
336 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
342 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
344 return kvm_make_all_cpus_request_except(kvm
, req
, NULL
);
346 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request
);
348 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
350 ++kvm
->stat
.generic
.remote_tlb_flush_requests
;
353 * We want to publish modifications to the page tables before reading
354 * mode. Pairs with a memory barrier in arch-specific code.
355 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
356 * and smp_mb in walk_shadow_page_lockless_begin/end.
357 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
359 * There is already an smp_mb__after_atomic() before
360 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
363 if (!kvm_arch_flush_remote_tlbs(kvm
)
364 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
365 ++kvm
->stat
.generic
.remote_tlb_flush
;
367 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
369 void kvm_flush_remote_tlbs_range(struct kvm
*kvm
, gfn_t gfn
, u64 nr_pages
)
371 if (!kvm_arch_flush_remote_tlbs_range(kvm
, gfn
, nr_pages
))
375 * Fall back to a flushing entire TLBs if the architecture range-based
376 * TLB invalidation is unsupported or can't be performed for whatever
379 kvm_flush_remote_tlbs(kvm
);
382 void kvm_flush_remote_tlbs_memslot(struct kvm
*kvm
,
383 const struct kvm_memory_slot
*memslot
)
386 * All current use cases for flushing the TLBs for a specific memslot
387 * are related to dirty logging, and many do the TLB flush out of
388 * mmu_lock. The interaction between the various operations on memslot
389 * must be serialized by slots_locks to ensure the TLB flush from one
390 * operation is observed by any other operation on the same memslot.
392 lockdep_assert_held(&kvm
->slots_lock
);
393 kvm_flush_remote_tlbs_range(kvm
, memslot
->base_gfn
, memslot
->npages
);
396 static void kvm_flush_shadow_all(struct kvm
*kvm
)
398 kvm_arch_flush_shadow_all(kvm
);
399 kvm_arch_guest_memory_reclaimed(kvm
);
402 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
403 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache
*mc
,
406 gfp_flags
|= mc
->gfp_zero
;
409 return kmem_cache_alloc(mc
->kmem_cache
, gfp_flags
);
411 return (void *)__get_free_page(gfp_flags
);
414 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int capacity
, int min
)
416 gfp_t gfp
= mc
->gfp_custom
? mc
->gfp_custom
: GFP_KERNEL_ACCOUNT
;
419 if (mc
->nobjs
>= min
)
422 if (unlikely(!mc
->objects
)) {
423 if (WARN_ON_ONCE(!capacity
))
426 mc
->objects
= kvmalloc_array(sizeof(void *), capacity
, gfp
);
430 mc
->capacity
= capacity
;
433 /* It is illegal to request a different capacity across topups. */
434 if (WARN_ON_ONCE(mc
->capacity
!= capacity
))
437 while (mc
->nobjs
< mc
->capacity
) {
438 obj
= mmu_memory_cache_alloc_obj(mc
, gfp
);
440 return mc
->nobjs
>= min
? 0 : -ENOMEM
;
441 mc
->objects
[mc
->nobjs
++] = obj
;
446 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int min
)
448 return __kvm_mmu_topup_memory_cache(mc
, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
, min
);
451 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache
*mc
)
456 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
460 kmem_cache_free(mc
->kmem_cache
, mc
->objects
[--mc
->nobjs
]);
462 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
471 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
)
475 if (WARN_ON(!mc
->nobjs
))
476 p
= mmu_memory_cache_alloc_obj(mc
, GFP_ATOMIC
| __GFP_ACCOUNT
);
478 p
= mc
->objects
[--mc
->nobjs
];
484 static void kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
486 mutex_init(&vcpu
->mutex
);
491 #ifndef __KVM_HAVE_ARCH_WQP
492 rcuwait_init(&vcpu
->wait
);
494 kvm_async_pf_vcpu_init(vcpu
);
496 kvm_vcpu_set_in_spin_loop(vcpu
, false);
497 kvm_vcpu_set_dy_eligible(vcpu
, false);
498 vcpu
->preempted
= false;
500 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
501 vcpu
->last_used_slot
= NULL
;
503 /* Fill the stats id string for the vcpu */
504 snprintf(vcpu
->stats_id
, sizeof(vcpu
->stats_id
), "kvm-%d/vcpu-%d",
505 task_pid_nr(current
), id
);
508 static void kvm_vcpu_destroy(struct kvm_vcpu
*vcpu
)
510 kvm_arch_vcpu_destroy(vcpu
);
511 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
514 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
515 * the vcpu->pid pointer, and at destruction time all file descriptors
518 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
520 free_page((unsigned long)vcpu
->run
);
521 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
524 void kvm_destroy_vcpus(struct kvm
*kvm
)
527 struct kvm_vcpu
*vcpu
;
529 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
530 kvm_vcpu_destroy(vcpu
);
531 xa_erase(&kvm
->vcpu_array
, i
);
534 atomic_set(&kvm
->online_vcpus
, 0);
536 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus
);
538 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
539 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
541 return container_of(mn
, struct kvm
, mmu_notifier
);
544 typedef bool (*gfn_handler_t
)(struct kvm
*kvm
, struct kvm_gfn_range
*range
);
546 typedef void (*on_lock_fn_t
)(struct kvm
*kvm
);
548 struct kvm_mmu_notifier_range
{
550 * 64-bit addresses, as KVM notifiers can operate on host virtual
551 * addresses (unsigned long) and guest physical addresses (64-bit).
555 union kvm_mmu_notifier_arg arg
;
556 gfn_handler_t handler
;
557 on_lock_fn_t on_lock
;
563 * The inner-most helper returns a tuple containing the return value from the
564 * arch- and action-specific handler, plus a flag indicating whether or not at
565 * least one memslot was found, i.e. if the handler found guest memory.
567 * Note, most notifiers are averse to booleans, so even though KVM tracks the
568 * return from arch code as a bool, outer helpers will cast it to an int. :-(
570 typedef struct kvm_mmu_notifier_return
{
576 * Use a dedicated stub instead of NULL to indicate that there is no callback
577 * function/handler. The compiler technically can't guarantee that a real
578 * function will have a non-zero address, and so it will generate code to
579 * check for !NULL, whereas comparing against a stub will be elided at compile
580 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
582 static void kvm_null_fn(void)
586 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
588 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG
;
590 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
591 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
592 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
594 node = interval_tree_iter_next(node, start, last)) \
596 static __always_inline kvm_mn_ret_t __kvm_handle_hva_range(struct kvm *kvm,
597 const struct kvm_mmu_notifier_range
*range
)
599 struct kvm_mmu_notifier_return r
= {
601 .found_memslot
= false,
603 struct kvm_gfn_range gfn_range
;
604 struct kvm_memory_slot
*slot
;
605 struct kvm_memslots
*slots
;
608 if (WARN_ON_ONCE(range
->end
<= range
->start
))
611 /* A null handler is allowed if and only if on_lock() is provided. */
612 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range
->on_lock
) &&
613 IS_KVM_NULL_FN(range
->handler
)))
616 idx
= srcu_read_lock(&kvm
->srcu
);
618 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
619 struct interval_tree_node
*node
;
621 slots
= __kvm_memslots(kvm
, i
);
622 kvm_for_each_memslot_in_hva_range(node
, slots
,
623 range
->start
, range
->end
- 1) {
624 unsigned long hva_start
, hva_end
;
626 slot
= container_of(node
, struct kvm_memory_slot
, hva_node
[slots
->node_idx
]);
627 hva_start
= max_t(unsigned long, range
->start
, slot
->userspace_addr
);
628 hva_end
= min_t(unsigned long, range
->end
,
629 slot
->userspace_addr
+ (slot
->npages
<< PAGE_SHIFT
));
632 * To optimize for the likely case where the address
633 * range is covered by zero or one memslots, don't
634 * bother making these conditional (to avoid writes on
635 * the second or later invocation of the handler).
637 gfn_range
.arg
= range
->arg
;
638 gfn_range
.may_block
= range
->may_block
;
641 * {gfn(page) | page intersects with [hva_start, hva_end)} =
642 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
644 gfn_range
.start
= hva_to_gfn_memslot(hva_start
, slot
);
645 gfn_range
.end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, slot
);
646 gfn_range
.slot
= slot
;
648 if (!r
.found_memslot
) {
649 r
.found_memslot
= true;
651 if (!IS_KVM_NULL_FN(range
->on_lock
))
654 if (IS_KVM_NULL_FN(range
->handler
))
657 r
.ret
|= range
->handler(kvm
, &gfn_range
);
661 if (range
->flush_on_ret
&& r
.ret
)
662 kvm_flush_remote_tlbs(kvm
);
667 srcu_read_unlock(&kvm
->srcu
, idx
);
672 static __always_inline
int kvm_handle_hva_range(struct mmu_notifier
*mn
,
675 union kvm_mmu_notifier_arg arg
,
676 gfn_handler_t handler
)
678 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
679 const struct kvm_mmu_notifier_range range
= {
684 .on_lock
= (void *)kvm_null_fn
,
685 .flush_on_ret
= true,
689 return __kvm_handle_hva_range(kvm
, &range
).ret
;
692 static __always_inline
int kvm_handle_hva_range_no_flush(struct mmu_notifier
*mn
,
695 gfn_handler_t handler
)
697 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
698 const struct kvm_mmu_notifier_range range
= {
702 .on_lock
= (void *)kvm_null_fn
,
703 .flush_on_ret
= false,
707 return __kvm_handle_hva_range(kvm
, &range
).ret
;
710 static bool kvm_change_spte_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
713 * Skipping invalid memslots is correct if and only change_pte() is
714 * surrounded by invalidate_range_{start,end}(), which is currently
715 * guaranteed by the primary MMU. If that ever changes, KVM needs to
716 * unmap the memslot instead of skipping the memslot to ensure that KVM
717 * doesn't hold references to the old PFN.
719 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
721 if (range
->slot
->flags
& KVM_MEMSLOT_INVALID
)
724 return kvm_set_spte_gfn(kvm
, range
);
727 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
728 struct mm_struct
*mm
,
729 unsigned long address
,
732 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
733 const union kvm_mmu_notifier_arg arg
= { .pte
= pte
};
735 trace_kvm_set_spte_hva(address
);
738 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
739 * If mmu_invalidate_in_progress is zero, then no in-progress
740 * invalidations, including this one, found a relevant memslot at
741 * start(); rechecking memslots here is unnecessary. Note, a false
742 * positive (count elevated by a different invalidation) is sub-optimal
743 * but functionally ok.
745 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
746 if (!READ_ONCE(kvm
->mmu_invalidate_in_progress
))
749 kvm_handle_hva_range(mn
, address
, address
+ 1, arg
, kvm_change_spte_gfn
);
752 void kvm_mmu_invalidate_begin(struct kvm
*kvm
)
754 lockdep_assert_held_write(&kvm
->mmu_lock
);
756 * The count increase must become visible at unlock time as no
757 * spte can be established without taking the mmu_lock and
758 * count is also read inside the mmu_lock critical section.
760 kvm
->mmu_invalidate_in_progress
++;
762 if (likely(kvm
->mmu_invalidate_in_progress
== 1)) {
763 kvm
->mmu_invalidate_range_start
= INVALID_GPA
;
764 kvm
->mmu_invalidate_range_end
= INVALID_GPA
;
768 void kvm_mmu_invalidate_range_add(struct kvm
*kvm
, gfn_t start
, gfn_t end
)
770 lockdep_assert_held_write(&kvm
->mmu_lock
);
772 WARN_ON_ONCE(!kvm
->mmu_invalidate_in_progress
);
774 if (likely(kvm
->mmu_invalidate_range_start
== INVALID_GPA
)) {
775 kvm
->mmu_invalidate_range_start
= start
;
776 kvm
->mmu_invalidate_range_end
= end
;
779 * Fully tracking multiple concurrent ranges has diminishing
780 * returns. Keep things simple and just find the minimal range
781 * which includes the current and new ranges. As there won't be
782 * enough information to subtract a range after its invalidate
783 * completes, any ranges invalidated concurrently will
784 * accumulate and persist until all outstanding invalidates
787 kvm
->mmu_invalidate_range_start
=
788 min(kvm
->mmu_invalidate_range_start
, start
);
789 kvm
->mmu_invalidate_range_end
=
790 max(kvm
->mmu_invalidate_range_end
, end
);
794 bool kvm_mmu_unmap_gfn_range(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
796 kvm_mmu_invalidate_range_add(kvm
, range
->start
, range
->end
);
797 return kvm_unmap_gfn_range(kvm
, range
);
800 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
801 const struct mmu_notifier_range
*range
)
803 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
804 const struct kvm_mmu_notifier_range hva_range
= {
805 .start
= range
->start
,
807 .handler
= kvm_mmu_unmap_gfn_range
,
808 .on_lock
= kvm_mmu_invalidate_begin
,
809 .flush_on_ret
= true,
810 .may_block
= mmu_notifier_range_blockable(range
),
813 trace_kvm_unmap_hva_range(range
->start
, range
->end
);
816 * Prevent memslot modification between range_start() and range_end()
817 * so that conditionally locking provides the same result in both
818 * functions. Without that guarantee, the mmu_invalidate_in_progress
819 * adjustments will be imbalanced.
821 * Pairs with the decrement in range_end().
823 spin_lock(&kvm
->mn_invalidate_lock
);
824 kvm
->mn_active_invalidate_count
++;
825 spin_unlock(&kvm
->mn_invalidate_lock
);
828 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
829 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
830 * each cache's lock. There are relatively few caches in existence at
831 * any given time, and the caches themselves can check for hva overlap,
832 * i.e. don't need to rely on memslot overlap checks for performance.
833 * Because this runs without holding mmu_lock, the pfn caches must use
834 * mn_active_invalidate_count (see above) instead of
835 * mmu_invalidate_in_progress.
837 gfn_to_pfn_cache_invalidate_start(kvm
, range
->start
, range
->end
,
838 hva_range
.may_block
);
841 * If one or more memslots were found and thus zapped, notify arch code
842 * that guest memory has been reclaimed. This needs to be done *after*
843 * dropping mmu_lock, as x86's reclaim path is slooooow.
845 if (__kvm_handle_hva_range(kvm
, &hva_range
).found_memslot
)
846 kvm_arch_guest_memory_reclaimed(kvm
);
851 void kvm_mmu_invalidate_end(struct kvm
*kvm
)
853 lockdep_assert_held_write(&kvm
->mmu_lock
);
856 * This sequence increase will notify the kvm page fault that
857 * the page that is going to be mapped in the spte could have
860 kvm
->mmu_invalidate_seq
++;
863 * The above sequence increase must be visible before the
864 * below count decrease, which is ensured by the smp_wmb above
865 * in conjunction with the smp_rmb in mmu_invalidate_retry().
867 kvm
->mmu_invalidate_in_progress
--;
868 KVM_BUG_ON(kvm
->mmu_invalidate_in_progress
< 0, kvm
);
871 * Assert that at least one range was added between start() and end().
872 * Not adding a range isn't fatal, but it is a KVM bug.
874 WARN_ON_ONCE(kvm
->mmu_invalidate_range_start
== INVALID_GPA
);
877 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
878 const struct mmu_notifier_range
*range
)
880 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
881 const struct kvm_mmu_notifier_range hva_range
= {
882 .start
= range
->start
,
884 .handler
= (void *)kvm_null_fn
,
885 .on_lock
= kvm_mmu_invalidate_end
,
886 .flush_on_ret
= false,
887 .may_block
= mmu_notifier_range_blockable(range
),
891 __kvm_handle_hva_range(kvm
, &hva_range
);
893 /* Pairs with the increment in range_start(). */
894 spin_lock(&kvm
->mn_invalidate_lock
);
895 wake
= (--kvm
->mn_active_invalidate_count
== 0);
896 spin_unlock(&kvm
->mn_invalidate_lock
);
899 * There can only be one waiter, since the wait happens under
903 rcuwait_wake_up(&kvm
->mn_memslots_update_rcuwait
);
906 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
907 struct mm_struct
*mm
,
911 trace_kvm_age_hva(start
, end
);
913 return kvm_handle_hva_range(mn
, start
, end
, KVM_MMU_NOTIFIER_NO_ARG
,
917 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
918 struct mm_struct
*mm
,
922 trace_kvm_age_hva(start
, end
);
925 * Even though we do not flush TLB, this will still adversely
926 * affect performance on pre-Haswell Intel EPT, where there is
927 * no EPT Access Bit to clear so that we have to tear down EPT
928 * tables instead. If we find this unacceptable, we can always
929 * add a parameter to kvm_age_hva so that it effectively doesn't
930 * do anything on clear_young.
932 * Also note that currently we never issue secondary TLB flushes
933 * from clear_young, leaving this job up to the regular system
934 * cadence. If we find this inaccurate, we might come up with a
935 * more sophisticated heuristic later.
