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 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_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 (*hva_handler_t
)(struct kvm
*kvm
, struct kvm_gfn_range
*range
);
544 typedef void (*on_lock_fn_t
)(struct kvm
*kvm
, unsigned long start
,
547 typedef void (*on_unlock_fn_t
)(struct kvm
*kvm
);
549 struct kvm_hva_range
{
552 union kvm_mmu_notifier_arg arg
;
553 hva_handler_t handler
;
554 on_lock_fn_t on_lock
;
555 on_unlock_fn_t on_unlock
;
561 * Use a dedicated stub instead of NULL to indicate that there is no callback
562 * function/handler. The compiler technically can't guarantee that a real
563 * function will have a non-zero address, and so it will generate code to
564 * check for !NULL, whereas comparing against a stub will be elided at compile
565 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
567 static void kvm_null_fn(void)
571 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
573 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG
;
575 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
576 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
577 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
579 node = interval_tree_iter_next(node, start, last)) \
581 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
582 const struct kvm_hva_range
*range
)
584 bool ret
= false, locked
= false;
585 struct kvm_gfn_range gfn_range
;
586 struct kvm_memory_slot
*slot
;
587 struct kvm_memslots
*slots
;
590 if (WARN_ON_ONCE(range
->end
<= range
->start
))
593 /* A null handler is allowed if and only if on_lock() is provided. */
594 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range
->on_lock
) &&
595 IS_KVM_NULL_FN(range
->handler
)))
598 idx
= srcu_read_lock(&kvm
->srcu
);
600 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
601 struct interval_tree_node
*node
;
603 slots
= __kvm_memslots(kvm
, i
);
604 kvm_for_each_memslot_in_hva_range(node
, slots
,
605 range
->start
, range
->end
- 1) {
606 unsigned long hva_start
, hva_end
;
608 slot
= container_of(node
, struct kvm_memory_slot
, hva_node
[slots
->node_idx
]);
609 hva_start
= max(range
->start
, slot
->userspace_addr
);
610 hva_end
= min(range
->end
, slot
->userspace_addr
+
611 (slot
->npages
<< PAGE_SHIFT
));
614 * To optimize for the likely case where the address
615 * range is covered by zero or one memslots, don't
616 * bother making these conditional (to avoid writes on
617 * the second or later invocation of the handler).
619 gfn_range
.arg
= range
->arg
;
620 gfn_range
.may_block
= range
->may_block
;
623 * {gfn(page) | page intersects with [hva_start, hva_end)} =
624 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
626 gfn_range
.start
= hva_to_gfn_memslot(hva_start
, slot
);
627 gfn_range
.end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, slot
);
628 gfn_range
.slot
= slot
;
633 if (!IS_KVM_NULL_FN(range
->on_lock
))
634 range
->on_lock(kvm
, range
->start
, range
->end
);
635 if (IS_KVM_NULL_FN(range
->handler
))
638 ret
|= range
->handler(kvm
, &gfn_range
);
642 if (range
->flush_on_ret
&& ret
)
643 kvm_flush_remote_tlbs(kvm
);
647 if (!IS_KVM_NULL_FN(range
->on_unlock
))
648 range
->on_unlock(kvm
);
651 srcu_read_unlock(&kvm
->srcu
, idx
);
653 /* The notifiers are averse to booleans. :-( */
657 static __always_inline
int kvm_handle_hva_range(struct mmu_notifier
*mn
,
660 union kvm_mmu_notifier_arg arg
,
661 hva_handler_t handler
)
663 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
664 const struct kvm_hva_range range
= {
669 .on_lock
= (void *)kvm_null_fn
,
670 .on_unlock
= (void *)kvm_null_fn
,
671 .flush_on_ret
= true,
675 return __kvm_handle_hva_range(kvm
, &range
);
678 static __always_inline
int kvm_handle_hva_range_no_flush(struct mmu_notifier
*mn
,
681 hva_handler_t handler
)
683 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
684 const struct kvm_hva_range range
= {
688 .on_lock
= (void *)kvm_null_fn
,
689 .on_unlock
= (void *)kvm_null_fn
,
690 .flush_on_ret
= false,
694 return __kvm_handle_hva_range(kvm
, &range
);
697 static bool kvm_change_spte_gfn(struct kvm
*kvm
, struct kvm_gfn_range
*range
)
700 * Skipping invalid memslots is correct if and only change_pte() is
701 * surrounded by invalidate_range_{start,end}(), which is currently
702 * guaranteed by the primary MMU. If that ever changes, KVM needs to
703 * unmap the memslot instead of skipping the memslot to ensure that KVM
704 * doesn't hold references to the old PFN.
706 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
708 if (range
->slot
->flags
& KVM_MEMSLOT_INVALID
)
711 return kvm_set_spte_gfn(kvm
, range
);
714 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
715 struct mm_struct
*mm
,
716 unsigned long address
,
719 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
720 const union kvm_mmu_notifier_arg arg
= { .pte
= pte
};
722 trace_kvm_set_spte_hva(address
);
725 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
726 * If mmu_invalidate_in_progress is zero, then no in-progress
727 * invalidations, including this one, found a relevant memslot at
728 * start(); rechecking memslots here is unnecessary. Note, a false
729 * positive (count elevated by a different invalidation) is sub-optimal
730 * but functionally ok.
732 WARN_ON_ONCE(!READ_ONCE(kvm
->mn_active_invalidate_count
));
733 if (!READ_ONCE(kvm
->mmu_invalidate_in_progress
))
736 kvm_handle_hva_range(mn
, address
, address
+ 1, arg
, kvm_change_spte_gfn
);
739 void kvm_mmu_invalidate_begin(struct kvm
*kvm
, unsigned long start
,
743 * The count increase must become visible at unlock time as no
744 * spte can be established without taking the mmu_lock and
745 * count is also read inside the mmu_lock critical section.
747 kvm
->mmu_invalidate_in_progress
++;
748 if (likely(kvm
->mmu_invalidate_in_progress
== 1)) {
749 kvm
->mmu_invalidate_range_start
= start
;
750 kvm
->mmu_invalidate_range_end
= end
;
753 * Fully tracking multiple concurrent ranges has diminishing
754 * returns. Keep things simple and just find the minimal range
755 * which includes the current and new ranges. As there won't be
756 * enough information to subtract a range after its invalidate
757 * completes, any ranges invalidated concurrently will
758 * accumulate and persist until all outstanding invalidates
761 kvm
->mmu_invalidate_range_start
=
762 min(kvm
->mmu_invalidate_range_start
, start
);
763 kvm
->mmu_invalidate_range_end
=
764 max(kvm
->mmu_invalidate_range_end
, end
);
768 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
769 const struct mmu_notifier_range
*range
)
771 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
772 const struct kvm_hva_range hva_range
= {
773 .start
= range
->start
,
775 .handler
= kvm_unmap_gfn_range
,
776 .on_lock
= kvm_mmu_invalidate_begin
,
777 .on_unlock
= kvm_arch_guest_memory_reclaimed
,
778 .flush_on_ret
= true,
779 .may_block
= mmu_notifier_range_blockable(range
),
782 trace_kvm_unmap_hva_range(range
->start
, range
->end
);
785 * Prevent memslot modification between range_start() and range_end()
786 * so that conditionally locking provides the same result in both
787 * functions. Without that guarantee, the mmu_invalidate_in_progress
788 * adjustments will be imbalanced.
790 * Pairs with the decrement in range_end().
792 spin_lock(&kvm
->mn_invalidate_lock
);
793 kvm
->mn_active_invalidate_count
++;
794 spin_unlock(&kvm
->mn_invalidate_lock
);
797 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
798 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
799 * each cache's lock. There are relatively few caches in existence at
800 * any given time, and the caches themselves can check for hva overlap,
801 * i.e. don't need to rely on memslot overlap checks for performance.
802 * Because this runs without holding mmu_lock, the pfn caches must use
803 * mn_active_invalidate_count (see above) instead of
804 * mmu_invalidate_in_progress.
806 gfn_to_pfn_cache_invalidate_start(kvm
, range
->start
, range
->end
,
807 hva_range
.may_block
);
809 __kvm_handle_hva_range(kvm
, &hva_range
);
814 void kvm_mmu_invalidate_end(struct kvm
*kvm
, unsigned long start
,
818 * This sequence increase will notify the kvm page fault that
819 * the page that is going to be mapped in the spte could have
822 kvm
->mmu_invalidate_seq
++;
825 * The above sequence increase must be visible before the
826 * below count decrease, which is ensured by the smp_wmb above
827 * in conjunction with the smp_rmb in mmu_invalidate_retry().
829 kvm
->mmu_invalidate_in_progress
--;
832 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
833 const struct mmu_notifier_range
*range
)
835 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
836 const struct kvm_hva_range hva_range
= {
837 .start
= range
->start
,
839 .handler
= (void *)kvm_null_fn
,
840 .on_lock
= kvm_mmu_invalidate_end
,
841 .on_unlock
= (void *)kvm_null_fn
,
842 .flush_on_ret
= false,
843 .may_block
= mmu_notifier_range_blockable(range
),
847 __kvm_handle_hva_range(kvm
, &hva_range
);
849 /* Pairs with the increment in range_start(). */
850 spin_lock(&kvm
->mn_invalidate_lock
);
851 wake
= (--kvm
->mn_active_invalidate_count
== 0);
852 spin_unlock(&kvm
->mn_invalidate_lock
);
855 * There can only be one waiter, since the wait happens under
859 rcuwait_wake_up(&kvm
->mn_memslots_update_rcuwait
);
861 BUG_ON(kvm
->mmu_invalidate_in_progress
< 0);
864 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
865 struct mm_struct
*mm
,
869 trace_kvm_age_hva(start
, end
);
871 return kvm_handle_hva_range(mn
, start
, end
, KVM_MMU_NOTIFIER_NO_ARG
,
875 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
876 struct mm_struct
*mm
,
880 trace_kvm_age_hva(start
, end
);
883 * Even though we do not flush TLB, this will still adversely
884 * affect performance on pre-Haswell Intel EPT, where there is
885 * no EPT Access Bit to clear so that we have to tear down EPT
886 * tables instead. If we find this unacceptable, we can always
887 * add a parameter to kvm_age_hva so that it effectively doesn't
888 * do anything on clear_young.
890 * Also note that currently we never issue secondary TLB flushes
891 * from clear_young, leaving this job up to the regular system
892 * cadence. If we find this inaccurate, we might come up with a
893 * more sophisticated heuristic later.
895 return kvm_handle_hva_range_no_flush(mn
, start
, end
, kvm_age_gfn
);
898 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
899 struct mm_struct
*mm
,
900 unsigned long address
)
902 trace_kvm_test_age_hva(address
);
904 return kvm_handle_hva_range_no_flush(mn
, address
, address
+ 1,
908 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
909 struct mm_struct
*mm
)
911 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
914 idx
= srcu_read_lock(&kvm
->srcu
);
915 kvm_flush_shadow_all(kvm
);
916 srcu_read_unlock(&kvm
->srcu
, idx
);
919 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
920 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
921 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
922 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
923 .clear_young
= kvm_mmu_notifier_clear_young
,
924 .test_young
= kvm_mmu_notifier_test_young
,
925 .change_pte
= kvm_mmu_notifier_change_pte
,
926 .release
= kvm_mmu_notifier_release
,
929 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
931 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
932 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
935 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
937 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
942 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
944 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
945 static int kvm_pm_notifier_call(struct notifier_block
*bl
,
949 struct kvm
*kvm
= container_of(bl
, struct kvm
, pm_notifier
);
951 return kvm_arch_pm_notifier(kvm
, state
);
954 static void kvm_init_pm_notifier(struct kvm
*kvm
)
956 kvm
->pm_notifier
.notifier_call
= kvm_pm_notifier_call
;
957 /* Suspend KVM before we suspend ftrace, RCU, etc. */
958 kvm
->pm_notifier
.priority
= INT_MAX
;
959 register_pm_notifier(&kvm
->pm_notifier
);
962 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
964 unregister_pm_notifier(&kvm
->pm_notifier
);
966 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
967 static void kvm_init_pm_notifier(struct kvm
*kvm
)
971 static void kvm_destroy_pm_notifier(struct kvm
*kvm
)
974 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
976 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
978 if (!memslot
->dirty_bitmap
)
981 kvfree(memslot
->dirty_bitmap
);
982 memslot
->dirty_bitmap
= NULL
;
985 /* This does not remove the slot from struct kvm_memslots data structures */
986 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
988 kvm_destroy_dirty_bitmap(slot
);
990 kvm_arch_free_memslot(kvm
, slot
);
995 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
997 struct hlist_node
*idnode
;
998 struct kvm_memory_slot
*memslot
;
1002 * The same memslot objects live in both active and inactive sets,
1003 * arbitrarily free using index '1' so the second invocation of this
1004 * function isn't operating over a structure with dangling pointers
1005 * (even though this function isn't actually touching them).
1007 if (!slots
->node_idx
)
1010 hash_for_each_safe(slots
->id_hash
, bkt
, idnode
, memslot
, id_node
[1])
1011 kvm_free_memslot(kvm
, memslot
);
1014 static umode_t
kvm_stats_debugfs_mode(const struct _kvm_stats_desc
*pdesc
)
1016 switch (pdesc
->desc
.flags
& KVM_STATS_TYPE_MASK
) {
1017 case KVM_STATS_TYPE_INSTANT
:
1019 case KVM_STATS_TYPE_CUMULATIVE
:
1020 case KVM_STATS_TYPE_PEAK
:
1027 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
1030 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1031 kvm_vcpu_stats_header
.num_desc
;
1033 if (IS_ERR(kvm
->debugfs_dentry
))
1036 debugfs_remove_recursive(kvm
->debugfs_dentry
);
1038 if (kvm
->debugfs_stat_data
) {
1039 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
1040 kfree(kvm
->debugfs_stat_data
[i
]);
1041 kfree(kvm
->debugfs_stat_data
);
1045 static int kvm_create_vm_debugfs(struct kvm
*kvm
, const char *fdname
)
1047 static DEFINE_MUTEX(kvm_debugfs_lock
);
1048 struct dentry
*dent
;
1049 char dir_name
[ITOA_MAX_LEN
* 2];
1050 struct kvm_stat_data
*stat_data
;
1051 const struct _kvm_stats_desc
*pdesc
;
1052 int i
, ret
= -ENOMEM
;
1053 int kvm_debugfs_num_entries
= kvm_vm_stats_header
.num_desc
+
1054 kvm_vcpu_stats_header
.num_desc
;
1056 if (!debugfs_initialized())
1059 snprintf(dir_name
, sizeof(dir_name
), "%d-%s", task_pid_nr(current
), fdname
);
1060 mutex_lock(&kvm_debugfs_lock
);
1061 dent
= debugfs_lookup(dir_name
, kvm_debugfs_dir
);
1063 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name
);
1065 mutex_unlock(&kvm_debugfs_lock
);
1068 dent
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
1069 mutex_unlock(&kvm_debugfs_lock
);
1073 kvm
->debugfs_dentry
= dent
;
1074 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
1075 sizeof(*kvm
->debugfs_stat_data
),
1076 GFP_KERNEL_ACCOUNT
);
1077 if (!kvm
->debugfs_stat_data
)
1080 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
1081 pdesc
= &kvm_vm_stats_desc
[i
];
1082 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1086 stat_data
->kvm
= kvm
;
1087 stat_data
->desc
= pdesc
;
1088 stat_data
->kind
= KVM_STAT_VM
;
1089 kvm
->debugfs_stat_data
[i
] = stat_data
;
1090 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1091 kvm
->debugfs_dentry
, stat_data
,
1095 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
1096 pdesc
= &kvm_vcpu_stats_desc
[i
];
1097 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
1101 stat_data
->kvm
= kvm
;
1102 stat_data
->desc
= pdesc
;
1103 stat_data
->kind
= KVM_STAT_VCPU
;
1104 kvm
->debugfs_stat_data
[i
+ kvm_vm_stats_header
.num_desc
] = stat_data
;
1105 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
1106 kvm
->debugfs_dentry
, stat_data
,
1110 ret
= kvm_arch_create_vm_debugfs(kvm
);
1116 kvm_destroy_vm_debugfs(kvm
);
1121 * Called after the VM is otherwise initialized, but just before adding it to
1124 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
1130 * Called just after removing the VM from the vm_list, but before doing any
1131 * other destruction.
