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>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
75 module_param(halt_poll_ns
, uint
, 0644);
76 EXPORT_SYMBOL_GPL(halt_poll_ns
);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow
= 2;
80 module_param(halt_poll_ns_grow
, uint
, 0644);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
83 /* The start value to grow halt_poll_ns from */
84 unsigned int halt_poll_ns_grow_start
= 10000; /* 10us */
85 module_param(halt_poll_ns_grow_start
, uint
, 0644);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start
);
88 /* Default resets per-vcpu halt_poll_ns . */
89 unsigned int halt_poll_ns_shrink
;
90 module_param(halt_poll_ns_shrink
, uint
, 0644);
91 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
96 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
99 DEFINE_MUTEX(kvm_lock
);
100 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
103 static cpumask_var_t cpus_hardware_enabled
;
104 static int kvm_usage_count
;
105 static atomic_t hardware_enable_failed
;
107 static struct kmem_cache
*kvm_vcpu_cache
;
109 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
110 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_running_vcpu
);
112 struct dentry
*kvm_debugfs_dir
;
113 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
115 static int kvm_debugfs_num_entries
;
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 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
149 __visible
bool kvm_rebooting
;
150 EXPORT_SYMBOL_GPL(kvm_rebooting
);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
155 static unsigned long long kvm_createvm_count
;
156 static unsigned long long kvm_active_vms
;
158 __weak
int kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
159 unsigned long start
, unsigned long end
, bool blockable
)
164 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn
)
167 * The metadata used by is_zone_device_page() to determine whether or
168 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
169 * the device has been pinned, e.g. by get_user_pages(). WARN if the
170 * page_count() is zero to help detect bad usage of this helper.
172 if (!pfn_valid(pfn
) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn
))))
175 return is_zone_device_page(pfn_to_page(pfn
));
178 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
181 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
182 * perspective they are "normal" pages, albeit with slightly different
186 return PageReserved(pfn_to_page(pfn
)) &&
188 !kvm_is_zone_device_pfn(pfn
);
193 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn
)
195 struct page
*page
= pfn_to_page(pfn
);
197 if (!PageTransCompoundMap(page
))
200 return is_transparent_hugepage(compound_head(page
));
204 * Switches to specified vcpu, until a matching vcpu_put()
206 void vcpu_load(struct kvm_vcpu
*vcpu
)
210 __this_cpu_write(kvm_running_vcpu
, vcpu
);
211 preempt_notifier_register(&vcpu
->preempt_notifier
);
212 kvm_arch_vcpu_load(vcpu
, cpu
);
215 EXPORT_SYMBOL_GPL(vcpu_load
);
217 void vcpu_put(struct kvm_vcpu
*vcpu
)
220 kvm_arch_vcpu_put(vcpu
);
221 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
222 __this_cpu_write(kvm_running_vcpu
, NULL
);
225 EXPORT_SYMBOL_GPL(vcpu_put
);
227 /* TODO: merge with kvm_arch_vcpu_should_kick */
228 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
230 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
233 * We need to wait for the VCPU to reenable interrupts and get out of
234 * READING_SHADOW_PAGE_TABLES mode.
236 if (req
& KVM_REQUEST_WAIT
)
237 return mode
!= OUTSIDE_GUEST_MODE
;
240 * Need to kick a running VCPU, but otherwise there is nothing to do.
242 return mode
== IN_GUEST_MODE
;
245 static void ack_flush(void *_completed
)
249 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
252 cpus
= cpu_online_mask
;
254 if (cpumask_empty(cpus
))
257 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
261 bool kvm_make_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
262 unsigned long *vcpu_bitmap
, cpumask_var_t tmp
)
265 struct kvm_vcpu
*vcpu
;
270 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
271 if (vcpu_bitmap
&& !test_bit(i
, vcpu_bitmap
))
274 kvm_make_request(req
, vcpu
);
277 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
280 if (tmp
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
281 kvm_request_needs_ipi(vcpu
, req
))
282 __cpumask_set_cpu(cpu
, tmp
);
285 called
= kvm_kick_many_cpus(tmp
, !!(req
& KVM_REQUEST_WAIT
));
291 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
296 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
298 called
= kvm_make_vcpus_request_mask(kvm
, req
, NULL
, cpus
);
300 free_cpumask_var(cpus
);
304 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
305 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
308 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
309 * kvm_make_all_cpus_request.
311 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
314 * We want to publish modifications to the page tables before reading
315 * mode. Pairs with a memory barrier in arch-specific code.
316 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
317 * and smp_mb in walk_shadow_page_lockless_begin/end.
318 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
320 * There is already an smp_mb__after_atomic() before
321 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
324 if (!kvm_arch_flush_remote_tlb(kvm
)
325 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
326 ++kvm
->stat
.remote_tlb_flush
;
327 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
329 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
332 void kvm_reload_remote_mmus(struct kvm
*kvm
)
334 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
337 static void kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
339 mutex_init(&vcpu
->mutex
);
344 init_swait_queue_head(&vcpu
->wq
);
345 kvm_async_pf_vcpu_init(vcpu
);
348 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
350 kvm_vcpu_set_in_spin_loop(vcpu
, false);
351 kvm_vcpu_set_dy_eligible(vcpu
, false);
352 vcpu
->preempted
= false;
354 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
357 void kvm_vcpu_destroy(struct kvm_vcpu
*vcpu
)
359 kvm_arch_vcpu_destroy(vcpu
);
362 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
363 * the vcpu->pid pointer, and at destruction time all file descriptors
366 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
368 free_page((unsigned long)vcpu
->run
);
369 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
371 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy
);
373 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
374 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
376 return container_of(mn
, struct kvm
, mmu_notifier
);
379 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
380 struct mm_struct
*mm
,
381 unsigned long address
,
384 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
387 idx
= srcu_read_lock(&kvm
->srcu
);
388 spin_lock(&kvm
->mmu_lock
);
389 kvm
->mmu_notifier_seq
++;
391 if (kvm_set_spte_hva(kvm
, address
, pte
))
392 kvm_flush_remote_tlbs(kvm
);
394 spin_unlock(&kvm
->mmu_lock
);
395 srcu_read_unlock(&kvm
->srcu
, idx
);
398 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
399 const struct mmu_notifier_range
*range
)
401 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
402 int need_tlb_flush
= 0, idx
;
405 idx
= srcu_read_lock(&kvm
->srcu
);
406 spin_lock(&kvm
->mmu_lock
);
408 * The count increase must become visible at unlock time as no
409 * spte can be established without taking the mmu_lock and
410 * count is also read inside the mmu_lock critical section.
412 kvm
->mmu_notifier_count
++;
413 need_tlb_flush
= kvm_unmap_hva_range(kvm
, range
->start
, range
->end
);
414 need_tlb_flush
|= kvm
->tlbs_dirty
;
415 /* we've to flush the tlb before the pages can be freed */
417 kvm_flush_remote_tlbs(kvm
);
419 spin_unlock(&kvm
->mmu_lock
);
421 ret
= kvm_arch_mmu_notifier_invalidate_range(kvm
, range
->start
,
423 mmu_notifier_range_blockable(range
));
425 srcu_read_unlock(&kvm
->srcu
, idx
);
430 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
431 const struct mmu_notifier_range
*range
)
433 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
435 spin_lock(&kvm
->mmu_lock
);
437 * This sequence increase will notify the kvm page fault that
438 * the page that is going to be mapped in the spte could have
441 kvm
->mmu_notifier_seq
++;
444 * The above sequence increase must be visible before the
445 * below count decrease, which is ensured by the smp_wmb above
446 * in conjunction with the smp_rmb in mmu_notifier_retry().
448 kvm
->mmu_notifier_count
--;
449 spin_unlock(&kvm
->mmu_lock
);
451 BUG_ON(kvm
->mmu_notifier_count
< 0);
454 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
455 struct mm_struct
*mm
,
459 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
462 idx
= srcu_read_lock(&kvm
->srcu
);
463 spin_lock(&kvm
->mmu_lock
);
465 young
= kvm_age_hva(kvm
, start
, end
);
467 kvm_flush_remote_tlbs(kvm
);
469 spin_unlock(&kvm
->mmu_lock
);
470 srcu_read_unlock(&kvm
->srcu
, idx
);
475 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
476 struct mm_struct
*mm
,
480 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
483 idx
= srcu_read_lock(&kvm
->srcu
);
484 spin_lock(&kvm
->mmu_lock
);
486 * Even though we do not flush TLB, this will still adversely
487 * affect performance on pre-Haswell Intel EPT, where there is
488 * no EPT Access Bit to clear so that we have to tear down EPT
489 * tables instead. If we find this unacceptable, we can always
490 * add a parameter to kvm_age_hva so that it effectively doesn't
491 * do anything on clear_young.
493 * Also note that currently we never issue secondary TLB flushes
494 * from clear_young, leaving this job up to the regular system
495 * cadence. If we find this inaccurate, we might come up with a
496 * more sophisticated heuristic later.