937 return kvm_handle_hva_range_no_flush(mn
, start
, end
, kvm_age_gfn
);
940 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
941 struct mm_struct
*mm
,
942 unsigned long address
)
944 trace_kvm_test_age_hva(address
);
946 return kvm_handle_hva_range_no_flush(mn
, address
, address
+ 1,
950 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
951 struct mm_struct
*mm
)
953 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
956 idx
= srcu_read_lock(&kvm
->srcu
);
957 kvm_flush_shadow_all(kvm
);
958 srcu_read_unlock(&kvm
->srcu
, idx
);
961 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
962 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
963 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
964 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
965 .clear_young
= kvm_mmu_notifier_clear_young
,
966 .test_young
= kvm_mmu_notifier_test_young
,
967 .change_pte
= kvm_mmu_notifier_change_pte
,
968 .release
= kvm_mmu_notifier_release
,
971 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
973 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
974 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
977 #else /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */
979 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
984 #endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */
986 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
987 static int kvm_pm_notifier_call(struct notifier_block
*bl
,
991 struct kvm
*kvm
= container_of(bl
, struct kvm
, pm_notifier
);
993 return kvm_arch_pm_notifier(kvm
, state
);
996 static void kvm_init_pm_notifier(struct kvm
*kvm
)
998 kvm
->pm_notifier
.notifier_call
= kvm_pm_notifier_call
;
999 /* Suspend KVM before we suspend ftrace, RCU, etc. */
1000 kvm
->pm_notifier
.priority
= INT_MAX
;
1001 register_pm_notifier(&kvm
->pm_notifier
);
1004 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
1006 unregister_pm_notifier(&kvm
->pm_notifier
);
1008 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
1009 static void kvm_init_pm_notifier(struct kvm
*kvm
)
1013 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
1016 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
1018 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1020 if (!memslot
->dirty_bitmap
)
1023 kvfree(memslot
->dirty_bitmap
);
1024 memslot
->dirty_bitmap
= NULL
;
1027 /* This does not remove the slot from struct kvm_memslots data structures */
1028 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
1030 if (slot
->flags
& KVM_MEM_GUEST_MEMFD
)
1031 kvm_gmem_unbind(slot
);
1033 kvm_destroy_dirty_bitmap(slot
);
1035 kvm_arch_free_memslot(kvm
, slot
);
1040 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
1042 struct hlist_node
*idnode
;
1043 struct kvm_memory_slot
*memslot
;
1047 * The same memslot objects live in both active and inactive sets,
1048 * arbitrarily free using index '1' so the second invocation of this
1049 * function isn't operating over a structure with dangling pointers
1050 * (even though this function isn't actually touching them).
1052 if (!slots
->node_idx
)
1055 hash_for_each_safe(slots
->id_hash
, bkt
, idnode
, memslot
, id_node
[1])
1056 kvm_free_memslot(kvm
, memslot
);
1059 static umode_t
kvm_stats_debugfs_mode(const struct _kvm_stats_desc
*pdesc
)
1061 switch (pdesc
->desc
.flags
& KVM_STATS_TYPE_MASK
) {
1062 case KVM_STATS_TYPE_INSTANT
:
1064 case KVM_STATS_TYPE_CUMULATIVE
:
1065 case KVM_STATS_TYPE_PEAK
:
1072 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
1075 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1076 kvm_vcpu_stats_header
.num_desc
;
1078 if (IS_ERR(kvm
->debugfs_dentry
))
1081 debugfs_remove_recursive(kvm
->debugfs_dentry
);
1083 if (kvm
->debugfs_stat_data
) {
1084 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
1085 kfree(kvm
->debugfs_stat_data
[i
]);
1086 kfree(kvm
->debugfs_stat_data
);
1090 static int kvm_create_vm_debugfs(struct kvm
*kvm
, const char *fdname
)
1092 static DEFINE_MUTEX(kvm_debugfs_lock
);
1093 struct dentry
*dent
;
1094 char dir_name
[ITOA_MAX_LEN
* 2];
1095 struct kvm_stat_data
*stat_data
;
1096 const struct _kvm_stats_desc
*pdesc
;
1097 int i
, ret
= -ENOMEM
;
1098 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1099 kvm_vcpu_stats_header
.num_desc
;
1101 if (!debugfs_initialized())
1104 snprintf(dir_name
, sizeof(dir_name
), "%d-%s", task_pid_nr(current
), fdname
);
1105 mutex_lock(&kvm_debugfs_lock
);
1106 dent
= debugfs_lookup(dir_name
, kvm_debugfs_dir
);
1108 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name
);
1110 mutex_unlock(&kvm_debugfs_lock
);
1113 dent
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
1114 mutex_unlock(&kvm_debugfs_lock
);
1118 kvm
->debugfs_dentry
= dent
;
1119 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
1120 sizeof(*kvm
->debugfs_stat_data
),
1121 GFP_KERNEL_ACCOUNT
);
1122 if (!kvm
->debugfs_stat_data
)
1125 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
1126 pdesc
= &kvm_vm_stats_desc
[i
];
1127 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1131 stat_data
->kvm
= kvm
;
1132 stat_data
->desc
= pdesc
;
1133 stat_data
->kind
= KVM_STAT_VM
;
1134 kvm
->debugfs_stat_data
[i
] = stat_data
;
1135 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1136 kvm
->debugfs_dentry
, stat_data
,
1140 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
1141 pdesc
= &kvm_vcpu_stats_desc
[i
];
1142 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1146 stat_data
->kvm
= kvm
;
1147 stat_data
->desc
= pdesc
;
1148 stat_data
->kind
= KVM_STAT_VCPU
;
1149 kvm
->debugfs_stat_data
[i
+ kvm_vm_stats_header
.num_desc
] = stat_data
;
1150 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1151 kvm
->debugfs_dentry
, stat_data
,
1155 ret
= kvm_arch_create_vm_debugfs(kvm
);
1161 kvm_destroy_vm_debugfs(kvm
);
1166 * Called after the VM is otherwise initialized, but just before adding it to
1169 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
1175 * Called just after removing the VM from the vm_list, but before doing any
1176 * other destruction.
1178 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
1183 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1184 * be setup already, so we can create arch-specific debugfs entries under it.
1185 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1186 * a per-arch destroy interface is not needed.
1188 int __weak
kvm_arch_create_vm_debugfs(struct kvm
*kvm
)
1193 static struct kvm
*kvm_create_vm(unsigned long type
, const char *fdname
)
1195 struct kvm
*kvm
= kvm_arch_alloc_vm();
1196 struct kvm_memslots
*slots
;
1201 return ERR_PTR(-ENOMEM
);
1203 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1204 __module_get(kvm_chardev_ops
.owner
);
1206 KVM_MMU_LOCK_INIT(kvm
);
1207 mmgrab(current
->mm
);
1208 kvm
->mm
= current
->mm
;
1209 kvm_eventfd_init(kvm
);
1210 mutex_init(&kvm
->lock
);
1211 mutex_init(&kvm
->irq_lock
);
1212 mutex_init(&kvm
->slots_lock
);
1213 mutex_init(&kvm
->slots_arch_lock
);
1214 spin_lock_init(&kvm
->mn_invalidate_lock
);
1215 rcuwait_init(&kvm
->mn_memslots_update_rcuwait
);
1216 xa_init(&kvm
->vcpu_array
);
1217 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1218 xa_init(&kvm
->mem_attr_array
);
1221 INIT_LIST_HEAD(&kvm
->gpc_list
);
1222 spin_lock_init(&kvm
->gpc_lock
);
1224 INIT_LIST_HEAD(&kvm
->devices
);
1225 kvm
->max_vcpus
= KVM_MAX_VCPUS
;
1227 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
1230 * Force subsequent debugfs file creations to fail if the VM directory
1231 * is not created (by kvm_create_vm_debugfs()).
1233 kvm
->debugfs_dentry
= ERR_PTR(-ENOENT
);
1235 snprintf(kvm
->stats_id
, sizeof(kvm
->stats_id
), "kvm-%d",
1236 task_pid_nr(current
));
1238 if (init_srcu_struct(&kvm
->srcu
))
1239 goto out_err_no_srcu
;
1240 if (init_srcu_struct(&kvm
->irq_srcu
))
1241 goto out_err_no_irq_srcu
;
1243 refcount_set(&kvm
->users_count
, 1);
1244 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
1245 for (j
= 0; j
< 2; j
++) {
1246 slots
= &kvm
->__memslots
[i
][j
];
1248 atomic_long_set(&slots
->last_used_slot
, (unsigned long)NULL
);
1249 slots
->hva_tree
= RB_ROOT_CACHED
;
1250 slots
->gfn_tree
= RB_ROOT
;
1251 hash_init(slots
->id_hash
);
1252 slots
->node_idx
= j
;
1254 /* Generations must be different for each address space. */
1255 slots
->generation
= i
;
1258 rcu_assign_pointer(kvm
->memslots
[i
], &kvm
->__memslots
[i
][0]);
1261 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1262 rcu_assign_pointer(kvm
->buses
[i
],
1263 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
1265 goto out_err_no_arch_destroy_vm
;
1268 r
= kvm_arch_init_vm(kvm
, type
);
1270 goto out_err_no_arch_destroy_vm
;
1272 r
= hardware_enable_all();
1274 goto out_err_no_disable
;
1276 #ifdef CONFIG_HAVE_KVM_IRQFD
1277 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
1280 r
= kvm_init_mmu_notifier(kvm
);
1282 goto out_err_no_mmu_notifier
;
1284 r
= kvm_coalesced_mmio_init(kvm
);
1286 goto out_no_coalesced_mmio
;
1288 r
= kvm_create_vm_debugfs(kvm
, fdname
);
1290 goto out_err_no_debugfs
;
1292 r
= kvm_arch_post_init_vm(kvm
);
1296 mutex_lock(&kvm_lock
);
1297 list_add(&kvm
->vm_list
, &vm_list
);
1298 mutex_unlock(&kvm_lock
);
1300 preempt_notifier_inc();
1301 kvm_init_pm_notifier(kvm
);
1306 kvm_destroy_vm_debugfs(kvm
);
1308 kvm_coalesced_mmio_free(kvm
);
1309 out_no_coalesced_mmio
:
1310 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1311 if (kvm
->mmu_notifier
.ops
)
1312 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
1314 out_err_no_mmu_notifier
:
1315 hardware_disable_all();
1317 kvm_arch_destroy_vm(kvm
);
1318 out_err_no_arch_destroy_vm
:
1319 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
1320 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
1321 kfree(kvm_get_bus(kvm
, i
));
1322 cleanup_srcu_struct(&kvm
->irq_srcu
);
1323 out_err_no_irq_srcu
:
1324 cleanup_srcu_struct(&kvm
->srcu
);
1326 kvm_arch_free_vm(kvm
);
1327 mmdrop(current
->mm
);
1328 module_put(kvm_chardev_ops
.owner
);
1332 static void kvm_destroy_devices(struct kvm
*kvm
)
1334 struct kvm_device
*dev
, *tmp
;
1337 * We do not need to take the kvm->lock here, because nobody else
1338 * has a reference to the struct kvm at this point and therefore
1339 * cannot access the devices list anyhow.
1341 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
1342 list_del(&dev
->vm_node
);
1343 dev
->ops
->destroy(dev
);
1347 static void kvm_destroy_vm(struct kvm
*kvm
)
1350 struct mm_struct
*mm
= kvm
->mm
;
1352 kvm_destroy_pm_notifier(kvm
);
1353 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
1354 kvm_destroy_vm_debugfs(kvm
);
1355 kvm_arch_sync_events(kvm
);
1356 mutex_lock(&kvm_lock
);
1357 list_del(&kvm
->vm_list
);
1358 mutex_unlock(&kvm_lock
);
1359 kvm_arch_pre_destroy_vm(kvm
);
1361 kvm_free_irq_routing(kvm
);
1362 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1363 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
1366 kvm_io_bus_destroy(bus
);
1367 kvm
->buses
[i
] = NULL
;
1369 kvm_coalesced_mmio_free(kvm
);
1370 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1371 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1373 * At this point, pending calls to invalidate_range_start()
1374 * have completed but no more MMU notifiers will run, so
1375 * mn_active_invalidate_count may remain unbalanced.
1376 * No threads can be waiting in kvm_swap_active_memslots() as the
1377 * last reference on KVM has been dropped, but freeing
1378 * memslots would deadlock without this manual intervention.
1380 * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU
1381 * notifier between a start() and end(), then there shouldn't be any
1382 * in-progress invalidations.
1384 WARN_ON(rcuwait_active(&kvm
->mn_memslots_update_rcuwait
));
1385 if (kvm
->mn_active_invalidate_count
)
1386 kvm
->mn_active_invalidate_count
= 0;
1388 WARN_ON(kvm
->mmu_invalidate_in_progress
);
1390 kvm_flush_shadow_all(kvm
);
1392 kvm_arch_destroy_vm(kvm
);
1393 kvm_destroy_devices(kvm
);
1394 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
1395 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][0]);
1396 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][1]);
1398 cleanup_srcu_struct(&kvm
->irq_srcu
);
1399 cleanup_srcu_struct(&kvm
->srcu
);
1400 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1401 xa_destroy(&kvm
->mem_attr_array
);
1403 kvm_arch_free_vm(kvm
);
1404 preempt_notifier_dec();
1405 hardware_disable_all();
1407 module_put(kvm_chardev_ops
.owner
);
1410 void kvm_get_kvm(struct kvm
*kvm
)
1412 refcount_inc(&kvm
->users_count
);
1414 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1417 * Make sure the vm is not during destruction, which is a safe version of
1418 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1420 bool kvm_get_kvm_safe(struct kvm
*kvm
)
1422 return refcount_inc_not_zero(&kvm
->users_count
);
1424 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe
);
1426 void kvm_put_kvm(struct kvm
*kvm
)
1428 if (refcount_dec_and_test(&kvm
->users_count
))
1429 kvm_destroy_vm(kvm
);
1431 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1434 * Used to put a reference that was taken on behalf of an object associated
1435 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1436 * of the new file descriptor fails and the reference cannot be transferred to
1437 * its final owner. In such cases, the caller is still actively using @kvm and
1438 * will fail miserably if the refcount unexpectedly hits zero.
1440 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
1442 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
1444 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
1446 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1448 struct kvm
*kvm
= filp
->private_data
;
1450 kvm_irqfd_release(kvm
);
1457 * Allocation size is twice as large as the actual dirty bitmap size.
1458 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1460 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1462 unsigned long dirty_bytes
= kvm_dirty_bitmap_bytes(memslot
);
1464 memslot
->dirty_bitmap
= __vcalloc(2, dirty_bytes
, GFP_KERNEL_ACCOUNT
);
1465 if (!memslot
->dirty_bitmap
)
1471 static struct kvm_memslots
*kvm_get_inactive_memslots(struct kvm
*kvm
, int as_id
)
1473 struct kvm_memslots
*active
= __kvm_memslots(kvm
, as_id
);
1474 int node_idx_inactive
= active
->node_idx
^ 1;
1476 return &kvm
->__memslots
[as_id
][node_idx_inactive
];
1480 * Helper to get the address space ID when one of memslot pointers may be NULL.
1481 * This also serves as a sanity that at least one of the pointers is non-NULL,
1482 * and that their address space IDs don't diverge.
1484 static int kvm_memslots_get_as_id(struct kvm_memory_slot
*a
,
1485 struct kvm_memory_slot
*b
)
1487 if (WARN_ON_ONCE(!a
&& !b
))
1495 WARN_ON_ONCE(a
->as_id
!= b
->as_id
);
1499 static void kvm_insert_gfn_node(struct kvm_memslots
*slots
,
1500 struct kvm_memory_slot
*slot
)
1502 struct rb_root
*gfn_tree
= &slots
->gfn_tree
;
1503 struct rb_node
**node
, *parent
;
1504 int idx
= slots
->node_idx
;
1507 for (node
= &gfn_tree
->rb_node
; *node
; ) {
1508 struct kvm_memory_slot
*tmp
;
1510 tmp
= container_of(*node
, struct kvm_memory_slot
, gfn_node
[idx
]);
1512 if (slot
->base_gfn
< tmp
->base_gfn
)
1513 node
= &(*node
)->rb_left
;
1514 else if (slot
->base_gfn
> tmp
->base_gfn
)
1515 node
= &(*node
)->rb_right
;
1520 rb_link_node(&slot
->gfn_node
[idx
], parent
, node
);
1521 rb_insert_color(&slot
->gfn_node
[idx
], gfn_tree
);
1524 static void kvm_erase_gfn_node(struct kvm_memslots
*slots
,
1525 struct kvm_memory_slot
*slot
)
1527 rb_erase(&slot
->gfn_node
[slots
->node_idx
], &slots
->gfn_tree
);
1530 static void kvm_replace_gfn_node(struct kvm_memslots
*slots
,
1531 struct kvm_memory_slot
*old
,
1532 struct kvm_memory_slot
*new)
1534 int idx
= slots
->node_idx
;
1536 WARN_ON_ONCE(old
->base_gfn
!= new->base_gfn
);
1538 rb_replace_node(&old
->gfn_node
[idx
], &new->gfn_node
[idx
],
1543 * Replace @old with @new in the inactive memslots.
1545 * With NULL @old this simply adds @new.
1546 * With NULL @new this simply removes @old.
1548 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1551 static void kvm_replace_memslot(struct kvm
*kvm
,
1552 struct kvm_memory_slot
*old
,
1553 struct kvm_memory_slot
*new)
1555 int as_id
= kvm_memslots_get_as_id(old
, new);
1556 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1557 int idx
= slots
->node_idx
;
1560 hash_del(&old
->id_node
[idx
]);
1561 interval_tree_remove(&old
->hva_node
[idx
], &slots
->hva_tree
);
1563 if ((long)old
== atomic_long_read(&slots
->last_used_slot
))
1564 atomic_long_set(&slots
->last_used_slot
, (long)new);
1567 kvm_erase_gfn_node(slots
, old
);
1573 * Initialize @new's hva range. Do this even when replacing an @old
1574 * slot, kvm_copy_memslot() deliberately does not touch node data.
1576 new->hva_node
[idx
].start
= new->userspace_addr
;
1577 new->hva_node
[idx
].last
= new->userspace_addr
+
1578 (new->npages
<< PAGE_SHIFT
) - 1;
1581 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1582 * hva_node needs to be swapped with remove+insert even though hva can't
1583 * change when replacing an existing slot.
1585 hash_add(slots
->id_hash
, &new->id_node
[idx
], new->id
);
1586 interval_tree_insert(&new->hva_node
[idx
], &slots
->hva_tree
);
1589 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1590 * switch the node in the gfn tree instead of removing the old and
1591 * inserting the new as two separate operations. Replacement is a
1592 * single O(1) operation versus two O(log(n)) operations for
1595 if (old
&& old
->base_gfn
== new->base_gfn
) {
1596 kvm_replace_gfn_node(slots
, old
, new);
1599 kvm_erase_gfn_node(slots
, old
);
1600 kvm_insert_gfn_node(slots
, new);
1605 * Flags that do not access any of the extra space of struct
1606 * kvm_userspace_memory_region2. KVM_SET_USER_MEMORY_REGION_V1_FLAGS
1607 * only allows these.
