2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
54 #include <linux/kthread.h>
56 #include <asm/processor.h>
58 #include <asm/ioctl.h>
59 #include <linux/uaccess.h>
60 #include <asm/pgtable.h>
62 #include "coalesced_mmio.h"
66 #define CREATE_TRACE_POINTS
67 #include <trace/events/kvm.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
77 module_param(halt_poll_ns
, uint
, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns
);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow
= 2;
82 module_param(halt_poll_ns_grow
, uint
, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
85 /* Default resets per-vcpu halt_poll_ns . */
86 unsigned int halt_poll_ns_shrink
;
87 module_param(halt_poll_ns_shrink
, uint
, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
93 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
96 DEFINE_MUTEX(kvm_lock
);
97 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
100 static cpumask_var_t cpus_hardware_enabled
;
101 static int kvm_usage_count
;
102 static atomic_t hardware_enable_failed
;
104 struct kmem_cache
*kvm_vcpu_cache
;
105 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
107 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
109 struct dentry
*kvm_debugfs_dir
;
110 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
112 static int kvm_debugfs_num_entries
;
113 static const struct file_operations
*stat_fops_per_vm
[];
115 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
117 #ifdef CONFIG_KVM_COMPAT
118 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
121 static int hardware_enable_all(void);
122 static void hardware_disable_all(void);
124 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
126 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
128 __visible
bool kvm_rebooting
;
129 EXPORT_SYMBOL_GPL(kvm_rebooting
);
131 static bool largepages_enabled
= true;
133 #define KVM_EVENT_CREATE_VM 0
134 #define KVM_EVENT_DESTROY_VM 1
135 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
136 static unsigned long long kvm_createvm_count
;
137 static unsigned long long kvm_active_vms
;
139 __weak
void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
140 unsigned long start
, unsigned long end
)
144 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
147 return PageReserved(pfn_to_page(pfn
));
153 * Switches to specified vcpu, until a matching vcpu_put()
155 int vcpu_load(struct kvm_vcpu
*vcpu
)
159 if (mutex_lock_killable(&vcpu
->mutex
))
162 preempt_notifier_register(&vcpu
->preempt_notifier
);
163 kvm_arch_vcpu_load(vcpu
, cpu
);
167 EXPORT_SYMBOL_GPL(vcpu_load
);
169 void vcpu_put(struct kvm_vcpu
*vcpu
)
172 kvm_arch_vcpu_put(vcpu
);
173 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
175 mutex_unlock(&vcpu
->mutex
);
177 EXPORT_SYMBOL_GPL(vcpu_put
);
179 /* TODO: merge with kvm_arch_vcpu_should_kick */
180 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
182 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
185 * We need to wait for the VCPU to reenable interrupts and get out of
186 * READING_SHADOW_PAGE_TABLES mode.
188 if (req
& KVM_REQUEST_WAIT
)
189 return mode
!= OUTSIDE_GUEST_MODE
;
192 * Need to kick a running VCPU, but otherwise there is nothing to do.
194 return mode
== IN_GUEST_MODE
;
197 static void ack_flush(void *_completed
)
201 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
204 cpus
= cpu_online_mask
;
206 if (cpumask_empty(cpus
))
209 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
213 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
218 struct kvm_vcpu
*vcpu
;
220 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
223 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
224 kvm_make_request(req
, vcpu
);
227 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
230 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
231 kvm_request_needs_ipi(vcpu
, req
))
232 __cpumask_set_cpu(cpu
, cpus
);
234 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
236 free_cpumask_var(cpus
);
240 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
241 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
244 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
245 * kvm_make_all_cpus_request.
247 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
250 * We want to publish modifications to the page tables before reading
251 * mode. Pairs with a memory barrier in arch-specific code.
252 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
253 * and smp_mb in walk_shadow_page_lockless_begin/end.
254 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
256 * There is already an smp_mb__after_atomic() before
257 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
260 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
261 ++kvm
->stat
.remote_tlb_flush
;
262 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
264 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
267 void kvm_reload_remote_mmus(struct kvm
*kvm
)
269 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
272 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
277 mutex_init(&vcpu
->mutex
);
282 init_swait_queue_head(&vcpu
->wq
);
283 kvm_async_pf_vcpu_init(vcpu
);
286 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
288 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
293 vcpu
->run
= page_address(page
);
295 kvm_vcpu_set_in_spin_loop(vcpu
, false);
296 kvm_vcpu_set_dy_eligible(vcpu
, false);
297 vcpu
->preempted
= false;
299 r
= kvm_arch_vcpu_init(vcpu
);
305 free_page((unsigned long)vcpu
->run
);
309 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
311 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
314 * no need for rcu_read_lock as VCPU_RUN is the only place that
315 * will change the vcpu->pid pointer and on uninit all file
316 * descriptors are already gone.
318 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
319 kvm_arch_vcpu_uninit(vcpu
);
320 free_page((unsigned long)vcpu
->run
);
322 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
324 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
325 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
327 return container_of(mn
, struct kvm
, mmu_notifier
);
330 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
331 struct mm_struct
*mm
,
332 unsigned long address
,
335 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
338 idx
= srcu_read_lock(&kvm
->srcu
);
339 spin_lock(&kvm
->mmu_lock
);
340 kvm
->mmu_notifier_seq
++;
341 kvm_set_spte_hva(kvm
, address
, pte
);
342 spin_unlock(&kvm
->mmu_lock
);
343 srcu_read_unlock(&kvm
->srcu
, idx
);
346 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
347 struct mm_struct
*mm
,
351 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
352 int need_tlb_flush
= 0, idx
;
354 idx
= srcu_read_lock(&kvm
->srcu
);
355 spin_lock(&kvm
->mmu_lock
);
357 * The count increase must become visible at unlock time as no
358 * spte can be established without taking the mmu_lock and
359 * count is also read inside the mmu_lock critical section.
361 kvm
->mmu_notifier_count
++;
362 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
363 need_tlb_flush
|= kvm
->tlbs_dirty
;
364 /* we've to flush the tlb before the pages can be freed */
366 kvm_flush_remote_tlbs(kvm
);
368 spin_unlock(&kvm
->mmu_lock
);
370 kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
);
372 srcu_read_unlock(&kvm
->srcu
, idx
);
375 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
376 struct mm_struct
*mm
,
380 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
382 spin_lock(&kvm
->mmu_lock
);
384 * This sequence increase will notify the kvm page fault that
385 * the page that is going to be mapped in the spte could have
388 kvm
->mmu_notifier_seq
++;
391 * The above sequence increase must be visible before the
392 * below count decrease, which is ensured by the smp_wmb above
393 * in conjunction with the smp_rmb in mmu_notifier_retry().
395 kvm
->mmu_notifier_count
--;
396 spin_unlock(&kvm
->mmu_lock
);
398 BUG_ON(kvm
->mmu_notifier_count
< 0);
401 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
402 struct mm_struct
*mm
,
406 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
409 idx
= srcu_read_lock(&kvm
->srcu
);
410 spin_lock(&kvm
->mmu_lock
);
412 young
= kvm_age_hva(kvm
, start
, end
);
414 kvm_flush_remote_tlbs(kvm
);
416 spin_unlock(&kvm
->mmu_lock
);
417 srcu_read_unlock(&kvm
->srcu
, idx
);
422 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
423 struct mm_struct
*mm
,
427 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
430 idx
= srcu_read_lock(&kvm
->srcu
);
431 spin_lock(&kvm
->mmu_lock
);
433 * Even though we do not flush TLB, this will still adversely
434 * affect performance on pre-Haswell Intel EPT, where there is
435 * no EPT Access Bit to clear so that we have to tear down EPT
436 * tables instead. If we find this unacceptable, we can always
437 * add a parameter to kvm_age_hva so that it effectively doesn't
438 * do anything on clear_young.
440 * Also note that currently we never issue secondary TLB flushes
441 * from clear_young, leaving this job up to the regular system
442 * cadence. If we find this inaccurate, we might come up with a
443 * more sophisticated heuristic later.
445 young
= kvm_age_hva(kvm
, start
, end
);
446 spin_unlock(&kvm
->mmu_lock
);
447 srcu_read_unlock(&kvm
->srcu
, idx
);
452 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
453 struct mm_struct
*mm
,
454 unsigned long address
)
456 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
459 idx
= srcu_read_lock(&kvm
->srcu
);
460 spin_lock(&kvm
->mmu_lock
);
461 young
= kvm_test_age_hva(kvm
, address
);
462 spin_unlock(&kvm
->mmu_lock
);
463 srcu_read_unlock(&kvm
->srcu
, idx
);
468 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
469 struct mm_struct
*mm
)
471 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
474 idx
= srcu_read_lock(&kvm
->srcu
);
475 kvm_arch_flush_shadow_all(kvm
);
476 srcu_read_unlock(&kvm
->srcu
, idx
);
479 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
480 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
481 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
482 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
483 .clear_young
= kvm_mmu_notifier_clear_young
,
484 .test_young
= kvm_mmu_notifier_test_young
,
485 .change_pte
= kvm_mmu_notifier_change_pte
,
486 .release
= kvm_mmu_notifier_release
,
489 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
491 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
492 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
495 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
497 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
502 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
504 static struct kvm_memslots
*kvm_alloc_memslots(void)
507 struct kvm_memslots
*slots
;
509 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
513 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
514 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
521 if (!memslot
->dirty_bitmap
)
524 kvfree(memslot
->dirty_bitmap
);
525 memslot
->dirty_bitmap
= NULL
;
529 * Free any memory in @free but not in @dont.
