1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 #include <linux/kvm_dirty_ring.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 /* The start value to grow halt_poll_ns from */
85 unsigned int halt_poll_ns_grow_start
= 10000; /* 10us */
86 module_param(halt_poll_ns_grow_start
, uint
, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start
);
89 /* Default resets per-vcpu halt_poll_ns . */
90 unsigned int halt_poll_ns_shrink
;
91 module_param(halt_poll_ns_shrink
, uint
, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink
);
97 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
100 DEFINE_MUTEX(kvm_lock
);
101 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
104 static cpumask_var_t cpus_hardware_enabled
;
105 static int kvm_usage_count
;
106 static atomic_t hardware_enable_failed
;
108 static struct kmem_cache
*kvm_vcpu_cache
;
110 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
111 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_running_vcpu
);
113 struct dentry
*kvm_debugfs_dir
;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
116 static int kvm_debugfs_num_entries
;
117 static const struct file_operations stat_fops_per_vm
;
119 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
121 #ifdef CONFIG_KVM_COMPAT
122 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
124 #define KVM_COMPAT(c) .compat_ioctl = (c)
127 * For architectures that don't implement a compat infrastructure,
128 * adopt a double line of defense:
129 * - Prevent a compat task from opening /dev/kvm
130 * - If the open has been done by a 64bit task, and the KVM fd
131 * passed to a compat task, let the ioctls fail.
133 static long kvm_no_compat_ioctl(struct file
*file
, unsigned int ioctl
,
134 unsigned long arg
) { return -EINVAL
; }
136 static int kvm_no_compat_open(struct inode
*inode
, struct file
*file
)
138 return is_compat_task() ? -ENODEV
: 0;
140 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
141 .open = kvm_no_compat_open
143 static int hardware_enable_all(void);
144 static void hardware_disable_all(void);
146 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
148 __visible
bool kvm_rebooting
;
149 EXPORT_SYMBOL_GPL(kvm_rebooting
);
151 #define KVM_EVENT_CREATE_VM 0
152 #define KVM_EVENT_DESTROY_VM 1
153 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
);
154 static unsigned long long kvm_createvm_count
;
155 static unsigned long long kvm_active_vms
;
157 __weak
void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
158 unsigned long start
, unsigned long end
)
162 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn
)
165 * The metadata used by is_zone_device_page() to determine whether or
166 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
167 * the device has been pinned, e.g. by get_user_pages(). WARN if the
168 * page_count() is zero to help detect bad usage of this helper.
170 if (!pfn_valid(pfn
) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn
))))
173 return is_zone_device_page(pfn_to_page(pfn
));
176 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
179 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
180 * perspective they are "normal" pages, albeit with slightly different
184 return PageReserved(pfn_to_page(pfn
)) &&
186 !kvm_is_zone_device_pfn(pfn
);
191 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn
)
193 struct page
*page
= pfn_to_page(pfn
);
195 if (!PageTransCompoundMap(page
))
198 return is_transparent_hugepage(compound_head(page
));
202 * Switches to specified vcpu, until a matching vcpu_put()
204 void vcpu_load(struct kvm_vcpu
*vcpu
)
208 __this_cpu_write(kvm_running_vcpu
, vcpu
);
209 preempt_notifier_register(&vcpu
->preempt_notifier
);
210 kvm_arch_vcpu_load(vcpu
, cpu
);
213 EXPORT_SYMBOL_GPL(vcpu_load
);
215 void vcpu_put(struct kvm_vcpu
*vcpu
)
218 kvm_arch_vcpu_put(vcpu
);
219 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
220 __this_cpu_write(kvm_running_vcpu
, NULL
);
223 EXPORT_SYMBOL_GPL(vcpu_put
);
225 /* TODO: merge with kvm_arch_vcpu_should_kick */
226 static bool kvm_request_needs_ipi(struct kvm_vcpu
*vcpu
, unsigned req
)
228 int mode
= kvm_vcpu_exiting_guest_mode(vcpu
);
231 * We need to wait for the VCPU to reenable interrupts and get out of
232 * READING_SHADOW_PAGE_TABLES mode.
234 if (req
& KVM_REQUEST_WAIT
)
235 return mode
!= OUTSIDE_GUEST_MODE
;
238 * Need to kick a running VCPU, but otherwise there is nothing to do.
240 return mode
== IN_GUEST_MODE
;
243 static void ack_flush(void *_completed
)
247 static inline bool kvm_kick_many_cpus(const struct cpumask
*cpus
, bool wait
)
250 cpus
= cpu_online_mask
;
252 if (cpumask_empty(cpus
))
255 smp_call_function_many(cpus
, ack_flush
, NULL
, wait
);
259 bool kvm_make_vcpus_request_mask(struct kvm
*kvm
, unsigned int req
,
260 struct kvm_vcpu
*except
,
261 unsigned long *vcpu_bitmap
, cpumask_var_t tmp
)
264 struct kvm_vcpu
*vcpu
;
269 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
270 if ((vcpu_bitmap
&& !test_bit(i
, vcpu_bitmap
)) ||
274 kvm_make_request(req
, vcpu
);
277 if (!(req
& KVM_REQUEST_NO_WAKEUP
) && kvm_vcpu_wake_up(vcpu
))
280 if (tmp
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
281 kvm_request_needs_ipi(vcpu
, req
))
282 __cpumask_set_cpu(cpu
, tmp
);
285 called
= kvm_kick_many_cpus(tmp
, !!(req
& KVM_REQUEST_WAIT
));
291 bool kvm_make_all_cpus_request_except(struct kvm
*kvm
, unsigned int req
,
292 struct kvm_vcpu
*except
)
297 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
299 called
= kvm_make_vcpus_request_mask(kvm
, req
, except
, NULL
, cpus
);
301 free_cpumask_var(cpus
);
305 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
307 return kvm_make_all_cpus_request_except(kvm
, req
, NULL
);
310 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
311 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
314 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
315 * kvm_make_all_cpus_request.