498 young
= kvm_age_hva(kvm
, start
, end
);
499 spin_unlock(&kvm
->mmu_lock
);
500 srcu_read_unlock(&kvm
->srcu
, idx
);
505 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
506 struct mm_struct
*mm
,
507 unsigned long address
)
509 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
512 idx
= srcu_read_lock(&kvm
->srcu
);
513 spin_lock(&kvm
->mmu_lock
);
514 young
= kvm_test_age_hva(kvm
, address
);
515 spin_unlock(&kvm
->mmu_lock
);
516 srcu_read_unlock(&kvm
->srcu
, idx
);
521 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
522 struct mm_struct
*mm
)
524 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
527 idx
= srcu_read_lock(&kvm
->srcu
);
528 kvm_arch_flush_shadow_all(kvm
);
529 srcu_read_unlock(&kvm
->srcu
, idx
);
532 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
533 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
534 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
535 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
536 .clear_young
= kvm_mmu_notifier_clear_young
,
537 .test_young
= kvm_mmu_notifier_test_young
,
538 .change_pte
= kvm_mmu_notifier_change_pte
,
539 .release
= kvm_mmu_notifier_release
,
542 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
544 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
545 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
548 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
550 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
555 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
557 static struct kvm_memslots
*kvm_alloc_memslots(void)
560 struct kvm_memslots
*slots
;
562 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL_ACCOUNT
);
566 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
567 slots
->id_to_index
[i
] = -1;
572 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
574 if (!memslot
->dirty_bitmap
)
577 kvfree(memslot
->dirty_bitmap
);
578 memslot
->dirty_bitmap
= NULL
;
581 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
583 kvm_destroy_dirty_bitmap(slot
);
585 kvm_arch_free_memslot(kvm
, slot
);
591 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
593 struct kvm_memory_slot
*memslot
;
598 kvm_for_each_memslot(memslot
, slots
)
599 kvm_free_memslot(kvm
, memslot
);
604 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
608 if (!kvm
->debugfs_dentry
)
611 debugfs_remove_recursive(kvm
->debugfs_dentry
);
613 if (kvm
->debugfs_stat_data
) {
614 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
615 kfree(kvm
->debugfs_stat_data
[i
]);
616 kfree(kvm
->debugfs_stat_data
);
620 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
622 char dir_name
[ITOA_MAX_LEN
* 2];
623 struct kvm_stat_data
*stat_data
;
624 struct kvm_stats_debugfs_item
*p
;
626 if (!debugfs_initialized())
629 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
630 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
632 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
633 sizeof(*kvm
->debugfs_stat_data
),
635 if (!kvm
->debugfs_stat_data
)
638 for (p
= debugfs_entries
; p
->name
; p
++) {
639 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
643 stat_data
->kvm
= kvm
;
644 stat_data
->dbgfs_item
= p
;
645 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
646 debugfs_create_file(p
->name
, KVM_DBGFS_GET_MODE(p
),
647 kvm
->debugfs_dentry
, stat_data
,
654 * Called after the VM is otherwise initialized, but just before adding it to
657 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
663 * Called just after removing the VM from the vm_list, but before doing any
666 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
670 static struct kvm
*kvm_create_vm(unsigned long type
)
672 struct kvm
*kvm
= kvm_arch_alloc_vm();
677 return ERR_PTR(-ENOMEM
);
679 spin_lock_init(&kvm
->mmu_lock
);
681 kvm
->mm
= current
->mm
;
682 kvm_eventfd_init(kvm
);
683 mutex_init(&kvm
->lock
);
684 mutex_init(&kvm
->irq_lock
);
685 mutex_init(&kvm
->slots_lock
);
686 INIT_LIST_HEAD(&kvm
->devices
);
688 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
690 if (init_srcu_struct(&kvm
->srcu
))
691 goto out_err_no_srcu
;
692 if (init_srcu_struct(&kvm
->irq_srcu
))
693 goto out_err_no_irq_srcu
;
695 refcount_set(&kvm
->users_count
, 1);
696 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
697 struct kvm_memslots
*slots
= kvm_alloc_memslots();
700 goto out_err_no_arch_destroy_vm
;
701 /* Generations must be different for each address space. */
702 slots
->generation
= i
;
703 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
706 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
707 rcu_assign_pointer(kvm
->buses
[i
],
708 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
710 goto out_err_no_arch_destroy_vm
;
713 r
= kvm_arch_init_vm(kvm
, type
);
715 goto out_err_no_arch_destroy_vm
;
717 r
= hardware_enable_all();
719 goto out_err_no_disable
;
721 #ifdef CONFIG_HAVE_KVM_IRQFD
722 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
725 r
= kvm_init_mmu_notifier(kvm
);
727 goto out_err_no_mmu_notifier
;
729 r
= kvm_arch_post_init_vm(kvm
);
733 mutex_lock(&kvm_lock
);
734 list_add(&kvm
->vm_list
, &vm_list
);
735 mutex_unlock(&kvm_lock
);
737 preempt_notifier_inc();
742 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
743 if (kvm
->mmu_notifier
.ops
)
744 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
746 out_err_no_mmu_notifier
:
747 hardware_disable_all();
749 kvm_arch_destroy_vm(kvm
);
750 out_err_no_arch_destroy_vm
:
751 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
752 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
753 kfree(kvm_get_bus(kvm
, i
));
754 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
755 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
756 cleanup_srcu_struct(&kvm
->irq_srcu
);
758 cleanup_srcu_struct(&kvm
->srcu
);
760 kvm_arch_free_vm(kvm
);
765 static void kvm_destroy_devices(struct kvm
*kvm
)
767 struct kvm_device
*dev
, *tmp
;
770 * We do not need to take the kvm->lock here, because nobody else
771 * has a reference to the struct kvm at this point and therefore
772 * cannot access the devices list anyhow.
774 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
775 list_del(&dev
->vm_node
);
776 dev
->ops
->destroy(dev
);
780 static void kvm_destroy_vm(struct kvm
*kvm
)
783 struct mm_struct
*mm
= kvm
->mm
;
785 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
786 kvm_destroy_vm_debugfs(kvm
);
787 kvm_arch_sync_events(kvm
);
788 mutex_lock(&kvm_lock
);
789 list_del(&kvm
->vm_list
);
790 mutex_unlock(&kvm_lock
);
791 kvm_arch_pre_destroy_vm(kvm
);
793 kvm_free_irq_routing(kvm
);
794 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
795 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
798 kvm_io_bus_destroy(bus
);
799 kvm
->buses
[i
] = NULL
;
801 kvm_coalesced_mmio_free(kvm
);
802 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
803 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
805 kvm_arch_flush_shadow_all(kvm
);
807 kvm_arch_destroy_vm(kvm
);
808 kvm_destroy_devices(kvm
);
809 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
810 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
811 cleanup_srcu_struct(&kvm
->irq_srcu
);
812 cleanup_srcu_struct(&kvm
->srcu
);
813 kvm_arch_free_vm(kvm
);
814 preempt_notifier_dec();
815 hardware_disable_all();
819 void kvm_get_kvm(struct kvm
*kvm
)
821 refcount_inc(&kvm
->users_count
);
823 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
825 void kvm_put_kvm(struct kvm
*kvm
)
827 if (refcount_dec_and_test(&kvm
->users_count
))
830 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
833 * Used to put a reference that was taken on behalf of an object associated
834 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
835 * of the new file descriptor fails and the reference cannot be transferred to
836 * its final owner. In such cases, the caller is still actively using @kvm and
837 * will fail miserably if the refcount unexpectedly hits zero.
839 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
841 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
843 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
845 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
847 struct kvm
*kvm
= filp
->private_data
;
849 kvm_irqfd_release(kvm
);
856 * Allocation size is twice as large as the actual dirty bitmap size.
857 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
859 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
861 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
863 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL_ACCOUNT
);
864 if (!memslot
->dirty_bitmap
)
871 * Delete a memslot by decrementing the number of used slots and shifting all
872 * other entries in the array forward one spot.
874 static inline void kvm_memslot_delete(struct kvm_memslots
*slots
,
875 struct kvm_memory_slot
*memslot
)
877 struct kvm_memory_slot
*mslots
= slots
->memslots
;
880 if (WARN_ON(slots
->id_to_index
[memslot
->id
] == -1))
885 if (atomic_read(&slots
->lru_slot
) >= slots
->used_slots
)
886 atomic_set(&slots
->lru_slot
, 0);
888 for (i
= slots
->id_to_index
[memslot
->id
]; i
< slots
->used_slots
; i
++) {
889 mslots
[i
] = mslots
[i
+ 1];
890 slots
->id_to_index
[mslots
[i
].id
] = i
;
892 mslots
[i
] = *memslot
;
893 slots
->id_to_index
[memslot
->id
] = -1;
897 * "Insert" a new memslot by incrementing the number of used slots. Returns
898 * the new slot's initial index into the memslots array.
900 static inline int kvm_memslot_insert_back(struct kvm_memslots
*slots
)
902 return slots
->used_slots
++;
906 * Move a changed memslot backwards in the array by shifting existing slots
907 * with a higher GFN toward the front of the array. Note, the changed memslot
908 * itself is not preserved in the array, i.e. not swapped at this time, only
909 * its new index into the array is tracked. Returns the changed memslot's
910 * current index into the memslots array.
912 static inline int kvm_memslot_move_backward(struct kvm_memslots
*slots
,
913 struct kvm_memory_slot
*memslot
)
915 struct kvm_memory_slot
*mslots
= slots
->memslots
;
918 if (WARN_ON_ONCE(slots
->id_to_index
[memslot
->id
] == -1) ||
919 WARN_ON_ONCE(!slots
->used_slots
))
923 * Move the target memslot backward in the array by shifting existing
924 * memslots with a higher GFN (than the target memslot) towards the
925 * front of the array.
927 for (i
= slots
->id_to_index
[memslot
->id
]; i
< slots
->used_slots
- 1; i
++) {
928 if (memslot
->base_gfn
> mslots
[i
+ 1].base_gfn
)
931 WARN_ON_ONCE(memslot
->base_gfn
== mslots
[i
+ 1].base_gfn
);
933 /* Shift the next memslot forward one and update its index. */
934 mslots
[i
] = mslots
[i
+ 1];
935 slots
->id_to_index
[mslots
[i
].id
] = i
;
941 * Move a changed memslot forwards in the array by shifting existing slots with
942 * a lower GFN toward the back of the array. Note, the changed memslot itself
943 * is not preserved in the array, i.e. not swapped at this time, only its new
944 * index into the array is tracked. Returns the changed memslot's final index
945 * into the memslots array.
947 static inline int kvm_memslot_move_forward(struct kvm_memslots
*slots
,
948 struct kvm_memory_slot
*memslot
,
951 struct kvm_memory_slot
*mslots
= slots
->memslots
;
954 for (i
= start
; i
> 0; i
--) {
955 if (memslot
->base_gfn
< mslots
[i
- 1].base_gfn
)
958 WARN_ON_ONCE(memslot
->base_gfn
== mslots
[i
- 1].base_gfn
);
960 /* Shift the next memslot back one and update its index. */
961 mslots
[i
] = mslots
[i
- 1];
962 slots
->id_to_index
[mslots
[i
].id
] = i
;
968 * Re-sort memslots based on their GFN to account for an added, deleted, or
969 * moved memslot. Sorting memslots by GFN allows using a binary search during
972 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
973 * at memslots[0] has the highest GFN.
975 * The sorting algorithm takes advantage of having initially sorted memslots
976 * and knowing the position of the changed memslot. Sorting is also optimized
977 * by not swapping the updated memslot and instead only shifting other memslots
978 * and tracking the new index for the update memslot. Only once its final
979 * index is known is the updated memslot copied into its position in the array.
981 * - When deleting a memslot, the deleted memslot simply needs to be moved to
982 * the end of the array.
984 * - When creating a memslot, the algorithm "inserts" the new memslot at the
985 * end of the array and then it forward to its correct location.
987 * - When moving a memslot, the algorithm first moves the updated memslot
988 * backward to handle the scenario where the memslot's GFN was changed to a
989 * lower value. update_memslots() then falls through and runs the same flow
990 * as creating a memslot to move the memslot forward to handle the scenario
991 * where its GFN was changed to a higher value.
993 * Note, slots are sorted from highest->lowest instead of lowest->highest for
994 * historical reasons. Originally, invalid memslots where denoted by having
995 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
996 * to the end of the array. The current algorithm uses dedicated logic to
997 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
999 * The other historical motiviation for highest->lowest was to improve the
1000 * performance of memslot lookup. KVM originally used a linear search starting
1001 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1002 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1003 * single memslot above the 4gb boundary. As the largest memslot is also the
1004 * most likely to be referenced, sorting it to the front of the array was
1005 * advantageous. The current binary search starts from the middle of the array
1006 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1008 static void update_memslots(struct kvm_memslots
*slots
,
1009 struct kvm_memory_slot
*memslot
,
1010 enum kvm_mr_change change
)
1014 if (change
== KVM_MR_DELETE
) {
1015 kvm_memslot_delete(slots
, memslot
);
1017 if (change
== KVM_MR_CREATE
)
1018 i
= kvm_memslot_insert_back(slots
);
1020 i
= kvm_memslot_move_backward(slots
, memslot
);
1021 i
= kvm_memslot_move_forward(slots
, memslot
, i
);
1024 * Copy the memslot to its new position in memslots and update
1025 * its index accordingly.
1027 slots
->memslots
[i
] = *memslot
;
1028 slots
->id_to_index
[memslot
->id
] = i
;
1032 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
1034 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1036 #ifdef __KVM_HAVE_READONLY_MEM
1037 valid_flags
|= KVM_MEM_READONLY
;
1040 if (mem
->flags
& ~valid_flags
)
1046 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
1047 int as_id
, struct kvm_memslots
*slots
)
1049 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
1050 u64 gen
= old_memslots
->generation
;
1052 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1053 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1055 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1056 synchronize_srcu_expedited(&kvm
->srcu
);
1059 * Increment the new memslot generation a second time, dropping the
1060 * update in-progress flag and incrementing the generation based on
1061 * the number of address spaces. This provides a unique and easily
1062 * identifiable generation number while the memslots are in flux.
1064 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1067 * Generations must be unique even across address spaces. We do not need
1068 * a global counter for that, instead the generation space is evenly split
1069 * across address spaces. For example, with two address spaces, address
1070 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1071 * use generations 1, 3, 5, ...