1609 #define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \
1610 (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY)
1612 static int check_memory_region_flags(struct kvm
*kvm
,
1613 const struct kvm_userspace_memory_region2
*mem
)
1615 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1617 if (kvm_arch_has_private_mem(kvm
))
1618 valid_flags
|= KVM_MEM_GUEST_MEMFD
;
1620 /* Dirty logging private memory is not currently supported. */
1621 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
)
1622 valid_flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
1624 #ifdef __KVM_HAVE_READONLY_MEM
1625 valid_flags
|= KVM_MEM_READONLY
;
1628 if (mem
->flags
& ~valid_flags
)
1634 static void kvm_swap_active_memslots(struct kvm
*kvm
, int as_id
)
1636 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1638 /* Grab the generation from the activate memslots. */
1639 u64 gen
= __kvm_memslots(kvm
, as_id
)->generation
;
1641 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1642 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1645 * Do not store the new memslots while there are invalidations in
1646 * progress, otherwise the locking in invalidate_range_start and
1647 * invalidate_range_end will be unbalanced.
1649 spin_lock(&kvm
->mn_invalidate_lock
);
1650 prepare_to_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1651 while (kvm
->mn_active_invalidate_count
) {
1652 set_current_state(TASK_UNINTERRUPTIBLE
);
1653 spin_unlock(&kvm
->mn_invalidate_lock
);
1655 spin_lock(&kvm
->mn_invalidate_lock
);
1657 finish_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1658 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1659 spin_unlock(&kvm
->mn_invalidate_lock
);
1662 * Acquired in kvm_set_memslot. Must be released before synchronize
1663 * SRCU below in order to avoid deadlock with another thread
1664 * acquiring the slots_arch_lock in an srcu critical section.
1666 mutex_unlock(&kvm
->slots_arch_lock
);
1668 synchronize_srcu_expedited(&kvm
->srcu
);
1671 * Increment the new memslot generation a second time, dropping the
1672 * update in-progress flag and incrementing the generation based on
1673 * the number of address spaces. This provides a unique and easily
1674 * identifiable generation number while the memslots are in flux.
1676 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1679 * Generations must be unique even across address spaces. We do not need
1680 * a global counter for that, instead the generation space is evenly split
1681 * across address spaces. For example, with two address spaces, address
1682 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1683 * use generations 1, 3, 5, ...
1685 gen
+= kvm_arch_nr_memslot_as_ids(kvm
);
1687 kvm_arch_memslots_updated(kvm
, gen
);
1689 slots
->generation
= gen
;
1692 static int kvm_prepare_memory_region(struct kvm
*kvm
,
1693 const struct kvm_memory_slot
*old
,
1694 struct kvm_memory_slot
*new,
1695 enum kvm_mr_change change
)
1700 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1701 * will be freed on "commit". If logging is enabled in both old and
1702 * new, reuse the existing bitmap. If logging is enabled only in the
1703 * new and KVM isn't using a ring buffer, allocate and initialize a
1706 if (change
!= KVM_MR_DELETE
) {
1707 if (!(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1708 new->dirty_bitmap
= NULL
;
1709 else if (old
&& old
->dirty_bitmap
)
1710 new->dirty_bitmap
= old
->dirty_bitmap
;
1711 else if (kvm_use_dirty_bitmap(kvm
)) {
1712 r
= kvm_alloc_dirty_bitmap(new);
1716 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1717 bitmap_set(new->dirty_bitmap
, 0, new->npages
);
1721 r
= kvm_arch_prepare_memory_region(kvm
, old
, new, change
);
1723 /* Free the bitmap on failure if it was allocated above. */
1724 if (r
&& new && new->dirty_bitmap
&& (!old
|| !old
->dirty_bitmap
))
1725 kvm_destroy_dirty_bitmap(new);
1730 static void kvm_commit_memory_region(struct kvm
*kvm
,
1731 struct kvm_memory_slot
*old
,
1732 const struct kvm_memory_slot
*new,
1733 enum kvm_mr_change change
)
1735 int old_flags
= old
? old
->flags
: 0;
1736 int new_flags
= new ? new->flags
: 0;
1738 * Update the total number of memslot pages before calling the arch
1739 * hook so that architectures can consume the result directly.
1741 if (change
== KVM_MR_DELETE
)
1742 kvm
->nr_memslot_pages
-= old
->npages
;
1743 else if (change
== KVM_MR_CREATE
)
1744 kvm
->nr_memslot_pages
+= new->npages
;
1746 if ((old_flags
^ new_flags
) & KVM_MEM_LOG_DIRTY_PAGES
) {
1747 int change
= (new_flags
& KVM_MEM_LOG_DIRTY_PAGES
) ? 1 : -1;
1748 atomic_set(&kvm
->nr_memslots_dirty_logging
,
1749 atomic_read(&kvm
->nr_memslots_dirty_logging
) + change
);
1752 kvm_arch_commit_memory_region(kvm
, old
, new, change
);
1756 /* Nothing more to do. */
1759 /* Free the old memslot and all its metadata. */
1760 kvm_free_memslot(kvm
, old
);
1763 case KVM_MR_FLAGS_ONLY
:
1765 * Free the dirty bitmap as needed; the below check encompasses
1766 * both the flags and whether a ring buffer is being used)
1768 if (old
->dirty_bitmap
&& !new->dirty_bitmap
)
1769 kvm_destroy_dirty_bitmap(old
);
1772 * The final quirk. Free the detached, old slot, but only its
1773 * memory, not any metadata. Metadata, including arch specific
1774 * data, may be reused by @new.
1784 * Activate @new, which must be installed in the inactive slots by the caller,
1785 * by swapping the active slots and then propagating @new to @old once @old is
1786 * unreachable and can be safely modified.
1788 * With NULL @old this simply adds @new to @active (while swapping the sets).
1789 * With NULL @new this simply removes @old from @active and frees it
1790 * (while also swapping the sets).
1792 static void kvm_activate_memslot(struct kvm
*kvm
,
1793 struct kvm_memory_slot
*old
,
1794 struct kvm_memory_slot
*new)
1796 int as_id
= kvm_memslots_get_as_id(old
, new);
1798 kvm_swap_active_memslots(kvm
, as_id
);
1800 /* Propagate the new memslot to the now inactive memslots. */
1801 kvm_replace_memslot(kvm
, old
, new);
1804 static void kvm_copy_memslot(struct kvm_memory_slot
*dest
,
1805 const struct kvm_memory_slot
*src
)
1807 dest
->base_gfn
= src
->base_gfn
;
1808 dest
->npages
= src
->npages
;
1809 dest
->dirty_bitmap
= src
->dirty_bitmap
;
1810 dest
->arch
= src
->arch
;
1811 dest
->userspace_addr
= src
->userspace_addr
;
1812 dest
->flags
= src
->flags
;
1814 dest
->as_id
= src
->as_id
;
1817 static void kvm_invalidate_memslot(struct kvm
*kvm
,
1818 struct kvm_memory_slot
*old
,
1819 struct kvm_memory_slot
*invalid_slot
)
1822 * Mark the current slot INVALID. As with all memslot modifications,
1823 * this must be done on an unreachable slot to avoid modifying the
1824 * current slot in the active tree.
1826 kvm_copy_memslot(invalid_slot
, old
);
1827 invalid_slot
->flags
|= KVM_MEMSLOT_INVALID
;
1828 kvm_replace_memslot(kvm
, old
, invalid_slot
);
1831 * Activate the slot that is now marked INVALID, but don't propagate
1832 * the slot to the now inactive slots. The slot is either going to be
1833 * deleted or recreated as a new slot.
1835 kvm_swap_active_memslots(kvm
, old
->as_id
);
1838 * From this point no new shadow pages pointing to a deleted, or moved,
1839 * memslot will be created. Validation of sp->gfn happens in:
1840 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1841 * - kvm_is_visible_gfn (mmu_check_root)
1843 kvm_arch_flush_shadow_memslot(kvm
, old
);
1844 kvm_arch_guest_memory_reclaimed(kvm
);
1846 /* Was released by kvm_swap_active_memslots(), reacquire. */
1847 mutex_lock(&kvm
->slots_arch_lock
);
1850 * Copy the arch-specific field of the newly-installed slot back to the
1851 * old slot as the arch data could have changed between releasing
1852 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1853 * above. Writers are required to retrieve memslots *after* acquiring
1854 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1856 old
->arch
= invalid_slot
->arch
;
1859 static void kvm_create_memslot(struct kvm
*kvm
,
1860 struct kvm_memory_slot
*new)
1862 /* Add the new memslot to the inactive set and activate. */
1863 kvm_replace_memslot(kvm
, NULL
, new);
1864 kvm_activate_memslot(kvm
, NULL
, new);
1867 static void kvm_delete_memslot(struct kvm
*kvm
,
1868 struct kvm_memory_slot
*old
,
1869 struct kvm_memory_slot
*invalid_slot
)
1872 * Remove the old memslot (in the inactive memslots) by passing NULL as
1873 * the "new" slot, and for the invalid version in the active slots.
1875 kvm_replace_memslot(kvm
, old
, NULL
);
1876 kvm_activate_memslot(kvm
, invalid_slot
, NULL
);
1879 static void kvm_move_memslot(struct kvm
*kvm
,
1880 struct kvm_memory_slot
*old
,
1881 struct kvm_memory_slot
*new,
1882 struct kvm_memory_slot
*invalid_slot
)
1885 * Replace the old memslot in the inactive slots, and then swap slots
1886 * and replace the current INVALID with the new as well.
1888 kvm_replace_memslot(kvm
, old
, new);
1889 kvm_activate_memslot(kvm
, invalid_slot
, new);
1892 static void kvm_update_flags_memslot(struct kvm
*kvm
,
1893 struct kvm_memory_slot
*old
,
1894 struct kvm_memory_slot
*new)
1897 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1898 * an intermediate step. Instead, the old memslot is simply replaced
1899 * with a new, updated copy in both memslot sets.
1901 kvm_replace_memslot(kvm
, old
, new);
1902 kvm_activate_memslot(kvm
, old
, new);
1905 static int kvm_set_memslot(struct kvm
*kvm
,
1906 struct kvm_memory_slot
*old
,
1907 struct kvm_memory_slot
*new,
1908 enum kvm_mr_change change
)
1910 struct kvm_memory_slot
*invalid_slot
;
1914 * Released in kvm_swap_active_memslots().
1916 * Must be held from before the current memslots are copied until after
1917 * the new memslots are installed with rcu_assign_pointer, then
1918 * released before the synchronize srcu in kvm_swap_active_memslots().
1920 * When modifying memslots outside of the slots_lock, must be held
1921 * before reading the pointer to the current memslots until after all
1922 * changes to those memslots are complete.
1924 * These rules ensure that installing new memslots does not lose
1925 * changes made to the previous memslots.
1927 mutex_lock(&kvm
->slots_arch_lock
);
1930 * Invalidate the old slot if it's being deleted or moved. This is
1931 * done prior to actually deleting/moving the memslot to allow vCPUs to
1932 * continue running by ensuring there are no mappings or shadow pages
1933 * for the memslot when it is deleted/moved. Without pre-invalidation
1934 * (and without a lock), a window would exist between effecting the
1935 * delete/move and committing the changes in arch code where KVM or a
1936 * guest could access a non-existent memslot.
1938 * Modifications are done on a temporary, unreachable slot. The old
1939 * slot needs to be preserved in case a later step fails and the
1940 * invalidation needs to be reverted.
1942 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1943 invalid_slot
= kzalloc(sizeof(*invalid_slot
), GFP_KERNEL_ACCOUNT
);
1944 if (!invalid_slot
) {
1945 mutex_unlock(&kvm
->slots_arch_lock
);
1948 kvm_invalidate_memslot(kvm
, old
, invalid_slot
);
1951 r
= kvm_prepare_memory_region(kvm
, old
, new, change
);
1954 * For DELETE/MOVE, revert the above INVALID change. No
1955 * modifications required since the original slot was preserved
1956 * in the inactive slots. Changing the active memslots also
1957 * release slots_arch_lock.
1959 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1960 kvm_activate_memslot(kvm
, invalid_slot
, old
);
1961 kfree(invalid_slot
);
1963 mutex_unlock(&kvm
->slots_arch_lock
);
1969 * For DELETE and MOVE, the working slot is now active as the INVALID
1970 * version of the old slot. MOVE is particularly special as it reuses
1971 * the old slot and returns a copy of the old slot (in working_slot).
1972 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1973 * old slot is detached but otherwise preserved.
1975 if (change
== KVM_MR_CREATE
)
1976 kvm_create_memslot(kvm
, new);
1977 else if (change
== KVM_MR_DELETE
)
1978 kvm_delete_memslot(kvm
, old
, invalid_slot
);
1979 else if (change
== KVM_MR_MOVE
)
1980 kvm_move_memslot(kvm
, old
, new, invalid_slot
);
1981 else if (change
== KVM_MR_FLAGS_ONLY
)
1982 kvm_update_flags_memslot(kvm
, old
, new);
1986 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1987 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1988 kfree(invalid_slot
);
1991 * No need to refresh new->arch, changes after dropping slots_arch_lock
1992 * will directly hit the final, active memslot. Architectures are
1993 * responsible for knowing that new->arch may be stale.
1995 kvm_commit_memory_region(kvm
, old
, new, change
);
2000 static bool kvm_check_memslot_overlap(struct kvm_memslots
*slots
, int id
,
2001 gfn_t start
, gfn_t end
)
2003 struct kvm_memslot_iter iter
;
2005 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, start
, end
) {
2006 if (iter
.slot
->id
!= id
)
2014 * Allocate some memory and give it an address in the guest physical address
2017 * Discontiguous memory is allowed, mostly for framebuffers.
2019 * Must be called holding kvm->slots_lock for write.
2021 int __kvm_set_memory_region(struct kvm
*kvm
,
2022 const struct kvm_userspace_memory_region2
*mem
)
2024 struct kvm_memory_slot
*old
, *new;
2025 struct kvm_memslots
*slots
;
2026 enum kvm_mr_change change
;
2027 unsigned long npages
;
2032 r
= check_memory_region_flags(kvm
, mem
);
2036 as_id
= mem
->slot
>> 16;
2037 id
= (u16
)mem
->slot
;
2039 /* General sanity checks */
2040 if ((mem
->memory_size
& (PAGE_SIZE
- 1)) ||
2041 (mem
->memory_size
!= (unsigned long)mem
->memory_size
))
2043 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
2045 /* We can read the guest memory with __xxx_user() later on. */
2046 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
2047 (mem
->userspace_addr
!= untagged_addr(mem
->userspace_addr
)) ||
2048 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
2051 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
&&
2052 (mem
->guest_memfd_offset
& (PAGE_SIZE
- 1) ||
2053 mem
->guest_memfd_offset
+ mem
->memory_size
< mem
->guest_memfd_offset
))
2055 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_MEM_SLOTS_NUM
)
2057 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
2059 if ((mem
->memory_size
>> PAGE_SHIFT
) > KVM_MEM_MAX_NR_PAGES
)
2062 slots
= __kvm_memslots(kvm
, as_id
);
2065 * Note, the old memslot (and the pointer itself!) may be invalidated
2066 * and/or destroyed by kvm_set_memslot().
2068 old
= id_to_memslot(slots
, id
);
2070 if (!mem
->memory_size
) {
2071 if (!old
|| !old
->npages
)
2074 if (WARN_ON_ONCE(kvm
->nr_memslot_pages
< old
->npages
))
2077 return kvm_set_memslot(kvm
, old
, NULL
, KVM_MR_DELETE
);
2080 base_gfn
= (mem
->guest_phys_addr
>> PAGE_SHIFT
);
2081 npages
= (mem
->memory_size
>> PAGE_SHIFT
);
2083 if (!old
|| !old
->npages
) {
2084 change
= KVM_MR_CREATE
;
2087 * To simplify KVM internals, the total number of pages across
2088 * all memslots must fit in an unsigned long.
2090 if ((kvm
->nr_memslot_pages
+ npages
) < kvm
->nr_memslot_pages
)
2092 } else { /* Modify an existing slot. */
2093 /* Private memslots are immutable, they can only be deleted. */
2094 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
)
2096 if ((mem
->userspace_addr
!= old
->userspace_addr
) ||
2097 (npages
!= old
->npages
) ||
2098 ((mem
->flags
^ old
->flags
) & KVM_MEM_READONLY
))
2101 if (base_gfn
!= old
->base_gfn
)
2102 change
= KVM_MR_MOVE
;
2103 else if (mem
->flags
!= old
->flags
)
2104 change
= KVM_MR_FLAGS_ONLY
;
2105 else /* Nothing to change. */
2109 if ((change
== KVM_MR_CREATE
|| change
== KVM_MR_MOVE
) &&
2110 kvm_check_memslot_overlap(slots
, id
, base_gfn
, base_gfn
+ npages
))
2113 /* Allocate a slot that will persist in the memslot. */
2114 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT
);
2120 new->base_gfn
= base_gfn
;
2121 new->npages
= npages
;
2122 new->flags
= mem
->flags
;
2123 new->userspace_addr
= mem
->userspace_addr
;
2124 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
) {
2125 r
= kvm_gmem_bind(kvm
, new, mem
->guest_memfd
, mem
->guest_memfd_offset
);
2130 r
= kvm_set_memslot(kvm
, old
, new, change
);
2137 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
)
2138 kvm_gmem_unbind(new);
2143 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
2145 int kvm_set_memory_region(struct kvm
*kvm
,
2146 const struct kvm_userspace_memory_region2
*mem
)
2150 mutex_lock(&kvm
->slots_lock
);
2151 r
= __kvm_set_memory_region(kvm
, mem
);
2152 mutex_unlock(&kvm
->slots_lock
);
2155 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
2157 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
2158 struct kvm_userspace_memory_region2
*mem
)
2160 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
2163 return kvm_set_memory_region(kvm
, mem
);
2166 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2168 * kvm_get_dirty_log - get a snapshot of dirty pages
2169 * @kvm: pointer to kvm instance
2170 * @log: slot id and address to which we copy the log
2171 * @is_dirty: set to '1' if any dirty pages were found
2172 * @memslot: set to the associated memslot, always valid on success
2174 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
2175 int *is_dirty
, struct kvm_memory_slot
**memslot
)
2177 struct kvm_memslots
*slots
;
2180 unsigned long any
= 0;
2182 /* Dirty ring tracking may be exclusive to dirty log tracking */
2183 if (!kvm_use_dirty_bitmap(kvm
))
2189 as_id
= log
->slot
>> 16;
2190 id
= (u16
)log
->slot
;
2191 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_USER_MEM_SLOTS
)
2194 slots
= __kvm_memslots(kvm
, as_id
);
2195 *memslot
= id_to_memslot(slots
, id
);
2196 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
2199 kvm_arch_sync_dirty_log(kvm
, *memslot
);
2201 n
= kvm_dirty_bitmap_bytes(*memslot
);
2203 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
2204 any
= (*memslot
)->dirty_bitmap
[i
];
2206 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
2213 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
2215 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2217 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2218 * and reenable dirty page tracking for the corresponding pages.