317 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
320 * We want to publish modifications to the page tables before reading
321 * mode. Pairs with a memory barrier in arch-specific code.
322 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
323 * and smp_mb in walk_shadow_page_lockless_begin/end.
324 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
326 * There is already an smp_mb__after_atomic() before
327 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
330 if (!kvm_arch_flush_remote_tlb(kvm
)
331 || kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
332 ++kvm
->stat
.remote_tlb_flush
;
333 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
335 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
338 void kvm_reload_remote_mmus(struct kvm
*kvm
)
340 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
343 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
344 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache
*mc
,
347 gfp_flags
|= mc
->gfp_zero
;
350 return kmem_cache_alloc(mc
->kmem_cache
, gfp_flags
);
352 return (void *)__get_free_page(gfp_flags
);
355 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*mc
, int min
)
359 if (mc
->nobjs
>= min
)
361 while (mc
->nobjs
< ARRAY_SIZE(mc
->objects
)) {
362 obj
= mmu_memory_cache_alloc_obj(mc
, GFP_KERNEL_ACCOUNT
);
364 return mc
->nobjs
>= min
? 0 : -ENOMEM
;
365 mc
->objects
[mc
->nobjs
++] = obj
;
370 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache
*mc
)
375 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
379 kmem_cache_free(mc
->kmem_cache
, mc
->objects
[--mc
->nobjs
]);
381 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
385 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
)
389 if (WARN_ON(!mc
->nobjs
))
390 p
= mmu_memory_cache_alloc_obj(mc
, GFP_ATOMIC
| __GFP_ACCOUNT
);
392 p
= mc
->objects
[--mc
->nobjs
];
398 static void kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
400 mutex_init(&vcpu
->mutex
);
405 rcuwait_init(&vcpu
->wait
);
406 kvm_async_pf_vcpu_init(vcpu
);
409 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
411 kvm_vcpu_set_in_spin_loop(vcpu
, false);
412 kvm_vcpu_set_dy_eligible(vcpu
, false);
413 vcpu
->preempted
= false;
415 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
418 void kvm_vcpu_destroy(struct kvm_vcpu
*vcpu
)
420 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
421 kvm_arch_vcpu_destroy(vcpu
);
424 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
425 * the vcpu->pid pointer, and at destruction time all file descriptors
428 put_pid(rcu_dereference_protected(vcpu
->pid
, 1));
430 free_page((unsigned long)vcpu
->run
);
431 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
433 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy
);
435 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
436 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
438 return container_of(mn
, struct kvm
, mmu_notifier
);
441 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier
*mn
,
442 struct mm_struct
*mm
,
443 unsigned long start
, unsigned long end
)
445 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
448 idx
= srcu_read_lock(&kvm
->srcu
);
449 kvm_arch_mmu_notifier_invalidate_range(kvm
, start
, end
);
450 srcu_read_unlock(&kvm
->srcu
, idx
);
453 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
454 struct mm_struct
*mm
,
455 unsigned long address
,
458 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
461 idx
= srcu_read_lock(&kvm
->srcu
);
462 spin_lock(&kvm
->mmu_lock
);
463 kvm
->mmu_notifier_seq
++;
465 if (kvm_set_spte_hva(kvm
, address
, pte
))
466 kvm_flush_remote_tlbs(kvm
);
468 spin_unlock(&kvm
->mmu_lock
);
469 srcu_read_unlock(&kvm
->srcu
, idx
);
472 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
473 const struct mmu_notifier_range
*range
)
475 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
476 int need_tlb_flush
= 0, idx
;
478 idx
= srcu_read_lock(&kvm
->srcu
);
479 spin_lock(&kvm
->mmu_lock
);
481 * The count increase must become visible at unlock time as no
482 * spte can be established without taking the mmu_lock and
483 * count is also read inside the mmu_lock critical section.
485 kvm
->mmu_notifier_count
++;
486 need_tlb_flush
= kvm_unmap_hva_range(kvm
, range
->start
, range
->end
,
488 need_tlb_flush
|= kvm
->tlbs_dirty
;
489 /* we've to flush the tlb before the pages can be freed */
491 kvm_flush_remote_tlbs(kvm
);
493 spin_unlock(&kvm
->mmu_lock
);
494 srcu_read_unlock(&kvm
->srcu
, idx
);
499 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
500 const struct mmu_notifier_range
*range
)
502 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
504 spin_lock(&kvm
->mmu_lock
);
506 * This sequence increase will notify the kvm page fault that
507 * the page that is going to be mapped in the spte could have
510 kvm
->mmu_notifier_seq
++;
513 * The above sequence increase must be visible before the
514 * below count decrease, which is ensured by the smp_wmb above
515 * in conjunction with the smp_rmb in mmu_notifier_retry().