1073 gen
+= KVM_ADDRESS_SPACE_NUM
;
1075 kvm_arch_memslots_updated(kvm
, gen
);
1077 slots
->generation
= gen
;
1079 return old_memslots
;
1083 * Note, at a minimum, the current number of used slots must be allocated, even
1084 * when deleting a memslot, as we need a complete duplicate of the memslots for
1085 * use when invalidating a memslot prior to deleting/moving the memslot.
1087 static struct kvm_memslots
*kvm_dup_memslots(struct kvm_memslots
*old
,
1088 enum kvm_mr_change change
)
1090 struct kvm_memslots
*slots
;
1091 size_t old_size
, new_size
;
1093 old_size
= sizeof(struct kvm_memslots
) +
1094 (sizeof(struct kvm_memory_slot
) * old
->used_slots
);
1096 if (change
== KVM_MR_CREATE
)
1097 new_size
= old_size
+ sizeof(struct kvm_memory_slot
);
1099 new_size
= old_size
;
1101 slots
= kvzalloc(new_size
, GFP_KERNEL_ACCOUNT
);
1103 memcpy(slots
, old
, old_size
);
1108 static int kvm_set_memslot(struct kvm
*kvm
,
1109 const struct kvm_userspace_memory_region
*mem
,
1110 struct kvm_memory_slot
*old
,
1111 struct kvm_memory_slot
*new, int as_id
,
1112 enum kvm_mr_change change
)
1114 struct kvm_memory_slot
*slot
;
1115 struct kvm_memslots
*slots
;
1118 slots
= kvm_dup_memslots(__kvm_memslots(kvm
, as_id
), change
);
1122 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1124 * Note, the INVALID flag needs to be in the appropriate entry
1125 * in the freshly allocated memslots, not in @old or @new.
1127 slot
= id_to_memslot(slots
, old
->id
);
1128 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1131 * We can re-use the old memslots, the only difference from the
1132 * newly installed memslots is the invalid flag, which will get
1133 * dropped by update_memslots anyway. We'll also revert to the
1134 * old memslots if preparing the new memory region fails.
1136 slots
= install_new_memslots(kvm
, as_id
, slots
);
1138 /* From this point no new shadow pages pointing to a deleted,
1139 * or moved, memslot will be created.
1141 * validation of sp->gfn happens in:
1142 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1143 * - kvm_is_visible_gfn (mmu_check_root)
1145 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1148 r
= kvm_arch_prepare_memory_region(kvm
, new, mem
, change
);
1152 update_memslots(slots
, new, change
);
1153 slots
= install_new_memslots(kvm
, as_id
, slots
);
1155 kvm_arch_commit_memory_region(kvm
, mem
, old
, new, change
);
1161 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1162 slots
= install_new_memslots(kvm
, as_id
, slots
);
1167 static int kvm_delete_memslot(struct kvm
*kvm
,
1168 const struct kvm_userspace_memory_region
*mem
,
1169 struct kvm_memory_slot
*old
, int as_id
)
1171 struct kvm_memory_slot
new;
1177 memset(&new, 0, sizeof(new));
1180 r
= kvm_set_memslot(kvm
, mem
, old
, &new, as_id
, KVM_MR_DELETE
);
1184 kvm_free_memslot(kvm
, old
);
1189 * Allocate some memory and give it an address in the guest physical address
1192 * Discontiguous memory is allowed, mostly for framebuffers.
1194 * Must be called holding kvm->slots_lock for write.
1196 int __kvm_set_memory_region(struct kvm
*kvm
,
1197 const struct kvm_userspace_memory_region
*mem
)
1199 struct kvm_memory_slot old
, new;
1200 struct kvm_memory_slot
*tmp
;
1201 enum kvm_mr_change change
;
1205 r
= check_memory_region_flags(mem
);
1209 as_id
= mem
->slot
>> 16;
1210 id
= (u16
)mem
->slot
;
1212 /* General sanity checks */
1213 if (mem
->memory_size
& (PAGE_SIZE
- 1))
1215 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
1217 /* We can read the guest memory with __xxx_user() later on. */
1218 if ((id
< KVM_USER_MEM_SLOTS
) &&
1219 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
1220 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
1223 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
1225 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
1229 * Make a full copy of the old memslot, the pointer will become stale
1230 * when the memslots are re-sorted by update_memslots(), and the old
1231 * memslot needs to be referenced after calling update_memslots(), e.g.
1232 * to free its resources and for arch specific behavior.
1234 tmp
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
1239 memset(&old
, 0, sizeof(old
));
1243 if (!mem
->memory_size
)
1244 return kvm_delete_memslot(kvm
, mem
, &old
, as_id
);
1247 new.base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
1248 new.npages
= mem
->memory_size
>> PAGE_SHIFT
;
1249 new.flags
= mem
->flags
;
1250 new.userspace_addr
= mem
->userspace_addr
;
1252 if (new.npages
> KVM_MEM_MAX_NR_PAGES
)
1256 change
= KVM_MR_CREATE
;
1257 new.dirty_bitmap
= NULL
;
1258 memset(&new.arch
, 0, sizeof(new.arch
));
1259 } else { /* Modify an existing slot. */
1260 if ((new.userspace_addr
!= old
.userspace_addr
) ||
1261 (new.npages
!= old
.npages
) ||
1262 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
1265 if (new.base_gfn
!= old
.base_gfn
)
1266 change
= KVM_MR_MOVE
;
1267 else if (new.flags
!= old
.flags
)
1268 change
= KVM_MR_FLAGS_ONLY
;
1269 else /* Nothing to change. */
1272 /* Copy dirty_bitmap and arch from the current memslot. */
1273 new.dirty_bitmap
= old
.dirty_bitmap
;
1274 memcpy(&new.arch
, &old
.arch
, sizeof(new.arch
));
1277 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1278 /* Check for overlaps */
1279 kvm_for_each_memslot(tmp
, __kvm_memslots(kvm
, as_id
)) {
1282 if (!((new.base_gfn
+ new.npages
<= tmp
->base_gfn
) ||
1283 (new.base_gfn
>= tmp
->base_gfn
+ tmp
->npages
)))
1288 /* Allocate/free page dirty bitmap as needed */
1289 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1290 new.dirty_bitmap
= NULL
;
1291 else if (!new.dirty_bitmap
) {
1292 r
= kvm_alloc_dirty_bitmap(&new);
1296 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1297 bitmap_set(new.dirty_bitmap
, 0, new.npages
);
1300 r
= kvm_set_memslot(kvm
, mem
, &old
, &new, as_id
, change
);
1304 if (old
.dirty_bitmap
&& !new.dirty_bitmap
)
1305 kvm_destroy_dirty_bitmap(&old
);
1309 if (new.dirty_bitmap
&& !old
.dirty_bitmap
)
1310 kvm_destroy_dirty_bitmap(&new);
1313 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1315 int kvm_set_memory_region(struct kvm
*kvm
,
1316 const struct kvm_userspace_memory_region
*mem
)
1320 mutex_lock(&kvm
->slots_lock
);
1321 r
= __kvm_set_memory_region(kvm
, mem
);
1322 mutex_unlock(&kvm
->slots_lock
);
1325 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1327 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1328 struct kvm_userspace_memory_region
*mem
)
1330 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1333 return kvm_set_memory_region(kvm
, mem
);
1336 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1338 * kvm_get_dirty_log - get a snapshot of dirty pages
1339 * @kvm: pointer to kvm instance
1340 * @log: slot id and address to which we copy the log
1341 * @is_dirty: set to '1' if any dirty pages were found
1342 * @memslot: set to the associated memslot, always valid on success
1344 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
1345 int *is_dirty
, struct kvm_memory_slot
**memslot
)
1347 struct kvm_memslots
*slots
;
1350 unsigned long any
= 0;
1355 as_id
= log
->slot
>> 16;
1356 id
= (u16
)log
->slot
;
1357 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1360 slots
= __kvm_memslots(kvm
, as_id
);
1361 *memslot
= id_to_memslot(slots
, id
);
1362 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
1365 kvm_arch_sync_dirty_log(kvm
, *memslot
);
1367 n
= kvm_dirty_bitmap_bytes(*memslot
);
1369 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1370 any
= (*memslot
)->dirty_bitmap
[i
];
1372 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
1379 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1381 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1383 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1384 * and reenable dirty page tracking for the corresponding pages.
1385 * @kvm: pointer to kvm instance
1386 * @log: slot id and address to which we copy the log
1388 * We need to keep it in mind that VCPU threads can write to the bitmap
1389 * concurrently. So, to avoid losing track of dirty pages we keep the
1392 * 1. Take a snapshot of the bit and clear it if needed.
1393 * 2. Write protect the corresponding page.
1394 * 3. Copy the snapshot to the userspace.
1395 * 4. Upon return caller flushes TLB's if needed.
1397 * Between 2 and 4, the guest may write to the page using the remaining TLB
1398 * entry. This is not a problem because the page is reported dirty using
1399 * the snapshot taken before and step 4 ensures that writes done after
1400 * exiting to userspace will be logged for the next call.
1403 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1405 struct kvm_memslots
*slots
;
1406 struct kvm_memory_slot
*memslot
;
1409 unsigned long *dirty_bitmap
;
1410 unsigned long *dirty_bitmap_buffer
;
1413 as_id
= log
->slot
>> 16;
1414 id
= (u16
)log
->slot
;
1415 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1418 slots
= __kvm_memslots(kvm
, as_id
);
1419 memslot
= id_to_memslot(slots
, id
);
1420 if (!memslot
|| !memslot
->dirty_bitmap
)
1423 dirty_bitmap
= memslot
->dirty_bitmap
;
1425 kvm_arch_sync_dirty_log(kvm
, memslot
);
1427 n
= kvm_dirty_bitmap_bytes(memslot
);
1429 if (kvm
->manual_dirty_log_protect
) {
1431 * Unlike kvm_get_dirty_log, we always return false in *flush,
1432 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1433 * is some code duplication between this function and
1434 * kvm_get_dirty_log, but hopefully all architecture
1435 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1436 * can be eliminated.
1438 dirty_bitmap_buffer
= dirty_bitmap
;
1440 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1441 memset(dirty_bitmap_buffer
, 0, n
);
1443 spin_lock(&kvm
->mmu_lock
);
1444 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1448 if (!dirty_bitmap
[i
])
1452 mask
= xchg(&dirty_bitmap
[i
], 0);
1453 dirty_bitmap_buffer
[i
] = mask
;
1455 offset
= i
* BITS_PER_LONG
;
1456 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1459 spin_unlock(&kvm
->mmu_lock
);
1463 kvm_arch_flush_remote_tlbs_memslot(kvm
, memslot
);
1465 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1472 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1473 * @kvm: kvm instance
1474 * @log: slot id and address to which we copy the log
1476 * Steps 1-4 below provide general overview of dirty page logging. See
1477 * kvm_get_dirty_log_protect() function description for additional details.
1479 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1480 * always flush the TLB (step 4) even if previous step failed and the dirty
1481 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1482 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1483 * writes will be marked dirty for next log read.
1485 * 1. Take a snapshot of the bit and clear it if needed.
1486 * 2. Write protect the corresponding page.
1487 * 3. Copy the snapshot to the userspace.
1488 * 4. Flush TLB's if needed.
1490 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1491 struct kvm_dirty_log
*log
)
1495 mutex_lock(&kvm
->slots_lock
);
1497 r
= kvm_get_dirty_log_protect(kvm
, log
);
1499 mutex_unlock(&kvm
->slots_lock
);
1504 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1505 * and reenable dirty page tracking for the corresponding pages.
