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>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
76 module_param(halt_poll_ns
, uint
, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns
);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow
= 2;
81 module_param(halt_poll_ns_grow
, uint
, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow
);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink
;
86 module_param(halt_poll_ns_shrink
, uint
, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_MUTEX(kvm_lock
);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
99 static cpumask_var_t cpus_hardware_enabled
;
100 static int kvm_usage_count
;
101 static atomic_t hardware_enable_failed
;
103 struct kmem_cache
*kvm_vcpu_cache
;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
106 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
108 struct dentry
*kvm_debugfs_dir
;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
111 static int kvm_debugfs_num_entries
;
112 static const struct file_operations
*stat_fops_per_vm
[];
114 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
125 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
127 __visible
bool kvm_rebooting
;
128 EXPORT_SYMBOL_GPL(kvm_rebooting
);
130 static bool largepages_enabled
= true;
132 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
135 static unsigned long long kvm_createvm_count
;
136 static unsigned long long kvm_active_vms
;
138 __weak
void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
139 unsigned long start
, unsigned long end
)
143 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
146 return PageReserved(pfn_to_page(pfn
));
152 * Switches to specified vcpu, until a matching vcpu_put()
154 int vcpu_load(struct kvm_vcpu
*vcpu
)
158 if (mutex_lock_killable(&vcpu
->mutex
))
161 preempt_notifier_register(&vcpu
->preempt_notifier
);
162 kvm_arch_vcpu_load(vcpu
, cpu
);
166 EXPORT_SYMBOL_GPL(vcpu_load
);
168 void vcpu_put(struct kvm_vcpu
*vcpu
)
171 kvm_arch_vcpu_put(vcpu
);
172 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
174 mutex_unlock(&vcpu
->mutex
);
176 EXPORT_SYMBOL_GPL(vcpu_put
);
178 /* TODO: merge with kvm_arch_vcpu_should_kick */
179 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
181 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
184 * We need to wait for the VCPU to reenable interrupts and get out of
185 * READING_SHADOW_PAGE_TABLES mode.
187 if (req
& KVM_REQUEST_WAIT
)
188 return mode
!= OUTSIDE_GUEST_MODE
;
191 * Need to kick a running VCPU, but otherwise there is nothing to do.
193 return mode
== IN_GUEST_MODE
;
196 static void ack_flush(void *_completed
)
200 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
203 cpus
= cpu_online_mask
;
205 if (cpumask_empty(cpus
))
208 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
212 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
217 struct kvm_vcpu
*vcpu
;
219 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
222 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
223 kvm_make_request(req
, vcpu
);
226 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
229 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
230 kvm_request_needs_ipi(vcpu
, req
))
231 __cpumask_set_cpu(cpu
, cpus
);
233 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
235 free_cpumask_var(cpus
);
239 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
240 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
243 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
244 * kvm_make_all_cpus_request.
246 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
249 * We want to publish modifications to the page tables before reading
250 * mode. Pairs with a memory barrier in arch-specific code.
251 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
252 * and smp_mb in walk_shadow_page_lockless_begin/end.
253 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
255 * There is already an smp_mb__after_atomic() before
256 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
259 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
260 ++kvm
->stat
.remote_tlb_flush
;
261 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
263 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
266 void kvm_reload_remote_mmus(struct kvm
*kvm
)
268 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
271 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
276 mutex_init(&vcpu
->mutex
);
281 init_swait_queue_head(&vcpu
->wq
);
282 kvm_async_pf_vcpu_init(vcpu
);
285 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
287 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
292 vcpu
->run
= page_address(page
);
294 kvm_vcpu_set_in_spin_loop(vcpu
, false);
295 kvm_vcpu_set_dy_eligible(vcpu
, false);
296 vcpu
->preempted
= false;
298 r
= kvm_arch_vcpu_init(vcpu
);
304 free_page((unsigned long)vcpu
->run
);
308 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
310 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
313 * no need for rcu_read_lock as VCPU_RUN is the only place that
314 * will change the vcpu->pid pointer and on uninit all file
315 * descriptors are already gone.
317 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
318 kvm_arch_vcpu_uninit(vcpu
);
319 free_page((unsigned long)vcpu
->run
);
321 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
323 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
324 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
326 return container_of(mn
, struct kvm
, mmu_notifier
);
329 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
330 struct mm_struct
*mm
,
331 unsigned long address
,
334 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
337 idx
= srcu_read_lock(&kvm
->srcu
);
338 spin_lock(&kvm
->mmu_lock
);
339 kvm
->mmu_notifier_seq
++;
340 kvm_set_spte_hva(kvm
, address
, pte
);
341 spin_unlock(&kvm
->mmu_lock
);
342 srcu_read_unlock(&kvm
->srcu
, idx
);
345 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
346 struct mm_struct
*mm
,
350 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
351 int need_tlb_flush
= 0, idx
;
353 idx
= srcu_read_lock(&kvm
->srcu
);
354 spin_lock(&kvm
->mmu_lock
);
356 * The count increase must become visible at unlock time as no
357 * spte can be established without taking the mmu_lock and
358 * count is also read inside the mmu_lock critical section.
360 kvm
->mmu_notifier_count
++;
361 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
362 need_tlb_flush
|= kvm
->tlbs_dirty
;
363 /* we've to flush the tlb before the pages can be freed */
365 kvm_flush_remote_tlbs(kvm
);
367 spin_unlock(&kvm
->mmu_lock
);
369 kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
);
371 srcu_read_unlock(&kvm
->srcu
, idx
);
374 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
375 struct mm_struct
*mm
,
379 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
381 spin_lock(&kvm
->mmu_lock
);
383 * This sequence increase will notify the kvm page fault that
384 * the page that is going to be mapped in the spte could have
387 kvm
->mmu_notifier_seq
++;
390 * The above sequence increase must be visible before the
391 * below count decrease, which is ensured by the smp_wmb above
392 * in conjunction with the smp_rmb in mmu_notifier_retry().
394 kvm
->mmu_notifier_count
--;
395 spin_unlock(&kvm
->mmu_lock
);
397 BUG_ON(kvm
->mmu_notifier_count
< 0);
400 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
401 struct mm_struct
*mm
,
405 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
408 idx
= srcu_read_lock(&kvm
->srcu
);
409 spin_lock(&kvm
->mmu_lock
);
411 young
= kvm_age_hva(kvm
, start
, end
);
413 kvm_flush_remote_tlbs(kvm
);
415 spin_unlock(&kvm
->mmu_lock
);
416 srcu_read_unlock(&kvm
->srcu
, idx
);
421 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
422 struct mm_struct
*mm
,
426 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
429 idx
= srcu_read_lock(&kvm
->srcu
);
430 spin_lock(&kvm
->mmu_lock
);
432 * Even though we do not flush TLB, this will still adversely
433 * affect performance on pre-Haswell Intel EPT, where there is
434 * no EPT Access Bit to clear so that we have to tear down EPT
435 * tables instead. If we find this unacceptable, we can always
436 * add a parameter to kvm_age_hva so that it effectively doesn't
437 * do anything on clear_young.
439 * Also note that currently we never issue secondary TLB flushes
440 * from clear_young, leaving this job up to the regular system
441 * cadence. If we find this inaccurate, we might come up with a
442 * more sophisticated heuristic later.
444 young
= kvm_age_hva(kvm
, start
, end
);
445 spin_unlock(&kvm
->mmu_lock
);
446 srcu_read_unlock(&kvm
->srcu
, idx
);
451 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
452 struct mm_struct
*mm
,
453 unsigned long address
)
455 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
458 idx
= srcu_read_lock(&kvm
->srcu
);
459 spin_lock(&kvm
->mmu_lock
);
460 young
= kvm_test_age_hva(kvm
, address
);
461 spin_unlock(&kvm
->mmu_lock
);
462 srcu_read_unlock(&kvm
->srcu
, idx
);
467 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
468 struct mm_struct
*mm
)
470 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
473 idx
= srcu_read_lock(&kvm
->srcu
);
474 kvm_arch_flush_shadow_all(kvm
);
475 srcu_read_unlock(&kvm
->srcu
, idx
);
478 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
479 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
480 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
481 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
482 .clear_young
= kvm_mmu_notifier_clear_young
,
483 .test_young
= kvm_mmu_notifier_test_young
,
484 .change_pte
= kvm_mmu_notifier_change_pte
,
485 .release
= kvm_mmu_notifier_release
,
488 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
490 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
491 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
494 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
496 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
501 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
503 static struct kvm_memslots
*kvm_alloc_memslots(void)
506 struct kvm_memslots
*slots
;
508 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
512 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
513 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
520 if (!memslot
->dirty_bitmap
)
523 kvfree(memslot
->dirty_bitmap
);
524 memslot
->dirty_bitmap
= NULL
;
528 * Free any memory in @free but not in @dont.
