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
,
119 #define KVM_COMPAT(c) .compat_ioctl = (c)
121 static long kvm_no_compat_ioctl(struct file
*file
, unsigned int ioctl
,
122 unsigned long arg
) { return -EINVAL
; }
123 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
125 static int hardware_enable_all(void);
126 static void hardware_disable_all(void);
128 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
130 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
132 __visible
bool kvm_rebooting
;
133 EXPORT_SYMBOL_GPL(kvm_rebooting
);
135 static bool largepages_enabled
= true;
137 #define KVM_EVENT_CREATE_VM 0
138 #define KVM_EVENT_DESTROY_VM 1
139 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
140 static unsigned long long kvm_createvm_count
;
141 static unsigned long long kvm_active_vms
;
143 __weak
int kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
144 unsigned long start
, unsigned long end
, bool blockable
)
149 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
152 return PageReserved(pfn_to_page(pfn
));
158 * Switches to specified vcpu, until a matching vcpu_put()
160 void vcpu_load(struct kvm_vcpu
*vcpu
)
163 preempt_notifier_register(&vcpu
->preempt_notifier
);
164 kvm_arch_vcpu_load(vcpu
, cpu
);
167 EXPORT_SYMBOL_GPL(vcpu_load
);
169 void vcpu_put(struct kvm_vcpu
*vcpu
)
172 kvm_arch_vcpu_put(vcpu
);
173 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
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_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
213 unsigned long *vcpu_bitmap
, cpumask_var_t tmp
)
216 struct kvm_vcpu
*vcpu
;
221 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
222 if (vcpu_bitmap
&& !test_bit(i
, vcpu_bitmap
))
225 kvm_make_request(req
, vcpu
);
228 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
231 if (tmp
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
232 kvm_request_needs_ipi(vcpu
, req
))
233 __cpumask_set_cpu(cpu
, tmp
);
236 called
= kvm_kick_many_cpus(tmp
, !!(req
& KVM_REQUEST_WAIT
));
242 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
247 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
249 called
= kvm_make_vcpus_request_mask(kvm
, req
, NULL
, cpus
);
251 free_cpumask_var(cpus
);
255 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
256 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
259 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
260 * kvm_make_all_cpus_request.
262 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
265 * We want to publish modifications to the page tables before reading
266 * mode. Pairs with a memory barrier in arch-specific code.
267 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
268 * and smp_mb in walk_shadow_page_lockless_begin/end.
269 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
271 * There is already an smp_mb__after_atomic() before
272 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
275 if (!kvm_arch_flush_remote_tlb(kvm
)
276 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
277 ++kvm
->stat
.remote_tlb_flush
;
278 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
280 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
283 void kvm_reload_remote_mmus(struct kvm
*kvm
)
285 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
288 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
293 mutex_init(&vcpu
->mutex
);
298 init_swait_queue_head(&vcpu
->wq
);
299 kvm_async_pf_vcpu_init(vcpu
);
302 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
304 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
309 vcpu
->run
= page_address(page
);
311 kvm_vcpu_set_in_spin_loop(vcpu
, false);
312 kvm_vcpu_set_dy_eligible(vcpu
, false);
313 vcpu
->preempted
= false;
315 r
= kvm_arch_vcpu_init(vcpu
);
321 free_page((unsigned long)vcpu
->run
);
325 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
327 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
330 * no need for rcu_read_lock as VCPU_RUN is the only place that
331 * will change the vcpu->pid pointer and on uninit all file
332 * descriptors are already gone.
334 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
335 kvm_arch_vcpu_uninit(vcpu
);
336 free_page((unsigned long)vcpu
->run
);
338 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
340 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
341 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
343 return container_of(mn
, struct kvm
, mmu_notifier
);
346 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
347 struct mm_struct
*mm
,
348 unsigned long address
,
351 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
354 idx
= srcu_read_lock(&kvm
->srcu
);
355 spin_lock(&kvm
->mmu_lock
);
356 kvm
->mmu_notifier_seq
++;
358 if (kvm_set_spte_hva(kvm
, address
, pte
))
359 kvm_flush_remote_tlbs(kvm
);
361 spin_unlock(&kvm
->mmu_lock
);
362 srcu_read_unlock(&kvm
->srcu
, idx
);
365 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
366 const struct mmu_notifier_range
*range
)
368 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
369 int need_tlb_flush
= 0, idx
;
372 idx
= srcu_read_lock(&kvm
->srcu
);
373 spin_lock(&kvm
->mmu_lock
);
375 * The count increase must become visible at unlock time as no
376 * spte can be established without taking the mmu_lock and
377 * count is also read inside the mmu_lock critical section.
379 kvm
->mmu_notifier_count
++;
380 need_tlb_flush
= kvm_unmap_hva_range(kvm
, range
->start
, range
->end
);
381 need_tlb_flush
|= kvm
->tlbs_dirty
;
382 /* we've to flush the tlb before the pages can be freed */
384 kvm_flush_remote_tlbs(kvm
);
386 spin_unlock(&kvm
->mmu_lock
);
388 ret
= kvm_arch_mmu_notifier_invalidate_range(kvm
, range
->start
,
389 range
->end
, range
->blockable
);
391 srcu_read_unlock(&kvm
->srcu
, idx
);
396 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
397 const struct mmu_notifier_range
*range
)
399 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
401 spin_lock(&kvm
->mmu_lock
);
403 * This sequence increase will notify the kvm page fault that
404 * the page that is going to be mapped in the spte could have
407 kvm
->mmu_notifier_seq
++;
410 * The above sequence increase must be visible before the
411 * below count decrease, which is ensured by the smp_wmb above
412 * in conjunction with the smp_rmb in mmu_notifier_retry().
414 kvm
->mmu_notifier_count
--;
415 spin_unlock(&kvm
->mmu_lock
);
417 BUG_ON(kvm
->mmu_notifier_count
< 0);
420 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
421 struct mm_struct
*mm
,
425 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
428 idx
= srcu_read_lock(&kvm
->srcu
);
429 spin_lock(&kvm
->mmu_lock
);
431 young
= kvm_age_hva(kvm
, start
, end
);
433 kvm_flush_remote_tlbs(kvm
);
435 spin_unlock(&kvm
->mmu_lock
);
436 srcu_read_unlock(&kvm
->srcu
, idx
);
441 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
442 struct mm_struct
*mm
,
446 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
449 idx
= srcu_read_lock(&kvm
->srcu
);
450 spin_lock(&kvm
->mmu_lock
);
452 * Even though we do not flush TLB, this will still adversely
453 * affect performance on pre-Haswell Intel EPT, where there is
454 * no EPT Access Bit to clear so that we have to tear down EPT
455 * tables instead. If we find this unacceptable, we can always
456 * add a parameter to kvm_age_hva so that it effectively doesn't
457 * do anything on clear_young.
459 * Also note that currently we never issue secondary TLB flushes
460 * from clear_young, leaving this job up to the regular system
461 * cadence. If we find this inaccurate, we might come up with a
462 * more sophisticated heuristic later.
464 young
= kvm_age_hva(kvm
, start
, end
);
465 spin_unlock(&kvm
->mmu_lock
);
466 srcu_read_unlock(&kvm
->srcu
, idx
);
471 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
472 struct mm_struct
*mm
,
473 unsigned long address
)
475 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
478 idx
= srcu_read_lock(&kvm
->srcu
);
479 spin_lock(&kvm
->mmu_lock
);
480 young
= kvm_test_age_hva(kvm
, address
);
481 spin_unlock(&kvm
->mmu_lock
);
482 srcu_read_unlock(&kvm
->srcu
, idx
);
487 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
488 struct mm_struct
*mm
)
490 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
493 idx
= srcu_read_lock(&kvm
->srcu
);
494 kvm_arch_flush_shadow_all(kvm
);
495 srcu_read_unlock(&kvm
->srcu
, idx
);
498 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
499 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
500 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
501 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
502 .clear_young
= kvm_mmu_notifier_clear_young
,
503 .test_young
= kvm_mmu_notifier_test_young
,
504 .change_pte
= kvm_mmu_notifier_change_pte
,
505 .release
= kvm_mmu_notifier_release
,
508 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
510 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
511 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
514 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
516 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
521 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
523 static struct kvm_memslots
*kvm_alloc_memslots(void)
526 struct kvm_memslots
*slots
;
528 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
532 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
533 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
538 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
540 if (!memslot
->dirty_bitmap
)
543 kvfree(memslot
->dirty_bitmap
);
544 memslot
->dirty_bitmap
= NULL
;
548 * Free any memory in @free but not in @dont.