1506 * @kvm: pointer to kvm instance
1507 * @log: slot id and address from which to fetch the bitmap of dirty pages
1509 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
1510 struct kvm_clear_dirty_log
*log
)
1512 struct kvm_memslots
*slots
;
1513 struct kvm_memory_slot
*memslot
;
1517 unsigned long *dirty_bitmap
;
1518 unsigned long *dirty_bitmap_buffer
;
1521 as_id
= log
->slot
>> 16;
1522 id
= (u16
)log
->slot
;
1523 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1526 if (log
->first_page
& 63)
1529 slots
= __kvm_memslots(kvm
, as_id
);
1530 memslot
= id_to_memslot(slots
, id
);
1531 if (!memslot
|| !memslot
->dirty_bitmap
)
1534 dirty_bitmap
= memslot
->dirty_bitmap
;
1536 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
1538 if (log
->first_page
> memslot
->npages
||
1539 log
->num_pages
> memslot
->npages
- log
->first_page
||
1540 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
1543 kvm_arch_sync_dirty_log(kvm
, memslot
);
1546 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1547 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
1550 spin_lock(&kvm
->mmu_lock
);
1551 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
1552 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
1553 i
++, offset
+= BITS_PER_LONG
) {
1554 unsigned long mask
= *dirty_bitmap_buffer
++;
1555 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
1559 mask
&= atomic_long_fetch_andnot(mask
, p
);
1562 * mask contains the bits that really have been cleared. This
1563 * never includes any bits beyond the length of the memslot (if
1564 * the length is not aligned to 64 pages), therefore it is not
1565 * a problem if userspace sets them in log->dirty_bitmap.
1569 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1573 spin_unlock(&kvm
->mmu_lock
);
1576 kvm_arch_flush_remote_tlbs_memslot(kvm
, memslot
);
1581 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
1582 struct kvm_clear_dirty_log
*log
)
1586 mutex_lock(&kvm
->slots_lock
);
1588 r
= kvm_clear_dirty_log_protect(kvm
, log
);
1590 mutex_unlock(&kvm
->slots_lock
);
1593 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1595 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1597 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1599 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1601 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1603 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1606 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1608 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1610 return kvm_is_visible_memslot(memslot
);
1612 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1614 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1616 struct vm_area_struct
*vma
;
1617 unsigned long addr
, size
;
1621 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
1622 if (kvm_is_error_hva(addr
))
1625 down_read(¤t
->mm
->mmap_sem
);
1626 vma
= find_vma(current
->mm
, addr
);
1630 size
= vma_kernel_pagesize(vma
);
1633 up_read(¤t
->mm
->mmap_sem
);
1638 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1640 return slot
->flags
& KVM_MEM_READONLY
;
1643 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1644 gfn_t
*nr_pages
, bool write
)
1646 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1647 return KVM_HVA_ERR_BAD
;
1649 if (memslot_is_readonly(slot
) && write
)
1650 return KVM_HVA_ERR_RO_BAD
;
1653 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1655 return __gfn_to_hva_memslot(slot
, gfn
);
1658 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1661 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1664 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1667 return gfn_to_hva_many(slot
, gfn
, NULL
);
1669 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1671 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1673 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1675 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1677 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1679 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1681 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1684 * Return the hva of a @gfn and the R/W attribute if possible.
1686 * @slot: the kvm_memory_slot which contains @gfn
1687 * @gfn: the gfn to be translated
1688 * @writable: used to return the read/write attribute of the @slot if the hva
1689 * is valid and @writable is not NULL
1691 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1692 gfn_t gfn
, bool *writable
)
1694 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1696 if (!kvm_is_error_hva(hva
) && writable
)
1697 *writable
= !memslot_is_readonly(slot
);
1702 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1704 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1706 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1709 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1711 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1713 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1716 static inline int check_user_page_hwpoison(unsigned long addr
)
1718 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1720 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1721 return rc
== -EHWPOISON
;
1725 * The fast path to get the writable pfn which will be stored in @pfn,
1726 * true indicates success, otherwise false is returned. It's also the
1727 * only part that runs if we can in atomic context.
1729 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
1730 bool *writable
, kvm_pfn_t
*pfn
)
1732 struct page
*page
[1];
1736 * Fast pin a writable pfn only if it is a write fault request
1737 * or the caller allows to map a writable pfn for a read fault
1740 if (!(write_fault
|| writable
))
1743 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1745 *pfn
= page_to_pfn(page
[0]);
1756 * The slow path to get the pfn of the specified host virtual address,
1757 * 1 indicates success, -errno is returned if error is detected.
1759 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1760 bool *writable
, kvm_pfn_t
*pfn
)
1762 unsigned int flags
= FOLL_HWPOISON
;
1769 *writable
= write_fault
;
1772 flags
|= FOLL_WRITE
;
1774 flags
|= FOLL_NOWAIT
;
1776 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
1780 /* map read fault as writable if possible */
1781 if (unlikely(!write_fault
) && writable
) {
1784 if (__get_user_pages_fast(addr
, 1, 1, &wpage
) == 1) {
1790 *pfn
= page_to_pfn(page
);
1794 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1796 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1799 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1805 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1806 unsigned long addr
, bool *async
,
1807 bool write_fault
, bool *writable
,
1813 r
= follow_pfn(vma
, addr
, &pfn
);
1816 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1817 * not call the fault handler, so do it here.
1819 bool unlocked
= false;
1820 r
= fixup_user_fault(current
, current
->mm
, addr
,
1821 (write_fault
? FAULT_FLAG_WRITE
: 0),
1828 r
= follow_pfn(vma
, addr
, &pfn
);
1838 * Get a reference here because callers of *hva_to_pfn* and
1839 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1840 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1841 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1842 * simply do nothing for reserved pfns.
1844 * Whoever called remap_pfn_range is also going to call e.g.
1845 * unmap_mapping_range before the underlying pages are freed,
1846 * causing a call to our MMU notifier.
1855 * Pin guest page in memory and return its pfn.
1856 * @addr: host virtual address which maps memory to the guest
1857 * @atomic: whether this function can sleep
1858 * @async: whether this function need to wait IO complete if the
1859 * host page is not in the memory
1860 * @write_fault: whether we should get a writable host page
1861 * @writable: whether it allows to map a writable host page for !@write_fault
1863 * The function will map a writable host page for these two cases:
1864 * 1): @write_fault = true
1865 * 2): @write_fault = false && @writable, @writable will tell the caller
1866 * whether the mapping is writable.
1868 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1869 bool write_fault
, bool *writable
)
1871 struct vm_area_struct
*vma
;
1875 /* we can do it either atomically or asynchronously, not both */
1876 BUG_ON(atomic
&& async
);
1878 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
1882 return KVM_PFN_ERR_FAULT
;
1884 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1888 down_read(¤t
->mm
->mmap_sem
);
1889 if (npages
== -EHWPOISON
||
1890 (!async
&& check_user_page_hwpoison(addr
))) {
1891 pfn
= KVM_PFN_ERR_HWPOISON
;
1896 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1899 pfn
= KVM_PFN_ERR_FAULT
;
1900 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1901 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1905 pfn
= KVM_PFN_ERR_FAULT
;
1907 if (async
&& vma_is_valid(vma
, write_fault
))
1909 pfn
= KVM_PFN_ERR_FAULT
;
1912 up_read(¤t
->mm
->mmap_sem
);
1916 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1917 bool atomic
, bool *async
, bool write_fault
,
1920 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1922 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1925 return KVM_PFN_ERR_RO_FAULT
;
1928 if (kvm_is_error_hva(addr
)) {
1931 return KVM_PFN_NOSLOT
;
1934 /* Do not map writable pfn in the readonly memslot. */
1935 if (writable
&& memslot_is_readonly(slot
)) {
1940 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1943 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1945 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1948 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1949 write_fault
, writable
);
1951 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1953 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1955 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1957 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1959 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1961 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1963 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1965 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1967 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1969 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1971 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1973 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1975 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1977 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1979 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1981 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1983 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1984 struct page
**pages
, int nr_pages
)
1989 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1990 if (kvm_is_error_hva(addr
))
1993 if (entry
< nr_pages
)
1996 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1998 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
2000 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
2002 if (is_error_noslot_pfn(pfn
))
2003 return KVM_ERR_PTR_BAD_PAGE
;
2005 if (kvm_is_reserved_pfn(pfn
)) {
2007 return KVM_ERR_PTR_BAD_PAGE
;
2010 return pfn_to_page(pfn
);
2013 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
2017 pfn
= gfn_to_pfn(kvm
, gfn
);
2019 return kvm_pfn_to_page(pfn
);
2021 EXPORT_SYMBOL_GPL(gfn_to_page
);
2023 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
, struct gfn_to_pfn_cache
*cache
)
2029 cache
->pfn
= cache
->gfn
= 0;
2032 kvm_release_pfn_dirty(pfn
);
2034 kvm_release_pfn_clean(pfn
);
2037 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2038 struct gfn_to_pfn_cache
*cache
, u64 gen
)
2040 kvm_release_pfn(cache
->pfn
, cache
->dirty
, cache
);
2042 cache
->pfn
= gfn_to_pfn_memslot(slot
, gfn
);
2044 cache
->dirty
= false;
2045 cache
->generation
= gen
;
2048 static int __kvm_map_gfn(struct kvm_memslots
*slots
, gfn_t gfn
,
2049 struct kvm_host_map
*map
,
2050 struct gfn_to_pfn_cache
*cache
,
2055 struct page
*page
= KVM_UNMAPPED_PAGE
;
2056 struct kvm_memory_slot
*slot
= __gfn_to_memslot(slots
, gfn
);
2057 u64 gen
= slots
->generation
;
2063 if (!cache
->pfn
|| cache
->gfn
!= gfn
||
2064 cache
->generation
!= gen
) {
2067 kvm_cache_gfn_to_pfn(slot
, gfn
, cache
, gen
);
2073 pfn
= gfn_to_pfn_memslot(slot
, gfn
);
2075 if (is_error_noslot_pfn(pfn
))
2078 if (pfn_valid(pfn
)) {
2079 page
= pfn_to_page(pfn
);
2081 hva
= kmap_atomic(page
);
2084 #ifdef CONFIG_HAS_IOMEM
2085 } else if (!atomic
) {
2086 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
2103 int kvm_map_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
,
2104 struct gfn_to_pfn_cache
*cache
, bool atomic
)
2106 return __kvm_map_gfn(kvm_memslots(vcpu
->kvm
), gfn
, map
,
2109 EXPORT_SYMBOL_GPL(kvm_map_gfn
);
2111 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
2113 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu
), gfn
, map
,
2116 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
2118 static void __kvm_unmap_gfn(struct kvm_memory_slot
*memslot
,
2119 struct kvm_host_map
*map
,
2120 struct gfn_to_pfn_cache
*cache
,
2121 bool dirty
, bool atomic
)
2129 if (map
->page
!= KVM_UNMAPPED_PAGE
) {
2131 kunmap_atomic(map
->hva
);
2135 #ifdef CONFIG_HAS_IOMEM
2139 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2143 mark_page_dirty_in_slot(memslot
, map
->gfn
);
2146 cache
->dirty
|= dirty
;
2148 kvm_release_pfn(map
->pfn
, dirty
, NULL
);
2154 int kvm_unmap_gfn(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
,
2155 struct gfn_to_pfn_cache
*cache
, bool dirty
, bool atomic
)
2157 __kvm_unmap_gfn(gfn_to_memslot(vcpu
->kvm
, map
->gfn
), map
,
2158 cache
, dirty
, atomic
);
2161 EXPORT_SYMBOL_GPL(kvm_unmap_gfn
);
2163 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
2165 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu
, map
->gfn
), map
, NULL
,
2168 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
2170 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2174 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
2176 return kvm_pfn_to_page(pfn
);
2178 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
2180 void kvm_release_page_clean(struct page
*page
)
2182 WARN_ON(is_error_page(page
));
2184 kvm_release_pfn_clean(page_to_pfn(page
));
2186 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
2188 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
2190 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
2191 put_page(pfn_to_page(pfn
));
2193 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
2195 void kvm_release_page_dirty(struct page
*page
)
2197 WARN_ON(is_error_page(page
));
2199 kvm_release_pfn_dirty(page_to_pfn(page
));
2201 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
2203 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
2205 kvm_set_pfn_dirty(pfn
);
2206 kvm_release_pfn_clean(pfn
);
2208 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
2210 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
2212 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
2213 SetPageDirty(pfn_to_page(pfn
));
2215 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
2217 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
2219 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
2220 mark_page_accessed(pfn_to_page(pfn
));
2222 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
2224 void kvm_get_pfn(kvm_pfn_t pfn
)
2226 if (!kvm_is_reserved_pfn(pfn
))
2227 get_page(pfn_to_page(pfn
));
2229 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
2231 static int next_segment(unsigned long len
, int offset
)
2233 if (len
> PAGE_SIZE
- offset
)
2234 return PAGE_SIZE
- offset
;
2239 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2240 void *data
, int offset
, int len
)
2245 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
2246 if (kvm_is_error_hva(addr
))
2248 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
2254 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
2257 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2259 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
2261 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
2263 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
2264 int offset
, int len
)
2266 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2268 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
2270 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
2272 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
2274 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2276 int offset
= offset_in_page(gpa
);
2279 while ((seg
= next_segment(len
, offset
)) != 0) {
2280 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
2290 EXPORT_SYMBOL_GPL(kvm_read_guest
);
2292 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
2294 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2296 int offset
= offset_in_page(gpa
);
2299 while ((seg
= next_segment(len
, offset
)) != 0) {
2300 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2310 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
2312 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2313 void *data
, int offset
, unsigned long len
)
2318 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
2319 if (kvm_is_error_hva(addr
))
2321 pagefault_disable();
2322 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
2329 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2330 void *data
, unsigned long len
)
2332 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2333 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2334 int offset
= offset_in_page(gpa
);
2336 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
2338 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
2340 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
2341 const void *data
, int offset
, int len
)
2346 addr
= gfn_to_hva_memslot(memslot
, gfn
);
2347 if (kvm_is_error_hva(addr
))
2349 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
2352 mark_page_dirty_in_slot(memslot
, gfn
);
2356 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
2357 const void *data
, int offset
, int len
)
2359 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2361 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
2363 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
2365 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
2366 const void *data
, int offset
, int len
)
2368 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2370 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
2372 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
2374 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
2377 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2379 int offset
= offset_in_page(gpa
);
2382 while ((seg
= next_segment(len
, offset
)) != 0) {
2383 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
2393 EXPORT_SYMBOL_GPL(kvm_write_guest
);
2395 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
2398 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2400 int offset
= offset_in_page(gpa
);
2403 while ((seg
= next_segment(len
, offset
)) != 0) {
2404 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2414 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
2416 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
2417 struct gfn_to_hva_cache
*ghc
,
2418 gpa_t gpa
, unsigned long len
)
2420 int offset
= offset_in_page(gpa
);
2421 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
2422 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
2423 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
2424 gfn_t nr_pages_avail
;
2426 /* Update ghc->generation before performing any error checks. */
2427 ghc
->generation
= slots
->generation
;
2429 if (start_gfn
> end_gfn
) {
2430 ghc
->hva
= KVM_HVA_ERR_BAD
;
2435 * If the requested region crosses two memslots, we still
2436 * verify that the entire region is valid here.