517 kvm
->mmu_notifier_count
--;
518 spin_unlock(&kvm
->mmu_lock
);
520 BUG_ON(kvm
->mmu_notifier_count
< 0);
523 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
524 struct mm_struct
*mm
,
528 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
531 idx
= srcu_read_lock(&kvm
->srcu
);
532 spin_lock(&kvm
->mmu_lock
);
534 young
= kvm_age_hva(kvm
, start
, end
);
536 kvm_flush_remote_tlbs(kvm
);
538 spin_unlock(&kvm
->mmu_lock
);
539 srcu_read_unlock(&kvm
->srcu
, idx
);
544 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
545 struct mm_struct
*mm
,
549 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
552 idx
= srcu_read_lock(&kvm
->srcu
);
553 spin_lock(&kvm
->mmu_lock
);
555 * Even though we do not flush TLB, this will still adversely
556 * affect performance on pre-Haswell Intel EPT, where there is
557 * no EPT Access Bit to clear so that we have to tear down EPT
558 * tables instead. If we find this unacceptable, we can always
559 * add a parameter to kvm_age_hva so that it effectively doesn't
560 * do anything on clear_young.
562 * Also note that currently we never issue secondary TLB flushes
563 * from clear_young, leaving this job up to the regular system
564 * cadence. If we find this inaccurate, we might come up with a
565 * more sophisticated heuristic later.
567 young
= kvm_age_hva(kvm
, start
, end
);
568 spin_unlock(&kvm
->mmu_lock
);
569 srcu_read_unlock(&kvm
->srcu
, idx
);
574 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
575 struct mm_struct
*mm
,
576 unsigned long address
)
578 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
581 idx
= srcu_read_lock(&kvm
->srcu
);
582 spin_lock(&kvm
->mmu_lock
);
583 young
= kvm_test_age_hva(kvm
, address
);
584 spin_unlock(&kvm
->mmu_lock
);
585 srcu_read_unlock(&kvm
->srcu
, idx
);
590 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
591 struct mm_struct
*mm
)
593 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
596 idx
= srcu_read_lock(&kvm
->srcu
);
597 kvm_arch_flush_shadow_all(kvm
);
598 srcu_read_unlock(&kvm
->srcu
, idx
);
601 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
602 .invalidate_range
= kvm_mmu_notifier_invalidate_range
,
603 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
604 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
605 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
606 .clear_young
= kvm_mmu_notifier_clear_young
,
607 .test_young
= kvm_mmu_notifier_test_young
,
608 .change_pte
= kvm_mmu_notifier_change_pte
,
609 .release
= kvm_mmu_notifier_release
,
612 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
614 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
615 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
618 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
620 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
625 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
627 static struct kvm_memslots
*kvm_alloc_memslots(void)
630 struct kvm_memslots
*slots
;
632 slots
= kvzalloc(sizeof(struct kvm_memslots
), GFP_KERNEL_ACCOUNT
);
636 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
637 slots
->id_to_index
[i
] = -1;
642 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
644 if (!memslot
->dirty_bitmap
)
647 kvfree(memslot
->dirty_bitmap
);
648 memslot
->dirty_bitmap
= NULL
;
651 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
653 kvm_destroy_dirty_bitmap(slot
);
655 kvm_arch_free_memslot(kvm
, slot
);
661 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
663 struct kvm_memory_slot
*memslot
;
668 kvm_for_each_memslot(memslot
, slots
)
669 kvm_free_memslot(kvm
, memslot
);
674 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
678 if (!kvm
->debugfs_dentry
)
681 debugfs_remove_recursive(kvm
->debugfs_dentry
);
683 if (kvm
->debugfs_stat_data
) {
684 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
685 kfree(kvm
->debugfs_stat_data
[i
]);
686 kfree(kvm
->debugfs_stat_data
);
690 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
692 char dir_name
[ITOA_MAX_LEN
* 2];
693 struct kvm_stat_data
*stat_data
;
694 struct kvm_stats_debugfs_item
*p
;
696 if (!debugfs_initialized())
699 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
700 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
, kvm_debugfs_dir
);
702 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
703 sizeof(*kvm
->debugfs_stat_data
),
705 if (!kvm
->debugfs_stat_data
)
708 for (p
= debugfs_entries
; p
->name
; p
++) {
709 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL_ACCOUNT
);
713 stat_data
->kvm
= kvm
;
714 stat_data
->dbgfs_item
= p
;
715 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
716 debugfs_create_file(p
->name
, KVM_DBGFS_GET_MODE(p
),
717 kvm