1133 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
1138 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1139 * be setup already, so we can create arch-specific debugfs entries under it.
1140 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1141 * a per-arch destroy interface is not needed.
1143 int __weak
kvm_arch_create_vm_debugfs(struct kvm
*kvm
)
1148 static struct kvm
*kvm_create_vm(unsigned long type
, const char *fdname
)
1150 struct kvm
*kvm
= kvm_arch_alloc_vm();
1151 struct kvm_memslots
*slots
;
1156 return ERR_PTR(-ENOMEM
);
1158 KVM_MMU_LOCK_INIT(kvm
);
1159 mmgrab(current
->mm
);
1160 kvm
->mm
= current
->mm
;
1161 kvm_eventfd_init(kvm
);
1162 mutex_init(&kvm
->lock
);
1163 mutex_init(&kvm
->irq_lock
);
1164 mutex_init(&kvm
->slots_lock
);
1165 mutex_init(&kvm
->slots_arch_lock
);
1166 spin_lock_init(&kvm
->mn_invalidate_lock
);
1167 rcuwait_init(&kvm
->mn_memslots_update_rcuwait
);
1168 xa_init(&kvm
->vcpu_array
);
1170 INIT_LIST_HEAD(&kvm
->gpc_list
);
1171 spin_lock_init(&kvm
->gpc_lock
);
1173 INIT_LIST_HEAD(&kvm
->devices
);
1174 kvm
->max_vcpus
= KVM_MAX_VCPUS
;
1176 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
1179 * Force subsequent debugfs file creations to fail if the VM directory
1180 * is not created (by kvm_create_vm_debugfs()).
1182 kvm
->debugfs_dentry
= ERR_PTR(-ENOENT
);
1184 snprintf(kvm
->stats_id
, sizeof(kvm
->stats_id
), "kvm-%d",
1185 task_pid_nr(current
));
1187 if (init_srcu_struct(&kvm
->srcu
))
1188 goto out_err_no_srcu
;
1189 if (init_srcu_struct(&kvm
->irq_srcu
))
1190 goto out_err_no_irq_srcu
;
1192 refcount_set(&kvm
->users_count
, 1);
1193 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
1194 for (j
= 0; j
< 2; j
++) {
1195 slots
= &kvm
->__memslots
[i
][j
];
1197 atomic_long_set(&slots
->last_used_slot
, (unsigned long)NULL
);
1198 slots
->hva_tree
= RB_ROOT_CACHED
;
1199 slots
->gfn_tree
= RB_ROOT
;
1200 hash_init(slots
->id_hash
);
1201 slots
->node_idx
= j
;
1203 /* Generations must be different for each address space. */
1204 slots
->generation
= i
;
1207 rcu_assign_pointer(kvm
->memslots
[i
], &kvm
->__memslots
[i
][0]);
1210 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1211 rcu_assign_pointer(kvm
->buses
[i
],
1212 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
1214 goto out_err_no_arch_destroy_vm
;
1217 r
= kvm_arch_init_vm(kvm
, type
);
1219 goto out_err_no_arch_destroy_vm
;
1221 r
= hardware_enable_all();
1223 goto out_err_no_disable
;
1225 #ifdef CONFIG_HAVE_KVM_IRQFD
1226 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
1229 r
= kvm_init_mmu_notifier(kvm
);
1231 goto out_err_no_mmu_notifier
;
1233 r
= kvm_coalesced_mmio_init(kvm
);
1235 goto out_no_coalesced_mmio
;
1237 r
= kvm_create_vm_debugfs(kvm
, fdname
);
1239 goto out_err_no_debugfs
;
1241 r
= kvm_arch_post_init_vm(kvm
);
1245 mutex_lock(&kvm_lock
);
1246 list_add(&kvm
->vm_list
, &vm_list
);
1247 mutex_unlock(&kvm_lock
);
1249 preempt_notifier_inc();
1250 kvm_init_pm_notifier(kvm
);
1255 kvm_destroy_vm_debugfs(kvm
);
1257 kvm_coalesced_mmio_free(kvm
);
1258 out_no_coalesced_mmio
:
1259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1260 if (kvm
->mmu_notifier
.ops
)
1261 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
1263 out_err_no_mmu_notifier
:
1264 hardware_disable_all();
1266 kvm_arch_destroy_vm(kvm
);
1267 out_err_no_arch_destroy_vm
:
1268 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
1269 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
1270 kfree(kvm_get_bus(kvm
, i
));
1271 cleanup_srcu_struct(&kvm
->irq_srcu
);
1272 out_err_no_irq_srcu
:
1273 cleanup_srcu_struct(&kvm
->srcu
);
1275 kvm_arch_free_vm(kvm
);
1276 mmdrop(current
->mm
);
1280 static void kvm_destroy_devices(struct kvm
*kvm
)
1282 struct kvm_device
*dev
, *tmp
;
1285 * We do not need to take the kvm->lock here, because nobody else
1286 * has a reference to the struct kvm at this point and therefore
1287 * cannot access the devices list anyhow.
1289 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
1290 list_del(&dev
->vm_node
);
1291 dev
->ops
->destroy(dev
);
1295 static void kvm_destroy_vm(struct kvm
*kvm
)
1298 struct mm_struct
*mm
= kvm
->mm
;
1300 kvm_destroy_pm_notifier(kvm
);
1301 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
1302 kvm_destroy_vm_debugfs(kvm
);
1303 kvm_arch_sync_events(kvm
);
1304 mutex_lock(&kvm_lock
);
1305 list_del(&kvm
->vm_list
);
1306 mutex_unlock(&kvm_lock
);
1307 kvm_arch_pre_destroy_vm(kvm
);
1309 kvm_free_irq_routing(kvm
);
1310 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
1311 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
1314 kvm_io_bus_destroy(bus
);
1315 kvm
->buses
[i
] = NULL
;
1317 kvm_coalesced_mmio_free(kvm
);
1318 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1319 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
1321 * At this point, pending calls to invalidate_range_start()
1322 * have completed but no more MMU notifiers will run, so
1323 * mn_active_invalidate_count may remain unbalanced.
1324 * No threads can be waiting in kvm_swap_active_memslots() as the
1325 * last reference on KVM has been dropped, but freeing
1326 * memslots would deadlock without this manual intervention.
1328 WARN_ON(rcuwait_active(&kvm
->mn_memslots_update_rcuwait
));
1329 kvm
->mn_active_invalidate_count
= 0;
1331 kvm_flush_shadow_all(kvm
);
1333 kvm_arch_destroy_vm(kvm
);
1334 kvm_destroy_devices(kvm
);
1335 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
1336 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][0]);
1337 kvm_free_memslots(kvm
, &kvm
->__memslots
[i
][1]);
1339 cleanup_srcu_struct(&kvm
->irq_srcu
);
1340 cleanup_srcu_struct(&kvm
->srcu
);
1341 kvm_arch_free_vm(kvm
);
1342 preempt_notifier_dec();
1343 hardware_disable_all();
1347 void kvm_get_kvm(struct kvm
*kvm
)
1349 refcount_inc(&kvm
->users_count
);
1351 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
1354 * Make sure the vm is not during destruction, which is a safe version of
1355 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1357 bool kvm_get_kvm_safe(struct kvm
*kvm
)
1359 return refcount_inc_not_zero(&kvm
->users_count
);
1361 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe
);
1363 void kvm_put_kvm(struct kvm
*kvm
)
1365 if (refcount_dec_and_test(&kvm
->users_count
))
1366 kvm_destroy_vm(kvm
);
1368 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
1371 * Used to put a reference that was taken on behalf of an object associated
1372 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1373 * of the new file descriptor fails and the reference cannot be transferred to
1374 * its final owner. In such cases, the caller is still actively using @kvm and
1375 * will fail miserably if the refcount unexpectedly hits zero.
1377 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
1379 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
1381 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
1383 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
1385 struct kvm
*kvm
= filp
->private_data
;
1387 kvm_irqfd_release(kvm
);
1394 * Allocation size is twice as large as the actual dirty bitmap size.
1395 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1397 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
1399 unsigned long dirty_bytes
= kvm_dirty_bitmap_bytes(memslot
);
1401 memslot
->dirty_bitmap
= __vcalloc(2, dirty_bytes
, GFP_KERNEL_ACCOUNT
);
1402 if (!memslot
->dirty_bitmap
)
1408 static struct kvm_memslots
*kvm_get_inactive_memslots(struct kvm
*kvm
, int as_id
)
1410 struct kvm_memslots
*active
= __kvm_memslots(kvm
, as_id
);
1411 int node_idx_inactive
= active
->node_idx
^ 1;
1413 return &kvm
->__memslots
[as_id
][node_idx_inactive
];
1417 * Helper to get the address space ID when one of memslot pointers may be NULL.
1418 * This also serves as a sanity that at least one of the pointers is non-NULL,
1419 * and that their address space IDs don't diverge.
1421 static int kvm_memslots_get_as_id(struct kvm_memory_slot
*a
,
1422 struct kvm_memory_slot
*b
)
1424 if (WARN_ON_ONCE(!a
&& !b
))
1432 WARN_ON_ONCE(a
->as_id
!= b
->as_id
);
1436 static void kvm_insert_gfn_node(struct kvm_memslots
*slots
,
1437 struct kvm_memory_slot
*slot
)
1439 struct rb_root
*gfn_tree
= &slots
->gfn_tree
;
1440 struct rb_node
**node
, *parent
;
1441 int idx
= slots
->node_idx
;
1444 for (node
= &gfn_tree
->rb_node
; *node
; ) {
1445 struct kvm_memory_slot
*tmp
;
1447 tmp
= container_of(*node
, struct kvm_memory_slot
, gfn_node
[idx
]);
1449 if (slot
->base_gfn
< tmp
->base_gfn
)
1450 node
= &(*node
)->rb_left
;
1451 else if (slot
->base_gfn
> tmp
->base_gfn
)
1452 node
= &(*node
)->rb_right
;
1457 rb_link_node(&slot
->gfn_node
[idx
], parent
, node
);
1458 rb_insert_color(&slot
->gfn_node
[idx
], gfn_tree
);
1461 static void kvm_erase_gfn_node(struct kvm_memslots
*slots
,
1462 struct kvm_memory_slot
*slot
)
1464 rb_erase(&slot
->gfn_node
[slots
->node_idx
], &slots
->gfn_tree
);
1467 static void kvm_replace_gfn_node(struct kvm_memslots
*slots
,
1468 struct kvm_memory_slot
*old
,
1469 struct kvm_memory_slot
*new)
1471 int idx
= slots
->node_idx
;
1473 WARN_ON_ONCE(old
->base_gfn
!= new->base_gfn
);
1475 rb_replace_node(&old
->gfn_node
[idx
], &new->gfn_node
[idx
],
1480 * Replace @old with @new in the inactive memslots.
1482 * With NULL @old this simply adds @new.
1483 * With NULL @new this simply removes @old.
1485 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1488 static void kvm_replace_memslot(struct kvm
*kvm
,
1489 struct kvm_memory_slot
*old
,
1490 struct kvm_memory_slot
*new)
1492 int as_id
= kvm_memslots_get_as_id(old
, new);
1493 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1494 int idx
= slots
->node_idx
;
1497 hash_del(&old
->id_node
[idx
]);
1498 interval_tree_remove(&old
->hva_node
[idx
], &slots
->hva_tree
);
1500 if ((long)old
== atomic_long_read(&slots
->last_used_slot
))
1501 atomic_long_set(&slots
->last_used_slot
, (long)new);
1504 kvm_erase_gfn_node(slots
, old
);
1510 * Initialize @new's hva range. Do this even when replacing an @old
1511 * slot, kvm_copy_memslot() deliberately does not touch node data.
1513 new->hva_node
[idx
].start
= new->userspace_addr
;
1514 new->hva_node
[idx
].last
= new->userspace_addr
+
1515 (new->npages
<< PAGE_SHIFT
) - 1;
1518 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1519 * hva_node needs to be swapped with remove+insert even though hva can't
1520 * change when replacing an existing slot.
1522 hash_add(slots
->id_hash
, &new->id_node
[idx
], new->id
);
1523 interval_tree_insert(&new->hva_node
[idx
], &slots
->hva_tree
);
1526 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1527 * switch the node in the gfn tree instead of removing the old and
1528 * inserting the new as two separate operations. Replacement is a
1529 * single O(1) operation versus two O(log(n)) operations for
1532 if (old
&& old
->base_gfn
== new->base_gfn
) {
1533 kvm_replace_gfn_node(slots
, old
, new);
1536 kvm_erase_gfn_node(slots
, old
);
1537 kvm_insert_gfn_node(slots
, new);
1541 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
1543 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1545 #ifdef __KVM_HAVE_READONLY_MEM
1546 valid_flags
|= KVM_MEM_READONLY
;
1549 if (mem
->flags
& ~valid_flags
)
1555 static void kvm_swap_active_memslots(struct kvm
*kvm
, int as_id
)
1557 struct kvm_memslots
*slots
= kvm_get_inactive_memslots(kvm
, as_id
);
1559 /* Grab the generation from the activate memslots. */
1560 u64 gen
= __kvm_memslots(kvm
, as_id
)->generation
;
1562 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1563 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1566 * Do not store the new memslots while there are invalidations in
1567 * progress, otherwise the locking in invalidate_range_start and
1568 * invalidate_range_end will be unbalanced.
1570 spin_lock(&kvm
->mn_invalidate_lock
);
1571 prepare_to_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1572 while (kvm
->mn_active_invalidate_count
) {
1573 set_current_state(TASK_UNINTERRUPTIBLE
);
1574 spin_unlock(&kvm
->mn_invalidate_lock
);
1576 spin_lock(&kvm
->mn_invalidate_lock
);
1578 finish_rcuwait(&kvm
->mn_memslots_update_rcuwait
);
1579 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1580 spin_unlock(&kvm
->mn_invalidate_lock
);
1583 * Acquired in kvm_set_memslot. Must be released before synchronize
1584 * SRCU below in order to avoid deadlock with another thread
1585 * acquiring the slots_arch_lock in an srcu critical section.
1587 mutex_unlock(&kvm
->slots_arch_lock
);
1589 synchronize_srcu_expedited(&kvm
->srcu
);
1592 * Increment the new memslot generation a second time, dropping the
1593 * update in-progress flag and incrementing the generation based on
1594 * the number of address spaces. This provides a unique and easily
1595 * identifiable generation number while the memslots are in flux.
1597 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1600 * Generations must be unique even across address spaces. We do not need
1601 * a global counter for that, instead the generation space is evenly split
1602 * across address spaces. For example, with two address spaces, address
1603 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1604 * use generations 1, 3, 5, ...
1606 gen
+= KVM_ADDRESS_SPACE_NUM
;
1608 kvm_arch_memslots_updated(kvm
, gen
);
1610 slots
->generation
= gen
;
1613 static int kvm_prepare_memory_region(struct kvm
*kvm
,
1614 const struct kvm_memory_slot
*old
,
1615 struct kvm_memory_slot
*new,
1616 enum kvm_mr_change change
)
1621 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1622 * will be freed on "commit". If logging is enabled in both old and
1623 * new, reuse the existing bitmap. If logging is enabled only in the
1624 * new and KVM isn't using a ring buffer, allocate and initialize a
1627 if (change
!= KVM_MR_DELETE
) {
1628 if (!(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1629 new->dirty_bitmap
= NULL
;
1630 else if (old
&& old
->dirty_bitmap
)
1631 new->dirty_bitmap
= old
->dirty_bitmap
;
1632 else if (kvm_use_dirty_bitmap(kvm
)) {
1633 r
= kvm_alloc_dirty_bitmap(new);
1637 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1638 bitmap_set(new->dirty_bitmap
, 0, new->npages
);
1642 r
= kvm_arch_prepare_memory_region(kvm
, old
, new, change
);
1644 /* Free the bitmap on failure if it was allocated above. */
1645 if (r
&& new && new->dirty_bitmap
&& (!old
|| !old
->dirty_bitmap
))
1646 kvm_destroy_dirty_bitmap(new);
1651 static void kvm_commit_memory_region(struct kvm
*kvm
,
1652 struct kvm_memory_slot
*old
,
1653 const struct kvm_memory_slot
*new,
1654 enum kvm_mr_change change
)
1656 int old_flags
= old
? old
->flags
: 0;
1657 int new_flags
= new ? new->flags
: 0;
1659 * Update the total number of memslot pages before calling the arch
1660 * hook so that architectures can consume the result directly.