2438 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
2439 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
2440 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
2442 if (kvm_is_error_hva(ghc
->hva
))
2446 /* Use the slow path for cross page reads and writes. */
2447 if (nr_pages_needed
== 1)
2450 ghc
->memslot
= NULL
;
2457 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2458 gpa_t gpa
, unsigned long len
)
2460 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2461 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
2463 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
2465 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2466 void *data
, unsigned int offset
,
2469 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2471 gpa_t gpa
= ghc
->gpa
+ offset
;
2473 BUG_ON(len
+ offset
> ghc
->len
);
2475 if (slots
->generation
!= ghc
->generation
) {
2476 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
2480 if (kvm_is_error_hva(ghc
->hva
))
2483 if (unlikely(!ghc
->memslot
))
2484 return kvm_write_guest(kvm
, gpa
, data
, len
);
2486 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2489 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2493 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2495 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2496 void *data
, unsigned long len
)
2498 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2500 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2502 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2503 void *data
, unsigned long len
)
2505 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2508 BUG_ON(len
> ghc
->len
);
2510 if (slots
->generation
!= ghc
->generation
) {
2511 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
2515 if (kvm_is_error_hva(ghc
->hva
))
2518 if (unlikely(!ghc
->memslot
))
2519 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2521 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2527 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2529 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2531 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2533 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2535 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2537 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2539 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2541 int offset
= offset_in_page(gpa
);
2544 while ((seg
= next_segment(len
, offset
)) != 0) {
2545 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2554 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2556 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2559 if (memslot
&& memslot
->dirty_bitmap
) {
2560 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2562 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2566 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2568 struct kvm_memory_slot
*memslot
;
2570 memslot
= gfn_to_memslot(kvm
, gfn
);
2571 mark_page_dirty_in_slot(memslot
, gfn
);
2573 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2575 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2577 struct kvm_memory_slot
*memslot
;
2579 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2580 mark_page_dirty_in_slot(memslot
, gfn
);
2582 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2584 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2586 if (!vcpu
->sigset_active
)
2590 * This does a lockless modification of ->real_blocked, which is fine
2591 * because, only current can change ->real_blocked and all readers of
2592 * ->real_blocked don't care as long ->real_blocked is always a subset
2595 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2598 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2600 if (!vcpu
->sigset_active
)
2603 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2604 sigemptyset(¤t
->real_blocked
);
2607 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2609 unsigned int old
, val
, grow
, grow_start
;
2611 old
= val
= vcpu
->halt_poll_ns
;
2612 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
2613 grow
= READ_ONCE(halt_poll_ns_grow
);
2618 if (val
< grow_start
)
2621 if (val
> halt_poll_ns
)
2624 vcpu
->halt_poll_ns
= val
;
2626 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2629 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2631 unsigned int old
, val
, shrink
;
2633 old
= val
= vcpu
->halt_poll_ns
;
2634 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2640 vcpu
->halt_poll_ns
= val
;
2641 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2644 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2647 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2649 if (kvm_arch_vcpu_runnable(vcpu
)) {
2650 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2653 if (kvm_cpu_has_pending_timer(vcpu
))
2655 if (signal_pending(current
))
2660 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2665 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2667 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2670 DECLARE_SWAITQUEUE(wait
);
2671 bool waited
= false;
2674 kvm_arch_vcpu_blocking(vcpu
);
2676 start
= cur
= ktime_get();
2677 if (vcpu
->halt_poll_ns
&& !kvm_arch_no_poll(vcpu
)) {
2678 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2680 ++vcpu
->stat
.halt_attempted_poll
;
2683 * This sets KVM_REQ_UNHALT if an interrupt
2686 if (kvm_vcpu_check_block(vcpu
) < 0) {
2687 ++vcpu
->stat
.halt_successful_poll
;
2688 if (!vcpu_valid_wakeup(vcpu
))
2689 ++vcpu
->stat
.halt_poll_invalid
;
2693 } while (single_task_running() && ktime_before(cur
, stop
));
2697 prepare_to_swait_exclusive(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2699 if (kvm_vcpu_check_block(vcpu
) < 0)
2706 finish_swait(&vcpu
->wq
, &wait
);
2709 kvm_arch_vcpu_unblocking(vcpu
);
2710 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2712 if (!kvm_arch_no_poll(vcpu
)) {
2713 if (!vcpu_valid_wakeup(vcpu
)) {
2714 shrink_halt_poll_ns(vcpu
);
2715 } else if (halt_poll_ns
) {
2716 if (block_ns
<= vcpu
->halt_poll_ns
)
2718 /* we had a long block, shrink polling */
2719 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2720 shrink_halt_poll_ns(vcpu
);
2721 /* we had a short halt and our poll time is too small */
2722 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2723 block_ns
< halt_poll_ns
)
2724 grow_halt_poll_ns(vcpu
);
2726 vcpu
->halt_poll_ns
= 0;
2730 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2731 kvm_arch_vcpu_block_finish(vcpu
);
2733 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2735 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2737 struct swait_queue_head
*wqp
;
2739 wqp
= kvm_arch_vcpu_wq(vcpu
);
2740 if (swq_has_sleeper(wqp
)) {
2742 WRITE_ONCE(vcpu
->ready
, true);
2743 ++vcpu
->stat
.halt_wakeup
;
2749 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2753 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2755 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2758 int cpu
= vcpu
->cpu
;
2760 if (kvm_vcpu_wake_up(vcpu
))
2764 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2765 if (kvm_arch_vcpu_should_kick(vcpu
))
2766 smp_send_reschedule(cpu
);
2769 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2770 #endif /* !CONFIG_S390 */
2772 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2775 struct task_struct
*task
= NULL
;
2779 pid
= rcu_dereference(target
->pid
);
2781 task
= get_pid_task(pid
, PIDTYPE_PID
);
2785 ret
= yield_to(task
, 1);
2786 put_task_struct(task
);
2790 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2793 * Helper that checks whether a VCPU is eligible for directed yield.
2794 * Most eligible candidate to yield is decided by following heuristics:
2796 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2797 * (preempted lock holder), indicated by @in_spin_loop.
2798 * Set at the beiginning and cleared at the end of interception/PLE handler.
2800 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2801 * chance last time (mostly it has become eligible now since we have probably
2802 * yielded to lockholder in last iteration. This is done by toggling
2803 * @dy_eligible each time a VCPU checked for eligibility.)
2805 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2806 * to preempted lock-holder could result in wrong VCPU selection and CPU
2807 * burning. Giving priority for a potential lock-holder increases lock
2810 * Since algorithm is based on heuristics, accessing another VCPU data without
2811 * locking does not harm. It may result in trying to yield to same VCPU, fail
2812 * and continue with next VCPU and so on.
2814 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2816 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2819 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2820 vcpu
->spin_loop
.dy_eligible
;
2822 if (vcpu
->spin_loop
.in_spin_loop
)
2823 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2832 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2833 * a vcpu_load/vcpu_put pair. However, for most architectures
2834 * kvm_arch_vcpu_runnable does not require vcpu_load.
2836 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
2838 return kvm_arch_vcpu_runnable(vcpu
);
2841 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
2843 if (kvm_arch_dy_runnable(vcpu
))
2846 #ifdef CONFIG_KVM_ASYNC_PF
2847 if (!list_empty_careful(&vcpu
->async_pf
.done
))
2854 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2856 struct kvm
*kvm
= me
->kvm
;
2857 struct kvm_vcpu
*vcpu
;
2858 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2864 kvm_vcpu_set_in_spin_loop(me
, true);
2866 * We boost the priority of a VCPU that is runnable but not
2867 * currently running, because it got preempted by something
2868 * else and called schedule in __vcpu_run. Hopefully that
2869 * VCPU is holding the lock that we need and will release it.
2870 * We approximate round-robin by starting at the last boosted VCPU.
