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 long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 * For architectures that don't implement a compat infrastructure,
127 * adopt a double line of defense:
128 * - Prevent a compat task from opening /dev/kvm
129 * - If the open has been done by a 64bit task, and the KVM fd
130 * passed to a compat task, let the ioctls fail.
132 static long kvm_no_compat_ioctl(struct file
*file
, unsigned int ioctl
,
133 unsigned long arg
) { return -EINVAL
; }
135 static int kvm_no_compat_open(struct inode
*inode
, struct file
*file
)
137 return is_compat_task() ? -ENODEV
: 0;
139 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
140 .open = kvm_no_compat_open
142 static int hardware_enable_all(void);
143 static void hardware_disable_all(void);
145 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
147 #define KVM_EVENT_CREATE_VM 0
148 #define KVM_EVENT_DESTROY_VM 1
149 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
150 static unsigned long long kvm_createvm_count
;
151 static unsigned long long kvm_active_vms
;
153 static DEFINE_PER_CPU(cpumask_var_t
, cpu_kick_mask
);
155 __weak
void kvm_arch_guest_memory_reclaimed(struct kvm
*kvm
)
159 bool kvm_is_zone_device_page(struct page
*page
)
162 * The metadata used by is_zone_device_page() to determine whether or
163 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
164 * the device has been pinned, e.g. by get_user_pages(). WARN if the
165 * page_count() is zero to help detect bad usage of this helper.
167 if (WARN_ON_ONCE(!page_count(page
)))
170 return is_zone_device_page(page
);
174 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
175 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
176 * is likely incomplete, it has been compiled purely through people wanting to
177 * back guest with a certain type of memory and encountering issues.
179 struct page
*kvm_pfn_to_refcounted_page(kvm_pfn_t pfn
)
186 page
= pfn_to_page(pfn
);
187 if (!PageReserved(page
))
190 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
191 if (is_zero_pfn(pfn
))
195 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
196 * perspective they are "normal" pages, albeit with slightly different
199 if (kvm_is_zone_device_page(page
))
206 * Switches to specified vcpu, until a matching vcpu_put()
208 void vcpu_load(struct kvm_vcpu
*vcpu
)
212 __this_cpu_write(kvm_running_vcpu
, vcpu
);
213 preempt_notifier_register(&vcpu
->preempt_notifier
);
214 kvm_arch_vcpu_load(vcpu
, cpu
);
217 EXPORT_SYMBOL_GPL(vcpu_load
);
219 void vcpu_put(struct kvm_vcpu
*vcpu
)
222 kvm_arch_vcpu_put(vcpu
);
223 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
224 __this_cpu_write(kvm_running_vcpu
, NULL
);
227 EXPORT_SYMBOL_GPL(vcpu_put
);
229 /* TODO: merge with kvm_arch_vcpu_should_kick */
230 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
232 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
235 * We need to wait for the VCPU to reenable interrupts and get out of
236 * READING_SHADOW_PAGE_TABLES mode.
238 if (req
& KVM_REQUEST_WAIT
)
239 return mode
!= OUTSIDE_GUEST_MODE
;
242 * Need to kick a running VCPU, but otherwise there is nothing to do.
244 return mode
== IN_GUEST_MODE
;
247 static void ack_kick(void *_completed
)
251 static inline bool kvm_kick_many_cpus(struct cpumask
*cpus
, bool wait
)
253 if (cpumask_empty(cpus
))
256 smp_call_function_many(cpus
, ack_kick
, NULL
, wait
);
260 static void kvm_make_vcpu_request(struct kvm_vcpu
*vcpu
, unsigned int req
,
261 struct cpumask
*tmp
, int current_cpu
)
265 if (likely(!(req
& KVM_REQUEST_NO_ACTION
)))
266 __kvm_make_request(req
, vcpu
);
268 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
272 * Note, the vCPU could get migrated to a different pCPU at any point
273 * after kvm_request_needs_ipi(), which could result in sending an IPI
274 * to the previous pCPU. But, that's OK because the purpose of the IPI
275 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
276 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
277 * after this point is also OK, as the requirement is only that KVM wait
278 * for vCPUs that were reading SPTEs _before_ any changes were
279 * finalized. See kvm_vcpu_kick() for more details on handling requests.
281 if (kvm_request_needs_ipi(vcpu
, req
)) {
282 cpu
= READ_ONCE(vcpu
->cpu
);
283 if (cpu
!= -1 && cpu
!= current_cpu
)
284 __cpumask_set_cpu(cpu
, tmp
);
288 bool kvm_make_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
289 unsigned long *vcpu_bitmap
)
291 struct kvm_vcpu
*vcpu
;
292 struct cpumask
*cpus
;
298 cpus
= this_cpu_cpumask_var_ptr(cpu_kick_mask
);
301 for_each_set_bit(i
, vcpu_bitmap
, KVM_MAX_VCPUS
) {
302 vcpu
= kvm_get_vcpu(kvm
, i
);
305 kvm_make_vcpu_request(vcpu
, req
, cpus
, me
);
308 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
314 bool kvm_make_all_cpus_request_except(struct kvm
*kvm
, unsigned int req
,
315 struct kvm_vcpu
*except
)
317 struct kvm_vcpu
*vcpu
;
318 struct cpumask
*cpus
;
325 cpus
= this_cpu_cpumask_var_ptr(cpu_kick_mask
);
328 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
331 kvm_make_vcpu_request(vcpu
, req
, cpus
, me
);
334 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
340 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
342 return kvm_make_all_cpus_request_except(kvm
, req
, NULL
);
344 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request
);
346 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
348 ++kvm
->stat
.generic
.remote_tlb_flush_requests
;
351 * We want to publish modifications to the page tables before reading
352 * mode. Pairs with a memory barrier in arch-specific code.
353 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
354 * and smp_mb in walk_shadow_page_lockless_begin/end.
355 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
357 * There is already an smp_mb__after_atomic() before
358 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
361 if (!kvm_arch_flush_remote_tlbs(kvm
)
362 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
363 ++kvm
->stat
.generic
.remote_tlb_flush
;
365 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
367 void kvm_flush_remote_tlbs_range(struct kvm
*kvm
, gfn_t gfn
, u64 nr_pages
)
369 if (!kvm_arch_flush_remote_tlbs_range(kvm
, gfn
, nr_pages
))
373 * Fall back to a flushing entire TLBs if the architecture range-based
374 * TLB invalidation is unsupported or can't be performed for whatever
377 kvm_flush_remote_tlbs(kvm
);
380 void kvm_flush_remote_tlbs_memslot(struct kvm
*kvm
,
381 const struct kvm_memory_slot
*memslot
)
384 * All current use cases for flushing the TLBs for a specific memslot
385 * are related to dirty logging, and many do the TLB flush out of
386 * mmu_lock. The interaction between the various operations on memslot
387 * must be serialized by slots_locks to ensure the TLB flush from one
388 * operation is observed by any other operation on the same memslot.
390 lockdep_assert_held(&kvm
->slots_lock
);
391 kvm_flush_remote_tlbs_range(kvm
, memslot
->base_gfn
, memslot
->npages
);
394 static void kvm_flush_shadow_all(struct kvm
*kvm
)
396 kvm_arch_flush_shadow_all(kvm
);
397 kvm_arch_guest_memory_reclaimed(kvm
);
400 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
401 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache
*mc
,
404 gfp_flags
|= mc
->gfp_zero
;
407 return kmem_cache_alloc(mc
->kmem_cache
, gfp_flags
);
409 return (void *)__get_free_page(gfp_flags
);
412 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int capacity
, int min
)
414 gfp_t gfp
= mc
->gfp_custom
? mc
->gfp_custom
: GFP_KERNEL_ACCOUNT
;
417 if (mc
->nobjs
>= min
)
420 if (unlikely(!mc
->objects
)) {
421 if (WARN_ON_ONCE(!capacity
))
424 mc
->objects
= kvmalloc_array(sizeof(void *), capacity
, gfp
);
428 mc
->capacity
= capacity
;
431 /* It is illegal to request a different capacity across topups. */
432 if (WARN_ON_ONCE(mc
->capacity
!= capacity
))
435 while (mc
->nobjs
< mc
->capacity
) {
436 obj
= mmu_memory_cache_alloc_obj(mc
, gfp
);
438 return mc
->nobjs
>= min
? 0 : -ENOMEM
;
439 mc
->objects
[mc
->nobjs
++] = obj
;
444 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int min
)
446 return __kvm_mmu_topup_memory_cache(mc
, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
, min
);
449 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache
*mc
)
454 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
458 kmem_cache_free(mc
->kmem_cache
, mc
->objects
[--mc
->nobjs
]);
460 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
469 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
)
473 if (WARN_ON(!mc
->nobjs
))
474 p
= mmu_memory_cache_alloc_obj(mc
, GFP_ATOMIC
| __GFP_ACCOUNT
);
476 p
= mc
->objects
[--mc
->nobjs
];
482 static void kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
484 mutex_init(&vcpu
->mutex
);
489 #ifndef __KVM_HAVE_ARCH_WQP
490 rcuwait_init(&vcpu
->wait
);
492 kvm_async_pf_vcpu_init(vcpu
);
494 kvm_vcpu_set_in_spin_loop(vcpu
, false);
495 kvm_vcpu_set_dy_eligible(vcpu
, false);
496 vcpu
->preempted
= false;
498 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
499 vcpu
->last_used_slot
= NULL
;
501 /* Fill the stats id string for the vcpu */
502 snprintf(vcpu
->stats_id
, sizeof(vcpu
->stats_id
), "kvm-%d/vcpu-%d",
503 task_pid_nr(current
), id
);
506 static void kvm_vcpu_destroy(struct kvm_vcpu
*vcpu
)
508 kvm_arch_vcpu_destroy(vcpu
);
509 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
512 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
513 * the vcpu->pid pointer, and at destruction time all file descriptors
516 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
518 free_page((unsigned long)vcpu
->run
);
519 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
522 void kvm_destroy_vcpus(struct kvm
*kvm
)
525 struct kvm_vcpu
*vcpu
;
527 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
528 kvm_vcpu_destroy(vcpu
);
529 xa_erase(&kvm
->vcpu_array
, i
);
532 atomic_set(&kvm
->online_vcpus
, 0);
534 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus
);
536 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
537 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
539 return container_of(mn
, struct kvm
, mmu_notifier
);
542 typedef bool (*gfn_handler_t
)(struct kvm
*kvm
, struct kvm_gfn_range
*range
);
544 typedef void (*on_lock_fn_t
)(struct kvm
*kvm
);
546 struct kvm_mmu_notifier_range
{
548 * 64-bit addresses, as KVM notifiers can operate on host virtual
549 * addresses (unsigned long) and guest physical addresses (64-bit).
553 union kvm_mmu_notifier_arg arg
;
554 gfn_handler_t handler
;
555 on_lock_fn_t on_lock
;
561 * The inner-most helper returns a tuple containing the return value from the
562 * arch- and action-specific handler, plus a flag indicating whether or not at
563 * least one memslot was found, i.e. if the handler found guest memory.
565 * Note, most notifiers are averse to booleans, so even though KVM tracks the
566 * return from arch code as a bool, outer helpers will cast it to an int. :-(
568 typedef struct kvm_mmu_notifier_return
{
574 * Use a dedicated stub instead of NULL to indicate that there is no callback
575 * function/handler. The compiler technically can't guarantee that a real
576 * function will have a non-zero address, and so it will generate code to
577 * check for !NULL, whereas comparing against a stub will be elided at compile
578 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
580 static void kvm_null_fn(void)
584 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
586 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG
;
588 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
589 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
590 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
592 node = interval_tree_iter_next(node, start, last)) \
594 static __always_inline kvm_mn_ret_t __kvm_handle_hva_range(struct kvm *kvm,
595 const struct kvm_mmu_notifier_range
*range
)
597 struct kvm_mmu_notifier_return r
= {
599 .found_memslot
= false,
601 struct kvm_gfn_range gfn_range
;
602 struct kvm_memory_slot
*slot
;
603 struct kvm_memslots
*slots
;
606 if (WARN_ON_ONCE(range
->end
<= range
->start
))
609 /* A null handler is allowed if and only if on_lock() is provided. */
610 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range
->on_lock
) &&
611 IS_KVM_NULL_FN(range
->handler
)))
614 idx
= srcu_read_lock(&kvm
->srcu
);
616 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
617 struct interval_tree_node
*node
;
619 slots
= __kvm_memslots(kvm
, i
);
620 kvm_for_each_memslot_in_hva_range(node
, slots
,
621 range
->start
, range
->end
- 1) {
622 unsigned long hva_start
, hva_end
;
624 slot
= container_of(node
, struct kvm_memory_slot
, hva_node
[slots
->node_idx
]);
625 hva_start
= max_t(unsigned long, range
->start
, slot
->userspace_addr
);
626 hva_end
= min_t(unsigned long, range
->end
,
627 slot
->userspace_addr
+ (slot
->npages
<< PAGE_SHIFT
));
630 * To optimize for the likely case where the address
631 * range is covered by zero or one memslots, don't
632 * bother making these conditional (to avoid writes on
633 * the second or later invocation of the handler).
635 gfn_range
.arg
= range
->arg
;
636 gfn_range
.may_block
= range
->may_block
;
639 * {gfn(page) | page intersects with [hva_start, hva_end)} =
640 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
642 gfn_range
.start
= hva_to_gfn_memslot(hva_start
, slot
);
643 gfn_range
.end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, slot
);
644 gfn_range
.slot
= slot
;
646 if (!r
.found_memslot
) {
647 r
.found_memslot
= true;
649 if (!IS_KVM_NULL_FN(range
->on_lock
))
652 if (IS_KVM_NULL_FN(range
->handler
))
655 r
.ret
|= range
->handler(kvm
, &gfn_range
);
659 if (range
->flush_on_ret
&& r
.ret
)
660 kvm_flush_remote_tlbs(kvm
);
665 srcu_read_unlock(&kvm
->srcu
, idx
);
670 static __always_inline
int kvm_handle_hva_range(struct mmu_notifier
*mn
,
673 union kvm_mmu_notifier_arg arg
,
674 gfn_handler_t handler
)
676 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
677 const struct kvm_mmu_notifier_range range
= {
682 .on_lock
= (void *)kvm_null_fn
,
683 .flush_on_ret
= true,
687 return __kvm_handle_hva_range(kvm
, &range
).ret
;
690 static __always_inline
int kvm_handle_hva_range_no_flush(struct mmu_notifier
*mn
,
693 gfn_handler_t handler
)
695 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
696 const struct kvm_mmu_notifier_range range
= {
700 .on_lock
= (void *)kvm_null_fn
,
701 .flush_on_ret
= false,
705 return __kvm_handle_hva_range(kvm
, &range
).ret
;
708 static bool kvm_change_spte_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
711 * Skipping invalid memslots is correct if and only change_pte() is
712 * surrounded by invalidate_range_{start,end}(), which is currently
713 * guaranteed by the primary MMU. If that ever changes, KVM needs to
714 * unmap the memslot instead of skipping the memslot to ensure that KVM
715 * doesn't hold references to the old PFN.
717 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
719 if (range
->slot
->flags
& KVM_MEMSLOT_INVALID
)
722 return kvm_set_spte_gfn(kvm
, range
);
725 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
726 struct mm_struct
*mm
,
727 unsigned long address
,
730 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
731 const union kvm_mmu_notifier_arg arg
= { .pte
= pte
};
733 trace_kvm_set_spte_hva(address
);
736 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
737 * If mmu_invalidate_in_progress is zero, then no in-progress
738 * invalidations, including this one, found a relevant memslot at
739 * start(); rechecking memslots here is unnecessary. Note, a false
740 * positive (count elevated by a different invalidation) is sub-optimal
741 * but functionally ok.
743 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
744 if (!READ_ONCE(kvm
->mmu_invalidate_in_progress
))
747 kvm_handle_hva_range(mn
, address
, address
+ 1, arg
, kvm_change_spte_gfn
);
750 void kvm_mmu_invalidate_begin(struct kvm
*kvm
)
752 lockdep_assert_held_write(&kvm
->mmu_lock
);
754 * The count increase must become visible at unlock time as no
755 * spte can be established without taking the mmu_lock and
756 * count is also read inside the mmu_lock critical section.
758 kvm
->mmu_invalidate_in_progress
++;
760 if (likely(kvm
->mmu_invalidate_in_progress
== 1)) {
761 kvm
->mmu_invalidate_range_start
= INVALID_GPA
;
762 kvm
->mmu_invalidate_range_end
= INVALID_GPA
;
766 void kvm_mmu_invalidate_range_add(struct kvm
*kvm
, gfn_t start
, gfn_t end
)
768 lockdep_assert_held_write(&kvm
->mmu_lock
);
770 WARN_ON_ONCE(!kvm
->mmu_invalidate_in_progress
);
772 if (likely(kvm
->mmu_invalidate_range_start
== INVALID_GPA
)) {
773 kvm
->mmu_invalidate_range_start
= start
;
774 kvm
->mmu_invalidate_range_end
= end
;
777 * Fully tracking multiple concurrent ranges has diminishing
778 * returns. Keep things simple and just find the minimal range
779 * which includes the current and new ranges. As there won't be
780 * enough information to subtract a range after its invalidate
781 * completes, any ranges invalidated concurrently will
782 * accumulate and persist until all outstanding invalidates
785 kvm
->mmu_invalidate_range_start
=
786 min(kvm
->mmu_invalidate_range_start
, start
);
787 kvm
->mmu_invalidate_range_end
=
788 max(kvm
->mmu_invalidate_range_end
, end
);
792 bool kvm_mmu_unmap_gfn_range(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
794 kvm_mmu_invalidate_range_add(kvm
, range
->start
, range
->end
);
795 return kvm_unmap_gfn_range(kvm
, range
);
798 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
799 const struct mmu_notifier_range
*range
)
801 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
802 const struct kvm_mmu_notifier_range hva_range
= {
803 .start
= range
->start
,
805 .handler
= kvm_mmu_unmap_gfn_range
,
806 .on_lock
= kvm_mmu_invalidate_begin
,
807 .flush_on_ret
= true,
808 .may_block
= mmu_notifier_range_blockable(range
),
811 trace_kvm_unmap_hva_range(range
->start
, range
->end
);
814 * Prevent memslot modification between range_start() and range_end()
815 * so that conditionally locking provides the same result in both
816 * functions. Without that guarantee, the mmu_invalidate_in_progress
817 * adjustments will be imbalanced.
819 * Pairs with the decrement in range_end().
821 spin_lock(&kvm
->mn_invalidate_lock
);
822 kvm
->mn_active_invalidate_count
++;
823 spin_unlock(&kvm
->mn_invalidate_lock
);
826 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
827 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
828 * each cache's lock. There are relatively few caches in existence at
829 * any given time, and the caches themselves can check for hva overlap,
830 * i.e. don't need to rely on memslot overlap checks for performance.