->debugfs_dentry
, stat_data
,
724 * Called after the VM is otherwise initialized, but just before adding it to
727 int __weak
kvm_arch_post_init_vm(struct kvm
*kvm
)
733 * Called just after removing the VM from the vm_list, but before doing any
736 void __weak
kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
740 static struct kvm
*kvm_create_vm(unsigned long type
)
742 struct kvm
*kvm
= kvm_arch_alloc_vm();
747 return ERR_PTR(-ENOMEM
);
749 spin_lock_init(&kvm
->mmu_lock
);
751 kvm
->mm
= current
->mm
;
752 kvm_eventfd_init(kvm
);
753 mutex_init(&kvm
->lock
);
754 mutex_init(&kvm
->irq_lock
);
755 mutex_init(&kvm
->slots_lock
);
756 INIT_LIST_HEAD(&kvm
->devices
);
758 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
760 if (init_srcu_struct(&kvm
->srcu
))
761 goto out_err_no_srcu
;
762 if (init_srcu_struct(&kvm
->irq_srcu
))
763 goto out_err_no_irq_srcu
;
765 refcount_set(&kvm
->users_count
, 1);
766 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
767 struct kvm_memslots
*slots
= kvm_alloc_memslots();
770 goto out_err_no_arch_destroy_vm
;
771 /* Generations must be different for each address space. */
772 slots
->generation
= i
;
773 rcu_assign_pointer(kvm
->memslots
[i
], slots
);
776 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
777 rcu_assign_pointer(kvm
->buses
[i
],
778 kzalloc(sizeof(struct kvm_io_bus
), GFP_KERNEL_ACCOUNT
));
780 goto out_err_no_arch_destroy_vm
;
783 kvm
->max_halt_poll_ns
= halt_poll_ns
;
785 r
= kvm_arch_init_vm(kvm
, type
);
787 goto out_err_no_arch_destroy_vm
;
789 r
= hardware_enable_all();
791 goto out_err_no_disable
;
793 #ifdef CONFIG_HAVE_KVM_IRQFD
794 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
797 r
= kvm_init_mmu_notifier(kvm
);
799 goto out_err_no_mmu_notifier
;
801 r
= kvm_arch_post_init_vm(kvm
);
805 mutex_lock(&kvm_lock
);
806 list_add(&kvm
->vm_list
, &vm_list
);
807 mutex_unlock(&kvm_lock
);
809 preempt_notifier_inc();
814 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
815 if (kvm
->mmu_notifier
.ops
)
816 mmu_notifier_unregister(&kvm
->mmu_notifier
, current
->mm
);
818 out_err_no_mmu_notifier
:
819 hardware_disable_all();
821 kvm_arch_destroy_vm(kvm
);
822 out_err_no_arch_destroy_vm
:
823 WARN_ON_ONCE(!refcount_dec_and_test(&kvm
->users_count
));
824 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
825 kfree(kvm_get_bus(kvm
, i
));
826 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
827 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
828 cleanup_srcu_struct(&kvm
->irq_srcu
);
830 cleanup_srcu_struct(&kvm
->srcu
);
832 kvm_arch_free_vm(kvm
);
837 static void kvm_destroy_devices(struct kvm
*kvm
)
839 struct kvm_device
*dev
, *tmp
;
842 * We do not need to take the kvm->lock here, because nobody else
843 * has a reference to the struct kvm at this point and therefore
844 * cannot access the devices list anyhow.
846 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
847 list_del(&dev
->vm_node
);
848 dev
->ops
->destroy(dev
);
852 static void kvm_destroy_vm(struct kvm
*kvm
)
855 struct mm_struct
*mm
= kvm
->mm
;
857 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM
, kvm
);
858 kvm_destroy_vm_debugfs(kvm
);
859 kvm_arch_sync_events(kvm
);
860 mutex_lock(&kvm_lock
);
861 list_del(&kvm
->vm_list
);
862 mutex_unlock(&kvm_lock
);
863 kvm_arch_pre_destroy_vm(kvm
);
865 kvm_free_irq_routing(kvm
);
866 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
867 struct kvm_io_bus
*bus
= kvm_get_bus(kvm
, i
);
870 kvm_io_bus_destroy(bus
);
871 kvm
->buses
[i
] = NULL
;
873 kvm_coalesced_mmio_free(kvm
);
874 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
875 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
877 kvm_arch_flush_shadow_all(kvm
);
879 kvm_arch_destroy_vm(kvm
);
880 kvm_destroy_devices(kvm
);
881 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
882 kvm_free_memslots(kvm
, __kvm_memslots(kvm
, i
));
883 cleanup_srcu_struct(&kvm
->irq_srcu
);
884 cleanup_srcu_struct(&kvm
->srcu
);
885 kvm_arch_free_vm(kvm
);
886 preempt_notifier_dec();
887 hardware_disable_all();
891 void kvm_get_kvm(struct kvm
*kvm
)
893 refcount_inc(&kvm
->users_count
);
895 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
897 void kvm_put_kvm(struct kvm
*kvm
)
899 if (refcount_dec_and_test(&kvm
->users_count
))
902 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
905 * Used to put a reference that was taken on behalf of an object associated
906 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
907 * of the new file descriptor fails and the reference cannot be transferred to
908 * its final owner. In such cases, the caller is still actively using @kvm and
909 * will fail miserably if the refcount unexpectedly hits zero.
