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_SPINLOCK(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 kvm_release_pfn_dirty(kvm_pfn_t pfn
);
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 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
142 return PageReserved(pfn_to_page(pfn
));
148 * Switches to specified vcpu, until a matching vcpu_put()
150 int vcpu_load(struct kvm_vcpu
*vcpu
)
154 if (mutex_lock_killable(&vcpu
->mutex
))
157 preempt_notifier_register(&vcpu
->preempt_notifier
);
158 kvm_arch_vcpu_load(vcpu
, cpu
);
162 EXPORT_SYMBOL_GPL(vcpu_load
);
164 void vcpu_put(struct kvm_vcpu
*vcpu
)
167 kvm_arch_vcpu_put(vcpu
);
168 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
170 mutex_unlock(&vcpu
->mutex
);
172 EXPORT_SYMBOL_GPL(vcpu_put
);
174 /* TODO: merge with kvm_arch_vcpu_should_kick */
175 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
177 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
180 * We need to wait for the VCPU to reenable interrupts and get out of
181 * READING_SHADOW_PAGE_TABLES mode.
183 if (req
& KVM_REQUEST_WAIT
)
184 return mode
!= OUTSIDE_GUEST_MODE
;
187 * Need to kick a running VCPU, but otherwise there is nothing to do.
189 return mode
== IN_GUEST_MODE
;
192 static void ack_flush(void *_completed
)
196 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
199 cpus
= cpu_online_mask
;
201 if (cpumask_empty(cpus
))
204 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
208 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
213 struct kvm_vcpu
*vcpu
;
215 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
218 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
219 kvm_make_request(req
, vcpu
);
222 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
225 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
226 kvm_request_needs_ipi(vcpu
, req
))
227 __cpumask_set_cpu(cpu
, cpus
);
229 called
= kvm_kick_many_cpus(cpus
, !!(req
& KVM_REQUEST_WAIT
));
231 free_cpumask_var(cpus
);
235 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
236 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
239 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
240 * kvm_make_all_cpus_request.
242 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
245 * We want to publish modifications to the page tables before reading
246 * mode. Pairs with a memory barrier in arch-specific code.
247 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
248 * and smp_mb in walk_shadow_page_lockless_begin/end.
249 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
251 * There is already an smp_mb__after_atomic() before
252 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
255 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
256 ++kvm
->stat
.remote_tlb_flush
;
257 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
259 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
262 void kvm_reload_remote_mmus(struct kvm
*kvm
)
264 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
267 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
272 mutex_init(&vcpu
->mutex
);
277 init_swait_queue_head(&vcpu
->wq
);
278 kvm_async_pf_vcpu_init(vcpu
);
281 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
283 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
288 vcpu
->run
= page_address(page
);
290 kvm_vcpu_set_in_spin_loop(vcpu
, false);
291 kvm_vcpu_set_dy_eligible(vcpu
, false);
292 vcpu
->preempted
= false;
294 r
= kvm_arch_vcpu_init(vcpu
);
300 free_page((unsigned long)vcpu
->run
);
304 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
306 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
309 * no need for rcu_read_lock as VCPU_RUN is the only place that
310 * will change the vcpu->pid pointer and on uninit all file
311 * descriptors are already gone.
313 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
314 kvm_arch_vcpu_uninit(vcpu
);
315 free_page((unsigned long)vcpu
->run
);
317 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
319 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
320 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
322 return container_of(mn
, struct kvm
, mmu_notifier
);
325 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
326 struct mm_struct
*mm
,
327 unsigned long address
)
329 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
330 int need_tlb_flush
, idx
;
333 * When ->invalidate_page runs, the linux pte has been zapped
334 * already but the page is still allocated until
335 * ->invalidate_page returns. So if we increase the sequence
336 * here the kvm page fault will notice if the spte can't be
337 * established because the page is going to be freed. If
338 * instead the kvm page fault establishes the spte before
339 * ->invalidate_page runs, kvm_unmap_hva will release it
342 * The sequence increase only need to be seen at spin_unlock
343 * time, and not at spin_lock time.
345 * Increasing the sequence after the spin_unlock would be
346 * unsafe because the kvm page fault could then establish the
347 * pte after kvm_unmap_hva returned, without noticing the page
348 * is going to be freed.
350 idx
= srcu_read_lock(&kvm
->srcu
);
351 spin_lock(&kvm
->mmu_lock
);
353 kvm
->mmu_notifier_seq
++;
354 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
355 /* we've to flush the tlb before the pages can be freed */
357 kvm_flush_remote_tlbs(kvm
);
359 spin_unlock(&kvm
->mmu_lock
);
361 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
363 srcu_read_unlock(&kvm
->srcu
, idx
);
366 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
367 struct mm_struct
*mm
,
368 unsigned long address
,
371 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
374 idx
= srcu_read_lock(&kvm
->srcu
);
375 spin_lock(&kvm
->mmu_lock
);
376 kvm
->mmu_notifier_seq
++;
377 kvm_set_spte_hva(kvm
, address
, pte
);
378 spin_unlock(&kvm
->mmu_lock
);
379 srcu_read_unlock(&kvm
->srcu
, idx
);
382 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
383 struct mm_struct
*mm
,
387 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
388 int need_tlb_flush
= 0, idx
;
390 idx
= srcu_read_lock(&kvm
->srcu
);
391 spin_lock(&kvm
->mmu_lock
);
393 * The count increase must become visible at unlock time as no
394 * spte can be established without taking the mmu_lock and
395 * count is also read inside the mmu_lock critical section.
397 kvm
->mmu_notifier_count
++;
398 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
399 need_tlb_flush
|= kvm
->tlbs_dirty
;
400 /* we've to flush the tlb before the pages can be freed */
402 kvm_flush_remote_tlbs(kvm
);
404 spin_unlock(&kvm
->mmu_lock
);
405 srcu_read_unlock(&kvm
->srcu
, idx
);
408 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
409 struct mm_struct
*mm
,
413 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
415 spin_lock(&kvm
->mmu_lock
);
417 * This sequence increase will notify the kvm page fault that
418 * the page that is going to be mapped in the spte could have
421 kvm
->mmu_notifier_seq
++;
424 * The above sequence increase must be visible before the
425 * below count decrease, which is ensured by the smp_wmb above
426 * in conjunction with the smp_rmb in mmu_notifier_retry().
428 kvm
->mmu_notifier_count
--;
429 spin_unlock(&kvm
->mmu_lock
);
431 BUG_ON(kvm
->mmu_notifier_count
< 0);
434 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
435 struct mm_struct
*mm
,
439 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
442 idx
= srcu_read_lock(&kvm
->srcu
);
443 spin_lock(&kvm
->mmu_lock
);
445 young
= kvm_age_hva(kvm
, start
, end
);
447 kvm_flush_remote_tlbs(kvm
);
449 spin_unlock(&kvm
->mmu_lock
);
450 srcu_read_unlock(&kvm
->srcu
, idx
);
455 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
456 struct mm_struct
*mm
,
460 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
463 idx
= srcu_read_lock(&kvm
->srcu
);
464 spin_lock(&kvm
->mmu_lock
);
466 * Even though we do not flush TLB, this will still adversely
467 * affect performance on pre-Haswell Intel EPT, where there is
468 * no EPT Access Bit to clear so that we have to tear down EPT
469 * tables instead. If we find this unacceptable, we can always
470 * add a parameter to kvm_age_hva so that it effectively doesn't
471 * do anything on clear_young.
473 * Also note that currently we never issue secondary TLB flushes
474 * from clear_young, leaving this job up to the regular system
475 * cadence. If we find this inaccurate, we might come up with a
476 * more sophisticated heuristic later.
478 young
= kvm_age_hva(kvm
, start
, end
);
479 spin_unlock(&kvm
->mmu_lock
);
480 srcu_read_unlock(&kvm
->srcu
, idx
);
485 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
486 struct mm_struct
*mm
,
487 unsigned long address
)
489 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
492 idx
= srcu_read_lock(&kvm
->srcu
);
493 spin_lock(&kvm
->mmu_lock
);
494 young
= kvm_test_age_hva(kvm
, address
);
495 spin_unlock(&kvm
->mmu_lock
);
496 srcu_read_unlock(&kvm
->srcu
, idx
);
501 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
502 struct mm_struct
*mm
)
504 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
507 idx
= srcu_read_lock(&kvm
->srcu
);
508 kvm_arch_flush_shadow_all(kvm
);
509 srcu_read_unlock(&kvm
->srcu
, idx
);
512 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
513 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
514 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
515 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
516 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
517 .clear_young
= kvm_mmu_notifier_clear_young
,
518 .test_young
= kvm_mmu_notifier_test_young
,
519 .change_pte
= kvm_mmu_notifier_change_pte
,
520 .release
= kvm_mmu_notifier_release
,
523 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
525 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
526 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
529 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
531 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
536 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
538 static struct kvm_memslots
*kvm_alloc_memslots(void)
541 struct kvm_memslots
*slots
;
543 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
547 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
548 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
553 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
555 if (!memslot
->dirty_bitmap
)
558 kvfree(memslot
->dirty_bitmap
);
559 memslot
->dirty_bitmap
= NULL
;
563 * Free any memory in @free but not in @dont.