1662 if (change
== KVM_MR_DELETE
)
1663 kvm
->nr_memslot_pages
-= old
->npages
;
1664 else if (change
== KVM_MR_CREATE
)
1665 kvm
->nr_memslot_pages
+= new->npages
;
1667 if ((old_flags
^ new_flags
) & KVM_MEM_LOG_DIRTY_PAGES
) {
1668 int change
= (new_flags
& KVM_MEM_LOG_DIRTY_PAGES
) ? 1 : -1;
1669 atomic_set(&kvm
->nr_memslots_dirty_logging
,
1670 atomic_read(&kvm
->nr_memslots_dirty_logging
) + change
);
1673 kvm_arch_commit_memory_region(kvm
, old
, new, change
);
1677 /* Nothing more to do. */
1680 /* Free the old memslot and all its metadata. */
1681 kvm_free_memslot(kvm
, old
);
1684 case KVM_MR_FLAGS_ONLY
:
1686 * Free the dirty bitmap as needed; the below check encompasses
1687 * both the flags and whether a ring buffer is being used)
1689 if (old
->dirty_bitmap
&& !new->dirty_bitmap
)
1690 kvm_destroy_dirty_bitmap(old
);
1693 * The final quirk. Free the detached, old slot, but only its
1694 * memory, not any metadata. Metadata, including arch specific
1695 * data, may be reused by @new.
1705 * Activate @new, which must be installed in the inactive slots by the caller,
1706 * by swapping the active slots and then propagating @new to @old once @old is
1707 * unreachable and can be safely modified.
1709 * With NULL @old this simply adds @new to @active (while swapping the sets).
1710 * With NULL @new this simply removes @old from @active and frees it
1711 * (while also swapping the sets).
1713 static void kvm_activate_memslot(struct kvm
*kvm
,
1714 struct kvm_memory_slot
*old
,
1715 struct kvm_memory_slot
*new)
1717 int as_id
= kvm_memslots_get_as_id(old
, new);
1719 kvm_swap_active_memslots(kvm
, as_id
);
1721 /* Propagate the new memslot to the now inactive memslots. */
1722 kvm_replace_memslot(kvm
, old
, new);
1725 static void kvm_copy_memslot(struct kvm_memory_slot
*dest
,
1726 const struct kvm_memory_slot
*src
)
1728 dest
->base_gfn
= src
->base_gfn
;
1729 dest
->npages
= src
->npages
;
1730 dest
->dirty_bitmap
= src
->dirty_bitmap
;
1731 dest
->arch
= src
->arch
;
1732 dest
->userspace_addr
= src
->userspace_addr
;
1733 dest
->flags
= src
->flags
;
1735 dest
->as_id
= src
->as_id
;
1738 static void kvm_invalidate_memslot(struct kvm
*kvm
,
1739 struct kvm_memory_slot
*old
,
1740 struct kvm_memory_slot
*invalid_slot
)
1743 * Mark the current slot INVALID. As with all memslot modifications,
1744 * this must be done on an unreachable slot to avoid modifying the
1745 * current slot in the active tree.
1747 kvm_copy_memslot(invalid_slot
, old
);
1748 invalid_slot
->flags
|= KVM_MEMSLOT_INVALID
;
1749 kvm_replace_memslot(kvm
, old
, invalid_slot
);
1752 * Activate the slot that is now marked INVALID, but don't propagate
1753 * the slot to the now inactive slots. The slot is either going to be
1754 * deleted or recreated as a new slot.
1756 kvm_swap_active_memslots(kvm
, old
->as_id
);
1759 * From this point no new shadow pages pointing to a deleted, or moved,
1760 * memslot will be created. Validation of sp->gfn happens in:
1761 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1762 * - kvm_is_visible_gfn (mmu_check_root)
1764 kvm_arch_flush_shadow_memslot(kvm
, old
);
1765 kvm_arch_guest_memory_reclaimed(kvm
);
1767 /* Was released by kvm_swap_active_memslots(), reacquire. */
1768 mutex_lock(&kvm
->slots_arch_lock
);
1771 * Copy the arch-specific field of the newly-installed slot back to the
1772 * old slot as the arch data could have changed between releasing
1773 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1774 * above. Writers are required to retrieve memslots *after* acquiring
1775 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1777 old
->arch
= invalid_slot
->arch
;
1780 static void kvm_create_memslot(struct kvm
*kvm
,
1781 struct kvm_memory_slot
*new)
1783 /* Add the new memslot to the inactive set and activate. */
1784 kvm_replace_memslot(kvm
, NULL
, new);
1785 kvm_activate_memslot(kvm
, NULL
, new);
1788 static void kvm_delete_memslot(struct kvm
*kvm
,
1789 struct kvm_memory_slot
*old
,
1790 struct kvm_memory_slot
*invalid_slot
)
1793 * Remove the old memslot (in the inactive memslots) by passing NULL as
1794 * the "new" slot, and for the invalid version in the active slots.
1796 kvm_replace_memslot(kvm
, old
, NULL
);
1797 kvm_activate_memslot(kvm
, invalid_slot
, NULL
);
1800 static void kvm_move_memslot(struct kvm
*kvm
,
1801 struct kvm_memory_slot
*old
,
1802 struct kvm_memory_slot
*new,
1803 struct kvm_memory_slot
*invalid_slot
)
1806 * Replace the old memslot in the inactive slots, and then swap slots
1807 * and replace the current INVALID with the new as well.
1809 kvm_replace_memslot(kvm
, old
, new);
1810 kvm_activate_memslot(kvm
, invalid_slot
, new);
1813 static void kvm_update_flags_memslot(struct kvm
*kvm
,
1814 struct kvm_memory_slot
*old
,
1815 struct kvm_memory_slot
*new)
1818 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1819 * an intermediate step. Instead, the old memslot is simply replaced
1820 * with a new, updated copy in both memslot sets.
1822 kvm_replace_memslot(kvm
, old
, new);
1823 kvm_activate_memslot(kvm
, old
, new);
1826 static int kvm_set_memslot(struct kvm
*kvm
,
1827 struct kvm_memory_slot
*old
,
1828 struct kvm_memory_slot
*new,
1829 enum kvm_mr_change change
)
1831 struct kvm_memory_slot
*invalid_slot
;
1835 * Released in kvm_swap_active_memslots().
1837 * Must be held from before the current memslots are copied until after
1838 * the new memslots are installed with rcu_assign_pointer, then
1839 * released before the synchronize srcu in kvm_swap_active_memslots().
1841 * When modifying memslots outside of the slots_lock, must be held
1842 * before reading the pointer to the current memslots until after all
1843 * changes to those memslots are complete.
1845 * These rules ensure that installing new memslots does not lose
1846 * changes made to the previous memslots.
1848 mutex_lock(&kvm
->slots_arch_lock
);
1851 * Invalidate the old slot if it's being deleted or moved. This is
1852 * done prior to actually deleting/moving the memslot to allow vCPUs to
1853 * continue running by ensuring there are no mappings or shadow pages
1854 * for the memslot when it is deleted/moved. Without pre-invalidation
1855 * (and without a lock), a window would exist between effecting the
1856 * delete/move and committing the changes in arch code where KVM or a
1857 * guest could access a non-existent memslot.
1859 * Modifications are done on a temporary, unreachable slot. The old
1860 * slot needs to be preserved in case a later step fails and the
1861 * invalidation needs to be reverted.
1863 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1864 invalid_slot
= kzalloc(sizeof(*invalid_slot
), GFP_KERNEL_ACCOUNT
);
1865 if (!invalid_slot
) {
1866 mutex_unlock(&kvm
->slots_arch_lock
);
1869 kvm_invalidate_memslot(kvm
, old
, invalid_slot
);
1872 r
= kvm_prepare_memory_region(kvm
, old
, new, change
);
1875 * For DELETE/MOVE, revert the above INVALID change. No
1876 * modifications required since the original slot was preserved
1877 * in the inactive slots. Changing the active memslots also
1878 * release slots_arch_lock.
1880 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1881 kvm_activate_memslot(kvm
, invalid_slot
, old
);
1882 kfree(invalid_slot
);
1884 mutex_unlock(&kvm
->slots_arch_lock
);
1890 * For DELETE and MOVE, the working slot is now active as the INVALID
1891 * version of the old slot. MOVE is particularly special as it reuses
1892 * the old slot and returns a copy of the old slot (in working_slot).
1893 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1894 * old slot is detached but otherwise preserved.
1896 if (change
== KVM_MR_CREATE
)
1897 kvm_create_memslot(kvm
, new);
1898 else if (change
== KVM_MR_DELETE
)
1899 kvm_delete_memslot(kvm
, old
, invalid_slot
);
1900 else if (change
== KVM_MR_MOVE
)
1901 kvm_move_memslot(kvm
, old
, new, invalid_slot
);
1902 else if (change
== KVM_MR_FLAGS_ONLY
)
1903 kvm_update_flags_memslot(kvm
, old
, new);
1907 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1908 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1909 kfree(invalid_slot
);
1912 * No need to refresh new->arch, changes after dropping slots_arch_lock
1913 * will directly hit the final, active memslot. Architectures are
1914 * responsible for knowing that new->arch may be stale.
1916 kvm_commit_memory_region(kvm
, old
, new, change
);
1921 static bool kvm_check_memslot_overlap(struct kvm_memslots
*slots
, int id
,
1922 gfn_t start
, gfn_t end
)
1924 struct kvm_memslot_iter iter
;
1926 kvm_for_each_memslot_in_gfn_range(&iter
, slots
, start
, end
) {
1927 if (iter
.slot
->id
!= id
)
1935 * Allocate some memory and give it an address in the guest physical address
1938 * Discontiguous memory is allowed, mostly for framebuffers.
1940 * Must be called holding kvm->slots_lock for write.
1942 int __kvm_set_memory_region(struct kvm
*kvm
,
1943 const struct kvm_userspace_memory_region
*mem
)
1945 struct kvm_memory_slot
*old
, *new;
1946 struct kvm_memslots
*slots
;
1947 enum kvm_mr_change change
;
1948 unsigned long npages
;
1953 r
= check_memory_region_flags(mem
);
1957 as_id
= mem
->slot
>> 16;
1958 id
= (u16
)mem
->slot
;
1960 /* General sanity checks */
1961 if ((mem
->memory_size
& (PAGE_SIZE
- 1)) ||
1962 (mem
->memory_size
!= (unsigned long)mem
->memory_size
))
1964 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
1966 /* We can read the guest memory with __xxx_user() later on. */
1967 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
1968 (mem
->userspace_addr
!= untagged_addr(mem
->userspace_addr
)) ||
1969 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
1972 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
1974 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
1976 if ((mem
->memory_size
>> PAGE_SHIFT
) > KVM_MEM_MAX_NR_PAGES
)
1979 slots
= __kvm_memslots(kvm
, as_id
);
1982 * Note, the old memslot (and the pointer itself!) may be invalidated
1983 * and/or destroyed by kvm_set_memslot().
1985 old
= id_to_memslot(slots
, id
);
1987 if (!mem
->memory_size
) {
1988 if (!old
|| !old
->npages
)
1991 if (WARN_ON_ONCE(kvm
->nr_memslot_pages
< old
->npages
))
1994 return kvm_set_memslot(kvm
, old
, NULL
, KVM_MR_DELETE
);
1997 base_gfn
= (mem
->guest_phys_addr
>> PAGE_SHIFT
);
1998 npages
= (mem
->memory_size
>> PAGE_SHIFT
);
2000 if (!old
|| !old
->npages
) {
2001 change
= KVM_MR_CREATE
;
2004 * To simplify KVM internals, the total number of pages across
2005 * all memslots must fit in an unsigned long.
2007 if ((kvm
->nr_memslot_pages
+ npages
) < kvm
->nr_memslot_pages
)
2009 } else { /* Modify an existing slot. */
2010 if ((mem
->userspace_addr
!= old
->userspace_addr
) ||
2011 (npages
!= old
->npages
) ||
2012 ((mem
->flags
^ old
->flags
) & KVM_MEM_READONLY
))
2015 if (base_gfn
!= old
->base_gfn
)
2016 change
= KVM_MR_MOVE
;
2017 else if (mem
->flags
!= old
->flags
)
2018 change
= KVM_MR_FLAGS_ONLY
;
2019 else /* Nothing to change. */
2023 if ((change
== KVM_MR_CREATE
|| change
== KVM_MR_MOVE
) &&
2024 kvm_check_memslot_overlap(slots
, id
, base_gfn
, base_gfn
+ npages
))
2027 /* Allocate a slot that will persist in the memslot. */
2028 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT
);
2034 new->base_gfn
= base_gfn
;
2035 new->npages
= npages
;
2036 new->flags
= mem
->flags
;
2037 new->userspace_addr
= mem
->userspace_addr
;
2039 r
= kvm_set_memslot(kvm
, old
, new, change
);
2044 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
2046 int kvm_set_memory_region(struct kvm
*kvm
,
2047 const struct kvm_userspace_memory_region
*mem
)
2051 mutex_lock(&kvm
->slots_lock
);
2052 r
= __kvm_set_memory_region(kvm
, mem
);
2053 mutex_unlock(&kvm
->slots_lock
);
2056 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
2058 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
2059 struct kvm_userspace_memory_region
*mem
)
2061 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
2064 return kvm_set_memory_region(kvm
, mem
);
2067 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2069 * kvm_get_dirty_log - get a snapshot of dirty pages
2070 * @kvm: pointer to kvm instance
2071 * @log: slot id and address to which we copy the log
2072 * @is_dirty: set to '1' if any dirty pages were found
2073 * @memslot: set to the associated memslot, always valid on success
2075 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
2076 int *is_dirty
, struct kvm_memory_slot
**memslot
)
2078 struct kvm_memslots
*slots
;
2081 unsigned long any
= 0;
2083 /* Dirty ring tracking may be exclusive to dirty log tracking */
2084 if (!kvm_use_dirty_bitmap(kvm
))
2090 as_id
= log
->slot
>> 16;
2091 id
= (u16
)log
->slot
;
2092 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2095 slots
= __kvm_memslots(kvm
, as_id
);
2096 *memslot
= id_to_memslot(slots
, id
);
2097 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
2100 kvm_arch_sync_dirty_log(kvm
, *memslot
);
2102 n
= kvm_dirty_bitmap_bytes(*memslot
);
2104 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
2105 any
= (*memslot
)->dirty_bitmap
[i
];
2107 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
2114 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
2116 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2118 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2119 * and reenable dirty page tracking for the corresponding pages.
2120 * @kvm: pointer to kvm instance
2121 * @log: slot id and address to which we copy the log
2123 * We need to keep it in mind that VCPU threads can write to the bitmap
2124 * concurrently. So, to avoid losing track of dirty pages we keep the
2127 * 1. Take a snapshot of the bit and clear it if needed.
2128 * 2. Write protect the corresponding page.
2129 * 3. Copy the snapshot to the userspace.
2130 * 4. Upon return caller flushes TLB's if needed.
2132 * Between 2 and 4, the guest may write to the page using the remaining TLB
2133 * entry. This is not a problem because the page is reported dirty using
2134 * the snapshot taken before and step 4 ensures that writes done after
2135 * exiting to userspace will be logged for the next call.
2138 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
2140 struct kvm_memslots
*slots
;
2141 struct kvm_memory_slot
*memslot
;
2144 unsigned long *dirty_bitmap
;
2145 unsigned long *dirty_bitmap_buffer
;
2148 /* Dirty ring tracking may be exclusive to dirty log tracking */
2149 if (!kvm_use_dirty_bitmap(kvm
))
2152 as_id
= log
->slot
>> 16;
2153 id
= (u16
)log
->slot
;
2154 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2157 slots
= __kvm_memslots(kvm
, as_id
);
2158 memslot
= id_to_memslot(slots
, id
);
2159 if (!memslot
|| !memslot
->dirty_bitmap
)
2162 dirty_bitmap
= memslot
->dirty_bitmap
;
2164 kvm_arch_sync_dirty_log(kvm
, memslot
);
2166 n
= kvm_dirty_bitmap_bytes(memslot
);
2168 if (kvm
->manual_dirty_log_protect
) {
2170 * Unlike kvm_get_dirty_log, we always return false in *flush,
2171 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2172 * is some code duplication between this function and
2173 * kvm_get_dirty_log, but hopefully all architecture
2174 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2175 * can be eliminated.