531 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
532 struct kvm_memory_slot
*dont
)
534 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
535 kvm_destroy_dirty_bitmap(free
);
537 kvm_arch_free_memslot(kvm
, free
, dont
);
542 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
544 struct kvm_memory_slot
*memslot
;
549 kvm_for_each_memslot(memslot
, slots
)
550 kvm_free_memslot(kvm
, memslot
, NULL
);
555 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
559 if (!kvm
->debugfs_dentry
)
562 debugfs_remove_recursive(kvm
->debugfs_dentry
);
564 if (kvm
->debugfs_stat_data
) {
565 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
566 kfree(kvm
->debugfs_stat_data
[i
]);
567 kfree(kvm
->debugfs_stat_data
);
571 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
573 char dir_name
[ITOA_MAX_LEN
* 2];
574 struct kvm_stat_data
*stat_data
;
575 struct kvm_stats_debugfs_item
*p
;
577 if (!debugfs_initialized())
580 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
581 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
583 if (!kvm
->debugfs_dentry
)
586 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
587 sizeof(*kvm
->debugfs_stat_data
),
589 if (!kvm
->debugfs_stat_data
)
592 for (p
= debugfs_entries
; p
->name
; p
++) {
593 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
597 stat_data
->kvm
= kvm
;
598 stat_data
->offset
= p
->offset
;
599 stat_data
->mode
= p
->mode
? p
->mode
: 0644;
600 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
601 if (!debugfs_create_file(p
->name
, stat_data
->mode
,
604 stat_fops_per_vm
[p
->kind
]))
610 static struct kvm
*kvm_create_vm(unsigned long type
)
613 struct kvm
*kvm
= kvm_arch_alloc_vm();
616 return ERR_PTR(-ENOMEM
);
618 spin_lock_init(&kvm
->mmu_lock
);
620 kvm
->mm
= current
->mm
;
621 kvm_eventfd_init(kvm
);
622 mutex_init(&kvm
->lock
);
623 mutex_init(&kvm
->irq_lock
);
624 mutex_init(&kvm
->slots_lock
);
625 refcount_set(&kvm
->users_count
, 1);
626 INIT_LIST_HEAD(&kvm
->devices
);
628 r
= kvm_arch_init_vm(kvm
, type
);
630 goto out_err_no_disable
;
632 r
= hardware_enable_all();
634 goto out_err_no_disable
;
636 #ifdef CONFIG_HAVE_KVM_IRQFD
637 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
640 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
643 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
644 struct kvm_memslots
*slots
= kvm_alloc_memslots();
646 goto out_err_no_srcu
;
648 * Generations must be different for each address space.
649 * Init kvm generation close to the maximum to easily test the
650 * code of handling generation number wrap-around.
652 slots
->generation
= i
* 2 - 150;
653 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
656 if (init_srcu_struct(&kvm
->srcu
))
657 goto out_err_no_srcu
;
658 if (init_srcu_struct(&kvm
->irq_srcu
))
659 goto out_err_no_irq_srcu
;
660 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
661 rcu_assign_pointer(kvm
->buses
[i
],
662 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
667 r
= kvm_init_mmu_notifier(kvm
);
671 mutex_lock(&kvm_lock
);
672 list_add(&kvm
->vm_list
, &vm_list
);
673 mutex_unlock(&kvm_lock
);
675 preempt_notifier_inc();
680 cleanup_srcu_struct(&kvm
->irq_srcu
);
682 cleanup_srcu_struct(&kvm
->srcu
);
684 hardware_disable_all();
686 refcount_set(&kvm
->users_count
, 0);
687 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
688 kfree(kvm_get_bus(kvm
, i
));
689 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
690 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
691 kvm_arch_free_vm(kvm
);
696 static void kvm_destroy_devices(struct kvm
*kvm
)
698 struct kvm_device
*dev
, *tmp
;
701 * We do not need to take the kvm->lock here, because nobody else
702 * has a reference to the struct kvm at this point and therefore
703 * cannot access the devices list anyhow.
705 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
706 list_del(&dev
->vm_node
);
707 dev
->ops
->destroy(dev
);
711 static void kvm_destroy_vm(struct kvm
*kvm
)
714 struct mm_struct
*mm
= kvm
->mm
;
716 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
717 kvm_destroy_vm_debugfs(kvm
);
718 kvm_arch_sync_events(kvm
);
719 mutex_lock(&kvm_lock
);
720 list_del(&kvm
->vm_list
);
721 mutex_unlock(&kvm_lock
);
722 kvm_free_irq_routing(kvm
);
723 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
724 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
727 kvm_io_bus_destroy(bus
);
728 kvm
->buses
[i
] = NULL
;
730 kvm_coalesced_mmio_free(kvm
);
731 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
732 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
734 kvm_arch_flush_shadow_all(kvm
);
736 kvm_arch_destroy_vm(kvm
);
737 kvm_destroy_devices(kvm
);
738 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
739 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
740 cleanup_srcu_struct(&kvm
->irq_srcu
);
741 cleanup_srcu_struct(&kvm
->srcu
);
742 kvm_arch_free_vm(kvm
);
743 preempt_notifier_dec();
744 hardware_disable_all();
748 void kvm_get_kvm(struct kvm
*kvm
)
750 refcount_inc(&kvm
->users_count
);
752 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
754 void kvm_put_kvm(struct kvm
*kvm
)
756 if (refcount_dec_and_test(&kvm
->users_count
))
759 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
762 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
764 struct kvm
*kvm
= filp
->private_data
;
766 kvm_irqfd_release(kvm
);
773 * Allocation size is twice as large as the actual dirty bitmap size.
774 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
776 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
778 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
780 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
781 if (!memslot
->dirty_bitmap
)
788 * Insert memslot and re-sort memslots based on their GFN,
789 * so binary search could be used to lookup GFN.
790 * Sorting algorithm takes advantage of having initially
791 * sorted array and known changed memslot position.
793 static void update_memslots(struct kvm_memslots
*slots
,
794 struct kvm_memory_slot
*new)
797 int i
= slots
->id_to_index
[id
];
798 struct kvm_memory_slot
*mslots
= slots
->memslots
;
800 WARN_ON(mslots
[i
].id
!= id
);
802 WARN_ON(!mslots
[i
].npages
);
803 if (mslots
[i
].npages
)
806 if (!mslots
[i
].npages
)
810 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
811 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
812 if (!mslots
[i
+ 1].npages
)
814 mslots
[i
] = mslots
[i
+ 1];
815 slots
->id_to_index
[mslots
[i
].id
] = i
;
820 * The ">=" is needed when creating a slot with base_gfn == 0,
821 * so that it moves before all those with base_gfn == npages == 0.
823 * On the other hand, if new->npages is zero, the above loop has
824 * already left i pointing to the beginning of the empty part of
825 * mslots, and the ">=" would move the hole backwards in this
826 * case---which is wrong. So skip the loop when deleting a slot.
830 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
831 mslots
[i
] = mslots
[i
- 1];
832 slots
->id_to_index
[mslots
[i
].id
] = i
;
836 WARN_ON_ONCE(i
!= slots
->used_slots
);
839 slots
->id_to_index
[mslots
[i
].id
] = i
;
842 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
844 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
846 #ifdef __KVM_HAVE_READONLY_MEM
847 valid_flags
|= KVM_MEM_READONLY
;
850 if (mem
->flags
& ~valid_flags
)
856 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
857 int as_id
, struct kvm_memslots
*slots
)
859 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
863 * Set the low bit in the generation, which disables SPTE caching
864 * until the end of synchronize_srcu_expedited.
866 WARN_ON(old_memslots
->generation
& 1);
867 slots
->generation
= old_memslots
->generation
+ 1;
869 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
870 synchronize_srcu_expedited(&kvm
->srcu
);
873 * Increment the new memslot generation a second time. This prevents
874 * vm exits that race with memslot updates from caching a memslot
875 * generation that will (potentially) be valid forever.
877 * Generations must be unique even across address spaces. We do not need
878 * a global counter for that, instead the generation space is evenly split
879 * across address spaces. For example, with two address spaces, address
880 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
881 * use generations 2, 6, 10, 14, ...
883 gen
= slots
->generation
+ KVM_ADDRESS_SPACE_NUM
* 2 - 1;
885 kvm_arch_memslots_updated(kvm
, gen
);
887 slots
->generation
= gen
;
893 * Allocate some memory and give it an address in the guest physical address
896 * Discontiguous memory is allowed, mostly for framebuffers.
898 * Must be called holding kvm->slots_lock for write.