2872 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2873 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2874 if (!pass
&& i
<= last_boosted_vcpu
) {
2875 i
= last_boosted_vcpu
;
2877 } else if (pass
&& i
> last_boosted_vcpu
)
2879 if (!READ_ONCE(vcpu
->ready
))
2883 if (swait_active(&vcpu
->wq
) && !vcpu_dy_runnable(vcpu
))
2885 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
2886 !kvm_arch_vcpu_in_kernel(vcpu
))
2888 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2891 yielded
= kvm_vcpu_yield_to(vcpu
);
2893 kvm
->last_boosted_vcpu
= i
;
2895 } else if (yielded
< 0) {
2902 kvm_vcpu_set_in_spin_loop(me
, false);
2904 /* Ensure vcpu is not eligible during next spinloop */
2905 kvm_vcpu_set_dy_eligible(me
, false);
2907 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2909 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
2911 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2914 if (vmf
->pgoff
== 0)
2915 page
= virt_to_page(vcpu
->run
);
2917 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2918 page
= virt_to_page(vcpu
->arch
.pio_data
);
2920 #ifdef CONFIG_KVM_MMIO
2921 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2922 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2925 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2931 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2932 .fault
= kvm_vcpu_fault
,
2935 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2937 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2941 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2943 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2945 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2946 kvm_put_kvm(vcpu
->kvm
);
2950 static struct file_operations kvm_vcpu_fops
= {
2951 .release
= kvm_vcpu_release
,
2952 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2953 .mmap
= kvm_vcpu_mmap
,
2954 .llseek
= noop_llseek
,
2955 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
2959 * Allocates an inode for the vcpu.
2961 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2963 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
2965 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
2966 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2969 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2971 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2972 char dir_name
[ITOA_MAX_LEN
* 2];
2974 if (!debugfs_initialized())
2977 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2978 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2979 vcpu
->kvm
->debugfs_dentry
);
2981 kvm_arch_create_vcpu_debugfs(vcpu
);
2986 * Creates some virtual cpus. Good luck creating more than one.
2988 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2991 struct kvm_vcpu
*vcpu
;
2994 if (id
>= KVM_MAX_VCPU_ID
)
2997 mutex_lock(&kvm
->lock
);
2998 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2999 mutex_unlock(&kvm
->lock
);
3003 kvm
->created_vcpus
++;
3004 mutex_unlock(&kvm
->lock
);
3006 r
= kvm_arch_vcpu_precreate(kvm
, id
);
3008 goto vcpu_decrement
;
3010 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
3013 goto vcpu_decrement
;
3016 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
3017 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
3022 vcpu
->run
= page_address(page
);
3024 kvm_vcpu_init(vcpu
, kvm
, id
);
3026 r
= kvm_arch_vcpu_create(vcpu
);
3028 goto vcpu_free_run_page
;
3030 mutex_lock(&kvm
->lock
);
3031 if (kvm_get_vcpu_by_id(kvm
, id
)) {
3033 goto unlock_vcpu_destroy
;
3036 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
3037 BUG_ON(kvm
->vcpus
[vcpu
->vcpu_idx
]);
3039 /* Now it's all set up, let userspace reach it */
3041 r
= create_vcpu_fd(vcpu
);
3043 kvm_put_kvm_no_destroy(kvm
);
3044 goto unlock_vcpu_destroy
;
3047 kvm
->vcpus
[vcpu
->vcpu_idx
] = vcpu
;
3050 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3051 * before kvm->online_vcpu's incremented value.
3054 atomic_inc(&kvm
->online_vcpus
);
3056 mutex_unlock(&kvm
->lock
);
3057 kvm_arch_vcpu_postcreate(vcpu
);
3058 kvm_create_vcpu_debugfs(vcpu
);
3061 unlock_vcpu_destroy
:
3062 mutex_unlock(&kvm
->lock
);
3063 kvm_arch_vcpu_destroy(vcpu
);
3065 free_page((unsigned long)vcpu
->run
);
3067 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
3069 mutex_lock(&kvm
->lock
);
3070 kvm
->created_vcpus
--;
3071 mutex_unlock(&kvm
->lock
);
3075 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
3078 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
3079 vcpu
->sigset_active
= 1;
3080 vcpu
->sigset
= *sigset
;
3082 vcpu
->sigset_active
= 0;
3086 static long kvm_vcpu_ioctl(struct file
*filp
,
3087 unsigned int ioctl
, unsigned long arg
)
3089 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3090 void __user
*argp
= (void __user
*)arg
;
3092 struct kvm_fpu
*fpu
= NULL
;
3093 struct kvm_sregs
*kvm_sregs
= NULL
;
3095 if (vcpu
->kvm
->mm
!= current
->mm
)
3098 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
3102 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3103 * execution; mutex_lock() would break them.
3105 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
3106 if (r
!= -ENOIOCTLCMD
)
3109 if (mutex_lock_killable(&vcpu
->mutex
))
3117 oldpid
= rcu_access_pointer(vcpu
->pid
);
3118 if (unlikely(oldpid
!= task_pid(current
))) {
3119 /* The thread running this VCPU changed. */
3122 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
3126 newpid
= get_task_pid(current
, PIDTYPE_PID
);
3127 rcu_assign_pointer(vcpu
->pid
, newpid
);
3132 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
3133 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
3136 case KVM_GET_REGS
: {
3137 struct kvm_regs
*kvm_regs
;
3140 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
3143 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
3147 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
3154 case KVM_SET_REGS
: {
3155 struct kvm_regs
*kvm_regs
;
3157 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
3158 if (IS_ERR(kvm_regs
)) {
3159 r
= PTR_ERR(kvm_regs
);
3162 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
3166 case KVM_GET_SREGS
: {
3167 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
3168 GFP_KERNEL_ACCOUNT
);
3172 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
3176 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
3181 case KVM_SET_SREGS
: {
3182 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
3183 if (IS_ERR(kvm_sregs
)) {
3184 r
= PTR_ERR(kvm_sregs
);
3188 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
3191 case KVM_GET_MP_STATE
: {
3192 struct kvm_mp_state mp_state
;
3194 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
3198 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
3203 case KVM_SET_MP_STATE
: {
3204 struct kvm_mp_state mp_state
;
3207 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
3209 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
3212 case KVM_TRANSLATE
: {
3213 struct kvm_translation tr
;
3216 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
3218 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
3222 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
3227 case KVM_SET_GUEST_DEBUG
: {
3228 struct kvm_guest_debug dbg
;
3231 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
3233 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
3236 case KVM_SET_SIGNAL_MASK
: {
3237 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
3238 struct kvm_signal_mask kvm_sigmask
;
3239 sigset_t sigset
, *p
;
3244 if (copy_from_user(&kvm_sigmask
, argp
,
3245 sizeof(kvm_sigmask
)))
3248 if (kvm_sigmask
.len
!= sizeof(sigset
))
3251 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
3256 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
3260 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
3264 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
3268 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
3274 fpu
= memdup_user(argp
, sizeof(*fpu
));
3280 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
3284 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
3287 mutex_unlock(&vcpu
->mutex
);
3293 #ifdef CONFIG_KVM_COMPAT
3294 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
3295 unsigned int ioctl
, unsigned long arg
)
3297 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3298 void __user
*argp
= compat_ptr(arg
);
3301 if (vcpu
->kvm
->mm
!= current
->mm
)
3305 case KVM_SET_SIGNAL_MASK
: {
3306 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
3307 struct kvm_signal_mask kvm_sigmask
;
3312 if (copy_from_user(&kvm_sigmask
, argp
,
3313 sizeof(kvm_sigmask
)))
3316 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
3319 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
3321 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
3323 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
3327 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
3335 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
3337 struct kvm_device
*dev
= filp
->private_data
;
3340 return dev
->ops
->mmap(dev
, vma
);
3345 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
3346 int (*accessor
)(struct kvm_device
*dev
,
3347 struct kvm_device_attr
*attr
),
3350 struct kvm_device_attr attr
;
3355 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
3358 return accessor(dev
, &attr
);
3361 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
3364 struct kvm_device
*dev
= filp
->private_data
;
3366 if (dev
->kvm
->mm
!= current
->mm
)
3370 case KVM_SET_DEVICE_ATTR
:
3371 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
3372 case KVM_GET_DEVICE_ATTR
:
3373 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
3374 case KVM_HAS_DEVICE_ATTR
:
3375 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
3377 if (dev
->ops
->ioctl
)
3378 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
3384 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
3386 struct kvm_device
*dev
= filp
->private_data
;
3387 struct kvm
*kvm
= dev
->kvm
;
3389 if (dev
->ops
->release
) {
3390 mutex_lock(&kvm
->lock
);
3391 list_del(&dev
->vm_node
);
3392 dev
->ops
->release(dev
);
3393 mutex_unlock(&kvm
->lock
);
3400 static const struct file_operations kvm_device_fops
= {
3401 .unlocked_ioctl
= kvm_device_ioctl
,
3402 .release
= kvm_device_release
,
3403 KVM_COMPAT(kvm_device_ioctl
),
3404 .mmap
= kvm_device_mmap
,
3407 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
3409 if (filp
->f_op
!= &kvm_device_fops
)
3412 return filp
->private_data
;
3415 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
3416 #ifdef CONFIG_KVM_MPIC
3417 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
3418 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
3422 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
3424 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
3427 if (kvm_device_ops_table
[type
] != NULL
)
3430 kvm_device_ops_table
[type
] = ops
;
3434 void kvm_unregister_device_ops(u32 type
)
3436 if (kvm_device_ops_table
[type
] != NULL
)
3437 kvm_device_ops_table
[type
] = NULL
;
3440 static int kvm_ioctl_create_device(struct kvm
*kvm
,
3441 struct kvm_create_device
*cd
)
3443 const struct kvm_device_ops
*ops
= NULL
;
3444 struct kvm_device
*dev
;
3445 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
3449 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
3452 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
3453 ops
= kvm_device_ops_table
[type
];
3460 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
3467 mutex_lock(&kvm
->lock
);
3468 ret
= ops
->create(dev
, type
);
3470 mutex_unlock(&kvm
->lock
);
3474 list_add(&dev
->vm_node
, &kvm
->devices
);
3475 mutex_unlock(&kvm
->lock
);
3481 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
3483 kvm_put_kvm_no_destroy(kvm
);
3484 mutex_lock(&kvm
->lock
);
3485 list_del(&dev
->vm_node
);
3486 mutex_unlock(&kvm
->lock
);
3495 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
3498 case KVM_CAP_USER_MEMORY
:
3499 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
3500 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
3501 case KVM_CAP_INTERNAL_ERROR_DATA
:
3502 #ifdef CONFIG_HAVE_KVM_MSI
3503 case KVM_CAP_SIGNAL_MSI
:
3505 #ifdef CONFIG_HAVE_KVM_IRQFD
3507 case KVM_CAP_IRQFD_RESAMPLE
:
3509 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
3510 case KVM_CAP_CHECK_EXTENSION_VM
:
3511 case KVM_CAP_ENABLE_CAP_VM
:
3513 #ifdef CONFIG_KVM_MMIO
3514 case KVM_CAP_COALESCED_MMIO
:
3515 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
3516 case KVM_CAP_COALESCED_PIO
:
3519 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3520 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
3521 return KVM_DIRTY_LOG_MANUAL_CAPS
;
3523 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3524 case KVM_CAP_IRQ_ROUTING
:
3525 return KVM_MAX_IRQ_ROUTES
;
3527 #if KVM_ADDRESS_SPACE_NUM > 1
3528 case KVM_CAP_MULTI_ADDRESS_SPACE
:
3529 return KVM_ADDRESS_SPACE_NUM
;
3531 case KVM_CAP_NR_MEMSLOTS
:
3532 return KVM_USER_MEM_SLOTS
;
3536 return kvm_vm_ioctl_check_extension(kvm
, arg
);
3539 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
3540 struct kvm_enable_cap
*cap
)
3545 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
3546 struct kvm_enable_cap
*cap
)
3549 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3550 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
3551 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
3553 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
3554 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
3556 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
3558 kvm
->manual_dirty_log_protect
= cap
->args
[0];
3563 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
3567 static long kvm_vm_ioctl(struct file
*filp
,
3568 unsigned int ioctl
, unsigned long arg
)