911 void kvm_put_kvm_no_destroy(struct kvm
*kvm
)
913 WARN_ON(refcount_dec_and_test(&kvm
->users_count
));
915 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy
);
917 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
919 struct kvm
*kvm
= filp
->private_data
;
921 kvm_irqfd_release(kvm
);
928 * Allocation size is twice as large as the actual dirty bitmap size.
929 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
931 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot
*memslot
)
933 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
935 memslot
->dirty_bitmap
= kvzalloc(dirty_bytes
, GFP_KERNEL_ACCOUNT
);
936 if (!memslot
->dirty_bitmap
)
943 * Delete a memslot by decrementing the number of used slots and shifting all
944 * other entries in the array forward one spot.
946 static inline void kvm_memslot_delete(struct kvm_memslots
*slots
,
947 struct kvm_memory_slot
*memslot
)
949 struct kvm_memory_slot
*mslots
= slots
->memslots
;
952 if (WARN_ON(slots
->id_to_index
[memslot
->id
] == -1))
957 if (atomic_read(&slots
->lru_slot
) >= slots
->used_slots
)
958 atomic_set(&slots
->lru_slot
, 0);
960 for (i
= slots
->id_to_index
[memslot
->id
]; i
< slots
->used_slots
; i
++) {
961 mslots
[i
] = mslots
[i
+ 1];
962 slots
->id_to_index
[mslots
[i
].id
] = i
;
964 mslots
[i
] = *memslot
;
965 slots
->id_to_index
[memslot
->id
] = -1;
969 * "Insert" a new memslot by incrementing the number of used slots. Returns
970 * the new slot's initial index into the memslots array.
972 static inline int kvm_memslot_insert_back(struct kvm_memslots
*slots
)
974 return slots
->used_slots
++;
978 * Move a changed memslot backwards in the array by shifting existing slots
979 * with a higher GFN toward the front of the array. Note, the changed memslot
980 * itself is not preserved in the array, i.e. not swapped at this time, only
981 * its new index into the array is tracked. Returns the changed memslot's
982 * current index into the memslots array.
984 static inline int kvm_memslot_move_backward(struct kvm_memslots
*slots
,
985 struct kvm_memory_slot
*memslot
)
987 struct kvm_memory_slot
*mslots
= slots
->memslots
;
990 if (WARN_ON_ONCE(slots
->id_to_index
[memslot
->id
] == -1) ||
991 WARN_ON_ONCE(!slots
->used_slots
))
995 * Move the target memslot backward in the array by shifting existing
996 * memslots with a higher GFN (than the target memslot) towards the
997 * front of the array.
999 for (i
= slots
->id_to_index
[memslot
->id
]; i
< slots
->used_slots
- 1; i
++) {
1000 if (memslot
->base_gfn
> mslots
[i
+ 1].base_gfn
)
1003 WARN_ON_ONCE(memslot
->base_gfn
== mslots
[i
+ 1].base_gfn
);
1005 /* Shift the next memslot forward one and update its index. */
1006 mslots
[i
] = mslots
[i
+ 1];
1007 slots
->id_to_index
[mslots
[i
].id
] = i
;
1013 * Move a changed memslot forwards in the array by shifting existing slots with
1014 * a lower GFN toward the back of the array. Note, the changed memslot itself
1015 * is not preserved in the array, i.e. not swapped at this time, only its new
1016 * index into the array is tracked. Returns the changed memslot's final index
1017 * into the memslots array.
1019 static inline int kvm_memslot_move_forward(struct kvm_memslots
*slots
,
1020 struct kvm_memory_slot
*memslot
,
1023 struct kvm_memory_slot
*mslots
= slots
->memslots
;
1026 for (i
= start
; i
> 0; i
--) {
1027 if (memslot
->base_gfn
< mslots
[i
- 1].base_gfn
)
1030 WARN_ON_ONCE(memslot
->base_gfn
== mslots
[i
- 1].base_gfn
);
1032 /* Shift the next memslot back one and update its index. */
1033 mslots
[i
] = mslots
[i
- 1];
1034 slots
->id_to_index
[mslots
[i
].id
] = i
;
1040 * Re-sort memslots based on their GFN to account for an added, deleted, or
1041 * moved memslot. Sorting memslots by GFN allows using a binary search during
1044 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1045 * at memslots[0] has the highest GFN.
1047 * The sorting algorithm takes advantage of having initially sorted memslots
1048 * and knowing the position of the changed memslot. Sorting is also optimized
1049 * by not swapping the updated memslot and instead only shifting other memslots
1050 * and tracking the new index for the update memslot. Only once its final
1051 * index is known is the updated memslot copied into its position in the array.
1053 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1054 * the end of the array.
1056 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1057 * end of the array and then it forward to its correct location.
1059 * - When moving a memslot, the algorithm first moves the updated memslot
1060 * backward to handle the scenario where the memslot's GFN was changed to a
1061 * lower value. update_memslots() then falls through and runs the same flow
1062 * as creating a memslot to move the memslot forward to handle the scenario
1063 * where its GFN was changed to a higher value.