900 int __kvm_set_memory_region(struct kvm
*kvm
,
901 const struct kvm_userspace_memory_region
*mem
)
905 unsigned long npages
;
906 struct kvm_memory_slot
*slot
;
907 struct kvm_memory_slot old
, new;
908 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
910 enum kvm_mr_change change
;
912 r
= check_memory_region_flags(mem
);
917 as_id
= mem
->slot
>> 16;
920 /* General sanity checks */
921 if (mem
->memory_size
& (PAGE_SIZE
- 1))
923 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
925 /* We can read the guest memory with __xxx_user() later on. */
926 if ((id
< KVM_USER_MEM_SLOTS
) &&
927 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
928 !access_ok(VERIFY_WRITE
,
929 (void __user
*)(unsigned long)mem
->userspace_addr
,
932 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
934 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
937 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
938 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
939 npages
= mem
->memory_size
>> PAGE_SHIFT
;
941 if (npages
> KVM_MEM_MAX_NR_PAGES
)
947 new.base_gfn
= base_gfn
;
949 new.flags
= mem
->flags
;
953 change
= KVM_MR_CREATE
;
954 else { /* Modify an existing slot. */
955 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
956 (npages
!= old
.npages
) ||
957 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
960 if (base_gfn
!= old
.base_gfn
)
961 change
= KVM_MR_MOVE
;
962 else if (new.flags
!= old
.flags
)
963 change
= KVM_MR_FLAGS_ONLY
;
964 else { /* Nothing to change. */
973 change
= KVM_MR_DELETE
;
978 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
979 /* Check for overlaps */
981 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
984 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
985 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
990 /* Free page dirty bitmap if unneeded */
991 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
992 new.dirty_bitmap
= NULL
;
995 if (change
== KVM_MR_CREATE
) {
996 new.userspace_addr
= mem
->userspace_addr
;
998 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1002 /* Allocate page dirty bitmap if needed */
1003 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1004 if (kvm_create_dirty_bitmap(&new) < 0)
1008 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1011 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1013 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1014 slot
= id_to_memslot(slots
, id
);
1015 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1017 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1019 /* From this point no new shadow pages pointing to a deleted,
1020 * or moved, memslot will be created.
1022 * validation of sp->gfn happens in:
1023 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1024 * - kvm_is_visible_gfn (mmu_check_roots)
1026 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1029 * We can re-use the old_memslots from above, the only difference
1030 * from the currently installed memslots is the invalid flag. This
1031 * will get overwritten by update_memslots anyway.
1033 slots
= old_memslots
;
1036 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1040 /* actual memory is freed via old in kvm_free_memslot below */
1041 if (change
== KVM_MR_DELETE
) {
1042 new.dirty_bitmap
= NULL
;
1043 memset(&new.arch
, 0, sizeof(new.arch
));
1046 update_memslots(slots
, &new);
1047 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1049 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1051 kvm_free_memslot(kvm
, &old
, &new);
1052 kvfree(old_memslots
);
1058 kvm_free_memslot(kvm
, &new, &old
);
1062 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1064 int kvm_set_memory_region(struct kvm
*kvm
,
1065 const struct kvm_userspace_memory_region
*mem
)
1069 mutex_lock(&kvm
->slots_lock
);
1070 r
= __kvm_set_memory_region(kvm
, mem
);
1071 mutex_unlock(&kvm
->slots_lock
);
1074 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1076 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1077 struct kvm_userspace_memory_region
*mem
)
1079 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1082 return kvm_set_memory_region(kvm
, mem
);
1085 int kvm_get_dirty_log(struct kvm
*kvm
,
1086 struct kvm_dirty_log
*log
, int *is_dirty
)
1088 struct kvm_memslots
*slots
;
1089 struct kvm_memory_slot
*memslot
;
1092 unsigned long any
= 0;
1094 as_id
= log
->slot
>> 16;
1095 id
= (u16
)log
->slot
;
1096 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1099 slots
= __kvm_memslots(kvm
, as_id
);
1100 memslot
= id_to_memslot(slots
, id
);
1101 if (!memslot
->dirty_bitmap
)
1104 n
= kvm_dirty_bitmap_bytes(memslot
);
1106 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1107 any
= memslot
->dirty_bitmap
[i
];
1109 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1116 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1118 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1120 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1121 * are dirty write protect them for next write.
1122 * @kvm: pointer to kvm instance
1123 * @log: slot id and address to which we copy the log
1124 * @is_dirty: flag set if any page is dirty
1126 * We need to keep it in mind that VCPU threads can write to the bitmap
1127 * concurrently. So, to avoid losing track of dirty pages we keep the
1130 * 1. Take a snapshot of the bit and clear it if needed.
1131 * 2. Write protect the corresponding page.
1132 * 3. Copy the snapshot to the userspace.
1133 * 4. Upon return caller flushes TLB's if needed.
1135 * Between 2 and 4, the guest may write to the page using the remaining TLB
1136 * entry. This is not a problem because the page is reported dirty using
1137 * the snapshot taken before and step 4 ensures that writes done after
1138 * exiting to userspace will be logged for the next call.
1141 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1142 struct kvm_dirty_log
*log
, bool *is_dirty
)
1144 struct kvm_memslots
*slots
;
1145 struct kvm_memory_slot
*memslot
;
1148 unsigned long *dirty_bitmap
;
1149 unsigned long *dirty_bitmap_buffer
;
1151 as_id
= log
->slot
>> 16;
1152 id
= (u16
)log
->slot
;
1153 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1156 slots
= __kvm_memslots(kvm
, as_id
);
1157 memslot
= id_to_memslot(slots
, id
);
1159 dirty_bitmap
= memslot
->dirty_bitmap
;
1163 n
= kvm_dirty_bitmap_bytes(memslot
);
1165 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1166 memset(dirty_bitmap_buffer
, 0, n
);
1168 spin_lock(&kvm
->mmu_lock
);
1170 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1174 if (!dirty_bitmap
[i
])
1179 mask
= xchg(&dirty_bitmap
[i
], 0);
1180 dirty_bitmap_buffer
[i
] = mask
;
1183 offset
= i
* BITS_PER_LONG
;
1184 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1189 spin_unlock(&kvm
->mmu_lock
);
1190 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1194 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1197 bool kvm_largepages_enabled(void)
1199 return largepages_enabled
;
1202 void kvm_disable_largepages(void)
1204 largepages_enabled
= false;
1206 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1208 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1210 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1212 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1214 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1216 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1219 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1221 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1223 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1224 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1229 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1231 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1233 struct vm_area_struct
*vma
;
1234 unsigned long addr
, size
;
1238 addr
= gfn_to_hva(kvm
, gfn
);
1239 if (kvm_is_error_hva(addr
))
1242 down_read(¤t
->mm
->mmap_sem
);
1243 vma
= find_vma(current
->mm
, addr
);
1247 size
= vma_kernel_pagesize(vma
);
1250 up_read(¤t
->mm
->mmap_sem
);
1255 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1257 return slot
->flags
& KVM_MEM_READONLY
;
1260 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1261 gfn_t
*nr_pages
, bool write
)
1263 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1264 return KVM_HVA_ERR_BAD
;
1266 if (memslot_is_readonly(slot
) && write
)
1267 return KVM_HVA_ERR_RO_BAD
;
1270 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1272 return __gfn_to_hva_memslot(slot
, gfn
);
1275 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1278 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1281 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1284 return gfn_to_hva_many(slot
, gfn
, NULL
);
1286 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1288 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1290 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1292 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1294 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1296 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1298 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1301 * If writable is set to false, the hva returned by this function is only
1302 * allowed to be read.
1304 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1305 gfn_t gfn
, bool *writable
)
1307 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1309 if (!kvm_is_error_hva(hva
) && writable
)
1310 *writable
= !memslot_is_readonly(slot
);
1315 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1317 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1319 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1322 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1324 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1326 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1329 static int get_user_page_nowait(unsigned long start
, int write
,
1332 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1335 flags
|= FOLL_WRITE
;
1337 return get_user_pages(start
, 1, flags
, page
, NULL
);
1340 static inline int check_user_page_hwpoison(unsigned long addr
)
1342 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1344 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1345 return rc
== -EHWPOISON
;
1349 * The atomic path to get the writable pfn which will be stored in @pfn,
1350 * true indicates success, otherwise false is returned.
1352 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1353 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1355 struct page
*page
[1];
1358 if (!(async
|| atomic
))
1362 * Fast pin a writable pfn only if it is a write fault request
1363 * or the caller allows to map a writable pfn for a read fault
1366 if (!(write_fault
|| writable
))
1369 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1371 *pfn
= page_to_pfn(page
[0]);
1382 * The slow path to get the pfn of the specified host virtual address,
1383 * 1 indicates success, -errno is returned if error is detected.
1385 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1386 bool *writable
, kvm_pfn_t
*pfn
)
1388 struct page
*page
[1];
1394 *writable
= write_fault
;
1397 down_read(¤t
->mm
->mmap_sem
);
1398 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1399 up_read(¤t
->mm
->mmap_sem
);
1401 unsigned int flags
= FOLL_HWPOISON
;
1404 flags
|= FOLL_WRITE
;
1406 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1411 /* map read fault as writable if possible */
1412 if (unlikely(!write_fault
) && writable
) {
1413 struct page
*wpage
[1];
1415 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1424 *pfn
= page_to_pfn(page
[0]);
1428 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1430 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1433 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1439 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1440 unsigned long addr
, bool *async
,
1441 bool write_fault
, bool *writable
,
1447 r
= follow_pfn(vma
, addr
, &pfn
);
1450 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1451 * not call the fault handler, so do it here.