530 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
531 struct kvm_memory_slot
*dont
)
533 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
534 kvm_destroy_dirty_bitmap(free
);
536 kvm_arch_free_memslot(kvm
, free
, dont
);
541 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
543 struct kvm_memory_slot
*memslot
;
548 kvm_for_each_memslot(memslot
, slots
)
549 kvm_free_memslot(kvm
, memslot
, NULL
);
554 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
558 if (!kvm
->debugfs_dentry
)
561 debugfs_remove_recursive(kvm
->debugfs_dentry
);
563 if (kvm
->debugfs_stat_data
) {
564 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
565 kfree(kvm
->debugfs_stat_data
[i
]);
566 kfree(kvm
->debugfs_stat_data
);
570 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
572 char dir_name
[ITOA_MAX_LEN
* 2];
573 struct kvm_stat_data
*stat_data
;
574 struct kvm_stats_debugfs_item
*p
;
576 if (!debugfs_initialized())
579 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
580 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
582 if (!kvm
->debugfs_dentry
)
585 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
586 sizeof(*kvm
->debugfs_stat_data
),
588 if (!kvm
->debugfs_stat_data
)
591 for (p
= debugfs_entries
; p
->name
; p
++) {
592 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
596 stat_data
->kvm
= kvm
;
597 stat_data
->offset
= p
->offset
;
598 stat_data
->mode
= p
->mode
? p
->mode
: 0644;
599 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
600 if (!debugfs_create_file(p
->name
, stat_data
->mode
,
603 stat_fops_per_vm
[p
->kind
]))
609 static struct kvm
*kvm_create_vm(unsigned long type
)
612 struct kvm
*kvm
= kvm_arch_alloc_vm();
615 return ERR_PTR(-ENOMEM
);
617 spin_lock_init(&kvm
->mmu_lock
);
619 kvm
->mm
= current
->mm
;
620 kvm_eventfd_init(kvm
);
621 mutex_init(&kvm
->lock
);
622 mutex_init(&kvm
->irq_lock
);
623 mutex_init(&kvm
->slots_lock
);
624 refcount_set(&kvm
->users_count
, 1);
625 INIT_LIST_HEAD(&kvm
->devices
);
627 r
= kvm_arch_init_vm(kvm
, type
);
629 goto out_err_no_disable
;
631 r
= hardware_enable_all();
633 goto out_err_no_disable
;
635 #ifdef CONFIG_HAVE_KVM_IRQFD
636 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
639 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
642 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
643 struct kvm_memslots
*slots
= kvm_alloc_memslots();
645 goto out_err_no_srcu
;
647 * Generations must be different for each address space.
648 * Init kvm generation close to the maximum to easily test the
649 * code of handling generation number wrap-around.
651 slots
->generation
= i
* 2 - 150;
652 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
655 if (init_srcu_struct(&kvm
->srcu
))
656 goto out_err_no_srcu
;
657 if (init_srcu_struct(&kvm
->irq_srcu
))
658 goto out_err_no_irq_srcu
;
659 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
660 rcu_assign_pointer(kvm
->buses
[i
],
661 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
666 r
= kvm_init_mmu_notifier(kvm
);
670 mutex_lock(&kvm_lock
);
671 list_add(&kvm
->vm_list
, &vm_list
);
672 mutex_unlock(&kvm_lock
);
674 preempt_notifier_inc();
679 cleanup_srcu_struct(&kvm
->irq_srcu
);
681 cleanup_srcu_struct(&kvm
->srcu
);
683 hardware_disable_all();
685 refcount_set(&kvm
->users_count
, 0);
686 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
687 kfree(kvm_get_bus(kvm
, i
));
688 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
689 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
690 kvm_arch_free_vm(kvm
);
695 static void kvm_destroy_devices(struct kvm
*kvm
)
697 struct kvm_device
*dev
, *tmp
;
700 * We do not need to take the kvm->lock here, because nobody else
701 * has a reference to the struct kvm at this point and therefore
702 * cannot access the devices list anyhow.
704 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
705 list_del(&dev
->vm_node
);
706 dev
->ops
->destroy(dev
);
710 static void kvm_destroy_vm(struct kvm
*kvm
)
713 struct mm_struct
*mm
= kvm
->mm
;
715 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
716 kvm_destroy_vm_debugfs(kvm
);
717 kvm_arch_sync_events(kvm
);
718 mutex_lock(&kvm_lock
);
719 list_del(&kvm
->vm_list
);
720 mutex_unlock(&kvm_lock
);
721 kvm_free_irq_routing(kvm
);
722 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
723 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
726 kvm_io_bus_destroy(bus
);
727 kvm
->buses
[i
] = NULL
;
729 kvm_coalesced_mmio_free(kvm
);
730 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
731 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
733 kvm_arch_flush_shadow_all(kvm
);
735 kvm_arch_destroy_vm(kvm
);
736 kvm_destroy_devices(kvm
);
737 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
738 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
739 cleanup_srcu_struct(&kvm
->irq_srcu
);
740 cleanup_srcu_struct(&kvm
->srcu
);
741 kvm_arch_free_vm(kvm
);
742 preempt_notifier_dec();
743 hardware_disable_all();
747 void kvm_get_kvm(struct kvm
*kvm
)
749 refcount_inc(&kvm
->users_count
);
751 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
753 void kvm_put_kvm(struct kvm
*kvm
)
755 if (refcount_dec_and_test(&kvm
->users_count
))
758 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
761 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
763 struct kvm
*kvm
= filp
->private_data
;
765 kvm_irqfd_release(kvm
);
772 * Allocation size is twice as large as the actual dirty bitmap size.
773 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
775 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
777 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
779 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
780 if (!memslot
->dirty_bitmap
)
787 * Insert memslot and re-sort memslots based on their GFN,
788 * so binary search could be used to lookup GFN.
789 * Sorting algorithm takes advantage of having initially
790 * sorted array and known changed memslot position.
792 static void update_memslots(struct kvm_memslots
*slots
,
793 struct kvm_memory_slot
*new)
796 int i
= slots
->id_to_index
[id
];
797 struct kvm_memory_slot
*mslots
= slots
->memslots
;
799 WARN_ON(mslots
[i
].id
!= id
);
801 WARN_ON(!mslots
[i
].npages
);
802 if (mslots
[i
].npages
)
805 if (!mslots
[i
].npages
)
809 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
810 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
811 if (!mslots
[i
+ 1].npages
)
813 mslots
[i
] = mslots
[i
+ 1];
814 slots
->id_to_index
[mslots
[i
].id
] = i
;
819 * The ">=" is needed when creating a slot with base_gfn == 0,
820 * so that it moves before all those with base_gfn == npages == 0.
822 * On the other hand, if new->npages is zero, the above loop has
823 * already left i pointing to the beginning of the empty part of
824 * mslots, and the ">=" would move the hole backwards in this
825 * case---which is wrong. So skip the loop when deleting a slot.
829 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
830 mslots
[i
] = mslots
[i
- 1];
831 slots
->id_to_index
[mslots
[i
].id
] = i
;
835 WARN_ON_ONCE(i
!= slots
->used_slots
);
838 slots
->id_to_index
[mslots
[i
].id
] = i
;
841 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
843 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
845 #ifdef __KVM_HAVE_READONLY_MEM
846 valid_flags
|= KVM_MEM_READONLY
;
849 if (mem
->flags
& ~valid_flags
)
855 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
856 int as_id
, struct kvm_memslots
*slots
)
858 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
862 * Set the low bit in the generation, which disables SPTE caching
863 * until the end of synchronize_srcu_expedited.
865 WARN_ON(old_memslots
->generation
& 1);
866 slots
->generation
= old_memslots
->generation
+ 1;
868 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
869 synchronize_srcu_expedited(&kvm
->srcu
);
872 * Increment the new memslot generation a second time. This prevents
873 * vm exits that race with memslot updates from caching a memslot
874 * generation that will (potentially) be valid forever.
876 * Generations must be unique even across address spaces. We do not need
877 * a global counter for that, instead the generation space is evenly split
878 * across address spaces. For example, with two address spaces, address
879 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
880 * use generations 2, 6, 10, 14, ...
882 gen
= slots
->generation
+ KVM_ADDRESS_SPACE_NUM
* 2 - 1;
884 kvm_arch_memslots_updated(kvm
, gen
);
886 slots
->generation
= gen
;
892 * Allocate some memory and give it an address in the guest physical address
895 * Discontiguous memory is allowed, mostly for framebuffers.
897 * Must be called holding kvm->slots_lock for write.