831 * Because this runs without holding mmu_lock, the pfn caches must use
832 * mn_active_invalidate_count (see above) instead of
833 * mmu_invalidate_in_progress.
835 gfn_to_pfn_cache_invalidate_start(kvm
, range
->start
, range
->end
,
836 hva_range
.may_block
);
839 * If one or more memslots were found and thus zapped, notify arch code
840 * that guest memory has been reclaimed. This needs to be done *after*
841 * dropping mmu_lock, as x86's reclaim path is slooooow.
843 if (__kvm_handle_hva_range(kvm
, &hva_range
).found_memslot
)
844 kvm_arch_guest_memory_reclaimed(kvm
);
849 void kvm_mmu_invalidate_end(struct kvm
*kvm
)
851 lockdep_assert_held_write(&kvm
->mmu_lock
);
854 * This sequence increase will notify the kvm page fault that
855 * the page that is going to be mapped in the spte could have
858 kvm
->mmu_invalidate_seq
++;
861 * The above sequence increase must be visible before the
862 * below count decrease, which is ensured by the smp_wmb above
863 * in conjunction with the smp_rmb in mmu_invalidate_retry().
865 kvm
->mmu_invalidate_in_progress
--;
866 KVM_BUG_ON(kvm
->mmu_invalidate_in_progress
< 0, kvm
);
869 * Assert that at least one range was added between start() and end().
870 * Not adding a range isn't fatal, but it is a KVM bug.
872 WARN_ON_ONCE(kvm
->mmu_invalidate_range_start
== INVALID_GPA
);
875 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
876 const struct mmu_notifier_range
*range
)
878 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
879 const struct kvm_mmu_notifier_range hva_range
= {
880 .start
= range
->start
,
882 .handler
= (void *)kvm_null_fn
,
883 .on_lock
= kvm_mmu_invalidate_end
,
884 .flush_on_ret
= false,
885 .may_block
= mmu_notifier_range_blockable(range
),
889 __kvm_handle_hva_range(kvm
, &hva_range
);
891 /* Pairs with the increment in range_start(). */
892 spin_lock(&kvm
->mn_invalidate_lock
);
893 wake
= (--kvm
->mn_active_invalidate_count
== 0);
894 spin_unlock(&kvm
->mn_invalidate_lock
);
897 * There can only be one waiter, since the wait happens under
901 rcuwait_wake_up(&kvm
->mn_memslots_update_rcuwait
);
904 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
905 struct mm_struct
*mm
,
909 trace_kvm_age_hva(start
, end
);
911 return kvm_handle_hva_range(mn
, start
, end
, KVM_MMU_NOTIFIER_NO_ARG
,
915 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
916 struct mm_struct
*mm
,
920 trace_kvm_age_hva(start
, end
);
923 * Even though we do not flush TLB, this will still adversely
924 * affect performance on pre-Haswell Intel EPT, where there is
925 * no EPT Access Bit to clear so that we have to tear down EPT
926 * tables instead. If we find this unacceptable, we can always
927 * add a parameter to kvm_age_hva so that it effectively doesn't
928 * do anything on clear_young.
930 * Also note that currently we never issue secondary TLB flushes
931 * from clear_young, leaving this job up to the regular system
932 * cadence. If we find this inaccurate, we might come up with a
933 * more sophisticated heuristic later.
935 return kvm_handle_hva_range_no_flush(mn
, start
, end
, kvm_age_gfn
);
938 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
939 struct mm_struct
*mm
,
940 unsigned long address
)
942 trace_kvm_test_age_hva(address
);
944 return kvm_handle_hva_range_no_flush(mn
, address
, address
+ 1,
948 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
949 struct mm_struct
*mm
)
951 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
954 idx
= srcu_read_lock(&kvm
->srcu
);
955 kvm_flush_shadow_all(kvm
);
956 srcu_read_unlock(&kvm
->srcu
, idx
);
959 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
960 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
961 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
962 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
963 .clear_young
= kvm_mmu_notifier_clear_young
,
964 .test_young
= kvm_mmu_notifier_test_young
,
965 .change_pte
= kvm_mmu_notifier_change_pte
,
966 .release
= kvm_mmu_notifier_release
,
969 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
971 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
972 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
975 #else /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */
977 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
982 #endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */
984 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
985 static int kvm_pm_notifier_call(struct notifier_block
*bl
,
989 struct kvm
*kvm
= container_of(bl
, struct kvm
, pm_notifier
);
991 return kvm_arch_pm_notifier(kvm
, state
);
994 static void kvm_init_pm_notifier(struct kvm
*kvm
)
996 kvm
->pm_notifier
.notifier_call
= kvm_pm_notifier_call
;
997 /* Suspend KVM before we suspend ftrace, RCU, etc. */
998 kvm
->pm_notifier
.priority
= INT_MAX
;
999 register_pm_notifier(&kvm
->pm_notifier
);
1002 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
1004 unregister_pm_notifier(&kvm
->pm_notifier
);
1006 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
1007 static void kvm_init_pm_notifier(struct kvm
*kvm
)
1011 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
1014 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
1016 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1018 if (!memslot
->dirty_bitmap
)
1021 kvfree(memslot
->dirty_bitmap
);
1022 memslot
->dirty_bitmap
= NULL
;
1025 /* This does not remove the slot from struct kvm_memslots data structures */
1026 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
1028 if (slot
->flags
& KVM_MEM_GUEST_MEMFD
)
1029 kvm_gmem_unbind(slot
);
1031 kvm_destroy_dirty_bitmap(slot
);
1033 kvm_arch_free_memslot(kvm
, slot
);
1038 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
1040 struct hlist_node
*idnode
;
1041 struct kvm_memory_slot
*memslot
;
1045 * The same memslot objects live in both active and inactive sets,
1046 * arbitrarily free using index '1' so the second invocation of this
1047 * function isn't operating over a structure with dangling pointers
1048 * (even though this function isn't actually touching them).
1050 if (!slots
->node_idx
)
1053 hash_for_each_safe(slots
->id_hash
, bkt
, idnode
, memslot
, id_node
[1])
1054 kvm_free_memslot(kvm
, memslot
);
1057 static umode_t
kvm_stats_debugfs_mode(const struct _kvm_stats_desc
*pdesc
)
1059 switch (pdesc
->desc
.flags
& KVM_STATS_TYPE_MASK
) {
1060 case KVM_STATS_TYPE_INSTANT
:
1062 case KVM_STATS_TYPE_CUMULATIVE
:
1063 case KVM_STATS_TYPE_PEAK
:
1070 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
1073 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1074 kvm_vcpu_stats_header
.num_desc
;
1076 if (IS_ERR(kvm
->debugfs_dentry
))
1079 debugfs_remove_recursive(kvm
->debugfs_dentry
);
1081 if (kvm
->debugfs_stat_data
) {
1082 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
1083 kfree(kvm
->debugfs_stat_data
[i
]);
1084 kfree(kvm
->debugfs_stat_data
);
1088 static int kvm_create_vm_debugfs(struct kvm
*kvm
, const char *fdname
)
1090 static DEFINE_MUTEX(kvm_debugfs_lock
);
1091 struct dentry
*dent
;
1092 char dir_name
[ITOA_MAX_LEN
* 2];
1093 struct kvm_stat_data
*stat_data
;
1094 const struct _kvm_stats_desc
*pdesc
;
1095 int i
, ret
= -ENOMEM
;
1096 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1097 kvm_vcpu_stats_header
.num_desc
;
1099 if (!debugfs_initialized())
1102 snprintf(dir_name
, sizeof(dir_name
), "%d-%s", task_pid_nr(current
), fdname
);
1103 mutex_lock(&kvm_debugfs_lock
);
1104 dent
= debugfs_lookup(dir_name
, kvm_debugfs_dir
);
1106 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name
);
1108 mutex_unlock(&kvm_debugfs_lock
);
1111 dent
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
1112 mutex_unlock(&kvm_debugfs_lock
);
1116 kvm
->debugfs_dentry
= dent
;
1117 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
1118 sizeof(*kvm
->debugfs_stat_data
),
1119 GFP_KERNEL_ACCOUNT
);
1120 if (!kvm
->debugfs_stat_data
)
1123 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
1124 pdesc
= &kvm_vm_stats_desc
[i
];
1125 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1129 stat_data
->kvm
= kvm
;
1130 stat_data
->desc
= pdesc
;
1131 stat_data
->kind
= KVM_STAT_VM
;
1132 kvm
->debugfs_stat_data
[i
] = stat_data
;
1133 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1134 kvm
->debugfs_dentry
, stat_data
,
1138 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
1139 pdesc
= &kvm_vcpu_stats_desc
[i
];
1140 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1144 stat_data
->kvm
= kvm
;
1145 stat_data
->desc
= pdesc
;
1146 stat_data
->kind
= KVM_STAT_VCPU
;
1147 kvm
->debugfs_stat_data
[i
+ kvm_vm_stats_header
.num_desc
] = stat_data
;
1148 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1149 kvm
->debugfs_dentry
, stat_data
,
1153 ret
= kvm_arch_create_vm_debugfs(kvm
);
1159 kvm_destroy_vm_debugfs(kvm
);
1164 * Called after the VM is otherwise initialized, but just before adding it to
1167 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
1173 * Called just after removing the VM from the vm_list, but before doing any
1174 * other destruction.
1176 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
1181 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1182 * be setup already, so we can create arch-specific debugfs entries under it.
1183 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1184 * a per-arch destroy interface is not needed.
1186 int __weak
kvm_arch_create_vm_debugfs(struct kvm
*kvm
)
1191 static struct kvm
*kvm_create_vm(unsigned long type
, const char *fdname
)
1193 struct kvm
*kvm
= kvm_arch_alloc_vm();
1194 struct kvm_memslots
*slots
;
1199 return ERR_PTR(-ENOMEM
);
1201 KVM_MMU_LOCK_INIT(kvm
);
1202 mmgrab(current
->mm
);
1203 kvm
->mm
= current
->mm
;
1204 kvm_eventfd_init(kvm
);
1205 mutex_init(&kvm
->lock
);
1206 mutex_init(&kvm
->irq_lock
);
1207 mutex_init(&kvm
->slots_lock
);
1208 mutex_init(&kvm
->slots_arch_lock
);
1209 spin_lock_init(&kvm
->mn_invalidate_lock
);
1210 rcuwait_init(&kvm
->mn_memslots_update_rcuwait
);
1211 xa_init(&kvm
->vcpu_array
);
1212 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1213 xa_init(&kvm
->mem_attr_array
);
1216 INIT_LIST_HEAD(&kvm
->gpc_list
);
1217 spin_lock_init(&kvm
->gpc_lock
);
1219 INIT_LIST_HEAD(&kvm
->devices
);
1220 kvm
->max_vcpus
= KVM_MAX_VCPUS
;
1222 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
1225 * Force subsequent debugfs file creations to fail if the VM directory
1226 * is not created (by kvm_create_vm_debugfs()).
1228 kvm
->debugfs_dentry
= ERR_PTR(-ENOENT
);
1230 snprintf(kvm
->stats_id
, sizeof(kvm
->stats_id
), "kvm-%d",
1231 task_pid_nr(current
));
1233 if (init_srcu_struct(&kvm
->srcu
))
1234 goto out_err_no_srcu
;
1235 if (init_srcu_struct(&kvm
->irq_srcu
))
1236 goto out_err_no_irq_srcu
;
1238 refcount_set(&kvm
->users_count
, 1);
1239 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
1240 for (j
= 0; j
< 2; j
++) {
1241 slots
= &kvm
->__memslots
[i
][j
];
1243 atomic_long_set(&slots
->last_used_slot
, (unsigned long)NULL
);
1244 slots
->hva_tree
= RB_ROOT_CACHED
;
1245 slots
->gfn_tree
= RB_ROOT
;
1246 hash_init(slots
->id_hash
);
1247 slots
->node_idx
= j
;
1249 /* Generations must be different for each address space. */
1250 slots
->generation
= i
;
1253 rcu_assign_pointer(kvm
->memslots
[i
], &kvm
->__memslots
[i
][0]);
1256 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1257 rcu_assign_pointer(kvm
->buses
[i
],
1258 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
1260 goto out_err_no_arch_destroy_vm
;
1263 r
= kvm_arch_init_vm(kvm
, type
);
1265 goto out_err_no_arch_destroy_vm
;
1267 r
= hardware_enable_all();
1269 goto out_err_no_disable
;
1271 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1272 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
1275 r
= kvm_init_mmu_notifier(kvm
);
1277 goto out_err_no_mmu_notifier
;
1279 r
= kvm_coalesced_mmio_init(kvm
);
1281 goto out_no_coalesced_mmio
;
1283 r
= kvm_create_vm_debugfs(kvm
, fdname
);
1285 goto out_err_no_debugfs
;
1287 r
= kvm_arch_post_init_vm(kvm
);
1291 mutex_lock(&kvm_lock
);
1292 list_add(&kvm
->vm_list
, &vm_list
);
1293 mutex_unlock(&kvm_lock
);
1295 preempt_notifier_inc();
1296 kvm_init_pm_notifier(kvm
);
1301 kvm_destroy_vm_debugfs(kvm
);
1303 kvm_coalesced_mmio_free(kvm
);
1304 out_no_coalesced_mmio
:
1305 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1306 if (kvm
->mmu_notifier
.ops
)
1307 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
1309 out_err_no_mmu_notifier
:
1310 hardware_disable_all();
1312 kvm_arch_destroy_vm(kvm
);
1313 out_err_no_arch_destroy_vm
:
1314 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
1315 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
1316 kfree(kvm_get_bus(kvm
, i
));
1317 cleanup_srcu_struct(&kvm
->irq_srcu
);
1318 out_err_no_irq_srcu
:
1319 cleanup_srcu_struct(&kvm
->srcu
);
1321 kvm_arch_free_vm(kvm
);
1322 mmdrop(current
->mm
);
1326 static void kvm_destroy_devices(struct kvm
*kvm
)
1328 struct kvm_device
*dev
, *tmp
;
1331 * We do not need to take the kvm->lock here, because nobody else
1332 * has a reference to the struct kvm at this point and therefore
1333 * cannot access the devices list anyhow.
1335 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
1336 list_del(&dev
->vm_node
);
1337 dev
->ops
->destroy(dev
);
1341 static void kvm_destroy_vm(struct kvm
*kvm
)
1344 struct mm_struct
*mm
= kvm
->mm
;
1346 kvm_destroy_pm_notifier(kvm
);
1347 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
1348 kvm_destroy_vm_debugfs(kvm
);
1349 kvm_arch_sync_events(kvm
);
1350 mutex_lock(&kvm_lock
);
1351 list_del(&kvm
->vm_list
);
1352 mutex_unlock(&kvm_lock
);
1353 kvm_arch_pre_destroy_vm(kvm
);
1355 kvm_free_irq_routing(kvm
);
1356 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1357 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
1360 kvm_io_bus_destroy(bus
);
1361 kvm
->buses
[i
] = NULL
;
1363 kvm_coalesced_mmio_free(kvm
);
1364 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1365 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1367 * At this point, pending calls to invalidate_range_start()
1368 * have completed but no more MMU notifiers will run, so
1369 * mn_active_invalidate_count may remain unbalanced.
1370 * No threads can be waiting in kvm_swap_active_memslots() as the
1371 * last reference on KVM has been dropped, but freeing
1372 * memslots would deadlock without this manual intervention.
1374 * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU
1375 * notifier between a start() and end(), then there shouldn't be any
1376 * in-progress invalidations.
1378 WARN_ON(rcuwait_active(&kvm
->mn_memslots_update_rcuwait
));
1379 if (kvm
->mn_active_invalidate_count
)
1380 kvm
->mn_active_invalidate_count
= 0;
1382 WARN_ON(kvm
->mmu_invalidate_in_progress
);
1384 kvm_flush_shadow_all(kvm
);
1386 kvm_arch_destroy_vm(kvm
);
1387 kvm_destroy_devices(kvm
);
1388 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
1389 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][0]);
1390 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][1]);
1392 cleanup_srcu_struct(&kvm
->irq_srcu
);
1393 cleanup_srcu_struct(&kvm
->srcu
);
1394 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1395 xa_destroy(&kvm
->mem_attr_array
);
1397 kvm_arch_free_vm(kvm
);
1398 preempt_notifier_dec();
1399 hardware_disable_all();
1403 void kvm_get_kvm(struct kvm
*kvm
)
1405 refcount_inc(&kvm
->users_count
);
1407 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1410 * Make sure the vm is not during destruction, which is a safe version of
1411 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1413 bool kvm_get_kvm_safe(struct kvm
*kvm
)
1415 return refcount_inc_not_zero(&kvm
->users_count
);
1417 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe
);
1419 void kvm_put_kvm(struct kvm
*kvm
)
1421 if (refcount_dec_and_test(&kvm
->users_count
))
1422 kvm_destroy_vm(kvm
);
1424 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1427 * Used to put a reference that was taken on behalf of an object associated
1428 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1429 * of the new file descriptor fails and the reference cannot be transferred to
1430 * its final owner. In such cases, the caller is still actively using @kvm and
1431 * will fail miserably if the refcount unexpectedly hits zero.
1433 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
1435 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
1437 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
1439 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1441 struct kvm
*kvm
= filp
->private_data
;
1443 kvm_irqfd_release(kvm
);
1450 * Allocation size is twice as large as the actual dirty bitmap size.
1451 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1453 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1455 unsigned long dirty_bytes
= kvm_dirty_bitmap_bytes(memslot
);
1457 memslot
->dirty_bitmap
= __vcalloc(2, dirty_bytes
, GFP_KERNEL_ACCOUNT
);
1458 if (!memslot
->dirty_bitmap
)
1464 static struct kvm_memslots
*kvm_get_inactive_memslots(struct kvm
*kvm
, int as_id
)
1466 struct kvm_memslots
*active
= __kvm_memslots(kvm
, as_id
);
1467 int node_idx_inactive
= active
->node_idx
^ 1;
1469 return &kvm
->__memslots
[as_id
][node_idx_inactive
];
1473 * Helper to get the address space ID when one of memslot pointers may be NULL.
1474 * This also serves as a sanity that at least one of the pointers is non-NULL,
1475 * and that their address space IDs don't diverge.