1453 bool unlocked
= false;
1454 r
= fixup_user_fault(current
, current
->mm
, addr
,
1455 (write_fault
? FAULT_FLAG_WRITE
: 0),
1462 r
= follow_pfn(vma
, addr
, &pfn
);
1472 * Get a reference here because callers of *hva_to_pfn* and
1473 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1474 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1475 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1476 * simply do nothing for reserved pfns.
1478 * Whoever called remap_pfn_range is also going to call e.g.
1479 * unmap_mapping_range before the underlying pages are freed,
1480 * causing a call to our MMU notifier.
1489 * Pin guest page in memory and return its pfn.
1490 * @addr: host virtual address which maps memory to the guest
1491 * @atomic: whether this function can sleep
1492 * @async: whether this function need to wait IO complete if the
1493 * host page is not in the memory
1494 * @write_fault: whether we should get a writable host page
1495 * @writable: whether it allows to map a writable host page for !@write_fault
1497 * The function will map a writable host page for these two cases:
1498 * 1): @write_fault = true
1499 * 2): @write_fault = false && @writable, @writable will tell the caller
1500 * whether the mapping is writable.
1502 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1503 bool write_fault
, bool *writable
)
1505 struct vm_area_struct
*vma
;
1509 /* we can do it either atomically or asynchronously, not both */
1510 BUG_ON(atomic
&& async
);
1512 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1516 return KVM_PFN_ERR_FAULT
;
1518 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1522 down_read(¤t
->mm
->mmap_sem
);
1523 if (npages
== -EHWPOISON
||
1524 (!async
&& check_user_page_hwpoison(addr
))) {
1525 pfn
= KVM_PFN_ERR_HWPOISON
;
1530 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1533 pfn
= KVM_PFN_ERR_FAULT
;
1534 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1535 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1539 pfn
= KVM_PFN_ERR_FAULT
;
1541 if (async
&& vma_is_valid(vma
, write_fault
))
1543 pfn
= KVM_PFN_ERR_FAULT
;
1546 up_read(¤t
->mm
->mmap_sem
);
1550 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1551 bool atomic
, bool *async
, bool write_fault
,
1554 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1556 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1559 return KVM_PFN_ERR_RO_FAULT
;
1562 if (kvm_is_error_hva(addr
)) {
1565 return KVM_PFN_NOSLOT
;
1568 /* Do not map writable pfn in the readonly memslot. */
1569 if (writable
&& memslot_is_readonly(slot
)) {
1574 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1577 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1579 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1582 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1583 write_fault
, writable
);
1585 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1587 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1589 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1591 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1593 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1595 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1597 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1599 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1601 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1603 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1605 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1607 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1609 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1611 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1613 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1615 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1617 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1619 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1621 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1623 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1624 struct page
**pages
, int nr_pages
)
1629 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1630 if (kvm_is_error_hva(addr
))
1633 if (entry
< nr_pages
)
1636 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1638 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1640 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1642 if (is_error_noslot_pfn(pfn
))
1643 return KVM_ERR_PTR_BAD_PAGE
;
1645 if (kvm_is_reserved_pfn(pfn
)) {
1647 return KVM_ERR_PTR_BAD_PAGE
;
1650 return pfn_to_page(pfn
);
1653 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1657 pfn
= gfn_to_pfn(kvm
, gfn
);
1659 return kvm_pfn_to_page(pfn
);
1661 EXPORT_SYMBOL_GPL(gfn_to_page
);
1663 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1667 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1669 return kvm_pfn_to_page(pfn
);
1671 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1673 void kvm_release_page_clean(struct page
*page
)
1675 WARN_ON(is_error_page(page
));
1677 kvm_release_pfn_clean(page_to_pfn(page
));
1679 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1681 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1683 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1684 put_page(pfn_to_page(pfn
));
1686 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1688 void kvm_release_page_dirty(struct page
*page
)
1690 WARN_ON(is_error_page(page
));
1692 kvm_release_pfn_dirty(page_to_pfn(page
));
1694 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1696 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1698 kvm_set_pfn_dirty(pfn
);
1699 kvm_release_pfn_clean(pfn
);
1701 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1703 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1705 if (!kvm_is_reserved_pfn(pfn
)) {
1706 struct page
*page
= pfn_to_page(pfn
);
1708 if (!PageReserved(page
))
1712 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1714 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1716 if (!kvm_is_reserved_pfn(pfn
))
1717 mark_page_accessed(pfn_to_page(pfn
));
1719 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1721 void kvm_get_pfn(kvm_pfn_t pfn
)
1723 if (!kvm_is_reserved_pfn(pfn
))
1724 get_page(pfn_to_page(pfn
));
1726 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1728 static int next_segment(unsigned long len
, int offset
)
1730 if (len
> PAGE_SIZE
- offset
)
1731 return PAGE_SIZE
- offset
;
1736 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1737 void *data
, int offset
, int len
)
1742 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1743 if (kvm_is_error_hva(addr
))
1745 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1751 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1754 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1756 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1758 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1760 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1761 int offset
, int len
)
1763 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1765 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1767 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1769 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1771 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1773 int offset
= offset_in_page(gpa
);
1776 while ((seg
= next_segment(len
, offset
)) != 0) {
1777 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1787 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1789 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1791 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1793 int offset
= offset_in_page(gpa
);
1796 while ((seg
= next_segment(len
, offset
)) != 0) {
1797 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1807 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1809 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1810 void *data
, int offset
, unsigned long len
)
1815 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1816 if (kvm_is_error_hva(addr
))
1818 pagefault_disable();
1819 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1826 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1829 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1830 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1831 int offset
= offset_in_page(gpa
);
1833 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1835 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1837 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1838 void *data
, unsigned long len
)
1840 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1841 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1842 int offset
= offset_in_page(gpa
);
1844 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1846 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1848 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1849 const void *data
, int offset
, int len
)
1854 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1855 if (kvm_is_error_hva(addr
))
1857 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1860 mark_page_dirty_in_slot(memslot
, gfn
);
1864 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1865 const void *data
, int offset
, int len
)
1867 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1869 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1871 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1873 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1874 const void *data
, int offset
, int len
)
1876 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1878 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1880 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1882 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1885 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1887 int offset
= offset_in_page(gpa
);
1890 while ((seg
= next_segment(len
, offset
)) != 0) {
1891 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1901 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1903 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1906 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1908 int offset
= offset_in_page(gpa
);
1911 while ((seg
= next_segment(len
, offset
)) != 0) {
1912 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1922 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1924 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1925 struct gfn_to_hva_cache
*ghc
,
1926 gpa_t gpa
, unsigned long len
)
1928 int offset
= offset_in_page(gpa
);
1929 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1930 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1931 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1932 gfn_t nr_pages_avail
;
1935 ghc
->generation
= slots
->generation
;
1937 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1938 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1939 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1943 * If the requested region crosses two memslots, we still
1944 * verify that the entire region is valid here.