565 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
566 struct kvm_memory_slot
*dont
)
568 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
569 kvm_destroy_dirty_bitmap(free
);
571 kvm_arch_free_memslot(kvm
, free
, dont
);
576 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
578 struct kvm_memory_slot
*memslot
;
583 kvm_for_each_memslot(memslot
, slots
)
584 kvm_free_memslot(kvm
, memslot
, NULL
);
589 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
593 if (!kvm
->debugfs_dentry
)
596 debugfs_remove_recursive(kvm
->debugfs_dentry
);
598 if (kvm
->debugfs_stat_data
) {
599 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
600 kfree(kvm
->debugfs_stat_data
[i
]);
601 kfree(kvm
->debugfs_stat_data
);
605 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
607 char dir_name
[ITOA_MAX_LEN
* 2];
608 struct kvm_stat_data
*stat_data
;
609 struct kvm_stats_debugfs_item
*p
;
611 if (!debugfs_initialized())
614 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
615 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
617 if (!kvm
->debugfs_dentry
)
620 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
621 sizeof(*kvm
->debugfs_stat_data
),
623 if (!kvm
->debugfs_stat_data
)
626 for (p
= debugfs_entries
; p
->name
; p
++) {
627 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
631 stat_data
->kvm
= kvm
;
632 stat_data
->offset
= p
->offset
;
633 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
634 if (!debugfs_create_file(p
->name
, 0644,
637 stat_fops_per_vm
[p
->kind
]))
643 static struct kvm
*kvm_create_vm(unsigned long type
)
646 struct kvm
*kvm
= kvm_arch_alloc_vm();
649 return ERR_PTR(-ENOMEM
);
651 spin_lock_init(&kvm
->mmu_lock
);
653 kvm
->mm
= current
->mm
;
654 kvm_eventfd_init(kvm
);
655 mutex_init(&kvm
->lock
);
656 mutex_init(&kvm
->irq_lock
);
657 mutex_init(&kvm
->slots_lock
);
658 refcount_set(&kvm
->users_count
, 1);
659 INIT_LIST_HEAD(&kvm
->devices
);
661 r
= kvm_arch_init_vm(kvm
, type
);
663 goto out_err_no_disable
;
665 r
= hardware_enable_all();
667 goto out_err_no_disable
;
669 #ifdef CONFIG_HAVE_KVM_IRQFD
670 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
673 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
676 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
677 struct kvm_memslots
*slots
= kvm_alloc_memslots();
679 goto out_err_no_srcu
;
681 * Generations must be different for each address space.
682 * Init kvm generation close to the maximum to easily test the
683 * code of handling generation number wrap-around.
685 slots
->generation
= i
* 2 - 150;
686 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
689 if (init_srcu_struct(&kvm
->srcu
))
690 goto out_err_no_srcu
;
691 if (init_srcu_struct(&kvm
->irq_srcu
))
692 goto out_err_no_irq_srcu
;
693 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
694 rcu_assign_pointer(kvm
->buses
[i
],
695 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
700 r
= kvm_init_mmu_notifier(kvm
);
704 spin_lock(&kvm_lock
);
705 list_add(&kvm
->vm_list
, &vm_list
);
706 spin_unlock(&kvm_lock
);
708 preempt_notifier_inc();
713 cleanup_srcu_struct(&kvm
->irq_srcu
);
715 cleanup_srcu_struct(&kvm
->srcu
);
717 hardware_disable_all();
719 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
720 kfree(kvm_get_bus(kvm
, i
));
721 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
722 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
723 kvm_arch_free_vm(kvm
);
728 static void kvm_destroy_devices(struct kvm
*kvm
)
730 struct kvm_device
*dev
, *tmp
;
733 * We do not need to take the kvm->lock here, because nobody else
734 * has a reference to the struct kvm at this point and therefore
735 * cannot access the devices list anyhow.
737 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
738 list_del(&dev
->vm_node
);
739 dev
->ops
->destroy(dev
);
743 static void kvm_destroy_vm(struct kvm
*kvm
)
746 struct mm_struct
*mm
= kvm
->mm
;
748 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
749 kvm_destroy_vm_debugfs(kvm
);
750 kvm_arch_sync_events(kvm
);
751 spin_lock(&kvm_lock
);
752 list_del(&kvm
->vm_list
);
753 spin_unlock(&kvm_lock
);
754 kvm_free_irq_routing(kvm
);
755 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
756 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
759 kvm_io_bus_destroy(bus
);
760 kvm
->buses
[i
] = NULL
;
762 kvm_coalesced_mmio_free(kvm
);
763 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
764 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
766 kvm_arch_flush_shadow_all(kvm
);
768 kvm_arch_destroy_vm(kvm
);
769 kvm_destroy_devices(kvm
);
770 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
771 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
772 cleanup_srcu_struct(&kvm
->irq_srcu
);
773 cleanup_srcu_struct(&kvm
->srcu
);
774 kvm_arch_free_vm(kvm
);
775 preempt_notifier_dec();
776 hardware_disable_all();
780 void kvm_get_kvm(struct kvm
*kvm
)
782 refcount_inc(&kvm
->users_count
);
784 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
786 void kvm_put_kvm(struct kvm
*kvm
)
788 if (refcount_dec_and_test(&kvm
->users_count
))
791 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
794 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
796 struct kvm
*kvm
= filp
->private_data
;
798 kvm_irqfd_release(kvm
);
805 * Allocation size is twice as large as the actual dirty bitmap size.
806 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
808 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
810 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
812 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
813 if (!memslot
->dirty_bitmap
)
820 * Insert memslot and re-sort memslots based on their GFN,
821 * so binary search could be used to lookup GFN.
822 * Sorting algorithm takes advantage of having initially
823 * sorted array and known changed memslot position.
825 static void update_memslots(struct kvm_memslots
*slots
,
826 struct kvm_memory_slot
*new)
829 int i
= slots
->id_to_index
[id
];
830 struct kvm_memory_slot
*mslots
= slots
->memslots
;
832 WARN_ON(mslots
[i
].id
!= id
);
834 WARN_ON(!mslots
[i
].npages
);
835 if (mslots
[i
].npages
)
838 if (!mslots
[i
].npages
)
842 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
843 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
844 if (!mslots
[i
+ 1].npages
)
846 mslots
[i
] = mslots
[i
+ 1];
847 slots
->id_to_index
[mslots
[i
].id
] = i
;
852 * The ">=" is needed when creating a slot with base_gfn == 0,
853 * so that it moves before all those with base_gfn == npages == 0.
855 * On the other hand, if new->npages is zero, the above loop has
856 * already left i pointing to the beginning of the empty part of
857 * mslots, and the ">=" would move the hole backwards in this
858 * case---which is wrong. So skip the loop when deleting a slot.
862 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
863 mslots
[i
] = mslots
[i
- 1];
864 slots
->id_to_index
[mslots
[i
].id
] = i
;
868 WARN_ON_ONCE(i
!= slots
->used_slots
);
871 slots
->id_to_index
[mslots
[i
].id
] = i
;
874 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
876 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
878 #ifdef __KVM_HAVE_READONLY_MEM
879 valid_flags
|= KVM_MEM_READONLY
;
882 if (mem
->flags
& ~valid_flags
)
888 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
889 int as_id
, struct kvm_memslots
*slots
)
891 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
894 * Set the low bit in the generation, which disables SPTE caching
895 * until the end of synchronize_srcu_expedited.
897 WARN_ON(old_memslots
->generation
& 1);
898 slots
->generation
= old_memslots
->generation
+ 1;
900 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
901 synchronize_srcu_expedited(&kvm
->srcu
);
904 * Increment the new memslot generation a second time. This prevents
905 * vm exits that race with memslot updates from caching a memslot
906 * generation that will (potentially) be valid forever.
908 * Generations must be unique even across address spaces. We do not need
909 * a global counter for that, instead the generation space is evenly split
910 * across address spaces. For example, with two address spaces, address
911 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
912 * use generations 2, 6, 10, 14, ...
914 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
916 kvm_arch_memslots_updated(kvm
, slots
);
922 * Allocate some memory and give it an address in the guest physical address
925 * Discontiguous memory is allowed, mostly for framebuffers.
927 * Must be called holding kvm->slots_lock for write.