550 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
551 struct kvm_memory_slot
*dont
)
553 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
554 kvm_destroy_dirty_bitmap(free
);
556 kvm_arch_free_memslot(kvm
, free
, dont
);
561 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
563 struct kvm_memory_slot
*memslot
;
568 kvm_for_each_memslot(memslot
, slots
)
569 kvm_free_memslot(kvm
, memslot
, NULL
);
574 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
578 if (!kvm
->debugfs_dentry
)
581 debugfs_remove_recursive(kvm
->debugfs_dentry
);
583 if (kvm
->debugfs_stat_data
) {
584 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
585 kfree(kvm
->debugfs_stat_data
[i
]);
586 kfree(kvm
->debugfs_stat_data
);
590 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
592 char dir_name
[ITOA_MAX_LEN
* 2];
593 struct kvm_stat_data
*stat_data
;
594 struct kvm_stats_debugfs_item
*p
;
596 if (!debugfs_initialized())
599 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
600 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
602 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
603 sizeof(*kvm
->debugfs_stat_data
),
605 if (!kvm
->debugfs_stat_data
)
608 for (p
= debugfs_entries
; p
->name
; p
++) {
609 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
613 stat_data
->kvm
= kvm
;
614 stat_data
->offset
= p
->offset
;
615 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
616 debugfs_create_file(p
->name
, 0644, kvm
->debugfs_dentry
,
617 stat_data
, stat_fops_per_vm
[p
->kind
]);
622 static struct kvm
*kvm_create_vm(unsigned long type
)
625 struct kvm
*kvm
= kvm_arch_alloc_vm();
628 return ERR_PTR(-ENOMEM
);
630 spin_lock_init(&kvm
->mmu_lock
);
632 kvm
->mm
= current
->mm
;
633 kvm_eventfd_init(kvm
);
634 mutex_init(&kvm
->lock
);
635 mutex_init(&kvm
->irq_lock
);
636 mutex_init(&kvm
->slots_lock
);
637 refcount_set(&kvm
->users_count
, 1);
638 INIT_LIST_HEAD(&kvm
->devices
);
640 r
= kvm_arch_init_vm(kvm
, type
);
642 goto out_err_no_disable
;
644 r
= hardware_enable_all();
646 goto out_err_no_disable
;
648 #ifdef CONFIG_HAVE_KVM_IRQFD
649 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
652 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
655 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
656 struct kvm_memslots
*slots
= kvm_alloc_memslots();
658 goto out_err_no_srcu
;
660 * Generations must be different for each address space.
661 * Init kvm generation close to the maximum to easily test the
662 * code of handling generation number wrap-around.
664 slots
->generation
= i
* 2 - 150;
665 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
668 if (init_srcu_struct(&kvm
->srcu
))
669 goto out_err_no_srcu
;
670 if (init_srcu_struct(&kvm
->irq_srcu
))
671 goto out_err_no_irq_srcu
;
672 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
673 rcu_assign_pointer(kvm
->buses
[i
],
674 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL
));
679 r
= kvm_init_mmu_notifier(kvm
);
683 spin_lock(&kvm_lock
);
684 list_add(&kvm
->vm_list
, &vm_list
);
685 spin_unlock(&kvm_lock
);
687 preempt_notifier_inc();
692 cleanup_srcu_struct(&kvm
->irq_srcu
);
694 cleanup_srcu_struct(&kvm
->srcu
);
696 hardware_disable_all();
698 refcount_set(&kvm
->users_count
, 0);
699 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
700 kfree(kvm_get_bus(kvm
, i
));
701 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
702 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
703 kvm_arch_free_vm(kvm
);
708 static void kvm_destroy_devices(struct kvm
*kvm
)
710 struct kvm_device
*dev
, *tmp
;
713 * We do not need to take the kvm->lock here, because nobody else
714 * has a reference to the struct kvm at this point and therefore
715 * cannot access the devices list anyhow.
717 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
718 list_del(&dev
->vm_node
);
719 dev
->ops
->destroy(dev
);
723 static void kvm_destroy_vm(struct kvm
*kvm
)
726 struct mm_struct
*mm
= kvm
->mm
;
728 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
729 kvm_destroy_vm_debugfs(kvm
);
730 kvm_arch_sync_events(kvm
);
731 spin_lock(&kvm_lock
);
732 list_del(&kvm
->vm_list
);
733 spin_unlock(&kvm_lock
);
734 kvm_free_irq_routing(kvm
);
735 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
736 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
739 kvm_io_bus_destroy(bus
);
740 kvm
->buses
[i
] = NULL
;
742 kvm_coalesced_mmio_free(kvm
);
743 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
744 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
746 kvm_arch_flush_shadow_all(kvm
);
748 kvm_arch_destroy_vm(kvm
);
749 kvm_destroy_devices(kvm
);
750 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
751 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
752 cleanup_srcu_struct(&kvm
->irq_srcu
);
753 cleanup_srcu_struct(&kvm
->srcu
);
754 kvm_arch_free_vm(kvm
);
755 preempt_notifier_dec();
756 hardware_disable_all();
760 void kvm_get_kvm(struct kvm
*kvm
)
762 refcount_inc(&kvm
->users_count
);
764 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
766 void kvm_put_kvm(struct kvm
*kvm
)
768 if (refcount_dec_and_test(&kvm
->users_count
))
771 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
774 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
776 struct kvm
*kvm
= filp
->private_data
;
778 kvm_irqfd_release(kvm
);
785 * Allocation size is twice as large as the actual dirty bitmap size.
786 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
788 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
790 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
792 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL
);
793 if (!memslot
->dirty_bitmap
)
800 * Insert memslot and re-sort memslots based on their GFN,
801 * so binary search could be used to lookup GFN.
802 * Sorting algorithm takes advantage of having initially
803 * sorted array and known changed memslot position.
805 static void update_memslots(struct kvm_memslots
*slots
,
806 struct kvm_memory_slot
*new,
807 enum kvm_mr_change change
)
810 int i
= slots
->id_to_index
[id
];
811 struct kvm_memory_slot
*mslots
= slots
->memslots
;
813 WARN_ON(mslots
[i
].id
!= id
);
817 WARN_ON(mslots
[i
].npages
|| !new->npages
);
821 WARN_ON(new->npages
|| !mslots
[i
].npages
);
827 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
828 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
829 if (!mslots
[i
+ 1].npages
)
831 mslots
[i
] = mslots
[i
+ 1];
832 slots
->id_to_index
[mslots
[i
].id
] = i
;
837 * The ">=" is needed when creating a slot with base_gfn == 0,
838 * so that it moves before all those with base_gfn == npages == 0.
840 * On the other hand, if new->npages is zero, the above loop has
841 * already left i pointing to the beginning of the empty part of
842 * mslots, and the ">=" would move the hole backwards in this
843 * case---which is wrong. So skip the loop when deleting a slot.
847 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
848 mslots
[i
] = mslots
[i
- 1];
849 slots
->id_to_index
[mslots
[i
].id
] = i
;
853 WARN_ON_ONCE(i
!= slots
->used_slots
);
856 slots
->id_to_index
[mslots
[i
].id
] = i
;
859 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
861 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
863 #ifdef __KVM_HAVE_READONLY_MEM
864 valid_flags
|= KVM_MEM_READONLY
;
867 if (mem
->flags
& ~valid_flags
)
873 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
874 int as_id
, struct kvm_memslots
*slots
)
876 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
879 * Set the low bit in the generation, which disables SPTE caching
880 * until the end of synchronize_srcu_expedited.
882 WARN_ON(old_memslots
->generation
& 1);
883 slots
->generation
= old_memslots
->generation
+ 1;
885 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
886 synchronize_srcu_expedited(&kvm
->srcu
);
889 * Increment the new memslot generation a second time. This prevents
890 * vm exits that race with memslot updates from caching a memslot
891 * generation that will (potentially) be valid forever.
893 * Generations must be unique even across address spaces. We do not need
894 * a global counter for that, instead the generation space is evenly split
895 * across address spaces. For example, with two address spaces, address
896 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
897 * use generations 2, 6, 10, 14, ...
899 slots
->generation
+= KVM_ADDRESS_SPACE_NUM
* 2 - 1;
901 kvm_arch_memslots_updated(kvm
, slots
);
907 * Allocate some memory and give it an address in the guest physical address
910 * Discontiguous memory is allowed, mostly for framebuffers.
912 * Must be called holding kvm->slots_lock for write.
914 int __kvm_set_memory_region(struct kvm
*kvm
,
915 const struct kvm_userspace_memory_region
*mem
)
919 unsigned long npages
;
920 struct kvm_memory_slot
*slot
;
921 struct kvm_memory_slot old
, new;
922 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
924 enum kvm_mr_change change
;
926 r
= check_memory_region_flags(mem
);
931 as_id
= mem
->slot
>> 16;
934 /* General sanity checks */
935 if (mem
->memory_size
& (PAGE_SIZE
- 1))
937 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
939 /* We can read the guest memory with __xxx_user() later on. */
940 if ((id
< KVM_USER_MEM_SLOTS
) &&
941 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
942 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
945 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
947 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
950 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
951 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
952 npages
= mem
->memory_size
>> PAGE_SHIFT
;
954 if (npages
> KVM_MEM_MAX_NR_PAGES
)
960 new.base_gfn
= base_gfn
;
962 new.flags
= mem
->flags
;
966 change
= KVM_MR_CREATE
;
967 else { /* Modify an existing slot. */
968 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
969 (npages
!= old
.npages
) ||
970 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
973 if (base_gfn
!= old
.base_gfn
)
974 change
= KVM_MR_MOVE
;
975 else if (new.flags
!= old
.flags
)
976 change
= KVM_MR_FLAGS_ONLY
;
977 else { /* Nothing to change. */
986 change
= KVM_MR_DELETE
;
991 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
992 /* Check for overlaps */
994 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
997 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
998 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
1003 /* Free page dirty bitmap if unneeded */
1004 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1005 new.dirty_bitmap
= NULL
;
1008 if (change
== KVM_MR_CREATE
) {
1009 new.userspace_addr
= mem
->userspace_addr
;
1011 if (kvm_arch_create_memslot(kvm
, &new, npages
))
1015 /* Allocate page dirty bitmap if needed */
1016 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
1017 if (kvm_create_dirty_bitmap(&new) < 0)
1021 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL
);
1024 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
1026 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
1027 slot
= id_to_memslot(slots
, id
);
1028 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1030 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1032 /* From this point no new shadow pages pointing to a deleted,
1033 * or moved, memslot will be created.