1065 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1066 * historical reasons. Originally, invalid memslots where denoted by having
1067 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1068 * to the end of the array. The current algorithm uses dedicated logic to
1069 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1071 * The other historical motiviation for highest->lowest was to improve the
1072 * performance of memslot lookup. KVM originally used a linear search starting
1073 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1074 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1075 * single memslot above the 4gb boundary. As the largest memslot is also the
1076 * most likely to be referenced, sorting it to the front of the array was
1077 * advantageous. The current binary search starts from the middle of the array
1078 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1080 static void update_memslots(struct kvm_memslots
*slots
,
1081 struct kvm_memory_slot
*memslot
,
1082 enum kvm_mr_change change
)
1086 if (change
== KVM_MR_DELETE
) {
1087 kvm_memslot_delete(slots
, memslot
);
1089 if (change
== KVM_MR_CREATE
)
1090 i
= kvm_memslot_insert_back(slots
);
1092 i
= kvm_memslot_move_backward(slots
, memslot
);
1093 i
= kvm_memslot_move_forward(slots
, memslot
, i
);
1096 * Copy the memslot to its new position in memslots and update
1097 * its index accordingly.
1099 slots
->memslots
[i
] = *memslot
;
1100 slots
->id_to_index
[memslot
->id
] = i
;
1104 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
1106 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
1108 #ifdef __KVM_HAVE_READONLY_MEM
1109 valid_flags
|= KVM_MEM_READONLY
;
1112 if (mem
->flags
& ~valid_flags
)
1118 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
1119 int as_id
, struct kvm_memslots
*slots
)
1121 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
1122 u64 gen
= old_memslots
->generation
;
1124 WARN_ON(gen
& KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
);
1125 slots
->generation
= gen
| KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1127 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
1128 synchronize_srcu_expedited(&kvm
->srcu
);
1131 * Increment the new memslot generation a second time, dropping the
1132 * update in-progress flag and incrementing the generation based on
1133 * the number of address spaces. This provides a unique and easily
1134 * identifiable generation number while the memslots are in flux.
1136 gen
= slots
->generation
& ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS
;
1139 * Generations must be unique even across address spaces. We do not need
1140 * a global counter for that, instead the generation space is evenly split
1141 * across address spaces. For example, with two address spaces, address
1142 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1143 * use generations 1, 3, 5, ...
1145 gen
+= KVM_ADDRESS_SPACE_NUM
;
1147 kvm_arch_memslots_updated(kvm
, gen
);
1149 slots
->generation
= gen
;
1151 return old_memslots
;
1155 * Note, at a minimum, the current number of used slots must be allocated, even
1156 * when deleting a memslot, as we need a complete duplicate of the memslots for
1157 * use when invalidating a memslot prior to deleting/moving the memslot.
1159 static struct kvm_memslots
*kvm_dup_memslots(struct kvm_memslots
*old
,
1160 enum kvm_mr_change change
)
1162 struct kvm_memslots
*slots
;
1163 size_t old_size
, new_size
;
1165 old_size
= sizeof(struct kvm_memslots
) +
1166 (sizeof(struct kvm_memory_slot
) * old
->used_slots
);
1168 if (change
== KVM_MR_CREATE
)
1169 new_size
= old_size
+ sizeof(struct kvm_memory_slot
);
1171 new_size
= old_size
;
1173 slots
= kvzalloc(new_size
, GFP_KERNEL_ACCOUNT
);
1175 memcpy(slots
, old
, old_size
);
1180 static int kvm_set_memslot(struct kvm
*kvm
,
1181 const struct kvm_userspace_memory_region
*mem
,
1182 struct kvm_memory_slot
*old
,
1183 struct kvm_memory_slot
*new, int as_id
,
1184 enum kvm_mr_change change
)
1186 struct kvm_memory_slot
*slot
;
1187 struct kvm_memslots
*slots
;
1190 slots
= kvm_dup_memslots(__kvm_memslots(kvm
, as_id
), change
);
1194 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
) {
1196 * Note, the INVALID flag needs to be in the appropriate entry
1197 * in the freshly allocated memslots, not in @old or @new.
1199 slot
= id_to_memslot(slots
, old
->id
);
1200 slot
->flags
|= KVM_MEMSLOT_INVALID
;
1203 * We can re-use the old memslots, the only difference from the
1204 * newly installed memslots is the invalid flag, which will get
1205 * dropped by update_memslots anyway. We'll also revert to the
1206 * old memslots if preparing the new memory region fails.
1208 slots
= install_new_memslots(kvm
, as_id
, slots
);
1210 /* From this point no new shadow pages pointing to a deleted,
1211 * or moved, memslot will be created.
1213 * validation of sp->gfn happens in:
1214 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1215 * - kvm_is_visible_gfn (mmu_check_root)
1217 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1220 r
= kvm_arch_prepare_memory_region(kvm
, new, mem
, change
);
1224 update_memslots(slots
, new, change
);
1225 slots
= install_new_memslots(kvm
, as_id
, slots
);
1227 kvm_arch_commit_memory_region(kvm
, mem
, old
, new, change
);
1233 if (change
== KVM_MR_DELETE
|| change
== KVM_MR_MOVE
)
1234 slots
= install_new_memslots(kvm
, as_id
, slots
);
1239 static int kvm_delete_memslot(struct kvm
*kvm
,
1240 const struct kvm_userspace_memory_region
*mem
,
1241 struct kvm_memory_slot
*old
, int as_id
)
1243 struct kvm_memory_slot
new;
1249 memset(&new, 0, sizeof(new));
1252 * This is only for debugging purpose; it should never be referenced
1253 * for a removed memslot.