929 int __kvm_set_memory_region(struct kvm
*kvm
,
930 const struct kvm_userspace_memory_region
*mem
)
934 unsigned long npages
;
935 struct kvm_memory_slot
*slot
;
936 struct kvm_memory_slot old
, new;
937 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
939 enum kvm_mr_change change
;
941 r
= check_memory_region_flags(mem
);
946 as_id
= mem
->slot
>> 16;
949 /* General sanity checks */
950 if (mem
->memory_size
& (PAGE_SIZE
- 1))
952 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
954 /* We can read the guest memory with __xxx_user() later on. */
955 if ((id
< KVM_USER_MEM_SLOTS
) &&
956 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
957 !access_ok(VERIFY_WRITE
,
958 (void __user
*)(unsigned long)mem
->userspace_addr
,
961 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
963 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
966 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
967 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
968 npages
= mem
->memory_size
>> PAGE_SHIFT
;
970 if (npages
> KVM_MEM_MAX_NR_PAGES
)
976 new.base_gfn
= base_gfn
;
978 new.flags
= mem
->flags
;
982 change
= KVM_MR_CREATE
;
983 else { /* Modify an existing slot. */
984 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
985 (npages
!= old
.npages
) ||
986 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
989 if (base_gfn
!= old
.base_gfn
)
990 change
= KVM_MR_MOVE
;
991 else if (new.flags
!= old
.flags
)
992 change
= KVM_MR_FLAGS_ONLY
;
993 else { /* Nothing to change. */
1002 change
= KVM_MR_DELETE
;
1007 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1008 /* Check for overlaps */
1010 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
1011 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
1014 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
1015 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
1020 /* Free page dirty bitmap if unneeded */
1021 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1022 new.dirty_bitmap
= NULL
;
1025 if (change
== KVM_MR_CREATE
) {
1026 new.userspace_addr
= mem
->userspace_addr
;
1028 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1032 /* Allocate page dirty bitmap if needed */
1033 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1034 if (kvm_create_dirty_bitmap(&new) < 0)
1038 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1041 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1043 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1044 slot
= id_to_memslot(slots
, id
);
1045 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1047 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1049 /* From this point no new shadow pages pointing to a deleted,
1050 * or moved, memslot will be created.
1052 * validation of sp->gfn happens in:
1053 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1054 * - kvm_is_visible_gfn (mmu_check_roots)
1056 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1059 * We can re-use the old_memslots from above, the only difference
1060 * from the currently installed memslots is the invalid flag. This
1061 * will get overwritten by update_memslots anyway.
1063 slots
= old_memslots
;
1066 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1070 /* actual memory is freed via old in kvm_free_memslot below */
1071 if (change
== KVM_MR_DELETE
) {
1072 new.dirty_bitmap
= NULL
;
1073 memset(&new.arch
, 0, sizeof(new.arch
));
1076 update_memslots(slots
, &new);
1077 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1079 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1081 kvm_free_memslot(kvm
, &old
, &new);
1082 kvfree(old_memslots
);
1088 kvm_free_memslot(kvm
, &new, &old
);
1092 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1094 int kvm_set_memory_region(struct kvm
*kvm
,
1095 const struct kvm_userspace_memory_region
*mem
)
1099 mutex_lock(&kvm
->slots_lock
);
1100 r
= __kvm_set_memory_region(kvm
, mem
);
1101 mutex_unlock(&kvm
->slots_lock
);
1104 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1106 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1107 struct kvm_userspace_memory_region
*mem
)
1109 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1112 return kvm_set_memory_region(kvm
, mem
);
1115 int kvm_get_dirty_log(struct kvm
*kvm
,
1116 struct kvm_dirty_log
*log
, int *is_dirty
)
1118 struct kvm_memslots
*slots
;
1119 struct kvm_memory_slot
*memslot
;
1122 unsigned long any
= 0;
1124 as_id
= log
->slot
>> 16;
1125 id
= (u16
)log
->slot
;
1126 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1129 slots
= __kvm_memslots(kvm
, as_id
);
1130 memslot
= id_to_memslot(slots
, id
);
1131 if (!memslot
->dirty_bitmap
)
1134 n
= kvm_dirty_bitmap_bytes(memslot
);
1136 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1137 any
= memslot
->dirty_bitmap
[i
];
1139 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1146 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1148 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1150 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1151 * are dirty write protect them for next write.
1152 * @kvm: pointer to kvm instance
1153 * @log: slot id and address to which we copy the log
1154 * @is_dirty: flag set if any page is dirty
1156 * We need to keep it in mind that VCPU threads can write to the bitmap
1157 * concurrently. So, to avoid losing track of dirty pages we keep the
1160 * 1. Take a snapshot of the bit and clear it if needed.
1161 * 2. Write protect the corresponding page.
1162 * 3. Copy the snapshot to the userspace.
1163 * 4. Upon return caller flushes TLB's if needed.
1165 * Between 2 and 4, the guest may write to the page using the remaining TLB
1166 * entry. This is not a problem because the page is reported dirty using
1167 * the snapshot taken before and step 4 ensures that writes done after
1168 * exiting to userspace will be logged for the next call.
1171 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1172 struct kvm_dirty_log
*log
, bool *is_dirty
)
1174 struct kvm_memslots
*slots
;
1175 struct kvm_memory_slot
*memslot
;
1178 unsigned long *dirty_bitmap
;
1179 unsigned long *dirty_bitmap_buffer
;
1181 as_id
= log
->slot
>> 16;
1182 id
= (u16
)log
->slot
;
1183 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1186 slots
= __kvm_memslots(kvm
, as_id
);
1187 memslot
= id_to_memslot(slots
, id
);
1189 dirty_bitmap
= memslot
->dirty_bitmap
;
1193 n
= kvm_dirty_bitmap_bytes(memslot
);
1195 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1196 memset(dirty_bitmap_buffer
, 0, n
);
1198 spin_lock(&kvm
->mmu_lock
);
1200 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1204 if (!dirty_bitmap
[i
])
1209 mask
= xchg(&dirty_bitmap
[i
], 0);
1210 dirty_bitmap_buffer
[i
] = mask
;
1213 offset
= i
* BITS_PER_LONG
;
1214 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1219 spin_unlock(&kvm
->mmu_lock
);
1220 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1224 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1227 bool kvm_largepages_enabled(void)
1229 return largepages_enabled
;
1232 void kvm_disable_largepages(void)
1234 largepages_enabled
= false;
1236 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1238 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1240 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1242 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1244 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1246 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1249 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1251 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1253 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1254 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1259 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1261 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1263 struct vm_area_struct
*vma
;
1264 unsigned long addr
, size
;
1268 addr
= gfn_to_hva(kvm
, gfn
);
1269 if (kvm_is_error_hva(addr
))
1272 down_read(¤t
->mm
->mmap_sem
);
1273 vma
= find_vma(current
->mm
, addr
);
1277 size
= vma_kernel_pagesize(vma
);
1280 up_read(¤t
->mm
->mmap_sem
);
1285 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1287 return slot
->flags
& KVM_MEM_READONLY
;
1290 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1291 gfn_t
*nr_pages
, bool write
)
1293 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1294 return KVM_HVA_ERR_BAD
;
1296 if (memslot_is_readonly(slot
) && write
)
1297 return KVM_HVA_ERR_RO_BAD
;
1300 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1302 return __gfn_to_hva_memslot(slot
, gfn
);
1305 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1308 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1311 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1314 return gfn_to_hva_many(slot
, gfn
, NULL
);
1316 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1318 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1320 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1322 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1324 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1326 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1328 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1331 * If writable is set to false, the hva returned by this function is only
1332 * allowed to be read.
1334 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1335 gfn_t gfn
, bool *writable
)
1337 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1339 if (!kvm_is_error_hva(hva
) && writable
)
1340 *writable
= !memslot_is_readonly(slot
);
1345 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1347 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1349 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1352 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1354 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1356 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1359 static int get_user_page_nowait(unsigned long start
, int write
,
1362 int flags
= FOLL_NOWAIT
| FOLL_HWPOISON
;
1365 flags
|= FOLL_WRITE
;
1367 return get_user_pages(start
, 1, flags
, page
, NULL
);
1370 static inline int check_user_page_hwpoison(unsigned long addr
)
1372 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1374 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1375 return rc
== -EHWPOISON
;
1379 * The atomic path to get the writable pfn which will be stored in @pfn,
1380 * true indicates success, otherwise false is returned.
1382 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1383 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1385 struct page
*page
[1];
1388 if (!(async
|| atomic
))
1392 * Fast pin a writable pfn only if it is a write fault request
1393 * or the caller allows to map a writable pfn for a read fault
1396 if (!(write_fault
|| writable
))
1399 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1401 *pfn
= page_to_pfn(page
[0]);
1412 * The slow path to get the pfn of the specified host virtual address,
1413 * 1 indicates success, -errno is returned if error is detected.