1477 static int kvm_memslots_get_as_id(struct kvm_memory_slot
*a
,
1478 struct kvm_memory_slot
*b
)
1480 if (WARN_ON_ONCE(!a
&& !b
))
1488 WARN_ON_ONCE(a
->as_id
!= b
->as_id
);
1492 static void kvm_insert_gfn_node(struct kvm_memslots
*slots
,
1493 struct kvm_memory_slot
*slot
)
1495 struct rb_root
*gfn_tree
= &slots
->gfn_tree
;
1496 struct rb_node
**node
, *parent
;
1497 int idx
= slots
->node_idx
;
1500 for (node
= &gfn_tree
->rb_node
; *node
; ) {
1501 struct kvm_memory_slot
*tmp
;
1503 tmp
= container_of(*node
, struct kvm_memory_slot
, gfn_node
[idx
]);
1505 if (slot
->base_gfn
< tmp
->base_gfn
)
1506 node
= &(*node
)->rb_left
;
1507 else if (slot
->base_gfn
> tmp
->base_gfn
)
1508 node
= &(*node
)->rb_right
;
1513 rb_link_node(&slot
->gfn_node
[idx
], parent
, node
);
1514 rb_insert_color(&slot
->gfn_node
[idx
], gfn_tree
);
1517 static void kvm_erase_gfn_node(struct kvm_memslots
*slots
,
1518 struct kvm_memory_slot
*slot
)
1520 rb_erase(&slot
->gfn_node
[slots
->node_idx
], &slots
->gfn_tree
);
1523 static void kvm_replace_gfn_node(struct kvm_memslots
*slots
,
1524 struct kvm_memory_slot
*old
,
1525 struct kvm_memory_slot
*new)
1527 int idx
= slots
->node_idx
;
1529 WARN_ON_ONCE(old
->base_gfn
!= new->base_gfn
);
1531 rb_replace_node(&old
->gfn_node
[idx
], &new->gfn_node
[idx
],
1536 * Replace @old with @new in the inactive memslots.
1538 * With NULL @old this simply adds @new.
1539 * With NULL @new this simply removes @old.
1541 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1544 static void kvm_replace_memslot(struct kvm
*kvm
,
1545 struct kvm_memory_slot
*old
,
1546 struct kvm_memory_slot
*new)
1548 int as_id
= kvm_memslots_get_as_id(old
, new);
1549 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1550 int idx
= slots
->node_idx
;
1553 hash_del(&old
->id_node
[idx
]);
1554 interval_tree_remove(&old
->hva_node
[idx
], &slots
->hva_tree
);
1556 if ((long)old
== atomic_long_read(&slots
->last_used_slot
))
1557 atomic_long_set(&slots
->last_used_slot
, (long)new);
1560 kvm_erase_gfn_node(slots
, old
);
1566 * Initialize @new's hva range. Do this even when replacing an @old
1567 * slot, kvm_copy_memslot() deliberately does not touch node data.
1569 new->hva_node
[idx
].start
= new->userspace_addr
;
1570 new->hva_node
[idx
].last
= new->userspace_addr
+
1571 (new->npages
<< PAGE_SHIFT
) - 1;
1574 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1575 * hva_node needs to be swapped with remove+insert even though hva can't
1576 * change when replacing an existing slot.
1578 hash_add(slots
->id_hash
, &new->id_node
[idx
], new->id
);
1579 interval_tree_insert(&new->hva_node
[idx
], &slots
->hva_tree
);
1582 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1583 * switch the node in the gfn tree instead of removing the old and
1584 * inserting the new as two separate operations. Replacement is a
1585 * single O(1) operation versus two O(log(n)) operations for
1588 if (old
&& old
->base_gfn
== new->base_gfn
) {
1589 kvm_replace_gfn_node(slots
, old
, new);
1592 kvm_erase_gfn_node(slots
, old
);
1593 kvm_insert_gfn_node(slots
, new);
1598 * Flags that do not access any of the extra space of struct
1599 * kvm_userspace_memory_region2. KVM_SET_USER_MEMORY_REGION_V1_FLAGS
1600 * only allows these.
1602 #define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \
1603 (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY)
1605 static int check_memory_region_flags(struct kvm
*kvm
,
1606 const struct kvm_userspace_memory_region2
*mem
)
1608 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1610 if (kvm_arch_has_private_mem(kvm
))
1611 valid_flags
|= KVM_MEM_GUEST_MEMFD
;
1613 /* Dirty logging private memory is not currently supported. */
1614 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
)
1615 valid_flags
&= ~KVM_MEM_LOG_DIRTY_PAGES
;
1617 #ifdef __KVM_HAVE_READONLY_MEM
1618 valid_flags
|= KVM_MEM_READONLY
;
1621 if (mem
->flags
& ~valid_flags
)
1627 static void kvm_swap_active_memslots(struct kvm
*kvm
, int as_id
)
1629 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1631 /* Grab the generation from the activate memslots. */
1632 u64 gen
= __kvm_memslots(kvm
, as_id
)->generation
;
1634 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1635 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1638 * Do not store the new memslots while there are invalidations in
1639 * progress, otherwise the locking in invalidate_range_start and
1640 * invalidate_range_end will be unbalanced.
1642 spin_lock(&kvm
->mn_invalidate_lock
);
1643 prepare_to_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1644 while (kvm
->mn_active_invalidate_count
) {
1645 set_current_state(TASK_UNINTERRUPTIBLE
);
1646 spin_unlock(&kvm
->mn_invalidate_lock
);
1648 spin_lock(&kvm
->mn_invalidate_lock
);
1650 finish_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1651 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1652 spin_unlock(&kvm
->mn_invalidate_lock
);
1655 * Acquired in kvm_set_memslot. Must be released before synchronize
1656 * SRCU below in order to avoid deadlock with another thread
1657 * acquiring the slots_arch_lock in an srcu critical section.
1659 mutex_unlock(&kvm
->slots_arch_lock
);
1661 synchronize_srcu_expedited(&kvm
->srcu
);
1664 * Increment the new memslot generation a second time, dropping the
1665 * update in-progress flag and incrementing the generation based on
1666 * the number of address spaces. This provides a unique and easily
1667 * identifiable generation number while the memslots are in flux.
1669 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1672 * Generations must be unique even across address spaces. We do not need
1673 * a global counter for that, instead the generation space is evenly split
1674 * across address spaces. For example, with two address spaces, address
1675 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1676 * use generations 1, 3, 5, ...
1678 gen
+= kvm_arch_nr_memslot_as_ids(kvm
);
1680 kvm_arch_memslots_updated(kvm
, gen
);
1682 slots
->generation
= gen
;
1685 static int kvm_prepare_memory_region(struct kvm
*kvm
,
1686 const struct kvm_memory_slot
*old
,
1687 struct kvm_memory_slot
*new,
1688 enum kvm_mr_change change
)
1693 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1694 * will be freed on "commit". If logging is enabled in both old and
1695 * new, reuse the existing bitmap. If logging is enabled only in the
1696 * new and KVM isn't using a ring buffer, allocate and initialize a
1699 if (change
!= KVM_MR_DELETE
) {
1700 if (!(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1701 new->dirty_bitmap
= NULL
;
1702 else if (old
&& old
->dirty_bitmap
)
1703 new->dirty_bitmap
= old
->dirty_bitmap
;
1704 else if (kvm_use_dirty_bitmap(kvm
)) {
1705 r
= kvm_alloc_dirty_bitmap(new);
1709 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1710 bitmap_set(new->dirty_bitmap
, 0, new->npages
);
1714 r
= kvm_arch_prepare_memory_region(kvm
, old
, new, change
);
1716 /* Free the bitmap on failure if it was allocated above. */
1717 if (r
&& new && new->dirty_bitmap
&& (!old
|| !old
->dirty_bitmap
))
1718 kvm_destroy_dirty_bitmap(new);
1723 static void kvm_commit_memory_region(struct kvm
*kvm
,
1724 struct kvm_memory_slot
*old
,
1725 const struct kvm_memory_slot
*new,
1726 enum kvm_mr_change change
)
1728 int old_flags
= old
? old
->flags
: 0;
1729 int new_flags
= new ? new->flags
: 0;
1731 * Update the total number of memslot pages before calling the arch
1732 * hook so that architectures can consume the result directly.
1734 if (change
== KVM_MR_DELETE
)
1735 kvm
->nr_memslot_pages
-= old
->npages
;
1736 else if (change
== KVM_MR_CREATE
)
1737 kvm
->nr_memslot_pages
+= new->npages
;
1739 if ((old_flags
^ new_flags
) & KVM_MEM_LOG_DIRTY_PAGES
) {
1740 int change
= (new_flags
& KVM_MEM_LOG_DIRTY_PAGES
) ? 1 : -1;
1741 atomic_set(&kvm
->nr_memslots_dirty_logging
,
1742 atomic_read(&kvm
->nr_memslots_dirty_logging
) + change
);
1745 kvm_arch_commit_memory_region(kvm
, old
, new, change
);
1749 /* Nothing more to do. */
1752 /* Free the old memslot and all its metadata. */
1753 kvm_free_memslot(kvm
, old
);
1756 case KVM_MR_FLAGS_ONLY
:
1758 * Free the dirty bitmap as needed; the below check encompasses
1759 * both the flags and whether a ring buffer is being used)
1761 if (old
->dirty_bitmap
&& !new->dirty_bitmap
)
1762 kvm_destroy_dirty_bitmap(old
);
1765 * The final quirk. Free the detached, old slot, but only its
1766 * memory, not any metadata. Metadata, including arch specific
1767 * data, may be reused by @new.
1777 * Activate @new, which must be installed in the inactive slots by the caller,
1778 * by swapping the active slots and then propagating @new to @old once @old is
1779 * unreachable and can be safely modified.
1781 * With NULL @old this simply adds @new to @active (while swapping the sets).
1782 * With NULL @new this simply removes @old from @active and frees it
1783 * (while also swapping the sets).
1785 static void kvm_activate_memslot(struct kvm
*kvm
,
1786 struct kvm_memory_slot
*old
,
1787 struct kvm_memory_slot
*new)
1789 int as_id
= kvm_memslots_get_as_id(old
, new);
1791 kvm_swap_active_memslots(kvm
, as_id
);
1793 /* Propagate the new memslot to the now inactive memslots. */
1794 kvm_replace_memslot(kvm
, old
, new);
1797 static void kvm_copy_memslot(struct kvm_memory_slot
*dest
,
1798 const struct kvm_memory_slot
*src
)
1800 dest
->base_gfn
= src
->base_gfn
;
1801 dest
->npages
= src
->npages
;
1802 dest
->dirty_bitmap
= src
->dirty_bitmap
;
1803 dest
->arch
= src
->arch
;
1804 dest
->userspace_addr
= src
->userspace_addr
;
1805 dest
->flags
= src
->flags
;
1807 dest
->as_id
= src
->as_id
;
1810 static void kvm_invalidate_memslot(struct kvm
*kvm
,
1811 struct kvm_memory_slot
*old
,
1812 struct kvm_memory_slot
*invalid_slot
)
1815 * Mark the current slot INVALID. As with all memslot modifications,
1816 * this must be done on an unreachable slot to avoid modifying the
1817 * current slot in the active tree.
1819 kvm_copy_memslot(invalid_slot
, old
);
1820 invalid_slot
->flags
|= KVM_MEMSLOT_INVALID
;
1821 kvm_replace_memslot(kvm
, old
, invalid_slot
);
1824 * Activate the slot that is now marked INVALID, but don't propagate
1825 * the slot to the now inactive slots. The slot is either going to be
1826 * deleted or recreated as a new slot.
1828 kvm_swap_active_memslots(kvm
, old
->as_id
);
1831 * From this point no new shadow pages pointing to a deleted, or moved,
1832 * memslot will be created. Validation of sp->gfn happens in:
1833 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1834 * - kvm_is_visible_gfn (mmu_check_root)
1836 kvm_arch_flush_shadow_memslot(kvm
, old
);
1837 kvm_arch_guest_memory_reclaimed(kvm
);
1839 /* Was released by kvm_swap_active_memslots(), reacquire. */
1840 mutex_lock(&kvm
->slots_arch_lock
);
1843 * Copy the arch-specific field of the newly-installed slot back to the
1844 * old slot as the arch data could have changed between releasing
1845 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1846 * above. Writers are required to retrieve memslots *after* acquiring
1847 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1849 old
->arch
= invalid_slot
->arch
;
1852 static void kvm_create_memslot(struct kvm
*kvm
,
1853 struct kvm_memory_slot
*new)
1855 /* Add the new memslot to the inactive set and activate. */
1856 kvm_replace_memslot(kvm
, NULL
, new);
1857 kvm_activate_memslot(kvm
, NULL
, new);
1860 static void kvm_delete_memslot(struct kvm
*kvm
,
1861 struct kvm_memory_slot
*old
,
1862 struct kvm_memory_slot
*invalid_slot
)
1865 * Remove the old memslot (in the inactive memslots) by passing NULL as
1866 * the "new" slot, and for the invalid version in the active slots.
1868 kvm_replace_memslot(kvm
, old
, NULL
);
1869 kvm_activate_memslot(kvm
, invalid_slot
, NULL
);
1872 static void kvm_move_memslot(struct kvm
*kvm
,
1873 struct kvm_memory_slot
*old
,
1874 struct kvm_memory_slot
*new,
1875 struct kvm_memory_slot
*invalid_slot
)
1878 * Replace the old memslot in the inactive slots, and then swap slots
1879 * and replace the current INVALID with the new as well.
1881 kvm_replace_memslot(kvm
, old
, new);
1882 kvm_activate_memslot(kvm
, invalid_slot
, new);
1885 static void kvm_update_flags_memslot(struct kvm
*kvm
,
1886 struct kvm_memory_slot
*old
,
1887 struct kvm_memory_slot
*new)
1890 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1891 * an intermediate step. Instead, the old memslot is simply replaced
1892 * with a new, updated copy in both memslot sets.
1894 kvm_replace_memslot(kvm
, old
, new);
1895 kvm_activate_memslot(kvm
, old
, new);
1898 static int kvm_set_memslot(struct kvm
*kvm
,
1899 struct kvm_memory_slot
*old
,
1900 struct kvm_memory_slot
*new,
1901 enum kvm_mr_change change
)
1903 struct kvm_memory_slot
*invalid_slot
;
1907 * Released in kvm_swap_active_memslots().
1909 * Must be held from before the current memslots are copied until after
1910 * the new memslots are installed with rcu_assign_pointer, then
1911 * released before the synchronize srcu in kvm_swap_active_memslots().
1913 * When modifying memslots outside of the slots_lock, must be held
1914 * before reading the pointer to the current memslots until after all
1915 * changes to those memslots are complete.
1917 * These rules ensure that installing new memslots does not lose
1918 * changes made to the previous memslots.
1920 mutex_lock(&kvm
->slots_arch_lock
);
1923 * Invalidate the old slot if it's being deleted or moved. This is
1924 * done prior to actually deleting/moving the memslot to allow vCPUs to
1925 * continue running by ensuring there are no mappings or shadow pages
1926 * for the memslot when it is deleted/moved. Without pre-invalidation
1927 * (and without a lock), a window would exist between effecting the
1928 * delete/move and committing the changes in arch code where KVM or a
1929 * guest could access a non-existent memslot.
1931 * Modifications are done on a temporary, unreachable slot. The old
1932 * slot needs to be preserved in case a later step fails and the
1933 * invalidation needs to be reverted.
1935 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1936 invalid_slot
= kzalloc(sizeof(*invalid_slot
), GFP_KERNEL_ACCOUNT
);
1937 if (!invalid_slot
) {
1938 mutex_unlock(&kvm
->slots_arch_lock
);
1941 kvm_invalidate_memslot(kvm
, old
, invalid_slot
);
1944 r
= kvm_prepare_memory_region(kvm
, old
, new, change
);
1947 * For DELETE/MOVE, revert the above INVALID change. No
1948 * modifications required since the original slot was preserved
1949 * in the inactive slots. Changing the active memslots also
1950 * release slots_arch_lock.
1952 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1953 kvm_activate_memslot(kvm
, invalid_slot
, old
);
1954 kfree(invalid_slot
);
1956 mutex_unlock(&kvm
->slots_arch_lock
);
1962 * For DELETE and MOVE, the working slot is now active as the INVALID
1963 * version of the old slot. MOVE is particularly special as it reuses
1964 * the old slot and returns a copy of the old slot (in working_slot).
1965 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1966 * old slot is detached but otherwise preserved.
1968 if (change
== KVM_MR_CREATE
)
1969 kvm_create_memslot(kvm
, new);
1970 else if (change
== KVM_MR_DELETE
)
1971 kvm_delete_memslot(kvm
, old
, invalid_slot
);
1972 else if (change
== KVM_MR_MOVE
)
1973 kvm_move_memslot(kvm
, old
, new, invalid_slot
);
1974 else if (change
== KVM_MR_FLAGS_ONLY
)
1975 kvm_update_flags_memslot(kvm
, old
, new);
1979 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1980 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1981 kfree(invalid_slot
);
1984 * No need to refresh new->arch, changes after dropping slots_arch_lock
1985 * will directly hit the final, active memslot. Architectures are
1986 * responsible for knowing that new->arch may be stale.
1988 kvm_commit_memory_region(kvm
, old
, new, change
);
1993 static bool kvm_check_memslot_overlap(struct kvm_memslots
*slots
, int id
,
1994 gfn_t start
, gfn_t end
)
1996 struct kvm_memslot_iter iter
;
1998 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, start
, end
) {
1999 if (iter
.slot
->id
!= id
)
2007 * Allocate some memory and give it an address in the guest physical address
2010 * Discontiguous memory is allowed, mostly for framebuffers.
2012 * Must be called holding kvm->slots_lock for write.
2014 int __kvm_set_memory_region(struct kvm
*kvm
,
2015 const struct kvm_userspace_memory_region2
*mem
)
2017 struct kvm_memory_slot
*old
, *new;
2018 struct kvm_memslots
*slots
;
2019 enum kvm_mr_change change
;
2020 unsigned long npages
;
2025 r
= check_memory_region_flags(kvm
, mem
);
2029 as_id
= mem
->slot
>> 16;
2030 id
= (u16
)mem
->slot
;
2032 /* General sanity checks */
2033 if ((mem
->memory_size
& (PAGE_SIZE
- 1)) ||
2034 (mem
->memory_size
!= (unsigned long)mem
->memory_size
))
2036 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
2038 /* We can read the guest memory with __xxx_user() later on. */
2039 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
2040 (mem
->userspace_addr
!= untagged_addr(mem
->userspace_addr
)) ||
2041 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
2044 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
&&
2045 (mem
->guest_memfd_offset
& (PAGE_SIZE
- 1) ||
2046 mem
->guest_memfd_offset
+ mem
->memory_size
< mem
->guest_memfd_offset
))
2048 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_MEM_SLOTS_NUM
)
2050 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
2052 if ((mem
->memory_size
>> PAGE_SHIFT
) > KVM_MEM_MAX_NR_PAGES
)
2055 slots
= __kvm_memslots(kvm
, as_id
);
2058 * Note, the old memslot (and the pointer itself!) may be invalidated
2059 * and/or destroyed by kvm_set_memslot().