1946 while (start_gfn
<= end_gfn
) {
1948 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1949 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1951 if (kvm_is_error_hva(ghc
->hva
))
1953 start_gfn
+= nr_pages_avail
;
1955 /* Use the slow path for cross page reads and writes. */
1956 ghc
->memslot
= NULL
;
1961 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1962 gpa_t gpa
, unsigned long len
)
1964 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1965 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1967 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1969 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1970 void *data
, unsigned int offset
,
1973 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1975 gpa_t gpa
= ghc
->gpa
+ offset
;
1977 BUG_ON(len
+ offset
> ghc
->len
);
1979 if (slots
->generation
!= ghc
->generation
)
1980 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1982 if (unlikely(!ghc
->memslot
))
1983 return kvm_write_guest(kvm
, gpa
, data
, len
);
1985 if (kvm_is_error_hva(ghc
->hva
))
1988 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
1991 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
1995 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
1997 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1998 void *data
, unsigned long len
)
2000 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2002 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2004 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2005 void *data
, unsigned long len
)
2007 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2010 BUG_ON(len
> ghc
->len
);
2012 if (slots
->generation
!= ghc
->generation
)
2013 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2015 if (unlikely(!ghc
->memslot
))
2016 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2018 if (kvm_is_error_hva(ghc
->hva
))
2021 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2027 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2029 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2031 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2033 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2035 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2037 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2039 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2041 int offset
= offset_in_page(gpa
);
2044 while ((seg
= next_segment(len
, offset
)) != 0) {
2045 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2054 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2056 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2059 if (memslot
&& memslot
->dirty_bitmap
) {
2060 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2062 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2066 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2068 struct kvm_memory_slot
*memslot
;
2070 memslot
= gfn_to_memslot(kvm
, gfn
);
2071 mark_page_dirty_in_slot(memslot
, gfn
);
2073 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2075 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2077 struct kvm_memory_slot
*memslot
;
2079 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2080 mark_page_dirty_in_slot(memslot
, gfn
);
2082 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2084 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2086 if (!vcpu
->sigset_active
)
2090 * This does a lockless modification of ->real_blocked, which is fine
2091 * because, only current can change ->real_blocked and all readers of
2092 * ->real_blocked don't care as long ->real_blocked is always a subset
2095 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2098 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2100 if (!vcpu
->sigset_active
)
2103 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2104 sigemptyset(¤t
->real_blocked
);
2107 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2109 unsigned int old
, val
, grow
;
2111 old
= val
= vcpu
->halt_poll_ns
;
2112 grow
= READ_ONCE(halt_poll_ns_grow
);
2114 if (val
== 0 && grow
)
2119 if (val
> halt_poll_ns
)
2122 vcpu
->halt_poll_ns
= val
;
2123 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2126 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2128 unsigned int old
, val
, shrink
;
2130 old
= val
= vcpu
->halt_poll_ns
;
2131 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2137 vcpu
->halt_poll_ns
= val
;
2138 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2141 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2143 if (kvm_arch_vcpu_runnable(vcpu
)) {
2144 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2147 if (kvm_cpu_has_pending_timer(vcpu
))
2149 if (signal_pending(current
))
2156 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2158 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2161 DECLARE_SWAITQUEUE(wait
);
2162 bool waited
= false;
2165 start
= cur
= ktime_get();
2166 if (vcpu
->halt_poll_ns
) {
2167 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2169 ++vcpu
->stat
.halt_attempted_poll
;
2172 * This sets KVM_REQ_UNHALT if an interrupt
2175 if (kvm_vcpu_check_block(vcpu
) < 0) {
2176 ++vcpu
->stat
.halt_successful_poll
;
2177 if (!vcpu_valid_wakeup(vcpu
))
2178 ++vcpu
->stat
.halt_poll_invalid
;
2182 } while (single_task_running() && ktime_before(cur
, stop
));
2185 kvm_arch_vcpu_blocking(vcpu
);
2188 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2190 if (kvm_vcpu_check_block(vcpu
) < 0)
2197 finish_swait(&vcpu
->wq
, &wait
);
2200 kvm_arch_vcpu_unblocking(vcpu
);
2202 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2204 if (!vcpu_valid_wakeup(vcpu
))
2205 shrink_halt_poll_ns(vcpu
);
2206 else if (halt_poll_ns
) {
2207 if (block_ns
<= vcpu
->halt_poll_ns
)
2209 /* we had a long block, shrink polling */
2210 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2211 shrink_halt_poll_ns(vcpu
);
2212 /* we had a short halt and our poll time is too small */
2213 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2214 block_ns
< halt_poll_ns
)
2215 grow_halt_poll_ns(vcpu
);
2217 vcpu
->halt_poll_ns
= 0;
2219 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2220 kvm_arch_vcpu_block_finish(vcpu
);
2222 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2224 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2226 struct swait_queue_head
*wqp
;
2228 wqp
= kvm_arch_vcpu_wq(vcpu
);
2229 if (swq_has_sleeper(wqp
)) {
2231 ++vcpu
->stat
.halt_wakeup
;
2237 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2241 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2243 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2246 int cpu
= vcpu
->cpu
;
2248 if (kvm_vcpu_wake_up(vcpu
))
2252 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2253 if (kvm_arch_vcpu_should_kick(vcpu
))
2254 smp_send_reschedule(cpu
);
2257 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2258 #endif /* !CONFIG_S390 */
2260 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2263 struct task_struct
*task
= NULL
;
2267 pid
= rcu_dereference(target
->pid
);
2269 task
= get_pid_task(pid
, PIDTYPE_PID
);
2273 ret
= yield_to(task
, 1);
2274 put_task_struct(task
);
2278 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2281 * Helper that checks whether a VCPU is eligible for directed yield.
2282 * Most eligible candidate to yield is decided by following heuristics:
2284 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2285 * (preempted lock holder), indicated by @in_spin_loop.
2286 * Set at the beiginning and cleared at the end of interception/PLE handler.
2288 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2289 * chance last time (mostly it has become eligible now since we have probably
2290 * yielded to lockholder in last iteration. This is done by toggling
2291 * @dy_eligible each time a VCPU checked for eligibility.)
2293 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2294 * to preempted lock-holder could result in wrong VCPU selection and CPU
2295 * burning. Giving priority for a potential lock-holder increases lock
2298 * Since algorithm is based on heuristics, accessing another VCPU data without
2299 * locking does not harm. It may result in trying to yield to same VCPU, fail
2300 * and continue with next VCPU and so on.
2302 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2304 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2307 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2308 vcpu
->spin_loop
.dy_eligible
;
2310 if (vcpu
->spin_loop
.in_spin_loop
)
2311 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2320 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2321 * a vcpu_load/vcpu_put pair. However, for most architectures
2322 * kvm_arch_vcpu_runnable does not require vcpu_load.
2324 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
2326 return kvm_arch_vcpu_runnable(vcpu
);
2329 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
2331 if (kvm_arch_dy_runnable(vcpu
))
2334 #ifdef CONFIG_KVM_ASYNC_PF
2335 if (!list_empty_careful(&vcpu
->async_pf
.done
))
2342 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2344 struct kvm
*kvm
= me
->kvm
;
2345 struct kvm_vcpu
*vcpu
;
2346 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2352 kvm_vcpu_set_in_spin_loop(me
, true);
2354 * We boost the priority of a VCPU that is runnable but not
2355 * currently running, because it got preempted by something
2356 * else and called schedule in __vcpu_run. Hopefully that
2357 * VCPU is holding the lock that we need and will release it.
2358 * We approximate round-robin by starting at the last boosted VCPU.
2360 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2361 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2362 if (!pass
&& i
<= last_boosted_vcpu
) {
2363 i
= last_boosted_vcpu
;
2365 } else if (pass
&& i
> last_boosted_vcpu
)
2367 if (!READ_ONCE(vcpu
->preempted
))
2371 if (swait_active(&vcpu
->wq
) && !vcpu_dy_runnable(vcpu
))
2373 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2375 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2378 yielded
= kvm_vcpu_yield_to(vcpu
);
2380 kvm
->last_boosted_vcpu
= i
;
2382 } else if (yielded
< 0) {
2389 kvm_vcpu_set_in_spin_loop(me
, false);
2391 /* Ensure vcpu is not eligible during next spinloop */
2392 kvm_vcpu_set_dy_eligible(me
, false);
2394 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2396 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2398 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2401 if (vmf
->pgoff
== 0)
2402 page
= virt_to_page(vcpu
->run
);
2404 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2405 page
= virt_to_page(vcpu
->arch
.pio_data
);
2407 #ifdef CONFIG_KVM_MMIO
2408 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2409 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2412 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2418 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2419 .fault
= kvm_vcpu_fault
,
2422 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2424 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2428 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2430 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2432 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2433 kvm_put_kvm(vcpu
->kvm
);
2437 static struct file_operations kvm_vcpu_fops
= {
2438 .release
= kvm_vcpu_release
,
2439 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2440 #ifdef CONFIG_KVM_COMPAT
2441 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2443 .mmap
= kvm_vcpu_mmap
,
2444 .llseek
= noop_llseek
,
2448 * Allocates an inode for the vcpu.
2450 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2452 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2455 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2457 char dir_name
[ITOA_MAX_LEN
* 2];
2460 if (!kvm_arch_has_vcpu_debugfs())
2463 if (!debugfs_initialized())
2466 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2467 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2468 vcpu
->kvm
->debugfs_dentry
);
2469 if (!vcpu
->debugfs_dentry
)
2472 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2474 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2482 * Creates some virtual cpus. Good luck creating more than one.
2484 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2487 struct kvm_vcpu
*vcpu
;
2489 if (id
>= KVM_MAX_VCPU_ID
)
2492 mutex_lock(&kvm
->lock
);
2493 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2494 mutex_unlock(&kvm
->lock
);
2498 kvm
->created_vcpus
++;
2499 mutex_unlock(&kvm
->lock
);
2501 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2504 goto vcpu_decrement
;
2507 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2509 r
= kvm_arch_vcpu_setup(vcpu
);
2513 r
= kvm_create_vcpu_debugfs(vcpu
);
2517 mutex_lock(&kvm
->lock
);
2518 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2520 goto unlock_vcpu_destroy
;
2523 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2525 /* Now it's all set up, let userspace reach it */
2527 r
= create_vcpu_fd(vcpu
);
2530 goto unlock_vcpu_destroy
;
2533 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2536 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2537 * before kvm->online_vcpu's incremented value.
2540 atomic_inc(&kvm
->online_vcpus
);
2542 mutex_unlock(&kvm
->lock
);
2543 kvm_arch_vcpu_postcreate(vcpu
);
2546 unlock_vcpu_destroy
:
2547 mutex_unlock(&kvm
->lock
);
2548 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2550 kvm_arch_vcpu_destroy(vcpu
);
2552 mutex_lock(&kvm
->lock
);
2553 kvm
->created_vcpus
--;
2554 mutex_unlock(&kvm
->lock
);
2558 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2561 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2562 vcpu
->sigset_active
= 1;
2563 vcpu
->sigset
= *sigset
;
2565 vcpu
->sigset_active
= 0;
2569 static long kvm_vcpu_ioctl(struct file
*filp
,
2570 unsigned int ioctl
, unsigned long arg
)
2572 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2573 void __user
*argp
= (void __user
*)arg
;
2575 struct kvm_fpu
*fpu
= NULL
;
2576 struct kvm_sregs
*kvm_sregs
= NULL
;
2578 if (vcpu
->kvm
->mm
!= current
->mm
)
2581 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2584 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2586 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2587 * so vcpu_load() would break it.