2219 * @kvm: pointer to kvm instance
2220 * @log: slot id and address to which we copy the log
2222 * We need to keep it in mind that VCPU threads can write to the bitmap
2223 * concurrently. So, to avoid losing track of dirty pages we keep the
2226 * 1. Take a snapshot of the bit and clear it if needed.
2227 * 2. Write protect the corresponding page.
2228 * 3. Copy the snapshot to the userspace.
2229 * 4. Upon return caller flushes TLB's if needed.
2231 * Between 2 and 4, the guest may write to the page using the remaining TLB
2232 * entry. This is not a problem because the page is reported dirty using
2233 * the snapshot taken before and step 4 ensures that writes done after
2234 * exiting to userspace will be logged for the next call.
2237 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
2239 struct kvm_memslots
*slots
;
2240 struct kvm_memory_slot
*memslot
;
2243 unsigned long *dirty_bitmap
;
2244 unsigned long *dirty_bitmap_buffer
;
2247 /* Dirty ring tracking may be exclusive to dirty log tracking */
2248 if (!kvm_use_dirty_bitmap(kvm
))
2251 as_id
= log
->slot
>> 16;
2252 id
= (u16
)log
->slot
;
2253 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_USER_MEM_SLOTS
)
2256 slots
= __kvm_memslots(kvm
, as_id
);
2257 memslot
= id_to_memslot(slots
, id
);
2258 if (!memslot
|| !memslot
->dirty_bitmap
)
2261 dirty_bitmap
= memslot
->dirty_bitmap
;
2263 kvm_arch_sync_dirty_log(kvm
, memslot
);
2265 n
= kvm_dirty_bitmap_bytes(memslot
);
2267 if (kvm
->manual_dirty_log_protect
) {
2269 * Unlike kvm_get_dirty_log, we always return false in *flush,
2270 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2271 * is some code duplication between this function and
2272 * kvm_get_dirty_log, but hopefully all architecture
2273 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2274 * can be eliminated.
2276 dirty_bitmap_buffer
= dirty_bitmap
;
2278 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2279 memset(dirty_bitmap_buffer
, 0, n
);
2282 for (i
= 0; i
< n
/ sizeof(long); i
++) {
2286 if (!dirty_bitmap
[i
])
2290 mask
= xchg(&dirty_bitmap
[i
], 0);
2291 dirty_bitmap_buffer
[i
] = mask
;
2293 offset
= i
* BITS_PER_LONG
;
2294 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2297 KVM_MMU_UNLOCK(kvm
);
2301 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2303 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
2310 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2311 * @kvm: kvm instance
2312 * @log: slot id and address to which we copy the log
2314 * Steps 1-4 below provide general overview of dirty page logging. See
2315 * kvm_get_dirty_log_protect() function description for additional details.
2317 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2318 * always flush the TLB (step 4) even if previous step failed and the dirty
2319 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2320 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2321 * writes will be marked dirty for next log read.
2323 * 1. Take a snapshot of the bit and clear it if needed.
2324 * 2. Write protect the corresponding page.
2325 * 3. Copy the snapshot to the userspace.
2326 * 4. Flush TLB's if needed.
2328 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2329 struct kvm_dirty_log
*log
)
2333 mutex_lock(&kvm
->slots_lock
);
2335 r
= kvm_get_dirty_log_protect(kvm
, log
);
2337 mutex_unlock(&kvm
->slots_lock
);
2342 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2343 * and reenable dirty page tracking for the corresponding pages.
2344 * @kvm: pointer to kvm instance
2345 * @log: slot id and address from which to fetch the bitmap of dirty pages
2347 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
2348 struct kvm_clear_dirty_log
*log
)
2350 struct kvm_memslots
*slots
;
2351 struct kvm_memory_slot
*memslot
;
2355 unsigned long *dirty_bitmap
;
2356 unsigned long *dirty_bitmap_buffer
;
2359 /* Dirty ring tracking may be exclusive to dirty log tracking */
2360 if (!kvm_use_dirty_bitmap(kvm
))
2363 as_id
= log
->slot
>> 16;
2364 id
= (u16
)log
->slot
;
2365 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_USER_MEM_SLOTS
)
2368 if (log
->first_page
& 63)
2371 slots
= __kvm_memslots(kvm
, as_id
);
2372 memslot
= id_to_memslot(slots
, id
);
2373 if (!memslot
|| !memslot
->dirty_bitmap
)
2376 dirty_bitmap
= memslot
->dirty_bitmap
;
2378 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
2380 if (log
->first_page
> memslot
->npages
||
2381 log
->num_pages
> memslot
->npages
- log
->first_page
||
2382 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
2385 kvm_arch_sync_dirty_log(kvm
, memslot
);
2388 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2389 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
2393 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
2394 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
2395 i
++, offset
+= BITS_PER_LONG
) {
2396 unsigned long mask
= *dirty_bitmap_buffer
++;
2397 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
2401 mask
&= atomic_long_fetch_andnot(mask
, p
);
2404 * mask contains the bits that really have been cleared. This
2405 * never includes any bits beyond the length of the memslot (if
2406 * the length is not aligned to 64 pages), therefore it is not
2407 * a problem if userspace sets them in log->dirty_bitmap.
2411 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2415 KVM_MMU_UNLOCK(kvm
);
2418 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2423 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
2424 struct kvm_clear_dirty_log
*log
)
2428 mutex_lock(&kvm
->slots_lock
);
2430 r
= kvm_clear_dirty_log_protect(kvm
, log
);
2432 mutex_unlock(&kvm
->slots_lock
);
2435 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2437 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
2439 * Returns true if _all_ gfns in the range [@start, @end) have attributes
2442 bool kvm_range_has_memory_attributes(struct kvm
*kvm
, gfn_t start
, gfn_t end
,
2443 unsigned long attrs
)
2445 XA_STATE(xas
, &kvm
->mem_attr_array
, start
);
2446 unsigned long index
;
2453 has_attrs
= !xas_find(&xas
, end
- 1);
2458 for (index
= start
; index
< end
; index
++) {
2460 entry
= xas_next(&xas
);
2461 } while (xas_retry(&xas
, entry
));
2463 if (xas
.xa_index
!= index
|| xa_to_value(entry
) != attrs
) {
2474 static u64
kvm_supported_mem_attributes(struct kvm
*kvm
)
2476 if (!kvm
|| kvm_arch_has_private_mem(kvm
))
2477 return KVM_MEMORY_ATTRIBUTE_PRIVATE
;
2482 static __always_inline
void kvm_handle_gfn_range(struct kvm
*kvm
,
2483 struct kvm_mmu_notifier_range
*range
)
2485 struct kvm_gfn_range gfn_range
;
2486 struct kvm_memory_slot
*slot
;
2487 struct kvm_memslots
*slots
;
2488 struct kvm_memslot_iter iter
;
2489 bool found_memslot
= false;
2493 gfn_range
.arg
= range
->arg
;
2494 gfn_range
.may_block
= range
->may_block
;
2496 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
2497 slots
= __kvm_memslots(kvm
, i
);
2499 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, range
->start
, range
->end
) {
2501 gfn_range
.slot
= slot
;
2503 gfn_range
.start
= max(range
->start
, slot
->base_gfn
);
2504 gfn_range
.end
= min(range
->end
, slot
->base_gfn
+ slot
->npages
);
2505 if (gfn_range
.start
>= gfn_range
.end
)
2508 if (!found_memslot
) {
2509 found_memslot
= true;
2511 if (!IS_KVM_NULL_FN(range
->on_lock
))
2512 range
->on_lock(kvm
);
2515 ret
|= range
->handler(kvm
, &gfn_range
);
2519 if (range
->flush_on_ret
&& ret
)
2520 kvm_flush_remote_tlbs(kvm
);
2523 KVM_MMU_UNLOCK(kvm
);
2526 static bool kvm_pre_set_memory_attributes(struct kvm
*kvm
,
2527 struct kvm_gfn_range
*range
)
2530 * Unconditionally add the range to the invalidation set, regardless of
2531 * whether or not the arch callback actually needs to zap SPTEs. E.g.
2532 * if KVM supports RWX attributes in the future and the attributes are
2533 * going from R=>RW, zapping isn't strictly necessary. Unconditionally
2534 * adding the range allows KVM to require that MMU invalidations add at
2535 * least one range between begin() and end(), e.g. allows KVM to detect
2536 * bugs where the add() is missed. Relaxing the rule *might* be safe,
2537 * but it's not obvious that allowing new mappings while the attributes
2538 * are in flux is desirable or worth the complexity.
2540 kvm_mmu_invalidate_range_add(kvm
, range
->start
, range
->end
);
2542 return kvm_arch_pre_set_memory_attributes(kvm
, range
);
2545 /* Set @attributes for the gfn range [@start, @end). */
2546 static int kvm_vm_set_mem_attributes(struct kvm
*kvm
, gfn_t start
, gfn_t end
,
2547 unsigned long attributes
)
2549 struct kvm_mmu_notifier_range pre_set_range
= {
2552 .handler
= kvm_pre_set_memory_attributes
,
2553 .on_lock
= kvm_mmu_invalidate_begin
,
2554 .flush_on_ret
= true,
2557 struct kvm_mmu_notifier_range post_set_range
= {
2560 .arg
.attributes
= attributes
,
2561 .handler
= kvm_arch_post_set_memory_attributes
,
2562 .on_lock
= kvm_mmu_invalidate_end
,
2569 entry
= attributes
? xa_mk_value(attributes
) : NULL
;
2571 mutex_lock(&kvm
->slots_lock
);
2573 /* Nothing to do if the entire range as the desired attributes. */
2574 if (kvm_range_has_memory_attributes(kvm
, start
, end
, attributes
))
2578 * Reserve memory ahead of time to avoid having to deal with failures
2579 * partway through setting the new attributes.
2581 for (i
= start
; i
< end
; i
++) {
2582 r
= xa_reserve(&kvm
->mem_attr_array
, i
, GFP_KERNEL_ACCOUNT
);
2587 kvm_handle_gfn_range(kvm
, &pre_set_range
);
2589 for (i
= start
; i
< end
; i
++) {
2590 r
= xa_err(xa_store(&kvm
->mem_attr_array
, i
, entry
,
2591 GFP_KERNEL_ACCOUNT
));
2595 kvm_handle_gfn_range(kvm
, &post_set_range
);
2598 mutex_unlock(&kvm
->slots_lock
);
2602 static int kvm_vm_ioctl_set_mem_attributes(struct kvm
*kvm
,
2603 struct kvm_memory_attributes
*attrs
)
2607 /* flags is currently not used. */
2610 if (attrs
->attributes
& ~kvm_supported_mem_attributes(kvm
))
2612 if (attrs
->size
== 0 || attrs
->address
+ attrs
->size
< attrs
->address
)
2614 if (!PAGE_ALIGNED(attrs
->address
) || !PAGE_ALIGNED(attrs
->size
))
2617 start
= attrs
->address
>> PAGE_SHIFT
;
2618 end
= (attrs
->address
+ attrs
->size
) >> PAGE_SHIFT
;
2621 * xarray tracks data using "unsigned long", and as a result so does
2622 * KVM. For simplicity, supports generic attributes only on 64-bit
2625 BUILD_BUG_ON(sizeof(attrs
->attributes
) != sizeof(unsigned long));
2627 return kvm_vm_set_mem_attributes(kvm
, start
, end
, attrs
->attributes
);
2629 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2631 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
2633 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
2635 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
2637 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2639 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2640 u64 gen
= slots
->generation
;
2641 struct kvm_memory_slot
*slot
;
2644 * This also protects against using a memslot from a different address space,
2645 * since different address spaces have different generation numbers.
2647 if (unlikely(gen
!= vcpu
->last_used_slot_gen
)) {
2648 vcpu
->last_used_slot
= NULL
;
2649 vcpu
->last_used_slot_gen
= gen
;
2652 slot
= try_get_memslot(vcpu
->last_used_slot
, gfn
);
2657 * Fall back to searching all memslots. We purposely use
2658 * search_memslots() instead of __gfn_to_memslot() to avoid
2659 * thrashing the VM-wide last_used_slot in kvm_memslots.
2661 slot
= search_memslots(slots
, gfn
, false);
2663 vcpu
->last_used_slot
= slot
;
2670 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
2672 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
2674 return kvm_is_visible_memslot(memslot
);
2676 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
2678 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2680 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2682 return kvm_is_visible_memslot(memslot
);
2684 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
2686 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2688 struct vm_area_struct
*vma
;
2689 unsigned long addr
, size
;
2693 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
2694 if (kvm_is_error_hva(addr
))
2697 mmap_read_lock(current
->mm
);
2698 vma
= find_vma(current
->mm
, addr
);
2702 size
= vma_kernel_pagesize(vma
);
2705 mmap_read_unlock(current
->mm
);
2710 static bool memslot_is_readonly(const struct kvm_memory_slot
*slot
)
2712 return slot
->flags
& KVM_MEM_READONLY
;
2715 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2716 gfn_t
*nr_pages
, bool write
)
2718 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
2719 return KVM_HVA_ERR_BAD
;
2721 if (memslot_is_readonly(slot
) && write
)
2722 return KVM_HVA_ERR_RO_BAD
;
2725 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
2727 return __gfn_to_hva_memslot(slot
, gfn
);
2730 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2733 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
2736 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
2739 return gfn_to_hva_many(slot
, gfn
, NULL
);
2741 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
2743 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
2745 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
2747 EXPORT_SYMBOL_GPL(gfn_to_hva
);
2749 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2751 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
2753 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
2756 * Return the hva of a @gfn and the R/W attribute if possible.
2758 * @slot: the kvm_memory_slot which contains @gfn
2759 * @gfn: the gfn to be translated
2760 * @writable: used to return the read/write attribute of the @slot if the hva
2761 * is valid and @writable is not NULL
2763 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
2764 gfn_t gfn
, bool *writable
)
2766 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
2768 if (!kvm_is_error_hva(hva
) && writable
)
2769 *writable
= !memslot_is_readonly(slot
);
2774 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
2776 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2778 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2781 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
2783 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2785 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2788 static inline int check_user_page_hwpoison(unsigned long addr
)
2790 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
2792 rc
= get_user_pages(addr
, 1, flags
, NULL
);
2793 return rc
== -EHWPOISON
;
2797 * The fast path to get the writable pfn which will be stored in @pfn,
2798 * true indicates success, otherwise false is returned. It's also the
2799 * only part that runs if we can in atomic context.
2801 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
2802 bool *writable
, kvm_pfn_t
*pfn
)
2804 struct page
*page
[1];
2807 * Fast pin a writable pfn only if it is a write fault request
2808 * or the caller allows to map a writable pfn for a read fault
2811 if (!(write_fault
|| writable
))
2814 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
2815 *pfn
= page_to_pfn(page
[0]);
2826 * The slow path to get the pfn of the specified host virtual address,
2827 * 1 indicates success, -errno is returned if error is detected.
2829 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
2830 bool interruptible
, bool *writable
, kvm_pfn_t
*pfn
)
2833 * When a VCPU accesses a page that is not mapped into the secondary
2834 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2835 * make progress. We always want to honor NUMA hinting faults in that
2836 * case, because GUP usage corresponds to memory accesses from the VCPU.
2837 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2838 * mapped into the secondary MMU and gets accessed by a VCPU.
2840 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2841 * implicitly honor NUMA hinting faults and don't need this flag.
2843 unsigned int flags
= FOLL_HWPOISON
| FOLL_HONOR_NUMA_FAULT
;
2850 *writable
= write_fault
;
2853 flags
|= FOLL_WRITE
;
2855 flags
|= FOLL_NOWAIT
;
2857 flags
|= FOLL_INTERRUPTIBLE
;
2859 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
2863 /* map read fault as writable if possible */
2864 if (unlikely(!write_fault
) && writable
) {
2867 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
2873 *pfn
= page_to_pfn(page
);
2877 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
2879 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
2882 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
2888 static int kvm_try_get_pfn(kvm_pfn_t pfn
)
2890 struct page
*page
= kvm_pfn_to_refcounted_page(pfn
);
2895 return get_page_unless_zero(page
);
2898 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
2899 unsigned long addr
, bool write_fault
,
2900 bool *writable
, kvm_pfn_t
*p_pfn
)
2908 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2911 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2912 * not call the fault handler, so do it here.
2914 bool unlocked
= false;
2915 r
= fixup_user_fault(current
->mm
, addr
,
2916 (write_fault
? FAULT_FLAG_WRITE
: 0),
2923 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2928 pte
= ptep_get(ptep
);
2930 if (write_fault
&& !pte_write(pte
)) {
2931 pfn
= KVM_PFN_ERR_RO_FAULT
;
2936 *writable
= pte_write(pte
);
2940 * Get a reference here because callers of *hva_to_pfn* and
2941 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2942 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2943 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2944 * simply do nothing for reserved pfns.
2946 * Whoever called remap_pfn_range is also going to call e.g.
2947 * unmap_mapping_range before the underlying pages are freed,
2948 * causing a call to our MMU notifier.