1257 r
= kvm_set_memslot(kvm
, mem
, old
, &new, as_id
, KVM_MR_DELETE
);
1261 kvm_free_memslot(kvm
, old
);
1266 * Allocate some memory and give it an address in the guest physical address
1269 * Discontiguous memory is allowed, mostly for framebuffers.
1271 * Must be called holding kvm->slots_lock for write.
1273 int __kvm_set_memory_region(struct kvm
*kvm
,
1274 const struct kvm_userspace_memory_region
*mem
)
1276 struct kvm_memory_slot old
, new;
1277 struct kvm_memory_slot
*tmp
;
1278 enum kvm_mr_change change
;
1282 r
= check_memory_region_flags(mem
);
1286 as_id
= mem
->slot
>> 16;
1287 id
= (u16
)mem
->slot
;
1289 /* General sanity checks */
1290 if (mem
->memory_size
& (PAGE_SIZE
- 1))
1292 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
1294 /* We can read the guest memory with __xxx_user() later on. */
1295 if ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
1296 !access_ok((void __user
*)(unsigned long)mem
->userspace_addr
,
1299 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
1301 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
1305 * Make a full copy of the old memslot, the pointer will become stale
1306 * when the memslots are re-sorted by update_memslots(), and the old
1307 * memslot needs to be referenced after calling update_memslots(), e.g.
1308 * to free its resources and for arch specific behavior.
1310 tmp
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
1315 memset(&old
, 0, sizeof(old
));
1319 if (!mem
->memory_size
)
1320 return kvm_delete_memslot(kvm
, mem
, &old
, as_id
);
1324 new.base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
1325 new.npages
= mem
->memory_size
>> PAGE_SHIFT
;
1326 new.flags
= mem
->flags
;
1327 new.userspace_addr
= mem
->userspace_addr
;
1329 if (new.npages
> KVM_MEM_MAX_NR_PAGES
)
1333 change
= KVM_MR_CREATE
;
1334 new.dirty_bitmap
= NULL
;
1335 memset(&new.arch
, 0, sizeof(new.arch
));
1336 } else { /* Modify an existing slot. */
1337 if ((new.userspace_addr
!= old
.userspace_addr
) ||
1338 (new.npages
!= old
.npages
) ||
1339 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
1342 if (new.base_gfn
!= old
.base_gfn
)
1343 change
= KVM_MR_MOVE
;
1344 else if (new.flags
!= old
.flags
)
1345 change
= KVM_MR_FLAGS_ONLY
;
1346 else /* Nothing to change. */
1349 /* Copy dirty_bitmap and arch from the current memslot. */
1350 new.dirty_bitmap
= old
.dirty_bitmap
;
1351 memcpy(&new.arch
, &old
.arch
, sizeof(new.arch
));
1354 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1355 /* Check for overlaps */
1356 kvm_for_each_memslot(tmp
, __kvm_memslots(kvm
, as_id
)) {
1359 if (!((new.base_gfn
+ new.npages
<= tmp
->base_gfn
) ||
1360 (new.base_gfn
>= tmp
->base_gfn
+ tmp
->npages
)))
1365 /* Allocate/free page dirty bitmap as needed */
1366 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
1367 new.dirty_bitmap
= NULL
;
1368 else if (!new.dirty_bitmap
&& !kvm
->dirty_ring_size
) {
1369 r
= kvm_alloc_dirty_bitmap(&new);
1373 if (kvm_dirty_log_manual_protect_and_init_set(kvm
))
1374 bitmap_set(new.dirty_bitmap
, 0, new.npages
);
1377 r
= kvm_set_memslot(kvm
, mem
, &old
, &new, as_id
, change
);
1381 if (old
.dirty_bitmap
&& !new.dirty_bitmap
)
1382 kvm_destroy_dirty_bitmap(&old
);
1386 if (new.dirty_bitmap
&& !old
.dirty_bitmap
)
1387 kvm_destroy_dirty_bitmap(&new);
1390 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1392 int kvm_set_memory_region(struct kvm
*kvm
,
1393 const struct kvm_userspace_memory_region
*mem
)
1397 mutex_lock(&kvm
->slots_lock
);
1398 r
= __kvm_set_memory_region(kvm
, mem
);
1399 mutex_unlock(&kvm
->slots_lock
);
1402 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1404 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1405 struct kvm_userspace_memory_region
*mem
)
1407 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1410 return kvm_set_memory_region(kvm
, mem
);
1413 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1415 * kvm_get_dirty_log - get a snapshot of dirty pages
1416 * @kvm: pointer to kvm instance
1417 * @log: slot id and address to which we copy the log
1418 * @is_dirty: set to '1' if any dirty pages were found
1419 * @memslot: set to the associated memslot, always valid on success
1421 int kvm_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
,
1422 int *is_dirty
, struct kvm_memory_slot
**memslot
)
1424 struct kvm_memslots
*slots
;
1427 unsigned long any
= 0;
1429 /* Dirty ring tracking is exclusive to dirty log tracking */
1430 if (kvm
->dirty_ring_size
)
1436 as_id
= log
->slot
>> 16;
1437 id
= (u16
)log
->slot
;
1438 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1441 slots
= __kvm_memslots(kvm
, as_id
);
1442 *memslot
= id_to_memslot(slots
, id
);
1443 if (!(*memslot
) || !(*memslot
)->dirty_bitmap
)
1446 kvm_arch_sync_dirty_log(kvm
, *memslot
);
1448 n
= kvm_dirty_bitmap_bytes(*memslot
);
1450 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1451 any
= (*memslot
)->dirty_bitmap
[i
];
1453 if (copy_to_user(log
->dirty_bitmap
, (*memslot
)->dirty_bitmap
, n
))
1460 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1462 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1464 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1465 * and reenable dirty page tracking for the corresponding pages.