2061 old
= id_to_memslot(slots
, id
);
2063 if (!mem
->memory_size
) {
2064 if (!old
|| !old
->npages
)
2067 if (WARN_ON_ONCE(kvm
->nr_memslot_pages
< old
->npages
))
2070 return kvm_set_memslot(kvm
, old
, NULL
, KVM_MR_DELETE
);
2073 base_gfn
= (mem
->guest_phys_addr
>> PAGE_SHIFT
);
2074 npages
= (mem
->memory_size
>> PAGE_SHIFT
);
2076 if (!old
|| !old
->npages
) {
2077 change
= KVM_MR_CREATE
;
2080 * To simplify KVM internals, the total number of pages across
2081 * all memslots must fit in an unsigned long.
2083 if ((kvm
->nr_memslot_pages
+ npages
) < kvm
->nr_memslot_pages
)
2085 } else { /* Modify an existing slot. */
2086 /* Private memslots are immutable, they can only be deleted. */
2087 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
)
2089 if ((mem
->userspace_addr
!= old
->userspace_addr
) ||
2090 (npages
!= old
->npages
) ||
2091 ((mem
->flags
^ old
->flags
) & KVM_MEM_READONLY
))
2094 if (base_gfn
!= old
->base_gfn
)
2095 change
= KVM_MR_MOVE
;
2096 else if (mem
->flags
!= old
->flags
)
2097 change
= KVM_MR_FLAGS_ONLY
;
2098 else /* Nothing to change. */
2102 if ((change
== KVM_MR_CREATE
|| change
== KVM_MR_MOVE
) &&
2103 kvm_check_memslot_overlap(slots
, id
, base_gfn
, base_gfn
+ npages
))
2106 /* Allocate a slot that will persist in the memslot. */
2107 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT
);
2113 new->base_gfn
= base_gfn
;
2114 new->npages
= npages
;
2115 new->flags
= mem
->flags
;
2116 new->userspace_addr
= mem
->userspace_addr
;
2117 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
) {
2118 r
= kvm_gmem_bind(kvm
, new, mem
->guest_memfd
, mem
->guest_memfd_offset
);
2123 r
= kvm_set_memslot(kvm
, old
, new, change
);
2130 if (mem
->flags
& KVM_MEM_GUEST_MEMFD
)
2131 kvm_gmem_unbind(new);
2136 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
2138 int kvm_set_memory_region(struct kvm
*kvm
,
2139 const struct kvm_userspace_memory_region2
*mem
)
2143 mutex_lock(&kvm
->slots_lock
);
2144 r
= __kvm_set_memory_region(kvm
, mem
);
2145 mutex_unlock(&kvm
->slots_lock
);
2148 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
2150 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
2151 struct kvm_userspace_memory_region2
*mem
)
2153 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
2156 return kvm_set_memory_region(kvm
, mem
);
2159 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2161 * kvm_get_dirty_log - get a snapshot of dirty pages
2162 * @kvm: pointer to kvm instance
2163 * @log: slot id and address to which we copy the log
2164 * @is_dirty: set to '1' if any dirty pages were found
2165 * @memslot: set to the associated memslot, always valid on success
2167 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
2168 int *is_dirty
, struct kvm_memory_slot
**memslot
)
2170 struct kvm_memslots
*slots
;
2173 unsigned long any
= 0;
2175 /* Dirty ring tracking may be exclusive to dirty log tracking */
2176 if (!kvm_use_dirty_bitmap(kvm
))
2182 as_id
= log
->slot
>> 16;
2183 id
= (u16
)log
->slot
;
2184 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_USER_MEM_SLOTS
)
2187 slots
= __kvm_memslots(kvm
, as_id
);
2188 *memslot
= id_to_memslot(slots
, id
);
2189 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
2192 kvm_arch_sync_dirty_log(kvm
, *memslot
);
2194 n
= kvm_dirty_bitmap_bytes(*memslot
);
2196 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
2197 any
= (*memslot
)->dirty_bitmap
[i
];
2199 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
2206 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
2208 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2210 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2211 * and reenable dirty page tracking for the corresponding pages.
2212 * @kvm: pointer to kvm instance
2213 * @log: slot id and address to which we copy the log
2215 * We need to keep it in mind that VCPU threads can write to the bitmap
2216 * concurrently. So, to avoid losing track of dirty pages we keep the
2219 * 1. Take a snapshot of the bit and clear it if needed.
2220 * 2. Write protect the corresponding page.
2221 * 3. Copy the snapshot to the userspace.
2222 * 4. Upon return caller flushes TLB's if needed.
2224 * Between 2 and 4, the guest may write to the page using the remaining TLB
2225 * entry. This is not a problem because the page is reported dirty using
2226 * the snapshot taken before and step 4 ensures that writes done after
2227 * exiting to userspace will be logged for the next call.
2230 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
2232 struct kvm_memslots
*slots
;
2233 struct kvm_memory_slot
*memslot
;
2236 unsigned long *dirty_bitmap
;
2237 unsigned long *dirty_bitmap_buffer
;
2240 /* Dirty ring tracking may be exclusive to dirty log tracking */
2241 if (!kvm_use_dirty_bitmap(kvm
))
2244 as_id
= log
->slot
>> 16;
2245 id
= (u16
)log
->slot
;
2246 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_USER_MEM_SLOTS
)
2249 slots
= __kvm_memslots(kvm
, as_id
);
2250 memslot
= id_to_memslot(slots
, id
);
2251 if (!memslot
|| !memslot
->dirty_bitmap
)
2254 dirty_bitmap
= memslot
->dirty_bitmap
;
2256 kvm_arch_sync_dirty_log(kvm
, memslot
);
2258 n
= kvm_dirty_bitmap_bytes(memslot
);
2260 if (kvm
->manual_dirty_log_protect
) {
2262 * Unlike kvm_get_dirty_log, we always return false in *flush,
2263 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2264 * is some code duplication between this function and
2265 * kvm_get_dirty_log, but hopefully all architecture
2266 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2267 * can be eliminated.
2269 dirty_bitmap_buffer
= dirty_bitmap
;
2271 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2272 memset(dirty_bitmap_buffer
, 0, n
);
2275 for (i
= 0; i
< n
/ sizeof(long); i
++) {
2279 if (!dirty_bitmap
[i
])
2283 mask
= xchg(&dirty_bitmap
[i
], 0);
2284 dirty_bitmap_buffer
[i
] = mask
;
2286 offset
= i
* BITS_PER_LONG
;
2287 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2290 KVM_MMU_UNLOCK(kvm
);
2294 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2296 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
2303 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2304 * @kvm: kvm instance
2305 * @log: slot id and address to which we copy the log
2307 * Steps 1-4 below provide general overview of dirty page logging. See
2308 * kvm_get_dirty_log_protect() function description for additional details.
2310 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2311 * always flush the TLB (step 4) even if previous step failed and the dirty
2312 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2313 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2314 * writes will be marked dirty for next log read.
2316 * 1. Take a snapshot of the bit and clear it if needed.
2317 * 2. Write protect the corresponding page.
2318 * 3. Copy the snapshot to the userspace.
2319 * 4. Flush TLB's if needed.
2321 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2322 struct kvm_dirty_log
*log
)
2326 mutex_lock(&kvm
->slots_lock
);
2328 r
= kvm_get_dirty_log_protect(kvm
, log
);
2330 mutex_unlock(&kvm
->slots_lock
);
2335 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2336 * and reenable dirty page tracking for the corresponding pages.
2337 * @kvm: pointer to kvm instance
2338 * @log: slot id and address from which to fetch the bitmap of dirty pages
2340 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
2341 struct kvm_clear_dirty_log
*log
)
2343 struct kvm_memslots
*slots
;
2344 struct kvm_memory_slot
*memslot
;
2348 unsigned long *dirty_bitmap
;
2349 unsigned long *dirty_bitmap_buffer
;
2352 /* Dirty ring tracking may be exclusive to dirty log tracking */
2353 if (!kvm_use_dirty_bitmap(kvm
))
2356 as_id
= log
->slot
>> 16;
2357 id
= (u16
)log
->slot
;
2358 if (as_id
>= kvm_arch_nr_memslot_as_ids(kvm
) || id
>= KVM_USER_MEM_SLOTS
)
2361 if (log
->first_page
& 63)
2364 slots
= __kvm_memslots(kvm
, as_id
);
2365 memslot
= id_to_memslot(slots
, id
);
2366 if (!memslot
|| !memslot
->dirty_bitmap
)
2369 dirty_bitmap
= memslot
->dirty_bitmap
;
2371 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
2373 if (log
->first_page
> memslot
->npages
||
2374 log
->num_pages
> memslot
->npages
- log
->first_page
||
2375 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
2378 kvm_arch_sync_dirty_log(kvm
, memslot
);
2381 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2382 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
2386 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
2387 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
2388 i
++, offset
+= BITS_PER_LONG
) {
2389 unsigned long mask
= *dirty_bitmap_buffer
++;
2390 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
2394 mask
&= atomic_long_fetch_andnot(mask
, p
);
2397 * mask contains the bits that really have been cleared. This
2398 * never includes any bits beyond the length of the memslot (if
2399 * the length is not aligned to 64 pages), therefore it is not
2400 * a problem if userspace sets them in log->dirty_bitmap.
2404 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2408 KVM_MMU_UNLOCK(kvm
);
2411 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2416 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
2417 struct kvm_clear_dirty_log
*log
)
2421 mutex_lock(&kvm
->slots_lock
);
2423 r
= kvm_clear_dirty_log_protect(kvm
, log
);
2425 mutex_unlock(&kvm
->slots_lock
);
2428 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2430 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
2432 * Returns true if _all_ gfns in the range [@start, @end) have attributes
2435 bool kvm_range_has_memory_attributes(struct kvm
*kvm
, gfn_t start
, gfn_t end
,
2436 unsigned long attrs
)
2438 XA_STATE(xas
, &kvm
->mem_attr_array
, start
);
2439 unsigned long index
;
2446 has_attrs
= !xas_find(&xas
, end
- 1);
2451 for (index
= start
; index
< end
; index
++) {
2453 entry
= xas_next(&xas
);
2454 } while (xas_retry(&xas
, entry
));
2456 if (xas
.xa_index
!= index
|| xa_to_value(entry
) != attrs
) {
2467 static u64
kvm_supported_mem_attributes(struct kvm
*kvm
)
2469 if (!kvm
|| kvm_arch_has_private_mem(kvm
))
2470 return KVM_MEMORY_ATTRIBUTE_PRIVATE
;
2475 static __always_inline
void kvm_handle_gfn_range(struct kvm
*kvm
,
2476 struct kvm_mmu_notifier_range
*range
)
2478 struct kvm_gfn_range gfn_range
;
2479 struct kvm_memory_slot
*slot
;
2480 struct kvm_memslots
*slots
;
2481 struct kvm_memslot_iter iter
;
2482 bool found_memslot
= false;
2486 gfn_range
.arg
= range
->arg
;
2487 gfn_range
.may_block
= range
->may_block
;
2489 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
2490 slots
= __kvm_memslots(kvm
, i
);
2492 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, range
->start
, range
->end
) {
2494 gfn_range
.slot
= slot
;
2496 gfn_range
.start
= max(range
->start
, slot
->base_gfn
);
2497 gfn_range
.end
= min(range
->end
, slot
->base_gfn
+ slot
->npages
);
2498 if (gfn_range
.start
>= gfn_range
.end
)
2501 if (!found_memslot
) {
2502 found_memslot
= true;
2504 if (!IS_KVM_NULL_FN(range
->on_lock
))
2505 range
->on_lock(kvm
);
2508 ret
|= range
->handler(kvm
, &gfn_range
);
2512 if (range
->flush_on_ret
&& ret
)
2513 kvm_flush_remote_tlbs(kvm
);
2516 KVM_MMU_UNLOCK(kvm
);
2519 static bool kvm_pre_set_memory_attributes(struct kvm
*kvm
,
2520 struct kvm_gfn_range
*range
)
2523 * Unconditionally add the range to the invalidation set, regardless of
2524 * whether or not the arch callback actually needs to zap SPTEs. E.g.
2525 * if KVM supports RWX attributes in the future and the attributes are
2526 * going from R=>RW, zapping isn't strictly necessary. Unconditionally
2527 * adding the range allows KVM to require that MMU invalidations add at
2528 * least one range between begin() and end(), e.g. allows KVM to detect
2529 * bugs where the add() is missed. Relaxing the rule *might* be safe,
2530 * but it's not obvious that allowing new mappings while the attributes
2531 * are in flux is desirable or worth the complexity.
2533 kvm_mmu_invalidate_range_add(kvm
, range
->start
, range
->end
);
2535 return kvm_arch_pre_set_memory_attributes(kvm
, range
);
2538 /* Set @attributes for the gfn range [@start, @end). */
2539 static int kvm_vm_set_mem_attributes(struct kvm
*kvm
, gfn_t start
, gfn_t end
,
2540 unsigned long attributes
)
2542 struct kvm_mmu_notifier_range pre_set_range
= {
2545 .handler
= kvm_pre_set_memory_attributes
,
2546 .on_lock
= kvm_mmu_invalidate_begin
,
2547 .flush_on_ret
= true,
2550 struct kvm_mmu_notifier_range post_set_range
= {
2553 .arg
.attributes
= attributes
,
2554 .handler
= kvm_arch_post_set_memory_attributes
,
2555 .on_lock
= kvm_mmu_invalidate_end
,
2562 entry
= attributes
? xa_mk_value(attributes
) : NULL
;
2564 mutex_lock(&kvm
->slots_lock
);
2566 /* Nothing to do if the entire range as the desired attributes. */
2567 if (kvm_range_has_memory_attributes(kvm
, start
, end
, attributes
))
2571 * Reserve memory ahead of time to avoid having to deal with failures
2572 * partway through setting the new attributes.
2574 for (i
= start
; i
< end
; i
++) {
2575 r
= xa_reserve(&kvm
->mem_attr_array
, i
, GFP_KERNEL_ACCOUNT
);
2580 kvm_handle_gfn_range(kvm
, &pre_set_range
);
2582 for (i
= start
; i
< end
; i
++) {
2583 r
= xa_err(xa_store(&kvm
->mem_attr_array
, i
, entry
,
2584 GFP_KERNEL_ACCOUNT
));
2588 kvm_handle_gfn_range(kvm
, &post_set_range
);
2591 mutex_unlock(&kvm
->slots_lock
);
2595 static int kvm_vm_ioctl_set_mem_attributes(struct kvm
*kvm
,
2596 struct kvm_memory_attributes
*attrs
)
2600 /* flags is currently not used. */
2603 if (attrs
->attributes
& ~kvm_supported_mem_attributes(kvm
))
2605 if (attrs
->size
== 0 || attrs
->address
+ attrs
->size
< attrs
->address
)
2607 if (!PAGE_ALIGNED(attrs
->address
) || !PAGE_ALIGNED(attrs
->size
))
2610 start
= attrs
->address
>> PAGE_SHIFT
;
2611 end
= (attrs
->address
+ attrs
->size
) >> PAGE_SHIFT
;
2614 * xarray tracks data using "unsigned long", and as a result so does
2615 * KVM. For simplicity, supports generic attributes only on 64-bit
2618 BUILD_BUG_ON(sizeof(attrs
->attributes
) != sizeof(unsigned long));
2620 return kvm_vm_set_mem_attributes(kvm
, start
, end
, attrs
->attributes
);
2622 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2624 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
2626 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
2628 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
2630 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2632 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2633 u64 gen
= slots
->generation
;
2634 struct kvm_memory_slot
*slot
;
2637 * This also protects against using a memslot from a different address space,
2638 * since different address spaces have different generation numbers.
2640 if (unlikely(gen
!= vcpu
->last_used_slot_gen
)) {
2641 vcpu
->last_used_slot
= NULL
;
2642 vcpu
->last_used_slot_gen
= gen
;
2645 slot
= try_get_memslot(vcpu
->last_used_slot
, gfn
);
2650 * Fall back to searching all memslots. We purposely use
2651 * search_memslots() instead of __gfn_to_memslot() to avoid
2652 * thrashing the VM-wide last_used_slot in kvm_memslots.
2654 slot
= search_memslots(slots
, gfn
, false);
2656 vcpu
->last_used_slot
= slot
;
2663 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
2665 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
2667 return kvm_is_visible_memslot(memslot
);
2669 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
2671 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2673 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2675 return kvm_is_visible_memslot(memslot
);
2677 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
2679 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2681 struct vm_area_struct
*vma
;
2682 unsigned long addr
, size
;
2686 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
2687 if (kvm_is_error_hva(addr
))
2690 mmap_read_lock(current
->mm
);
2691 vma
= find_vma(current
->mm
, addr
);
2695 size
= vma_kernel_pagesize(vma
);
2698 mmap_read_unlock(current
->mm
);
2703 static bool memslot_is_readonly(const struct kvm_memory_slot
*slot
)
2705 return slot
->flags
& KVM_MEM_READONLY
;
2708 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2709 gfn_t
*nr_pages
, bool write
)
2711 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
2712 return KVM_HVA_ERR_BAD
;
2714 if (memslot_is_readonly(slot
) && write
)
2715 return KVM_HVA_ERR_RO_BAD
;
2718 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
2720 return __gfn_to_hva_memslot(slot
, gfn
);
2723 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2726 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
2729 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
2732 return gfn_to_hva_many(slot
, gfn
, NULL
);
2734 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
2736 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
2738 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
2740 EXPORT_SYMBOL_GPL(gfn_to_hva
);
2742 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2744 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
2746 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
2749 * Return the hva of a @gfn and the R/W attribute if possible.
2751 * @slot: the kvm_memory_slot which contains @gfn
2752 * @gfn: the gfn to be translated
2753 * @writable: used to return the read/write attribute of the @slot if the hva
2754 * is valid and @writable is not NULL
2756 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
2757 gfn_t gfn
, bool *writable
)
2759 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
2761 if (!kvm_is_error_hva(hva
) && writable
)
2762 *writable
= !memslot_is_readonly(slot
);
2767 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
2769 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2771 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2774 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
2776 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2778 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2781 static inline int check_user_page_hwpoison(unsigned long addr
)
2783 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
2785 rc
= get_user_pages(addr
, 1, flags
, NULL
);
2786 return rc
== -EHWPOISON
;
2790 * The fast path to get the writable pfn which will be stored in @pfn,
2791 * true indicates success, otherwise false is returned. It's also the
2792 * only part that runs if we can in atomic context.