2177 dirty_bitmap_buffer
= dirty_bitmap
;
2179 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2180 memset(dirty_bitmap_buffer
, 0, n
);
2183 for (i
= 0; i
< n
/ sizeof(long); i
++) {
2187 if (!dirty_bitmap
[i
])
2191 mask
= xchg(&dirty_bitmap
[i
], 0);
2192 dirty_bitmap_buffer
[i
] = mask
;
2194 offset
= i
* BITS_PER_LONG
;
2195 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2198 KVM_MMU_UNLOCK(kvm
);
2202 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2204 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
2211 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2212 * @kvm: kvm instance
2213 * @log: slot id and address to which we copy the log
2215 * Steps 1-4 below provide general overview of dirty page logging. See
2216 * kvm_get_dirty_log_protect() function description for additional details.
2218 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2219 * always flush the TLB (step 4) even if previous step failed and the dirty
2220 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2221 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2222 * writes will be marked dirty for next log read.
2224 * 1. Take a snapshot of the bit and clear it if needed.
2225 * 2. Write protect the corresponding page.
2226 * 3. Copy the snapshot to the userspace.
2227 * 4. Flush TLB's if needed.
2229 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2230 struct kvm_dirty_log
*log
)
2234 mutex_lock(&kvm
->slots_lock
);
2236 r
= kvm_get_dirty_log_protect(kvm
, log
);
2238 mutex_unlock(&kvm
->slots_lock
);
2243 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2244 * and reenable dirty page tracking for the corresponding pages.
2245 * @kvm: pointer to kvm instance
2246 * @log: slot id and address from which to fetch the bitmap of dirty pages
2248 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
2249 struct kvm_clear_dirty_log
*log
)
2251 struct kvm_memslots
*slots
;
2252 struct kvm_memory_slot
*memslot
;
2256 unsigned long *dirty_bitmap
;
2257 unsigned long *dirty_bitmap_buffer
;
2260 /* Dirty ring tracking may be exclusive to dirty log tracking */
2261 if (!kvm_use_dirty_bitmap(kvm
))
2264 as_id
= log
->slot
>> 16;
2265 id
= (u16
)log
->slot
;
2266 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
2269 if (log
->first_page
& 63)
2272 slots
= __kvm_memslots(kvm
, as_id
);
2273 memslot
= id_to_memslot(slots
, id
);
2274 if (!memslot
|| !memslot
->dirty_bitmap
)
2277 dirty_bitmap
= memslot
->dirty_bitmap
;
2279 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
2281 if (log
->first_page
> memslot
->npages
||
2282 log
->num_pages
> memslot
->npages
- log
->first_page
||
2283 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
2286 kvm_arch_sync_dirty_log(kvm
, memslot
);
2289 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
2290 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
2294 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
2295 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
2296 i
++, offset
+= BITS_PER_LONG
) {
2297 unsigned long mask
= *dirty_bitmap_buffer
++;
2298 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
2302 mask
&= atomic_long_fetch_andnot(mask
, p
);
2305 * mask contains the bits that really have been cleared. This
2306 * never includes any bits beyond the length of the memslot (if
2307 * the length is not aligned to 64 pages), therefore it is not
2308 * a problem if userspace sets them in log->dirty_bitmap.
2312 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
2316 KVM_MMU_UNLOCK(kvm
);
2319 kvm_flush_remote_tlbs_memslot(kvm
, memslot
);
2324 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
2325 struct kvm_clear_dirty_log
*log
)
2329 mutex_lock(&kvm
->slots_lock
);
2331 r
= kvm_clear_dirty_log_protect(kvm
, log
);
2333 mutex_unlock(&kvm
->slots_lock
);
2336 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2338 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
2340 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
2342 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
2344 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2346 struct kvm_memslots
*slots
= kvm_vcpu_memslots(vcpu
);
2347 u64 gen
= slots
->generation
;
2348 struct kvm_memory_slot
*slot
;
2351 * This also protects against using a memslot from a different address space,
2352 * since different address spaces have different generation numbers.
2354 if (unlikely(gen
!= vcpu
->last_used_slot_gen
)) {
2355 vcpu
->last_used_slot
= NULL
;
2356 vcpu
->last_used_slot_gen
= gen
;
2359 slot
= try_get_memslot(vcpu
->last_used_slot
, gfn
);
2364 * Fall back to searching all memslots. We purposely use
2365 * search_memslots() instead of __gfn_to_memslot() to avoid
2366 * thrashing the VM-wide last_used_slot in kvm_memslots.
2368 slot
= search_memslots(slots
, gfn
, false);
2370 vcpu
->last_used_slot
= slot
;
2377 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
2379 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
2381 return kvm_is_visible_memslot(memslot
);
2383 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
2385 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2387 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2389 return kvm_is_visible_memslot(memslot
);
2391 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
2393 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2395 struct vm_area_struct
*vma
;
2396 unsigned long addr
, size
;
2400 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
2401 if (kvm_is_error_hva(addr
))
2404 mmap_read_lock(current
->mm
);
2405 vma
= find_vma(current
->mm
, addr
);
2409 size
= vma_kernel_pagesize(vma
);
2412 mmap_read_unlock(current
->mm
);
2417 static bool memslot_is_readonly(const struct kvm_memory_slot
*slot
)
2419 return slot
->flags
& KVM_MEM_READONLY
;
2422 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2423 gfn_t
*nr_pages
, bool write
)
2425 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
2426 return KVM_HVA_ERR_BAD
;
2428 if (memslot_is_readonly(slot
) && write
)
2429 return KVM_HVA_ERR_RO_BAD
;
2432 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
2434 return __gfn_to_hva_memslot(slot
, gfn
);
2437 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2440 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
2443 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
2446 return gfn_to_hva_many(slot
, gfn
, NULL
);
2448 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
2450 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
2452 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
2454 EXPORT_SYMBOL_GPL(gfn_to_hva
);
2456 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2458 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
2460 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
2463 * Return the hva of a @gfn and the R/W attribute if possible.
2465 * @slot: the kvm_memory_slot which contains @gfn
2466 * @gfn: the gfn to be translated
2467 * @writable: used to return the read/write attribute of the @slot if the hva
2468 * is valid and @writable is not NULL
2470 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
2471 gfn_t gfn
, bool *writable
)
2473 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
2475 if (!kvm_is_error_hva(hva
) && writable
)
2476 *writable
= !memslot_is_readonly(slot
);
2481 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
2483 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2485 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2488 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
2490 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2492 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
2495 static inline int check_user_page_hwpoison(unsigned long addr
)
2497 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
2499 rc
= get_user_pages(addr
, 1, flags
, NULL
);
2500 return rc
== -EHWPOISON
;
2504 * The fast path to get the writable pfn which will be stored in @pfn,
2505 * true indicates success, otherwise false is returned. It's also the
2506 * only part that runs if we can in atomic context.
2508 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
2509 bool *writable
, kvm_pfn_t
*pfn
)
2511 struct page
*page
[1];
2514 * Fast pin a writable pfn only if it is a write fault request
2515 * or the caller allows to map a writable pfn for a read fault
2518 if (!(write_fault
|| writable
))
2521 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
2522 *pfn
= page_to_pfn(page
[0]);
2533 * The slow path to get the pfn of the specified host virtual address,
2534 * 1 indicates success, -errno is returned if error is detected.
2536 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
2537 bool interruptible
, bool *writable
, kvm_pfn_t
*pfn
)
2540 * When a VCPU accesses a page that is not mapped into the secondary
2541 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2542 * make progress. We always want to honor NUMA hinting faults in that
2543 * case, because GUP usage corresponds to memory accesses from the VCPU.
2544 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2545 * mapped into the secondary MMU and gets accessed by a VCPU.
2547 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2548 * implicitly honor NUMA hinting faults and don't need this flag.
2550 unsigned int flags
= FOLL_HWPOISON
| FOLL_HONOR_NUMA_FAULT
;
2557 *writable
= write_fault
;
2560 flags
|= FOLL_WRITE
;
2562 flags
|= FOLL_NOWAIT
;
2564 flags
|= FOLL_INTERRUPTIBLE
;
2566 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
2570 /* map read fault as writable if possible */
2571 if (unlikely(!write_fault
) && writable
) {
2574 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
2580 *pfn
= page_to_pfn(page
);
2584 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
2586 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
2589 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
2595 static int kvm_try_get_pfn(kvm_pfn_t pfn
)
2597 struct page
*page
= kvm_pfn_to_refcounted_page(pfn
);
2602 return get_page_unless_zero(page
);
2605 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
2606 unsigned long addr
, bool write_fault
,
2607 bool *writable
, kvm_pfn_t
*p_pfn
)
2615 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2618 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2619 * not call the fault handler, so do it here.
2621 bool unlocked
= false;
2622 r
= fixup_user_fault(current
->mm
, addr
,
2623 (write_fault
? FAULT_FLAG_WRITE
: 0),
2630 r
= follow_pte(vma
->vm_mm
, addr
, &ptep
, &ptl
);
2635 pte
= ptep_get(ptep
);
2637 if (write_fault
&& !pte_write(pte
)) {
2638 pfn
= KVM_PFN_ERR_RO_FAULT
;
2643 *writable
= pte_write(pte
);
2647 * Get a reference here because callers of *hva_to_pfn* and
2648 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2649 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2650 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2651 * simply do nothing for reserved pfns.
2653 * Whoever called remap_pfn_range is also going to call e.g.
2654 * unmap_mapping_range before the underlying pages are freed,
2655 * causing a call to our MMU notifier.
2657 * Certain IO or PFNMAP mappings can be backed with valid
2658 * struct pages, but be allocated without refcounting e.g.,
2659 * tail pages of non-compound higher order allocations, which
2660 * would then underflow the refcount when the caller does the
2661 * required put_page. Don't allow those pages here.
2663 if (!kvm_try_get_pfn(pfn
))
2667 pte_unmap_unlock(ptep
, ptl
);
2674 * Pin guest page in memory and return its pfn.
2675 * @addr: host virtual address which maps memory to the guest
2676 * @atomic: whether this function can sleep
2677 * @interruptible: whether the process can be interrupted by non-fatal signals
2678 * @async: whether this function need to wait IO complete if the
2679 * host page is not in the memory
2680 * @write_fault: whether we should get a writable host page
2681 * @writable: whether it allows to map a writable host page for !@write_fault
2683 * The function will map a writable host page for these two cases:
2684 * 1): @write_fault = true
2685 * 2): @write_fault = false && @writable, @writable will tell the caller
2686 * whether the mapping is writable.
2688 kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool interruptible
,
2689 bool *async
, bool write_fault
, bool *writable
)
2691 struct vm_area_struct
*vma
;
2695 /* we can do it either atomically or asynchronously, not both */
2696 BUG_ON(atomic
&& async
);
2698 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
2702 return KVM_PFN_ERR_FAULT
;
2704 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, interruptible
,
2708 if (npages
== -EINTR
)
2709 return KVM_PFN_ERR_SIGPENDING
;
2711 mmap_read_lock(current
->mm
);
2712 if (npages
== -EHWPOISON
||
2713 (!async
&& check_user_page_hwpoison(addr
))) {
2714 pfn
= KVM_PFN_ERR_HWPOISON
;
2719 vma
= vma_lookup(current
->mm
, addr
);
2722 pfn
= KVM_PFN_ERR_FAULT
;
2723 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
2724 r
= hva_to_pfn_remapped(vma
, addr
, write_fault
, writable
, &pfn
);
2728 pfn
= KVM_PFN_ERR_FAULT
;
2730 if (async
&& vma_is_valid(vma
, write_fault
))
2732 pfn
= KVM_PFN_ERR_FAULT
;
2735 mmap_read_unlock(current
->mm
);
2739 kvm_pfn_t
__gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
,
2740 bool atomic
, bool interruptible
, bool *async
,
2741 bool write_fault
, bool *writable
, hva_t
*hva
)
2743 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
2748 if (addr
== KVM_HVA_ERR_RO_BAD
) {
2751 return KVM_PFN_ERR_RO_FAULT
;
2754 if (kvm_is_error_hva(addr
)) {
2757 return KVM_PFN_NOSLOT
;
2760 /* Do not map writable pfn in the readonly memslot. */
2761 if (writable
&& memslot_is_readonly(slot
)) {
2766 return hva_to_pfn(addr
, atomic
, interruptible
, async
, write_fault
,
2769 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
2771 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
2774 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, false,
2775 NULL
, write_fault
, writable
, NULL
);
2777 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
2779 kvm_pfn_t
gfn_to_pfn_memslot(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
2781 return __gfn_to_pfn_memslot(slot
, gfn
, false, false, NULL
, true,
2784 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
2786 kvm_pfn_t
gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot
*slot
, gfn_t gfn
)
2788 return __gfn_to_pfn_memslot(slot
, gfn
, true, false, NULL
, true,
2791 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
2793 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2795 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2797 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
2799 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
2801 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
2803 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
2805 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2807 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2809 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
2811 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2812 struct page
**pages
, int nr_pages
)
2817 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
2818 if (kvm_is_error_hva(addr
))
2821 if (entry
< nr_pages
)
2824 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
2826 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
2829 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2830 * backed by 'struct page'. A valid example is if the backing memslot is
2831 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2832 * been elevated by gfn_to_pfn().
2834 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
2839 pfn
= gfn_to_pfn(kvm
, gfn
);
2841 if (is_error_noslot_pfn(pfn
))
2842 return KVM_ERR_PTR_BAD_PAGE
;
2844 page
= kvm_pfn_to_refcounted_page(pfn
);
2846 return KVM_ERR_PTR_BAD_PAGE
;
2850 EXPORT_SYMBOL_GPL(gfn_to_page
);
2852 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
)
2855 kvm_release_pfn_dirty(pfn
);
2857 kvm_release_pfn_clean(pfn
);
2860 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
2864 struct page
*page
= KVM_UNMAPPED_PAGE
;
2869 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2870 if (is_error_noslot_pfn(pfn
))
2873 if (pfn_valid(pfn
)) {
2874 page
= pfn_to_page(pfn
);
2876 #ifdef CONFIG_HAS_IOMEM
2878 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
2892 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
2894 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
2902 if (map
->page
!= KVM_UNMAPPED_PAGE
)
2904 #ifdef CONFIG_HAS_IOMEM
2910 kvm_vcpu_mark_page_dirty(vcpu
, map
->gfn
);
2912 kvm_release_pfn(map
->pfn
, dirty
);
2917 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
2919 static bool kvm_is_ad_tracked_page(struct page
*page
)
2922 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2923 * touched (e.g. set dirty) except by its owner".
2925 return !PageReserved(page
);
2928 static void kvm_set_page_dirty(struct page
*page
)
2930 if (kvm_is_ad_tracked_page(page
))
2934 static void kvm_set_page_accessed(struct page
*page
)
2936 if (kvm_is_ad_tracked_page(page
))
2937 mark_page_accessed(page
);
2940 void kvm_release_page_clean(struct page
*page
)
2942 WARN_ON(is_error_page(page
));
2944 kvm_set_page_accessed(page
);
2947 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
2949 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
2953 if (is_error_noslot_pfn(pfn
))
2956 page
= kvm_pfn_to_refcounted_page(pfn
);
2960 kvm_release_page_clean(page
);
2962 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
2964 void kvm_release_page_dirty(struct page
*page
)
2966 WARN_ON(is_error_page(page
));
2968 kvm_set_page_dirty(page
);
2969 kvm_release_page_clean(page
);
2971 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
2973 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
2977 if (is_error_noslot_pfn(pfn
))
2980 page
= kvm_pfn_to_refcounted_page(pfn
);
2984 kvm_release_page_dirty(page
);
2986 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
2989 * Note, checking for an error/noslot pfn is the caller's responsibility when
2990 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2991 * "set" helpers are not to be used when the pfn might point at garbage.