1035 * validation of sp->gfn happens in:
1036 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1037 * - kvm_is_visible_gfn (mmu_check_roots)
1039 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1042 * We can re-use the old_memslots from above, the only difference
1043 * from the currently installed memslots is the invalid flag. This
1044 * will get overwritten by update_memslots anyway.
1046 slots
= old_memslots
;
1049 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1053 /* actual memory is freed via old in kvm_free_memslot below */
1054 if (change
== KVM_MR_DELETE
) {
1055 new.dirty_bitmap
= NULL
;
1056 memset(&new.arch
, 0, sizeof(new.arch
));
1059 update_memslots(slots
, &new, change
);
1060 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1062 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1064 kvm_free_memslot(kvm
, &old
, &new);
1065 kvfree(old_memslots
);
1071 kvm_free_memslot(kvm
, &new, &old
);
1075 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1077 int kvm_set_memory_region(struct kvm
*kvm
,
1078 const struct kvm_userspace_memory_region
*mem
)
1082 mutex_lock(&kvm
->slots_lock
);
1083 r
= __kvm_set_memory_region(kvm
, mem
);
1084 mutex_unlock(&kvm
->slots_lock
);
1087 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1089 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1090 struct kvm_userspace_memory_region
*mem
)
1092 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1095 return kvm_set_memory_region(kvm
, mem
);
1098 int kvm_get_dirty_log(struct kvm
*kvm
,
1099 struct kvm_dirty_log
*log
, int *is_dirty
)
1101 struct kvm_memslots
*slots
;
1102 struct kvm_memory_slot
*memslot
;
1105 unsigned long any
= 0;
1107 as_id
= log
->slot
>> 16;
1108 id
= (u16
)log
->slot
;
1109 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1112 slots
= __kvm_memslots(kvm
, as_id
);
1113 memslot
= id_to_memslot(slots
, id
);
1114 if (!memslot
->dirty_bitmap
)
1117 n
= kvm_dirty_bitmap_bytes(memslot
);
1119 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1120 any
= memslot
->dirty_bitmap
[i
];
1122 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1129 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1131 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1133 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1134 * and reenable dirty page tracking for the corresponding pages.
1135 * @kvm: pointer to kvm instance
1136 * @log: slot id and address to which we copy the log
1137 * @is_dirty: flag set if any page is dirty
1139 * We need to keep it in mind that VCPU threads can write to the bitmap
1140 * concurrently. So, to avoid losing track of dirty pages we keep the
1143 * 1. Take a snapshot of the bit and clear it if needed.
1144 * 2. Write protect the corresponding page.
1145 * 3. Copy the snapshot to the userspace.
1146 * 4. Upon return caller flushes TLB's if needed.
1148 * Between 2 and 4, the guest may write to the page using the remaining TLB
1149 * entry. This is not a problem because the page is reported dirty using
1150 * the snapshot taken before and step 4 ensures that writes done after
1151 * exiting to userspace will be logged for the next call.
1154 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1155 struct kvm_dirty_log
*log
, bool *flush
)
1157 struct kvm_memslots
*slots
;
1158 struct kvm_memory_slot
*memslot
;
1161 unsigned long *dirty_bitmap
;
1162 unsigned long *dirty_bitmap_buffer
;
1164 as_id
= log
->slot
>> 16;
1165 id
= (u16
)log
->slot
;
1166 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1169 slots
= __kvm_memslots(kvm
, as_id
);
1170 memslot
= id_to_memslot(slots
, id
);
1172 dirty_bitmap
= memslot
->dirty_bitmap
;
1176 n
= kvm_dirty_bitmap_bytes(memslot
);
1178 if (kvm
->manual_dirty_log_protect
) {
1180 * Unlike kvm_get_dirty_log, we always return false in *flush,
1181 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1182 * is some code duplication between this function and
1183 * kvm_get_dirty_log, but hopefully all architecture
1184 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1185 * can be eliminated.
1187 dirty_bitmap_buffer
= dirty_bitmap
;
1189 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1190 memset(dirty_bitmap_buffer
, 0, n
);
1192 spin_lock(&kvm
->mmu_lock
);
1193 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1197 if (!dirty_bitmap
[i
])
1201 mask
= xchg(&dirty_bitmap
[i
], 0);
1202 dirty_bitmap_buffer
[i
] = mask
;
1205 offset
= i
* BITS_PER_LONG
;
1206 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1210 spin_unlock(&kvm
->mmu_lock
);
1213 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1217 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1220 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1221 * and reenable dirty page tracking for the corresponding pages.
1222 * @kvm: pointer to kvm instance
1223 * @log: slot id and address from which to fetch the bitmap of dirty pages
1225 int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
1226 struct kvm_clear_dirty_log
*log
, bool *flush
)
1228 struct kvm_memslots
*slots
;
1229 struct kvm_memory_slot
*memslot
;
1233 unsigned long *dirty_bitmap
;
1234 unsigned long *dirty_bitmap_buffer
;
1236 as_id
= log
->slot
>> 16;
1237 id
= (u16
)log
->slot
;
1238 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1241 if ((log
->first_page
& 63) || (log
->num_pages
& 63))
1244 slots
= __kvm_memslots(kvm
, as_id
);
1245 memslot
= id_to_memslot(slots
, id
);
1247 dirty_bitmap
= memslot
->dirty_bitmap
;
1251 n
= kvm_dirty_bitmap_bytes(memslot
);
1253 if (log
->first_page
> memslot
->npages
||
1254 log
->num_pages
> memslot
->npages
- log
->first_page
)
1258 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1259 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
1262 spin_lock(&kvm
->mmu_lock
);
1263 for (offset
= log
->first_page
,
1264 i
= offset
/ BITS_PER_LONG
, n
= log
->num_pages
/ BITS_PER_LONG
; n
--;
1265 i
++, offset
+= BITS_PER_LONG
) {
1266 unsigned long mask
= *dirty_bitmap_buffer
++;
1267 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
1271 mask
&= atomic_long_fetch_andnot(mask
, p
);
1274 * mask contains the bits that really have been cleared. This
1275 * never includes any bits beyond the length of the memslot (if
1276 * the length is not aligned to 64 pages), therefore it is not
1277 * a problem if userspace sets them in log->dirty_bitmap.
1281 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1285 spin_unlock(&kvm
->mmu_lock
);
1289 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect
);
1292 bool kvm_largepages_enabled(void)
1294 return largepages_enabled
;
1297 void kvm_disable_largepages(void)
1299 largepages_enabled
= false;
1301 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1303 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1305 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1307 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1309 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1311 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1314 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1316 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1318 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1319 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1324 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1326 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1328 struct vm_area_struct
*vma
;
1329 unsigned long addr
, size
;
1333 addr
= gfn_to_hva(kvm
, gfn
);
1334 if (kvm_is_error_hva(addr
))
1337 down_read(¤t
->mm
->mmap_sem
);
1338 vma
= find_vma(current
->mm
, addr
);
1342 size
= vma_kernel_pagesize(vma
);
1345 up_read(¤t
->mm
->mmap_sem
);
1350 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1352 return slot
->flags
& KVM_MEM_READONLY
;
1355 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1356 gfn_t
*nr_pages
, bool write
)
1358 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1359 return KVM_HVA_ERR_BAD
;
1361 if (memslot_is_readonly(slot
) && write
)
1362 return KVM_HVA_ERR_RO_BAD
;
1365 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1367 return __gfn_to_hva_memslot(slot
, gfn
);
1370 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1373 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1376 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1379 return gfn_to_hva_many(slot
, gfn
, NULL
);
1381 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1383 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1385 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1387 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1389 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1391 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1393 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1396 * Return the hva of a @gfn and the R/W attribute if possible.
1398 * @slot: the kvm_memory_slot which contains @gfn
1399 * @gfn: the gfn to be translated
1400 * @writable: used to return the read/write attribute of the @slot if the hva
1401 * is valid and @writable is not NULL
1403 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1404 gfn_t gfn
, bool *writable
)
1406 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1408 if (!kvm_is_error_hva(hva
) && writable
)
1409 *writable
= !memslot_is_readonly(slot
);
1414 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1416 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1418 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1421 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1423 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1425 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1428 static inline int check_user_page_hwpoison(unsigned long addr
)
1430 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1432 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1433 return rc
== -EHWPOISON
;
1437 * The fast path to get the writable pfn which will be stored in @pfn,
1438 * true indicates success, otherwise false is returned. It's also the
1439 * only part that runs if we can are in atomic context.
1441 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
1442 bool *writable
, kvm_pfn_t
*pfn
)
1444 struct page
*page
[1];
1448 * Fast pin a writable pfn only if it is a write fault request
1449 * or the caller allows to map a writable pfn for a read fault
1452 if (!(write_fault
|| writable
))
1455 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1457 *pfn
= page_to_pfn(page
[0]);
1468 * The slow path to get the pfn of the specified host virtual address,
1469 * 1 indicates success, -errno is returned if error is detected.