2794 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
2795 bool *writable
, kvm_pfn_t
*pfn
)
2797 struct page
*page
[1];
2800 * Fast pin a writable pfn only if it is a write fault request
2801 * or the caller allows to map a writable pfn for a read fault
2804 if (!(write_fault
|| writable
))
2807 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
2808 *pfn
= page_to_pfn(page
[0]);
2819 * The slow path to get the pfn of the specified host virtual address,
2820 * 1 indicates success, -errno is returned if error is detected.
2822 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
2823 bool interruptible
, bool *writable
, kvm_pfn_t
*pfn
)
2826 * When a VCPU accesses a page that is not mapped into the secondary
2827 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2828 * make progress. We always want to honor NUMA hinting faults in that
2829 * case, because GUP usage corresponds to memory accesses from the VCPU.
2830 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2831 * mapped into the secondary MMU and gets accessed by a VCPU.
2833 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2834 * implicitly honor NUMA hinting faults and don't need this flag.
2836 unsigned int flags
= FOLL_HWPOISON
| FOLL_HONOR_NUMA_FAULT
;
2843 *writable
= write_fault
;
2846 flags
|= FOLL_WRITE
;
2848 flags
|= FOLL_NOWAIT
;
2850 flags
|= FOLL_INTERRUPTIBLE
;
2852 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
2856 /* map read fault as writable if possible */
2857 if (unlikely(!write_fault
) && writable
) {
2860 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
2866 *pfn
= page_to_pfn(page
);
2870 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
2872 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
2875 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
2881 static int kvm_try_get_pfn(kvm_pfn_t pfn
)
2883 struct page
*page
= kvm_pfn_to_refcounted_page(pfn
);
2888 return get_page_unless_zero(page
);
2891 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
2892 unsigned long addr
, bool write_fault
,
2893 bool *writable
, kvm_pfn_t
*p_pfn
)
2901 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2904 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2905 * not call the fault handler, so do it here.
2907 bool unlocked
= false;
2908 r
= fixup_user_fault(current
->mm
, addr
,
2909 (write_fault
? FAULT_FLAG_WRITE
: 0),
2916 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2921 pte
= ptep_get(ptep
);
2923 if (write_fault
&& !pte_write(pte
)) {
2924 pfn
= KVM_PFN_ERR_RO_FAULT
;
2929 *writable
= pte_write(pte
);
2933 * Get a reference here because callers of *hva_to_pfn* and
2934 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2935 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2936 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2937 * simply do nothing for reserved pfns.
2939 * Whoever called remap_pfn_range is also going to call e.g.
2940 * unmap_mapping_range before the underlying pages are freed,
2941 * causing a call to our MMU notifier.
2943 * Certain IO or PFNMAP mappings can be backed with valid
2944 * struct pages, but be allocated without refcounting e.g.,
2945 * tail pages of non-compound higher order allocations, which
2946 * would then underflow the refcount when the caller does the
2947 * required put_page. Don't allow those pages here.
2949 if (!kvm_try_get_pfn(pfn
))
2953 pte_unmap_unlock(ptep
, ptl
);
2960 * Pin guest page in memory and return its pfn.
2961 * @addr: host virtual address which maps memory to the guest
2962 * @atomic: whether this function can sleep
2963 * @interruptible: whether the process can be interrupted by non-fatal signals
2964 * @async: whether this function need to wait IO complete if the
2965 * host page is not in the memory
2966 * @write_fault: whether we should get a writable host page
2967 * @writable: whether it allows to map a writable host page for !@write_fault
2969 * The function will map a writable host page for these two cases:
2970 * 1): @write_fault = true
2971 * 2): @write_fault = false && @writable, @writable will tell the caller
2972 * whether the mapping is writable.
2974 kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool interruptible
,
2975 bool *async
, bool write_fault
, bool *writable
)
2977 struct vm_area_struct
*vma
;
2981 /* we can do it either atomically or asynchronously, not both */
2982 BUG_ON(atomic
&& async
);
2984 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
2988 return KVM_PFN_ERR_FAULT
;
2990 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, interruptible
,
2994 if (npages
== -EINTR
)
2995 return KVM_PFN_ERR_SIGPENDING
;
2997 mmap_read_lock(current
->mm
);
2998 if (npages
== -EHWPOISON
||
2999 (!async
&& check_user_page_hwpoison(addr
))) {
3000 pfn
= KVM_PFN_ERR_HWPOISON
;
3005 vma
= vma_lookup(current
->mm
, addr
);
3008 pfn
= KVM_PFN_ERR_FAULT
;
3009 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
3010 r
= hva_to_pfn_remapped(vma
, addr
, write_fault
, writable
, &pfn
);
3014 pfn
= KVM_PFN_ERR_FAULT
;
3016 if (async
&& vma_is_valid(vma
, write_fault
))
3018 pfn
= KVM_PFN_ERR_FAULT
;
3021 mmap_read_unlock(current
->mm
);
3025 kvm_pfn_t
__gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
3026 bool atomic
, bool interruptible
, bool *async
,
3027 bool write_fault
, bool *writable
, hva_t
*hva
)
3029 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
3034 if (addr
== KVM_HVA_ERR_RO_BAD
) {
3037 return KVM_PFN_ERR_RO_FAULT
;
3040 if (kvm_is_error_hva(addr
)) {
3043 return KVM_PFN_NOSLOT
;
3046 /* Do not map writable pfn in the readonly memslot. */
3047 if (writable
&& memslot_is_readonly(slot
)) {
3052 return hva_to_pfn(addr
, atomic
, interruptible
, async
, write_fault
,
3055 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
3057 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
3060 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, false,
3061 NULL
, write_fault
, writable
, NULL
);
3063 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
3065 kvm_pfn_t
gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
3067 return __gfn_to_pfn_memslot(slot
, gfn
, false, false, NULL
, true,
3070 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
3072 kvm_pfn_t
gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
3074 return __gfn_to_pfn_memslot(slot
, gfn
, true, false, NULL
, true,
3077 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
3079 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3081 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
3083 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
3085 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
3087 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
3089 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
3091 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3093 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
3095 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
3097 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3098 struct page
**pages
, int nr_pages
)
3103 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
3104 if (kvm_is_error_hva(addr
))
3107 if (entry
< nr_pages
)
3110 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
3112 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
3115 * Do not use this helper unless you are absolutely certain the gfn _must_ be
3116 * backed by 'struct page'. A valid example is if the backing memslot is
3117 * controlled by KVM. Note, if the returned page is valid, it's refcount has
3118 * been elevated by gfn_to_pfn().
3120 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
3125 pfn
= gfn_to_pfn(kvm
, gfn
);
3127 if (is_error_noslot_pfn(pfn
))
3128 return KVM_ERR_PTR_BAD_PAGE
;
3130 page
= kvm_pfn_to_refcounted_page(pfn
);
3132 return KVM_ERR_PTR_BAD_PAGE
;
3136 EXPORT_SYMBOL_GPL(gfn_to_page
);
3138 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
)
3141 kvm_release_pfn_dirty(pfn
);
3143 kvm_release_pfn_clean(pfn
);
3146 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
3150 struct page
*page
= KVM_UNMAPPED_PAGE
;
3155 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
3156 if (is_error_noslot_pfn(pfn
))
3159 if (pfn_valid(pfn
)) {
3160 page
= pfn_to_page(pfn
);
3162 #ifdef CONFIG_HAS_IOMEM
3164 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
3178 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
3180 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
3188 if (map
->page
!= KVM_UNMAPPED_PAGE
)
3190 #ifdef CONFIG_HAS_IOMEM
3196 kvm_vcpu_mark_page_dirty(vcpu
, map
->gfn
);
3198 kvm_release_pfn(map
->pfn
, dirty
);
3203 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
3205 static bool kvm_is_ad_tracked_page(struct page
*page
)
3208 * Per page-flags.h, pages tagged PG_reserved "should in general not be
3209 * touched (e.g. set dirty) except by its owner".
3211 return !PageReserved(page
);
3214 static void kvm_set_page_dirty(struct page
*page
)
3216 if (kvm_is_ad_tracked_page(page
))
3220 static void kvm_set_page_accessed(struct page
*page
)
3222 if (kvm_is_ad_tracked_page(page
))
3223 mark_page_accessed(page
);
3226 void kvm_release_page_clean(struct page
*page
)
3228 WARN_ON(is_error_page(page
));
3230 kvm_set_page_accessed(page
);
3233 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
3235 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
3239 if (is_error_noslot_pfn(pfn
))
3242 page
= kvm_pfn_to_refcounted_page(pfn
);
3246 kvm_release_page_clean(page
);
3248 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
3250 void kvm_release_page_dirty(struct page
*page
)
3252 WARN_ON(is_error_page(page
));
3254 kvm_set_page_dirty(page
);
3255 kvm_release_page_clean(page
);
3257 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
3259 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
3263 if (is_error_noslot_pfn(pfn
))
3266 page
= kvm_pfn_to_refcounted_page(pfn
);
3270 kvm_release_page_dirty(page
);
3272 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
3275 * Note, checking for an error/noslot pfn is the caller's responsibility when
3276 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
3277 * "set" helpers are not to be used when the pfn might point at garbage.
3279 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
3281 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3285 kvm_set_page_dirty(pfn_to_page(pfn
));
3287 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
3289 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
3291 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3295 kvm_set_page_accessed(pfn_to_page(pfn
));
3297 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
3299 static int next_segment(unsigned long len
, int offset
)
3301 if (len
> PAGE_SIZE
- offset
)
3302 return PAGE_SIZE
- offset
;
3307 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3308 void *data
, int offset
, int len
)
3313 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3314 if (kvm_is_error_hva(addr
))
3316 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
3322 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
3325 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3327 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3329 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
3331 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
3332 int offset
, int len
)
3334 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3336 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3338 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
3340 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
3342 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3344 int offset
= offset_in_page(gpa
);
3347 while ((seg
= next_segment(len
, offset
)) != 0) {
3348 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
3358 EXPORT_SYMBOL_GPL(kvm_read_guest
);
3360 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
3362 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3364 int offset
= offset_in_page(gpa
);
3367 while ((seg
= next_segment(len
, offset
)) != 0) {
3368 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3378 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
3380 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3381 void *data
, int offset
, unsigned long len
)
3386 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3387 if (kvm_is_error_hva(addr
))
3389 pagefault_disable();
3390 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
3397 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3398 void *data
, unsigned long len
)
3400 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3401 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3402 int offset
= offset_in_page(gpa
);
3404 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
3406 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
3408 static int __kvm_write_guest_page(struct kvm
*kvm
,
3409 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
3410 const void *data
, int offset
, int len
)
3415 addr
= gfn_to_hva_memslot(memslot
, gfn
);
3416 if (kvm_is_error_hva(addr
))
3418 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
3421 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3425 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
3426 const void *data
, int offset
, int len
)
3428 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3430 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
3432 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
3434 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
3435 const void *data
, int offset
, int len
)
3437 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3439 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
3441 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
3443 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
3446 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3448 int offset
= offset_in_page(gpa
);
3451 while ((seg
= next_segment(len
, offset
)) != 0) {
3452 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
3462 EXPORT_SYMBOL_GPL(kvm_write_guest
);
3464 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
3467 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3469 int offset
= offset_in_page(gpa
);
3472 while ((seg
= next_segment(len
, offset
)) != 0) {
3473 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3483 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
3485 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
3486 struct gfn_to_hva_cache
*ghc
,
3487 gpa_t gpa
, unsigned long len
)
3489 int offset
= offset_in_page(gpa
);
3490 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
3491 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
3492 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
3493 gfn_t nr_pages_avail
;
3495 /* Update ghc->generation before performing any error checks. */
3496 ghc
->generation
= slots
->generation
;
3498 if (start_gfn
> end_gfn
) {
3499 ghc
->hva
= KVM_HVA_ERR_BAD
;
3504 * If the requested region crosses two memslots, we still
3505 * verify that the entire region is valid here.
3507 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
3508 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
3509 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
3511 if (kvm_is_error_hva(ghc
->hva
))
3515 /* Use the slow path for cross page reads and writes. */
3516 if (nr_pages_needed
== 1)
3519 ghc
->memslot
= NULL
;
3526 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3527 gpa_t gpa
, unsigned long len
)
3529 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3530 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
3532 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
3534 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3535 void *data
, unsigned int offset
,
3538 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3540 gpa_t gpa
= ghc
->gpa
+ offset
;
3542 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3545 if (slots
->generation
!= ghc
->generation
) {
3546 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3550 if (kvm_is_error_hva(ghc
->hva
))
3553 if (unlikely(!ghc
->memslot
))
3554 return kvm_write_guest(kvm
, gpa
, data
, len
);
3556 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
3559 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
3563 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
3565 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3566 void *data
, unsigned long len
)
3568 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3570 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
3572 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3573 void *data
, unsigned int offset
,
3576 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3578 gpa_t gpa
= ghc
->gpa
+ offset
;
3580 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3583 if (slots
->generation
!= ghc
->generation
) {
3584 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3588 if (kvm_is_error_hva(ghc
->hva
))
3591 if (unlikely(!ghc
->memslot
))
3592 return kvm_read_guest(kvm
, gpa
, data
, len
);
3594 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
3600 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
3602 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3603 void *data
, unsigned long len
)
3605 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3607 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
3609 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
3611 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3612 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3614 int offset
= offset_in_page(gpa
);
3617 while ((seg
= next_segment(len
, offset
)) != 0) {
3618 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
3627 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
3629 void mark_page_dirty_in_slot(struct kvm
*kvm
,
3630 const struct kvm_memory_slot
*memslot
,
3633 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
3635 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3636 if (WARN_ON_ONCE(vcpu
&& vcpu
->kvm
!= kvm
))
3639 WARN_ON_ONCE(!vcpu
&& !kvm_arch_allow_write_without_running_vcpu(kvm
));
3642 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
3643 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
3644 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
3646 if (kvm
->dirty_ring_size
&& vcpu
)
3647 kvm_dirty_ring_push(vcpu
, slot
, rel_gfn
);
3648 else if (memslot
->dirty_bitmap
)
3649 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
3652 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
3654 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
3656 struct kvm_memory_slot
*memslot
;
3658 memslot
= gfn_to_memslot(kvm
, gfn
);
3659 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3661 EXPORT_SYMBOL_GPL(mark_page_dirty
);
3663 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3665 struct kvm_memory_slot
*memslot
;
3667 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3668 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
3670 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
3672 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
3674 if (!vcpu
->sigset_active
)
3678 * This does a lockless modification of ->real_blocked, which is fine
3679 * because, only current can change ->real_blocked and all readers of
3680 * ->real_blocked don't care as long ->real_blocked is always a subset
3683 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
3686 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
3688 if (!vcpu
->sigset_active
)
3691 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
3692 sigemptyset(¤t
->real_blocked
);
3695 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3697 unsigned int old
, val
, grow
, grow_start
;
3699 old
= val
= vcpu
->halt_poll_ns
;
3700 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3701 grow
= READ_ONCE(halt_poll_ns_grow
);
3706 if (val
< grow_start
)
3709 vcpu
->halt_poll_ns
= val
;
3711 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
3714 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3716 unsigned int old
, val
, shrink
, grow_start
;
3718 old
= val
= vcpu
->halt_poll_ns
;
3719 shrink
= READ_ONCE(halt_poll_ns_shrink
);
3720 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3726 if (val
< grow_start
)
3729 vcpu
->halt_poll_ns
= val
;
3730 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
3733 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
3736 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3738 if (kvm_arch_vcpu_runnable(vcpu
))
3740 if (kvm_cpu_has_pending_timer(vcpu
))
3742 if (signal_pending(current
))
3744 if (kvm_check_request(KVM_REQ_UNBLOCK
, vcpu
))
3749 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3754 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3755 * pending. This is mostly used when halting a vCPU, but may also be used
3756 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3758 bool kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
3760 struct rcuwait
*wait
= kvm_arch_vcpu_get_wait(vcpu
);
3761 bool waited
= false;
3763 vcpu
->stat
.generic
.blocking
= 1;
3766 kvm_arch_vcpu_blocking(vcpu
);
3767 prepare_to_rcuwait(wait
);
3771 set_current_state(TASK_INTERRUPTIBLE
);
3773 if (kvm_vcpu_check_block(vcpu
) < 0)
3781 finish_rcuwait(wait
);
3782 kvm_arch_vcpu_unblocking(vcpu
);
3785 vcpu
->stat
.generic
.blocking
= 0;
3790 static inline void update_halt_poll_stats(struct kvm_vcpu
*vcpu
, ktime_t start
,
3791 ktime_t end
, bool success
)
3793 struct kvm_vcpu_stat_generic
*stats
= &vcpu
->stat
.generic
;
3794 u64 poll_ns
= ktime_to_ns(ktime_sub(end
, start
));
3796 ++vcpu
->stat
.generic
.halt_attempted_poll
;
3799 ++vcpu
->stat
.generic
.halt_successful_poll
;
3801 if (!vcpu_valid_wakeup(vcpu
))
3802 ++vcpu
->stat
.generic
.halt_poll_invalid
;
3804 stats
->halt_poll_success_ns
+= poll_ns
;
3805 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_success_hist
, poll_ns
);
3807 stats
->halt_poll_fail_ns
+= poll_ns
;
3808 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_fail_hist
, poll_ns
);
3812 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3814 struct kvm
*kvm
= vcpu
->kvm
;
3816 if (kvm
->override_halt_poll_ns
) {
3818 * Ensure kvm->max_halt_poll_ns is not read before
3819 * kvm->override_halt_poll_ns.
3821 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3824 return READ_ONCE(kvm
->max_halt_poll_ns
);
3827 return READ_ONCE(halt_poll_ns
);
3831 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3832 * polling is enabled, busy wait for a short time before blocking to avoid the
3833 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3836 void kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
3838 unsigned int max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3839 bool halt_poll_allowed
= !kvm_arch_no_poll(vcpu
);
3840 ktime_t start
, cur
, poll_end
;
3841 bool waited
= false;
3845 if (vcpu
->halt_poll_ns
> max_halt_poll_ns
)
3846 vcpu
->halt_poll_ns
= max_halt_poll_ns
;
3848 do_halt_poll
= halt_poll_allowed
&& vcpu
->halt_poll_ns
;
3850 start
= cur
= poll_end
= ktime_get();
3852 ktime_t stop
= ktime_add_ns(start
, vcpu
->halt_poll_ns
);
3855 if (kvm_vcpu_check_block(vcpu
) < 0)
3858 poll_end
= cur
= ktime_get();
3859 } while (kvm_vcpu_can_poll(cur
, stop
));
3862 waited
= kvm_vcpu_block(vcpu
);
3866 vcpu
->stat
.generic
.halt_wait_ns
+=
3867 ktime_to_ns(cur
) - ktime_to_ns(poll_end
);
3868 KVM_STATS_LOG_HIST_UPDATE(vcpu
->stat
.generic
.halt_wait_hist
,
3869 ktime_to_ns(cur
) - ktime_to_ns(poll_end
));
3872 /* The total time the vCPU was "halted", including polling time. */
3873 halt_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
3876 * Note, halt-polling is considered successful so long as the vCPU was
3877 * never actually scheduled out, i.e. even if the wake event arrived
3878 * after of the halt-polling loop itself, but before the full wait.