1415 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1416 bool *writable
, kvm_pfn_t
*pfn
)
1418 struct page
*page
[1];
1424 *writable
= write_fault
;
1427 down_read(¤t
->mm
->mmap_sem
);
1428 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1429 up_read(¤t
->mm
->mmap_sem
);
1431 unsigned int flags
= FOLL_HWPOISON
;
1434 flags
|= FOLL_WRITE
;
1436 npages
= get_user_pages_unlocked(addr
, 1, page
, flags
);
1441 /* map read fault as writable if possible */
1442 if (unlikely(!write_fault
) && writable
) {
1443 struct page
*wpage
[1];
1445 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1454 *pfn
= page_to_pfn(page
[0]);
1458 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1460 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1463 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1469 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1470 unsigned long addr
, bool *async
,
1471 bool write_fault
, kvm_pfn_t
*p_pfn
)
1476 r
= follow_pfn(vma
, addr
, &pfn
);
1479 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1480 * not call the fault handler, so do it here.
1482 bool unlocked
= false;
1483 r
= fixup_user_fault(current
, current
->mm
, addr
,
1484 (write_fault
? FAULT_FLAG_WRITE
: 0),
1491 r
= follow_pfn(vma
, addr
, &pfn
);
1499 * Get a reference here because callers of *hva_to_pfn* and
1500 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1501 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1502 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1503 * simply do nothing for reserved pfns.
1505 * Whoever called remap_pfn_range is also going to call e.g.
1506 * unmap_mapping_range before the underlying pages are freed,
1507 * causing a call to our MMU notifier.
1516 * Pin guest page in memory and return its pfn.
1517 * @addr: host virtual address which maps memory to the guest
1518 * @atomic: whether this function can sleep
1519 * @async: whether this function need to wait IO complete if the
1520 * host page is not in the memory
1521 * @write_fault: whether we should get a writable host page
1522 * @writable: whether it allows to map a writable host page for !@write_fault
1524 * The function will map a writable host page for these two cases:
1525 * 1): @write_fault = true
1526 * 2): @write_fault = false && @writable, @writable will tell the caller
1527 * whether the mapping is writable.
1529 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1530 bool write_fault
, bool *writable
)
1532 struct vm_area_struct
*vma
;
1536 /* we can do it either atomically or asynchronously, not both */
1537 BUG_ON(atomic
&& async
);
1539 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1543 return KVM_PFN_ERR_FAULT
;
1545 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1549 down_read(¤t
->mm
->mmap_sem
);
1550 if (npages
== -EHWPOISON
||
1551 (!async
&& check_user_page_hwpoison(addr
))) {
1552 pfn
= KVM_PFN_ERR_HWPOISON
;
1557 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1560 pfn
= KVM_PFN_ERR_FAULT
;
1561 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1562 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, &pfn
);
1566 pfn
= KVM_PFN_ERR_FAULT
;
1568 if (async
&& vma_is_valid(vma
, write_fault
))
1570 pfn
= KVM_PFN_ERR_FAULT
;
1573 up_read(¤t
->mm
->mmap_sem
);
1577 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1578 bool atomic
, bool *async
, bool write_fault
,
1581 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1583 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1586 return KVM_PFN_ERR_RO_FAULT
;
1589 if (kvm_is_error_hva(addr
)) {
1592 return KVM_PFN_NOSLOT
;
1595 /* Do not map writable pfn in the readonly memslot. */
1596 if (writable
&& memslot_is_readonly(slot
)) {
1601 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1604 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1606 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1609 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1610 write_fault
, writable
);
1612 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1614 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1616 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1618 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1620 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1622 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1624 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1626 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1628 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1630 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1632 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1634 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1636 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1638 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1640 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1642 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1644 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1646 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1648 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1650 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1651 struct page
**pages
, int nr_pages
)
1656 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1657 if (kvm_is_error_hva(addr
))
1660 if (entry
< nr_pages
)
1663 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1665 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1667 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1669 if (is_error_noslot_pfn(pfn
))
1670 return KVM_ERR_PTR_BAD_PAGE
;
1672 if (kvm_is_reserved_pfn(pfn
)) {
1674 return KVM_ERR_PTR_BAD_PAGE
;
1677 return pfn_to_page(pfn
);
1680 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1684 pfn
= gfn_to_pfn(kvm
, gfn
);
1686 return kvm_pfn_to_page(pfn
);
1688 EXPORT_SYMBOL_GPL(gfn_to_page
);
1690 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1694 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1696 return kvm_pfn_to_page(pfn
);
1698 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1700 void kvm_release_page_clean(struct page
*page
)
1702 WARN_ON(is_error_page(page
));
1704 kvm_release_pfn_clean(page_to_pfn(page
));
1706 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1708 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1710 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1711 put_page(pfn_to_page(pfn
));
1713 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1715 void kvm_release_page_dirty(struct page
*page
)
1717 WARN_ON(is_error_page(page
));
1719 kvm_release_pfn_dirty(page_to_pfn(page
));
1721 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1723 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1725 kvm_set_pfn_dirty(pfn
);
1726 kvm_release_pfn_clean(pfn
);
1729 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1731 if (!kvm_is_reserved_pfn(pfn
)) {
1732 struct page
*page
= pfn_to_page(pfn
);
1734 if (!PageReserved(page
))
1738 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1740 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1742 if (!kvm_is_reserved_pfn(pfn
))
1743 mark_page_accessed(pfn_to_page(pfn
));
1745 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1747 void kvm_get_pfn(kvm_pfn_t pfn
)
1749 if (!kvm_is_reserved_pfn(pfn
))
1750 get_page(pfn_to_page(pfn
));
1752 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1754 static int next_segment(unsigned long len
, int offset
)
1756 if (len
> PAGE_SIZE
- offset
)
1757 return PAGE_SIZE
- offset
;
1762 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1763 void *data
, int offset
, int len
)
1768 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1769 if (kvm_is_error_hva(addr
))
1771 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1777 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1780 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1782 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1784 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1786 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1787 int offset
, int len
)
1789 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1791 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1793 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1795 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1797 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1799 int offset
= offset_in_page(gpa
);
1802 while ((seg
= next_segment(len
, offset
)) != 0) {
1803 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1813 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1815 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1817 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1819 int offset
= offset_in_page(gpa
);
1822 while ((seg
= next_segment(len
, offset
)) != 0) {
1823 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1833 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1835 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1836 void *data
, int offset
, unsigned long len
)
1841 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1842 if (kvm_is_error_hva(addr
))
1844 pagefault_disable();
1845 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1852 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1855 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1856 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1857 int offset
= offset_in_page(gpa
);
1859 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1861 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1863 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1864 void *data
, unsigned long len
)
1866 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1867 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1868 int offset
= offset_in_page(gpa
);
1870 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1872 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1874 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1875 const void *data
, int offset
, int len
)
1880 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1881 if (kvm_is_error_hva(addr
))
1883 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1886 mark_page_dirty_in_slot(memslot
, gfn
);
1890 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1891 const void *data
, int offset
, int len
)
1893 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1895 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1897 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1899 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1900 const void *data
, int offset
, int len
)
1902 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1904 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1906 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1908 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1911 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1913 int offset
= offset_in_page(gpa
);
1916 while ((seg
= next_segment(len
, offset
)) != 0) {
1917 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1927 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1929 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1932 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1934 int offset
= offset_in_page(gpa
);
1937 while ((seg
= next_segment(len
, offset
)) != 0) {
1938 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1948 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1950 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
1951 struct gfn_to_hva_cache
*ghc
,
1952 gpa_t gpa
, unsigned long len
)
1954 int offset
= offset_in_page(gpa
);
1955 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1956 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1957 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1958 gfn_t nr_pages_avail
;
1961 ghc
->generation
= slots
->generation
;
1963 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1964 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1965 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1969 * If the requested region crosses two memslots, we still
1970 * verify that the entire region is valid here.