1471 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1472 bool *writable
, kvm_pfn_t
*pfn
)
1474 unsigned int flags
= FOLL_HWPOISON
;
1481 *writable
= write_fault
;
1484 flags
|= FOLL_WRITE
;
1486 flags
|= FOLL_NOWAIT
;
1488 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
1492 /* map read fault as writable if possible */
1493 if (unlikely(!write_fault
) && writable
) {
1496 if (__get_user_pages_fast(addr
, 1, 1, &wpage
) == 1) {
1502 *pfn
= page_to_pfn(page
);
1506 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1508 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1511 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1517 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1518 unsigned long addr
, bool *async
,
1519 bool write_fault
, bool *writable
,
1525 r
= follow_pfn(vma
, addr
, &pfn
);
1528 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1529 * not call the fault handler, so do it here.
1531 bool unlocked
= false;
1532 r
= fixup_user_fault(current
, current
->mm
, addr
,
1533 (write_fault
? FAULT_FLAG_WRITE
: 0),
1540 r
= follow_pfn(vma
, addr
, &pfn
);
1550 * Get a reference here because callers of *hva_to_pfn* and
1551 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1552 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1553 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1554 * simply do nothing for reserved pfns.
1556 * Whoever called remap_pfn_range is also going to call e.g.
1557 * unmap_mapping_range before the underlying pages are freed,
1558 * causing a call to our MMU notifier.
1567 * Pin guest page in memory and return its pfn.
1568 * @addr: host virtual address which maps memory to the guest
1569 * @atomic: whether this function can sleep
1570 * @async: whether this function need to wait IO complete if the
1571 * host page is not in the memory
1572 * @write_fault: whether we should get a writable host page
1573 * @writable: whether it allows to map a writable host page for !@write_fault
1575 * The function will map a writable host page for these two cases:
1576 * 1): @write_fault = true
1577 * 2): @write_fault = false && @writable, @writable will tell the caller
1578 * whether the mapping is writable.
1580 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1581 bool write_fault
, bool *writable
)
1583 struct vm_area_struct
*vma
;
1587 /* we can do it either atomically or asynchronously, not both */
1588 BUG_ON(atomic
&& async
);
1590 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
1594 return KVM_PFN_ERR_FAULT
;
1596 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1600 down_read(¤t
->mm
->mmap_sem
);
1601 if (npages
== -EHWPOISON
||
1602 (!async
&& check_user_page_hwpoison(addr
))) {
1603 pfn
= KVM_PFN_ERR_HWPOISON
;
1608 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1611 pfn
= KVM_PFN_ERR_FAULT
;
1612 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1613 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
1617 pfn
= KVM_PFN_ERR_FAULT
;
1619 if (async
&& vma_is_valid(vma
, write_fault
))
1621 pfn
= KVM_PFN_ERR_FAULT
;
1624 up_read(¤t
->mm
->mmap_sem
);
1628 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1629 bool atomic
, bool *async
, bool write_fault
,
1632 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1634 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1637 return KVM_PFN_ERR_RO_FAULT
;
1640 if (kvm_is_error_hva(addr
)) {
1643 return KVM_PFN_NOSLOT
;
1646 /* Do not map writable pfn in the readonly memslot. */
1647 if (writable
&& memslot_is_readonly(slot
)) {
1652 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1655 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1657 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1660 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1661 write_fault
, writable
);
1663 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1665 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1667 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1669 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1671 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1673 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1675 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1677 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1679 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1681 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1683 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1685 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1687 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1689 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1691 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1693 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1695 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1697 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1699 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1701 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1702 struct page
**pages
, int nr_pages
)
1707 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1708 if (kvm_is_error_hva(addr
))
1711 if (entry
< nr_pages
)
1714 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1716 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1718 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1720 if (is_error_noslot_pfn(pfn
))
1721 return KVM_ERR_PTR_BAD_PAGE
;
1723 if (kvm_is_reserved_pfn(pfn
)) {
1725 return KVM_ERR_PTR_BAD_PAGE
;
1728 return pfn_to_page(pfn
);
1731 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1735 pfn
= gfn_to_pfn(kvm
, gfn
);
1737 return kvm_pfn_to_page(pfn
);
1739 EXPORT_SYMBOL_GPL(gfn_to_page
);
1741 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1745 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1747 return kvm_pfn_to_page(pfn
);
1749 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1751 void kvm_release_page_clean(struct page
*page
)
1753 WARN_ON(is_error_page(page
));
1755 kvm_release_pfn_clean(page_to_pfn(page
));
1757 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1759 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1761 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1762 put_page(pfn_to_page(pfn
));
1764 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1766 void kvm_release_page_dirty(struct page
*page
)
1768 WARN_ON(is_error_page(page
));
1770 kvm_release_pfn_dirty(page_to_pfn(page
));
1772 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1774 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1776 kvm_set_pfn_dirty(pfn
);
1777 kvm_release_pfn_clean(pfn
);
1779 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
1781 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1783 if (!kvm_is_reserved_pfn(pfn
)) {
1784 struct page
*page
= pfn_to_page(pfn
);
1786 if (!PageReserved(page
))
1790 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1792 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1794 if (!kvm_is_reserved_pfn(pfn
))
1795 mark_page_accessed(pfn_to_page(pfn
));
1797 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1799 void kvm_get_pfn(kvm_pfn_t pfn
)
1801 if (!kvm_is_reserved_pfn(pfn
))
1802 get_page(pfn_to_page(pfn
));
1804 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1806 static int next_segment(unsigned long len
, int offset
)
1808 if (len
> PAGE_SIZE
- offset
)
1809 return PAGE_SIZE
- offset
;
1814 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1815 void *data
, int offset
, int len
)
1820 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1821 if (kvm_is_error_hva(addr
))
1823 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1829 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1832 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1834 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1836 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1838 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1839 int offset
, int len
)
1841 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1843 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1845 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1847 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1849 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1851 int offset
= offset_in_page(gpa
);
1854 while ((seg
= next_segment(len
, offset
)) != 0) {
1855 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1865 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1867 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1869 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1871 int offset
= offset_in_page(gpa
);
1874 while ((seg
= next_segment(len
, offset
)) != 0) {
1875 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1885 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1887 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1888 void *data
, int offset
, unsigned long len
)
1893 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1894 if (kvm_is_error_hva(addr
))
1896 pagefault_disable();
1897 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1904 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1907 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1908 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1909 int offset
= offset_in_page(gpa
);
1911 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1913 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1915 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1916 void *data
, unsigned long len
)
1918 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1919 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1920 int offset
= offset_in_page(gpa
);
1922 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1924 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1926 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1927 const void *data
, int offset
, int len
)
1932 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1933 if (kvm_is_error_hva(addr
))
1935 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1938 mark_page_dirty_in_slot(memslot
, gfn
);
1942 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1943 const void *data
, int offset
, int len
)
1945 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1947 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1949 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1951 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1952 const void *data
, int offset
, int len
)
1954 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1956 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1958 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1960 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1963 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1965 int offset
= offset_in_page(gpa
);
1968 while ((seg
= next_segment(len
, offset
)) != 0) {
1969 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1979 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1981 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1984 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1986 int offset
= offset_in_page(gpa
);
1989 while ((seg
= next_segment(len
, offset
)) != 0) {
1990 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2000 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
2002 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
2003 struct gfn_to_hva_cache
*ghc
,
2004 gpa_t gpa
, unsigned long len
)
2006 int offset
= offset_in_page(gpa
);
2007 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
2008 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
2009 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
2010 gfn_t nr_pages_avail
;
2011 int r
= start_gfn
<= end_gfn
? 0 : -EINVAL
;
2014 ghc
->generation
= slots
->generation
;
2016 ghc
->hva
= KVM_HVA_ERR_BAD
;
2019 * If the requested region crosses two memslots, we still
2020 * verify that the entire region is valid here.