3881 update_halt_poll_stats(vcpu
, start
, poll_end
, !waited
);
3883 if (halt_poll_allowed
) {
3884 /* Recompute the max halt poll time in case it changed. */
3885 max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3887 if (!vcpu_valid_wakeup(vcpu
)) {
3888 shrink_halt_poll_ns(vcpu
);
3889 } else if (max_halt_poll_ns
) {
3890 if (halt_ns
<= vcpu
->halt_poll_ns
)
3892 /* we had a long block, shrink polling */
3893 else if (vcpu
->halt_poll_ns
&&
3894 halt_ns
> max_halt_poll_ns
)
3895 shrink_halt_poll_ns(vcpu
);
3896 /* we had a short halt and our poll time is too small */
3897 else if (vcpu
->halt_poll_ns
< max_halt_poll_ns
&&
3898 halt_ns
< max_halt_poll_ns
)
3899 grow_halt_poll_ns(vcpu
);
3901 vcpu
->halt_poll_ns
= 0;
3905 trace_kvm_vcpu_wakeup(halt_ns
, waited
, vcpu_valid_wakeup(vcpu
));
3907 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
3909 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
3911 if (__kvm_vcpu_wake_up(vcpu
)) {
3912 WRITE_ONCE(vcpu
->ready
, true);
3913 ++vcpu
->stat
.generic
.halt_wakeup
;
3919 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
3923 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3925 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3929 if (kvm_vcpu_wake_up(vcpu
))
3934 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3935 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3936 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3937 * within the vCPU thread itself.
3939 if (vcpu
== __this_cpu_read(kvm_running_vcpu
)) {
3940 if (vcpu
->mode
== IN_GUEST_MODE
)
3941 WRITE_ONCE(vcpu
->mode
, EXITING_GUEST_MODE
);
3946 * Note, the vCPU could get migrated to a different pCPU at any point
3947 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3948 * IPI to the previous pCPU. But, that's ok because the purpose of the
3949 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3950 * vCPU also requires it to leave IN_GUEST_MODE.
3952 if (kvm_arch_vcpu_should_kick(vcpu
)) {
3953 cpu
= READ_ONCE(vcpu
->cpu
);
3954 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
3955 smp_send_reschedule(cpu
);
3960 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
3961 #endif /* !CONFIG_S390 */
3963 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
3966 struct task_struct
*task
= NULL
;
3970 pid
= rcu_dereference(target
->pid
);
3972 task
= get_pid_task(pid
, PIDTYPE_PID
);
3976 ret
= yield_to(task
, 1);
3977 put_task_struct(task
);
3981 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
3984 * Helper that checks whether a VCPU is eligible for directed yield.
3985 * Most eligible candidate to yield is decided by following heuristics:
3987 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3988 * (preempted lock holder), indicated by @in_spin_loop.
3989 * Set at the beginning and cleared at the end of interception/PLE handler.
3991 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3992 * chance last time (mostly it has become eligible now since we have probably
3993 * yielded to lockholder in last iteration. This is done by toggling
3994 * @dy_eligible each time a VCPU checked for eligibility.)
3996 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3997 * to preempted lock-holder could result in wrong VCPU selection and CPU
3998 * burning. Giving priority for a potential lock-holder increases lock
4001 * Since algorithm is based on heuristics, accessing another VCPU data without
4002 * locking does not harm. It may result in trying to yield to same VCPU, fail
4003 * and continue with next VCPU and so on.
4005 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
4007 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
4010 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
4011 vcpu
->spin_loop
.dy_eligible
;
4013 if (vcpu
->spin_loop
.in_spin_loop
)
4014 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
4023 * Unlike kvm_arch_vcpu_runnable, this function is called outside
4024 * a vcpu_load/vcpu_put pair. However, for most architectures
4025 * kvm_arch_vcpu_runnable does not require vcpu_load.
4027 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
4029 return kvm_arch_vcpu_runnable(vcpu
);
4032 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
4034 if (kvm_arch_dy_runnable(vcpu
))
4037 #ifdef CONFIG_KVM_ASYNC_PF
4038 if (!list_empty_careful(&vcpu
->async_pf
.done
))
4045 bool __weak
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu
*vcpu
)
4050 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
4052 struct kvm
*kvm
= me
->kvm
;
4053 struct kvm_vcpu
*vcpu
;
4054 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
4060 kvm_vcpu_set_in_spin_loop(me
, true);
4062 * We boost the priority of a VCPU that is runnable but not
4063 * currently running, because it got preempted by something
4064 * else and called schedule in __vcpu_run. Hopefully that
4065 * VCPU is holding the lock that we need and will release it.
4066 * We approximate round-robin by starting at the last boosted VCPU.
4068 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
4069 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
4070 if (!pass
&& i
<= last_boosted_vcpu
) {
4071 i
= last_boosted_vcpu
;
4073 } else if (pass
&& i
> last_boosted_vcpu
)
4075 if (!READ_ONCE(vcpu
->ready
))
4079 if (kvm_vcpu_is_blocking(vcpu
) && !vcpu_dy_runnable(vcpu
))
4081 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
4082 !kvm_arch_dy_has_pending_interrupt(vcpu
) &&
4083 !kvm_arch_vcpu_in_kernel(vcpu
))
4085 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
4088 yielded
= kvm_vcpu_yield_to(vcpu
);
4090 kvm
->last_boosted_vcpu
= i
;
4092 } else if (yielded
< 0) {
4099 kvm_vcpu_set_in_spin_loop(me
, false);
4101 /* Ensure vcpu is not eligible during next spinloop */
4102 kvm_vcpu_set_dy_eligible(me
, false);
4104 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
4106 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
4108 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4109 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
4110 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
4111 kvm
->dirty_ring_size
/ PAGE_SIZE
);
4117 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
4119 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
4122 if (vmf
->pgoff
== 0)
4123 page
= virt_to_page(vcpu
->run
);
4125 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
4126 page
= virt_to_page(vcpu
->arch
.pio_data
);
4128 #ifdef CONFIG_KVM_MMIO
4129 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
4130 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
4132 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
4133 page
= kvm_dirty_ring_get_page(
4135 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
4137 return kvm_arch_vcpu_fault(vcpu
, vmf
);
4143 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
4144 .fault
= kvm_vcpu_fault
,
4147 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
4149 struct kvm_vcpu
*vcpu
= file
->private_data
;
4150 unsigned long pages
= vma_pages(vma
);
4152 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
4153 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
4154 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
4157 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
4161 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
4163 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4165 kvm_put_kvm(vcpu
->kvm
);
4169 static struct file_operations kvm_vcpu_fops
= {
4170 .release
= kvm_vcpu_release
,
4171 .unlocked_ioctl
= kvm_vcpu_ioctl
,
4172 .mmap
= kvm_vcpu_mmap
,
4173 .llseek
= noop_llseek
,
4174 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
4178 * Allocates an inode for the vcpu.
4180 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
4182 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
4184 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
4185 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
4188 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
4189 static int vcpu_get_pid(void *data
, u64
*val
)
4191 struct kvm_vcpu
*vcpu
= data
;
4194 *val
= pid_nr(rcu_dereference(vcpu
->pid
));
4199 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops
, vcpu_get_pid
, NULL
, "%llu\n");
4201 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
4203 struct dentry
*debugfs_dentry
;
4204 char dir_name
[ITOA_MAX_LEN
* 2];
4206 if (!debugfs_initialized())
4209 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
4210 debugfs_dentry
= debugfs_create_dir(dir_name
,
4211 vcpu
->kvm
->debugfs_dentry
);
4212 debugfs_create_file("pid", 0444, debugfs_dentry
, vcpu
,
4213 &vcpu_get_pid_fops
);
4215 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
4220 * Creates some virtual cpus. Good luck creating more than one.
4222 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
4225 struct kvm_vcpu
*vcpu
;
4228 if (id
>= KVM_MAX_VCPU_IDS
)
4231 mutex_lock(&kvm
->lock
);
4232 if (kvm
->created_vcpus
>= kvm
->max_vcpus
) {
4233 mutex_unlock(&kvm
->lock
);
4237 r
= kvm_arch_vcpu_precreate(kvm
, id
);
4239 mutex_unlock(&kvm
->lock
);
4243 kvm
->created_vcpus
++;
4244 mutex_unlock(&kvm
->lock
);
4246 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL_ACCOUNT
);
4249 goto vcpu_decrement
;
4252 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
4253 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
4258 vcpu
->run
= page_address(page
);
4260 kvm_vcpu_init(vcpu
, kvm
, id
);
4262 r
= kvm_arch_vcpu_create(vcpu
);
4264 goto vcpu_free_run_page
;
4266 if (kvm
->dirty_ring_size
) {
4267 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
4268 id
, kvm
->dirty_ring_size
);
4270 goto arch_vcpu_destroy
;
4273 mutex_lock(&kvm
->lock
);
4275 #ifdef CONFIG_LOCKDEP
4276 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
4277 mutex_lock(&vcpu
->mutex
);
4278 mutex_unlock(&vcpu
->mutex
);
4281 if (kvm_get_vcpu_by_id(kvm
, id
)) {
4283 goto unlock_vcpu_destroy
;
4286 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
4287 r
= xa_reserve(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, GFP_KERNEL_ACCOUNT
);
4289 goto unlock_vcpu_destroy
;
4291 /* Now it's all set up, let userspace reach it */
4293 r
= create_vcpu_fd(vcpu
);
4295 goto kvm_put_xa_release
;
4297 if (KVM_BUG_ON(xa_store(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, vcpu
, 0), kvm
)) {
4299 goto kvm_put_xa_release
;
4303 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4304 * pointer before kvm->online_vcpu's incremented value.
4307 atomic_inc(&kvm
->online_vcpus
);
4309 mutex_unlock(&kvm
->lock
);
4310 kvm_arch_vcpu_postcreate(vcpu
);
4311 kvm_create_vcpu_debugfs(vcpu
);
4315 kvm_put_kvm_no_destroy(kvm
);
4316 xa_release(&kvm
->vcpu_array
, vcpu
->vcpu_idx
);
4317 unlock_vcpu_destroy
:
4318 mutex_unlock(&kvm
->lock
);
4319 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
4321 kvm_arch_vcpu_destroy(vcpu
);
4323 free_page((unsigned long)vcpu
->run
);
4325 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
4327 mutex_lock(&kvm
->lock
);
4328 kvm
->created_vcpus
--;
4329 mutex_unlock(&kvm
->lock
);
4333 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
4336 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
4337 vcpu
->sigset_active
= 1;
4338 vcpu
->sigset
= *sigset
;
4340 vcpu
->sigset_active
= 0;
4344 static ssize_t
kvm_vcpu_stats_read(struct file
*file
, char __user
*user_buffer
,
4345 size_t size
, loff_t
*offset
)
4347 struct kvm_vcpu
*vcpu
= file
->private_data
;
4349 return kvm_stats_read(vcpu
->stats_id
, &kvm_vcpu_stats_header
,
4350 &kvm_vcpu_stats_desc
[0], &vcpu
->stat
,
4351 sizeof(vcpu
->stat
), user_buffer
, size
, offset
);
4354 static int kvm_vcpu_stats_release(struct inode
*inode
, struct file
*file
)
4356 struct kvm_vcpu
*vcpu
= file
->private_data
;
4358 kvm_put_kvm(vcpu
->kvm
);
4362 static const struct file_operations kvm_vcpu_stats_fops
= {
4363 .owner
= THIS_MODULE
,
4364 .read
= kvm_vcpu_stats_read
,
4365 .release
= kvm_vcpu_stats_release
,
4366 .llseek
= noop_llseek
,
4369 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu
*vcpu
)
4373 char name
[15 + ITOA_MAX_LEN
+ 1];
4375 snprintf(name
, sizeof(name
), "kvm-vcpu-stats:%d", vcpu
->vcpu_id
);
4377 fd
= get_unused_fd_flags(O_CLOEXEC
);
4381 file
= anon_inode_getfile(name
, &kvm_vcpu_stats_fops
, vcpu
, O_RDONLY
);
4384 return PTR_ERR(file
);
4387 kvm_get_kvm(vcpu
->kvm
);
4389 file
->f_mode
|= FMODE_PREAD
;
4390 fd_install(fd
, file
);
4395 static long kvm_vcpu_ioctl(struct file
*filp
,
4396 unsigned int ioctl
, unsigned long arg
)
4398 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4399 void __user
*argp
= (void __user
*)arg
;
4401 struct kvm_fpu
*fpu
= NULL
;
4402 struct kvm_sregs
*kvm_sregs
= NULL
;
4404 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4407 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
4411 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4412 * execution; mutex_lock() would break them.
4414 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
4415 if (r
!= -ENOIOCTLCMD
)
4418 if (mutex_lock_killable(&vcpu
->mutex
))
4426 oldpid
= rcu_access_pointer(vcpu
->pid
);
4427 if (unlikely(oldpid
!= task_pid(current
))) {
4428 /* The thread running this VCPU changed. */
4431 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
4435 newpid
= get_task_pid(current
, PIDTYPE_PID
);
4436 rcu_assign_pointer(vcpu
->pid
, newpid
);
4441 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
4442 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
4445 case KVM_GET_REGS
: {
4446 struct kvm_regs
*kvm_regs
;
4449 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
4452 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
4456 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
4463 case KVM_SET_REGS
: {
4464 struct kvm_regs
*kvm_regs
;
4466 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
4467 if (IS_ERR(kvm_regs
)) {
4468 r
= PTR_ERR(kvm_regs
);
4471 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
4475 case KVM_GET_SREGS
: {
4476 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
4477 GFP_KERNEL_ACCOUNT
);
4481 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
4485 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
4490 case KVM_SET_SREGS
: {
4491 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
4492 if (IS_ERR(kvm_sregs
)) {
4493 r
= PTR_ERR(kvm_sregs
);
4497 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
4500 case KVM_GET_MP_STATE
: {
4501 struct kvm_mp_state mp_state
;
4503 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
4507 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
4512 case KVM_SET_MP_STATE
: {
4513 struct kvm_mp_state mp_state
;
4516 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
4518 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
4521 case KVM_TRANSLATE
: {
4522 struct kvm_translation tr
;
4525 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
4527 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
4531 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
4536 case KVM_SET_GUEST_DEBUG
: {
4537 struct kvm_guest_debug dbg
;
4540 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
4542 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
4545 case KVM_SET_SIGNAL_MASK
: {
4546 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4547 struct kvm_signal_mask kvm_sigmask
;
4548 sigset_t sigset
, *p
;
4553 if (copy_from_user(&kvm_sigmask
, argp
,
4554 sizeof(kvm_sigmask
)))
4557 if (kvm_sigmask
.len
!= sizeof(sigset
))
4560 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
4565 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
4569 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
4573 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
4577 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
4583 fpu
= memdup_user(argp
, sizeof(*fpu
));
4589 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
4592 case KVM_GET_STATS_FD
: {
4593 r
= kvm_vcpu_ioctl_get_stats_fd(vcpu
);
4597 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
4600 mutex_unlock(&vcpu
->mutex
);
4606 #ifdef CONFIG_KVM_COMPAT
4607 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
4608 unsigned int ioctl
, unsigned long arg
)
4610 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4611 void __user
*argp
= compat_ptr(arg
);
4614 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4618 case KVM_SET_SIGNAL_MASK
: {
4619 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4620 struct kvm_signal_mask kvm_sigmask
;
4625 if (copy_from_user(&kvm_sigmask
, argp
,
4626 sizeof(kvm_sigmask
)))
4629 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
4632 if (get_compat_sigset(&sigset
,
4633 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
4635 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
4637 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
4641 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
4649 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
4651 struct kvm_device
*dev
= filp
->private_data
;
4654 return dev
->ops
->mmap(dev
, vma
);
4659 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
4660 int (*accessor
)(struct kvm_device
*dev
,
4661 struct kvm_device_attr
*attr
),
4664 struct kvm_device_attr attr
;
4669 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
4672 return accessor(dev
, &attr
);
4675 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
4678 struct kvm_device
*dev
= filp
->private_data
;
4680 if (dev
->kvm
->mm
!= current
->mm
|| dev
->kvm
->vm_dead
)
4684 case KVM_SET_DEVICE_ATTR
:
4685 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
4686 case KVM_GET_DEVICE_ATTR
:
4687 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
4688 case KVM_HAS_DEVICE_ATTR
:
4689 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
4691 if (dev
->ops
->ioctl
)
4692 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
4698 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
4700 struct kvm_device
*dev
= filp
->private_data
;
4701 struct kvm
*kvm
= dev
->kvm
;
4703 if (dev
->ops
->release
) {
4704 mutex_lock(&kvm
->lock
);
4705 list_del(&dev
->vm_node
);
4706 dev
->ops
->release(dev
);
4707 mutex_unlock(&kvm
->lock
);
4714 static struct file_operations kvm_device_fops
= {
4715 .unlocked_ioctl
= kvm_device_ioctl
,
4716 .release
= kvm_device_release
,
4717 KVM_COMPAT(kvm_device_ioctl
),
4718 .mmap
= kvm_device_mmap
,
4721 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
4723 if (filp
->f_op
!= &kvm_device_fops
)
4726 return filp
->private_data
;
4729 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
4730 #ifdef CONFIG_KVM_MPIC
4731 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
4732 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
4736 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
4738 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
4741 if (kvm_device_ops_table
[type
] != NULL
)
4744 kvm_device_ops_table
[type
] = ops
;
4748 void kvm_unregister_device_ops(u32 type
)
4750 if (kvm_device_ops_table
[type
] != NULL
)
4751 kvm_device_ops_table
[type
] = NULL
;
4754 static int kvm_ioctl_create_device(struct kvm
*kvm
,
4755 struct kvm_create_device
*cd
)
4757 const struct kvm_device_ops
*ops
;
4758 struct kvm_device
*dev
;
4759 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
4763 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
4766 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
4767 ops
= kvm_device_ops_table
[type
];
4774 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
4781 mutex_lock(&kvm
->lock
);
4782 ret
= ops
->create(dev
, type
);
4784 mutex_unlock(&kvm
->lock
);
4788 list_add(&dev
->vm_node
, &kvm
->devices
);
4789 mutex_unlock(&kvm
->lock
);
4795 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
4797 kvm_put_kvm_no_destroy(kvm
);
4798 mutex_lock(&kvm
->lock
);
4799 list_del(&dev
->vm_node
);
4802 mutex_unlock(&kvm
->lock
);
4812 static int kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
4815 case KVM_CAP_USER_MEMORY
:
4816 case KVM_CAP_USER_MEMORY2
:
4817 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
4818 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
4819 case KVM_CAP_INTERNAL_ERROR_DATA
:
4820 #ifdef CONFIG_HAVE_KVM_MSI
4821 case KVM_CAP_SIGNAL_MSI
:
4823 #ifdef CONFIG_HAVE_KVM_IRQCHIP
4826 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
4827 case KVM_CAP_CHECK_EXTENSION_VM
:
4828 case KVM_CAP_ENABLE_CAP_VM
:
4829 case KVM_CAP_HALT_POLL
:
4831 #ifdef CONFIG_KVM_MMIO
4832 case KVM_CAP_COALESCED_MMIO
:
4833 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
4834 case KVM_CAP_COALESCED_PIO
:
4837 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4838 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
4839 return KVM_DIRTY_LOG_MANUAL_CAPS
;
4841 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4842 case KVM_CAP_IRQ_ROUTING
:
4843 return KVM_MAX_IRQ_ROUTES
;
4845 #if KVM_MAX_NR_ADDRESS_SPACES > 1
4846 case KVM_CAP_MULTI_ADDRESS_SPACE
:
4848 return kvm_arch_nr_memslot_as_ids(kvm
);
4849 return KVM_MAX_NR_ADDRESS_SPACES
;
4851 case KVM_CAP_NR_MEMSLOTS
:
4852 return KVM_USER_MEM_SLOTS
;
4853 case KVM_CAP_DIRTY_LOG_RING
:
4854 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4855 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4859 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4860 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4861 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4865 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4866 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
:
4868 case KVM_CAP_BINARY_STATS_FD
:
4869 case KVM_CAP_SYSTEM_EVENT_DATA
:
4871 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
4872 case KVM_CAP_MEMORY_ATTRIBUTES
:
4873 return kvm_supported_mem_attributes(kvm
);
4875 #ifdef CONFIG_KVM_PRIVATE_MEM
4876 case KVM_CAP_GUEST_MEMFD
:
4877 return !kvm
|| kvm_arch_has_private_mem(kvm
);
4882 return kvm_vm_ioctl_check_extension(kvm
, arg
);
4885 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
4889 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
4892 /* the size should be power of 2 */
4893 if (!size
|| (size
& (size
- 1)))
4896 /* Should be bigger to keep the reserved entries, or a page */
4897 if (size
< kvm_dirty_ring_get_rsvd_entries() *
4898 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
4901 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
4902 sizeof(struct kvm_dirty_gfn
))
4905 /* We only allow it to set once */
4906 if (kvm
->dirty_ring_size
)
4909 mutex_lock(&kvm
->lock
);
4911 if (kvm
->created_vcpus
) {
4912 /* We don't allow to change this value after vcpu created */
4915 kvm
->dirty_ring_size
= size
;
4919 mutex_unlock(&kvm
->lock
);
4923 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
4926 struct kvm_vcpu
*vcpu
;
4929 if (!kvm
->dirty_ring_size
)
4932 mutex_lock(&kvm
->slots_lock
);
4934 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4935 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
4937 mutex_unlock(&kvm
->slots_lock
);
4940 kvm_flush_remote_tlbs(kvm
);
4945 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4946 struct kvm_enable_cap
*cap
)
4951 bool kvm_are_all_memslots_empty(struct kvm
*kvm
)
4955 lockdep_assert_held(&kvm
->slots_lock
);
4957 for (i
= 0; i
< kvm_arch_nr_memslot_as_ids(kvm
); i
++) {
4958 if (!kvm_memslots_empty(__kvm_memslots(kvm
, i
)))
4964 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty
);
4966 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
4967 struct kvm_enable_cap
*cap
)
4970 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4971 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
4972 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
4974 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
4975 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
4977 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
4979 kvm
->manual_dirty_log_protect
= cap
->args
[0];
4983 case KVM_CAP_HALT_POLL
: {
4984 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
4987 kvm
->max_halt_poll_ns
= cap
->args
[0];
4990 * Ensure kvm->override_halt_poll_ns does not become visible
4991 * before kvm->max_halt_poll_ns.