899 int __kvm_set_memory_region(struct kvm
*kvm
,
900 const struct kvm_userspace_memory_region
*mem
)
904 unsigned long npages
;
905 struct kvm_memory_slot
*slot
;
906 struct kvm_memory_slot old
, new;
907 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
909 enum kvm_mr_change change
;
911 r
= check_memory_region_flags(mem
);
916 as_id
= mem
->slot
>> 16;
919 /* General sanity checks */
920 if (mem
->memory_size
& (PAGE_SIZE
- 1))
922 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
924 /* We can read the guest memory with __xxx_user() later on. */
925 if ((id
< KVM_USER_MEM_SLOTS
) &&
926 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
927 !access_ok(VERIFY_WRITE
,
928 (void __user
*)(unsigned long)mem
->userspace_addr
,
931 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
933 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
936 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
937 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
938 npages
= mem
->memory_size
>> PAGE_SHIFT
;
940 if (npages
> KVM_MEM_MAX_NR_PAGES
)
946 new.base_gfn
= base_gfn
;
948 new.flags
= mem
->flags
;
952 change
= KVM_MR_CREATE
;
953 else { /* Modify an existing slot. */
954 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
955 (npages
!= old
.npages
) ||
956 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
959 if (base_gfn
!= old
.base_gfn
)
960 change
= KVM_MR_MOVE
;
961 else if (new.flags
!= old
.flags
)
962 change
= KVM_MR_FLAGS_ONLY
;
963 else { /* Nothing to change. */
972 change
= KVM_MR_DELETE
;
977 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
978 /* Check for overlaps */
980 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
983 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
984 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
989 /* Free page dirty bitmap if unneeded */
990 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
991 new.dirty_bitmap
= NULL
;
994 if (change
== KVM_MR_CREATE
) {
995 new.userspace_addr
= mem
->userspace_addr
;
997 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1001 /* Allocate page dirty bitmap if needed */
1002 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1003 if (kvm_create_dirty_bitmap(&new) < 0)
1007 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1010 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1012 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1013 slot
= id_to_memslot(slots
, id
);
1014 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1016 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1018 /* From this point no new shadow pages pointing to a deleted,
1019 * or moved, memslot will be created.
1021 * validation of sp->gfn happens in:
1022 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1023 * - kvm_is_visible_gfn (mmu_check_roots)
1025 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1028 * We can re-use the old_memslots from above, the only difference
1029 * from the currently installed memslots is the invalid flag. This
1030 * will get overwritten by update_memslots anyway.
1032 slots
= old_memslots
;
1035 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1039 /* actual memory is freed via old in kvm_free_memslot below */
1040 if (change
== KVM_MR_DELETE
) {
1041 new.dirty_bitmap
= NULL
;
1042 memset(&new.arch
, 0, sizeof(new.arch
));
1045 update_memslots(slots
, &new);
1046 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1048 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1050 kvm_free_memslot(kvm
, &old
, &new);
1051 kvfree(old_memslots
);
1057 kvm_free_memslot(kvm
, &new, &old
);
1061 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1063 int kvm_set_memory_region(struct kvm
*kvm
,
1064 const struct kvm_userspace_memory_region
*mem
)
1068 mutex_lock(&kvm
->slots_lock
);
1069 r
= __kvm_set_memory_region(kvm
, mem
);
1070 mutex_unlock(&kvm
->slots_lock
);
1073 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1075 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1076 struct kvm_userspace_memory_region
*mem
)
1078 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1081 return kvm_set_memory_region(kvm
, mem
);
1084 int kvm_get_dirty_log(struct kvm
*kvm
,
1085 struct kvm_dirty_log
*log
, int *is_dirty
)
1087 struct kvm_memslots
*slots
;
1088 struct kvm_memory_slot
*memslot
;
1091 unsigned long any
= 0;
1093 as_id
= log
->slot
>> 16;
1094 id
= (u16
)log
->slot
;
1095 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1098 slots
= __kvm_memslots(kvm
, as_id
);
1099 memslot
= id_to_memslot(slots
, id
);
1100 if (!memslot
->dirty_bitmap
)
1103 n
= kvm_dirty_bitmap_bytes(memslot
);
1105 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1106 any
= memslot
->dirty_bitmap
[i
];
1108 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1115 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1117 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1119 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1120 * are dirty write protect them for next write.
1121 * @kvm: pointer to kvm instance
1122 * @log: slot id and address to which we copy the log
1123 * @is_dirty: flag set if any page is dirty
1125 * We need to keep it in mind that VCPU threads can write to the bitmap
1126 * concurrently. So, to avoid losing track of dirty pages we keep the
1129 * 1. Take a snapshot of the bit and clear it if needed.
1130 * 2. Write protect the corresponding page.
1131 * 3. Copy the snapshot to the userspace.
1132 * 4. Upon return caller flushes TLB's if needed.
1134 * Between 2 and 4, the guest may write to the page using the remaining TLB
1135 * entry. This is not a problem because the page is reported dirty using
1136 * the snapshot taken before and step 4 ensures that writes done after
1137 * exiting to userspace will be logged for the next call.
1140 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1141 struct kvm_dirty_log
*log
, bool *is_dirty
)
1143 struct kvm_memslots
*slots
;
1144 struct kvm_memory_slot
*memslot
;
1147 unsigned long *dirty_bitmap
;
1148 unsigned long *dirty_bitmap_buffer
;
1150 as_id
= log
->slot
>> 16;
1151 id
= (u16
)log
->slot
;
1152 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1155 slots
= __kvm_memslots(kvm
, as_id
);
1156 memslot
= id_to_memslot(slots
, id
);
1158 dirty_bitmap
= memslot
->dirty_bitmap
;
1162 n
= kvm_dirty_bitmap_bytes(memslot
);
1164 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1165 memset(dirty_bitmap_buffer
, 0, n
);
1167 spin_lock(&kvm
->mmu_lock
);
1169 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1173 if (!dirty_bitmap
[i
])
1178 mask
= xchg(&dirty_bitmap
[i
], 0);
1179 dirty_bitmap_buffer
[i
] = mask
;
1182 offset
= i
* BITS_PER_LONG
;
1183 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1188 spin_unlock(&kvm
->mmu_lock
);
1189 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1193 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1196 bool kvm_largepages_enabled(void)
1198 return largepages_enabled
;
1201 void kvm_disable_largepages(void)
1203 largepages_enabled
= false;
1205 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1207 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1209 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1211 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1213 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1215 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1218 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1220 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1222 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1223 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1228 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1230 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1232 struct vm_area_struct
*vma
;
1233 unsigned long addr
, size
;
1237 addr
= gfn_to_hva(kvm
, gfn
);
1238 if (kvm_is_error_hva(addr
))
1241 down_read(¤t
->mm
->mmap_sem
);
1242 vma
= find_vma(current
->mm
, addr
);
1246 size
= vma_kernel_pagesize(vma
);
1249 up_read(¤t
->mm
->mmap_sem
);
1254 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1256 return slot
->flags
& KVM_MEM_READONLY
;
1259 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1260 gfn_t
*nr_pages
, bool write
)
1262 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1263 return KVM_HVA_ERR_BAD
;
1265 if (memslot_is_readonly(slot
) && write
)
1266 return KVM_HVA_ERR_RO_BAD
;
1269 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1271 return __gfn_to_hva_memslot(slot
, gfn
);
1274 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1277 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1280 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1283 return gfn_to_hva_many(slot
, gfn
, NULL
);
1285 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1287 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1289 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1291 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1293 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1295 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1297 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1300 * If writable is set to false, the hva returned by this function is only
1301 * allowed to be read.
1303 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1304 gfn_t gfn
, bool *writable
)
1306 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1308 if (!kvm_is_error_hva(hva
) && writable
)
1309 *writable
= !memslot_is_readonly(slot
);
1314 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1316 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1318 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1321 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1323 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1325 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1328 static int get_user_page_nowait(unsigned long start
, int write
,
1331 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1334 flags
|= FOLL_WRITE
;
1336 return get_user_pages(start
, 1, flags
, page
, NULL
);
1339 static inline int check_user_page_hwpoison(unsigned long addr
)
1341 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1343 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1344 return rc
== -EHWPOISON
;
1348 * The atomic path to get the writable pfn which will be stored in @pfn,
1349 * true indicates success, otherwise false is returned.
1351 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1352 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1354 struct page
*page
[1];
1357 if (!(async
|| atomic
))
1361 * Fast pin a writable pfn only if it is a write fault request
1362 * or the caller allows to map a writable pfn for a read fault
1365 if (!(write_fault
|| writable
))
1368 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1370 *pfn
= page_to_pfn(page
[0]);
1381 * The slow path to get the pfn of the specified host virtual address,
1382 * 1 indicates success, -errno is returned if error is detected.
1384 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1385 bool *writable
, kvm_pfn_t
*pfn
)
1387 struct page
*page
[1];
1393 *writable
= write_fault
;
1396 down_read(¤t
->mm
->mmap_sem
);
1397 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1398 up_read(¤t
->mm
->mmap_sem
);
1400 unsigned int flags
= FOLL_HWPOISON
;
1403 flags
|= FOLL_WRITE
;
1405 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1410 /* map read fault as writable if possible */
1411 if (unlikely(!write_fault
) && writable
) {
1412 struct page
*wpage
[1];
1414 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1423 *pfn
= page_to_pfn(page
[0]);
1427 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1429 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1432 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1438 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1439 unsigned long addr
, bool *async
,
1440 bool write_fault
, bool *writable
,
1446 r
= follow_pfn(vma
, addr
, &pfn
);
1449 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1450 * not call the fault handler, so do it here.