2022 while (!r
&& start_gfn
<= end_gfn
) {
2023 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
2024 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
2026 if (kvm_is_error_hva(ghc
->hva
))
2028 start_gfn
+= nr_pages_avail
;
2031 /* Use the slow path for cross page reads and writes. */
2032 if (!r
&& nr_pages_needed
== 1)
2035 ghc
->memslot
= NULL
;
2040 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2041 gpa_t gpa
, unsigned long len
)
2043 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2044 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
2046 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
2048 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2049 void *data
, unsigned int offset
,
2052 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2054 gpa_t gpa
= ghc
->gpa
+ offset
;
2056 BUG_ON(len
+ offset
> ghc
->len
);
2058 if (slots
->generation
!= ghc
->generation
)
2059 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2061 if (unlikely(!ghc
->memslot
))
2062 return kvm_write_guest(kvm
, gpa
, data
, len
);
2064 if (kvm_is_error_hva(ghc
->hva
))
2067 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2070 mark_page_dirty_in_slot(ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2074 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2076 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2077 void *data
, unsigned long len
)
2079 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2081 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2083 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2084 void *data
, unsigned long len
)
2086 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2089 BUG_ON(len
> ghc
->len
);
2091 if (slots
->generation
!= ghc
->generation
)
2092 __kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
);
2094 if (unlikely(!ghc
->memslot
))
2095 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2097 if (kvm_is_error_hva(ghc
->hva
))
2100 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2106 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2108 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2110 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2112 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2114 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2116 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2118 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2120 int offset
= offset_in_page(gpa
);
2123 while ((seg
= next_segment(len
, offset
)) != 0) {
2124 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2133 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2135 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2138 if (memslot
&& memslot
->dirty_bitmap
) {
2139 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2141 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2145 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2147 struct kvm_memory_slot
*memslot
;
2149 memslot
= gfn_to_memslot(kvm
, gfn
);
2150 mark_page_dirty_in_slot(memslot
, gfn
);
2152 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2154 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2156 struct kvm_memory_slot
*memslot
;
2158 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2159 mark_page_dirty_in_slot(memslot
, gfn
);
2161 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2163 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2165 if (!vcpu
->sigset_active
)
2169 * This does a lockless modification of ->real_blocked, which is fine
2170 * because, only current can change ->real_blocked and all readers of
2171 * ->real_blocked don't care as long ->real_blocked is always a subset
2174 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2177 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2179 if (!vcpu
->sigset_active
)
2182 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2183 sigemptyset(¤t
->real_blocked
);
2186 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2188 unsigned int old
, val
, grow
;
2190 old
= val
= vcpu
->halt_poll_ns
;
2191 grow
= READ_ONCE(halt_poll_ns_grow
);
2193 if (val
== 0 && grow
)
2198 if (val
> halt_poll_ns
)
2201 vcpu
->halt_poll_ns
= val
;
2202 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2205 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2207 unsigned int old
, val
, shrink
;
2209 old
= val
= vcpu
->halt_poll_ns
;
2210 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2216 vcpu
->halt_poll_ns
= val
;
2217 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2220 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2223 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2225 if (kvm_arch_vcpu_runnable(vcpu
)) {
2226 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2229 if (kvm_cpu_has_pending_timer(vcpu
))
2231 if (signal_pending(current
))
2236 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2241 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2243 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2246 DECLARE_SWAITQUEUE(wait
);
2247 bool waited
= false;
2250 start
= cur
= ktime_get();
2251 if (vcpu
->halt_poll_ns
) {
2252 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2254 ++vcpu
->stat
.halt_attempted_poll
;
2257 * This sets KVM_REQ_UNHALT if an interrupt
2260 if (kvm_vcpu_check_block(vcpu
) < 0) {
2261 ++vcpu
->stat
.halt_successful_poll
;
2262 if (!vcpu_valid_wakeup(vcpu
))
2263 ++vcpu
->stat
.halt_poll_invalid
;
2267 } while (single_task_running() && ktime_before(cur
, stop
));
2270 kvm_arch_vcpu_blocking(vcpu
);
2273 prepare_to_swait_exclusive(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2275 if (kvm_vcpu_check_block(vcpu
) < 0)
2282 finish_swait(&vcpu
->wq
, &wait
);
2285 kvm_arch_vcpu_unblocking(vcpu
);
2287 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2289 if (!vcpu_valid_wakeup(vcpu
))
2290 shrink_halt_poll_ns(vcpu
);
2291 else if (halt_poll_ns
) {
2292 if (block_ns
<= vcpu
->halt_poll_ns
)
2294 /* we had a long block, shrink polling */
2295 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2296 shrink_halt_poll_ns(vcpu
);
2297 /* we had a short halt and our poll time is too small */
2298 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2299 block_ns
< halt_poll_ns
)
2300 grow_halt_poll_ns(vcpu
);
2302 vcpu
->halt_poll_ns
= 0;
2304 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2305 kvm_arch_vcpu_block_finish(vcpu
);
2307 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2309 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2311 struct swait_queue_head
*wqp
;
2313 wqp
= kvm_arch_vcpu_wq(vcpu
);
2314 if (swq_has_sleeper(wqp
)) {
2316 ++vcpu
->stat
.halt_wakeup
;
2322 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2326 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2328 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2331 int cpu
= vcpu
->cpu
;
2333 if (kvm_vcpu_wake_up(vcpu
))
2337 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2338 if (kvm_arch_vcpu_should_kick(vcpu
))
2339 smp_send_reschedule(cpu
);
2342 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2343 #endif /* !CONFIG_S390 */
2345 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2348 struct task_struct
*task
= NULL
;
2352 pid
= rcu_dereference(target
->pid
);
2354 task
= get_pid_task(pid
, PIDTYPE_PID
);
2358 ret
= yield_to(task
, 1);
2359 put_task_struct(task
);
2363 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2366 * Helper that checks whether a VCPU is eligible for directed yield.
2367 * Most eligible candidate to yield is decided by following heuristics:
2369 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2370 * (preempted lock holder), indicated by @in_spin_loop.
2371 * Set at the beiginning and cleared at the end of interception/PLE handler.
2373 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2374 * chance last time (mostly it has become eligible now since we have probably
2375 * yielded to lockholder in last iteration. This is done by toggling
2376 * @dy_eligible each time a VCPU checked for eligibility.)
2378 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2379 * to preempted lock-holder could result in wrong VCPU selection and CPU
2380 * burning. Giving priority for a potential lock-holder increases lock
2383 * Since algorithm is based on heuristics, accessing another VCPU data without
2384 * locking does not harm. It may result in trying to yield to same VCPU, fail
2385 * and continue with next VCPU and so on.
2387 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2389 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2392 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2393 vcpu
->spin_loop
.dy_eligible
;
2395 if (vcpu
->spin_loop
.in_spin_loop
)
2396 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2404 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2406 struct kvm
*kvm
= me
->kvm
;
2407 struct kvm_vcpu
*vcpu
;
2408 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2414 kvm_vcpu_set_in_spin_loop(me
, true);
2416 * We boost the priority of a VCPU that is runnable but not
2417 * currently running, because it got preempted by something
2418 * else and called schedule in __vcpu_run. Hopefully that
2419 * VCPU is holding the lock that we need and will release it.
2420 * We approximate round-robin by starting at the last boosted VCPU.
2422 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2423 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2424 if (!pass
&& i
<= last_boosted_vcpu
) {
2425 i
= last_boosted_vcpu
;
2427 } else if (pass
&& i
> last_boosted_vcpu
)
2429 if (!READ_ONCE(vcpu
->preempted
))
2433 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2435 if (yield_to_kernel_mode
&& !kvm_arch_vcpu_in_kernel(vcpu
))
2437 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2440 yielded
= kvm_vcpu_yield_to(vcpu
);
2442 kvm
->last_boosted_vcpu
= i
;
2444 } else if (yielded
< 0) {
2451 kvm_vcpu_set_in_spin_loop(me
, false);
2453 /* Ensure vcpu is not eligible during next spinloop */
2454 kvm_vcpu_set_dy_eligible(me
, false);
2456 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2458 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
2460 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
2463 if (vmf
->pgoff
== 0)
2464 page
= virt_to_page(vcpu
->run
);
2466 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2467 page
= virt_to_page(vcpu
->arch
.pio_data
);
2469 #ifdef CONFIG_KVM_MMIO
2470 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2471 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2474 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2480 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2481 .fault
= kvm_vcpu_fault
,
2484 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2486 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2490 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2492 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2494 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2495 kvm_put_kvm(vcpu
->kvm
);
2499 static struct file_operations kvm_vcpu_fops
= {
2500 .release
= kvm_vcpu_release
,
2501 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2502 .mmap
= kvm_vcpu_mmap
,
2503 .llseek
= noop_llseek
,
2504 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
2508 * Allocates an inode for the vcpu.
2510 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2512 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
2514 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
2515 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2518 static int kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
2520 char dir_name
[ITOA_MAX_LEN
* 2];
2523 if (!kvm_arch_has_vcpu_debugfs())
2526 if (!debugfs_initialized())
2529 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
2530 vcpu
->debugfs_dentry
= debugfs_create_dir(dir_name
,
2531 vcpu
->kvm
->debugfs_dentry
);
2532 if (!vcpu
->debugfs_dentry
)
2535 ret
= kvm_arch_create_vcpu_debugfs(vcpu
);
2537 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2545 * Creates some virtual cpus. Good luck creating more than one.
2547 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2550 struct kvm_vcpu
*vcpu
;
2552 if (id
>= KVM_MAX_VCPU_ID
)
2555 mutex_lock(&kvm
->lock
);
2556 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2557 mutex_unlock(&kvm
->lock
);
2561 kvm
->created_vcpus
++;
2562 mutex_unlock(&kvm
->lock
);
2564 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2567 goto vcpu_decrement
;
2570 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2572 r
= kvm_arch_vcpu_setup(vcpu
);
2576 r
= kvm_create_vcpu_debugfs(vcpu
);
2580 mutex_lock(&kvm
->lock
);
2581 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2583 goto unlock_vcpu_destroy
;
2586 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2588 /* Now it's all set up, let userspace reach it */
2590 r
= create_vcpu_fd(vcpu
);
2593 goto unlock_vcpu_destroy
;
2596 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2599 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2600 * before kvm->online_vcpu's incremented value.