4993 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4996 kvm
->override_halt_poll_ns
= true;
5000 case KVM_CAP_DIRTY_LOG_RING
:
5001 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
5002 if (!kvm_vm_ioctl_check_extension_generic(kvm
, cap
->cap
))
5005 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
5006 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
: {
5009 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
) ||
5010 !kvm
->dirty_ring_size
|| cap
->flags
)
5013 mutex_lock(&kvm
->slots_lock
);
5016 * For simplicity, allow enabling ring+bitmap if and only if
5017 * there are no memslots, e.g. to ensure all memslots allocate
5018 * a bitmap after the capability is enabled.
5020 if (kvm_are_all_memslots_empty(kvm
)) {
5021 kvm
->dirty_ring_with_bitmap
= true;
5025 mutex_unlock(&kvm
->slots_lock
);
5030 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
5034 static ssize_t
kvm_vm_stats_read(struct file
*file
, char __user
*user_buffer
,
5035 size_t size
, loff_t
*offset
)
5037 struct kvm
*kvm
= file
->private_data
;
5039 return kvm_stats_read(kvm
->stats_id
, &kvm_vm_stats_header
,
5040 &kvm_vm_stats_desc
[0], &kvm
->stat
,
5041 sizeof(kvm
->stat
), user_buffer
, size
, offset
);
5044 static int kvm_vm_stats_release(struct inode
*inode
, struct file
*file
)
5046 struct kvm
*kvm
= file
->private_data
;
5052 static const struct file_operations kvm_vm_stats_fops
= {
5053 .owner
= THIS_MODULE
,
5054 .read
= kvm_vm_stats_read
,
5055 .release
= kvm_vm_stats_release
,
5056 .llseek
= noop_llseek
,
5059 static int kvm_vm_ioctl_get_stats_fd(struct kvm
*kvm
)
5064 fd
= get_unused_fd_flags(O_CLOEXEC
);
5068 file
= anon_inode_getfile("kvm-vm-stats",
5069 &kvm_vm_stats_fops
, kvm
, O_RDONLY
);
5072 return PTR_ERR(file
);
5077 file
->f_mode
|= FMODE_PREAD
;
5078 fd_install(fd
, file
);
5083 #define SANITY_CHECK_MEM_REGION_FIELD(field) \
5085 BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) != \
5086 offsetof(struct kvm_userspace_memory_region2, field)); \
5087 BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) != \
5088 sizeof_field(struct kvm_userspace_memory_region2, field)); \
5091 static long kvm_vm_ioctl(struct file
*filp
,
5092 unsigned int ioctl
, unsigned long arg
)
5094 struct kvm
*kvm
= filp
->private_data
;
5095 void __user
*argp
= (void __user
*)arg
;
5098 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5101 case KVM_CREATE_VCPU
:
5102 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
5104 case KVM_ENABLE_CAP
: {
5105 struct kvm_enable_cap cap
;
5108 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
5110 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
5113 case KVM_SET_USER_MEMORY_REGION2
:
5114 case KVM_SET_USER_MEMORY_REGION
: {
5115 struct kvm_userspace_memory_region2 mem
;
5118 if (ioctl
== KVM_SET_USER_MEMORY_REGION
) {
5120 * Fields beyond struct kvm_userspace_memory_region shouldn't be
5121 * accessed, but avoid leaking kernel memory in case of a bug.
5123 memset(&mem
, 0, sizeof(mem
));
5124 size
= sizeof(struct kvm_userspace_memory_region
);
5126 size
= sizeof(struct kvm_userspace_memory_region2
);
5129 /* Ensure the common parts of the two structs are identical. */
5130 SANITY_CHECK_MEM_REGION_FIELD(slot
);
5131 SANITY_CHECK_MEM_REGION_FIELD(flags
);
5132 SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr
);
5133 SANITY_CHECK_MEM_REGION_FIELD(memory_size
);
5134 SANITY_CHECK_MEM_REGION_FIELD(userspace_addr
);
5137 if (copy_from_user(&mem
, argp
, size
))
5141 if (ioctl
== KVM_SET_USER_MEMORY_REGION
&&
5142 (mem
.flags
& ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS
))
5145 r
= kvm_vm_ioctl_set_memory_region(kvm
, &mem
);
5148 case KVM_GET_DIRTY_LOG
: {
5149 struct kvm_dirty_log log
;
5152 if (copy_from_user(&log
, argp
, sizeof(log
)))
5154 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5157 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5158 case KVM_CLEAR_DIRTY_LOG
: {
5159 struct kvm_clear_dirty_log log
;
5162 if (copy_from_user(&log
, argp
, sizeof(log
)))
5164 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5168 #ifdef CONFIG_KVM_MMIO
5169 case KVM_REGISTER_COALESCED_MMIO
: {
5170 struct kvm_coalesced_mmio_zone zone
;
5173 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
5175 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
5178 case KVM_UNREGISTER_COALESCED_MMIO
: {
5179 struct kvm_coalesced_mmio_zone zone
;
5182 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
5184 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
5189 struct kvm_irqfd data
;
5192 if (copy_from_user(&data
, argp
, sizeof(data
)))
5194 r
= kvm_irqfd(kvm
, &data
);
5197 case KVM_IOEVENTFD
: {
5198 struct kvm_ioeventfd data
;
5201 if (copy_from_user(&data
, argp
, sizeof(data
)))
5203 r
= kvm_ioeventfd(kvm
, &data
);
5206 #ifdef CONFIG_HAVE_KVM_MSI
5207 case KVM_SIGNAL_MSI
: {
5211 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
5213 r
= kvm_send_userspace_msi(kvm
, &msi
);
5217 #ifdef __KVM_HAVE_IRQ_LINE
5218 case KVM_IRQ_LINE_STATUS
:
5219 case KVM_IRQ_LINE
: {
5220 struct kvm_irq_level irq_event
;
5223 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
5226 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
5227 ioctl
== KVM_IRQ_LINE_STATUS
);
5232 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
5233 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
5241 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
5242 case KVM_SET_GSI_ROUTING
: {
5243 struct kvm_irq_routing routing
;
5244 struct kvm_irq_routing __user
*urouting
;
5245 struct kvm_irq_routing_entry
*entries
= NULL
;
5248 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
5251 if (!kvm_arch_can_set_irq_routing(kvm
))
5253 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
5259 entries
= vmemdup_user(urouting
->entries
,
5260 array_size(sizeof(*entries
),
5262 if (IS_ERR(entries
)) {
5263 r
= PTR_ERR(entries
);
5267 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
5272 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
5273 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
5274 case KVM_SET_MEMORY_ATTRIBUTES
: {
5275 struct kvm_memory_attributes attrs
;
5278 if (copy_from_user(&attrs
, argp
, sizeof(attrs
)))
5281 r
= kvm_vm_ioctl_set_mem_attributes(kvm
, &attrs
);
5284 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
5285 case KVM_CREATE_DEVICE
: {
5286 struct kvm_create_device cd
;
5289 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
5292 r
= kvm_ioctl_create_device(kvm
, &cd
);
5297 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
5303 case KVM_CHECK_EXTENSION
:
5304 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
5306 case KVM_RESET_DIRTY_RINGS
:
5307 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
5309 case KVM_GET_STATS_FD
:
5310 r
= kvm_vm_ioctl_get_stats_fd(kvm
);
5312 #ifdef CONFIG_KVM_PRIVATE_MEM
5313 case KVM_CREATE_GUEST_MEMFD
: {
5314 struct kvm_create_guest_memfd guest_memfd
;
5317 if (copy_from_user(&guest_memfd
, argp
, sizeof(guest_memfd
)))
5320 r
= kvm_gmem_create(kvm
, &guest_memfd
);
5325 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
5331 #ifdef CONFIG_KVM_COMPAT
5332 struct compat_kvm_dirty_log
{
5336 compat_uptr_t dirty_bitmap
; /* one bit per page */
5341 struct compat_kvm_clear_dirty_log
{
5346 compat_uptr_t dirty_bitmap
; /* one bit per page */
5351 long __weak
kvm_arch_vm_compat_ioctl(struct file
*filp
, unsigned int ioctl
,
5357 static long kvm_vm_compat_ioctl(struct file
*filp
,
5358 unsigned int ioctl
, unsigned long arg
)
5360 struct kvm
*kvm
= filp
->private_data
;
5363 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5366 r
= kvm_arch_vm_compat_ioctl(filp
, ioctl
, arg
);
5371 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5372 case KVM_CLEAR_DIRTY_LOG
: {
5373 struct compat_kvm_clear_dirty_log compat_log
;
5374 struct kvm_clear_dirty_log log
;
5376 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5377 sizeof(compat_log
)))
5379 log
.slot
= compat_log
.slot
;
5380 log
.num_pages
= compat_log
.num_pages
;
5381 log
.first_page
= compat_log
.first_page
;
5382 log
.padding2
= compat_log
.padding2
;
5383 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5385 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5389 case KVM_GET_DIRTY_LOG
: {
5390 struct compat_kvm_dirty_log compat_log
;
5391 struct kvm_dirty_log log
;
5393 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5394 sizeof(compat_log
)))
5396 log
.slot
= compat_log
.slot
;
5397 log
.padding1
= compat_log
.padding1
;
5398 log
.padding2
= compat_log
.padding2
;
5399 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5401 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5405 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
5411 static struct file_operations kvm_vm_fops
= {
5412 .release
= kvm_vm_release
,
5413 .unlocked_ioctl
= kvm_vm_ioctl
,
5414 .llseek
= noop_llseek
,
5415 KVM_COMPAT(kvm_vm_compat_ioctl
),
5418 bool file_is_kvm(struct file
*file
)
5420 return file
&& file
->f_op
== &kvm_vm_fops
;
5422 EXPORT_SYMBOL_GPL(file_is_kvm
);
5424 static int kvm_dev_ioctl_create_vm(unsigned long type
)
5426 char fdname
[ITOA_MAX_LEN
+ 1];
5431 fd
= get_unused_fd_flags(O_CLOEXEC
);
5435 snprintf(fdname
, sizeof(fdname
), "%d", fd
);
5437 kvm
= kvm_create_vm(type
, fdname
);
5443 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
5450 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5451 * already set, with ->release() being kvm_vm_release(). In error
5452 * cases it will be called by the final fput(file) and will take
5453 * care of doing kvm_put_kvm(kvm).
5455 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
5457 fd_install(fd
, file
);
5467 static long kvm_dev_ioctl(struct file
*filp
,
5468 unsigned int ioctl
, unsigned long arg
)
5473 case KVM_GET_API_VERSION
:
5476 r
= KVM_API_VERSION
;
5479 r
= kvm_dev_ioctl_create_vm(arg
);
5481 case KVM_CHECK_EXTENSION
:
5482 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
5484 case KVM_GET_VCPU_MMAP_SIZE
:
5487 r
= PAGE_SIZE
; /* struct kvm_run */
5489 r
+= PAGE_SIZE
; /* pio data page */
5491 #ifdef CONFIG_KVM_MMIO
5492 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
5496 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
5502 static struct file_operations kvm_chardev_ops
= {
5503 .unlocked_ioctl
= kvm_dev_ioctl
,
5504 .llseek
= noop_llseek
,
5505 KVM_COMPAT(kvm_dev_ioctl
),
5508 static struct miscdevice kvm_dev
= {
5514 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5515 __visible
bool kvm_rebooting
;
5516 EXPORT_SYMBOL_GPL(kvm_rebooting
);
5518 static DEFINE_PER_CPU(bool, hardware_enabled
);
5519 static int kvm_usage_count
;
5521 static int __hardware_enable_nolock(void)
5523 if (__this_cpu_read(hardware_enabled
))
5526 if (kvm_arch_hardware_enable()) {
5527 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5528 raw_smp_processor_id());
5532 __this_cpu_write(hardware_enabled
, true);
5536 static void hardware_enable_nolock(void *failed
)
5538 if (__hardware_enable_nolock())
5542 static int kvm_online_cpu(unsigned int cpu
)
5547 * Abort the CPU online process if hardware virtualization cannot
5548 * be enabled. Otherwise running VMs would encounter unrecoverable
5549 * errors when scheduled to this CPU.
5551 mutex_lock(&kvm_lock
);
5552 if (kvm_usage_count
)
5553 ret
= __hardware_enable_nolock();
5554 mutex_unlock(&kvm_lock
);
5558 static void hardware_disable_nolock(void *junk
)
5561 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5562 * hardware, not just CPUs that successfully enabled hardware!
5564 if (!__this_cpu_read(hardware_enabled
))
5567 kvm_arch_hardware_disable();
5569 __this_cpu_write(hardware_enabled
, false);
5572 static int kvm_offline_cpu(unsigned int cpu
)
5574 mutex_lock(&kvm_lock
);
5575 if (kvm_usage_count
)
5576 hardware_disable_nolock(NULL
);
5577 mutex_unlock(&kvm_lock
);
5581 static void hardware_disable_all_nolock(void)
5583 BUG_ON(!kvm_usage_count
);
5586 if (!kvm_usage_count
)
5587 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5590 static void hardware_disable_all(void)
5593 mutex_lock(&kvm_lock
);
5594 hardware_disable_all_nolock();
5595 mutex_unlock(&kvm_lock
);
5599 static int hardware_enable_all(void)
5601 atomic_t failed
= ATOMIC_INIT(0);
5605 * Do not enable hardware virtualization if the system is going down.
5606 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5607 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5608 * after kvm_reboot() is called. Note, this relies on system_state
5609 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5610 * hook instead of registering a dedicated reboot notifier (the latter
5611 * runs before system_state is updated).
5613 if (system_state
== SYSTEM_HALT
|| system_state
== SYSTEM_POWER_OFF
||
5614 system_state
== SYSTEM_RESTART
)
5618 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5619 * is called, and so on_each_cpu() between them includes the CPU that
5620 * is being onlined. As a result, hardware_enable_nolock() may get
5621 * invoked before kvm_online_cpu(), which also enables hardware if the
5622 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5623 * enable hardware multiple times.
5626 mutex_lock(&kvm_lock
);
5631 if (kvm_usage_count
== 1) {
5632 on_each_cpu(hardware_enable_nolock
, &failed
, 1);
5634 if (atomic_read(&failed
)) {
5635 hardware_disable_all_nolock();
5640 mutex_unlock(&kvm_lock
);
5646 static void kvm_shutdown(void)
5649 * Disable hardware virtualization and set kvm_rebooting to indicate
5650 * that KVM has asynchronously disabled hardware virtualization, i.e.
5651 * that relevant errors and exceptions aren't entirely unexpected.
5652 * Some flavors of hardware virtualization need to be disabled before
5653 * transferring control to firmware (to perform shutdown/reboot), e.g.