1972 while (start_gfn
<= end_gfn
) {
1973 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
1974 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1976 if (kvm_is_error_hva(ghc
->hva
))
1978 start_gfn
+= nr_pages_avail
;
1980 /* Use the slow path for cross page reads and writes. */
1981 ghc
->memslot
= NULL
;
1986 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1987 gpa_t gpa
, unsigned long len
)
1989 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1990 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
1992 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1994 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1995 void *data
, int offset
, unsigned long len
)
1997 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1999 gpa_t gpa
= ghc
->gpa
+ offset
;
2001 BUG_ON(len
+ offset
> ghc
->len
);
2003 if (slots
->generation
!= ghc
->generation
)
2004 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2006 if (unlikely(!ghc
->memslot
))
2007 return kvm_write_guest(kvm
, gpa
, data
, len
);
2009 if (kvm_is_error_hva(ghc
->hva
))
2012 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2015 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2019 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2021 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2022 void *data
, unsigned long len
)
2024 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2026 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2028 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2029 void *data
, unsigned long len
)
2031 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2034 BUG_ON(len
> ghc
->len
);
2036 if (slots
->generation
!= ghc
->generation
)
2037 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2039 if (unlikely(!ghc
->memslot
))
2040 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2042 if (kvm_is_error_hva(ghc
->hva
))
2045 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2051 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2053 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2055 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2057 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2059 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2061 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2063 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2065 int offset
= offset_in_page(gpa
);
2068 while ((seg
= next_segment(len
, offset
)) != 0) {
2069 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2078 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2080 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2083 if (memslot
&& memslot
->dirty_bitmap
) {
2084 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2086 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2090 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2092 struct kvm_memory_slot
*memslot
;
2094 memslot
= gfn_to_memslot(kvm
, gfn
);
2095 mark_page_dirty_in_slot(memslot
, gfn
);
2097 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2099 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2101 struct kvm_memory_slot
*memslot
;
2103 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2104 mark_page_dirty_in_slot(memslot
, gfn
);
2106 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2108 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2110 unsigned int old
, val
, grow
;
2112 old
= val
= vcpu
->halt_poll_ns
;
2113 grow
= READ_ONCE(halt_poll_ns_grow
);
2115 if (val
== 0 && grow
)
2120 if (val
> halt_poll_ns
)
2123 vcpu
->halt_poll_ns
= val
;
2124 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2127 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2129 unsigned int old
, val
, shrink
;
2131 old
= val
= vcpu
->halt_poll_ns
;
2132 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2138 vcpu
->halt_poll_ns
= val
;
2139 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2142 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2144 if (kvm_arch_vcpu_runnable(vcpu
)) {
2145 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2148 if (kvm_cpu_has_pending_timer(vcpu
))
2150 if (signal_pending(current
))
2157 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2159 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2162 DECLARE_SWAITQUEUE(wait
);
2163 bool waited
= false;
2166 start
= cur
= ktime_get();
2167 if (vcpu
->halt_poll_ns
) {
2168 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2170 ++vcpu
->stat
.halt_attempted_poll
;
2173 * This sets KVM_REQ_UNHALT if an interrupt
2176 if (kvm_vcpu_check_block(vcpu
) < 0) {
2177 ++vcpu
->stat
.halt_successful_poll
;
2178 if (!vcpu_valid_wakeup(vcpu
))
2179 ++vcpu
->stat
.halt_poll_invalid
;
2183 } while (single_task_running() && ktime_before(cur
, stop
));
2186 kvm_arch_vcpu_blocking(vcpu
);
2189 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2191 if (kvm_vcpu_check_block(vcpu
) < 0)
2198 finish_swait(&vcpu
->wq
, &wait
);
2201 kvm_arch_vcpu_unblocking(vcpu
);
2203 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2205 if (!vcpu_valid_wakeup(vcpu
))
2206 shrink_halt_poll_ns(vcpu
);
2207 else if (halt_poll_ns
) {
2208 if (block_ns
<= vcpu
->halt_poll_ns
)
2210 /* we had a long block, shrink polling */
2211 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2212 shrink_halt_poll_ns(vcpu
);
2213 /* we had a short halt and our poll time is too small */
2214 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2215 block_ns
< halt_poll_ns
)
2216 grow_halt_poll_ns(vcpu
);
2218 vcpu
->halt_poll_ns
= 0;
2220 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2221 kvm_arch_vcpu_block_finish(vcpu
);
2223 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2225 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2227 struct swait_queue_head
*wqp
;
2229 wqp
= kvm_arch_vcpu_wq(vcpu
);
2230 if (swait_active(wqp
)) {
2232 ++vcpu
->stat
.halt_wakeup
;
2238 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2242 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2244 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2247 int cpu
= vcpu
->cpu
;
2249 if (kvm_vcpu_wake_up(vcpu
))
2253 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2254 if (kvm_arch_vcpu_should_kick(vcpu
))
2255 smp_send_reschedule(cpu
);
2258 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2259 #endif /* !CONFIG_S390 */
2261 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2264 struct task_struct
*task
= NULL
;
2268 pid
= rcu_dereference(target
->pid
);
2270 task
= get_pid_task(pid
, PIDTYPE_PID
);
2274 ret
= yield_to(task
, 1);
2275 put_task_struct(task
);
2279 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2282 * Helper that checks whether a VCPU is eligible for directed yield.
2283 * Most eligible candidate to yield is decided by following heuristics:
2285 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2286 * (preempted lock holder), indicated by @in_spin_loop.
2287 * Set at the beiginning and cleared at the end of interception/PLE handler.
2289 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2290 * chance last time (mostly it has become eligible now since we have probably
2291 * yielded to lockholder in last iteration. This is done by toggling
2292 * @dy_eligible each time a VCPU checked for eligibility.)
2294 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2295 * to preempted lock-holder could result in wrong VCPU selection and CPU
2296 * burning. Giving priority for a potential lock-holder increases lock
2299 * Since algorithm is based on heuristics, accessing another VCPU data without
2300 * locking does not harm. It may result in trying to yield to same VCPU, fail
2301 * and continue with next VCPU and so on.
2303 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2305 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2308 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2309 vcpu
->spin_loop
.dy_eligible
;
2311 if (vcpu
->spin_loop
.in_spin_loop
)
2312 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2320 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2322 struct kvm
*kvm
= me
->kvm
;
2323 struct kvm_vcpu
*vcpu
;
2324 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2330 kvm_vcpu_set_in_spin_loop(me
, true);
2332 * We boost the priority of a VCPU that is runnable but not
2333 * currently running, because it got preempted by something
2334 * else and called schedule in __vcpu_run. Hopefully that
2335 * VCPU is holding the lock that we need and will release it.
2336 * We approximate round-robin by starting at the last boosted VCPU.
2338 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2339 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2340 if (!pass
&& i
<= last_boosted_vcpu
) {
2341 i
= last_boosted_vcpu
;
2343 } else if (pass
&& i
> last_boosted_vcpu
)
2345 if (!ACCESS_ONCE(vcpu
->preempted
))
2349 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2351 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2354 yielded
= kvm_vcpu_yield_to(vcpu
);
2356 kvm
->last_boosted_vcpu
= i
;
2358 } else if (yielded
< 0) {
2365 kvm_vcpu_set_in_spin_loop(me
, false);
2367 /* Ensure vcpu is not eligible during next spinloop */
2368 kvm_vcpu_set_dy_eligible(me
, false);
2370 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2372 static int kvm_vcpu_fault(struct vm_fault
*vmf
)
2374 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2377 if (vmf
->pgoff
== 0)
2378 page
= virt_to_page(vcpu
->run
);
2380 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2381 page
= virt_to_page(vcpu
->arch
.pio_data
);
2383 #ifdef CONFIG_KVM_MMIO
2384 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2385 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2388 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2394 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2395 .fault
= kvm_vcpu_fault
,
2398 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2400 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2404 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2406 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2408 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2409 kvm_put_kvm(vcpu
->kvm
);
2413 static struct file_operations kvm_vcpu_fops
= {
2414 .release
= kvm_vcpu_release
,
2415 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2416 #ifdef CONFIG_KVM_COMPAT
2417 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2419 .mmap
= kvm_vcpu_mmap
,
2420 .llseek
= noop_llseek
,
2424 * Allocates an inode for the vcpu.
2426 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2428 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2431 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2433 char dir_name
[ITOA_MAX_LEN
* 2];
2436 if (!kvm_arch_has_vcpu_debugfs())
2439 if (!debugfs_initialized())
2442 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2443 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2444 vcpu
->kvm
->debugfs_dentry
);
2445 if (!vcpu
->debugfs_dentry
)
2448 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2450 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2458 * Creates some virtual cpus. Good luck creating more than one.
2460 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2463 struct kvm_vcpu
*vcpu
;
2465 if (id
>= KVM_MAX_VCPU_ID
)
2468 mutex_lock(&kvm
->lock
);
2469 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2470 mutex_unlock(&kvm
->lock
);
2474 kvm
->created_vcpus
++;
2475 mutex_unlock(&kvm
->lock
);
2477 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2480 goto vcpu_decrement
;
2483 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2485 r
= kvm_arch_vcpu_setup(vcpu
);
2489 r
= kvm_create_vcpu_debugfs(vcpu
);
2493 mutex_lock(&kvm
->lock
);
2494 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2496 goto unlock_vcpu_destroy
;
2499 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2501 /* Now it's all set up, let userspace reach it */
2503 r
= create_vcpu_fd(vcpu
);
2506 goto unlock_vcpu_destroy
;
2509 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2512 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2513 * before kvm->online_vcpu's incremented value.