2603 atomic_inc(&kvm
->online_vcpus
);
2605 mutex_unlock(&kvm
->lock
);
2606 kvm_arch_vcpu_postcreate(vcpu
);
2609 unlock_vcpu_destroy
:
2610 mutex_unlock(&kvm
->lock
);
2611 debugfs_remove_recursive(vcpu
->debugfs_dentry
);
2613 kvm_arch_vcpu_destroy(vcpu
);
2615 mutex_lock(&kvm
->lock
);
2616 kvm
->created_vcpus
--;
2617 mutex_unlock(&kvm
->lock
);
2621 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2624 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2625 vcpu
->sigset_active
= 1;
2626 vcpu
->sigset
= *sigset
;
2628 vcpu
->sigset_active
= 0;
2632 static long kvm_vcpu_ioctl(struct file
*filp
,
2633 unsigned int ioctl
, unsigned long arg
)
2635 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2636 void __user
*argp
= (void __user
*)arg
;
2638 struct kvm_fpu
*fpu
= NULL
;
2639 struct kvm_sregs
*kvm_sregs
= NULL
;
2641 if (vcpu
->kvm
->mm
!= current
->mm
)
2644 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2648 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2649 * execution; mutex_lock() would break them.
2651 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
2652 if (r
!= -ENOIOCTLCMD
)
2655 if (mutex_lock_killable(&vcpu
->mutex
))
2663 oldpid
= rcu_access_pointer(vcpu
->pid
);
2664 if (unlikely(oldpid
!= task_pid(current
))) {
2665 /* The thread running this VCPU changed. */
2668 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
2672 newpid
= get_task_pid(current
, PIDTYPE_PID
);
2673 rcu_assign_pointer(vcpu
->pid
, newpid
);
2678 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2679 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2682 case KVM_GET_REGS
: {
2683 struct kvm_regs
*kvm_regs
;
2686 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2689 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2693 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2700 case KVM_SET_REGS
: {
2701 struct kvm_regs
*kvm_regs
;
2704 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2705 if (IS_ERR(kvm_regs
)) {
2706 r
= PTR_ERR(kvm_regs
);
2709 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2713 case KVM_GET_SREGS
: {
2714 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2718 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2722 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2727 case KVM_SET_SREGS
: {
2728 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2729 if (IS_ERR(kvm_sregs
)) {
2730 r
= PTR_ERR(kvm_sregs
);
2734 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2737 case KVM_GET_MP_STATE
: {
2738 struct kvm_mp_state mp_state
;
2740 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2744 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2749 case KVM_SET_MP_STATE
: {
2750 struct kvm_mp_state mp_state
;
2753 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2755 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2758 case KVM_TRANSLATE
: {
2759 struct kvm_translation tr
;
2762 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2764 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2768 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2773 case KVM_SET_GUEST_DEBUG
: {
2774 struct kvm_guest_debug dbg
;
2777 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2779 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2782 case KVM_SET_SIGNAL_MASK
: {
2783 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2784 struct kvm_signal_mask kvm_sigmask
;
2785 sigset_t sigset
, *p
;
2790 if (copy_from_user(&kvm_sigmask
, argp
,
2791 sizeof(kvm_sigmask
)))
2794 if (kvm_sigmask
.len
!= sizeof(sigset
))
2797 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2802 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2806 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2810 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2814 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2820 fpu
= memdup_user(argp
, sizeof(*fpu
));
2826 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2830 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2833 mutex_unlock(&vcpu
->mutex
);
2839 #ifdef CONFIG_KVM_COMPAT
2840 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2841 unsigned int ioctl
, unsigned long arg
)
2843 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2844 void __user
*argp
= compat_ptr(arg
);
2847 if (vcpu
->kvm
->mm
!= current
->mm
)
2851 case KVM_SET_SIGNAL_MASK
: {
2852 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2853 struct kvm_signal_mask kvm_sigmask
;
2858 if (copy_from_user(&kvm_sigmask
, argp
,
2859 sizeof(kvm_sigmask
)))
2862 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
2865 if (get_compat_sigset(&sigset
, (void *)sigmask_arg
->sigset
))
2867 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2869 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2873 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2881 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2882 int (*accessor
)(struct kvm_device
*dev
,
2883 struct kvm_device_attr
*attr
),
2886 struct kvm_device_attr attr
;
2891 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2894 return accessor(dev
, &attr
);
2897 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2900 struct kvm_device
*dev
= filp
->private_data
;
2903 case KVM_SET_DEVICE_ATTR
:
2904 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2905 case KVM_GET_DEVICE_ATTR
:
2906 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2907 case KVM_HAS_DEVICE_ATTR
:
2908 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2910 if (dev
->ops
->ioctl
)
2911 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2917 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2919 struct kvm_device
*dev
= filp
->private_data
;
2920 struct kvm
*kvm
= dev
->kvm
;
2926 static const struct file_operations kvm_device_fops
= {
2927 .unlocked_ioctl
= kvm_device_ioctl
,
2928 .release
= kvm_device_release
,
2929 KVM_COMPAT(kvm_device_ioctl
),
2932 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2934 if (filp
->f_op
!= &kvm_device_fops
)
2937 return filp
->private_data
;
2940 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2941 #ifdef CONFIG_KVM_MPIC
2942 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2943 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2947 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2949 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2952 if (kvm_device_ops_table
[type
] != NULL
)
2955 kvm_device_ops_table
[type
] = ops
;
2959 void kvm_unregister_device_ops(u32 type
)
2961 if (kvm_device_ops_table
[type
] != NULL
)
2962 kvm_device_ops_table
[type
] = NULL
;
2965 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2966 struct kvm_create_device
*cd
)
2968 struct kvm_device_ops
*ops
= NULL
;
2969 struct kvm_device
*dev
;
2970 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2973 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2976 ops
= kvm_device_ops_table
[cd
->type
];
2983 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2990 mutex_lock(&kvm
->lock
);
2991 ret
= ops
->create(dev
, cd
->type
);
2993 mutex_unlock(&kvm
->lock
);
2997 list_add(&dev
->vm_node
, &kvm
->devices
);
2998 mutex_unlock(&kvm
->lock
);
3004 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
3007 mutex_lock(&kvm
->lock
);
3008 list_del(&dev
->vm_node
);
3009 mutex_unlock(&kvm
->lock
);
3018 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
3021 case KVM_CAP_USER_MEMORY
:
3022 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
3023 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
3024 case KVM_CAP_INTERNAL_ERROR_DATA
:
3025 #ifdef CONFIG_HAVE_KVM_MSI
3026 case KVM_CAP_SIGNAL_MSI
:
3028 #ifdef CONFIG_HAVE_KVM_IRQFD
3030 case KVM_CAP_IRQFD_RESAMPLE
:
3032 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
3033 case KVM_CAP_CHECK_EXTENSION_VM
:
3034 case KVM_CAP_ENABLE_CAP_VM
:
3035 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3036 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
:
3039 #ifdef CONFIG_KVM_MMIO
3040 case KVM_CAP_COALESCED_MMIO
:
3041 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
3042 case KVM_CAP_COALESCED_PIO
:
3045 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3046 case KVM_CAP_IRQ_ROUTING
:
3047 return KVM_MAX_IRQ_ROUTES
;
3049 #if KVM_ADDRESS_SPACE_NUM > 1
3050 case KVM_CAP_MULTI_ADDRESS_SPACE
:
3051 return KVM_ADDRESS_SPACE_NUM
;
3053 case KVM_CAP_MAX_VCPU_ID
:
3054 return KVM_MAX_VCPU_ID
;
3058 return kvm_vm_ioctl_check_extension(kvm
, arg
);
3061 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
3062 struct kvm_enable_cap
*cap
)
3067 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
3068 struct kvm_enable_cap
*cap
)
3071 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3072 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT
:
3073 if (cap
->flags
|| (cap
->args
[0] & ~1))
3075 kvm
->manual_dirty_log_protect
= cap
->args
[0];
3079 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
3083 static long kvm_vm_ioctl(struct file
*filp
,
3084 unsigned int ioctl
, unsigned long arg
)
3086 struct kvm
*kvm
= filp
->private_data
;
3087 void __user
*argp
= (void __user
*)arg
;
3090 if (kvm
->mm
!= current
->mm
)
3093 case KVM_CREATE_VCPU
:
3094 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3096 case KVM_ENABLE_CAP
: {
3097 struct kvm_enable_cap cap
;
3100 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3102 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
3105 case KVM_SET_USER_MEMORY_REGION
: {
3106 struct kvm_userspace_memory_region kvm_userspace_mem
;
3109 if (copy_from_user(&kvm_userspace_mem
, argp
,
3110 sizeof(kvm_userspace_mem
)))
3113 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3116 case KVM_GET_DIRTY_LOG
: {
3117 struct kvm_dirty_log log
;
3120 if (copy_from_user(&log
, argp
, sizeof(log
)))
3122 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3125 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3126 case KVM_CLEAR_DIRTY_LOG
: {
3127 struct kvm_clear_dirty_log log
;
3130 if (copy_from_user(&log
, argp
, sizeof(log
)))
3132 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
3136 #ifdef CONFIG_KVM_MMIO
3137 case KVM_REGISTER_COALESCED_MMIO
: {
3138 struct kvm_coalesced_mmio_zone zone
;
3141 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3143 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3146 case KVM_UNREGISTER_COALESCED_MMIO
: {
3147 struct kvm_coalesced_mmio_zone zone
;
3150 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3152 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3157 struct kvm_irqfd data
;
3160 if (copy_from_user(&data
, argp
, sizeof(data
)))
3162 r
= kvm_irqfd(kvm
, &data
);
3165 case KVM_IOEVENTFD
: {
3166 struct kvm_ioeventfd data
;
3169 if (copy_from_user(&data
, argp
, sizeof(data
)))
3171 r
= kvm_ioeventfd(kvm
, &data
);
3174 #ifdef CONFIG_HAVE_KVM_MSI
3175 case KVM_SIGNAL_MSI
: {
3179 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3181 r
= kvm_send_userspace_msi(kvm
, &msi
);
3185 #ifdef __KVM_HAVE_IRQ_LINE
3186 case KVM_IRQ_LINE_STATUS
:
3187 case KVM_IRQ_LINE
: {
3188 struct kvm_irq_level irq_event
;
3191 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3194 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3195 ioctl
== KVM_IRQ_LINE_STATUS
);
3200 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3201 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3209 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3210 case KVM_SET_GSI_ROUTING
: {