1452 bool unlocked
= false;
1453 r
= fixup_user_fault(current
, current
->mm
, addr
,
1454 (write_fault
? FAULT_FLAG_WRITE
: 0),
1461 r
= follow_pfn(vma
, addr
, &pfn
);
1471 * Get a reference here because callers of *hva_to_pfn* and
1472 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1473 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1474 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1475 * simply do nothing for reserved pfns.
1477 * Whoever called remap_pfn_range is also going to call e.g.
1478 * unmap_mapping_range before the underlying pages are freed,
1479 * causing a call to our MMU notifier.
1488 * Pin guest page in memory and return its pfn.
1489 * @addr: host virtual address which maps memory to the guest
1490 * @atomic: whether this function can sleep
1491 * @async: whether this function need to wait IO complete if the
1492 * host page is not in the memory
1493 * @write_fault: whether we should get a writable host page
1494 * @writable: whether it allows to map a writable host page for !@write_fault
1496 * The function will map a writable host page for these two cases:
1497 * 1): @write_fault = true
1498 * 2): @write_fault = false && @writable, @writable will tell the caller
1499 * whether the mapping is writable.
1501 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1502 bool write_fault
, bool *writable
)
1504 struct vm_area_struct
*vma
;
1508 /* we can do it either atomically or asynchronously, not both */
1509 BUG_ON(atomic
&& async
);
1511 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1515 return KVM_PFN_ERR_FAULT
;
1517 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1521 down_read(¤t
->mm
->mmap_sem
);
1522 if (npages
== -EHWPOISON
||
1523 (!async
&& check_user_page_hwpoison(addr
))) {
1524 pfn
= KVM_PFN_ERR_HWPOISON
;
1529 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1532 pfn
= KVM_PFN_ERR_FAULT
;
1533 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1534 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1538 pfn
= KVM_PFN_ERR_FAULT
;
1540 if (async
&& vma_is_valid(vma
, write_fault
))
1542 pfn
= KVM_PFN_ERR_FAULT
;
1545 up_read(¤t
->mm
->mmap_sem
);
1549 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1550 bool atomic
, bool *async
, bool write_fault
,
1553 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1555 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1558 return KVM_PFN_ERR_RO_FAULT
;
1561 if (kvm_is_error_hva(addr
)) {
1564 return KVM_PFN_NOSLOT
;
1567 /* Do not map writable pfn in the readonly memslot. */
1568 if (writable
&& memslot_is_readonly(slot
)) {
1573 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1576 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1578 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1581 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1582 write_fault
, writable
);
1584 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1586 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1588 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1592 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1594 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1596 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1598 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1600 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1602 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1604 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1606 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1608 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1610 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1612 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1614 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1616 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1618 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1620 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1622 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1623 struct page
**pages
, int nr_pages
)
1628 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1629 if (kvm_is_error_hva(addr
))
1632 if (entry
< nr_pages
)
1635 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1637 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1639 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1641 if (is_error_noslot_pfn(pfn
))
1642 return KVM_ERR_PTR_BAD_PAGE
;
1644 if (kvm_is_reserved_pfn(pfn
)) {
1646 return KVM_ERR_PTR_BAD_PAGE
;
1649 return pfn_to_page(pfn
);
1652 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1656 pfn
= gfn_to_pfn(kvm
, gfn
);
1658 return kvm_pfn_to_page(pfn
);
1660 EXPORT_SYMBOL_GPL(gfn_to_page
);
1662 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1666 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1668 return kvm_pfn_to_page(pfn
);
1670 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1672 void kvm_release_page_clean(struct page
*page
)
1674 WARN_ON(is_error_page(page
));
1676 kvm_release_pfn_clean(page_to_pfn(page
));
1678 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1680 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1682 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1683 put_page(pfn_to_page(pfn
));
1685 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1687 void kvm_release_page_dirty(struct page
*page
)
1689 WARN_ON(is_error_page(page
));
1691 kvm_release_pfn_dirty(page_to_pfn(page
));
1693 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1695 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1697 kvm_set_pfn_dirty(pfn
);
1698 kvm_release_pfn_clean(pfn
);
1700 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1702 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1704 if (!kvm_is_reserved_pfn(pfn
)) {
1705 struct page
*page
= pfn_to_page(pfn
);
1707 if (!PageReserved(page
))
1711 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1713 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1715 if (!kvm_is_reserved_pfn(pfn
))
1716 mark_page_accessed(pfn_to_page(pfn
));
1718 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1720 void kvm_get_pfn(kvm_pfn_t pfn
)
1722 if (!kvm_is_reserved_pfn(pfn
))
1723 get_page(pfn_to_page(pfn
));
1725 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1727 static int next_segment(unsigned long len
, int offset
)
1729 if (len
> PAGE_SIZE
- offset
)
1730 return PAGE_SIZE
- offset
;
1735 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1736 void *data
, int offset
, int len
)
1741 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1742 if (kvm_is_error_hva(addr
))
1744 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1750 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1753 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1755 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1757 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1759 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1760 int offset
, int len
)
1762 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1764 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1766 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1768 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1770 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1772 int offset
= offset_in_page(gpa
);
1775 while ((seg
= next_segment(len
, offset
)) != 0) {
1776 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1786 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1788 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1790 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1792 int offset
= offset_in_page(gpa
);
1795 while ((seg
= next_segment(len
, offset
)) != 0) {
1796 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1806 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1808 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1809 void *data
, int offset
, unsigned long len
)
1814 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1815 if (kvm_is_error_hva(addr
))
1817 pagefault_disable();
1818 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1825 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1828 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1829 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1830 int offset
= offset_in_page(gpa
);
1832 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1834 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1836 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1837 void *data
, unsigned long len
)
1839 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1840 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1841 int offset
= offset_in_page(gpa
);
1843 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1845 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1847 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1848 const void *data
, int offset
, int len
)
1853 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1854 if (kvm_is_error_hva(addr
))
1856 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1859 mark_page_dirty_in_slot(memslot
, gfn
);
1863 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1864 const void *data
, int offset
, int len
)
1866 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1868 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1870 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1872 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1873 const void *data
, int offset
, int len
)
1875 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1877 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1879 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1881 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1884 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1886 int offset
= offset_in_page(gpa
);
1889 while ((seg
= next_segment(len
, offset
)) != 0) {
1890 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1900 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1902 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1905 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1907 int offset
= offset_in_page(gpa
);
1910 while ((seg
= next_segment(len
, offset
)) != 0) {
1911 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1921 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1923 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1924 struct gfn_to_hva_cache
*ghc
,
1925 gpa_t gpa
, unsigned long len
)
1927 int offset
= offset_in_page(gpa
);
1928 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1929 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1930 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1931 gfn_t nr_pages_avail
;
1934 ghc
->generation
= slots
->generation
;
1936 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1937 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1938 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1942 * If the requested region crosses two memslots, we still
1943 * verify that the entire region is valid here.