2993 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
2995 if (WARN_ON(is_error_noslot_pfn(pfn
)))
2999 kvm_set_page_dirty(pfn_to_page(pfn
));
3001 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
3003 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
3005 if (WARN_ON(is_error_noslot_pfn(pfn
)))
3009 kvm_set_page_accessed(pfn_to_page(pfn
));
3011 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
3013 static int next_segment(unsigned long len
, int offset
)
3015 if (len
> PAGE_SIZE
- offset
)
3016 return PAGE_SIZE
- offset
;
3021 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3022 void *data
, int offset
, int len
)
3027 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3028 if (kvm_is_error_hva(addr
))
3030 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
3036 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
3039 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3041 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3043 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
3045 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
3046 int offset
, int len
)
3048 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3050 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
3052 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
3054 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
3056 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3058 int offset
= offset_in_page(gpa
);
3061 while ((seg
= next_segment(len
, offset
)) != 0) {
3062 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
3072 EXPORT_SYMBOL_GPL(kvm_read_guest
);
3074 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
3076 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3078 int offset
= offset_in_page(gpa
);
3081 while ((seg
= next_segment(len
, offset
)) != 0) {
3082 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3092 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
3094 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
3095 void *data
, int offset
, unsigned long len
)
3100 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
3101 if (kvm_is_error_hva(addr
))
3103 pagefault_disable();
3104 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
3111 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3112 void *data
, unsigned long len
)
3114 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3115 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3116 int offset
= offset_in_page(gpa
);
3118 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
3120 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
3122 static int __kvm_write_guest_page(struct kvm
*kvm
,
3123 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
3124 const void *data
, int offset
, int len
)
3129 addr
= gfn_to_hva_memslot(memslot
, gfn
);
3130 if (kvm_is_error_hva(addr
))
3132 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
3135 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3139 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
3140 const void *data
, int offset
, int len
)
3142 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
3144 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
3146 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
3148 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
3149 const void *data
, int offset
, int len
)
3151 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3153 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
3155 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
3157 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
3160 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3162 int offset
= offset_in_page(gpa
);
3165 while ((seg
= next_segment(len
, offset
)) != 0) {
3166 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
3176 EXPORT_SYMBOL_GPL(kvm_write_guest
);
3178 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
3181 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3183 int offset
= offset_in_page(gpa
);
3186 while ((seg
= next_segment(len
, offset
)) != 0) {
3187 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
3197 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
3199 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
3200 struct gfn_to_hva_cache
*ghc
,
3201 gpa_t gpa
, unsigned long len
)
3203 int offset
= offset_in_page(gpa
);
3204 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
3205 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
3206 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
3207 gfn_t nr_pages_avail
;
3209 /* Update ghc->generation before performing any error checks. */
3210 ghc
->generation
= slots
->generation
;
3212 if (start_gfn
> end_gfn
) {
3213 ghc
->hva
= KVM_HVA_ERR_BAD
;
3218 * If the requested region crosses two memslots, we still
3219 * verify that the entire region is valid here.
3221 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
3222 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
3223 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
3225 if (kvm_is_error_hva(ghc
->hva
))
3229 /* Use the slow path for cross page reads and writes. */
3230 if (nr_pages_needed
== 1)
3233 ghc
->memslot
= NULL
;
3240 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3241 gpa_t gpa
, unsigned long len
)
3243 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3244 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
3246 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
3248 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3249 void *data
, unsigned int offset
,
3252 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3254 gpa_t gpa
= ghc
->gpa
+ offset
;
3256 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3259 if (slots
->generation
!= ghc
->generation
) {
3260 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3264 if (kvm_is_error_hva(ghc
->hva
))
3267 if (unlikely(!ghc
->memslot
))
3268 return kvm_write_guest(kvm
, gpa
, data
, len
);
3270 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
3273 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
3277 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
3279 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3280 void *data
, unsigned long len
)
3282 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3284 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
3286 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3287 void *data
, unsigned int offset
,
3290 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
3292 gpa_t gpa
= ghc
->gpa
+ offset
;
3294 if (WARN_ON_ONCE(len
+ offset
> ghc
->len
))
3297 if (slots
->generation
!= ghc
->generation
) {
3298 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
3302 if (kvm_is_error_hva(ghc
->hva
))
3305 if (unlikely(!ghc
->memslot
))
3306 return kvm_read_guest(kvm
, gpa
, data
, len
);
3308 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
3314 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
3316 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
3317 void *data
, unsigned long len
)
3319 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
3321 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
3323 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
3325 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3326 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3328 int offset
= offset_in_page(gpa
);
3331 while ((seg
= next_segment(len
, offset
)) != 0) {
3332 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
3341 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
3343 void mark_page_dirty_in_slot(struct kvm
*kvm
,
3344 const struct kvm_memory_slot
*memslot
,
3347 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
3349 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3350 if (WARN_ON_ONCE(vcpu
&& vcpu
->kvm
!= kvm
))
3353 WARN_ON_ONCE(!vcpu
&& !kvm_arch_allow_write_without_running_vcpu(kvm
));
3356 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
3357 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
3358 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
3360 if (kvm
->dirty_ring_size
&& vcpu
)
3361 kvm_dirty_ring_push(vcpu
, slot
, rel_gfn
);
3362 else if (memslot
->dirty_bitmap
)
3363 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
3366 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
3368 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
3370 struct kvm_memory_slot
*memslot
;
3372 memslot
= gfn_to_memslot(kvm
, gfn
);
3373 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
3375 EXPORT_SYMBOL_GPL(mark_page_dirty
);
3377 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3379 struct kvm_memory_slot
*memslot
;
3381 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
3382 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
3384 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
3386 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
3388 if (!vcpu
->sigset_active
)
3392 * This does a lockless modification of ->real_blocked, which is fine
3393 * because, only current can change ->real_blocked and all readers of
3394 * ->real_blocked don't care as long ->real_blocked is always a subset
3397 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
3400 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
3402 if (!vcpu
->sigset_active
)
3405 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
3406 sigemptyset(¤t
->real_blocked
);
3409 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3411 unsigned int old
, val
, grow
, grow_start
;
3413 old
= val
= vcpu
->halt_poll_ns
;
3414 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3415 grow
= READ_ONCE(halt_poll_ns_grow
);
3420 if (val
< grow_start
)
3423 vcpu
->halt_poll_ns
= val
;
3425 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
3428 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3430 unsigned int old
, val
, shrink
, grow_start
;
3432 old
= val
= vcpu
->halt_poll_ns
;
3433 shrink
= READ_ONCE(halt_poll_ns_shrink
);
3434 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
3440 if (val
< grow_start
)
3443 vcpu
->halt_poll_ns
= val
;
3444 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
3447 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
3450 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3452 if (kvm_arch_vcpu_runnable(vcpu
))
3454 if (kvm_cpu_has_pending_timer(vcpu
))
3456 if (signal_pending(current
))
3458 if (kvm_check_request(KVM_REQ_UNBLOCK
, vcpu
))
3463 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3468 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3469 * pending. This is mostly used when halting a vCPU, but may also be used
3470 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3472 bool kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
3474 struct rcuwait
*wait
= kvm_arch_vcpu_get_wait(vcpu
);
3475 bool waited
= false;
3477 vcpu
->stat
.generic
.blocking
= 1;
3480 kvm_arch_vcpu_blocking(vcpu
);
3481 prepare_to_rcuwait(wait
);
3485 set_current_state(TASK_INTERRUPTIBLE
);
3487 if (kvm_vcpu_check_block(vcpu
) < 0)
3495 finish_rcuwait(wait
);
3496 kvm_arch_vcpu_unblocking(vcpu
);
3499 vcpu
->stat
.generic
.blocking
= 0;
3504 static inline void update_halt_poll_stats(struct kvm_vcpu
*vcpu
, ktime_t start
,
3505 ktime_t end
, bool success
)
3507 struct kvm_vcpu_stat_generic
*stats
= &vcpu
->stat
.generic
;
3508 u64 poll_ns
= ktime_to_ns(ktime_sub(end
, start
));
3510 ++vcpu
->stat
.generic
.halt_attempted_poll
;
3513 ++vcpu
->stat
.generic
.halt_successful_poll
;
3515 if (!vcpu_valid_wakeup(vcpu
))
3516 ++vcpu
->stat
.generic
.halt_poll_invalid
;
3518 stats
->halt_poll_success_ns
+= poll_ns
;
3519 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_success_hist
, poll_ns
);
3521 stats
->halt_poll_fail_ns
+= poll_ns
;
3522 KVM_STATS_LOG_HIST_UPDATE(stats
->halt_poll_fail_hist
, poll_ns
);
3526 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu
*vcpu
)
3528 struct kvm
*kvm
= vcpu
->kvm
;
3530 if (kvm
->override_halt_poll_ns
) {
3532 * Ensure kvm->max_halt_poll_ns is not read before
3533 * kvm->override_halt_poll_ns.
3535 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3538 return READ_ONCE(kvm
->max_halt_poll_ns
);
3541 return READ_ONCE(halt_poll_ns
);
3545 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3546 * polling is enabled, busy wait for a short time before blocking to avoid the
3547 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3550 void kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
3552 unsigned int max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3553 bool halt_poll_allowed
= !kvm_arch_no_poll(vcpu
);
3554 ktime_t start
, cur
, poll_end
;
3555 bool waited
= false;
3559 if (vcpu
->halt_poll_ns
> max_halt_poll_ns
)
3560 vcpu
->halt_poll_ns
= max_halt_poll_ns
;
3562 do_halt_poll
= halt_poll_allowed
&& vcpu
->halt_poll_ns
;
3564 start
= cur
= poll_end
= ktime_get();
3566 ktime_t stop
= ktime_add_ns(start
, vcpu
->halt_poll_ns
);
3569 if (kvm_vcpu_check_block(vcpu
) < 0)
3572 poll_end
= cur
= ktime_get();
3573 } while (kvm_vcpu_can_poll(cur
, stop
));
3576 waited
= kvm_vcpu_block(vcpu
);
3580 vcpu
->stat
.generic
.halt_wait_ns
+=
3581 ktime_to_ns(cur
) - ktime_to_ns(poll_end
);
3582 KVM_STATS_LOG_HIST_UPDATE(vcpu
->stat
.generic
.halt_wait_hist
,
3583 ktime_to_ns(cur
) - ktime_to_ns(poll_end
));
3586 /* The total time the vCPU was "halted", including polling time. */
3587 halt_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
3590 * Note, halt-polling is considered successful so long as the vCPU was
3591 * never actually scheduled out, i.e. even if the wake event arrived
3592 * after of the halt-polling loop itself, but before the full wait.
3595 update_halt_poll_stats(vcpu
, start
, poll_end
, !waited
);
3597 if (halt_poll_allowed
) {
3598 /* Recompute the max halt poll time in case it changed. */
3599 max_halt_poll_ns
= kvm_vcpu_max_halt_poll_ns(vcpu
);
3601 if (!vcpu_valid_wakeup(vcpu
)) {
3602 shrink_halt_poll_ns(vcpu
);
3603 } else if (max_halt_poll_ns
) {
3604 if (halt_ns
<= vcpu
->halt_poll_ns
)
3606 /* we had a long block, shrink polling */
3607 else if (vcpu
->halt_poll_ns
&&
3608 halt_ns
> max_halt_poll_ns
)
3609 shrink_halt_poll_ns(vcpu
);
3610 /* we had a short halt and our poll time is too small */
3611 else if (vcpu
->halt_poll_ns
< max_halt_poll_ns
&&
3612 halt_ns
< max_halt_poll_ns
)
3613 grow_halt_poll_ns(vcpu
);
3615 vcpu
->halt_poll_ns
= 0;
3619 trace_kvm_vcpu_wakeup(halt_ns
, waited
, vcpu_valid_wakeup(vcpu
));
3621 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
3623 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
3625 if (__kvm_vcpu_wake_up(vcpu
)) {
3626 WRITE_ONCE(vcpu
->ready
, true);
3627 ++vcpu
->stat
.generic
.halt_wakeup
;
3633 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
3637 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3639 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3643 if (kvm_vcpu_wake_up(vcpu
))
3648 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3649 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3650 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3651 * within the vCPU thread itself.
3653 if (vcpu
== __this_cpu_read(kvm_running_vcpu
)) {
3654 if (vcpu
->mode
== IN_GUEST_MODE
)
3655 WRITE_ONCE(vcpu
->mode
, EXITING_GUEST_MODE
);
3660 * Note, the vCPU could get migrated to a different pCPU at any point
3661 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3662 * IPI to the previous pCPU. But, that's ok because the purpose of the
3663 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3664 * vCPU also requires it to leave IN_GUEST_MODE.
3666 if (kvm_arch_vcpu_should_kick(vcpu
)) {
3667 cpu
= READ_ONCE(vcpu
->cpu
);
3668 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
3669 smp_send_reschedule(cpu
);
3674 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
3675 #endif /* !CONFIG_S390 */
3677 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
3680 struct task_struct
*task
= NULL
;
3684 pid
= rcu_dereference(target
->pid
);
3686 task
= get_pid_task(pid
, PIDTYPE_PID
);
3690 ret
= yield_to(task
, 1);
3691 put_task_struct(task
);
3695 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
3698 * Helper that checks whether a VCPU is eligible for directed yield.
3699 * Most eligible candidate to yield is decided by following heuristics:
3701 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3702 * (preempted lock holder), indicated by @in_spin_loop.
3703 * Set at the beginning and cleared at the end of interception/PLE handler.
3705 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3706 * chance last time (mostly it has become eligible now since we have probably
3707 * yielded to lockholder in last iteration. This is done by toggling
3708 * @dy_eligible each time a VCPU checked for eligibility.)
3710 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3711 * to preempted lock-holder could result in wrong VCPU selection and CPU
3712 * burning. Giving priority for a potential lock-holder increases lock
3715 * Since algorithm is based on heuristics, accessing another VCPU data without
3716 * locking does not harm. It may result in trying to yield to same VCPU, fail
3717 * and continue with next VCPU and so on.
3719 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
3721 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3724 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
3725 vcpu
->spin_loop
.dy_eligible
;
3727 if (vcpu
->spin_loop
.in_spin_loop
)
3728 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
3737 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3738 * a vcpu_load/vcpu_put pair. However, for most architectures
3739 * kvm_arch_vcpu_runnable does not require vcpu_load.
3741 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
3743 return kvm_arch_vcpu_runnable(vcpu
);
3746 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
3748 if (kvm_arch_dy_runnable(vcpu
))
3751 #ifdef CONFIG_KVM_ASYNC_PF
3752 if (!list_empty_careful(&vcpu
->async_pf
.done
))
3759 bool __weak
kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu
*vcpu
)
3764 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
3766 struct kvm
*kvm
= me
->kvm
;
3767 struct kvm_vcpu
*vcpu
;
3768 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
3774 kvm_vcpu_set_in_spin_loop(me
, true);
3776 * We boost the priority of a VCPU that is runnable but not
3777 * currently running, because it got preempted by something
3778 * else and called schedule in __vcpu_run. Hopefully that
3779 * VCPU is holding the lock that we need and will release it.
3780 * We approximate round-robin by starting at the last boosted VCPU.