3570 struct kvm
*kvm
= filp
->private_data
;
3571 void __user
*argp
= (void __user
*)arg
;
3574 if (kvm
->mm
!= current
->mm
)
3577 case KVM_CREATE_VCPU
:
3578 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3580 case KVM_ENABLE_CAP
: {
3581 struct kvm_enable_cap cap
;
3584 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3586 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
3589 case KVM_SET_USER_MEMORY_REGION
: {
3590 struct kvm_userspace_memory_region kvm_userspace_mem
;
3593 if (copy_from_user(&kvm_userspace_mem
, argp
,
3594 sizeof(kvm_userspace_mem
)))
3597 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3600 case KVM_GET_DIRTY_LOG
: {
3601 struct kvm_dirty_log log
;
3604 if (copy_from_user(&log
, argp
, sizeof(log
)))
3606 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3609 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3610 case KVM_CLEAR_DIRTY_LOG
: {
3611 struct kvm_clear_dirty_log log
;
3614 if (copy_from_user(&log
, argp
, sizeof(log
)))
3616 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
3620 #ifdef CONFIG_KVM_MMIO
3621 case KVM_REGISTER_COALESCED_MMIO
: {
3622 struct kvm_coalesced_mmio_zone zone
;
3625 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3627 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3630 case KVM_UNREGISTER_COALESCED_MMIO
: {
3631 struct kvm_coalesced_mmio_zone zone
;
3634 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3636 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3641 struct kvm_irqfd data
;
3644 if (copy_from_user(&data
, argp
, sizeof(data
)))
3646 r
= kvm_irqfd(kvm
, &data
);
3649 case KVM_IOEVENTFD
: {
3650 struct kvm_ioeventfd data
;
3653 if (copy_from_user(&data
, argp
, sizeof(data
)))
3655 r
= kvm_ioeventfd(kvm
, &data
);
3658 #ifdef CONFIG_HAVE_KVM_MSI
3659 case KVM_SIGNAL_MSI
: {
3663 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3665 r
= kvm_send_userspace_msi(kvm
, &msi
);
3669 #ifdef __KVM_HAVE_IRQ_LINE
3670 case KVM_IRQ_LINE_STATUS
:
3671 case KVM_IRQ_LINE
: {
3672 struct kvm_irq_level irq_event
;
3675 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3678 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3679 ioctl
== KVM_IRQ_LINE_STATUS
);
3684 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3685 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3693 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3694 case KVM_SET_GSI_ROUTING
: {
3695 struct kvm_irq_routing routing
;
3696 struct kvm_irq_routing __user
*urouting
;
3697 struct kvm_irq_routing_entry
*entries
= NULL
;
3700 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3703 if (!kvm_arch_can_set_irq_routing(kvm
))
3705 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3711 entries
= vmalloc(array_size(sizeof(*entries
),
3717 if (copy_from_user(entries
, urouting
->entries
,
3718 routing
.nr
* sizeof(*entries
)))
3719 goto out_free_irq_routing
;
3721 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3723 out_free_irq_routing
:
3727 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3728 case KVM_CREATE_DEVICE
: {
3729 struct kvm_create_device cd
;
3732 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3735 r
= kvm_ioctl_create_device(kvm
, &cd
);
3740 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3746 case KVM_CHECK_EXTENSION
:
3747 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3750 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3756 #ifdef CONFIG_KVM_COMPAT
3757 struct compat_kvm_dirty_log
{
3761 compat_uptr_t dirty_bitmap
; /* one bit per page */
3766 static long kvm_vm_compat_ioctl(struct file
*filp
,
3767 unsigned int ioctl
, unsigned long arg
)
3769 struct kvm
*kvm
= filp
->private_data
;
3772 if (kvm
->mm
!= current
->mm
)
3775 case KVM_GET_DIRTY_LOG
: {
3776 struct compat_kvm_dirty_log compat_log
;
3777 struct kvm_dirty_log log
;
3779 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3780 sizeof(compat_log
)))
3782 log
.slot
= compat_log
.slot
;
3783 log
.padding1
= compat_log
.padding1
;
3784 log
.padding2
= compat_log
.padding2
;
3785 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3787 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3791 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3797 static struct file_operations kvm_vm_fops
= {
3798 .release
= kvm_vm_release
,
3799 .unlocked_ioctl
= kvm_vm_ioctl
,
3800 .llseek
= noop_llseek
,
3801 KVM_COMPAT(kvm_vm_compat_ioctl
),
3804 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3810 kvm
= kvm_create_vm(type
);
3812 return PTR_ERR(kvm
);
3813 #ifdef CONFIG_KVM_MMIO
3814 r
= kvm_coalesced_mmio_init(kvm
);
3818 r
= get_unused_fd_flags(O_CLOEXEC
);
3822 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3830 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3831 * already set, with ->release() being kvm_vm_release(). In error
3832 * cases it will be called by the final fput(file) and will take
3833 * care of doing kvm_put_kvm(kvm).
3835 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3840 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3842 fd_install(r
, file
);
3850 static long kvm_dev_ioctl(struct file
*filp
,
3851 unsigned int ioctl
, unsigned long arg
)
3856 case KVM_GET_API_VERSION
:
3859 r
= KVM_API_VERSION
;
3862 r
= kvm_dev_ioctl_create_vm(arg
);
3864 case KVM_CHECK_EXTENSION
:
3865 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3867 case KVM_GET_VCPU_MMAP_SIZE
:
3870 r
= PAGE_SIZE
; /* struct kvm_run */
3872 r
+= PAGE_SIZE
; /* pio data page */
3874 #ifdef CONFIG_KVM_MMIO
3875 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3878 case KVM_TRACE_ENABLE
:
3879 case KVM_TRACE_PAUSE
:
3880 case KVM_TRACE_DISABLE
:
3884 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3890 static struct file_operations kvm_chardev_ops
= {
3891 .unlocked_ioctl
= kvm_dev_ioctl
,
3892 .llseek
= noop_llseek
,
3893 KVM_COMPAT(kvm_dev_ioctl
),
3896 static struct miscdevice kvm_dev
= {
3902 static void hardware_enable_nolock(void *junk
)
3904 int cpu
= raw_smp_processor_id();
3907 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3910 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3912 r
= kvm_arch_hardware_enable();
3915 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3916 atomic_inc(&hardware_enable_failed
);
3917 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3921 static int kvm_starting_cpu(unsigned int cpu
)
3923 raw_spin_lock(&kvm_count_lock
);
3924 if (kvm_usage_count
)
3925 hardware_enable_nolock(NULL
);
3926 raw_spin_unlock(&kvm_count_lock
);
3930 static void hardware_disable_nolock(void *junk
)
3932 int cpu
= raw_smp_processor_id();
3934 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3936 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3937 kvm_arch_hardware_disable();
3940 static int kvm_dying_cpu(unsigned int cpu
)
3942 raw_spin_lock(&kvm_count_lock
);
3943 if (kvm_usage_count
)
3944 hardware_disable_nolock(NULL
);
3945 raw_spin_unlock(&kvm_count_lock
);
3949 static void hardware_disable_all_nolock(void)
3951 BUG_ON(!kvm_usage_count
);
3954 if (!kvm_usage_count
)
3955 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3958 static void hardware_disable_all(void)
3960 raw_spin_lock(&kvm_count_lock
);
3961 hardware_disable_all_nolock();
3962 raw_spin_unlock(&kvm_count_lock
);
3965 static int hardware_enable_all(void)
3969 raw_spin_lock(&kvm_count_lock
);
3972 if (kvm_usage_count
== 1) {
3973 atomic_set(&hardware_enable_failed
, 0);
3974 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3976 if (atomic_read(&hardware_enable_failed
)) {
3977 hardware_disable_all_nolock();
3982 raw_spin_unlock(&kvm_count_lock
);
3987 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3991 * Some (well, at least mine) BIOSes hang on reboot if
3994 * And Intel TXT required VMX off for all cpu when system shutdown.
3996 pr_info("kvm: exiting hardware virtualization\n");
3997 kvm_rebooting
= true;
3998 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4002 static struct notifier_block kvm_reboot_notifier
= {
4003 .notifier_call
= kvm_reboot
,
4007 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
4011 for (i
= 0; i
< bus
->dev_count
; i
++) {
4012 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
4014 kvm_iodevice_destructor(pos
);
4019 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
4020 const struct kvm_io_range
*r2
)
4022 gpa_t addr1
= r1
->addr
;
4023 gpa_t addr2
= r2
->addr
;
4028 /* If r2->len == 0, match the exact address. If r2->len != 0,
4029 * accept any overlapping write. Any order is acceptable for
4030 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4031 * we process all of them.
4044 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
4046 return kvm_io_bus_cmp(p1
, p2
);
4049 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
4050 gpa_t addr
, int len
)
4052 struct kvm_io_range
*range
, key
;
4055 key
= (struct kvm_io_range
) {
4060 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
4061 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
4065 off
= range
- bus
->range
;
4067 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
4073 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
4074 struct kvm_io_range
*range
, const void *val
)
4078 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
4082 while (idx
< bus
->dev_count
&&
4083 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
4084 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
4093 /* kvm_io_bus_write - called under kvm->slots_lock */
4094 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
4095 int len
, const void *val
)
4097 struct kvm_io_bus
*bus
;
4098 struct kvm_io_range range
;
4101 range
= (struct kvm_io_range
) {
4106 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4109 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
4110 return r
< 0 ? r
: 0;
4112 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
4114 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4115 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
4116 gpa_t addr
, int len
, const void *val
, long cookie
)
4118 struct kvm_io_bus
*bus
;
4119 struct kvm_io_range range
;
4121 range
= (struct kvm_io_range
) {
4126 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4130 /* First try the device referenced by cookie. */
4131 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
4132 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
4133 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
4138 * cookie contained garbage; fall back to search and return the
4139 * correct cookie value.
4141 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
4144 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
4145 struct kvm_io_range
*range
, void *val
)
4149 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
4153 while (idx
< bus
->dev_count
&&
4154 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
4155 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
4164 /* kvm_io_bus_read - called under kvm->slots_lock */
4165 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
4168 struct kvm_io_bus
*bus
;
4169 struct kvm_io_range range
;
4172 range
= (struct kvm_io_range
) {
4177 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4180 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
4181 return r
< 0 ? r
: 0;
4184 /* Caller must hold slots_lock. */
4185 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
4186 int len
, struct kvm_io_device
*dev
)
4189 struct kvm_io_bus
*new_bus
, *bus
;
4190 struct kvm_io_range range
;
4192 bus
= kvm_get_bus(kvm
, bus_idx
);
4196 /* exclude ioeventfd which is limited by maximum fd */
4197 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
4200 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
4201 GFP_KERNEL_ACCOUNT
);
4205 range
= (struct kvm_io_range
) {
4211 for (i
= 0; i
< bus
->dev_count
; i
++)
4212 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
4215 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
4216 new_bus
->dev_count
++;
4217 new_bus
->range
[i
] = range
;
4218 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
4219 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
4220 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
4221 synchronize_srcu_expedited(&kvm
->srcu
);
4227 /* Caller must hold slots_lock. */
4228 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
4229 struct kvm_io_device
*dev
)
4232 struct kvm_io_bus
*new_bus
, *bus
;
4234 bus
= kvm_get_bus(kvm
, bus_idx
);
4238 for (i
= 0; i
< bus
->dev_count
; i
++)
4239 if (bus
->range
[i
].dev
== dev
) {
4243 if (i
== bus
->dev_count
)
4246 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
4247 GFP_KERNEL_ACCOUNT
);
4249 pr_err("kvm: failed to shrink bus, removing it completely\n");
4253 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
4254 new_bus
->dev_count
--;
4255 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
4256 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
4259 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
4260 synchronize_srcu_expedited(&kvm
->srcu
);
4265 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
4268 struct kvm_io_bus
*bus
;
4269 int dev_idx
, srcu_idx
;
4270 struct kvm_io_device
*iodev
= NULL
;
4272 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
4274 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
4278 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
4282 iodev
= bus
->range
[dev_idx
].dev
;
4285 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
4289 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
4291 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
4292 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
4295 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
4298 /* The debugfs files are a reference to the kvm struct which
4299 * is still valid when kvm_destroy_vm is called.