1945 while (start_gfn
<= end_gfn
) {
1947 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1948 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1950 if (kvm_is_error_hva(ghc
->hva
))
1952 start_gfn
+= nr_pages_avail
;
1954 /* Use the slow path for cross page reads and writes. */
1955 ghc
->memslot
= NULL
;
1960 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1961 gpa_t gpa
, unsigned long len
)
1963 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1964 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1966 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1968 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1969 void *data
, unsigned int offset
,
1972 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1974 gpa_t gpa
= ghc
->gpa
+ offset
;
1976 BUG_ON(len
+ offset
> ghc
->len
);
1978 if (slots
->generation
!= ghc
->generation
)
1979 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
1981 if (unlikely(!ghc
->memslot
))
1982 return kvm_write_guest(kvm
, gpa
, data
, len
);
1984 if (kvm_is_error_hva(ghc
->hva
))
1987 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
1990 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
1994 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
1996 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1997 void *data
, unsigned long len
)
1999 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2001 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2003 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2004 void *data
, unsigned long len
)
2006 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2009 BUG_ON(len
> ghc
->len
);
2011 if (slots
->generation
!= ghc
->generation
)
2012 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2014 if (unlikely(!ghc
->memslot
))
2015 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2017 if (kvm_is_error_hva(ghc
->hva
))
2020 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2026 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2028 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2030 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2032 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2034 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2036 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2038 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2040 int offset
= offset_in_page(gpa
);
2043 while ((seg
= next_segment(len
, offset
)) != 0) {
2044 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2053 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2055 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2058 if (memslot
&& memslot
->dirty_bitmap
) {
2059 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2061 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2065 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2067 struct kvm_memory_slot
*memslot
;
2069 memslot
= gfn_to_memslot(kvm
, gfn
);
2070 mark_page_dirty_in_slot(memslot
, gfn
);
2072 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2074 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2076 struct kvm_memory_slot
*memslot
;
2078 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2079 mark_page_dirty_in_slot(memslot
, gfn
);
2081 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2083 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2085 if (!vcpu
->sigset_active
)
2089 * This does a lockless modification of ->real_blocked, which is fine
2090 * because, only current can change ->real_blocked and all readers of
2091 * ->real_blocked don't care as long ->real_blocked is always a subset
2094 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2097 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2099 if (!vcpu
->sigset_active
)
2102 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2103 sigemptyset(¤t
->real_blocked
);
2106 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2108 unsigned int old
, val
, grow
;
2110 old
= val
= vcpu
->halt_poll_ns
;
2111 grow
= READ_ONCE(halt_poll_ns_grow
);
2113 if (val
== 0 && grow
)
2118 if (val
> halt_poll_ns
)
2121 vcpu
->halt_poll_ns
= val
;
2122 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2125 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2127 unsigned int old
, val
, shrink
;
2129 old
= val
= vcpu
->halt_poll_ns
;
2130 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2136 vcpu
->halt_poll_ns
= val
;
2137 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2140 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2142 if (kvm_arch_vcpu_runnable(vcpu
)) {
2143 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2146 if (kvm_cpu_has_pending_timer(vcpu
))
2148 if (signal_pending(current
))
2155 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2157 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2160 DECLARE_SWAITQUEUE(wait
);
2161 bool waited
= false;
2164 start
= cur
= ktime_get();
2165 if (vcpu
->halt_poll_ns
) {
2166 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2168 ++vcpu
->stat
.halt_attempted_poll
;
2171 * This sets KVM_REQ_UNHALT if an interrupt
2174 if (kvm_vcpu_check_block(vcpu
) < 0) {
2175 ++vcpu
->stat
.halt_successful_poll
;
2176 if (!vcpu_valid_wakeup(vcpu
))
2177 ++vcpu
->stat
.halt_poll_invalid
;
2181 } while (single_task_running() && ktime_before(cur
, stop
));
2184 kvm_arch_vcpu_blocking(vcpu
);
2187 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2189 if (kvm_vcpu_check_block(vcpu
) < 0)
2196 finish_swait(&vcpu
->wq
, &wait
);
2199 kvm_arch_vcpu_unblocking(vcpu
);
2201 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2203 if (!vcpu_valid_wakeup(vcpu
))
2204 shrink_halt_poll_ns(vcpu
);
2205 else if (halt_poll_ns
) {
2206 if (block_ns
<= vcpu
->halt_poll_ns
)
2208 /* we had a long block, shrink polling */
2209 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2210 shrink_halt_poll_ns(vcpu
);
2211 /* we had a short halt and our poll time is too small */
2212 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2213 block_ns
< halt_poll_ns
)
2214 grow_halt_poll_ns(vcpu
);
2216 vcpu
->halt_poll_ns
= 0;
2218 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2219 kvm_arch_vcpu_block_finish(vcpu
);
2221 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2223 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2225 struct swait_queue_head
*wqp
;
2227 wqp
= kvm_arch_vcpu_wq(vcpu
);
2228 if (swq_has_sleeper(wqp
)) {
2230 ++vcpu
->stat
.halt_wakeup
;
2236 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2240 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2242 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2245 int cpu
= vcpu
->cpu
;
2247 if (kvm_vcpu_wake_up(vcpu
))
2251 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2252 if (kvm_arch_vcpu_should_kick(vcpu
))
2253 smp_send_reschedule(cpu
);
2256 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2257 #endif /* !CONFIG_S390 */
2259 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2262 struct task_struct
*task
= NULL
;
2266 pid
= rcu_dereference(target
->pid
);
2268 task
= get_pid_task(pid
, PIDTYPE_PID
);
2272 ret
= yield_to(task
, 1);
2273 put_task_struct(task
);
2277 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2280 * Helper that checks whether a VCPU is eligible for directed yield.
2281 * Most eligible candidate to yield is decided by following heuristics:
2283 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2284 * (preempted lock holder), indicated by @in_spin_loop.
2285 * Set at the beiginning and cleared at the end of interception/PLE handler.
2287 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2288 * chance last time (mostly it has become eligible now since we have probably
2289 * yielded to lockholder in last iteration. This is done by toggling
2290 * @dy_eligible each time a VCPU checked for eligibility.)
2292 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2293 * to preempted lock-holder could result in wrong VCPU selection and CPU
2294 * burning. Giving priority for a potential lock-holder increases lock
2297 * Since algorithm is based on heuristics, accessing another VCPU data without
2298 * locking does not harm. It may result in trying to yield to same VCPU, fail
2299 * and continue with next VCPU and so on.
2301 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2303 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2306 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2307 vcpu
->spin_loop
.dy_eligible
;
2309 if (vcpu
->spin_loop
.in_spin_loop
)
2310 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2319 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2320 * a vcpu_load/vcpu_put pair. However, for most architectures
2321 * kvm_arch_vcpu_runnable does not require vcpu_load.
2323 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
2325 return kvm_arch_vcpu_runnable(vcpu
);
2328 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
2330 if (kvm_arch_dy_runnable(vcpu
))
2333 #ifdef CONFIG_KVM_ASYNC_PF
2334 if (!list_empty_careful(&vcpu
->async_pf
.done
))
2341 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2343 struct kvm
*kvm
= me
->kvm
;
2344 struct kvm_vcpu
*vcpu
;
2345 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2351 kvm_vcpu_set_in_spin_loop(me
, true);
2353 * We boost the priority of a VCPU that is runnable but not
2354 * currently running, because it got preempted by something
2355 * else and called schedule in __vcpu_run. Hopefully that
2356 * VCPU is holding the lock that we need and will release it.
2357 * We approximate round-robin by starting at the last boosted VCPU.
2359 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2360 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2361 if (!pass
&& i
<= last_boosted_vcpu
) {
2362 i
= last_boosted_vcpu
;
2364 } else if (pass
&& i
> last_boosted_vcpu
)
2366 if (!READ_ONCE(vcpu
->preempted
))
2370 if (swait_active(&vcpu
->wq
) && !vcpu_dy_runnable(vcpu
))
2372 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2374 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2377 yielded
= kvm_vcpu_yield_to(vcpu
);
2379 kvm
->last_boosted_vcpu
= i
;
2381 } else if (yielded
< 0) {
2388 kvm_vcpu_set_in_spin_loop(me
, false);
2390 /* Ensure vcpu is not eligible during next spinloop */
2391 kvm_vcpu_set_dy_eligible(me
, false);
2393 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2395 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2397 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2400 if (vmf
->pgoff
== 0)
2401 page
= virt_to_page(vcpu
->run
);
2403 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2404 page
= virt_to_page(vcpu
->arch
.pio_data
);
2406 #ifdef CONFIG_KVM_MMIO
2407 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2408 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2411 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2417 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2418 .fault
= kvm_vcpu_fault
,
2421 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2423 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2427 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2429 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2431 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2432 kvm_put_kvm(vcpu
->kvm
);
2436 static struct file_operations kvm_vcpu_fops
= {
2437 .release
= kvm_vcpu_release
,
2438 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2439 #ifdef CONFIG_KVM_COMPAT
2440 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2442 .mmap
= kvm_vcpu_mmap
,
2443 .llseek
= noop_llseek
,
2447 * Allocates an inode for the vcpu.
2449 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2451 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2454 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2456 char dir_name
[ITOA_MAX_LEN
* 2];
2459 if (!kvm_arch_has_vcpu_debugfs())
2462 if (!debugfs_initialized())
2465 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2466 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2467 vcpu
->kvm
->debugfs_dentry
);
2468 if (!vcpu
->debugfs_dentry
)
2471 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2473 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2481 * Creates some virtual cpus. Good luck creating more than one.
2483 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2486 struct kvm_vcpu
*vcpu
;
2488 if (id
>= KVM_MAX_VCPU_ID
)
2491 mutex_lock(&kvm
->lock
);
2492 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2493 mutex_unlock(&kvm
->lock
);
2497 kvm
->created_vcpus
++;
2498 mutex_unlock(&kvm
->lock
);
2500 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2503 goto vcpu_decrement
;
2506 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2508 r
= kvm_arch_vcpu_setup(vcpu
);
2512 r
= kvm_create_vcpu_debugfs(vcpu
);
2516 mutex_lock(&kvm
->lock
);
2517 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2519 goto unlock_vcpu_destroy
;
2522 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2524 /* Now it's all set up, let userspace reach it */
2526 r
= create_vcpu_fd(vcpu
);
2529 goto unlock_vcpu_destroy
;
2532 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2535 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2536 * before kvm->online_vcpu's incremented value.