3782 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
3783 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3784 if (!pass
&& i
<= last_boosted_vcpu
) {
3785 i
= last_boosted_vcpu
;
3787 } else if (pass
&& i
> last_boosted_vcpu
)
3789 if (!READ_ONCE(vcpu
->ready
))
3793 if (kvm_vcpu_is_blocking(vcpu
) && !vcpu_dy_runnable(vcpu
))
3795 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
3796 !kvm_arch_dy_has_pending_interrupt(vcpu
) &&
3797 !kvm_arch_vcpu_in_kernel(vcpu
))
3799 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
3802 yielded
= kvm_vcpu_yield_to(vcpu
);
3804 kvm
->last_boosted_vcpu
= i
;
3806 } else if (yielded
< 0) {
3813 kvm_vcpu_set_in_spin_loop(me
, false);
3815 /* Ensure vcpu is not eligible during next spinloop */
3816 kvm_vcpu_set_dy_eligible(me
, false);
3818 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
3820 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
3822 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3823 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
3824 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
3825 kvm
->dirty_ring_size
/ PAGE_SIZE
);
3831 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
3833 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
3836 if (vmf
->pgoff
== 0)
3837 page
= virt_to_page(vcpu
->run
);
3839 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
3840 page
= virt_to_page(vcpu
->arch
.pio_data
);
3842 #ifdef CONFIG_KVM_MMIO
3843 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
3844 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
3846 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
3847 page
= kvm_dirty_ring_get_page(
3849 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
3851 return kvm_arch_vcpu_fault(vcpu
, vmf
);
3857 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
3858 .fault
= kvm_vcpu_fault
,
3861 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3863 struct kvm_vcpu
*vcpu
= file
->private_data
;
3864 unsigned long pages
= vma_pages(vma
);
3866 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
3867 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
3868 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
3871 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
3875 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
3877 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3879 kvm_put_kvm(vcpu
->kvm
);
3883 static struct file_operations kvm_vcpu_fops
= {
3884 .release
= kvm_vcpu_release
,
3885 .unlocked_ioctl
= kvm_vcpu_ioctl
,
3886 .mmap
= kvm_vcpu_mmap
,
3887 .llseek
= noop_llseek
,
3888 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
3892 * Allocates an inode for the vcpu.
3894 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
3896 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
3898 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
3899 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
3902 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3903 static int vcpu_get_pid(void *data
, u64
*val
)
3905 struct kvm_vcpu
*vcpu
= data
;
3908 *val
= pid_nr(rcu_dereference(vcpu
->pid
));
3913 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops
, vcpu_get_pid
, NULL
, "%llu\n");
3915 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
3917 struct dentry
*debugfs_dentry
;
3918 char dir_name
[ITOA_MAX_LEN
* 2];
3920 if (!debugfs_initialized())
3923 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
3924 debugfs_dentry
= debugfs_create_dir(dir_name
,
3925 vcpu
->kvm
->debugfs_dentry
);
3926 debugfs_create_file("pid", 0444, debugfs_dentry
, vcpu
,
3927 &vcpu_get_pid_fops
);
3929 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
3934 * Creates some virtual cpus. Good luck creating more than one.
3936 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
3939 struct kvm_vcpu
*vcpu
;
3942 if (id
>= KVM_MAX_VCPU_IDS
)
3945 mutex_lock(&kvm
->lock
);
3946 if (kvm
->created_vcpus
>= kvm
->max_vcpus
) {
3947 mutex_unlock(&kvm
->lock
);
3951 r
= kvm_arch_vcpu_precreate(kvm
, id
);
3953 mutex_unlock(&kvm
->lock
);
3957 kvm
->created_vcpus
++;
3958 mutex_unlock(&kvm
->lock
);
3960 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL_ACCOUNT
);
3963 goto vcpu_decrement
;
3966 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
3967 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
3972 vcpu
->run
= page_address(page
);
3974 kvm_vcpu_init(vcpu
, kvm
, id
);
3976 r
= kvm_arch_vcpu_create(vcpu
);
3978 goto vcpu_free_run_page
;
3980 if (kvm
->dirty_ring_size
) {
3981 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
3982 id
, kvm
->dirty_ring_size
);
3984 goto arch_vcpu_destroy
;
3987 mutex_lock(&kvm
->lock
);
3989 #ifdef CONFIG_LOCKDEP
3990 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
3991 mutex_lock(&vcpu
->mutex
);
3992 mutex_unlock(&vcpu
->mutex
);
3995 if (kvm_get_vcpu_by_id(kvm
, id
)) {
3997 goto unlock_vcpu_destroy
;
4000 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
4001 r
= xa_reserve(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, GFP_KERNEL_ACCOUNT
);
4003 goto unlock_vcpu_destroy
;
4005 /* Now it's all set up, let userspace reach it */
4007 r
= create_vcpu_fd(vcpu
);
4009 goto kvm_put_xa_release
;
4011 if (KVM_BUG_ON(xa_store(&kvm
->vcpu_array
, vcpu
->vcpu_idx
, vcpu
, 0), kvm
)) {
4013 goto kvm_put_xa_release
;
4017 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4018 * pointer before kvm->online_vcpu's incremented value.
4021 atomic_inc(&kvm
->online_vcpus
);
4023 mutex_unlock(&kvm
->lock
);
4024 kvm_arch_vcpu_postcreate(vcpu
);
4025 kvm_create_vcpu_debugfs(vcpu
);
4029 kvm_put_kvm_no_destroy(kvm
);
4030 xa_release(&kvm
->vcpu_array
, vcpu
->vcpu_idx
);
4031 unlock_vcpu_destroy
:
4032 mutex_unlock(&kvm
->lock
);
4033 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
4035 kvm_arch_vcpu_destroy(vcpu
);
4037 free_page((unsigned long)vcpu
->run
);
4039 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
4041 mutex_lock(&kvm
->lock
);
4042 kvm
->created_vcpus
--;
4043 mutex_unlock(&kvm
->lock
);
4047 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
4050 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
4051 vcpu
->sigset_active
= 1;
4052 vcpu
->sigset
= *sigset
;
4054 vcpu
->sigset_active
= 0;
4058 static ssize_t
kvm_vcpu_stats_read(struct file
*file
, char __user
*user_buffer
,
4059 size_t size
, loff_t
*offset
)
4061 struct kvm_vcpu
*vcpu
= file
->private_data
;
4063 return kvm_stats_read(vcpu
->stats_id
, &kvm_vcpu_stats_header
,
4064 &kvm_vcpu_stats_desc
[0], &vcpu
->stat
,
4065 sizeof(vcpu
->stat
), user_buffer
, size
, offset
);
4068 static int kvm_vcpu_stats_release(struct inode
*inode
, struct file
*file
)
4070 struct kvm_vcpu
*vcpu
= file
->private_data
;
4072 kvm_put_kvm(vcpu
->kvm
);
4076 static const struct file_operations kvm_vcpu_stats_fops
= {
4077 .owner
= THIS_MODULE
,
4078 .read
= kvm_vcpu_stats_read
,
4079 .release
= kvm_vcpu_stats_release
,
4080 .llseek
= noop_llseek
,
4083 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu
*vcpu
)
4087 char name
[15 + ITOA_MAX_LEN
+ 1];
4089 snprintf(name
, sizeof(name
), "kvm-vcpu-stats:%d", vcpu
->vcpu_id
);
4091 fd
= get_unused_fd_flags(O_CLOEXEC
);
4095 file
= anon_inode_getfile(name
, &kvm_vcpu_stats_fops
, vcpu
, O_RDONLY
);
4098 return PTR_ERR(file
);
4101 kvm_get_kvm(vcpu
->kvm
);
4103 file
->f_mode
|= FMODE_PREAD
;
4104 fd_install(fd
, file
);
4109 static long kvm_vcpu_ioctl(struct file
*filp
,
4110 unsigned int ioctl
, unsigned long arg
)
4112 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4113 void __user
*argp
= (void __user
*)arg
;
4115 struct kvm_fpu
*fpu
= NULL
;
4116 struct kvm_sregs
*kvm_sregs
= NULL
;
4118 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4121 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
4125 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4126 * execution; mutex_lock() would break them.
4128 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
4129 if (r
!= -ENOIOCTLCMD
)
4132 if (mutex_lock_killable(&vcpu
->mutex
))
4140 oldpid
= rcu_access_pointer(vcpu
->pid
);
4141 if (unlikely(oldpid
!= task_pid(current
))) {
4142 /* The thread running this VCPU changed. */
4145 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
4149 newpid
= get_task_pid(current
, PIDTYPE_PID
);
4150 rcu_assign_pointer(vcpu
->pid
, newpid
);
4155 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
4156 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
4159 case KVM_GET_REGS
: {
4160 struct kvm_regs
*kvm_regs
;
4163 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
4166 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
4170 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
4177 case KVM_SET_REGS
: {
4178 struct kvm_regs
*kvm_regs
;
4180 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
4181 if (IS_ERR(kvm_regs
)) {
4182 r
= PTR_ERR(kvm_regs
);
4185 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
4189 case KVM_GET_SREGS
: {
4190 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
4191 GFP_KERNEL_ACCOUNT
);
4195 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
4199 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
4204 case KVM_SET_SREGS
: {
4205 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
4206 if (IS_ERR(kvm_sregs
)) {
4207 r
= PTR_ERR(kvm_sregs
);
4211 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
4214 case KVM_GET_MP_STATE
: {
4215 struct kvm_mp_state mp_state
;
4217 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
4221 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
4226 case KVM_SET_MP_STATE
: {
4227 struct kvm_mp_state mp_state
;
4230 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
4232 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
4235 case KVM_TRANSLATE
: {
4236 struct kvm_translation tr
;
4239 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
4241 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
4245 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
4250 case KVM_SET_GUEST_DEBUG
: {
4251 struct kvm_guest_debug dbg
;
4254 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
4256 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
4259 case KVM_SET_SIGNAL_MASK
: {
4260 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4261 struct kvm_signal_mask kvm_sigmask
;
4262 sigset_t sigset
, *p
;
4267 if (copy_from_user(&kvm_sigmask
, argp
,
4268 sizeof(kvm_sigmask
)))
4271 if (kvm_sigmask
.len
!= sizeof(sigset
))
4274 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
4279 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
4283 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
4287 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
4291 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
4297 fpu
= memdup_user(argp
, sizeof(*fpu
));
4303 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
4306 case KVM_GET_STATS_FD
: {
4307 r
= kvm_vcpu_ioctl_get_stats_fd(vcpu
);
4311 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
4314 mutex_unlock(&vcpu
->mutex
);
4320 #ifdef CONFIG_KVM_COMPAT
4321 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
4322 unsigned int ioctl
, unsigned long arg
)
4324 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4325 void __user
*argp
= compat_ptr(arg
);
4328 if (vcpu
->kvm
->mm
!= current
->mm
|| vcpu
->kvm
->vm_dead
)
4332 case KVM_SET_SIGNAL_MASK
: {
4333 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
4334 struct kvm_signal_mask kvm_sigmask
;
4339 if (copy_from_user(&kvm_sigmask
, argp
,
4340 sizeof(kvm_sigmask
)))
4343 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
4346 if (get_compat_sigset(&sigset
,
4347 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
4349 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
4351 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
4355 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
4363 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
4365 struct kvm_device
*dev
= filp
->private_data
;
4368 return dev
->ops
->mmap(dev
, vma
);
4373 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
4374 int (*accessor
)(struct kvm_device
*dev
,
4375 struct kvm_device_attr
*attr
),
4378 struct kvm_device_attr attr
;
4383 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
4386 return accessor(dev
, &attr
);
4389 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
4392 struct kvm_device
*dev
= filp
->private_data
;
4394 if (dev
->kvm
->mm
!= current
->mm
|| dev
->kvm
->vm_dead
)
4398 case KVM_SET_DEVICE_ATTR
:
4399 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
4400 case KVM_GET_DEVICE_ATTR
:
4401 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
4402 case KVM_HAS_DEVICE_ATTR
:
4403 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
4405 if (dev
->ops
->ioctl
)
4406 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
4412 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
4414 struct kvm_device
*dev
= filp
->private_data
;
4415 struct kvm
*kvm
= dev
->kvm
;
4417 if (dev
->ops
->release
) {
4418 mutex_lock(&kvm
->lock
);
4419 list_del(&dev
->vm_node
);
4420 dev
->ops
->release(dev
);
4421 mutex_unlock(&kvm
->lock
);
4428 static struct file_operations kvm_device_fops
= {
4429 .unlocked_ioctl
= kvm_device_ioctl
,
4430 .release
= kvm_device_release
,
4431 KVM_COMPAT(kvm_device_ioctl
),
4432 .mmap
= kvm_device_mmap
,
4435 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
4437 if (filp
->f_op
!= &kvm_device_fops
)
4440 return filp
->private_data
;
4443 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
4444 #ifdef CONFIG_KVM_MPIC
4445 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
4446 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
4450 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
4452 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
4455 if (kvm_device_ops_table
[type
] != NULL
)
4458 kvm_device_ops_table
[type
] = ops
;
4462 void kvm_unregister_device_ops(u32 type
)
4464 if (kvm_device_ops_table
[type
] != NULL
)
4465 kvm_device_ops_table
[type
] = NULL
;
4468 static int kvm_ioctl_create_device(struct kvm
*kvm
,
4469 struct kvm_create_device
*cd
)
4471 const struct kvm_device_ops
*ops
;
4472 struct kvm_device
*dev
;
4473 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
4477 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
4480 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
4481 ops
= kvm_device_ops_table
[type
];
4488 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
4495 mutex_lock(&kvm
->lock
);
4496 ret
= ops
->create(dev
, type
);
4498 mutex_unlock(&kvm
->lock
);
4502 list_add(&dev
->vm_node
, &kvm
->devices
);
4503 mutex_unlock(&kvm
->lock
);
4509 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
4511 kvm_put_kvm_no_destroy(kvm
);
4512 mutex_lock(&kvm
->lock
);
4513 list_del(&dev
->vm_node
);
4516 mutex_unlock(&kvm
->lock
);
4526 static int kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
4529 case KVM_CAP_USER_MEMORY
:
4530 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
4531 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
4532 case KVM_CAP_INTERNAL_ERROR_DATA
:
4533 #ifdef CONFIG_HAVE_KVM_MSI
4534 case KVM_CAP_SIGNAL_MSI
:
4536 #ifdef CONFIG_HAVE_KVM_IRQFD
4539 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
4540 case KVM_CAP_CHECK_EXTENSION_VM
:
4541 case KVM_CAP_ENABLE_CAP_VM
:
4542 case KVM_CAP_HALT_POLL
:
4544 #ifdef CONFIG_KVM_MMIO
4545 case KVM_CAP_COALESCED_MMIO
:
4546 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
4547 case KVM_CAP_COALESCED_PIO
:
4550 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4551 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
4552 return KVM_DIRTY_LOG_MANUAL_CAPS
;
4554 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4555 case KVM_CAP_IRQ_ROUTING
:
4556 return KVM_MAX_IRQ_ROUTES
;
4558 #if KVM_ADDRESS_SPACE_NUM > 1
4559 case KVM_CAP_MULTI_ADDRESS_SPACE
:
4560 return KVM_ADDRESS_SPACE_NUM
;
4562 case KVM_CAP_NR_MEMSLOTS
:
4563 return KVM_USER_MEM_SLOTS
;
4564 case KVM_CAP_DIRTY_LOG_RING
:
4565 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4566 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4570 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4571 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4572 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
4576 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4577 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
:
4579 case KVM_CAP_BINARY_STATS_FD
:
4580 case KVM_CAP_SYSTEM_EVENT_DATA
:
4585 return kvm_vm_ioctl_check_extension(kvm
, arg
);
4588 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
4592 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
4595 /* the size should be power of 2 */
4596 if (!size
|| (size
& (size
- 1)))
4599 /* Should be bigger to keep the reserved entries, or a page */
4600 if (size
< kvm_dirty_ring_get_rsvd_entries() *
4601 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
4604 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
4605 sizeof(struct kvm_dirty_gfn
))
4608 /* We only allow it to set once */
4609 if (kvm
->dirty_ring_size
)
4612 mutex_lock(&kvm
->lock
);
4614 if (kvm
->created_vcpus
) {
4615 /* We don't allow to change this value after vcpu created */
4618 kvm
->dirty_ring_size
= size
;
4622 mutex_unlock(&kvm
->lock
);
4626 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
4629 struct kvm_vcpu
*vcpu
;
4632 if (!kvm
->dirty_ring_size
)
4635 mutex_lock(&kvm
->slots_lock
);
4637 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4638 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
4640 mutex_unlock(&kvm
->slots_lock
);
4643 kvm_flush_remote_tlbs(kvm
);
4648 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4649 struct kvm_enable_cap
*cap
)
4654 bool kvm_are_all_memslots_empty(struct kvm
*kvm
)
4658 lockdep_assert_held(&kvm
->slots_lock
);
4660 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
4661 if (!kvm_memslots_empty(__kvm_memslots(kvm
, i
)))
4667 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty
);
4669 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
4670 struct kvm_enable_cap
*cap
)
4673 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4674 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
4675 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
4677 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
4678 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
4680 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
4682 kvm
->manual_dirty_log_protect
= cap
->args
[0];
4686 case KVM_CAP_HALT_POLL
: {
4687 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
4690 kvm
->max_halt_poll_ns
= cap
->args
[0];
4693 * Ensure kvm->override_halt_poll_ns does not become visible
4694 * before kvm->max_halt_poll_ns.