3211 struct kvm_irq_routing routing
;
3212 struct kvm_irq_routing __user
*urouting
;
3213 struct kvm_irq_routing_entry
*entries
= NULL
;
3216 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3219 if (!kvm_arch_can_set_irq_routing(kvm
))
3221 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3227 entries
= vmalloc(array_size(sizeof(*entries
),
3233 if (copy_from_user(entries
, urouting
->entries
,
3234 routing
.nr
* sizeof(*entries
)))
3235 goto out_free_irq_routing
;
3237 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3239 out_free_irq_routing
:
3243 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3244 case KVM_CREATE_DEVICE
: {
3245 struct kvm_create_device cd
;
3248 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3251 r
= kvm_ioctl_create_device(kvm
, &cd
);
3256 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3262 case KVM_CHECK_EXTENSION
:
3263 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3266 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3272 #ifdef CONFIG_KVM_COMPAT
3273 struct compat_kvm_dirty_log
{
3277 compat_uptr_t dirty_bitmap
; /* one bit per page */
3282 static long kvm_vm_compat_ioctl(struct file
*filp
,
3283 unsigned int ioctl
, unsigned long arg
)
3285 struct kvm
*kvm
= filp
->private_data
;
3288 if (kvm
->mm
!= current
->mm
)
3291 case KVM_GET_DIRTY_LOG
: {
3292 struct compat_kvm_dirty_log compat_log
;
3293 struct kvm_dirty_log log
;
3295 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3296 sizeof(compat_log
)))
3298 log
.slot
= compat_log
.slot
;
3299 log
.padding1
= compat_log
.padding1
;
3300 log
.padding2
= compat_log
.padding2
;
3301 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3303 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3307 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3313 static struct file_operations kvm_vm_fops
= {
3314 .release
= kvm_vm_release
,
3315 .unlocked_ioctl
= kvm_vm_ioctl
,
3316 .llseek
= noop_llseek
,
3317 KVM_COMPAT(kvm_vm_compat_ioctl
),
3320 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3326 kvm
= kvm_create_vm(type
);
3328 return PTR_ERR(kvm
);
3329 #ifdef CONFIG_KVM_MMIO
3330 r
= kvm_coalesced_mmio_init(kvm
);
3334 r
= get_unused_fd_flags(O_CLOEXEC
);
3338 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3346 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3347 * already set, with ->release() being kvm_vm_release(). In error
3348 * cases it will be called by the final fput(file) and will take
3349 * care of doing kvm_put_kvm(kvm).
3351 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3356 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
3358 fd_install(r
, file
);
3366 static long kvm_dev_ioctl(struct file
*filp
,
3367 unsigned int ioctl
, unsigned long arg
)
3372 case KVM_GET_API_VERSION
:
3375 r
= KVM_API_VERSION
;
3378 r
= kvm_dev_ioctl_create_vm(arg
);
3380 case KVM_CHECK_EXTENSION
:
3381 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3383 case KVM_GET_VCPU_MMAP_SIZE
:
3386 r
= PAGE_SIZE
; /* struct kvm_run */
3388 r
+= PAGE_SIZE
; /* pio data page */
3390 #ifdef CONFIG_KVM_MMIO
3391 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3394 case KVM_TRACE_ENABLE
:
3395 case KVM_TRACE_PAUSE
:
3396 case KVM_TRACE_DISABLE
:
3400 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3406 static struct file_operations kvm_chardev_ops
= {
3407 .unlocked_ioctl
= kvm_dev_ioctl
,
3408 .llseek
= noop_llseek
,
3409 KVM_COMPAT(kvm_dev_ioctl
),
3412 static struct miscdevice kvm_dev
= {
3418 static void hardware_enable_nolock(void *junk
)
3420 int cpu
= raw_smp_processor_id();
3423 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3426 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3428 r
= kvm_arch_hardware_enable();
3431 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3432 atomic_inc(&hardware_enable_failed
);
3433 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3437 static int kvm_starting_cpu(unsigned int cpu
)
3439 raw_spin_lock(&kvm_count_lock
);
3440 if (kvm_usage_count
)
3441 hardware_enable_nolock(NULL
);
3442 raw_spin_unlock(&kvm_count_lock
);
3446 static void hardware_disable_nolock(void *junk
)
3448 int cpu
= raw_smp_processor_id();
3450 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3452 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3453 kvm_arch_hardware_disable();
3456 static int kvm_dying_cpu(unsigned int cpu
)
3458 raw_spin_lock(&kvm_count_lock
);
3459 if (kvm_usage_count
)
3460 hardware_disable_nolock(NULL
);
3461 raw_spin_unlock(&kvm_count_lock
);
3465 static void hardware_disable_all_nolock(void)
3467 BUG_ON(!kvm_usage_count
);
3470 if (!kvm_usage_count
)
3471 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3474 static void hardware_disable_all(void)
3476 raw_spin_lock(&kvm_count_lock
);
3477 hardware_disable_all_nolock();
3478 raw_spin_unlock(&kvm_count_lock
);
3481 static int hardware_enable_all(void)
3485 raw_spin_lock(&kvm_count_lock
);
3488 if (kvm_usage_count
== 1) {
3489 atomic_set(&hardware_enable_failed
, 0);
3490 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3492 if (atomic_read(&hardware_enable_failed
)) {
3493 hardware_disable_all_nolock();
3498 raw_spin_unlock(&kvm_count_lock
);
3503 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3507 * Some (well, at least mine) BIOSes hang on reboot if
3510 * And Intel TXT required VMX off for all cpu when system shutdown.
3512 pr_info("kvm: exiting hardware virtualization\n");
3513 kvm_rebooting
= true;
3514 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3518 static struct notifier_block kvm_reboot_notifier
= {
3519 .notifier_call
= kvm_reboot
,
3523 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3527 for (i
= 0; i
< bus
->dev_count
; i
++) {
3528 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3530 kvm_iodevice_destructor(pos
);
3535 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3536 const struct kvm_io_range
*r2
)
3538 gpa_t addr1
= r1
->addr
;
3539 gpa_t addr2
= r2
->addr
;
3544 /* If r2->len == 0, match the exact address. If r2->len != 0,
3545 * accept any overlapping write. Any order is acceptable for
3546 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3547 * we process all of them.
3560 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3562 return kvm_io_bus_cmp(p1
, p2
);
3565 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3566 gpa_t addr
, int len
)
3568 struct kvm_io_range
*range
, key
;
3571 key
= (struct kvm_io_range
) {
3576 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3577 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3581 off
= range
- bus
->range
;
3583 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3589 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3590 struct kvm_io_range
*range
, const void *val
)
3594 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3598 while (idx
< bus
->dev_count
&&
3599 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3600 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3609 /* kvm_io_bus_write - called under kvm->slots_lock */
3610 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3611 int len
, const void *val
)
3613 struct kvm_io_bus
*bus
;
3614 struct kvm_io_range range
;
3617 range
= (struct kvm_io_range
) {
3622 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3625 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3626 return r
< 0 ? r
: 0;
3629 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3630 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3631 gpa_t addr
, int len
, const void *val
, long cookie
)
3633 struct kvm_io_bus
*bus
;
3634 struct kvm_io_range range
;
3636 range
= (struct kvm_io_range
) {
3641 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3645 /* First try the device referenced by cookie. */
3646 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3647 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3648 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3653 * cookie contained garbage; fall back to search and return the
3654 * correct cookie value.
3656 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3659 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3660 struct kvm_io_range
*range
, void *val
)
3664 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3668 while (idx
< bus
->dev_count
&&
3669 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3670 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3678 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3680 /* kvm_io_bus_read - called under kvm->slots_lock */
3681 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3684 struct kvm_io_bus
*bus
;
3685 struct kvm_io_range range
;
3688 range
= (struct kvm_io_range
) {
3693 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3696 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3697 return r
< 0 ? r
: 0;
3701 /* Caller must hold slots_lock. */
3702 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3703 int len
, struct kvm_io_device
*dev
)
3706 struct kvm_io_bus
*new_bus
, *bus
;
3707 struct kvm_io_range range
;
3709 bus
= kvm_get_bus(kvm
, bus_idx
);
3713 /* exclude ioeventfd which is limited by maximum fd */
3714 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3717 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3718 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3722 range
= (struct kvm_io_range
) {
3728 for (i
= 0; i
< bus
->dev_count
; i
++)
3729 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
3732 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3733 new_bus
->dev_count
++;
3734 new_bus
->range
[i
] = range
;
3735 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
3736 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3737 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3738 synchronize_srcu_expedited(&kvm
->srcu
);
3744 /* Caller must hold slots_lock. */
3745 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3746 struct kvm_io_device
*dev
)
3749 struct kvm_io_bus
*new_bus
, *bus
;
3751 bus
= kvm_get_bus(kvm
, bus_idx
);
3755 for (i
= 0; i
< bus
->dev_count
; i
++)
3756 if (bus
->range
[i
].dev
== dev
) {
3760 if (i
== bus
->dev_count
)
3763 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3764 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3766 pr_err("kvm: failed to shrink bus, removing it completely\n");
3770 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3771 new_bus
->dev_count
--;
3772 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3773 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3776 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3777 synchronize_srcu_expedited(&kvm
->srcu
);
3782 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3785 struct kvm_io_bus
*bus
;
3786 int dev_idx
, srcu_idx
;
3787 struct kvm_io_device
*iodev
= NULL
;
3789 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3791 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3795 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3799 iodev
= bus
->range
[dev_idx
].dev
;
3802 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3806 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3808 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3809 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3812 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3815 /* The debugfs files are a reference to the kvm struct which
3816 * is still valid when kvm_destroy_vm is called.