2516 atomic_inc(&kvm
->online_vcpus
);
2518 mutex_unlock(&kvm
->lock
);
2519 kvm_arch_vcpu_postcreate(vcpu
);
2522 unlock_vcpu_destroy
:
2523 mutex_unlock(&kvm
->lock
);
2524 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2526 kvm_arch_vcpu_destroy(vcpu
);
2528 mutex_lock(&kvm
->lock
);
2529 kvm
->created_vcpus
--;
2530 mutex_unlock(&kvm
->lock
);
2534 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2537 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2538 vcpu
->sigset_active
= 1;
2539 vcpu
->sigset
= *sigset
;
2541 vcpu
->sigset_active
= 0;
2545 static long kvm_vcpu_ioctl(struct file
*filp
,
2546 unsigned int ioctl
, unsigned long arg
)
2548 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2549 void __user
*argp
= (void __user
*)arg
;
2551 struct kvm_fpu
*fpu
= NULL
;
2552 struct kvm_sregs
*kvm_sregs
= NULL
;
2554 if (vcpu
->kvm
->mm
!= current
->mm
)
2557 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2560 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2562 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2563 * so vcpu_load() would break it.
2565 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2566 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2570 r
= vcpu_load(vcpu
);
2579 oldpid
= rcu_access_pointer(vcpu
->pid
);
2580 if (unlikely(oldpid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2581 /* The thread running this VCPU changed. */
2582 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2584 rcu_assign_pointer(vcpu
->pid
, newpid
);
2589 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2590 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2593 case KVM_GET_REGS
: {
2594 struct kvm_regs
*kvm_regs
;
2597 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2600 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2604 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2611 case KVM_SET_REGS
: {
2612 struct kvm_regs
*kvm_regs
;
2615 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2616 if (IS_ERR(kvm_regs
)) {
2617 r
= PTR_ERR(kvm_regs
);
2620 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2624 case KVM_GET_SREGS
: {
2625 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2629 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2633 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2638 case KVM_SET_SREGS
: {
2639 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2640 if (IS_ERR(kvm_sregs
)) {
2641 r
= PTR_ERR(kvm_sregs
);
2645 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2648 case KVM_GET_MP_STATE
: {
2649 struct kvm_mp_state mp_state
;
2651 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2655 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2660 case KVM_SET_MP_STATE
: {
2661 struct kvm_mp_state mp_state
;
2664 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2666 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2669 case KVM_TRANSLATE
: {
2670 struct kvm_translation tr
;
2673 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2675 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2679 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2684 case KVM_SET_GUEST_DEBUG
: {
2685 struct kvm_guest_debug dbg
;
2688 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2690 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2693 case KVM_SET_SIGNAL_MASK
: {
2694 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2695 struct kvm_signal_mask kvm_sigmask
;
2696 sigset_t sigset
, *p
;
2701 if (copy_from_user(&kvm_sigmask
, argp
,
2702 sizeof(kvm_sigmask
)))
2705 if (kvm_sigmask
.len
!= sizeof(sigset
))
2708 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2713 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2717 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2721 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2725 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2731 fpu
= memdup_user(argp
, sizeof(*fpu
));
2737 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2741 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2750 #ifdef CONFIG_KVM_COMPAT
2751 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2752 unsigned int ioctl
, unsigned long arg
)
2754 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2755 void __user
*argp
= compat_ptr(arg
);
2758 if (vcpu
->kvm
->mm
!= current
->mm
)
2762 case KVM_SET_SIGNAL_MASK
: {
2763 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2764 struct kvm_signal_mask kvm_sigmask
;
2765 compat_sigset_t csigset
;
2770 if (copy_from_user(&kvm_sigmask
, argp
,
2771 sizeof(kvm_sigmask
)))
2774 if (kvm_sigmask
.len
!= sizeof(csigset
))
2777 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2780 sigset_from_compat(&sigset
, &csigset
);
2781 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2783 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2787 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2795 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2796 int (*accessor
)(struct kvm_device
*dev
,
2797 struct kvm_device_attr
*attr
),
2800 struct kvm_device_attr attr
;
2805 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2808 return accessor(dev
, &attr
);
2811 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2814 struct kvm_device
*dev
= filp
->private_data
;
2817 case KVM_SET_DEVICE_ATTR
:
2818 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2819 case KVM_GET_DEVICE_ATTR
:
2820 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2821 case KVM_HAS_DEVICE_ATTR
:
2822 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2824 if (dev
->ops
->ioctl
)
2825 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2831 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2833 struct kvm_device
*dev
= filp
->private_data
;
2834 struct kvm
*kvm
= dev
->kvm
;
2840 static const struct file_operations kvm_device_fops
= {
2841 .unlocked_ioctl
= kvm_device_ioctl
,
2842 #ifdef CONFIG_KVM_COMPAT
2843 .compat_ioctl
= kvm_device_ioctl
,
2845 .release
= kvm_device_release
,
2848 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2850 if (filp
->f_op
!= &kvm_device_fops
)
2853 return filp
->private_data
;
2856 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2857 #ifdef CONFIG_KVM_MPIC
2858 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2859 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2863 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2865 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2868 if (kvm_device_ops_table
[type
] != NULL
)
2871 kvm_device_ops_table
[type
] = ops
;
2875 void kvm_unregister_device_ops(u32 type
)
2877 if (kvm_device_ops_table
[type
] != NULL
)
2878 kvm_device_ops_table
[type
] = NULL
;
2881 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2882 struct kvm_create_device
*cd
)
2884 struct kvm_device_ops
*ops
= NULL
;
2885 struct kvm_device
*dev
;
2886 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2889 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2892 ops
= kvm_device_ops_table
[cd
->type
];
2899 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2906 mutex_lock(&kvm
->lock
);
2907 ret
= ops
->create(dev
, cd
->type
);
2909 mutex_unlock(&kvm
->lock
);
2913 list_add(&dev
->vm_node
, &kvm
->devices
);
2914 mutex_unlock(&kvm
->lock
);
2919 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2921 mutex_lock(&kvm
->lock
);
2922 list_del(&dev
->vm_node
);
2923 mutex_unlock(&kvm
->lock
);
2933 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2936 case KVM_CAP_USER_MEMORY
:
2937 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2938 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2939 case KVM_CAP_INTERNAL_ERROR_DATA
:
2940 #ifdef CONFIG_HAVE_KVM_MSI
2941 case KVM_CAP_SIGNAL_MSI
:
2943 #ifdef CONFIG_HAVE_KVM_IRQFD
2945 case KVM_CAP_IRQFD_RESAMPLE
:
2947 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2948 case KVM_CAP_CHECK_EXTENSION_VM
:
2950 #ifdef CONFIG_KVM_MMIO
2951 case KVM_CAP_COALESCED_MMIO
:
2952 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
2954 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2955 case KVM_CAP_IRQ_ROUTING
:
2956 return KVM_MAX_IRQ_ROUTES
;
2958 #if KVM_ADDRESS_SPACE_NUM > 1
2959 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2960 return KVM_ADDRESS_SPACE_NUM
;
2962 case KVM_CAP_MAX_VCPU_ID
:
2963 return KVM_MAX_VCPU_ID
;
2967 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2970 static long kvm_vm_ioctl(struct file
*filp
,
2971 unsigned int ioctl
, unsigned long arg
)
2973 struct kvm
*kvm
= filp
->private_data
;
2974 void __user
*argp
= (void __user
*)arg
;
2977 if (kvm
->mm
!= current
->mm
)
2980 case KVM_CREATE_VCPU
:
2981 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2983 case KVM_SET_USER_MEMORY_REGION
: {
2984 struct kvm_userspace_memory_region kvm_userspace_mem
;
2987 if (copy_from_user(&kvm_userspace_mem
, argp
,
2988 sizeof(kvm_userspace_mem
)))
2991 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2994 case KVM_GET_DIRTY_LOG
: {
2995 struct kvm_dirty_log log
;
2998 if (copy_from_user(&log
, argp
, sizeof(log
)))
3000 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3003 #ifdef CONFIG_KVM_MMIO
3004 case KVM_REGISTER_COALESCED_MMIO
: {
3005 struct kvm_coalesced_mmio_zone zone
;
3008 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3010 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3013 case KVM_UNREGISTER_COALESCED_MMIO
: {
3014 struct kvm_coalesced_mmio_zone zone
;
3017 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3019 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3024 struct kvm_irqfd data
;
3027 if (copy_from_user(&data
, argp
, sizeof(data
)))
3029 r
= kvm_irqfd(kvm
, &data
);
3032 case KVM_IOEVENTFD
: {
3033 struct kvm_ioeventfd data
;
3036 if (copy_from_user(&data
, argp
, sizeof(data
)))
3038 r
= kvm_ioeventfd(kvm
, &data
);
3041 #ifdef CONFIG_HAVE_KVM_MSI
3042 case KVM_SIGNAL_MSI
: {
3046 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3048 r
= kvm_send_userspace_msi(kvm
, &msi
);
3052 #ifdef __KVM_HAVE_IRQ_LINE
3053 case KVM_IRQ_LINE_STATUS
:
3054 case KVM_IRQ_LINE
: {
3055 struct kvm_irq_level irq_event
;
3058 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3061 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3062 ioctl
== KVM_IRQ_LINE_STATUS
);
3067 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3068 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3076 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3077 case KVM_SET_GSI_ROUTING
: {
3078 struct kvm_irq_routing routing
;
3079 struct kvm_irq_routing __user
*urouting
;
3080 struct kvm_irq_routing_entry
*entries
= NULL
;
3083 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3086 if (!kvm_arch_can_set_irq_routing(kvm
))
3088 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3094 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3099 if (copy_from_user(entries
, urouting
->entries
,
3100 routing
.nr
* sizeof(*entries
)))
3101 goto out_free_irq_routing
;