4696 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4699 kvm
->override_halt_poll_ns
= true;
4703 case KVM_CAP_DIRTY_LOG_RING
:
4704 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL
:
4705 if (!kvm_vm_ioctl_check_extension_generic(kvm
, cap
->cap
))
4708 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
4709 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP
: {
4712 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
) ||
4713 !kvm
->dirty_ring_size
|| cap
->flags
)
4716 mutex_lock(&kvm
->slots_lock
);
4719 * For simplicity, allow enabling ring+bitmap if and only if
4720 * there are no memslots, e.g. to ensure all memslots allocate
4721 * a bitmap after the capability is enabled.
4723 if (kvm_are_all_memslots_empty(kvm
)) {
4724 kvm
->dirty_ring_with_bitmap
= true;
4728 mutex_unlock(&kvm
->slots_lock
);
4733 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
4737 static ssize_t
kvm_vm_stats_read(struct file
*file
, char __user
*user_buffer
,
4738 size_t size
, loff_t
*offset
)
4740 struct kvm
*kvm
= file
->private_data
;
4742 return kvm_stats_read(kvm
->stats_id
, &kvm_vm_stats_header
,
4743 &kvm_vm_stats_desc
[0], &kvm
->stat
,
4744 sizeof(kvm
->stat
), user_buffer
, size
, offset
);
4747 static int kvm_vm_stats_release(struct inode
*inode
, struct file
*file
)
4749 struct kvm
*kvm
= file
->private_data
;
4755 static const struct file_operations kvm_vm_stats_fops
= {
4756 .owner
= THIS_MODULE
,
4757 .read
= kvm_vm_stats_read
,
4758 .release
= kvm_vm_stats_release
,
4759 .llseek
= noop_llseek
,
4762 static int kvm_vm_ioctl_get_stats_fd(struct kvm
*kvm
)
4767 fd
= get_unused_fd_flags(O_CLOEXEC
);
4771 file
= anon_inode_getfile("kvm-vm-stats",
4772 &kvm_vm_stats_fops
, kvm
, O_RDONLY
);
4775 return PTR_ERR(file
);
4780 file
->f_mode
|= FMODE_PREAD
;
4781 fd_install(fd
, file
);
4786 static long kvm_vm_ioctl(struct file
*filp
,
4787 unsigned int ioctl
, unsigned long arg
)
4789 struct kvm
*kvm
= filp
->private_data
;
4790 void __user
*argp
= (void __user
*)arg
;
4793 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
4796 case KVM_CREATE_VCPU
:
4797 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
4799 case KVM_ENABLE_CAP
: {
4800 struct kvm_enable_cap cap
;
4803 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4805 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
4808 case KVM_SET_USER_MEMORY_REGION
: {
4809 struct kvm_userspace_memory_region kvm_userspace_mem
;
4812 if (copy_from_user(&kvm_userspace_mem
, argp
,
4813 sizeof(kvm_userspace_mem
)))
4816 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
4819 case KVM_GET_DIRTY_LOG
: {
4820 struct kvm_dirty_log log
;
4823 if (copy_from_user(&log
, argp
, sizeof(log
)))
4825 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
4828 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4829 case KVM_CLEAR_DIRTY_LOG
: {
4830 struct kvm_clear_dirty_log log
;
4833 if (copy_from_user(&log
, argp
, sizeof(log
)))
4835 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
4839 #ifdef CONFIG_KVM_MMIO
4840 case KVM_REGISTER_COALESCED_MMIO
: {
4841 struct kvm_coalesced_mmio_zone zone
;
4844 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4846 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
4849 case KVM_UNREGISTER_COALESCED_MMIO
: {
4850 struct kvm_coalesced_mmio_zone zone
;
4853 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
4855 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
4860 struct kvm_irqfd data
;
4863 if (copy_from_user(&data
, argp
, sizeof(data
)))
4865 r
= kvm_irqfd(kvm
, &data
);
4868 case KVM_IOEVENTFD
: {
4869 struct kvm_ioeventfd data
;
4872 if (copy_from_user(&data
, argp
, sizeof(data
)))
4874 r
= kvm_ioeventfd(kvm
, &data
);
4877 #ifdef CONFIG_HAVE_KVM_MSI
4878 case KVM_SIGNAL_MSI
: {
4882 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
4884 r
= kvm_send_userspace_msi(kvm
, &msi
);
4888 #ifdef __KVM_HAVE_IRQ_LINE
4889 case KVM_IRQ_LINE_STATUS
:
4890 case KVM_IRQ_LINE
: {
4891 struct kvm_irq_level irq_event
;
4894 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
4897 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
4898 ioctl
== KVM_IRQ_LINE_STATUS
);
4903 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
4904 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
4912 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4913 case KVM_SET_GSI_ROUTING
: {
4914 struct kvm_irq_routing routing
;
4915 struct kvm_irq_routing __user
*urouting
;
4916 struct kvm_irq_routing_entry
*entries
= NULL
;
4919 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
4922 if (!kvm_arch_can_set_irq_routing(kvm
))
4924 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
4930 entries
= vmemdup_user(urouting
->entries
,
4931 array_size(sizeof(*entries
),
4933 if (IS_ERR(entries
)) {
4934 r
= PTR_ERR(entries
);
4938 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
4943 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4944 case KVM_CREATE_DEVICE
: {
4945 struct kvm_create_device cd
;
4948 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
4951 r
= kvm_ioctl_create_device(kvm
, &cd
);
4956 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
4962 case KVM_CHECK_EXTENSION
:
4963 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
4965 case KVM_RESET_DIRTY_RINGS
:
4966 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
4968 case KVM_GET_STATS_FD
:
4969 r
= kvm_vm_ioctl_get_stats_fd(kvm
);
4972 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
4978 #ifdef CONFIG_KVM_COMPAT
4979 struct compat_kvm_dirty_log
{
4983 compat_uptr_t dirty_bitmap
; /* one bit per page */
4988 struct compat_kvm_clear_dirty_log
{
4993 compat_uptr_t dirty_bitmap
; /* one bit per page */
4998 long __weak
kvm_arch_vm_compat_ioctl(struct file
*filp
, unsigned int ioctl
,
5004 static long kvm_vm_compat_ioctl(struct file
*filp
,
5005 unsigned int ioctl
, unsigned long arg
)
5007 struct kvm
*kvm
= filp
->private_data
;
5010 if (kvm
->mm
!= current
->mm
|| kvm
->vm_dead
)
5013 r
= kvm_arch_vm_compat_ioctl(filp
, ioctl
, arg
);
5018 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5019 case KVM_CLEAR_DIRTY_LOG
: {
5020 struct compat_kvm_clear_dirty_log compat_log
;
5021 struct kvm_clear_dirty_log log
;
5023 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5024 sizeof(compat_log
)))
5026 log
.slot
= compat_log
.slot
;
5027 log
.num_pages
= compat_log
.num_pages
;
5028 log
.first_page
= compat_log
.first_page
;
5029 log
.padding2
= compat_log
.padding2
;
5030 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5032 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
5036 case KVM_GET_DIRTY_LOG
: {
5037 struct compat_kvm_dirty_log compat_log
;
5038 struct kvm_dirty_log log
;
5040 if (copy_from_user(&compat_log
, (void __user
*)arg
,
5041 sizeof(compat_log
)))
5043 log
.slot
= compat_log
.slot
;
5044 log
.padding1
= compat_log
.padding1
;
5045 log
.padding2
= compat_log
.padding2
;
5046 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
5048 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
5052 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
5058 static struct file_operations kvm_vm_fops
= {
5059 .release
= kvm_vm_release
,
5060 .unlocked_ioctl
= kvm_vm_ioctl
,
5061 .llseek
= noop_llseek
,
5062 KVM_COMPAT(kvm_vm_compat_ioctl
),
5065 bool file_is_kvm(struct file
*file
)
5067 return file
&& file
->f_op
== &kvm_vm_fops
;
5069 EXPORT_SYMBOL_GPL(file_is_kvm
);
5071 static int kvm_dev_ioctl_create_vm(unsigned long type
)
5073 char fdname
[ITOA_MAX_LEN
+ 1];
5078 fd
= get_unused_fd_flags(O_CLOEXEC
);
5082 snprintf(fdname
, sizeof(fdname
), "%d", fd
);
5084 kvm
= kvm_create_vm(type
, fdname
);
5090 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
5097 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5098 * already set, with ->release() being kvm_vm_release(). In error
5099 * cases it will be called by the final fput(file) and will take
5100 * care of doing kvm_put_kvm(kvm).
5102 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
5104 fd_install(fd
, file
);
5114 static long kvm_dev_ioctl(struct file
*filp
,
5115 unsigned int ioctl
, unsigned long arg
)
5120 case KVM_GET_API_VERSION
:
5123 r
= KVM_API_VERSION
;
5126 r
= kvm_dev_ioctl_create_vm(arg
);
5128 case KVM_CHECK_EXTENSION
:
5129 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
5131 case KVM_GET_VCPU_MMAP_SIZE
:
5134 r
= PAGE_SIZE
; /* struct kvm_run */
5136 r
+= PAGE_SIZE
; /* pio data page */
5138 #ifdef CONFIG_KVM_MMIO
5139 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
5142 case KVM_TRACE_ENABLE
:
5143 case KVM_TRACE_PAUSE
:
5144 case KVM_TRACE_DISABLE
:
5148 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
5154 static struct file_operations kvm_chardev_ops
= {
5155 .unlocked_ioctl
= kvm_dev_ioctl
,
5156 .llseek
= noop_llseek
,
5157 KVM_COMPAT(kvm_dev_ioctl
),
5160 static struct miscdevice kvm_dev
= {
5166 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5167 __visible
bool kvm_rebooting
;
5168 EXPORT_SYMBOL_GPL(kvm_rebooting
);
5170 static DEFINE_PER_CPU(bool, hardware_enabled
);
5171 static int kvm_usage_count
;
5173 static int __hardware_enable_nolock(void)
5175 if (__this_cpu_read(hardware_enabled
))
5178 if (kvm_arch_hardware_enable()) {
5179 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5180 raw_smp_processor_id());
5184 __this_cpu_write(hardware_enabled
, true);
5188 static void hardware_enable_nolock(void *failed
)
5190 if (__hardware_enable_nolock())
5194 static int kvm_online_cpu(unsigned int cpu
)
5199 * Abort the CPU online process if hardware virtualization cannot
5200 * be enabled. Otherwise running VMs would encounter unrecoverable
5201 * errors when scheduled to this CPU.
5203 mutex_lock(&kvm_lock
);
5204 if (kvm_usage_count
)
5205 ret
= __hardware_enable_nolock();
5206 mutex_unlock(&kvm_lock
);
5210 static void hardware_disable_nolock(void *junk
)
5213 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5214 * hardware, not just CPUs that successfully enabled hardware!
5216 if (!__this_cpu_read(hardware_enabled
))
5219 kvm_arch_hardware_disable();
5221 __this_cpu_write(hardware_enabled
, false);
5224 static int kvm_offline_cpu(unsigned int cpu
)
5226 mutex_lock(&kvm_lock
);
5227 if (kvm_usage_count
)
5228 hardware_disable_nolock(NULL
);
5229 mutex_unlock(&kvm_lock
);
5233 static void hardware_disable_all_nolock(void)
5235 BUG_ON(!kvm_usage_count
);
5238 if (!kvm_usage_count
)
5239 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5242 static void hardware_disable_all(void)
5245 mutex_lock(&kvm_lock
);
5246 hardware_disable_all_nolock();
5247 mutex_unlock(&kvm_lock
);
5251 static int hardware_enable_all(void)
5253 atomic_t failed
= ATOMIC_INIT(0);
5257 * Do not enable hardware virtualization if the system is going down.
5258 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5259 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5260 * after kvm_reboot() is called. Note, this relies on system_state
5261 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5262 * hook instead of registering a dedicated reboot notifier (the latter
5263 * runs before system_state is updated).
5265 if (system_state
== SYSTEM_HALT
|| system_state
== SYSTEM_POWER_OFF
||
5266 system_state
== SYSTEM_RESTART
)
5270 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5271 * is called, and so on_each_cpu() between them includes the CPU that
5272 * is being onlined. As a result, hardware_enable_nolock() may get
5273 * invoked before kvm_online_cpu(), which also enables hardware if the
5274 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5275 * enable hardware multiple times.
5278 mutex_lock(&kvm_lock
);
5283 if (kvm_usage_count
== 1) {
5284 on_each_cpu(hardware_enable_nolock
, &failed
, 1);
5286 if (atomic_read(&failed
)) {
5287 hardware_disable_all_nolock();
5292 mutex_unlock(&kvm_lock
);
5298 static void kvm_shutdown(void)
5301 * Disable hardware virtualization and set kvm_rebooting to indicate
5302 * that KVM has asynchronously disabled hardware virtualization, i.e.
5303 * that relevant errors and exceptions aren't entirely unexpected.
5304 * Some flavors of hardware virtualization need to be disabled before
5305 * transferring control to firmware (to perform shutdown/reboot), e.g.
5306 * on x86, virtualization can block INIT interrupts, which are used by
5307 * firmware to pull APs back under firmware control. Note, this path
5308 * is used for both shutdown and reboot scenarios, i.e. neither name is
5309 * 100% comprehensive.
5311 pr_info("kvm: exiting hardware virtualization\n");
5312 kvm_rebooting
= true;
5313 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5316 static int kvm_suspend(void)
5319 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5320 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5321 * is stable. Assert that kvm_lock is not held to ensure the system
5322 * isn't suspended while KVM is enabling hardware. Hardware enabling
5323 * can be preempted, but the task cannot be frozen until it has dropped
5324 * all locks (userspace tasks are frozen via a fake signal).
5326 lockdep_assert_not_held(&kvm_lock
);
5327 lockdep_assert_irqs_disabled();
5329 if (kvm_usage_count
)
5330 hardware_disable_nolock(NULL
);
5334 static void kvm_resume(void)
5336 lockdep_assert_not_held(&kvm_lock
);
5337 lockdep_assert_irqs_disabled();
5339 if (kvm_usage_count
)
5340 WARN_ON_ONCE(__hardware_enable_nolock());
5343 static struct syscore_ops kvm_syscore_ops
= {
5344 .suspend
= kvm_suspend
,
5345 .resume
= kvm_resume
,
5346 .shutdown
= kvm_shutdown
,
5348 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5349 static int hardware_enable_all(void)
5354 static void hardware_disable_all(void)
5358 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5360 static void kvm_iodevice_destructor(struct kvm_io_device
*dev
)
5362 if (dev
->ops
->destructor
)
5363 dev
->ops
->destructor(dev
);
5366 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
5370 for (i
= 0; i
< bus
->dev_count
; i
++) {
5371 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
5373 kvm_iodevice_destructor(pos
);
5378 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
5379 const struct kvm_io_range
*r2
)
5381 gpa_t addr1
= r1
->addr
;
5382 gpa_t addr2
= r2
->addr
;
5387 /* If r2->len == 0, match the exact address. If r2->len != 0,
5388 * accept any overlapping write. Any order is acceptable for
5389 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5390 * we process all of them.
5403 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
5405 return kvm_io_bus_cmp(p1
, p2
);
5408 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
5409 gpa_t addr
, int len
)
5411 struct kvm_io_range
*range
, key
;
5414 key
= (struct kvm_io_range
) {
5419 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
5420 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
5424 off
= range
- bus
->range
;
5426 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
5432 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5433 struct kvm_io_range
*range
, const void *val
)
5437 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5441 while (idx
< bus
->dev_count
&&
5442 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5443 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5452 /* kvm_io_bus_write - called under kvm->slots_lock */
5453 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5454 int len
, const void *val
)
5456 struct kvm_io_bus
*bus
;
5457 struct kvm_io_range range
;
5460 range
= (struct kvm_io_range
) {
5465 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5468 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5469 return r
< 0 ? r
: 0;
5471 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
5473 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5474 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
5475 gpa_t addr
, int len
, const void *val
, long cookie
)
5477 struct kvm_io_bus
*bus
;
5478 struct kvm_io_range range
;
5480 range
= (struct kvm_io_range
) {
5485 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5489 /* First try the device referenced by cookie. */
5490 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
5491 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
5492 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
5497 * cookie contained garbage; fall back to search and return the
5498 * correct cookie value.