5654 * on x86, virtualization can block INIT interrupts, which are used by
5655 * firmware to pull APs back under firmware control. Note, this path
5656 * is used for both shutdown and reboot scenarios, i.e. neither name is
5657 * 100% comprehensive.
5659 pr_info("kvm: exiting hardware virtualization\n");
5660 kvm_rebooting
= true;
5661 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5664 static int kvm_suspend(void)
5667 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5668 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5669 * is stable. Assert that kvm_lock is not held to ensure the system
5670 * isn't suspended while KVM is enabling hardware. Hardware enabling
5671 * can be preempted, but the task cannot be frozen until it has dropped
5672 * all locks (userspace tasks are frozen via a fake signal).
5674 lockdep_assert_not_held(&kvm_lock
);
5675 lockdep_assert_irqs_disabled();
5677 if (kvm_usage_count
)
5678 hardware_disable_nolock(NULL
);
5682 static void kvm_resume(void)
5684 lockdep_assert_not_held(&kvm_lock
);
5685 lockdep_assert_irqs_disabled();
5687 if (kvm_usage_count
)
5688 WARN_ON_ONCE(__hardware_enable_nolock());
5691 static struct syscore_ops kvm_syscore_ops
= {
5692 .suspend
= kvm_suspend
,
5693 .resume
= kvm_resume
,
5694 .shutdown
= kvm_shutdown
,
5696 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5697 static int hardware_enable_all(void)
5702 static void hardware_disable_all(void)
5706 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5708 static void kvm_iodevice_destructor(struct kvm_io_device
*dev
)
5710 if (dev
->ops
->destructor
)
5711 dev
->ops
->destructor(dev
);
5714 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
5718 for (i
= 0; i
< bus
->dev_count
; i
++) {
5719 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
5721 kvm_iodevice_destructor(pos
);
5726 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
5727 const struct kvm_io_range
*r2
)
5729 gpa_t addr1
= r1
->addr
;
5730 gpa_t addr2
= r2
->addr
;
5735 /* If r2->len == 0, match the exact address. If r2->len != 0,
5736 * accept any overlapping write. Any order is acceptable for
5737 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5738 * we process all of them.
5751 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
5753 return kvm_io_bus_cmp(p1
, p2
);
5756 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
5757 gpa_t addr
, int len
)
5759 struct kvm_io_range
*range
, key
;
5762 key
= (struct kvm_io_range
) {
5767 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
5768 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
5772 off
= range
- bus
->range
;
5774 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
5780 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5781 struct kvm_io_range
*range
, const void *val
)
5785 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5789 while (idx
< bus
->dev_count
&&
5790 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5791 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5800 /* kvm_io_bus_write - called under kvm->slots_lock */
5801 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5802 int len
, const void *val
)
5804 struct kvm_io_bus
*bus
;
5805 struct kvm_io_range range
;
5808 range
= (struct kvm_io_range
) {
5813 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5816 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5817 return r
< 0 ? r
: 0;
5819 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
5821 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5822 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
5823 gpa_t addr
, int len
, const void *val
, long cookie
)
5825 struct kvm_io_bus
*bus
;
5826 struct kvm_io_range range
;
5828 range
= (struct kvm_io_range
) {
5833 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5837 /* First try the device referenced by cookie. */
5838 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
5839 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
5840 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
5845 * cookie contained garbage; fall back to search and return the
5846 * correct cookie value.
5848 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5851 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5852 struct kvm_io_range
*range
, void *val
)
5856 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5860 while (idx
< bus
->dev_count
&&
5861 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5862 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5871 /* kvm_io_bus_read - called under kvm->slots_lock */
5872 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5875 struct kvm_io_bus
*bus
;
5876 struct kvm_io_range range
;
5879 range
= (struct kvm_io_range
) {
5884 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5887 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
5888 return r
< 0 ? r
: 0;
5891 /* Caller must hold slots_lock. */
5892 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
5893 int len
, struct kvm_io_device
*dev
)
5896 struct kvm_io_bus
*new_bus
, *bus
;
5897 struct kvm_io_range range
;
5899 bus
= kvm_get_bus(kvm
, bus_idx
);
5903 /* exclude ioeventfd which is limited by maximum fd */
5904 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
5907 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
5908 GFP_KERNEL_ACCOUNT
);
5912 range
= (struct kvm_io_range
) {
5918 for (i
= 0; i
< bus
->dev_count
; i
++)
5919 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
5922 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
5923 new_bus
->dev_count
++;
5924 new_bus
->range
[i
] = range
;
5925 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
5926 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
5927 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5928 synchronize_srcu_expedited(&kvm
->srcu
);
5934 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5935 struct kvm_io_device
*dev
)
5938 struct kvm_io_bus
*new_bus
, *bus
;
5940 lockdep_assert_held(&kvm
->slots_lock
);
5942 bus
= kvm_get_bus(kvm
, bus_idx
);
5946 for (i
= 0; i
< bus
->dev_count
; i
++) {
5947 if (bus
->range
[i
].dev
== dev
) {
5952 if (i
== bus
->dev_count
)
5955 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
5956 GFP_KERNEL_ACCOUNT
);
5958 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
5959 new_bus
->dev_count
--;
5960 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
5961 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
5964 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5965 synchronize_srcu_expedited(&kvm
->srcu
);
5968 * If NULL bus is installed, destroy the old bus, including all the
5969 * attached devices. Otherwise, destroy the caller's device only.
5972 pr_err("kvm: failed to shrink bus, removing it completely\n");
5973 kvm_io_bus_destroy(bus
);
5977 kvm_iodevice_destructor(dev
);
5982 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5985 struct kvm_io_bus
*bus
;
5986 int dev_idx
, srcu_idx
;
5987 struct kvm_io_device
*iodev
= NULL
;
5989 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5991 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
5995 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
5999 iodev
= bus
->range
[dev_idx
].dev
;
6002 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
6006 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
6008 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
6009 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
6013 struct kvm_stat_data
*stat_data
= inode
->i_private
;
6016 * The debugfs files are a reference to the kvm struct which
6017 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
6018 * avoids the race between open and the removal of the debugfs directory.
6020 if (!kvm_get_kvm_safe(stat_data
->kvm
))
6023 ret
= simple_attr_open(inode
, file
, get
,
6024 kvm_stats_debugfs_mode(stat_data
->desc
) & 0222
6027 kvm_put_kvm(stat_data
->kvm
);
6032 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
6034 struct kvm_stat_data
*stat_data
= inode
->i_private
;
6036 simple_attr_release(inode
, file
);
6037 kvm_put_kvm(stat_data
->kvm
);
6042 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
6044 *val
= *(u64
*)((void *)(&kvm
->stat
) + offset
);
6049 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
6051 *(u64
*)((void *)(&kvm
->stat
) + offset
) = 0;
6056 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
6059 struct kvm_vcpu
*vcpu
;
6063 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6064 *val
+= *(u64
*)((void *)(&vcpu
->stat
) + offset
);
6069 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
6072 struct kvm_vcpu
*vcpu
;
6074 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6075 *(u64
*)((void *)(&vcpu
->stat
) + offset
) = 0;
6080 static int kvm_stat_data_get(void *data
, u64
*val
)
6083 struct kvm_stat_data
*stat_data
= data
;
6085 switch (stat_data
->kind
) {
6087 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
6088 stat_data
->desc
->desc
.offset
, val
);
6091 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
6092 stat_data
->desc
->desc
.offset
, val
);
6099 static int kvm_stat_data_clear(void *data
, u64 val
)
6102 struct kvm_stat_data
*stat_data
= data
;
6107 switch (stat_data
->kind
) {
6109 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
6110 stat_data
->desc
->desc
.offset
);
6113 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
6114 stat_data
->desc
->desc
.offset
);
6121 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
6123 __simple_attr_check_format("%llu\n", 0ull);
6124 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
6125 kvm_stat_data_clear
, "%llu\n");
6128 static const struct file_operations stat_fops_per_vm
= {
6129 .owner
= THIS_MODULE
,
6130 .open
= kvm_stat_data_open
,
6131 .release
= kvm_debugfs_release
,
6132 .read
= simple_attr_read
,
6133 .write
= simple_attr_write
,
6134 .llseek
= no_llseek
,
6137 static int vm_stat_get(void *_offset
, u64
*val
)
6139 unsigned offset
= (long)_offset
;
6144 mutex_lock(&kvm_lock
);
6145 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6146 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
6149 mutex_unlock(&kvm_lock
);
6153 static int vm_stat_clear(void *_offset
, u64 val
)
6155 unsigned offset
= (long)_offset
;
6161 mutex_lock(&kvm_lock
);
6162 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6163 kvm_clear_stat_per_vm(kvm
, offset
);
6165 mutex_unlock(&kvm_lock
);
6170 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
6171 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops
, vm_stat_get
, NULL
, "%llu\n");
6173 static int vcpu_stat_get(void *_offset
, u64
*val
)
6175 unsigned offset
= (long)_offset
;
6180 mutex_lock(&kvm_lock
);
6181 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6182 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
6185 mutex_unlock(&kvm_lock
);
6189 static int vcpu_stat_clear(void *_offset
, u64 val
)
6191 unsigned offset
= (long)_offset
;
6197 mutex_lock(&kvm_lock
);
6198 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6199 kvm_clear_stat_per_vcpu(kvm
, offset
);
6201 mutex_unlock(&kvm_lock
);
6206 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
6208 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops
, vcpu_stat_get
, NULL
, "%llu\n");
6210 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
6212 struct kobj_uevent_env
*env
;
6213 unsigned long long created
, active
;
6215 if (!kvm_dev
.this_device
|| !kvm
)
6218 mutex_lock(&kvm_lock
);
6219 if (type
== KVM_EVENT_CREATE_VM
) {
6220 kvm_createvm_count
++;
6222 } else if (type
== KVM_EVENT_DESTROY_VM
) {
6225 created
= kvm_createvm_count
;
6226 active
= kvm_active_vms
;
6227 mutex_unlock(&kvm_lock
);
6229 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
6233 add_uevent_var(env
, "CREATED=%llu", created
);
6234 add_uevent_var(env
, "COUNT=%llu", active
);
6236 if (type
== KVM_EVENT_CREATE_VM
) {
6237 add_uevent_var(env
, "EVENT=create");
6238 kvm
->userspace_pid
= task_pid_nr(current
);
6239 } else if (type
== KVM_EVENT_DESTROY_VM
) {
6240 add_uevent_var(env
, "EVENT=destroy");
6242 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
6244 if (!IS_ERR(kvm
->debugfs_dentry
)) {
6245 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
6248 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
6250 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
6254 /* no need for checks, since we are adding at most only 5 keys */
6255 env
->envp
[env
->envp_idx
++] = NULL
;
6256 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
6260 static void kvm_init_debug(void)
6262 const struct file_operations
*fops
;
6263 const struct _kvm_stats_desc
*pdesc
;
6266 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
6268 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
6269 pdesc
= &kvm_vm_stats_desc
[i
];
6270 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
6271 fops
= &vm_stat_fops
;
6273 fops
= &vm_stat_readonly_fops
;
6274 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
6276 (void *)(long)pdesc
->desc
.offset
, fops
);
6279 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
6280 pdesc
= &kvm_vcpu_stats_desc
[i
];
6281 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
6282 fops
= &vcpu_stat_fops
;
6284 fops
= &vcpu_stat_readonly_fops
;
6285 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
6287 (void *)(long)pdesc
->desc
.offset
, fops
);
6292 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
6294 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
6297 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
6299 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
6301 WRITE_ONCE(vcpu
->preempted
, false);
6302 WRITE_ONCE(vcpu
->ready
, false);
6304 __this_cpu_write(kvm_running_vcpu
, vcpu
);
6305 kvm_arch_sched_in(vcpu
, cpu
);
6306 kvm_arch_vcpu_load(vcpu
, cpu
);
6309 static void kvm_sched_out(struct preempt_notifier
*pn
,
6310 struct task_struct
*next
)
6312 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
6314 if (current
->on_rq
) {
6315 WRITE_ONCE(vcpu
->preempted
, true);
6316 WRITE_ONCE(vcpu
->ready
, true);
6318 kvm_arch_vcpu_put(vcpu
);
6319 __this_cpu_write(kvm_running_vcpu
, NULL
);
6323 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
6325 * We can disable preemption locally around accessing the per-CPU variable,
6326 * and use the resolved vcpu pointer after enabling preemption again,
6327 * because even if the current thread is migrated to another CPU, reading
6328 * the per-CPU value later will give us the same value as we update the
6329 * per-CPU variable in the preempt notifier handlers.
6331 struct kvm_vcpu
*kvm_get_running_vcpu(void)
6333 struct kvm_vcpu
*vcpu
;
6336 vcpu
= __this_cpu_read(kvm_running_vcpu
);
6341 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
6344 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
6346 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
6348 return &kvm_running_vcpu
;
6351 #ifdef CONFIG_GUEST_PERF_EVENTS
6352 static unsigned int kvm_guest_state(void)
6354 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6357 if (!kvm_arch_pmi_in_guest(vcpu
))
6360 state
= PERF_GUEST_ACTIVE
;
6361 if (!kvm_arch_vcpu_in_kernel(vcpu
))
6362 state
|= PERF_GUEST_USER
;
6367 static unsigned long kvm_guest_get_ip(void)
6369 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6371 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6372 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu
)))
6375 return kvm_arch_vcpu_get_ip(vcpu
);
6378 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6379 .state
= kvm_guest_state
,
6380 .get_ip
= kvm_guest_get_ip
,
6381 .handle_intel_pt_intr
= NULL
,
6384 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler
)(void))
6386 kvm_guest_cbs
.handle_intel_pt_intr
= pt_intr_handler
;
6387 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6389 void kvm_unregister_perf_callbacks(void)
6391 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6395 int kvm_init(unsigned vcpu_size
, unsigned vcpu_align
, struct module
*module
)
6400 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6401 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE
, "kvm/cpu:online",
6402 kvm_online_cpu
, kvm_offline_cpu
);
6406 register_syscore_ops(&kvm_syscore_ops
);
6409 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6411 vcpu_align
= __alignof__(struct kvm_vcpu
);
6413 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
6415 offsetof(struct kvm_vcpu
, arch
),
6416 offsetofend(struct kvm_vcpu
, stats_id
)
6417 - offsetof(struct kvm_vcpu
, arch
),
6419 if (!kvm_vcpu_cache
) {
6421 goto err_vcpu_cache
;
6424 for_each_possible_cpu(cpu
) {
6425 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask
, cpu
),
6426 GFP_KERNEL
, cpu_to_node(cpu
))) {
6428 goto err_cpu_kick_mask
;
6432 r
= kvm_irqfd_init();
6436 r
= kvm_async_pf_init();
6440 kvm_chardev_ops
.owner
= module
;
6441 kvm_vm_fops
.owner
= module
;
6442 kvm_vcpu_fops
.owner
= module
;
6443 kvm_device_fops
.owner
= module
;
6445 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
6446 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
6450 r
= kvm_vfio_ops_init();
6451 if (WARN_ON_ONCE(r
))
6454 kvm_gmem_init(module
);
6457 * Registration _must_ be the very last thing done, as this exposes
6458 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6460 r
= misc_register(&kvm_dev
);
6462 pr_err("kvm: misc device register failed\n");
6469 kvm_vfio_ops_exit();
6471 kvm_async_pf_deinit();
6476 for_each_possible_cpu(cpu
)
6477 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6478 kmem_cache_destroy(kvm_vcpu_cache
);
6480 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6481 unregister_syscore_ops(&kvm_syscore_ops
);
6482 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6486 EXPORT_SYMBOL_GPL(kvm_init
);
6493 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6494 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6495 * to KVM while the module is being stopped.
6497 misc_deregister(&kvm_dev
);
6499 debugfs_remove_recursive(kvm_debugfs_dir
);
6500 for_each_possible_cpu(cpu
)
6501 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6502 kmem_cache_destroy(kvm_vcpu_cache
);
6503 kvm_vfio_ops_exit();
6504 kvm_async_pf_deinit();
6505 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6506 unregister_syscore_ops(&kvm_syscore_ops
);
6507 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6511 EXPORT_SYMBOL_GPL(kvm_exit
);
6513 struct kvm_vm_worker_thread_context
{
6515 struct task_struct
*parent
;
6516 struct completion init_done
;
6517 kvm_vm_thread_fn_t thread_fn
;
6522 static int kvm_vm_worker_thread(void *context
)
6525 * The init_context is allocated on the stack of the parent thread, so
6526 * we have to locally copy anything that is needed beyond initialization
6528 struct kvm_vm_worker_thread_context
*init_context
= context
;
6529 struct task_struct
*parent
;
6530 struct kvm
*kvm
= init_context
->kvm
;
6531 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
6532 uintptr_t data
= init_context
->data
;
6535 err
= kthread_park(current
);
6536 /* kthread_park(current) is never supposed to return an error */
6541 err
= cgroup_attach_task_all(init_context
->parent
, current
);
6543 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6548 set_user_nice(current
, task_nice(init_context
->parent
));
6551 init_context
->err
= err
;
6552 complete(&init_context
->init_done
);
6553 init_context
= NULL
;
6558 /* Wait to be woken up by the spawner before proceeding. */
6561 if (!kthread_should_stop())
6562 err
= thread_fn(kvm
, data
);
6566 * Move kthread back to its original cgroup to prevent it lingering in
6567 * the cgroup of the VM process, after the latter finishes its
6570 * kthread_stop() waits on the 'exited' completion condition which is
6571 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6572 * kthread is removed from the cgroup in the cgroup_exit() which is
6573 * called after the exit_mm(). This causes the kthread_stop() to return
6574 * before the kthread actually quits the cgroup.
6577 parent
= rcu_dereference(current
->real_parent
);
6578 get_task_struct(parent
);
6580 cgroup_attach_task_all(parent
, current
);
6581 put_task_struct(parent
);
6586 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
6587 uintptr_t data
, const char *name
,
6588 struct task_struct
**thread_ptr
)
6590 struct kvm_vm_worker_thread_context init_context
= {};
6591 struct task_struct
*thread
;
6594 init_context
.kvm
= kvm
;
6595 init_context
.parent
= current
;
6596 init_context
.thread_fn
= thread_fn
;
6597 init_context
.data
= data
;
6598 init_completion(&init_context
.init_done
);
6600 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
6601 "%s-%d", name
, task_pid_nr(current
));
6603 return PTR_ERR(thread
);
6605 /* kthread_run is never supposed to return NULL */
6606 WARN_ON(thread
== NULL
);
6608 wait_for_completion(&init_context
.init_done
);
6610 if (!init_context
.err
)
6611 *thread_ptr
= thread
;
6613 return init_context
.err
;