2539 atomic_inc(&kvm
->online_vcpus
);
2541 mutex_unlock(&kvm
->lock
);
2542 kvm_arch_vcpu_postcreate(vcpu
);
2545 unlock_vcpu_destroy
:
2546 mutex_unlock(&kvm
->lock
);
2547 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2549 kvm_arch_vcpu_destroy(vcpu
);
2551 mutex_lock(&kvm
->lock
);
2552 kvm
->created_vcpus
--;
2553 mutex_unlock(&kvm
->lock
);
2557 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2560 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2561 vcpu
->sigset_active
= 1;
2562 vcpu
->sigset
= *sigset
;
2564 vcpu
->sigset_active
= 0;
2568 static long kvm_vcpu_ioctl(struct file
*filp
,
2569 unsigned int ioctl
, unsigned long arg
)
2571 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2572 void __user
*argp
= (void __user
*)arg
;
2574 struct kvm_fpu
*fpu
= NULL
;
2575 struct kvm_sregs
*kvm_sregs
= NULL
;
2577 if (vcpu
->kvm
->mm
!= current
->mm
)
2580 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2583 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2585 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2586 * so vcpu_load() would break it.
2588 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2589 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2593 r
= vcpu_load(vcpu
);
2602 oldpid
= rcu_access_pointer(vcpu
->pid
);
2603 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2604 /* The thread running this VCPU changed. */
2605 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2607 rcu_assign_pointer(vcpu
->pid
, newpid
);
2612 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2613 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2616 case KVM_GET_REGS
: {
2617 struct kvm_regs
*kvm_regs
;
2620 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2623 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2627 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2634 case KVM_SET_REGS
: {
2635 struct kvm_regs
*kvm_regs
;
2638 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2639 if (IS_ERR(kvm_regs
)) {
2640 r
= PTR_ERR(kvm_regs
);
2643 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2647 case KVM_GET_SREGS
: {
2648 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2652 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2656 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2661 case KVM_SET_SREGS
: {
2662 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2663 if (IS_ERR(kvm_sregs
)) {
2664 r
= PTR_ERR(kvm_sregs
);
2668 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2671 case KVM_GET_MP_STATE
: {
2672 struct kvm_mp_state mp_state
;
2674 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2678 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2683 case KVM_SET_MP_STATE
: {
2684 struct kvm_mp_state mp_state
;
2687 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2689 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2692 case KVM_TRANSLATE
: {
2693 struct kvm_translation tr
;
2696 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2698 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2702 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2707 case KVM_SET_GUEST_DEBUG
: {
2708 struct kvm_guest_debug dbg
;
2711 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2713 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2716 case KVM_SET_SIGNAL_MASK
: {
2717 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2718 struct kvm_signal_mask kvm_sigmask
;
2719 sigset_t sigset
, *p
;
2724 if (copy_from_user(&kvm_sigmask
, argp
,
2725 sizeof(kvm_sigmask
)))
2728 if (kvm_sigmask
.len
!= sizeof(sigset
))
2731 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2736 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2740 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2744 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2748 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2754 fpu
= memdup_user(argp
, sizeof(*fpu
));
2760 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2764 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2773 #ifdef CONFIG_KVM_COMPAT
2774 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2775 unsigned int ioctl
, unsigned long arg
)
2777 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2778 void __user
*argp
= compat_ptr(arg
);
2781 if (vcpu
->kvm
->mm
!= current
->mm
)
2785 case KVM_SET_SIGNAL_MASK
: {
2786 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2787 struct kvm_signal_mask kvm_sigmask
;
2792 if (copy_from_user(&kvm_sigmask
, argp
,
2793 sizeof(kvm_sigmask
)))
2796 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2799 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2801 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2803 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2807 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2815 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2816 int (*accessor
)(struct kvm_device
*dev
,
2817 struct kvm_device_attr
*attr
),
2820 struct kvm_device_attr attr
;
2825 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2828 return accessor(dev
, &attr
);
2831 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2834 struct kvm_device
*dev
= filp
->private_data
;
2836 if (dev
->kvm
->mm
!= current
->mm
)
2840 case KVM_SET_DEVICE_ATTR
:
2841 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2842 case KVM_GET_DEVICE_ATTR
:
2843 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2844 case KVM_HAS_DEVICE_ATTR
:
2845 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2847 if (dev
->ops
->ioctl
)
2848 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2854 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2856 struct kvm_device
*dev
= filp
->private_data
;
2857 struct kvm
*kvm
= dev
->kvm
;
2863 static const struct file_operations kvm_device_fops
= {
2864 .unlocked_ioctl
= kvm_device_ioctl
,
2865 #ifdef CONFIG_KVM_COMPAT
2866 .compat_ioctl
= kvm_device_ioctl
,
2868 .release
= kvm_device_release
,
2871 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2873 if (filp
->f_op
!= &kvm_device_fops
)
2876 return filp
->private_data
;
2879 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2880 #ifdef CONFIG_KVM_MPIC
2881 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2882 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2886 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2888 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2891 if (kvm_device_ops_table
[type
] != NULL
)
2894 kvm_device_ops_table
[type
] = ops
;
2898 void kvm_unregister_device_ops(u32 type
)
2900 if (kvm_device_ops_table
[type
] != NULL
)
2901 kvm_device_ops_table
[type
] = NULL
;
2904 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2905 struct kvm_create_device
*cd
)
2907 struct kvm_device_ops
*ops
= NULL
;
2908 struct kvm_device
*dev
;
2909 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2913 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2916 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
2917 ops
= kvm_device_ops_table
[type
];
2924 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2931 mutex_lock(&kvm
->lock
);
2932 ret
= ops
->create(dev
, type
);
2934 mutex_unlock(&kvm
->lock
);
2938 list_add(&dev
->vm_node
, &kvm
->devices
);
2939 mutex_unlock(&kvm
->lock
);
2945 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2948 mutex_lock(&kvm
->lock
);
2949 list_del(&dev
->vm_node
);
2950 mutex_unlock(&kvm
->lock
);
2959 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2962 case KVM_CAP_USER_MEMORY
:
2963 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2964 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2965 case KVM_CAP_INTERNAL_ERROR_DATA
:
2966 #ifdef CONFIG_HAVE_KVM_MSI
2967 case KVM_CAP_SIGNAL_MSI
:
2969 #ifdef CONFIG_HAVE_KVM_IRQFD
2971 case KVM_CAP_IRQFD_RESAMPLE
:
2973 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2974 case KVM_CAP_CHECK_EXTENSION_VM
:
2976 #ifdef CONFIG_KVM_MMIO
2977 case KVM_CAP_COALESCED_MMIO
:
2978 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2980 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2981 case KVM_CAP_IRQ_ROUTING
:
2982 return KVM_MAX_IRQ_ROUTES
;
2984 #if KVM_ADDRESS_SPACE_NUM > 1
2985 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2986 return KVM_ADDRESS_SPACE_NUM
;
2991 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2994 static long kvm_vm_ioctl(struct file
*filp
,
2995 unsigned int ioctl
, unsigned long arg
)
2997 struct kvm
*kvm
= filp
->private_data
;
2998 void __user
*argp
= (void __user
*)arg
;
3001 if (kvm
->mm
!= current
->mm
)
3004 case KVM_CREATE_VCPU
:
3005 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3007 case KVM_SET_USER_MEMORY_REGION
: {
3008 struct kvm_userspace_memory_region kvm_userspace_mem
;
3011 if (copy_from_user(&kvm_userspace_mem
, argp
,
3012 sizeof(kvm_userspace_mem
)))
3015 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3018 case KVM_GET_DIRTY_LOG
: {
3019 struct kvm_dirty_log log
;
3022 if (copy_from_user(&log
, argp
, sizeof(log
)))
3024 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3027 #ifdef CONFIG_KVM_MMIO
3028 case KVM_REGISTER_COALESCED_MMIO
: {
3029 struct kvm_coalesced_mmio_zone zone
;
3032 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3034 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3037 case KVM_UNREGISTER_COALESCED_MMIO
: {
3038 struct kvm_coalesced_mmio_zone zone
;
3041 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3043 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3048 struct kvm_irqfd data
;
3051 if (copy_from_user(&data
, argp
, sizeof(data
)))
3053 r
= kvm_irqfd(kvm
, &data
);
3056 case KVM_IOEVENTFD
: {
3057 struct kvm_ioeventfd data
;
3060 if (copy_from_user(&data
, argp
, sizeof(data
)))
3062 r
= kvm_ioeventfd(kvm
, &data
);
3065 #ifdef CONFIG_HAVE_KVM_MSI
3066 case KVM_SIGNAL_MSI
: {
3070 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3072 r
= kvm_send_userspace_msi(kvm
, &msi
);
3076 #ifdef __KVM_HAVE_IRQ_LINE
3077 case KVM_IRQ_LINE_STATUS
:
3078 case KVM_IRQ_LINE
: {
3079 struct kvm_irq_level irq_event
;
3082 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3085 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3086 ioctl
== KVM_IRQ_LINE_STATUS
);
3091 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3092 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3100 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3101 case KVM_SET_GSI_ROUTING
: {
3102 struct kvm_irq_routing routing
;
3103 struct kvm_irq_routing __user
*urouting
;
3104 struct kvm_irq_routing_entry
*entries
= NULL
;
3107 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3110 if (!kvm_arch_can_set_irq_routing(kvm
))
3112 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3118 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3123 if (copy_from_user(entries
, urouting
->entries
,
3124 routing
.nr
* sizeof(*entries
)))
3125 goto out_free_irq_routing
;
3127 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3129 out_free_irq_routing
:
3133 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3134 case KVM_CREATE_DEVICE
: {
3135 struct kvm_create_device cd
;
3138 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3141 r
= kvm_ioctl_create_device(kvm
, &cd
);
3146 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3152 case KVM_CHECK_EXTENSION
:
3153 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3156 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3162 #ifdef CONFIG_KVM_COMPAT
3163 struct compat_kvm_dirty_log
{
3167 compat_uptr_t dirty_bitmap
; /* one bit per page */
3172 static long kvm_vm_compat_ioctl(struct file
*filp
,
3173 unsigned int ioctl
, unsigned long arg
)
3175 struct kvm
*kvm
= filp
->private_data
;
3178 if (kvm
->mm
!= current
->mm
)
3181 case KVM_GET_DIRTY_LOG
: {
3182 struct compat_kvm_dirty_log compat_log
;
3183 struct kvm_dirty_log log
;
3185 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3186 sizeof(compat_log
)))
3188 log
.slot
= compat_log
.slot
;
3189 log
.padding1
= compat_log
.padding1
;
3190 log
.padding2
= compat_log
.padding2
;
3191 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3193 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3197 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3203 static struct file_operations kvm_vm_fops
= {
3204 .release
= kvm_vm_release
,
3205 .unlocked_ioctl
= kvm_vm_ioctl
,
3206 #ifdef CONFIG_KVM_COMPAT
3207 .compat_ioctl
= kvm_vm_compat_ioctl
,
3209 .llseek
= noop_llseek
,
3212 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3218 kvm
= kvm_create_vm(type
);
3220 return PTR_ERR(kvm
);
3221 #ifdef CONFIG_KVM_MMIO
3222 r
= kvm_coalesced_mmio_init(kvm
);
3228 r
= get_unused_fd_flags(O_CLOEXEC
);
3233 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3237 return PTR_ERR(file
);
3241 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3242 * already set, with ->release() being kvm_vm_release(). In error
3243 * cases it will be called by the final fput(file) and will take
3244 * care of doing kvm_put_kvm(kvm).