2589 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2590 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2594 r
= vcpu_load(vcpu
);
2603 oldpid
= rcu_access_pointer(vcpu
->pid
);
2604 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2605 /* The thread running this VCPU changed. */
2606 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2608 rcu_assign_pointer(vcpu
->pid
, newpid
);
2613 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2614 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2617 case KVM_GET_REGS
: {
2618 struct kvm_regs
*kvm_regs
;
2621 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2624 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2628 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2635 case KVM_SET_REGS
: {
2636 struct kvm_regs
*kvm_regs
;
2639 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2640 if (IS_ERR(kvm_regs
)) {
2641 r
= PTR_ERR(kvm_regs
);
2644 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2648 case KVM_GET_SREGS
: {
2649 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2653 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2657 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2662 case KVM_SET_SREGS
: {
2663 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2664 if (IS_ERR(kvm_sregs
)) {
2665 r
= PTR_ERR(kvm_sregs
);
2669 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2672 case KVM_GET_MP_STATE
: {
2673 struct kvm_mp_state mp_state
;
2675 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2679 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2684 case KVM_SET_MP_STATE
: {
2685 struct kvm_mp_state mp_state
;
2688 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2690 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2693 case KVM_TRANSLATE
: {
2694 struct kvm_translation tr
;
2697 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2699 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2703 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2708 case KVM_SET_GUEST_DEBUG
: {
2709 struct kvm_guest_debug dbg
;
2712 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2714 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2717 case KVM_SET_SIGNAL_MASK
: {
2718 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2719 struct kvm_signal_mask kvm_sigmask
;
2720 sigset_t sigset
, *p
;
2725 if (copy_from_user(&kvm_sigmask
, argp
,
2726 sizeof(kvm_sigmask
)))
2729 if (kvm_sigmask
.len
!= sizeof(sigset
))
2732 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2737 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2741 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2745 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2749 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2755 fpu
= memdup_user(argp
, sizeof(*fpu
));
2761 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2765 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2774 #ifdef CONFIG_KVM_COMPAT
2775 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2776 unsigned int ioctl
, unsigned long arg
)
2778 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2779 void __user
*argp
= compat_ptr(arg
);
2782 if (vcpu
->kvm
->mm
!= current
->mm
)
2786 case KVM_SET_SIGNAL_MASK
: {
2787 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2788 struct kvm_signal_mask kvm_sigmask
;
2793 if (copy_from_user(&kvm_sigmask
, argp
,
2794 sizeof(kvm_sigmask
)))
2797 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2800 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2802 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2804 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2808 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2816 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2817 int (*accessor
)(struct kvm_device
*dev
,
2818 struct kvm_device_attr
*attr
),
2821 struct kvm_device_attr attr
;
2826 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2829 return accessor(dev
, &attr
);
2832 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2835 struct kvm_device
*dev
= filp
->private_data
;
2837 if (dev
->kvm
->mm
!= current
->mm
)
2841 case KVM_SET_DEVICE_ATTR
:
2842 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2843 case KVM_GET_DEVICE_ATTR
:
2844 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2845 case KVM_HAS_DEVICE_ATTR
:
2846 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2848 if (dev
->ops
->ioctl
)
2849 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2855 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2857 struct kvm_device
*dev
= filp
->private_data
;
2858 struct kvm
*kvm
= dev
->kvm
;
2864 static const struct file_operations kvm_device_fops
= {
2865 .unlocked_ioctl
= kvm_device_ioctl
,
2866 #ifdef CONFIG_KVM_COMPAT
2867 .compat_ioctl
= kvm_device_ioctl
,
2869 .release
= kvm_device_release
,
2872 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2874 if (filp
->f_op
!= &kvm_device_fops
)
2877 return filp
->private_data
;
2880 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2881 #ifdef CONFIG_KVM_MPIC
2882 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2883 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2887 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2889 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2892 if (kvm_device_ops_table
[type
] != NULL
)
2895 kvm_device_ops_table
[type
] = ops
;
2899 void kvm_unregister_device_ops(u32 type
)
2901 if (kvm_device_ops_table
[type
] != NULL
)
2902 kvm_device_ops_table
[type
] = NULL
;
2905 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2906 struct kvm_create_device
*cd
)
2908 struct kvm_device_ops
*ops
= NULL
;
2909 struct kvm_device
*dev
;
2910 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2914 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2917 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
2918 ops
= kvm_device_ops_table
[type
];
2925 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2932 mutex_lock(&kvm
->lock
);
2933 ret
= ops
->create(dev
, type
);
2935 mutex_unlock(&kvm
->lock
);
2939 list_add(&dev
->vm_node
, &kvm
->devices
);
2940 mutex_unlock(&kvm
->lock
);
2946 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2949 mutex_lock(&kvm
->lock
);
2950 list_del(&dev
->vm_node
);
2951 mutex_unlock(&kvm
->lock
);
2960 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2963 case KVM_CAP_USER_MEMORY
:
2964 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2965 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2966 case KVM_CAP_INTERNAL_ERROR_DATA
:
2967 #ifdef CONFIG_HAVE_KVM_MSI
2968 case KVM_CAP_SIGNAL_MSI
:
2970 #ifdef CONFIG_HAVE_KVM_IRQFD
2972 case KVM_CAP_IRQFD_RESAMPLE
:
2974 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2975 case KVM_CAP_CHECK_EXTENSION_VM
:
2977 #ifdef CONFIG_KVM_MMIO
2978 case KVM_CAP_COALESCED_MMIO
:
2979 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2981 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2982 case KVM_CAP_IRQ_ROUTING
:
2983 return KVM_MAX_IRQ_ROUTES
;
2985 #if KVM_ADDRESS_SPACE_NUM > 1
2986 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2987 return KVM_ADDRESS_SPACE_NUM
;
2992 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2995 static long kvm_vm_ioctl(struct file
*filp
,
2996 unsigned int ioctl
, unsigned long arg
)
2998 struct kvm
*kvm
= filp
->private_data
;
2999 void __user
*argp
= (void __user
*)arg
;
3002 if (kvm
->mm
!= current
->mm
)
3005 case KVM_CREATE_VCPU
:
3006 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3008 case KVM_SET_USER_MEMORY_REGION
: {
3009 struct kvm_userspace_memory_region kvm_userspace_mem
;
3012 if (copy_from_user(&kvm_userspace_mem
, argp
,
3013 sizeof(kvm_userspace_mem
)))
3016 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3019 case KVM_GET_DIRTY_LOG
: {
3020 struct kvm_dirty_log log
;
3023 if (copy_from_user(&log
, argp
, sizeof(log
)))
3025 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3028 #ifdef CONFIG_KVM_MMIO
3029 case KVM_REGISTER_COALESCED_MMIO
: {
3030 struct kvm_coalesced_mmio_zone zone
;
3033 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3035 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3038 case KVM_UNREGISTER_COALESCED_MMIO
: {
3039 struct kvm_coalesced_mmio_zone zone
;
3042 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3044 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3049 struct kvm_irqfd data
;
3052 if (copy_from_user(&data
, argp
, sizeof(data
)))
3054 r
= kvm_irqfd(kvm
, &data
);
3057 case KVM_IOEVENTFD
: {
3058 struct kvm_ioeventfd data
;
3061 if (copy_from_user(&data
, argp
, sizeof(data
)))
3063 r
= kvm_ioeventfd(kvm
, &data
);
3066 #ifdef CONFIG_HAVE_KVM_MSI
3067 case KVM_SIGNAL_MSI
: {
3071 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3073 r
= kvm_send_userspace_msi(kvm
, &msi
);
3077 #ifdef __KVM_HAVE_IRQ_LINE
3078 case KVM_IRQ_LINE_STATUS
:
3079 case KVM_IRQ_LINE
: {
3080 struct kvm_irq_level irq_event
;
3083 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3086 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3087 ioctl
== KVM_IRQ_LINE_STATUS
);
3092 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3093 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3101 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3102 case KVM_SET_GSI_ROUTING
: {
3103 struct kvm_irq_routing routing
;
3104 struct kvm_irq_routing __user
*urouting
;
3105 struct kvm_irq_routing_entry
*entries
= NULL
;
3108 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3111 if (!kvm_arch_can_set_irq_routing(kvm
))
3113 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3119 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3124 if (copy_from_user(entries
, urouting
->entries
,
3125 routing
.nr
* sizeof(*entries
)))
3126 goto out_free_irq_routing
;
3128 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3130 out_free_irq_routing
:
3134 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3135 case KVM_CREATE_DEVICE
: {
3136 struct kvm_create_device cd
;
3139 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3142 r
= kvm_ioctl_create_device(kvm
, &cd
);
3147 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3153 case KVM_CHECK_EXTENSION
:
3154 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3157 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3163 #ifdef CONFIG_KVM_COMPAT
3164 struct compat_kvm_dirty_log
{
3168 compat_uptr_t dirty_bitmap
; /* one bit per page */
3173 static long kvm_vm_compat_ioctl(struct file
*filp
,
3174 unsigned int ioctl
, unsigned long arg
)
3176 struct kvm
*kvm
= filp
->private_data
;
3179 if (kvm
->mm
!= current
->mm
)
3182 case KVM_GET_DIRTY_LOG
: {
3183 struct compat_kvm_dirty_log compat_log
;
3184 struct kvm_dirty_log log
;
3186 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3187 sizeof(compat_log
)))
3189 log
.slot
= compat_log
.slot
;
3190 log
.padding1
= compat_log
.padding1
;
3191 log
.padding2
= compat_log
.padding2
;
3192 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3194 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3198 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3204 static struct file_operations kvm_vm_fops
= {
3205 .release
= kvm_vm_release
,
3206 .unlocked_ioctl
= kvm_vm_ioctl
,
3207 #ifdef CONFIG_KVM_COMPAT
3208 .compat_ioctl
= kvm_vm_compat_ioctl
,
3210 .llseek
= noop_llseek
,
3213 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3219 kvm
= kvm_create_vm(type
);
3221 return PTR_ERR(kvm
);
3222 #ifdef CONFIG_KVM_MMIO
3223 r
= kvm_coalesced_mmio_init(kvm
);
3229 r
= get_unused_fd_flags(O_CLOEXEC
);
3234 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3238 return PTR_ERR(file
);
3242 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3243 * already set, with ->release() being kvm_vm_release(). In error
3244 * cases it will be called by the final fput(file) and will take
3245 * care of doing kvm_put_kvm(kvm).