3103 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3105 out_free_irq_routing
:
3109 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3110 case KVM_CREATE_DEVICE
: {
3111 struct kvm_create_device cd
;
3114 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3117 r
= kvm_ioctl_create_device(kvm
, &cd
);
3122 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3128 case KVM_CHECK_EXTENSION
:
3129 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3132 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3138 #ifdef CONFIG_KVM_COMPAT
3139 struct compat_kvm_dirty_log
{
3143 compat_uptr_t dirty_bitmap
; /* one bit per page */
3148 static long kvm_vm_compat_ioctl(struct file
*filp
,
3149 unsigned int ioctl
, unsigned long arg
)
3151 struct kvm
*kvm
= filp
->private_data
;
3154 if (kvm
->mm
!= current
->mm
)
3157 case KVM_GET_DIRTY_LOG
: {
3158 struct compat_kvm_dirty_log compat_log
;
3159 struct kvm_dirty_log log
;
3161 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3162 sizeof(compat_log
)))
3164 log
.slot
= compat_log
.slot
;
3165 log
.padding1
= compat_log
.padding1
;
3166 log
.padding2
= compat_log
.padding2
;
3167 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3169 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3173 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3179 static struct file_operations kvm_vm_fops
= {
3180 .release
= kvm_vm_release
,
3181 .unlocked_ioctl
= kvm_vm_ioctl
,
3182 #ifdef CONFIG_KVM_COMPAT
3183 .compat_ioctl
= kvm_vm_compat_ioctl
,
3185 .llseek
= noop_llseek
,
3188 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3194 kvm
= kvm_create_vm(type
);
3196 return PTR_ERR(kvm
);
3197 #ifdef CONFIG_KVM_MMIO
3198 r
= kvm_coalesced_mmio_init(kvm
);
3204 r
= get_unused_fd_flags(O_CLOEXEC
);
3209 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3213 return PTR_ERR(file
);
3217 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3218 * already set, with ->release() being kvm_vm_release(). In error
3219 * cases it will be called by the final fput(file) and will take
3220 * care of doing kvm_put_kvm(kvm).
3222 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3227 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3229 fd_install(r
, file
);
3233 static long kvm_dev_ioctl(struct file
*filp
,
3234 unsigned int ioctl
, unsigned long arg
)
3239 case KVM_GET_API_VERSION
:
3242 r
= KVM_API_VERSION
;
3245 r
= kvm_dev_ioctl_create_vm(arg
);
3247 case KVM_CHECK_EXTENSION
:
3248 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3250 case KVM_GET_VCPU_MMAP_SIZE
:
3253 r
= PAGE_SIZE
; /* struct kvm_run */
3255 r
+= PAGE_SIZE
; /* pio data page */
3257 #ifdef CONFIG_KVM_MMIO
3258 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3261 case KVM_TRACE_ENABLE
:
3262 case KVM_TRACE_PAUSE
:
3263 case KVM_TRACE_DISABLE
:
3267 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3273 static struct file_operations kvm_chardev_ops
= {
3274 .unlocked_ioctl
= kvm_dev_ioctl
,
3275 .compat_ioctl
= kvm_dev_ioctl
,
3276 .llseek
= noop_llseek
,
3279 static struct miscdevice kvm_dev
= {
3285 static void hardware_enable_nolock(void *junk
)
3287 int cpu
= raw_smp_processor_id();
3290 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3293 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3295 r
= kvm_arch_hardware_enable();
3298 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3299 atomic_inc(&hardware_enable_failed
);
3300 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3304 static int kvm_starting_cpu(unsigned int cpu
)
3306 raw_spin_lock(&kvm_count_lock
);
3307 if (kvm_usage_count
)
3308 hardware_enable_nolock(NULL
);
3309 raw_spin_unlock(&kvm_count_lock
);
3313 static void hardware_disable_nolock(void *junk
)
3315 int cpu
= raw_smp_processor_id();
3317 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3319 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3320 kvm_arch_hardware_disable();
3323 static int kvm_dying_cpu(unsigned int cpu
)
3325 raw_spin_lock(&kvm_count_lock
);
3326 if (kvm_usage_count
)
3327 hardware_disable_nolock(NULL
);
3328 raw_spin_unlock(&kvm_count_lock
);
3332 static void hardware_disable_all_nolock(void)
3334 BUG_ON(!kvm_usage_count
);
3337 if (!kvm_usage_count
)
3338 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3341 static void hardware_disable_all(void)
3343 raw_spin_lock(&kvm_count_lock
);
3344 hardware_disable_all_nolock();
3345 raw_spin_unlock(&kvm_count_lock
);
3348 static int hardware_enable_all(void)
3352 raw_spin_lock(&kvm_count_lock
);
3355 if (kvm_usage_count
== 1) {
3356 atomic_set(&hardware_enable_failed
, 0);
3357 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3359 if (atomic_read(&hardware_enable_failed
)) {
3360 hardware_disable_all_nolock();
3365 raw_spin_unlock(&kvm_count_lock
);
3370 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3374 * Some (well, at least mine) BIOSes hang on reboot if
3377 * And Intel TXT required VMX off for all cpu when system shutdown.
3379 pr_info("kvm: exiting hardware virtualization\n");
3380 kvm_rebooting
= true;
3381 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3385 static struct notifier_block kvm_reboot_notifier
= {
3386 .notifier_call
= kvm_reboot
,
3390 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3394 for (i
= 0; i
< bus
->dev_count
; i
++) {
3395 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3397 kvm_iodevice_destructor(pos
);
3402 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3403 const struct kvm_io_range
*r2
)
3405 gpa_t addr1
= r1
->addr
;
3406 gpa_t addr2
= r2
->addr
;
3411 /* If r2->len == 0, match the exact address. If r2->len != 0,
3412 * accept any overlapping write. Any order is acceptable for
3413 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3414 * we process all of them.
3427 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3429 return kvm_io_bus_cmp(p1
, p2
);
3432 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3433 gpa_t addr
, int len
)
3435 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3441 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3442 kvm_io_bus_sort_cmp
, NULL
);
3447 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3448 gpa_t addr
, int len
)
3450 struct kvm_io_range
*range
, key
;
3453 key
= (struct kvm_io_range
) {
3458 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3459 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3463 off
= range
- bus
->range
;
3465 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3471 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3472 struct kvm_io_range
*range
, const void *val
)
3476 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3480 while (idx
< bus
->dev_count
&&
3481 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3482 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3491 /* kvm_io_bus_write - called under kvm->slots_lock */
3492 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3493 int len
, const void *val
)
3495 struct kvm_io_bus
*bus
;
3496 struct kvm_io_range range
;
3499 range
= (struct kvm_io_range
) {
3504 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3507 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3508 return r
< 0 ? r
: 0;
3511 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3512 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3513 gpa_t addr
, int len
, const void *val
, long cookie
)
3515 struct kvm_io_bus
*bus
;
3516 struct kvm_io_range range
;
3518 range
= (struct kvm_io_range
) {
3523 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3527 /* First try the device referenced by cookie. */
3528 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3529 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3530 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3535 * cookie contained garbage; fall back to search and return the
3536 * correct cookie value.
3538 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3541 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3542 struct kvm_io_range
*range
, void *val
)
3546 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3550 while (idx
< bus
->dev_count
&&
3551 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3552 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3560 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3562 /* kvm_io_bus_read - called under kvm->slots_lock */
3563 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3566 struct kvm_io_bus
*bus
;
3567 struct kvm_io_range range
;
3570 range
= (struct kvm_io_range
) {
3575 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3578 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3579 return r
< 0 ? r
: 0;
3583 /* Caller must hold slots_lock. */
3584 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3585 int len
, struct kvm_io_device
*dev
)
3587 struct kvm_io_bus
*new_bus
, *bus
;
3589 bus
= kvm_get_bus(kvm
, bus_idx
);
3593 /* exclude ioeventfd which is limited by maximum fd */
3594 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3597 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3598 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3601 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3602 sizeof(struct kvm_io_range
)));
3603 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3604 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3605 synchronize_srcu_expedited(&kvm
->srcu
);
3611 /* Caller must hold slots_lock. */
3612 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3613 struct kvm_io_device
*dev
)
3616 struct kvm_io_bus
*new_bus
, *bus
;
3618 bus
= kvm_get_bus(kvm
, bus_idx
);
3622 for (i
= 0; i
< bus
->dev_count
; i
++)
3623 if (bus
->range
[i
].dev
== dev
) {
3627 if (i
== bus
->dev_count
)
3630 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3631 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3633 pr_err("kvm: failed to shrink bus, removing it completely\n");
3637 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3638 new_bus
->dev_count
--;
3639 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3640 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3643 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3644 synchronize_srcu_expedited(&kvm
->srcu
);
3649 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3652 struct kvm_io_bus
*bus
;
3653 int dev_idx
, srcu_idx
;
3654 struct kvm_io_device
*iodev
= NULL
;
3656 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3658 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3662 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3666 iodev
= bus
->range
[dev_idx
].dev
;
3669 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3673 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3675 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3676 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3679 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3682 /* The debugfs files are a reference to the kvm struct which
3683 * is still valid when kvm_destroy_vm is called.