2950 * Certain IO or PFNMAP mappings can be backed with valid
2951 * struct pages, but be allocated without refcounting e.g.,
2952 * tail pages of non-compound higher order allocations, which
2953 * would then underflow the refcount when the caller does the
2954 * required put_page. Don't allow those pages here.
2956 if (!kvm_try_get_pfn(pfn
))
2960 pte_unmap_unlock(ptep
, ptl
);
2967 * Pin guest page in memory and return its pfn.
2968 * @addr: host virtual address which maps memory to the guest
2969 * @atomic: whether this function can sleep
2970 * @interruptible: whether the process can be interrupted by non-fatal signals
2971 * @async: whether this function need to wait IO complete if the
2972 * host page is not in the memory
2973 * @write_fault: whether we should get a writable host page
2974 * @writable: whether it allows to map a writable host page for !@write_fault
2976 * The function will map a writable host page for these two cases:
2977 * 1): @write_fault = true
2978 * 2): @write_fault = false && @writable, @writable will tell the caller
2979 * whether the mapping is writable.
2981 kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool interruptible
,
2982 bool *async
, bool write_fault
, bool *writable
)
2984 struct vm_area_struct
*vma
;
2988 /* we can do it either atomically or asynchronously, not both */
2989 BUG_ON(atomic
&& async
);
2991 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
2995 return KVM_PFN_ERR_FAULT
;
2997 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, interruptible
,
3001 if (npages
== -EINTR
)
3002 return KVM_PFN_ERR_SIGPENDING
;
3004 mmap_read_lock(current
->mm
);
3005 if (npages
== -EHWPOISON
||
3006 (!async
&& check_user_page_hwpoison(addr
))) {
3007 pfn
= KVM_PFN_ERR_HWPOISON
;
3012 vma
= vma_lookup(current
->mm
, addr
);
3015 pfn
= KVM_PFN_ERR_FAULT
;
3016 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
3017 r
= hva_to_pfn_remapped(vma
, addr
, write_fault
, writable
, &pfn
);
3021 pfn
= KVM_PFN_ERR_FAULT
;
3023 if (async
&& vma_is_valid(vma
, write_fault
))
3025 pfn
= KVM_PFN_ERR_FAULT
;
3028 mmap_read_unlock(current
->mm
);
3032 kvm_pfn_t
__gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
3033 bool atomic
, bool interruptible
, bool *async
,
3034 bool write_fault
, bool *writable
, hva_t
*hva
)
3036 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
3041 if (addr
== KVM_HVA_ERR_RO_BAD
) {
3044 return KVM_PFN_ERR_RO_FAULT
;
3047 if (kvm_is_error_hva(addr
)) {
3050 return KVM_PFN_NOSLOT
;
3053 /* Do not map writable pfn in the readonly memslot. */
3054 if (writable
&& memslot_is_readonly(slot
)) {
3059 return hva_to_pfn(addr
, atomic
, interruptible
, async
, write_fault
,
3062 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
3064 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
3067 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, false,
3068 NULL
, write_fault
, writable
, NULL
);
3070 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
3072 kvm_pfn_t
gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
3074 return __gfn_to_pfn_memslot(slot
, gfn
, false, false, NULL
, true,
3077 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
3079 kvm_pfn_t
gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
3081 return __gfn_to_pfn_memslot(slot
, gfn
, true, false, NULL
, true,
3084 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
3086 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3088 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
3090 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
3092 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
3094 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
3096 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
3098 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3100 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
3102 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
3104 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3105 struct page
**pages
, int nr_pages
)
3110 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
3111 if (kvm_is_error_hva(addr
))
3114 if (entry
< nr_pages
)
3117 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
3119 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
3122 * Do not use this helper unless you are absolutely certain the gfn _must_ be
3123 * backed by 'struct page'. A valid example is if the backing memslot is
3124 * controlled by KVM. Note, if the returned page is valid, it's refcount has
3125 * been elevated by gfn_to_pfn().
3127 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
3132 pfn
= gfn_to_pfn(kvm
, gfn
);
3134 if (is_error_noslot_pfn(pfn
))
3135 return KVM_ERR_PTR_BAD_PAGE
;
3137 page
= kvm_pfn_to_refcounted_page(pfn
);
3139 return KVM_ERR_PTR_BAD_PAGE
;
3143 EXPORT_SYMBOL_GPL(gfn_to_page
);
3145 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
)
3148 kvm_release_pfn_dirty(pfn
);
3150 kvm_release_pfn_clean(pfn
);
3153 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
3157 struct page
*page
= KVM_UNMAPPED_PAGE
;
3162 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
3163 if (is_error_noslot_pfn(pfn
))
3166 if (pfn_valid(pfn
)) {
3167 page
= pfn_to_page(pfn
);
3169 #ifdef CONFIG_HAS_IOMEM
3171 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
3185 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
3187 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
3195 if (map
->page
!= KVM_UNMAPPED_PAGE
)
3197 #ifdef CONFIG_HAS_IOMEM
3203 kvm_vcpu_mark_page_dirty(vcpu
, map
->gfn
);
3205 kvm_release_pfn(map
->pfn
, dirty
);
3210 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
3212 static bool kvm_is_ad_tracked_page(struct page
*page
)
3215 * Per page-flags.h, pages tagged PG_reserved "should in general not be
3216 * touched (e.g. set dirty) except by its owner".
3218 return !PageReserved(page
);
3221 static void kvm_set_page_dirty(struct page
*page
)
3223 if (kvm_is_ad_tracked_page(page
))
3227 static void kvm_set_page_accessed(struct page
*page
)
3229 if (kvm_is_ad_tracked_page(page
))
3230 mark_page_accessed(page
);
3233 void kvm_release_page_clean(struct page
*page
)
3235 WARN_ON(is_error_page(page
));
3237 kvm_set_page_accessed(page
);
3240 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
3242 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
3246 if (is_error_noslot_pfn(pfn
))
3249 page
= kvm_pfn_to_refcounted_page(pfn
);
3253 kvm_release_page_clean(page
);
3255 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
3257 void kvm_release_page_dirty(struct page
*page
)
3259 WARN_ON(is_error_page(page
));
3261 kvm_set_page_dirty(page
);
3262 kvm_release_page_clean(page
);
3264 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
3266 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
3270 if (is_error_noslot_pfn(pfn
))
3273 page
= kvm_pfn_to_refcounted_page(pfn
);
3277 kvm_release_page_dirty(page
);
3279 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
3282 * Note, checking for an error/noslot pfn is the caller's responsibility when
3283 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
3284 * "set" helpers are not to be used when the pfn might point at garbage.
3286 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
3288 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3292 kvm_set_page_dirty(pfn_to_page(pfn
));
3294 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
3296 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
3298 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3302 kvm_set_page_accessed(pfn_to_page(pfn
));
3304 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
3306 static int next_segment(unsigned long len
, int offset
)
3308 if (len
> PAGE_SIZE
- offset
)
3309 return PAGE_SIZE
- offset
;
3314 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3315 void *data
, int offset
, int len
)
3320 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3321 if (kvm_is_error_hva(addr
))
3323 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
3329 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
3332 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3334 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3336 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
3338 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
3339 int offset
, int len
)
3341 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3343 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3345 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
3347 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
3349 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3351 int offset
= offset_in_page(gpa
);
3354 while ((seg
= next_segment(len
, offset
)) != 0) {
3355 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
3365 EXPORT_SYMBOL_GPL(kvm_read_guest
);
3367 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
3369 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3371 int offset
= offset_in_page(gpa
);
3374 while ((seg
= next_segment(len
, offset
)) != 0) {
3375 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3385 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
3387 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3388 void *data
, int offset
, unsigned long len
)
3393 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3394 if (kvm_is_error_hva(addr
))
3396 pagefault_disable();
3397 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
3404 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3405 void *data
, unsigned long len
)
3407 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3408 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3409 int offset
= offset_in_page(gpa
);
3411 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
3413 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
3415 static int __kvm_write_guest_page(struct kvm
*kvm
,
3416 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
3417 const void *data
, int offset
, int len
)
3422 addr
= gfn_to_hva_memslot(memslot
, gfn
);
3423 if (kvm_is_error_hva(addr
))
3425 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
3428 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3432 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
3433 const void *data
, int offset
, int len
)
3435 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3437 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
3439 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
3441 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
3442 const void *data
, int offset
, int len
)
3444 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3446 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
3448 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
3450 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
3453 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3455 int offset
= offset_in_page(gpa
);
3458 while ((seg
= next_segment(len
, offset
)) != 0) {
3459 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
3469 EXPORT_SYMBOL_GPL(kvm_write_guest
);
3471 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
3474 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3476 int offset
= offset_in_page(gpa
);
3479 while ((seg
= next_segment(len
, offset
)) != 0) {
3480 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3490 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
3492 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
3493 struct gfn_to_hva_cache
*ghc
,
3494 gpa_t gpa
, unsigned long len
)
3496 int offset
= offset_in_page(gpa
);
3497 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
3498 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
3499 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
3500 gfn_t nr_pages_avail
;
3502 /* Update ghc->generation before performing any error checks. */
3503 ghc
->generation
= slots
->generation
;
3505 if (start_gfn
> end_gfn
) {
3506 ghc
->hva
= KVM_HVA_ERR_BAD
;
3511 * If the requested region crosses two memslots, we still
3512 * verify that the entire region is valid here.
3514 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
3515 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
3516 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
3518 if (kvm_is_error_hva(ghc
->hva
))
3522 /* Use the slow path for cross page reads and writes. */
3523 if (nr_pages_needed
== 1)
3526 ghc
->memslot
= NULL
;
3533 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3534 gpa_t gpa
, unsigned long len
)
3536 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3537 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
3539 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
3541 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3542 void *data
, unsigned int offset
,
3545 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3547 gpa_t gpa
= ghc
->gpa
+ offset
;
3549 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3552 if (slots
->generation
!= ghc
->generation
) {
3553 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3557 if (kvm_is_error_hva(ghc
->hva
))
3560 if (unlikely(!ghc
->memslot
))
3561 return kvm_write_guest(kvm
, gpa
, data
, len
);
3563 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
3566 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
3570 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
3572 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3573 void *data
, unsigned long len
)
3575 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3577 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
3579 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3580 void *data
, unsigned int offset
,
3583 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3585 gpa_t gpa
= ghc
->gpa
+ offset
;
3587 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3590 if (slots
->generation
!= ghc
->generation
) {
3591 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3595 if (kvm_is_error_hva(ghc
->hva
))
3598 if (unlikely(!ghc
->memslot
))
3599 return kvm_read_guest(kvm
, gpa
, data
, len
);
3601 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
3607 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
3609 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3610 void *data
, unsigned long len
)
3612 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3614 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
3616 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
3618 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3619 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3621 int offset
= offset_in_page(gpa
);
3624 while ((seg
= next_segment(len
, offset
)) != 0) {
3625 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
3634 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
3636 void mark_page_dirty_in_slot(struct kvm
*kvm
,
3637 const struct kvm_memory_slot
*memslot
,
3640 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
3642 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3643 if (WARN_ON_ONCE(vcpu
&& vcpu
->kvm
!= kvm
))
3646 WARN_ON_ONCE(!vcpu
&& !kvm_arch_allow_write_without_running_vcpu(kvm
));
3649 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
3650 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
3651 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
3653 if (kvm
->dirty_ring_size
&& vcpu
)
3654 kvm_dirty_ring_push(vcpu
, slot
, rel_gfn
);
3655 else if (memslot
->dirty_bitmap
)
3656 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
3659 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
3661 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
3663 struct kvm_memory_slot
*memslot
;
3665 memslot
= gfn_to_memslot(kvm
, gfn
);
3666 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3668 EXPORT_SYMBOL_GPL(mark_page_dirty
);
3670 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3672 struct kvm_memory_slot
*memslot
;
3674 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3675 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
3677 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
3679 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
3681 if (!vcpu
->sigset_active
)
3685 * This does a lockless modification of ->real_blocked, which is fine
3686 * because, only current can change ->real_blocked and all readers of
3687 * ->real_blocked don't care as long ->real_blocked is always a subset
3690 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
3693 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
3695 if (!vcpu
->sigset_active
)
3698 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
3699 sigemptyset(¤t
->real_blocked
);
3702 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3704 unsigned int old
, val
, grow
, grow_start
;
3706 old
= val
= vcpu
->halt_poll_ns
;
3707 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3708 grow
= READ_ONCE(halt_poll_ns_grow
);
3713 if (val
< grow_start
)
3716 vcpu
->halt_poll_ns
= val
;
3718 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
3721 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3723 unsigned int old
, val
, shrink
, grow_start
;
3725 old
= val
= vcpu
->halt_poll_ns
;
3726 shrink
= READ_ONCE(halt_poll_ns_shrink
);
3727 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3733 if (val
< grow_start
)
3736 vcpu
->halt_poll_ns
= val
;
3737 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
3740 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
3743 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3745 if (kvm_arch_vcpu_runnable(vcpu
))
3747 if (kvm_cpu_has_pending_timer(vcpu
))
3749 if (signal_pending(current
))
3751 if (kvm_check_request(KVM_REQ_UNBLOCK
, vcpu
))
3756 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3761 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3762 * pending. This is mostly used when halting a vCPU, but may also be used
3763 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3765 bool kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
3767 struct rcuwait
*wait
= kvm_arch_vcpu_get_wait(vcpu
);
3768 bool waited
= false;
3770 vcpu
->stat
.generic
.blocking
= 1;
3773 kvm_arch_vcpu_blocking(vcpu
);
3774 prepare_to_rcuwait(wait
);
3778 set_current_state(TASK_INTERRUPTIBLE
);
3780 if (kvm_vcpu_check_block(vcpu
) < 0)
3788 finish_rcuwait(wait
);
3789 kvm_arch_vcpu_unblocking(vcpu
);
3792 vcpu
->stat
.generic
.blocking
= 0;
3797 static inline void update_halt_poll_stats(struct kvm_vcpu
*vcpu
, ktime_t start
,
3798 ktime_t end
, bool success
)
3800 struct kvm_vcpu_stat_generic
*stats
= &vcpu
->stat
.generic
;
3801 u64 poll_ns
= ktime_to_ns(ktime_sub(end
, start
));
3803 ++vcpu
->stat
.generic
.halt_attempted_poll
;
3806 ++vcpu
->stat
.generic
.halt_successful_poll
;
3808 if (!vcpu_valid_wakeup(vcpu
))
3809 ++vcpu
->stat
.generic
.halt_poll_invalid
;
3811 stats
->halt_poll_success_ns
+= poll_ns
;
3812 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_success_hist
, poll_ns
);
3814 stats
->halt_poll_fail_ns
+= poll_ns
;
3815 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_fail_hist
, poll_ns
);
3819 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3821 struct kvm
*kvm
= vcpu
->kvm
;
3823 if (kvm
->override_halt_poll_ns
) {
3825 * Ensure kvm->max_halt_poll_ns is not read before
3826 * kvm->override_halt_poll_ns.
3828 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3831 return READ_ONCE(kvm
->max_halt_poll_ns
);
3834 return READ_ONCE(halt_poll_ns
);
3838 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3839 * polling is enabled, busy wait for a short time before blocking to avoid the
3840 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3843 void kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
3845 unsigned int max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3846 bool halt_poll_allowed
= !kvm_arch_no_poll(vcpu
);
3847 ktime_t start
, cur
, poll_end
;
3848 bool waited
= false;
3852 if (vcpu
->halt_poll_ns
> max_halt_poll_ns
)
3853 vcpu
->halt_poll_ns
= max_halt_poll_ns
;
3855 do_halt_poll
= halt_poll_allowed
&& vcpu
->halt_poll_ns
;
3857 start
= cur
= poll_end
= ktime_get();
3859 ktime_t stop
= ktime_add_ns(start
, vcpu
->halt_poll_ns
);
3862 if (kvm_vcpu_check_block(vcpu
) < 0)
3865 poll_end
= cur
= ktime_get();
3866 } while (kvm_vcpu_can_poll(cur
, stop
));
3869 waited
= kvm_vcpu_block(vcpu
);
3873 vcpu
->stat
.generic
.halt_wait_ns
+=
3874 ktime_to_ns(cur
) - ktime_to_ns(poll_end
);
3875 KVM_STATS_LOG_HIST_UPDATE(vcpu
->stat
.generic
.halt_wait_hist
,
3876 ktime_to_ns(cur
) - ktime_to_ns(poll_end
));
3879 /* The total time the vCPU was "halted", including polling time. */
3880 halt_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
3883 * Note, halt-polling is considered successful so long as the vCPU was
3884 * never actually scheduled out, i.e. even if the wake event arrived
3885 * after of the halt-polling loop itself, but before the full wait.
3888 update_halt_poll_stats(vcpu
, start
, poll_end
, !waited
);
3890 if (halt_poll_allowed
) {
3891 /* Recompute the max halt poll time in case it changed. */
3892 max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3894 if (!vcpu_valid_wakeup(vcpu
)) {
3895 shrink_halt_poll_ns(vcpu
);
3896 } else if (max_halt_poll_ns
) {
3897 if (halt_ns
<= vcpu
->halt_poll_ns
)
3899 /* we had a long block, shrink polling */
3900 else if (vcpu
->halt_poll_ns
&&
3901 halt_ns
> max_halt_poll_ns
)
3902 shrink_halt_poll_ns(vcpu
);
3903 /* we had a short halt and our poll time is too small */
3904 else if (vcpu
->halt_poll_ns
< max_halt_poll_ns
&&
3905 halt_ns
< max_halt_poll_ns
)
3906 grow_halt_poll_ns(vcpu
);
3908 vcpu
->halt_poll_ns
= 0;
3912 trace_kvm_vcpu_wakeup(halt_ns
, waited
, vcpu_valid_wakeup(vcpu
));
3914 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
3916 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
3918 if (__kvm_vcpu_wake_up(vcpu
)) {
3919 WRITE_ONCE(vcpu
->ready
, true);
3920 ++vcpu
->stat
.generic
.halt_wakeup
;
3926 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
3930 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3932 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3936 if (kvm_vcpu_wake_up(vcpu
))
3941 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3942 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3943 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3944 * within the vCPU thread itself.