3247 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3252 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3254 fd_install(r
, file
);
3258 static long kvm_dev_ioctl(struct file
*filp
,
3259 unsigned int ioctl
, unsigned long arg
)
3264 case KVM_GET_API_VERSION
:
3267 r
= KVM_API_VERSION
;
3270 r
= kvm_dev_ioctl_create_vm(arg
);
3272 case KVM_CHECK_EXTENSION
:
3273 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3275 case KVM_GET_VCPU_MMAP_SIZE
:
3278 r
= PAGE_SIZE
; /* struct kvm_run */
3280 r
+= PAGE_SIZE
; /* pio data page */
3282 #ifdef CONFIG_KVM_MMIO
3283 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3286 case KVM_TRACE_ENABLE
:
3287 case KVM_TRACE_PAUSE
:
3288 case KVM_TRACE_DISABLE
:
3292 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3298 static struct file_operations kvm_chardev_ops
= {
3299 .unlocked_ioctl
= kvm_dev_ioctl
,
3300 .compat_ioctl
= kvm_dev_ioctl
,
3301 .llseek
= noop_llseek
,
3304 static struct miscdevice kvm_dev
= {
3310 static void hardware_enable_nolock(void *junk
)
3312 int cpu
= raw_smp_processor_id();
3315 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3318 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3320 r
= kvm_arch_hardware_enable();
3323 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3324 atomic_inc(&hardware_enable_failed
);
3325 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3329 static int kvm_starting_cpu(unsigned int cpu
)
3331 raw_spin_lock(&kvm_count_lock
);
3332 if (kvm_usage_count
)
3333 hardware_enable_nolock(NULL
);
3334 raw_spin_unlock(&kvm_count_lock
);
3338 static void hardware_disable_nolock(void *junk
)
3340 int cpu
= raw_smp_processor_id();
3342 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3344 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3345 kvm_arch_hardware_disable();
3348 static int kvm_dying_cpu(unsigned int cpu
)
3350 raw_spin_lock(&kvm_count_lock
);
3351 if (kvm_usage_count
)
3352 hardware_disable_nolock(NULL
);
3353 raw_spin_unlock(&kvm_count_lock
);
3357 static void hardware_disable_all_nolock(void)
3359 BUG_ON(!kvm_usage_count
);
3362 if (!kvm_usage_count
)
3363 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3366 static void hardware_disable_all(void)
3368 raw_spin_lock(&kvm_count_lock
);
3369 hardware_disable_all_nolock();
3370 raw_spin_unlock(&kvm_count_lock
);
3373 static int hardware_enable_all(void)
3377 raw_spin_lock(&kvm_count_lock
);
3380 if (kvm_usage_count
== 1) {
3381 atomic_set(&hardware_enable_failed
, 0);
3382 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3384 if (atomic_read(&hardware_enable_failed
)) {
3385 hardware_disable_all_nolock();
3390 raw_spin_unlock(&kvm_count_lock
);
3395 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3399 * Some (well, at least mine) BIOSes hang on reboot if
3402 * And Intel TXT required VMX off for all cpu when system shutdown.
3404 pr_info("kvm: exiting hardware virtualization\n");
3405 kvm_rebooting
= true;
3406 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3410 static struct notifier_block kvm_reboot_notifier
= {
3411 .notifier_call
= kvm_reboot
,
3415 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3419 for (i
= 0; i
< bus
->dev_count
; i
++) {
3420 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3422 kvm_iodevice_destructor(pos
);
3427 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3428 const struct kvm_io_range
*r2
)
3430 gpa_t addr1
= r1
->addr
;
3431 gpa_t addr2
= r2
->addr
;
3436 /* If r2->len == 0, match the exact address. If r2->len != 0,
3437 * accept any overlapping write. Any order is acceptable for
3438 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3439 * we process all of them.
3452 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3454 return kvm_io_bus_cmp(p1
, p2
);
3457 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3458 gpa_t addr
, int len
)
3460 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3466 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3467 kvm_io_bus_sort_cmp
, NULL
);
3472 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3473 gpa_t addr
, int len
)
3475 struct kvm_io_range
*range
, key
;
3478 key
= (struct kvm_io_range
) {
3483 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3484 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3488 off
= range
- bus
->range
;
3490 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3496 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3497 struct kvm_io_range
*range
, const void *val
)
3501 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3505 while (idx
< bus
->dev_count
&&
3506 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3507 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3516 /* kvm_io_bus_write - called under kvm->slots_lock */
3517 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3518 int len
, const void *val
)
3520 struct kvm_io_bus
*bus
;
3521 struct kvm_io_range range
;
3524 range
= (struct kvm_io_range
) {
3529 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3532 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3533 return r
< 0 ? r
: 0;
3536 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3537 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3538 gpa_t addr
, int len
, const void *val
, long cookie
)
3540 struct kvm_io_bus
*bus
;
3541 struct kvm_io_range range
;
3543 range
= (struct kvm_io_range
) {
3548 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3552 /* First try the device referenced by cookie. */
3553 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3554 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3555 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3560 * cookie contained garbage; fall back to search and return the
3561 * correct cookie value.
3563 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3566 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3567 struct kvm_io_range
*range
, void *val
)
3571 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3575 while (idx
< bus
->dev_count
&&
3576 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3577 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3585 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3587 /* kvm_io_bus_read - called under kvm->slots_lock */
3588 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3591 struct kvm_io_bus
*bus
;
3592 struct kvm_io_range range
;
3595 range
= (struct kvm_io_range
) {
3600 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3603 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3604 return r
< 0 ? r
: 0;
3608 /* Caller must hold slots_lock. */
3609 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3610 int len
, struct kvm_io_device
*dev
)
3612 struct kvm_io_bus
*new_bus
, *bus
;
3614 bus
= kvm_get_bus(kvm
, bus_idx
);
3618 /* exclude ioeventfd which is limited by maximum fd */
3619 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3622 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3623 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3626 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3627 sizeof(struct kvm_io_range
)));
3628 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3629 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3630 synchronize_srcu_expedited(&kvm
->srcu
);
3636 /* Caller must hold slots_lock. */
3637 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3638 struct kvm_io_device
*dev
)
3641 struct kvm_io_bus
*new_bus
, *bus
;
3643 bus
= kvm_get_bus(kvm
, bus_idx
);
3647 for (i
= 0; i
< bus
->dev_count
; i
++)
3648 if (bus
->range
[i
].dev
== dev
) {
3652 if (i
== bus
->dev_count
)
3655 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3656 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3658 pr_err("kvm: failed to shrink bus, removing it completely\n");
3662 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3663 new_bus
->dev_count
--;
3664 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3665 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3668 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3669 synchronize_srcu_expedited(&kvm
->srcu
);
3674 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3677 struct kvm_io_bus
*bus
;
3678 int dev_idx
, srcu_idx
;
3679 struct kvm_io_device
*iodev
= NULL
;
3681 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3683 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3687 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3691 iodev
= bus
->range
[dev_idx
].dev
;
3694 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3698 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3700 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3701 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3704 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3707 /* The debugfs files are a reference to the kvm struct which
3708 * is still valid when kvm_destroy_vm is called.
3709 * To avoid the race between open and the removal of the debugfs
3710 * directory we test against the users count.