4300 * To avoid the race between open and the removal of the debugfs
4301 * directory we test against the users count.
4303 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
4306 if (simple_attr_open(inode
, file
, get
,
4307 KVM_DBGFS_GET_MODE(stat_data
->dbgfs_item
) & 0222
4310 kvm_put_kvm(stat_data
->kvm
);
4317 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
4319 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
4322 simple_attr_release(inode
, file
);
4323 kvm_put_kvm(stat_data
->kvm
);
4328 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
4330 *val
= *(ulong
*)((void *)kvm
+ offset
);
4335 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
4337 *(ulong
*)((void *)kvm
+ offset
) = 0;
4342 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
4345 struct kvm_vcpu
*vcpu
;
4349 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4350 *val
+= *(u64
*)((void *)vcpu
+ offset
);
4355 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
4358 struct kvm_vcpu
*vcpu
;
4360 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4361 *(u64
*)((void *)vcpu
+ offset
) = 0;
4366 static int kvm_stat_data_get(void *data
, u64
*val
)
4369 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
4371 switch (stat_data
->dbgfs_item
->kind
) {
4373 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
4374 stat_data
->dbgfs_item
->offset
, val
);
4377 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
4378 stat_data
->dbgfs_item
->offset
, val
);
4385 static int kvm_stat_data_clear(void *data
, u64 val
)
4388 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
4393 switch (stat_data
->dbgfs_item
->kind
) {
4395 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
4396 stat_data
->dbgfs_item
->offset
);
4399 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
4400 stat_data
->dbgfs_item
->offset
);
4407 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
4409 __simple_attr_check_format("%llu\n", 0ull);
4410 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
4411 kvm_stat_data_clear
, "%llu\n");
4414 static const struct file_operations stat_fops_per_vm
= {
4415 .owner
= THIS_MODULE
,
4416 .open
= kvm_stat_data_open
,
4417 .release
= kvm_debugfs_release
,
4418 .read
= simple_attr_read
,
4419 .write
= simple_attr_write
,
4420 .llseek
= no_llseek
,
4423 static int vm_stat_get(void *_offset
, u64
*val
)
4425 unsigned offset
= (long)_offset
;
4430 mutex_lock(&kvm_lock
);
4431 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4432 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
4435 mutex_unlock(&kvm_lock
);
4439 static int vm_stat_clear(void *_offset
, u64 val
)
4441 unsigned offset
= (long)_offset
;
4447 mutex_lock(&kvm_lock
);
4448 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4449 kvm_clear_stat_per_vm(kvm
, offset
);
4451 mutex_unlock(&kvm_lock
);
4456 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
4458 static int vcpu_stat_get(void *_offset
, u64
*val
)
4460 unsigned offset
= (long)_offset
;
4465 mutex_lock(&kvm_lock
);
4466 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4467 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
4470 mutex_unlock(&kvm_lock
);
4474 static int vcpu_stat_clear(void *_offset
, u64 val
)
4476 unsigned offset
= (long)_offset
;
4482 mutex_lock(&kvm_lock
);
4483 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4484 kvm_clear_stat_per_vcpu(kvm
, offset
);
4486 mutex_unlock(&kvm_lock
);
4491 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
4494 static const struct file_operations
*stat_fops
[] = {
4495 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
4496 [KVM_STAT_VM
] = &vm_stat_fops
,
4499 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
4501 struct kobj_uevent_env
*env
;
4502 unsigned long long created
, active
;
4504 if (!kvm_dev
.this_device
|| !kvm
)
4507 mutex_lock(&kvm_lock
);
4508 if (type
== KVM_EVENT_CREATE_VM
) {
4509 kvm_createvm_count
++;
4511 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4514 created
= kvm_createvm_count
;
4515 active
= kvm_active_vms
;
4516 mutex_unlock(&kvm_lock
);
4518 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
4522 add_uevent_var(env
, "CREATED=%llu", created
);
4523 add_uevent_var(env
, "COUNT=%llu", active
);
4525 if (type
== KVM_EVENT_CREATE_VM
) {
4526 add_uevent_var(env
, "EVENT=create");
4527 kvm
->userspace_pid
= task_pid_nr(current
);
4528 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4529 add_uevent_var(env
, "EVENT=destroy");
4531 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
4533 if (!IS_ERR_OR_NULL(kvm
->debugfs_dentry
)) {
4534 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
4537 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
4539 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
4543 /* no need for checks, since we are adding at most only 5 keys */
4544 env
->envp
[env
->envp_idx
++] = NULL
;
4545 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
4549 static void kvm_init_debug(void)
4551 struct kvm_stats_debugfs_item
*p
;
4553 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
4555 kvm_debugfs_num_entries
= 0;
4556 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
4557 debugfs_create_file(p
->name
, KVM_DBGFS_GET_MODE(p
),
4558 kvm_debugfs_dir
, (void *)(long)p
->offset
,
4559 stat_fops
[p
->kind
]);
4563 static int kvm_suspend(void)
4565 if (kvm_usage_count
)
4566 hardware_disable_nolock(NULL
);
4570 static void kvm_resume(void)
4572 if (kvm_usage_count
) {
4573 #ifdef CONFIG_LOCKDEP
4574 WARN_ON(lockdep_is_held(&kvm_count_lock
));
4576 hardware_enable_nolock(NULL
);
4580 static struct syscore_ops kvm_syscore_ops
= {
4581 .suspend
= kvm_suspend
,
4582 .resume
= kvm_resume
,
4586 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4588 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4591 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4593 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4595 WRITE_ONCE(vcpu
->preempted
, false);
4596 WRITE_ONCE(vcpu
->ready
, false);
4598 __this_cpu_write(kvm_running_vcpu
, vcpu
);
4599 kvm_arch_sched_in(vcpu
, cpu
);
4600 kvm_arch_vcpu_load(vcpu
, cpu
);
4603 static void kvm_sched_out(struct preempt_notifier
*pn
,
4604 struct task_struct
*next
)
4606 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4608 if (current
->state
== TASK_RUNNING
) {
4609 WRITE_ONCE(vcpu
->preempted
, true);
4610 WRITE_ONCE(vcpu
->ready
, true);
4612 kvm_arch_vcpu_put(vcpu
);
4613 __this_cpu_write(kvm_running_vcpu
, NULL
);
4617 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4619 * We can disable preemption locally around accessing the per-CPU variable,
4620 * and use the resolved vcpu pointer after enabling preemption again,
4621 * because even if the current thread is migrated to another CPU, reading
4622 * the per-CPU value later will give us the same value as we update the
4623 * per-CPU variable in the preempt notifier handlers.
4625 struct kvm_vcpu
*kvm_get_running_vcpu(void)
4627 struct kvm_vcpu
*vcpu
;
4630 vcpu
= __this_cpu_read(kvm_running_vcpu
);
4637 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4639 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
4641 return &kvm_running_vcpu
;
4644 struct kvm_cpu_compat_check
{
4649 static void check_processor_compat(void *data
)
4651 struct kvm_cpu_compat_check
*c
= data
;
4653 *c
->ret
= kvm_arch_check_processor_compat(c
->opaque
);
4656 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4657 struct module
*module
)
4659 struct kvm_cpu_compat_check c
;
4663 r
= kvm_arch_init(opaque
);
4668 * kvm_arch_init makes sure there's at most one caller
4669 * for architectures that support multiple implementations,
4670 * like intel and amd on x86.
4671 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4672 * conflicts in case kvm is already setup for another implementation.
4674 r
= kvm_irqfd_init();
4678 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4683 r
= kvm_arch_hardware_setup(opaque
);
4689 for_each_online_cpu(cpu
) {
4690 smp_call_function_single(cpu
, check_processor_compat
, &c
, 1);
4695 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4696 kvm_starting_cpu
, kvm_dying_cpu
);
4699 register_reboot_notifier(&kvm_reboot_notifier
);
4701 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4703 vcpu_align
= __alignof__(struct kvm_vcpu
);
4705 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
4707 offsetof(struct kvm_vcpu
, arch
),
4708 sizeof_field(struct kvm_vcpu
, arch
),
4710 if (!kvm_vcpu_cache
) {
4715 r
= kvm_async_pf_init();
4719 kvm_chardev_ops
.owner
= module
;
4720 kvm_vm_fops
.owner
= module
;
4721 kvm_vcpu_fops
.owner
= module
;
4723 r
= misc_register(&kvm_dev
);
4725 pr_err("kvm: misc device register failed\n");
4729 register_syscore_ops(&kvm_syscore_ops
);
4731 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4732 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4736 r
= kvm_vfio_ops_init();
4742 kvm_async_pf_deinit();
4744 kmem_cache_destroy(kvm_vcpu_cache
);
4746 unregister_reboot_notifier(&kvm_reboot_notifier
);
4747 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4749 kvm_arch_hardware_unsetup();
4751 free_cpumask_var(cpus_hardware_enabled
);
4759 EXPORT_SYMBOL_GPL(kvm_init
);
4763 debugfs_remove_recursive(kvm_debugfs_dir
);
4764 misc_deregister(&kvm_dev
);
4765 kmem_cache_destroy(kvm_vcpu_cache
);
4766 kvm_async_pf_deinit();
4767 unregister_syscore_ops(&kvm_syscore_ops
);
4768 unregister_reboot_notifier(&kvm_reboot_notifier
);
4769 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4770 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4771 kvm_arch_hardware_unsetup();
4774 free_cpumask_var(cpus_hardware_enabled
);
4775 kvm_vfio_ops_exit();
4777 EXPORT_SYMBOL_GPL(kvm_exit
);
4779 struct kvm_vm_worker_thread_context
{
4781 struct task_struct
*parent
;
4782 struct completion init_done
;
4783 kvm_vm_thread_fn_t thread_fn
;
4788 static int kvm_vm_worker_thread(void *context
)
4791 * The init_context is allocated on the stack of the parent thread, so
4792 * we have to locally copy anything that is needed beyond initialization
4794 struct kvm_vm_worker_thread_context
*init_context
= context
;
4795 struct kvm
*kvm
= init_context
->kvm
;
4796 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
4797 uintptr_t data
= init_context
->data
;
4800 err
= kthread_park(current
);
4801 /* kthread_park(current) is never supposed to return an error */
4806 err
= cgroup_attach_task_all(init_context
->parent
, current
);
4808 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4813 set_user_nice(current
, task_nice(init_context
->parent
));
4816 init_context
->err
= err
;
4817 complete(&init_context
->init_done
);
4818 init_context
= NULL
;
4823 /* Wait to be woken up by the spawner before proceeding. */
4826 if (!kthread_should_stop())
4827 err
= thread_fn(kvm
, data
);
4832 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
4833 uintptr_t data
, const char *name
,
4834 struct task_struct
**thread_ptr
)
4836 struct kvm_vm_worker_thread_context init_context
= {};
4837 struct task_struct
*thread
;
4840 init_context
.kvm
= kvm
;
4841 init_context
.parent
= current
;
4842 init_context
.thread_fn
= thread_fn
;
4843 init_context
.data
= data
;
4844 init_completion(&init_context
.init_done
);
4846 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
4847 "%s-%d", name
, task_pid_nr(current
));
4849 return PTR_ERR(thread
);
4851 /* kthread_run is never supposed to return NULL */
4852 WARN_ON(thread
== NULL
);
4854 wait_for_completion(&init_context
.init_done
);
4856 if (!init_context
.err
)
4857 *thread_ptr
= thread
;
4859 return init_context
.err
;