3246 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3251 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3253 fd_install(r
, file
);
3257 static long kvm_dev_ioctl(struct file
*filp
,
3258 unsigned int ioctl
, unsigned long arg
)
3263 case KVM_GET_API_VERSION
:
3266 r
= KVM_API_VERSION
;
3269 r
= kvm_dev_ioctl_create_vm(arg
);
3271 case KVM_CHECK_EXTENSION
:
3272 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3274 case KVM_GET_VCPU_MMAP_SIZE
:
3277 r
= PAGE_SIZE
; /* struct kvm_run */
3279 r
+= PAGE_SIZE
; /* pio data page */
3281 #ifdef CONFIG_KVM_MMIO
3282 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3285 case KVM_TRACE_ENABLE
:
3286 case KVM_TRACE_PAUSE
:
3287 case KVM_TRACE_DISABLE
:
3291 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3297 static struct file_operations kvm_chardev_ops
= {
3298 .unlocked_ioctl
= kvm_dev_ioctl
,
3299 .compat_ioctl
= kvm_dev_ioctl
,
3300 .llseek
= noop_llseek
,
3303 static struct miscdevice kvm_dev
= {
3309 static void hardware_enable_nolock(void *junk
)
3311 int cpu
= raw_smp_processor_id();
3314 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3317 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3319 r
= kvm_arch_hardware_enable();
3322 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3323 atomic_inc(&hardware_enable_failed
);
3324 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3328 static int kvm_starting_cpu(unsigned int cpu
)
3330 raw_spin_lock(&kvm_count_lock
);
3331 if (kvm_usage_count
)
3332 hardware_enable_nolock(NULL
);
3333 raw_spin_unlock(&kvm_count_lock
);
3337 static void hardware_disable_nolock(void *junk
)
3339 int cpu
= raw_smp_processor_id();
3341 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3343 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3344 kvm_arch_hardware_disable();
3347 static int kvm_dying_cpu(unsigned int cpu
)
3349 raw_spin_lock(&kvm_count_lock
);
3350 if (kvm_usage_count
)
3351 hardware_disable_nolock(NULL
);
3352 raw_spin_unlock(&kvm_count_lock
);
3356 static void hardware_disable_all_nolock(void)
3358 BUG_ON(!kvm_usage_count
);
3361 if (!kvm_usage_count
)
3362 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3365 static void hardware_disable_all(void)
3367 raw_spin_lock(&kvm_count_lock
);
3368 hardware_disable_all_nolock();
3369 raw_spin_unlock(&kvm_count_lock
);
3372 static int hardware_enable_all(void)
3376 raw_spin_lock(&kvm_count_lock
);
3379 if (kvm_usage_count
== 1) {
3380 atomic_set(&hardware_enable_failed
, 0);
3381 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3383 if (atomic_read(&hardware_enable_failed
)) {
3384 hardware_disable_all_nolock();
3389 raw_spin_unlock(&kvm_count_lock
);
3394 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3398 * Some (well, at least mine) BIOSes hang on reboot if
3401 * And Intel TXT required VMX off for all cpu when system shutdown.
3403 pr_info("kvm: exiting hardware virtualization\n");
3404 kvm_rebooting
= true;
3405 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3409 static struct notifier_block kvm_reboot_notifier
= {
3410 .notifier_call
= kvm_reboot
,
3414 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3418 for (i
= 0; i
< bus
->dev_count
; i
++) {
3419 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3421 kvm_iodevice_destructor(pos
);
3426 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3427 const struct kvm_io_range
*r2
)
3429 gpa_t addr1
= r1
->addr
;
3430 gpa_t addr2
= r2
->addr
;
3435 /* If r2->len == 0, match the exact address. If r2->len != 0,
3436 * accept any overlapping write. Any order is acceptable for
3437 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3438 * we process all of them.
3451 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3453 return kvm_io_bus_cmp(p1
, p2
);
3456 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3457 gpa_t addr
, int len
)
3459 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3465 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3466 kvm_io_bus_sort_cmp
, NULL
);
3471 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3472 gpa_t addr
, int len
)
3474 struct kvm_io_range
*range
, key
;
3477 key
= (struct kvm_io_range
) {
3482 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3483 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3487 off
= range
- bus
->range
;
3489 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3495 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3496 struct kvm_io_range
*range
, const void *val
)
3500 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3504 while (idx
< bus
->dev_count
&&
3505 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3506 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3515 /* kvm_io_bus_write - called under kvm->slots_lock */
3516 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3517 int len
, const void *val
)
3519 struct kvm_io_bus
*bus
;
3520 struct kvm_io_range range
;
3523 range
= (struct kvm_io_range
) {
3528 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3531 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3532 return r
< 0 ? r
: 0;
3535 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3536 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3537 gpa_t addr
, int len
, const void *val
, long cookie
)
3539 struct kvm_io_bus
*bus
;
3540 struct kvm_io_range range
;
3542 range
= (struct kvm_io_range
) {
3547 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3551 /* First try the device referenced by cookie. */
3552 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3553 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3554 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3559 * cookie contained garbage; fall back to search and return the
3560 * correct cookie value.
3562 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3565 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3566 struct kvm_io_range
*range
, void *val
)
3570 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3574 while (idx
< bus
->dev_count
&&
3575 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3576 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3584 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3586 /* kvm_io_bus_read - called under kvm->slots_lock */
3587 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3590 struct kvm_io_bus
*bus
;
3591 struct kvm_io_range range
;
3594 range
= (struct kvm_io_range
) {
3599 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3602 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3603 return r
< 0 ? r
: 0;
3607 /* Caller must hold slots_lock. */
3608 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3609 int len
, struct kvm_io_device
*dev
)
3611 struct kvm_io_bus
*new_bus
, *bus
;
3613 bus
= kvm_get_bus(kvm
, bus_idx
);
3617 /* exclude ioeventfd which is limited by maximum fd */
3618 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3621 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3622 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3625 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3626 sizeof(struct kvm_io_range
)));
3627 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3628 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3629 synchronize_srcu_expedited(&kvm
->srcu
);
3635 /* Caller must hold slots_lock. */
3636 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3637 struct kvm_io_device
*dev
)
3640 struct kvm_io_bus
*new_bus
, *bus
;
3642 bus
= kvm_get_bus(kvm
, bus_idx
);
3646 for (i
= 0; i
< bus
->dev_count
; i
++)
3647 if (bus
->range
[i
].dev
== dev
) {
3651 if (i
== bus
->dev_count
)
3654 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3655 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3657 pr_err("kvm: failed to shrink bus, removing it completely\n");
3661 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3662 new_bus
->dev_count
--;
3663 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3664 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3667 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3668 synchronize_srcu_expedited(&kvm
->srcu
);
3673 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3676 struct kvm_io_bus
*bus
;
3677 int dev_idx
, srcu_idx
;
3678 struct kvm_io_device
*iodev
= NULL
;
3680 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3682 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3686 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3690 iodev
= bus
->range
[dev_idx
].dev
;
3693 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3697 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3699 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3700 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3703 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3706 /* The debugfs files are a reference to the kvm struct which
3707 * is still valid when kvm_destroy_vm is called.