3946 if (vcpu
== __this_cpu_read(kvm_running_vcpu
)) {
3947 if (vcpu
->mode
== IN_GUEST_MODE
)
3948 WRITE_ONCE(vcpu
->mode
, EXITING_GUEST_MODE
);
3953 * Note, the vCPU could get migrated to a different pCPU at any point
3954 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3955 * IPI to the previous pCPU. But, that's ok because the purpose of the
3956 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3957 * vCPU also requires it to leave IN_GUEST_MODE.
3959 if (kvm_arch_vcpu_should_kick(vcpu
)) {
3960 cpu
= READ_ONCE(vcpu
->cpu
);
3961 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
3962 smp_send_reschedule(cpu
);
3967 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
3968 #endif /* !CONFIG_S390 */
3970 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
3973 struct task_struct
*task
= NULL
;
3977 pid
= rcu_dereference(target
->pid
);
3979 task
= get_pid_task(pid
, PIDTYPE_PID
);
3983 ret
= yield_to(task
, 1);
3984 put_task_struct(task
);
3988 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
3991 * Helper that checks whether a VCPU is eligible for directed yield.
3992 * Most eligible candidate to yield is decided by following heuristics:
3994 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3995 * (preempted lock holder), indicated by @in_spin_loop.
3996 * Set at the beginning and cleared at the end of interception/PLE handler.
3998 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3999 * chance last time (mostly it has become eligible now since we have probably
4000 * yielded to lockholder in last iteration. This is done by toggling
4001 * @dy_eligible each time a VCPU checked for eligibility.)
4003 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
4004 * to preempted lock-holder could result in wrong VCPU selection and CPU
4005 * burning. Giving priority for a potential lock-holder increases lock
4008 * Since algorithm is based on heuristics, accessing another VCPU data without
4009 * locking does not harm. It may result in trying to yield to same VCPU, fail
4010 * and continue with next VCPU and so on.
4012 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
4014 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
4017 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
4018 vcpu
->spin_loop
.dy_eligible
;
4020 if (vcpu
->spin_loop
.in_spin_loop
)
4021 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
4030 * Unlike kvm_arch_vcpu_runnable, this function is called outside
4031 * a vcpu_load/vcpu_put pair. However, for most architectures
4032 * kvm_arch_vcpu_runnable does not require vcpu_load.
4034 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
4036 return kvm_arch_vcpu_runnable(vcpu
);
4039 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
4041 if (kvm_arch_dy_runnable(vcpu
))
4044 #ifdef CONFIG_KVM_ASYNC_PF
4045 if (!list_empty_careful(&vcpu
->async_pf
.done
))
4052 bool __weak
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu
*vcpu
)
4057 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
4059 struct kvm
*kvm
= me
->kvm
;
4060 struct kvm_vcpu
*vcpu
;
4061 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
4067 kvm_vcpu_set_in_spin_loop(me
, true);
4069 * We boost the priority of a VCPU that is runnable but not
4070 * currently running, because it got preempted by something
4071 * else and called schedule in __vcpu_run. Hopefully that
4072 * VCPU is holding the lock that we need and will release it.
4073 * We approximate round-robin by starting at the last boosted VCPU.
4075 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
4076 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
4077 if (!pass
&& i
<= last_boosted_vcpu
) {
4078 i
= last_boosted_vcpu
;
4080 } else if (pass
&& i
> last_boosted_vcpu
)
4082 if (!READ_ONCE(vcpu
->ready
))
4086 if (kvm_vcpu_is_blocking(vcpu
) && !vcpu_dy_runnable(vcpu
))
4088 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
4089 !kvm_arch_dy_has_pending_interrupt(vcpu
) &&
4090 !kvm_arch_vcpu_in_kernel(vcpu
))
4092 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
4095 yielded
= kvm_vcpu_yield_to(vcpu
);
4097 kvm
->last_boosted_vcpu
= i
;
4099 } else if (yielded
< 0) {
4106 kvm_vcpu_set_in_spin_loop(me
, false);
4108 /* Ensure vcpu is not eligible during next spinloop */
4109 kvm_vcpu_set_dy_eligible(me
, false);
4111 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
4113 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
4115 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4116 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
4117 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
4118 kvm
->dirty_ring_size
/ PAGE_SIZE
);
4124 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
4126 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
4129 if (vmf
->pgoff
== 0)
4130 page
= virt_to_page(vcpu
->run
);
4132 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
4133 page
= virt_to_page(vcpu
->arch
.pio_data
);
4135 #ifdef CONFIG_KVM_MMIO
4136 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
4137 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
4139 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
4140 page
= kvm_dirty_ring_get_page(
4142 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
4144 return kvm_arch_vcpu_fault(vcpu
, vmf
);
4150 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
4151 .fault
= kvm_vcpu_fault
,
4154 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
4156 struct kvm_vcpu
*vcpu
= file
->private_data
;
4157 unsigned long pages
= vma_pages(vma
);
4159 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
4160 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
4161 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
4164 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
4168 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
4170 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4172 kvm_put_kvm(vcpu
->kvm
);
4176 static const struct file_operations kvm_vcpu_fops
= {
4177 .release
= kvm_vcpu_release
,
4178 .unlocked_ioctl
= kvm_vcpu_ioctl
,
4179 .mmap
= kvm_vcpu_mmap
,
4180 .llseek
= noop_llseek
,
4181 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
4185 * Allocates an inode for the vcpu.
4187 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
4189 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
4191 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
4192 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
4195 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
4196 static int vcpu_get_pid(void *data
, u64
*val
)
4198 struct kvm_vcpu
*vcpu
= data
;
4201 *val
= pid_nr(rcu_dereference(vcpu
->pid
));
4206 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops
, vcpu_get_pid
, NULL
, "%llu\n");
4208 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
4210 struct dentry
*debugfs_dentry
;
4211 char dir_name
[ITOA_MAX_LEN
* 2];
4213 if (!debugfs_initialized())
4216 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
4217 debugfs_dentry
= debugfs_create_dir(dir_name
,
4218 vcpu
->kvm
->debugfs_dentry
);
4219 debugfs_create_file("pid", 0444, debugfs_dentry
, vcpu
,
4220 &vcpu_get_pid_fops
);
4222 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
4227 * Creates some virtual cpus. Good luck creating more than one.
4229 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
4232 struct kvm_vcpu
*vcpu
;
4235 if (id
>= KVM_MAX_VCPU_IDS
)
4238 mutex_lock(&kvm
->lock
);
4239 if (kvm
->created_vcpus
>= kvm
->max_vcpus
) {
4240 mutex_unlock(&kvm
->lock
);
4244 r
= kvm_arch_vcpu_precreate(kvm
, id
);
4246 mutex_unlock(&kvm
->lock
);
4250 kvm
->created_vcpus
++;
4251 mutex_unlock(&kvm
->lock
);
4253 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL_ACCOUNT
);
4256 goto vcpu_decrement
;
4259 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
4260 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
4265 vcpu
->run
= page_address(page
);
4267 kvm_vcpu_init(vcpu
, kvm
, id
);
4269 r
= kvm_arch_vcpu_create(vcpu
);
4271 goto vcpu_free_run_page
;
4273 if (kvm
->dirty_ring_size
) {
4274 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
4275 id
, kvm
->dirty_ring_size
);
4277 goto arch_vcpu_destroy
;
4280 mutex_lock(&kvm
->lock
);
4282 #ifdef CONFIG_LOCKDEP
4283 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
4284 mutex_lock(&vcpu
->mutex
);
4285 mutex_unlock(&vcpu
->mutex
);
4288 if (kvm_get_vcpu_by_id(kvm
, id
)) {
4290 goto unlock_vcpu_destroy
;
4293 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
4294 r
= xa_reserve(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, GFP_KERNEL_ACCOUNT
);
4296 goto unlock_vcpu_destroy
;
4298 /* Now it's all set up, let userspace reach it */
4300 r
= create_vcpu_fd(vcpu
);
4302 goto kvm_put_xa_release
;
4304 if (KVM_BUG_ON(xa_store(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, vcpu
, 0), kvm
)) {
4306 goto kvm_put_xa_release
;
4310 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4311 * pointer before kvm->online_vcpu's incremented value.
4314 atomic_inc(&kvm
->online_vcpus
);
4316 mutex_unlock(&kvm
->lock
);
4317 kvm_arch_vcpu_postcreate(vcpu
);
4318 kvm_create_vcpu_debugfs(vcpu
);
4322 kvm_put_kvm_no_destroy(kvm
);
4323 xa_release(&kvm
->vcpu_array
, vcpu
->vcpu_idx
);
4324 unlock_vcpu_destroy
:
4325 mutex_unlock(&kvm
->lock
);
4326 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
4328 kvm_arch_vcpu_destroy(vcpu
);
4330 free_page((unsigned long)vcpu
->run
);
4332 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
4334 mutex_lock(&kvm
->lock
);
4335 kvm
->created_vcpus
--;
4336 mutex_unlock(&kvm
->lock
);
4340 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
4343 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
4344 vcpu
->sigset_active
= 1;
4345 vcpu
->sigset
= *sigset
;
4347 vcpu
->sigset_active
= 0;
4351 static ssize_t
kvm_vcpu_stats_read(struct file
*file
, char __user
*user_buffer
,
4352 size_t size
, loff_t
*offset
)
4354 struct kvm_vcpu
*vcpu
= file
->private_data
;
4356 return kvm_stats_read(vcpu
->stats_id
, &kvm_vcpu_stats_header
,
4357 &kvm_vcpu_stats_desc
[0], &vcpu
->stat
,
4358 sizeof(vcpu
->stat
), user_buffer
, size
, offset
);
4361 static int kvm_vcpu_stats_release(struct inode
*inode
, struct file
*file
)
4363 struct kvm_vcpu
*vcpu
= file
->private_data
;
4365 kvm_put_kvm(vcpu
->kvm
);
4369 static const struct file_operations kvm_vcpu_stats_fops
= {
4370 .read
= kvm_vcpu_stats_read
,
4371 .release
= kvm_vcpu_stats_release
,
4372 .llseek
= noop_llseek
,
4375 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu
*vcpu
)
4379 char name
[15 + ITOA_MAX_LEN
+ 1];
4381 snprintf(name
, sizeof(name
), "kvm-vcpu-stats:%d", vcpu
->vcpu_id
);
4383 fd
= get_unused_fd_flags(O_CLOEXEC
);
4387 file
= anon_inode_getfile(name
, &kvm_vcpu_stats_fops
, vcpu
, O_RDONLY
);
4390 return PTR_ERR(file
);
4393 kvm_get_kvm(vcpu
->kvm
);
4395 file
->f_mode
|= FMODE_PREAD
;
4396 fd_install(fd
, file
);
4401 static long kvm_vcpu_ioctl(struct file
*filp
,
4402 unsigned int ioctl
, unsigned long arg
)
4404 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4405 void __user
*argp
= (void __user
*)arg
;
4407 struct kvm_fpu
*fpu
= NULL
;
4408 struct kvm_sregs
*kvm_sregs
= NULL
;
4410 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4413 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
4417 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4418 * execution; mutex_lock() would break them.
4420 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
4421 if (r
!= -ENOIOCTLCMD
)
4424 if (mutex_lock_killable(&vcpu
->mutex
))
4432 oldpid
= rcu_access_pointer(vcpu
->pid
);
4433 if (unlikely(oldpid
!= task_pid(current
))) {
4434 /* The thread running this VCPU changed. */
4437 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
4441 newpid
= get_task_pid(current
, PIDTYPE_PID
);
4442 rcu_assign_pointer(vcpu
->pid
, newpid
);
4447 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
4448 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
4451 case KVM_GET_REGS
: {
4452 struct kvm_regs
*kvm_regs
;
4455 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
4458 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
4462 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
4469 case KVM_SET_REGS
: {
4470 struct kvm_regs
*kvm_regs
;
4472 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
4473 if (IS_ERR(kvm_regs
)) {
4474 r
= PTR_ERR(kvm_regs
);
4477 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
4481 case KVM_GET_SREGS
: {
4482 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
4483 GFP_KERNEL_ACCOUNT
);
4487 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
4491 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
4496 case KVM_SET_SREGS
: {
4497 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
4498 if (IS_ERR(kvm_sregs
)) {
4499 r
= PTR_ERR(kvm_sregs
);
4503 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
4506 case KVM_GET_MP_STATE
: {
4507 struct kvm_mp_state mp_state
;
4509 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
4513 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
4518 case KVM_SET_MP_STATE
: {
4519 struct kvm_mp_state mp_state
;
4522 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
4524 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
4527 case KVM_TRANSLATE
: {
4528 struct kvm_translation tr
;
4531 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
4533 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
4537 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
4542 case KVM_SET_GUEST_DEBUG
: {
4543 struct kvm_guest_debug dbg
;
4546 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
4548 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
4551 case KVM_SET_SIGNAL_MASK
: {
4552 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4553 struct kvm_signal_mask kvm_sigmask
;
4554 sigset_t sigset
, *p
;
4559 if (copy_from_user(&kvm_sigmask
, argp
,
4560 sizeof(kvm_sigmask
)))
4563 if (kvm_sigmask
.len
!= sizeof(sigset
))
4566 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
4571 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
4575 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
4579 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
4583 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
4589 fpu
= memdup_user(argp
, sizeof(*fpu
));
4595 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
4598 case KVM_GET_STATS_FD
: {
4599 r
= kvm_vcpu_ioctl_get_stats_fd(vcpu
);
4603 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
4606 mutex_unlock(&vcpu
->mutex
);
4612 #ifdef CONFIG_KVM_COMPAT
4613 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
4614 unsigned int ioctl
, unsigned long arg
)
4616 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4617 void __user
*argp
= compat_ptr(arg
);
4620 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4624 case KVM_SET_SIGNAL_MASK
: {
4625 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4626 struct kvm_signal_mask kvm_sigmask
;
4631 if (copy_from_user(&kvm_sigmask
, argp
,
4632 sizeof(kvm_sigmask
)))
4635 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
4638 if (get_compat_sigset(&sigset
,
4639 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
4641 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
4643 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
4647 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
4655 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
4657 struct kvm_device
*dev
= filp
->private_data
;
4660 return dev
->ops
->mmap(dev
, vma
);
4665 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
4666 int (*accessor
)(struct kvm_device
*dev
,
4667 struct kvm_device_attr
*attr
),
4670 struct kvm_device_attr attr
;
4675 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
4678 return accessor(dev
, &attr
);
4681 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
4684 struct kvm_device
*dev
= filp
->private_data
;
4686 if (dev
->kvm
->mm
!= current
->mm
|| dev
->kvm
->vm_dead
)
4690 case KVM_SET_DEVICE_ATTR
:
4691 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
4692 case KVM_GET_DEVICE_ATTR
:
4693 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
4694 case KVM_HAS_DEVICE_ATTR
:
4695 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
4697 if (dev
->ops
->ioctl
)
4698 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
4704 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
4706 struct kvm_device
*dev
= filp
->private_data
;
4707 struct kvm
*kvm
= dev
->kvm
;
4709 if (dev
->ops
->release
) {
4710 mutex_lock(&kvm
->lock
);
4711 list_del(&dev
->vm_node
);
4712 dev
->ops
->release(dev
);
4713 mutex_unlock(&kvm
->lock
);
4720 static const struct file_operations kvm_device_fops
= {
4721 .unlocked_ioctl
= kvm_device_ioctl
,
4722 .release
= kvm_device_release
,
4723 KVM_COMPAT(kvm_device_ioctl
),
4724 .mmap
= kvm_device_mmap
,
4727 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
4729 if (filp
->f_op
!= &kvm_device_fops
)
4732 return filp
->private_data
;
4735 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
4736 #ifdef CONFIG_KVM_MPIC
4737 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
4738 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
4742 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
4744 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
4747 if (kvm_device_ops_table
[type
] != NULL
)
4750 kvm_device_ops_table
[type
] = ops
;
4754 void kvm_unregister_device_ops(u32 type
)
4756 if (kvm_device_ops_table
[type
] != NULL
)
4757 kvm_device_ops_table
[type
] = NULL
;
4760 static int kvm_ioctl_create_device(struct kvm
*kvm
,
4761 struct kvm_create_device
*cd
)
4763 const struct kvm_device_ops
*ops
;
4764 struct kvm_device
*dev
;
4765 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
4769 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
4772 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
4773 ops
= kvm_device_ops_table
[type
];
4780 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
4787 mutex_lock(&kvm
->lock
);
4788 ret
= ops
->create(dev
, type
);
4790 mutex_unlock(&kvm
->lock
);
4794 list_add(&dev
->vm_node
, &kvm
->devices
);
4795 mutex_unlock(&kvm
->lock
);
4801 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
4803 kvm_put_kvm_no_destroy(kvm
);
4804 mutex_lock(&kvm
->lock
);
4805 list_del(&dev
->vm_node
);
4808 mutex_unlock(&kvm
->lock
);
4818 static int kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
4821 case KVM_CAP_USER_MEMORY
:
4822 case KVM_CAP_USER_MEMORY2
:
4823 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
4824 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
4825 case KVM_CAP_INTERNAL_ERROR_DATA
:
4826 #ifdef CONFIG_HAVE_KVM_MSI
4827 case KVM_CAP_SIGNAL_MSI
:
4829 #ifdef CONFIG_HAVE_KVM_IRQFD
4832 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
4833 case KVM_CAP_CHECK_EXTENSION_VM
:
4834 case KVM_CAP_ENABLE_CAP_VM
:
4835 case KVM_CAP_HALT_POLL
:
4837 #ifdef CONFIG_KVM_MMIO
4838 case KVM_CAP_COALESCED_MMIO
:
4839 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
4840 case KVM_CAP_COALESCED_PIO
:
4843 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4844 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
4845 return KVM_DIRTY_LOG_MANUAL_CAPS
;
4847 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4848 case KVM_CAP_IRQ_ROUTING
:
4849 return KVM_MAX_IRQ_ROUTES
;
4851 #if KVM_MAX_NR_ADDRESS_SPACES > 1
4852 case KVM_CAP_MULTI_ADDRESS_SPACE
:
4854 return kvm_arch_nr_memslot_as_ids(kvm
);
4855 return KVM_MAX_NR_ADDRESS_SPACES
;
4857 case KVM_CAP_NR_MEMSLOTS
:
4858 return KVM_USER_MEM_SLOTS
;
4859 case KVM_CAP_DIRTY_LOG_RING
:
4860 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4861 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4865 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4866 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4867 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4871 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4872 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
:
4874 case KVM_CAP_BINARY_STATS_FD
:
4875 case KVM_CAP_SYSTEM_EVENT_DATA
:
4876 case KVM_CAP_DEVICE_CTRL
:
4878 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
4879 case KVM_CAP_MEMORY_ATTRIBUTES
:
4880 return kvm_supported_mem_attributes(kvm
);
4882 #ifdef CONFIG_KVM_PRIVATE_MEM
4883 case KVM_CAP_GUEST_MEMFD
:
4884 return !kvm
|| kvm_arch_has_private_mem(kvm
);
4889 return kvm_vm_ioctl_check_extension(kvm
, arg
);
4892 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
4896 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
4899 /* the size should be power of 2 */
4900 if (!size
|| (size
& (size
- 1)))
4903 /* Should be bigger to keep the reserved entries, or a page */
4904 if (size
< kvm_dirty_ring_get_rsvd_entries() *
4905 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
4908 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
4909 sizeof(struct kvm_dirty_gfn
))
4912 /* We only allow it to set once */
4913 if (kvm
->dirty_ring_size
)
4916 mutex_lock(&kvm
->lock
);
4918 if (kvm
->created_vcpus
) {
4919 /* We don't allow to change this value after vcpu created */
4922 kvm
->dirty_ring_size
= size
;
4926 mutex_unlock(&kvm
->lock
);
4930 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
4933 struct kvm_vcpu
*vcpu
;
4936 if (!kvm
->dirty_ring_size
)
4939 mutex_lock(&kvm
->slots_lock
);
4941 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4942 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
4944 mutex_unlock(&kvm
->slots_lock
);
4947 kvm_flush_remote_tlbs(kvm
);
4952 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4953 struct kvm_enable_cap
*cap
)
4958 bool kvm_are_all_memslots_empty(struct kvm
*kvm
)
4962 lockdep_assert_held(&kvm
->slots_lock
);
4964 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
4965 if (!kvm_memslots_empty(__kvm_memslots(kvm
, i
)))
4971 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty
);
4973 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
4974 struct kvm_enable_cap
*cap
)
4977 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4978 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
4979 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
4981 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
4982 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
4984 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
4986 kvm
->manual_dirty_log_protect
= cap
->args
[0];
4990 case KVM_CAP_HALT_POLL
: {
4991 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
4994 kvm
->max_halt_poll_ns
= cap
->args
[0];
4997 * Ensure kvm->override_halt_poll_ns does not become visible
4998 * before kvm->max_halt_poll_ns.