3712 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3715 if (simple_attr_open(inode
, file
, get
,
3716 stat_data
->mode
& S_IWUGO
? set
: NULL
,
3718 kvm_put_kvm(stat_data
->kvm
);
3725 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3727 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3730 simple_attr_release(inode
, file
);
3731 kvm_put_kvm(stat_data
->kvm
);
3736 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3738 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3740 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3745 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3747 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3752 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3757 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3759 __simple_attr_check_format("%llu\n", 0ull);
3760 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3761 vm_stat_clear_per_vm
, "%llu\n");
3764 static const struct file_operations vm_stat_get_per_vm_fops
= {
3765 .owner
= THIS_MODULE
,
3766 .open
= vm_stat_get_per_vm_open
,
3767 .release
= kvm_debugfs_release
,
3768 .read
= simple_attr_read
,
3769 .write
= simple_attr_write
,
3770 .llseek
= no_llseek
,
3773 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3776 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3777 struct kvm_vcpu
*vcpu
;
3781 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3782 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3787 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3790 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3791 struct kvm_vcpu
*vcpu
;
3796 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3797 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3802 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3804 __simple_attr_check_format("%llu\n", 0ull);
3805 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3806 vcpu_stat_clear_per_vm
, "%llu\n");
3809 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3810 .owner
= THIS_MODULE
,
3811 .open
= vcpu_stat_get_per_vm_open
,
3812 .release
= kvm_debugfs_release
,
3813 .read
= simple_attr_read
,
3814 .write
= simple_attr_write
,
3815 .llseek
= no_llseek
,
3818 static const struct file_operations
*stat_fops_per_vm
[] = {
3819 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3820 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3823 static int vm_stat_get(void *_offset
, u64
*val
)
3825 unsigned offset
= (long)_offset
;
3827 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3831 mutex_lock(&kvm_lock
);
3832 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3834 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3837 mutex_unlock(&kvm_lock
);
3841 static int vm_stat_clear(void *_offset
, u64 val
)
3843 unsigned offset
= (long)_offset
;
3845 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3850 mutex_lock(&kvm_lock
);
3851 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3853 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3855 mutex_unlock(&kvm_lock
);
3860 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3862 static int vcpu_stat_get(void *_offset
, u64
*val
)
3864 unsigned offset
= (long)_offset
;
3866 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3870 mutex_lock(&kvm_lock
);
3871 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3873 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3876 mutex_unlock(&kvm_lock
);
3880 static int vcpu_stat_clear(void *_offset
, u64 val
)
3882 unsigned offset
= (long)_offset
;
3884 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3889 mutex_lock(&kvm_lock
);
3890 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3892 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3894 mutex_unlock(&kvm_lock
);
3899 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3902 static const struct file_operations
*stat_fops
[] = {
3903 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3904 [KVM_STAT_VM
] = &vm_stat_fops
,
3907 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3909 struct kobj_uevent_env
*env
;
3910 unsigned long long created
, active
;
3912 if (!kvm_dev
.this_device
|| !kvm
)
3915 mutex_lock(&kvm_lock
);
3916 if (type
== KVM_EVENT_CREATE_VM
) {
3917 kvm_createvm_count
++;
3919 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3922 created
= kvm_createvm_count
;
3923 active
= kvm_active_vms
;
3924 mutex_unlock(&kvm_lock
);
3926 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3930 add_uevent_var(env
, "CREATED=%llu", created
);
3931 add_uevent_var(env
, "COUNT=%llu", active
);
3933 if (type
== KVM_EVENT_CREATE_VM
) {
3934 add_uevent_var(env
, "EVENT=create");
3935 kvm
->userspace_pid
= task_pid_nr(current
);
3936 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3937 add_uevent_var(env
, "EVENT=destroy");
3939 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3941 if (kvm
->debugfs_dentry
) {
3942 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3945 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3947 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3951 /* no need for checks, since we are adding at most only 5 keys */
3952 env
->envp
[env
->envp_idx
++] = NULL
;
3953 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3957 static int kvm_init_debug(void)
3960 struct kvm_stats_debugfs_item
*p
;
3962 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3963 if (kvm_debugfs_dir
== NULL
)
3966 kvm_debugfs_num_entries
= 0;
3967 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3968 int mode
= p
->mode
? p
->mode
: 0644;
3969 if (!debugfs_create_file(p
->name
, mode
, kvm_debugfs_dir
,
3970 (void *)(long)p
->offset
,
3971 stat_fops
[p
->kind
]))
3978 debugfs_remove_recursive(kvm_debugfs_dir
);
3983 static int kvm_suspend(void)
3985 if (kvm_usage_count
)
3986 hardware_disable_nolock(NULL
);
3990 static void kvm_resume(void)
3992 if (kvm_usage_count
) {
3993 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3994 hardware_enable_nolock(NULL
);
3998 static struct syscore_ops kvm_syscore_ops
= {
3999 .suspend
= kvm_suspend
,
4000 .resume
= kvm_resume
,
4004 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4006 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4009 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4011 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4013 if (vcpu
->preempted
)
4014 vcpu
->preempted
= false;
4016 kvm_arch_sched_in(vcpu
, cpu
);
4018 kvm_arch_vcpu_load(vcpu
, cpu
);
4021 static void kvm_sched_out(struct preempt_notifier
*pn
,
4022 struct task_struct
*next
)
4024 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4026 if (current
->state
== TASK_RUNNING
)
4027 vcpu
->preempted
= true;
4028 kvm_arch_vcpu_put(vcpu
);
4031 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4032 struct module
*module
)
4037 r
= kvm_arch_init(opaque
);
4042 * kvm_arch_init makes sure there's at most one caller
4043 * for architectures that support multiple implementations,
4044 * like intel and amd on x86.
4045 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4046 * conflicts in case kvm is already setup for another implementation.
4048 r
= kvm_irqfd_init();
4052 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4057 r
= kvm_arch_hardware_setup();
4061 for_each_online_cpu(cpu
) {
4062 smp_call_function_single(cpu
,
4063 kvm_arch_check_processor_compat
,
4069 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4070 kvm_starting_cpu
, kvm_dying_cpu
);
4073 register_reboot_notifier(&kvm_reboot_notifier
);
4075 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4077 vcpu_align
= __alignof__(struct kvm_vcpu
);
4078 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4079 SLAB_ACCOUNT
, NULL
);
4080 if (!kvm_vcpu_cache
) {
4085 r
= kvm_async_pf_init();
4089 kvm_chardev_ops
.owner
= module
;
4090 kvm_vm_fops
.owner
= module
;
4091 kvm_vcpu_fops
.owner
= module
;
4093 r
= misc_register(&kvm_dev
);
4095 pr_err("kvm: misc device register failed\n");
4099 register_syscore_ops(&kvm_syscore_ops
);
4101 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4102 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4104 r
= kvm_init_debug();
4106 pr_err("kvm: create debugfs files failed\n");
4110 r
= kvm_vfio_ops_init();
4116 unregister_syscore_ops(&kvm_syscore_ops
);
4117 misc_deregister(&kvm_dev
);
4119 kvm_async_pf_deinit();
4121 kmem_cache_destroy(kvm_vcpu_cache
);
4123 unregister_reboot_notifier(&kvm_reboot_notifier
);
4124 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4127 kvm_arch_hardware_unsetup();
4129 free_cpumask_var(cpus_hardware_enabled
);
4137 EXPORT_SYMBOL_GPL(kvm_init
);
4141 debugfs_remove_recursive(kvm_debugfs_dir
);
4142 misc_deregister(&kvm_dev
);
4143 kmem_cache_destroy(kvm_vcpu_cache
);
4144 kvm_async_pf_deinit();
4145 unregister_syscore_ops(&kvm_syscore_ops
);
4146 unregister_reboot_notifier(&kvm_reboot_notifier
);
4147 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4148 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4149 kvm_arch_hardware_unsetup();
4152 free_cpumask_var(cpus_hardware_enabled
);
4153 kvm_vfio_ops_exit();
4155 EXPORT_SYMBOL_GPL(kvm_exit
);
4157 struct kvm_vm_worker_thread_context
{
4159 struct task_struct
*parent
;
4160 struct completion init_done
;
4161 kvm_vm_thread_fn_t thread_fn
;
4166 static int kvm_vm_worker_thread(void *context
)
4169 * The init_context is allocated on the stack of the parent thread, so
4170 * we have to locally copy anything that is needed beyond initialization
4172 struct kvm_vm_worker_thread_context
*init_context
= context
;
4173 struct kvm
*kvm
= init_context
->kvm
;
4174 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
4175 uintptr_t data
= init_context
->data
;
4178 err
= kthread_park(current
);
4179 /* kthread_park(current) is never supposed to return an error */
4184 err
= cgroup_attach_task_all(init_context
->parent
, current
);
4186 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4191 set_user_nice(current
, task_nice(init_context
->parent
));
4194 init_context
->err
= err
;
4195 complete(&init_context
->init_done
);
4196 init_context
= NULL
;
4201 /* Wait to be woken up by the spawner before proceeding. */
4204 if (!kthread_should_stop())
4205 err
= thread_fn(kvm
, data
);
4210 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
4211 uintptr_t data
, const char *name
,
4212 struct task_struct
**thread_ptr
)
4214 struct kvm_vm_worker_thread_context init_context
= {};
4215 struct task_struct
*thread
;
4218 init_context
.kvm
= kvm
;
4219 init_context
.parent
= current
;
4220 init_context
.thread_fn
= thread_fn
;
4221 init_context
.data
= data
;
4222 init_completion(&init_context
.init_done
);
4224 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
4225 "%s-%d", name
, task_pid_nr(current
));
4227 return PTR_ERR(thread
);
4229 /* kthread_run is never supposed to return NULL */
4230 WARN_ON(thread
== NULL
);
4232 wait_for_completion(&init_context
.init_done
);
4234 if (!init_context
.err
)
4235 *thread_ptr
= thread
;
4237 return init_context
.err
;