3708 * To avoid the race between open and the removal of the debugfs
3709 * directory we test against the users count.
3711 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3714 if (simple_attr_open(inode
, file
, get
,
3715 stat_data
->mode
& S_IWUGO
? set
: NULL
,
3717 kvm_put_kvm(stat_data
->kvm
);
3724 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3726 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3729 simple_attr_release(inode
, file
);
3730 kvm_put_kvm(stat_data
->kvm
);
3735 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3737 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3739 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3744 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3746 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3751 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3756 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3758 __simple_attr_check_format("%llu\n", 0ull);
3759 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3760 vm_stat_clear_per_vm
, "%llu\n");
3763 static const struct file_operations vm_stat_get_per_vm_fops
= {
3764 .owner
= THIS_MODULE
,
3765 .open
= vm_stat_get_per_vm_open
,
3766 .release
= kvm_debugfs_release
,
3767 .read
= simple_attr_read
,
3768 .write
= simple_attr_write
,
3769 .llseek
= no_llseek
,
3772 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3775 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3776 struct kvm_vcpu
*vcpu
;
3780 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3781 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3786 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3789 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3790 struct kvm_vcpu
*vcpu
;
3795 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3796 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3801 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3803 __simple_attr_check_format("%llu\n", 0ull);
3804 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3805 vcpu_stat_clear_per_vm
, "%llu\n");
3808 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3809 .owner
= THIS_MODULE
,
3810 .open
= vcpu_stat_get_per_vm_open
,
3811 .release
= kvm_debugfs_release
,
3812 .read
= simple_attr_read
,
3813 .write
= simple_attr_write
,
3814 .llseek
= no_llseek
,
3817 static const struct file_operations
*stat_fops_per_vm
[] = {
3818 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3819 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3822 static int vm_stat_get(void *_offset
, u64
*val
)
3824 unsigned offset
= (long)_offset
;
3826 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3830 mutex_lock(&kvm_lock
);
3831 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3833 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3836 mutex_unlock(&kvm_lock
);
3840 static int vm_stat_clear(void *_offset
, u64 val
)
3842 unsigned offset
= (long)_offset
;
3844 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3849 mutex_lock(&kvm_lock
);
3850 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3852 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3854 mutex_unlock(&kvm_lock
);
3859 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3861 static int vcpu_stat_get(void *_offset
, u64
*val
)
3863 unsigned offset
= (long)_offset
;
3865 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3869 mutex_lock(&kvm_lock
);
3870 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3872 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3875 mutex_unlock(&kvm_lock
);
3879 static int vcpu_stat_clear(void *_offset
, u64 val
)
3881 unsigned offset
= (long)_offset
;
3883 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3888 mutex_lock(&kvm_lock
);
3889 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3891 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3893 mutex_unlock(&kvm_lock
);
3898 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3901 static const struct file_operations
*stat_fops
[] = {
3902 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3903 [KVM_STAT_VM
] = &vm_stat_fops
,
3906 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3908 struct kobj_uevent_env
*env
;
3909 unsigned long long created
, active
;
3911 if (!kvm_dev
.this_device
|| !kvm
)
3914 mutex_lock(&kvm_lock
);
3915 if (type
== KVM_EVENT_CREATE_VM
) {
3916 kvm_createvm_count
++;
3918 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3921 created
= kvm_createvm_count
;
3922 active
= kvm_active_vms
;
3923 mutex_unlock(&kvm_lock
);
3925 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3929 add_uevent_var(env
, "CREATED=%llu", created
);
3930 add_uevent_var(env
, "COUNT=%llu", active
);
3932 if (type
== KVM_EVENT_CREATE_VM
) {
3933 add_uevent_var(env
, "EVENT=create");
3934 kvm
->userspace_pid
= task_pid_nr(current
);
3935 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3936 add_uevent_var(env
, "EVENT=destroy");
3938 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3940 if (kvm
->debugfs_dentry
) {
3941 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3944 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3946 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3950 /* no need for checks, since we are adding at most only 5 keys */
3951 env
->envp
[env
->envp_idx
++] = NULL
;
3952 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3956 static int kvm_init_debug(void)
3959 struct kvm_stats_debugfs_item
*p
;
3961 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3962 if (kvm_debugfs_dir
== NULL
)
3965 kvm_debugfs_num_entries
= 0;
3966 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3967 int mode
= p
->mode
? p
->mode
: 0644;
3968 if (!debugfs_create_file(p
->name
, mode
, kvm_debugfs_dir
,
3969 (void *)(long)p
->offset
,
3970 stat_fops
[p
->kind
]))
3977 debugfs_remove_recursive(kvm_debugfs_dir
);
3982 static int kvm_suspend(void)
3984 if (kvm_usage_count
)
3985 hardware_disable_nolock(NULL
);
3989 static void kvm_resume(void)
3991 if (kvm_usage_count
) {
3992 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3993 hardware_enable_nolock(NULL
);
3997 static struct syscore_ops kvm_syscore_ops
= {
3998 .suspend
= kvm_suspend
,
3999 .resume
= kvm_resume
,
4003 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4005 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4008 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4010 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4012 if (vcpu
->preempted
)
4013 vcpu
->preempted
= false;
4015 kvm_arch_sched_in(vcpu
, cpu
);
4017 kvm_arch_vcpu_load(vcpu
, cpu
);
4020 static void kvm_sched_out(struct preempt_notifier
*pn
,
4021 struct task_struct
*next
)
4023 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4025 if (current
->state
== TASK_RUNNING
)
4026 vcpu
->preempted
= true;
4027 kvm_arch_vcpu_put(vcpu
);
4030 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4031 struct module
*module
)
4036 r
= kvm_arch_init(opaque
);
4041 * kvm_arch_init makes sure there's at most one caller
4042 * for architectures that support multiple implementations,
4043 * like intel and amd on x86.
4044 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4045 * conflicts in case kvm is already setup for another implementation.
4047 r
= kvm_irqfd_init();
4051 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4056 r
= kvm_arch_hardware_setup();
4060 for_each_online_cpu(cpu
) {
4061 smp_call_function_single(cpu
,
4062 kvm_arch_check_processor_compat
,
4068 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4069 kvm_starting_cpu
, kvm_dying_cpu
);
4072 register_reboot_notifier(&kvm_reboot_notifier
);
4074 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4076 vcpu_align
= __alignof__(struct kvm_vcpu
);
4077 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4078 SLAB_ACCOUNT
, NULL
);
4079 if (!kvm_vcpu_cache
) {
4084 r
= kvm_async_pf_init();
4088 kvm_chardev_ops
.owner
= module
;
4089 kvm_vm_fops
.owner
= module
;
4090 kvm_vcpu_fops
.owner
= module
;
4092 r
= misc_register(&kvm_dev
);
4094 pr_err("kvm: misc device register failed\n");
4098 register_syscore_ops(&kvm_syscore_ops
);
4100 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4101 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4103 r
= kvm_init_debug();
4105 pr_err("kvm: create debugfs files failed\n");
4109 r
= kvm_vfio_ops_init();
4115 unregister_syscore_ops(&kvm_syscore_ops
);
4116 misc_deregister(&kvm_dev
);
4118 kvm_async_pf_deinit();
4120 kmem_cache_destroy(kvm_vcpu_cache
);
4122 unregister_reboot_notifier(&kvm_reboot_notifier
);
4123 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4126 kvm_arch_hardware_unsetup();
4128 free_cpumask_var(cpus_hardware_enabled
);
4136 EXPORT_SYMBOL_GPL(kvm_init
);
4140 debugfs_remove_recursive(kvm_debugfs_dir
);
4141 misc_deregister(&kvm_dev
);
4142 kmem_cache_destroy(kvm_vcpu_cache
);
4143 kvm_async_pf_deinit();
4144 unregister_syscore_ops(&kvm_syscore_ops
);
4145 unregister_reboot_notifier(&kvm_reboot_notifier
);
4146 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4147 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4148 kvm_arch_hardware_unsetup();
4151 free_cpumask_var(cpus_hardware_enabled
);
4152 kvm_vfio_ops_exit();
4154 EXPORT_SYMBOL_GPL(kvm_exit
);