5000 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
5003 kvm
->override_halt_poll_ns
= true;
5007 case KVM_CAP_DIRTY_LOG_RING
:
5008 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
5009 if (!kvm_vm_ioctl_check_extension_generic(kvm
, cap
->cap
))
5012 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
5013 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
: {
5016 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
) ||
5017 !kvm
->dirty_ring_size
|| cap
->flags
)
5020 mutex_lock(&kvm
->slots_lock
);
5023 * For simplicity, allow enabling ring+bitmap if and only if
5024 * there are no memslots, e.g. to ensure all memslots allocate
5025 * a bitmap after the capability is enabled.
5027 if (kvm_are_all_memslots_empty(kvm
)) {
5028 kvm
->dirty_ring_with_bitmap
= true;
5032 mutex_unlock(&kvm
->slots_lock
);
5037 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
5041 static ssize_t
kvm_vm_stats_read(struct file
*file
, char __user
*user_buffer
,
5042 size_t size
, loff_t
*offset
)
5044 struct kvm
*kvm
= file
->private_data
;
5046 return kvm_stats_read(kvm
->stats_id
, &kvm_vm_stats_header
,
5047 &kvm_vm_stats_desc
[0], &kvm
->stat
,
5048 sizeof(kvm
->stat
), user_buffer
, size
, offset
);
5051 static int kvm_vm_stats_release(struct inode
*inode
, struct file
*file
)
5053 struct kvm
*kvm
= file
->private_data
;
5059 static const struct file_operations kvm_vm_stats_fops
= {
5060 .read
= kvm_vm_stats_read
,
5061 .release
= kvm_vm_stats_release
,
5062 .llseek
= noop_llseek
,
5065 static int kvm_vm_ioctl_get_stats_fd(struct kvm
*kvm
)
5070 fd
= get_unused_fd_flags(O_CLOEXEC
);
5074 file
= anon_inode_getfile("kvm-vm-stats",
5075 &kvm_vm_stats_fops
, kvm
, O_RDONLY
);
5078 return PTR_ERR(file
);
5083 file
->f_mode
|= FMODE_PREAD
;
5084 fd_install(fd
, file
);
5089 #define SANITY_CHECK_MEM_REGION_FIELD(field) \
5091 BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) != \
5092 offsetof(struct kvm_userspace_memory_region2, field)); \
5093 BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) != \
5094 sizeof_field(struct kvm_userspace_memory_region2, field)); \
5097 static long kvm_vm_ioctl(struct file
*filp
,
5098 unsigned int ioctl
, unsigned long arg
)
5100 struct kvm
*kvm
= filp
->private_data
;
5101 void __user
*argp
= (void __user
*)arg
;
5104 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5107 case KVM_CREATE_VCPU
:
5108 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
5110 case KVM_ENABLE_CAP
: {
5111 struct kvm_enable_cap cap
;
5114 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
5116 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
5119 case KVM_SET_USER_MEMORY_REGION2
:
5120 case KVM_SET_USER_MEMORY_REGION
: {
5121 struct kvm_userspace_memory_region2 mem
;
5124 if (ioctl
== KVM_SET_USER_MEMORY_REGION
) {
5126 * Fields beyond struct kvm_userspace_memory_region shouldn't be
5127 * accessed, but avoid leaking kernel memory in case of a bug.
5129 memset(&mem
, 0, sizeof(mem
));
5130 size
= sizeof(struct kvm_userspace_memory_region
);
5132 size
= sizeof(struct kvm_userspace_memory_region2
);
5135 /* Ensure the common parts of the two structs are identical. */
5136 SANITY_CHECK_MEM_REGION_FIELD(slot
);
5137 SANITY_CHECK_MEM_REGION_FIELD(flags
);
5138 SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr
);
5139 SANITY_CHECK_MEM_REGION_FIELD(memory_size
);
5140 SANITY_CHECK_MEM_REGION_FIELD(userspace_addr
);
5143 if (copy_from_user(&mem
, argp
, size
))
5147 if (ioctl
== KVM_SET_USER_MEMORY_REGION
&&
5148 (mem
.flags
& ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS
))
5151 r
= kvm_vm_ioctl_set_memory_region(kvm
, &mem
);
5154 case KVM_GET_DIRTY_LOG
: {
5155 struct kvm_dirty_log log
;
5158 if (copy_from_user(&log
, argp
, sizeof(log
)))
5160 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5163 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5164 case KVM_CLEAR_DIRTY_LOG
: {
5165 struct kvm_clear_dirty_log log
;
5168 if (copy_from_user(&log
, argp
, sizeof(log
)))
5170 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5174 #ifdef CONFIG_KVM_MMIO
5175 case KVM_REGISTER_COALESCED_MMIO
: {
5176 struct kvm_coalesced_mmio_zone zone
;
5179 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
5181 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
5184 case KVM_UNREGISTER_COALESCED_MMIO
: {
5185 struct kvm_coalesced_mmio_zone zone
;
5188 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
5190 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
5195 struct kvm_irqfd data
;
5198 if (copy_from_user(&data
, argp
, sizeof(data
)))
5200 r
= kvm_irqfd(kvm
, &data
);
5203 case KVM_IOEVENTFD
: {
5204 struct kvm_ioeventfd data
;
5207 if (copy_from_user(&data
, argp
, sizeof(data
)))
5209 r
= kvm_ioeventfd(kvm
, &data
);
5212 #ifdef CONFIG_HAVE_KVM_MSI
5213 case KVM_SIGNAL_MSI
: {
5217 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
5219 r
= kvm_send_userspace_msi(kvm
, &msi
);
5223 #ifdef __KVM_HAVE_IRQ_LINE
5224 case KVM_IRQ_LINE_STATUS
:
5225 case KVM_IRQ_LINE
: {
5226 struct kvm_irq_level irq_event
;
5229 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
5232 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
5233 ioctl
== KVM_IRQ_LINE_STATUS
);
5238 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
5239 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
5247 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
5248 case KVM_SET_GSI_ROUTING
: {
5249 struct kvm_irq_routing routing
;
5250 struct kvm_irq_routing __user
*urouting
;
5251 struct kvm_irq_routing_entry
*entries
= NULL
;
5254 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
5257 if (!kvm_arch_can_set_irq_routing(kvm
))
5259 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
5265 entries
= vmemdup_array_user(urouting
->entries
,
5266 routing
.nr
, sizeof(*entries
));
5267 if (IS_ERR(entries
)) {
5268 r
= PTR_ERR(entries
);
5272 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
5277 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
5278 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
5279 case KVM_SET_MEMORY_ATTRIBUTES
: {
5280 struct kvm_memory_attributes attrs
;
5283 if (copy_from_user(&attrs
, argp
, sizeof(attrs
)))
5286 r
= kvm_vm_ioctl_set_mem_attributes(kvm
, &attrs
);
5289 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
5290 case KVM_CREATE_DEVICE
: {
5291 struct kvm_create_device cd
;
5294 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
5297 r
= kvm_ioctl_create_device(kvm
, &cd
);
5302 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
5308 case KVM_CHECK_EXTENSION
:
5309 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
5311 case KVM_RESET_DIRTY_RINGS
:
5312 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
5314 case KVM_GET_STATS_FD
:
5315 r
= kvm_vm_ioctl_get_stats_fd(kvm
);
5317 #ifdef CONFIG_KVM_PRIVATE_MEM
5318 case KVM_CREATE_GUEST_MEMFD
: {
5319 struct kvm_create_guest_memfd guest_memfd
;
5322 if (copy_from_user(&guest_memfd
, argp
, sizeof(guest_memfd
)))
5325 r
= kvm_gmem_create(kvm
, &guest_memfd
);
5330 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
5336 #ifdef CONFIG_KVM_COMPAT
5337 struct compat_kvm_dirty_log
{
5341 compat_uptr_t dirty_bitmap
; /* one bit per page */
5346 struct compat_kvm_clear_dirty_log
{
5351 compat_uptr_t dirty_bitmap
; /* one bit per page */
5356 long __weak
kvm_arch_vm_compat_ioctl(struct file
*filp
, unsigned int ioctl
,
5362 static long kvm_vm_compat_ioctl(struct file
*filp
,
5363 unsigned int ioctl
, unsigned long arg
)
5365 struct kvm
*kvm
= filp
->private_data
;
5368 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5371 r
= kvm_arch_vm_compat_ioctl(filp
, ioctl
, arg
);
5376 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5377 case KVM_CLEAR_DIRTY_LOG
: {
5378 struct compat_kvm_clear_dirty_log compat_log
;
5379 struct kvm_clear_dirty_log log
;
5381 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5382 sizeof(compat_log
)))
5384 log
.slot
= compat_log
.slot
;
5385 log
.num_pages
= compat_log
.num_pages
;
5386 log
.first_page
= compat_log
.first_page
;
5387 log
.padding2
= compat_log
.padding2
;
5388 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5390 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5394 case KVM_GET_DIRTY_LOG
: {
5395 struct compat_kvm_dirty_log compat_log
;
5396 struct kvm_dirty_log log
;
5398 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5399 sizeof(compat_log
)))
5401 log
.slot
= compat_log
.slot
;
5402 log
.padding1
= compat_log
.padding1
;
5403 log
.padding2
= compat_log
.padding2
;
5404 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5406 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5410 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
5416 static const struct file_operations kvm_vm_fops
= {
5417 .release
= kvm_vm_release
,
5418 .unlocked_ioctl
= kvm_vm_ioctl
,
5419 .llseek
= noop_llseek
,
5420 KVM_COMPAT(kvm_vm_compat_ioctl
),
5423 bool file_is_kvm(struct file
*file
)
5425 return file
&& file
->f_op
== &kvm_vm_fops
;
5427 EXPORT_SYMBOL_GPL(file_is_kvm
);
5429 static int kvm_dev_ioctl_create_vm(unsigned long type
)
5431 char fdname
[ITOA_MAX_LEN
+ 1];
5436 fd
= get_unused_fd_flags(O_CLOEXEC
);
5440 snprintf(fdname
, sizeof(fdname
), "%d", fd
);
5442 kvm
= kvm_create_vm(type
, fdname
);
5448 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
5455 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5456 * already set, with ->release() being kvm_vm_release(). In error
5457 * cases it will be called by the final fput(file) and will take
5458 * care of doing kvm_put_kvm(kvm).
5460 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
5462 fd_install(fd
, file
);
5472 static long kvm_dev_ioctl(struct file
*filp
,
5473 unsigned int ioctl
, unsigned long arg
)
5478 case KVM_GET_API_VERSION
:
5481 r
= KVM_API_VERSION
;
5484 r
= kvm_dev_ioctl_create_vm(arg
);
5486 case KVM_CHECK_EXTENSION
:
5487 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
5489 case KVM_GET_VCPU_MMAP_SIZE
:
5492 r
= PAGE_SIZE
; /* struct kvm_run */
5494 r
+= PAGE_SIZE
; /* pio data page */
5496 #ifdef CONFIG_KVM_MMIO
5497 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
5500 case KVM_TRACE_ENABLE
:
5501 case KVM_TRACE_PAUSE
:
5502 case KVM_TRACE_DISABLE
:
5506 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
5512 static struct file_operations kvm_chardev_ops
= {
5513 .unlocked_ioctl
= kvm_dev_ioctl
,
5514 .llseek
= noop_llseek
,
5515 KVM_COMPAT(kvm_dev_ioctl
),
5518 static struct miscdevice kvm_dev
= {
5524 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5525 __visible
bool kvm_rebooting
;
5526 EXPORT_SYMBOL_GPL(kvm_rebooting
);
5528 static DEFINE_PER_CPU(bool, hardware_enabled
);
5529 static int kvm_usage_count
;
5531 static int __hardware_enable_nolock(void)
5533 if (__this_cpu_read(hardware_enabled
))
5536 if (kvm_arch_hardware_enable()) {
5537 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5538 raw_smp_processor_id());
5542 __this_cpu_write(hardware_enabled
, true);
5546 static void hardware_enable_nolock(void *failed
)
5548 if (__hardware_enable_nolock())
5552 static int kvm_online_cpu(unsigned int cpu
)
5557 * Abort the CPU online process if hardware virtualization cannot
5558 * be enabled. Otherwise running VMs would encounter unrecoverable
5559 * errors when scheduled to this CPU.
5561 mutex_lock(&kvm_lock
);
5562 if (kvm_usage_count
)
5563 ret
= __hardware_enable_nolock();
5564 mutex_unlock(&kvm_lock
);
5568 static void hardware_disable_nolock(void *junk
)
5571 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5572 * hardware, not just CPUs that successfully enabled hardware!
5574 if (!__this_cpu_read(hardware_enabled
))
5577 kvm_arch_hardware_disable();
5579 __this_cpu_write(hardware_enabled
, false);
5582 static int kvm_offline_cpu(unsigned int cpu
)
5584 mutex_lock(&kvm_lock
);
5585 if (kvm_usage_count
)
5586 hardware_disable_nolock(NULL
);
5587 mutex_unlock(&kvm_lock
);
5591 static void hardware_disable_all_nolock(void)
5593 BUG_ON(!kvm_usage_count
);
5596 if (!kvm_usage_count
)
5597 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5600 static void hardware_disable_all(void)
5603 mutex_lock(&kvm_lock
);
5604 hardware_disable_all_nolock();
5605 mutex_unlock(&kvm_lock
);
5609 static int hardware_enable_all(void)
5611 atomic_t failed
= ATOMIC_INIT(0);
5615 * Do not enable hardware virtualization if the system is going down.
5616 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5617 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5618 * after kvm_reboot() is called. Note, this relies on system_state
5619 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5620 * hook instead of registering a dedicated reboot notifier (the latter
5621 * runs before system_state is updated).
5623 if (system_state
== SYSTEM_HALT
|| system_state
== SYSTEM_POWER_OFF
||
5624 system_state
== SYSTEM_RESTART
)
5628 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5629 * is called, and so on_each_cpu() between them includes the CPU that
5630 * is being onlined. As a result, hardware_enable_nolock() may get
5631 * invoked before kvm_online_cpu(), which also enables hardware if the
5632 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5633 * enable hardware multiple times.
5636 mutex_lock(&kvm_lock
);
5641 if (kvm_usage_count
== 1) {
5642 on_each_cpu(hardware_enable_nolock
, &failed
, 1);
5644 if (atomic_read(&failed
)) {
5645 hardware_disable_all_nolock();
5650 mutex_unlock(&kvm_lock
);
5656 static void kvm_shutdown(void)
5659 * Disable hardware virtualization and set kvm_rebooting to indicate
5660 * that KVM has asynchronously disabled hardware virtualization, i.e.
5661 * that relevant errors and exceptions aren't entirely unexpected.
5662 * Some flavors of hardware virtualization need to be disabled before
5663 * transferring control to firmware (to perform shutdown/reboot), e.g.
5664 * on x86, virtualization can block INIT interrupts, which are used by
5665 * firmware to pull APs back under firmware control. Note, this path
5666 * is used for both shutdown and reboot scenarios, i.e. neither name is
5667 * 100% comprehensive.
5669 pr_info("kvm: exiting hardware virtualization\n");
5670 kvm_rebooting
= true;
5671 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5674 static int kvm_suspend(void)
5677 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5678 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5679 * is stable. Assert that kvm_lock is not held to ensure the system
5680 * isn't suspended while KVM is enabling hardware. Hardware enabling
5681 * can be preempted, but the task cannot be frozen until it has dropped
5682 * all locks (userspace tasks are frozen via a fake signal).