3684 * To avoid the race between open and the removal of the debugfs
3685 * directory we test against the users count.
3687 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3690 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3691 kvm_put_kvm(stat_data
->kvm
);
3698 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3700 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3703 simple_attr_release(inode
, file
);
3704 kvm_put_kvm(stat_data
->kvm
);
3709 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3711 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3713 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3718 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3720 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3725 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3730 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3732 __simple_attr_check_format("%llu\n", 0ull);
3733 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3734 vm_stat_clear_per_vm
, "%llu\n");
3737 static const struct file_operations vm_stat_get_per_vm_fops
= {
3738 .owner
= THIS_MODULE
,
3739 .open
= vm_stat_get_per_vm_open
,
3740 .release
= kvm_debugfs_release
,
3741 .read
= simple_attr_read
,
3742 .write
= simple_attr_write
,
3743 .llseek
= no_llseek
,
3746 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3749 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3750 struct kvm_vcpu
*vcpu
;
3754 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3755 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3760 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3763 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3764 struct kvm_vcpu
*vcpu
;
3769 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3770 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3775 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3777 __simple_attr_check_format("%llu\n", 0ull);
3778 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3779 vcpu_stat_clear_per_vm
, "%llu\n");
3782 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3783 .owner
= THIS_MODULE
,
3784 .open
= vcpu_stat_get_per_vm_open
,
3785 .release
= kvm_debugfs_release
,
3786 .read
= simple_attr_read
,
3787 .write
= simple_attr_write
,
3788 .llseek
= no_llseek
,
3791 static const struct file_operations
*stat_fops_per_vm
[] = {
3792 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3793 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3796 static int vm_stat_get(void *_offset
, u64
*val
)
3798 unsigned offset
= (long)_offset
;
3800 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3804 spin_lock(&kvm_lock
);
3805 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3807 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3810 spin_unlock(&kvm_lock
);
3814 static int vm_stat_clear(void *_offset
, u64 val
)
3816 unsigned offset
= (long)_offset
;
3818 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3823 spin_lock(&kvm_lock
);
3824 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3826 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3828 spin_unlock(&kvm_lock
);
3833 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3835 static int vcpu_stat_get(void *_offset
, u64
*val
)
3837 unsigned offset
= (long)_offset
;
3839 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3843 spin_lock(&kvm_lock
);
3844 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3846 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3849 spin_unlock(&kvm_lock
);
3853 static int vcpu_stat_clear(void *_offset
, u64 val
)
3855 unsigned offset
= (long)_offset
;
3857 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3862 spin_lock(&kvm_lock
);
3863 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3865 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
3867 spin_unlock(&kvm_lock
);
3872 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
3875 static const struct file_operations
*stat_fops
[] = {
3876 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3877 [KVM_STAT_VM
] = &vm_stat_fops
,
3880 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
3882 struct kobj_uevent_env
*env
;
3883 unsigned long long created
, active
;
3885 if (!kvm_dev
.this_device
|| !kvm
)
3888 spin_lock(&kvm_lock
);
3889 if (type
== KVM_EVENT_CREATE_VM
) {
3890 kvm_createvm_count
++;
3892 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3895 created
= kvm_createvm_count
;
3896 active
= kvm_active_vms
;
3897 spin_unlock(&kvm_lock
);
3899 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
3903 add_uevent_var(env
, "CREATED=%llu", created
);
3904 add_uevent_var(env
, "COUNT=%llu", active
);
3906 if (type
== KVM_EVENT_CREATE_VM
) {
3907 add_uevent_var(env
, "EVENT=create");
3908 kvm
->userspace_pid
= task_pid_nr(current
);
3909 } else if (type
== KVM_EVENT_DESTROY_VM
) {
3910 add_uevent_var(env
, "EVENT=destroy");
3912 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
3914 if (kvm
->debugfs_dentry
) {
3915 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3918 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
3920 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
3924 /* no need for checks, since we are adding at most only 5 keys */
3925 env
->envp
[env
->envp_idx
++] = NULL
;
3926 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
3930 static int kvm_init_debug(void)
3933 struct kvm_stats_debugfs_item
*p
;
3935 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3936 if (kvm_debugfs_dir
== NULL
)
3939 kvm_debugfs_num_entries
= 0;
3940 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3941 if (!debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
3942 (void *)(long)p
->offset
,
3943 stat_fops
[p
->kind
]))
3950 debugfs_remove_recursive(kvm_debugfs_dir
);
3955 static int kvm_suspend(void)
3957 if (kvm_usage_count
)
3958 hardware_disable_nolock(NULL
);
3962 static void kvm_resume(void)
3964 if (kvm_usage_count
) {
3965 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3966 hardware_enable_nolock(NULL
);
3970 static struct syscore_ops kvm_syscore_ops
= {
3971 .suspend
= kvm_suspend
,
3972 .resume
= kvm_resume
,
3976 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3978 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3981 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3983 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3985 if (vcpu
->preempted
)
3986 vcpu
->preempted
= false;
3988 kvm_arch_sched_in(vcpu
, cpu
);
3990 kvm_arch_vcpu_load(vcpu
, cpu
);
3993 static void kvm_sched_out(struct preempt_notifier
*pn
,
3994 struct task_struct
*next
)
3996 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3998 if (current
->state
== TASK_RUNNING
)
3999 vcpu
->preempted
= true;
4000 kvm_arch_vcpu_put(vcpu
);
4003 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4004 struct module
*module
)
4009 r
= kvm_arch_init(opaque
);
4014 * kvm_arch_init makes sure there's at most one caller
4015 * for architectures that support multiple implementations,
4016 * like intel and amd on x86.
4017 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4018 * conflicts in case kvm is already setup for another implementation.
4020 r
= kvm_irqfd_init();
4024 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4029 r
= kvm_arch_hardware_setup();
4033 for_each_online_cpu(cpu
) {
4034 smp_call_function_single(cpu
,
4035 kvm_arch_check_processor_compat
,
4041 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4042 kvm_starting_cpu
, kvm_dying_cpu
);
4045 register_reboot_notifier(&kvm_reboot_notifier
);
4047 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4049 vcpu_align
= __alignof__(struct kvm_vcpu
);
4050 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
4052 if (!kvm_vcpu_cache
) {
4057 r
= kvm_async_pf_init();
4061 kvm_chardev_ops
.owner
= module
;
4062 kvm_vm_fops
.owner
= module
;
4063 kvm_vcpu_fops
.owner
= module
;
4065 r
= misc_register(&kvm_dev
);
4067 pr_err("kvm: misc device register failed\n");
4071 register_syscore_ops(&kvm_syscore_ops
);
4073 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4074 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4076 r
= kvm_init_debug();
4078 pr_err("kvm: create debugfs files failed\n");
4082 r
= kvm_vfio_ops_init();
4088 unregister_syscore_ops(&kvm_syscore_ops
);
4089 misc_deregister(&kvm_dev
);
4091 kvm_async_pf_deinit();
4093 kmem_cache_destroy(kvm_vcpu_cache
);
4095 unregister_reboot_notifier(&kvm_reboot_notifier
);
4096 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4099 kvm_arch_hardware_unsetup();
4101 free_cpumask_var(cpus_hardware_enabled
);
4109 EXPORT_SYMBOL_GPL(kvm_init
);
4113 debugfs_remove_recursive(kvm_debugfs_dir
);
4114 misc_deregister(&kvm_dev
);
4115 kmem_cache_destroy(kvm_vcpu_cache
);
4116 kvm_async_pf_deinit();
4117 unregister_syscore_ops(&kvm_syscore_ops
);
4118 unregister_reboot_notifier(&kvm_reboot_notifier
);
4119 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4120 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4121 kvm_arch_hardware_unsetup();
4124 free_cpumask_var(cpus_hardware_enabled
);
4125 kvm_vfio_ops_exit();
4127 EXPORT_SYMBOL_GPL(kvm_exit
);