3817 * To avoid the race between open and the removal of the debugfs
3818 * directory we test against the users count.
3820 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
3823 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3824 kvm_put_kvm(stat_data
->kvm
);
3831 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3833 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3836 simple_attr_release(inode
, file
);
3837 kvm_put_kvm(stat_data
->kvm
);
3842 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3844 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3846 *val
= *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3851 static int vm_stat_clear_per_vm(void *data
, u64 val
)
3853 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3858 *(ulong
*)((void *)stat_data
->kvm
+ stat_data
->offset
) = 0;
3863 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3865 __simple_attr_check_format("%llu\n", 0ull);
3866 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3867 vm_stat_clear_per_vm
, "%llu\n");
3870 static const struct file_operations vm_stat_get_per_vm_fops
= {
3871 .owner
= THIS_MODULE
,
3872 .open
= vm_stat_get_per_vm_open
,
3873 .release
= kvm_debugfs_release
,
3874 .read
= simple_attr_read
,
3875 .write
= simple_attr_write
,
3876 .llseek
= no_llseek
,
3879 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3882 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3883 struct kvm_vcpu
*vcpu
;
3887 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3888 *val
+= *(u64
*)((void *)vcpu
+ stat_data
->offset
);
3893 static int vcpu_stat_clear_per_vm(void *data
, u64 val
)
3896 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3897 struct kvm_vcpu
*vcpu
;
3902 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3903 *(u64
*)((void *)vcpu
+ stat_data
->offset
) = 0;
3908 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3910 __simple_attr_check_format("%llu\n", 0ull);
3911 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3912 vcpu_stat_clear_per_vm
, "%llu\n");
3915 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3916 .owner
= THIS_MODULE
,
3917 .open
= vcpu_stat_get_per_vm_open
,
3918 .release
= kvm_debugfs_release
,
3919 .read
= simple_attr_read
,
3920 .write
= simple_attr_write
,
3921 .llseek
= no_llseek
,
3924 static const struct file_operations
*stat_fops_per_vm
[] = {
3925 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3926 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3929 static int vm_stat_get(void *_offset
, u64
*val
)
3931 unsigned offset
= (long)_offset
;
3933 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3937 spin_lock(&kvm_lock
);
3938 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3940 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3943 spin_unlock(&kvm_lock
);
3947 static int vm_stat_clear(void *_offset
, u64 val
)
3949 unsigned offset
= (long)_offset
;
3951 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3956 spin_lock(&kvm_lock
);
3957 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3959 vm_stat_clear_per_vm((void *)&stat_tmp
, 0);
3961 spin_unlock(&kvm_lock
);
3966 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
3968 static int vcpu_stat_get(void *_offset
, u64
*val
)
3970 unsigned offset
= (long)_offset
;
3972 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3976 spin_lock(&kvm_lock
);
3977 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3979 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3982 spin_unlock(&kvm_lock
);
3986 static int vcpu_stat_clear(void *_offset
, u64 val
)
3988 unsigned offset
= (long)_offset
;
3990 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3995 spin_lock(&kvm_lock
);
3996 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3998 vcpu_stat_clear_per_vm((void *)&stat_tmp
, 0);
4000 spin_unlock(&kvm_lock
);
4005 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
4008 static const struct file_operations
*stat_fops
[] = {
4009 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
4010 [KVM_STAT_VM
] = &vm_stat_fops
,
4013 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
4015 struct kobj_uevent_env
*env
;
4016 unsigned long long created
, active
;
4018 if (!kvm_dev
.this_device
|| !kvm
)
4021 spin_lock(&kvm_lock
);
4022 if (type
== KVM_EVENT_CREATE_VM
) {
4023 kvm_createvm_count
++;
4025 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4028 created
= kvm_createvm_count
;
4029 active
= kvm_active_vms
;
4030 spin_unlock(&kvm_lock
);
4032 env
= kzalloc(sizeof(*env
), GFP_KERNEL
);
4036 add_uevent_var(env
, "CREATED=%llu", created
);
4037 add_uevent_var(env
, "COUNT=%llu", active
);
4039 if (type
== KVM_EVENT_CREATE_VM
) {
4040 add_uevent_var(env
, "EVENT=create");
4041 kvm
->userspace_pid
= task_pid_nr(current
);
4042 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4043 add_uevent_var(env
, "EVENT=destroy");
4045 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
4047 if (kvm
->debugfs_dentry
) {
4048 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL
);
4051 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
4053 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
4057 /* no need for checks, since we are adding at most only 5 keys */
4058 env
->envp
[env
->envp_idx
++] = NULL
;
4059 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
4063 static void kvm_init_debug(void)
4065 struct kvm_stats_debugfs_item
*p
;
4067 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
4069 kvm_debugfs_num_entries
= 0;
4070 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
4071 debugfs_create_file(p
->name
, 0644, kvm_debugfs_dir
,
4072 (void *)(long)p
->offset
,
4073 stat_fops
[p
->kind
]);
4077 static int kvm_suspend(void)
4079 if (kvm_usage_count
)
4080 hardware_disable_nolock(NULL
);
4084 static void kvm_resume(void)
4086 if (kvm_usage_count
) {
4087 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
4088 hardware_enable_nolock(NULL
);
4092 static struct syscore_ops kvm_syscore_ops
= {
4093 .suspend
= kvm_suspend
,
4094 .resume
= kvm_resume
,
4098 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4100 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4103 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4105 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4107 if (vcpu
->preempted
)
4108 vcpu
->preempted
= false;
4110 kvm_arch_sched_in(vcpu
, cpu
);
4112 kvm_arch_vcpu_load(vcpu
, cpu
);
4115 static void kvm_sched_out(struct preempt_notifier
*pn
,
4116 struct task_struct
*next
)
4118 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4120 if (current
->state
== TASK_RUNNING
)
4121 vcpu
->preempted
= true;
4122 kvm_arch_vcpu_put(vcpu
);
4125 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4126 struct module
*module
)
4131 r
= kvm_arch_init(opaque
);
4136 * kvm_arch_init makes sure there's at most one caller
4137 * for architectures that support multiple implementations,
4138 * like intel and amd on x86.
4139 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4140 * conflicts in case kvm is already setup for another implementation.
4142 r
= kvm_irqfd_init();
4146 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4151 r
= kvm_arch_hardware_setup();
4155 for_each_online_cpu(cpu
) {
4156 smp_call_function_single(cpu
,
4157 kvm_arch_check_processor_compat
,
4163 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4164 kvm_starting_cpu
, kvm_dying_cpu
);
4167 register_reboot_notifier(&kvm_reboot_notifier
);
4169 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4171 vcpu_align
= __alignof__(struct kvm_vcpu
);
4173 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
4175 offsetof(struct kvm_vcpu
, arch
),
4176 sizeof_field(struct kvm_vcpu
, arch
),
4178 if (!kvm_vcpu_cache
) {
4183 r
= kvm_async_pf_init();
4187 kvm_chardev_ops
.owner
= module
;
4188 kvm_vm_fops
.owner
= module
;
4189 kvm_vcpu_fops
.owner
= module
;
4191 r
= misc_register(&kvm_dev
);
4193 pr_err("kvm: misc device register failed\n");
4197 register_syscore_ops(&kvm_syscore_ops
);
4199 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4200 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4204 r
= kvm_vfio_ops_init();
4210 kvm_async_pf_deinit();
4212 kmem_cache_destroy(kvm_vcpu_cache
);
4214 unregister_reboot_notifier(&kvm_reboot_notifier
);
4215 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4218 kvm_arch_hardware_unsetup();
4220 free_cpumask_var(cpus_hardware_enabled
);
4228 EXPORT_SYMBOL_GPL(kvm_init
);
4232 debugfs_remove_recursive(kvm_debugfs_dir
);
4233 misc_deregister(&kvm_dev
);
4234 kmem_cache_destroy(kvm_vcpu_cache
);
4235 kvm_async_pf_deinit();
4236 unregister_syscore_ops(&kvm_syscore_ops
);
4237 unregister_reboot_notifier(&kvm_reboot_notifier
);
4238 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4239 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4240 kvm_arch_hardware_unsetup();
4243 free_cpumask_var(cpus_hardware_enabled
);
4244 kvm_vfio_ops_exit();
4246 EXPORT_SYMBOL_GPL(kvm_exit
);