5684 lockdep_assert_not_held(&kvm_lock
);
5685 lockdep_assert_irqs_disabled();
5687 if (kvm_usage_count
)
5688 hardware_disable_nolock(NULL
);
5692 static void kvm_resume(void)
5694 lockdep_assert_not_held(&kvm_lock
);
5695 lockdep_assert_irqs_disabled();
5697 if (kvm_usage_count
)
5698 WARN_ON_ONCE(__hardware_enable_nolock());
5701 static struct syscore_ops kvm_syscore_ops
= {
5702 .suspend
= kvm_suspend
,
5703 .resume
= kvm_resume
,
5704 .shutdown
= kvm_shutdown
,
5706 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5707 static int hardware_enable_all(void)
5712 static void hardware_disable_all(void)
5716 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5718 static void kvm_iodevice_destructor(struct kvm_io_device
*dev
)
5720 if (dev
->ops
->destructor
)
5721 dev
->ops
->destructor(dev
);
5724 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
5728 for (i
= 0; i
< bus
->dev_count
; i
++) {
5729 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
5731 kvm_iodevice_destructor(pos
);
5736 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
5737 const struct kvm_io_range
*r2
)
5739 gpa_t addr1
= r1
->addr
;
5740 gpa_t addr2
= r2
->addr
;
5745 /* If r2->len == 0, match the exact address. If r2->len != 0,
5746 * accept any overlapping write. Any order is acceptable for
5747 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5748 * we process all of them.
5761 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
5763 return kvm_io_bus_cmp(p1
, p2
);
5766 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
5767 gpa_t addr
, int len
)
5769 struct kvm_io_range
*range
, key
;
5772 key
= (struct kvm_io_range
) {
5777 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
5778 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
5782 off
= range
- bus
->range
;
5784 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
5790 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5791 struct kvm_io_range
*range
, const void *val
)
5795 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5799 while (idx
< bus
->dev_count
&&
5800 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5801 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5810 /* kvm_io_bus_write - called under kvm->slots_lock */
5811 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5812 int len
, const void *val
)
5814 struct kvm_io_bus
*bus
;
5815 struct kvm_io_range range
;
5818 range
= (struct kvm_io_range
) {
5823 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5826 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5827 return r
< 0 ? r
: 0;
5829 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
5831 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5832 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
5833 gpa_t addr
, int len
, const void *val
, long cookie
)
5835 struct kvm_io_bus
*bus
;
5836 struct kvm_io_range range
;
5838 range
= (struct kvm_io_range
) {
5843 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5847 /* First try the device referenced by cookie. */
5848 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
5849 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
5850 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
5855 * cookie contained garbage; fall back to search and return the
5856 * correct cookie value.
5858 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5861 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5862 struct kvm_io_range
*range
, void *val
)
5866 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5870 while (idx
< bus
->dev_count
&&
5871 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5872 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5881 /* kvm_io_bus_read - called under kvm->slots_lock */
5882 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5885 struct kvm_io_bus
*bus
;
5886 struct kvm_io_range range
;
5889 range
= (struct kvm_io_range
) {
5894 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5897 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
5898 return r
< 0 ? r
: 0;
5901 /* Caller must hold slots_lock. */
5902 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
5903 int len
, struct kvm_io_device
*dev
)
5906 struct kvm_io_bus
*new_bus
, *bus
;
5907 struct kvm_io_range range
;
5909 bus
= kvm_get_bus(kvm
, bus_idx
);
5913 /* exclude ioeventfd which is limited by maximum fd */
5914 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
5917 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
5918 GFP_KERNEL_ACCOUNT
);
5922 range
= (struct kvm_io_range
) {
5928 for (i
= 0; i
< bus
->dev_count
; i
++)
5929 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
5932 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
5933 new_bus
->dev_count
++;
5934 new_bus
->range
[i
] = range
;
5935 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
5936 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
5937 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5938 synchronize_srcu_expedited(&kvm
->srcu
);
5944 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5945 struct kvm_io_device
*dev
)
5948 struct kvm_io_bus
*new_bus
, *bus
;
5950 lockdep_assert_held(&kvm
->slots_lock
);
5952 bus
= kvm_get_bus(kvm
, bus_idx
);
5956 for (i
= 0; i
< bus
->dev_count
; i
++) {
5957 if (bus
->range
[i
].dev
== dev
) {
5962 if (i
== bus
->dev_count
)
5965 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
5966 GFP_KERNEL_ACCOUNT
);
5968 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
5969 new_bus
->dev_count
--;
5970 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
5971 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
5974 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5975 synchronize_srcu_expedited(&kvm
->srcu
);
5978 * If NULL bus is installed, destroy the old bus, including all the
5979 * attached devices. Otherwise, destroy the caller's device only.
5982 pr_err("kvm: failed to shrink bus, removing it completely\n");
5983 kvm_io_bus_destroy(bus
);
5987 kvm_iodevice_destructor(dev
);
5992 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5995 struct kvm_io_bus
*bus
;
5996 int dev_idx
, srcu_idx
;
5997 struct kvm_io_device
*iodev
= NULL
;
5999 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
6001 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
6005 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
6009 iodev
= bus
->range
[dev_idx
].dev
;
6012 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
6016 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
6018 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
6019 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
6023 struct kvm_stat_data
*stat_data
= inode
->i_private
;
6026 * The debugfs files are a reference to the kvm struct which
6027 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
6028 * avoids the race between open and the removal of the debugfs directory.
6030 if (!kvm_get_kvm_safe(stat_data
->kvm
))
6033 ret
= simple_attr_open(inode
, file
, get
,
6034 kvm_stats_debugfs_mode(stat_data
->desc
) & 0222
6037 kvm_put_kvm(stat_data
->kvm
);
6042 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
6044 struct kvm_stat_data
*stat_data
= inode
->i_private
;
6046 simple_attr_release(inode
, file
);
6047 kvm_put_kvm(stat_data
->kvm
);
6052 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
6054 *val
= *(u64
*)((void *)(&kvm
->stat
) + offset
);
6059 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
6061 *(u64
*)((void *)(&kvm
->stat
) + offset
) = 0;
6066 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
6069 struct kvm_vcpu
*vcpu
;
6073 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6074 *val
+= *(u64
*)((void *)(&vcpu
->stat
) + offset
);
6079 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
6082 struct kvm_vcpu
*vcpu
;
6084 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6085 *(u64
*)((void *)(&vcpu
->stat
) + offset
) = 0;
6090 static int kvm_stat_data_get(void *data
, u64
*val
)
6093 struct kvm_stat_data
*stat_data
= data
;
6095 switch (stat_data
->kind
) {
6097 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
6098 stat_data
->desc
->desc
.offset
, val
);
6101 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
6102 stat_data
->desc
->desc
.offset
, val
);
6109 static int kvm_stat_data_clear(void *data
, u64 val
)
6112 struct kvm_stat_data
*stat_data
= data
;
6117 switch (stat_data
->kind
) {
6119 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
6120 stat_data
->desc
->desc
.offset
);
6123 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
6124 stat_data
->desc
->desc
.offset
);
6131 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
6133 __simple_attr_check_format("%llu\n", 0ull);
6134 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
6135 kvm_stat_data_clear
, "%llu\n");
6138 static const struct file_operations stat_fops_per_vm
= {
6139 .owner
= THIS_MODULE
,
6140 .open
= kvm_stat_data_open
,
6141 .release
= kvm_debugfs_release
,
6142 .read
= simple_attr_read
,
6143 .write
= simple_attr_write
,
6144 .llseek
= no_llseek
,
6147 static int vm_stat_get(void *_offset
, u64
*val
)
6149 unsigned offset
= (long)_offset
;
6154 mutex_lock(&kvm_lock
);
6155 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6156 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
6159 mutex_unlock(&kvm_lock
);
6163 static int vm_stat_clear(void *_offset
, u64 val
)
6165 unsigned offset
= (long)_offset
;
6171 mutex_lock(&kvm_lock
);
6172 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6173 kvm_clear_stat_per_vm(kvm
, offset
);
6175 mutex_unlock(&kvm_lock
);
6180 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
6181 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops
, vm_stat_get
, NULL
, "%llu\n");
6183 static int vcpu_stat_get(void *_offset
, u64
*val
)
6185 unsigned offset
= (long)_offset
;
6190 mutex_lock(&kvm_lock
);
6191 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6192 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
6195 mutex_unlock(&kvm_lock
);
6199 static int vcpu_stat_clear(void *_offset
, u64 val
)
6201 unsigned offset
= (long)_offset
;
6207 mutex_lock(&kvm_lock
);
6208 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6209 kvm_clear_stat_per_vcpu(kvm
, offset
);
6211 mutex_unlock(&kvm_lock
);
6216 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
6218 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops
, vcpu_stat_get
, NULL
, "%llu\n");
6220 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
6222 struct kobj_uevent_env
*env
;
6223 unsigned long long created
, active
;
6225 if (!kvm_dev
.this_device
|| !kvm
)
6228 mutex_lock(&kvm_lock
);
6229 if (type
== KVM_EVENT_CREATE_VM
) {
6230 kvm_createvm_count
++;
6232 } else if (type
== KVM_EVENT_DESTROY_VM
) {
6235 created
= kvm_createvm_count
;
6236 active
= kvm_active_vms
;
6237 mutex_unlock(&kvm_lock
);
6239 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
6243 add_uevent_var(env
, "CREATED=%llu", created
);
6244 add_uevent_var(env
, "COUNT=%llu", active
);
6246 if (type
== KVM_EVENT_CREATE_VM
) {
6247 add_uevent_var(env
, "EVENT=create");
6248 kvm
->userspace_pid
= task_pid_nr(current
);
6249 } else if (type
== KVM_EVENT_DESTROY_VM
) {
6250 add_uevent_var(env
, "EVENT=destroy");
6252 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
6254 if (!IS_ERR(kvm
->debugfs_dentry
)) {
6255 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
6258 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
6260 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
6264 /* no need for checks, since we are adding at most only 5 keys */
6265 env
->envp
[env
->envp_idx
++] = NULL
;
6266 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
6270 static void kvm_init_debug(void)
6272 const struct file_operations
*fops
;
6273 const struct _kvm_stats_desc
*pdesc
;
6276 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
6278 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
6279 pdesc
= &kvm_vm_stats_desc
[i
];
6280 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
6281 fops
= &vm_stat_fops
;
6283 fops
= &vm_stat_readonly_fops
;
6284 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
6286 (void *)(long)pdesc
->desc
.offset
, fops
);
6289 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
6290 pdesc
= &kvm_vcpu_stats_desc
[i
];
6291 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
6292 fops
= &vcpu_stat_fops
;
6294 fops
= &vcpu_stat_readonly_fops
;
6295 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
6297 (void *)(long)pdesc
->desc
.offset
, fops
);
6302 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
6304 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
6307 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
6309 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
6311 WRITE_ONCE(vcpu
->preempted
, false);
6312 WRITE_ONCE(vcpu
->ready
, false);
6314 __this_cpu_write(kvm_running_vcpu
, vcpu
);
6315 kvm_arch_sched_in(vcpu
, cpu
);
6316 kvm_arch_vcpu_load(vcpu
, cpu
);
6319 static void kvm_sched_out(struct preempt_notifier
*pn
,
6320 struct task_struct
*next
)
6322 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
6324 if (current
->on_rq
) {
6325 WRITE_ONCE(vcpu
->preempted
, true);
6326 WRITE_ONCE(vcpu
->ready
, true);
6328 kvm_arch_vcpu_put(vcpu
);
6329 __this_cpu_write(kvm_running_vcpu
, NULL
);
6333 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
6335 * We can disable preemption locally around accessing the per-CPU variable,
6336 * and use the resolved vcpu pointer after enabling preemption again,
6337 * because even if the current thread is migrated to another CPU, reading
6338 * the per-CPU value later will give us the same value as we update the
6339 * per-CPU variable in the preempt notifier handlers.
6341 struct kvm_vcpu
*kvm_get_running_vcpu(void)
6343 struct kvm_vcpu
*vcpu
;
6346 vcpu
= __this_cpu_read(kvm_running_vcpu
);
6351 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
6354 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
6356 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
6358 return &kvm_running_vcpu
;
6361 #ifdef CONFIG_GUEST_PERF_EVENTS
6362 static unsigned int kvm_guest_state(void)
6364 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6367 if (!kvm_arch_pmi_in_guest(vcpu
))
6370 state
= PERF_GUEST_ACTIVE
;
6371 if (!kvm_arch_vcpu_in_kernel(vcpu
))
6372 state
|= PERF_GUEST_USER
;
6377 static unsigned long kvm_guest_get_ip(void)
6379 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6381 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6382 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu
)))
6385 return kvm_arch_vcpu_get_ip(vcpu
);
6388 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6389 .state
= kvm_guest_state
,
6390 .get_ip
= kvm_guest_get_ip
,
6391 .handle_intel_pt_intr
= NULL
,
6394 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler
)(void))
6396 kvm_guest_cbs
.handle_intel_pt_intr
= pt_intr_handler
;
6397 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6399 void kvm_unregister_perf_callbacks(void)
6401 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6405 int kvm_init(unsigned vcpu_size
, unsigned vcpu_align
, struct module
*module
)
6410 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6411 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE
, "kvm/cpu:online",
6412 kvm_online_cpu
, kvm_offline_cpu
);
6416 register_syscore_ops(&kvm_syscore_ops
);
6419 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6421 vcpu_align
= __alignof__(struct kvm_vcpu
);
6423 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
6425 offsetof(struct kvm_vcpu
, arch
),
6426 offsetofend(struct kvm_vcpu
, stats_id
)
6427 - offsetof(struct kvm_vcpu
, arch
),
6429 if (!kvm_vcpu_cache
) {
6431 goto err_vcpu_cache
;
6434 for_each_possible_cpu(cpu
) {
6435 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask
, cpu
),
6436 GFP_KERNEL
, cpu_to_node(cpu
))) {
6438 goto err_cpu_kick_mask
;
6442 r
= kvm_irqfd_init();
6446 r
= kvm_async_pf_init();
6450 kvm_chardev_ops
.owner
= module
;
6452 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
6453 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
6457 r
= kvm_vfio_ops_init();
6458 if (WARN_ON_ONCE(r
))
6461 kvm_gmem_init(module
);
6464 * Registration _must_ be the very last thing done, as this exposes
6465 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6467 r
= misc_register(&kvm_dev
);
6469 pr_err("kvm: misc device register failed\n");
6476 kvm_vfio_ops_exit();
6478 kvm_async_pf_deinit();
6483 for_each_possible_cpu(cpu
)
6484 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6485 kmem_cache_destroy(kvm_vcpu_cache
);
6487 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6488 unregister_syscore_ops(&kvm_syscore_ops
);
6489 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6493 EXPORT_SYMBOL_GPL(kvm_init
);
6500 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6501 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6502 * to KVM while the module is being stopped.
6504 misc_deregister(&kvm_dev
);
6506 debugfs_remove_recursive(kvm_debugfs_dir
);
6507 for_each_possible_cpu(cpu
)
6508 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6509 kmem_cache_destroy(kvm_vcpu_cache
);
6510 kvm_vfio_ops_exit();
6511 kvm_async_pf_deinit();
6512 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6513 unregister_syscore_ops(&kvm_syscore_ops
);
6514 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6518 EXPORT_SYMBOL_GPL(kvm_exit
);
6520 struct kvm_vm_worker_thread_context
{
6522 struct task_struct
*parent
;
6523 struct completion init_done
;
6524 kvm_vm_thread_fn_t thread_fn
;
6529 static int kvm_vm_worker_thread(void *context
)
6532 * The init_context is allocated on the stack of the parent thread, so
6533 * we have to locally copy anything that is needed beyond initialization
6535 struct kvm_vm_worker_thread_context
*init_context
= context
;
6536 struct task_struct
*parent
;
6537 struct kvm
*kvm
= init_context
->kvm
;
6538 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
6539 uintptr_t data
= init_context
->data
;
6542 err
= kthread_park(current
);
6543 /* kthread_park(current) is never supposed to return an error */
6548 err
= cgroup_attach_task_all(init_context
->parent
, current
);
6550 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6555 set_user_nice(current
, task_nice(init_context
->parent
));
6558 init_context
->err
= err
;
6559 complete(&init_context
->init_done
);
6560 init_context
= NULL
;
6565 /* Wait to be woken up by the spawner before proceeding. */
6568 if (!kthread_should_stop())
6569 err
= thread_fn(kvm
, data
);
6573 * Move kthread back to its original cgroup to prevent it lingering in
6574 * the cgroup of the VM process, after the latter finishes its
6577 * kthread_stop() waits on the 'exited' completion condition which is
6578 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6579 * kthread is removed from the cgroup in the cgroup_exit() which is
6580 * called after the exit_mm(). This causes the kthread_stop() to return
6581 * before the kthread actually quits the cgroup.
6584 parent
= rcu_dereference(current
->real_parent
);
6585 get_task_struct(parent
);
6587 cgroup_attach_task_all(parent
, current
);
6588 put_task_struct(parent
);
6593 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
6594 uintptr_t data
, const char *name
,
6595 struct task_struct
**thread_ptr
)
6597 struct kvm_vm_worker_thread_context init_context
= {};
6598 struct task_struct
*thread
;
6601 init_context
.kvm
= kvm
;
6602 init_context
.parent
= current
;
6603 init_context
.thread_fn
= thread_fn
;
6604 init_context
.data
= data
;
6605 init_completion(&init_context
.init_done
);
6607 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
6608 "%s-%d", name
, task_pid_nr(current
));
6610 return PTR_ERR(thread
);
6612 /* kthread_run is never supposed to return NULL */
6613 WARN_ON(thread
== NULL
);
6615 wait_for_completion(&init_context
.init_done
);
6617 if (!init_context
.err
)
6618 *thread_ptr
= thread
;
6620 return init_context
.err
;