5500 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
5503 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
5504 struct kvm_io_range
*range
, void *val
)
5508 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
5512 while (idx
< bus
->dev_count
&&
5513 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
5514 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
5523 /* kvm_io_bus_read - called under kvm->slots_lock */
5524 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
5527 struct kvm_io_bus
*bus
;
5528 struct kvm_io_range range
;
5531 range
= (struct kvm_io_range
) {
5536 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
5539 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
5540 return r
< 0 ? r
: 0;
5543 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
5544 int len
, struct kvm_io_device
*dev
)
5547 struct kvm_io_bus
*new_bus
, *bus
;
5548 struct kvm_io_range range
;
5550 lockdep_assert_held(&kvm
->slots_lock
);
5552 bus
= kvm_get_bus(kvm
, bus_idx
);
5556 /* exclude ioeventfd which is limited by maximum fd */
5557 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
5560 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
5561 GFP_KERNEL_ACCOUNT
);
5565 range
= (struct kvm_io_range
) {
5571 for (i
= 0; i
< bus
->dev_count
; i
++)
5572 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
5575 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
5576 new_bus
->dev_count
++;
5577 new_bus
->range
[i
] = range
;
5578 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
5579 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
5580 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5581 synchronize_srcu_expedited(&kvm
->srcu
);
5587 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5588 struct kvm_io_device
*dev
)
5591 struct kvm_io_bus
*new_bus
, *bus
;
5593 lockdep_assert_held(&kvm
->slots_lock
);
5595 bus
= kvm_get_bus(kvm
, bus_idx
);
5599 for (i
= 0; i
< bus
->dev_count
; i
++) {
5600 if (bus
->range
[i
].dev
== dev
) {
5605 if (i
== bus
->dev_count
)
5608 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
5609 GFP_KERNEL_ACCOUNT
);
5611 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
5612 new_bus
->dev_count
--;
5613 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
5614 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
5617 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
5618 synchronize_srcu_expedited(&kvm
->srcu
);
5621 * If NULL bus is installed, destroy the old bus, including all the
5622 * attached devices. Otherwise, destroy the caller's device only.
5625 pr_err("kvm: failed to shrink bus, removing it completely\n");
5626 kvm_io_bus_destroy(bus
);
5630 kvm_iodevice_destructor(dev
);
5635 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
5638 struct kvm_io_bus
*bus
;
5639 int dev_idx
, srcu_idx
;
5640 struct kvm_io_device
*iodev
= NULL
;
5642 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
5644 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
5648 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
5652 iodev
= bus
->range
[dev_idx
].dev
;
5655 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
5659 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
5661 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
5662 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
5666 struct kvm_stat_data
*stat_data
= inode
->i_private
;
5669 * The debugfs files are a reference to the kvm struct which
5670 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5671 * avoids the race between open and the removal of the debugfs directory.
5673 if (!kvm_get_kvm_safe(stat_data
->kvm
))
5676 ret
= simple_attr_open(inode
, file
, get
,
5677 kvm_stats_debugfs_mode(stat_data
->desc
) & 0222
5680 kvm_put_kvm(stat_data
->kvm
);
5685 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
5687 struct kvm_stat_data
*stat_data
= inode
->i_private
;
5689 simple_attr_release(inode
, file
);
5690 kvm_put_kvm(stat_data
->kvm
);
5695 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
5697 *val
= *(u64
*)((void *)(&kvm
->stat
) + offset
);
5702 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
5704 *(u64
*)((void *)(&kvm
->stat
) + offset
) = 0;
5709 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
5712 struct kvm_vcpu
*vcpu
;
5716 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5717 *val
+= *(u64
*)((void *)(&vcpu
->stat
) + offset
);
5722 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
5725 struct kvm_vcpu
*vcpu
;
5727 kvm_for_each_vcpu(i
, vcpu
, kvm
)
5728 *(u64
*)((void *)(&vcpu
->stat
) + offset
) = 0;
5733 static int kvm_stat_data_get(void *data
, u64
*val
)
5736 struct kvm_stat_data
*stat_data
= data
;
5738 switch (stat_data
->kind
) {
5740 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
5741 stat_data
->desc
->desc
.offset
, val
);
5744 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
5745 stat_data
->desc
->desc
.offset
, val
);
5752 static int kvm_stat_data_clear(void *data
, u64 val
)
5755 struct kvm_stat_data
*stat_data
= data
;
5760 switch (stat_data
->kind
) {
5762 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
5763 stat_data
->desc
->desc
.offset
);
5766 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
5767 stat_data
->desc
->desc
.offset
);
5774 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
5776 __simple_attr_check_format("%llu\n", 0ull);
5777 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
5778 kvm_stat_data_clear
, "%llu\n");
5781 static const struct file_operations stat_fops_per_vm
= {
5782 .owner
= THIS_MODULE
,
5783 .open
= kvm_stat_data_open
,
5784 .release
= kvm_debugfs_release
,
5785 .read
= simple_attr_read
,
5786 .write
= simple_attr_write
,
5787 .llseek
= no_llseek
,
5790 static int vm_stat_get(void *_offset
, u64
*val
)
5792 unsigned offset
= (long)_offset
;
5797 mutex_lock(&kvm_lock
);
5798 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5799 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
5802 mutex_unlock(&kvm_lock
);
5806 static int vm_stat_clear(void *_offset
, u64 val
)
5808 unsigned offset
= (long)_offset
;
5814 mutex_lock(&kvm_lock
);
5815 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5816 kvm_clear_stat_per_vm(kvm
, offset
);
5818 mutex_unlock(&kvm_lock
);
5823 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
5824 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops
, vm_stat_get
, NULL
, "%llu\n");
5826 static int vcpu_stat_get(void *_offset
, u64
*val
)
5828 unsigned offset
= (long)_offset
;
5833 mutex_lock(&kvm_lock
);
5834 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5835 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
5838 mutex_unlock(&kvm_lock
);
5842 static int vcpu_stat_clear(void *_offset
, u64 val
)
5844 unsigned offset
= (long)_offset
;
5850 mutex_lock(&kvm_lock
);
5851 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
5852 kvm_clear_stat_per_vcpu(kvm
, offset
);
5854 mutex_unlock(&kvm_lock
);
5859 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
5861 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops
, vcpu_stat_get
, NULL
, "%llu\n");
5863 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
5865 struct kobj_uevent_env
*env
;
5866 unsigned long long created
, active
;
5868 if (!kvm_dev
.this_device
|| !kvm
)
5871 mutex_lock(&kvm_lock
);
5872 if (type
== KVM_EVENT_CREATE_VM
) {
5873 kvm_createvm_count
++;
5875 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5878 created
= kvm_createvm_count
;
5879 active
= kvm_active_vms
;
5880 mutex_unlock(&kvm_lock
);
5882 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
5886 add_uevent_var(env
, "CREATED=%llu", created
);
5887 add_uevent_var(env
, "COUNT=%llu", active
);
5889 if (type
== KVM_EVENT_CREATE_VM
) {
5890 add_uevent_var(env
, "EVENT=create");
5891 kvm
->userspace_pid
= task_pid_nr(current
);
5892 } else if (type
== KVM_EVENT_DESTROY_VM
) {
5893 add_uevent_var(env
, "EVENT=destroy");
5895 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
5897 if (!IS_ERR(kvm
->debugfs_dentry
)) {
5898 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
5901 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
5903 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
5907 /* no need for checks, since we are adding at most only 5 keys */
5908 env
->envp
[env
->envp_idx
++] = NULL
;
5909 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
5913 static void kvm_init_debug(void)
5915 const struct file_operations
*fops
;
5916 const struct _kvm_stats_desc
*pdesc
;
5919 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
5921 for (i
= 0; i
< kvm_vm_stats_header
.num_desc
; ++i
) {
5922 pdesc
= &kvm_vm_stats_desc
[i
];
5923 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
5924 fops
= &vm_stat_fops
;
5926 fops
= &vm_stat_readonly_fops
;
5927 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
5929 (void *)(long)pdesc
->desc
.offset
, fops
);
5932 for (i
= 0; i
< kvm_vcpu_stats_header
.num_desc
; ++i
) {
5933 pdesc
= &kvm_vcpu_stats_desc
[i
];
5934 if (kvm_stats_debugfs_mode(pdesc
) & 0222)
5935 fops
= &vcpu_stat_fops
;
5937 fops
= &vcpu_stat_readonly_fops
;
5938 debugfs_create_file(pdesc
->name
, kvm_stats_debugfs_mode(pdesc
),
5940 (void *)(long)pdesc
->desc
.offset
, fops
);
5945 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
5947 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
5950 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
5952 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
5954 WRITE_ONCE(vcpu
->preempted
, false);
5955 WRITE_ONCE(vcpu
->ready
, false);
5957 __this_cpu_write(kvm_running_vcpu
, vcpu
);
5958 kvm_arch_sched_in(vcpu
, cpu
);
5959 kvm_arch_vcpu_load(vcpu
, cpu
);
5962 static void kvm_sched_out(struct preempt_notifier
*pn
,
5963 struct task_struct
*next
)
5965 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
5967 if (current
->on_rq
) {
5968 WRITE_ONCE(vcpu
->preempted
, true);
5969 WRITE_ONCE(vcpu
->ready
, true);
5971 kvm_arch_vcpu_put(vcpu
);
5972 __this_cpu_write(kvm_running_vcpu
, NULL
);
5976 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5978 * We can disable preemption locally around accessing the per-CPU variable,
5979 * and use the resolved vcpu pointer after enabling preemption again,
5980 * because even if the current thread is migrated to another CPU, reading
5981 * the per-CPU value later will give us the same value as we update the
5982 * per-CPU variable in the preempt notifier handlers.
5984 struct kvm_vcpu
*kvm_get_running_vcpu(void)
5986 struct kvm_vcpu
*vcpu
;
5989 vcpu
= __this_cpu_read(kvm_running_vcpu
);
5994 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
5997 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5999 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
6001 return &kvm_running_vcpu
;
6004 #ifdef CONFIG_GUEST_PERF_EVENTS
6005 static unsigned int kvm_guest_state(void)
6007 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6010 if (!kvm_arch_pmi_in_guest(vcpu
))
6013 state
= PERF_GUEST_ACTIVE
;
6014 if (!kvm_arch_vcpu_in_kernel(vcpu
))
6015 state
|= PERF_GUEST_USER
;
6020 static unsigned long kvm_guest_get_ip(void)
6022 struct kvm_vcpu
*vcpu
= kvm_get_running_vcpu();
6024 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6025 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu
)))
6028 return kvm_arch_vcpu_get_ip(vcpu
);
6031 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6032 .state
= kvm_guest_state
,
6033 .get_ip
= kvm_guest_get_ip
,
6034 .handle_intel_pt_intr
= NULL
,
6037 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler
)(void))
6039 kvm_guest_cbs
.handle_intel_pt_intr
= pt_intr_handler
;
6040 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6042 void kvm_unregister_perf_callbacks(void)
6044 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6048 int kvm_init(unsigned vcpu_size
, unsigned vcpu_align
, struct module
*module
)
6053 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6054 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE
, "kvm/cpu:online",
6055 kvm_online_cpu
, kvm_offline_cpu
);
6059 register_syscore_ops(&kvm_syscore_ops
);
6062 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6064 vcpu_align
= __alignof__(struct kvm_vcpu
);
6066 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
6068 offsetof(struct kvm_vcpu
, arch
),
6069 offsetofend(struct kvm_vcpu
, stats_id
)
6070 - offsetof(struct kvm_vcpu
, arch
),
6072 if (!kvm_vcpu_cache
) {
6074 goto err_vcpu_cache
;
6077 for_each_possible_cpu(cpu
) {
6078 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask
, cpu
),
6079 GFP_KERNEL
, cpu_to_node(cpu
))) {
6081 goto err_cpu_kick_mask
;
6085 r
= kvm_irqfd_init();
6089 r
= kvm_async_pf_init();
6093 kvm_chardev_ops
.owner
= module
;
6094 kvm_vm_fops
.owner
= module
;
6095 kvm_vcpu_fops
.owner
= module
;
6096 kvm_device_fops
.owner
= module
;
6098 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
6099 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
6103 r
= kvm_vfio_ops_init();
6104 if (WARN_ON_ONCE(r
))
6108 * Registration _must_ be the very last thing done, as this exposes
6109 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6111 r
= misc_register(&kvm_dev
);
6113 pr_err("kvm: misc device register failed\n");
6120 kvm_vfio_ops_exit();
6122 kvm_async_pf_deinit();
6127 for_each_possible_cpu(cpu
)
6128 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6129 kmem_cache_destroy(kvm_vcpu_cache
);
6131 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6132 unregister_syscore_ops(&kvm_syscore_ops
);
6133 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6137 EXPORT_SYMBOL_GPL(kvm_init
);
6144 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6145 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6146 * to KVM while the module is being stopped.
6148 misc_deregister(&kvm_dev
);
6150 debugfs_remove_recursive(kvm_debugfs_dir
);
6151 for_each_possible_cpu(cpu
)
6152 free_cpumask_var(per_cpu(cpu_kick_mask
, cpu
));
6153 kmem_cache_destroy(kvm_vcpu_cache
);
6154 kvm_vfio_ops_exit();
6155 kvm_async_pf_deinit();
6156 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6157 unregister_syscore_ops(&kvm_syscore_ops
);
6158 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE
);
6162 EXPORT_SYMBOL_GPL(kvm_exit
);
6164 struct kvm_vm_worker_thread_context
{
6166 struct task_struct
*parent
;
6167 struct completion init_done
;
6168 kvm_vm_thread_fn_t thread_fn
;
6173 static int kvm_vm_worker_thread(void *context
)
6176 * The init_context is allocated on the stack of the parent thread, so
6177 * we have to locally copy anything that is needed beyond initialization
6179 struct kvm_vm_worker_thread_context
*init_context
= context
;
6180 struct task_struct
*parent
;
6181 struct kvm
*kvm
= init_context
->kvm
;
6182 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
6183 uintptr_t data
= init_context
->data
;
6186 err
= kthread_park(current
);
6187 /* kthread_park(current) is never supposed to return an error */
6192 err
= cgroup_attach_task_all(init_context
->parent
, current
);
6194 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6199 set_user_nice(current
, task_nice(init_context
->parent
));
6202 init_context
->err
= err
;
6203 complete(&init_context
->init_done
);
6204 init_context
= NULL
;
6209 /* Wait to be woken up by the spawner before proceeding. */
6212 if (!kthread_should_stop())
6213 err
= thread_fn(kvm
, data
);
6217 * Move kthread back to its original cgroup to prevent it lingering in
6218 * the cgroup of the VM process, after the latter finishes its
6221 * kthread_stop() waits on the 'exited' completion condition which is
6222 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6223 * kthread is removed from the cgroup in the cgroup_exit() which is
6224 * called after the exit_mm(). This causes the kthread_stop() to return
6225 * before the kthread actually quits the cgroup.
6228 parent
= rcu_dereference(current
->real_parent
);
6229 get_task_struct(parent
);
6231 cgroup_attach_task_all(parent
, current
);
6232 put_task_struct(parent
);
6237 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
6238 uintptr_t data
, const char *name
,
6239 struct task_struct
**thread_ptr
)
6241 struct kvm_vm_worker_thread_context init_context
= {};
6242 struct task_struct
*thread
;
6245 init_context
.kvm
= kvm
;
6246 init_context
.parent
= current
;
6247 init_context
.thread_fn
= thread_fn
;
6248 init_context
.data
= data
;
6249 init_completion(&init_context
.init_done
);
6251 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
6252 "%s-%d", name
, task_pid_nr(current
));
6254 return PTR_ERR(thread
);
6256 /* kthread_run is never supposed to return NULL */
6257 WARN_ON(thread
== NULL
);
6259 wait_for_completion(&init_context
.init_done
);
6261 if (!init_context
.err
)
6262 *thread_ptr
= thread
;
6264 return init_context
.err
;