1466 * @kvm: pointer to kvm instance
1467 * @log: slot id and address to which we copy the log
1469 * We need to keep it in mind that VCPU threads can write to the bitmap
1470 * concurrently. So, to avoid losing track of dirty pages we keep the
1473 * 1. Take a snapshot of the bit and clear it if needed.
1474 * 2. Write protect the corresponding page.
1475 * 3. Copy the snapshot to the userspace.
1476 * 4. Upon return caller flushes TLB's if needed.
1478 * Between 2 and 4, the guest may write to the page using the remaining TLB
1479 * entry. This is not a problem because the page is reported dirty using
1480 * the snapshot taken before and step 4 ensures that writes done after
1481 * exiting to userspace will be logged for the next call.
1484 static int kvm_get_dirty_log_protect(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1486 struct kvm_memslots
*slots
;
1487 struct kvm_memory_slot
*memslot
;
1490 unsigned long *dirty_bitmap
;
1491 unsigned long *dirty_bitmap_buffer
;
1494 /* Dirty ring tracking is exclusive to dirty log tracking */
1495 if (kvm
->dirty_ring_size
)
1498 as_id
= log
->slot
>> 16;
1499 id
= (u16
)log
->slot
;
1500 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1503 slots
= __kvm_memslots(kvm
, as_id
);
1504 memslot
= id_to_memslot(slots
, id
);
1505 if (!memslot
|| !memslot
->dirty_bitmap
)
1508 dirty_bitmap
= memslot
->dirty_bitmap
;
1510 kvm_arch_sync_dirty_log(kvm
, memslot
);
1512 n
= kvm_dirty_bitmap_bytes(memslot
);
1514 if (kvm
->manual_dirty_log_protect
) {
1516 * Unlike kvm_get_dirty_log, we always return false in *flush,
1517 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1518 * is some code duplication between this function and
1519 * kvm_get_dirty_log, but hopefully all architecture
1520 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1521 * can be eliminated.
1523 dirty_bitmap_buffer
= dirty_bitmap
;
1525 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1526 memset(dirty_bitmap_buffer
, 0, n
);
1528 spin_lock(&kvm
->mmu_lock
);
1529 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1533 if (!dirty_bitmap
[i
])
1537 mask
= xchg(&dirty_bitmap
[i
], 0);
1538 dirty_bitmap_buffer
[i
] = mask
;
1540 offset
= i
* BITS_PER_LONG
;
1541 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1544 spin_unlock(&kvm
->mmu_lock
);
1548 kvm_arch_flush_remote_tlbs_memslot(kvm
, memslot
);
1550 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1557 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1558 * @kvm: kvm instance
1559 * @log: slot id and address to which we copy the log
1561 * Steps 1-4 below provide general overview of dirty page logging. See
1562 * kvm_get_dirty_log_protect() function description for additional details.
1564 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1565 * always flush the TLB (step 4) even if previous step failed and the dirty
1566 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1567 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1568 * writes will be marked dirty for next log read.
1570 * 1. Take a snapshot of the bit and clear it if needed.
1571 * 2. Write protect the corresponding page.
1572 * 3. Copy the snapshot to the userspace.
1573 * 4. Flush TLB's if needed.
1575 static int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1576 struct kvm_dirty_log
*log
)
1580 mutex_lock(&kvm
->slots_lock
);
1582 r
= kvm_get_dirty_log_protect(kvm
, log
);
1584 mutex_unlock(&kvm
->slots_lock
);
1589 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1590 * and reenable dirty page tracking for the corresponding pages.
1591 * @kvm: pointer to kvm instance
1592 * @log: slot id and address from which to fetch the bitmap of dirty pages
1594 static int kvm_clear_dirty_log_protect(struct kvm
*kvm
,
1595 struct kvm_clear_dirty_log
*log
)
1597 struct kvm_memslots
*slots
;
1598 struct kvm_memory_slot
*memslot
;
1602 unsigned long *dirty_bitmap
;
1603 unsigned long *dirty_bitmap_buffer
;
1606 /* Dirty ring tracking is exclusive to dirty log tracking */
1607 if (kvm
->dirty_ring_size
)
1610 as_id
= log
->slot
>> 16;
1611 id
= (u16
)log
->slot
;
1612 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1615 if (log
->first_page
& 63)
1618 slots
= __kvm_memslots(kvm
, as_id
);
1619 memslot
= id_to_memslot(slots
, id
);
1620 if (!memslot
|| !memslot
->dirty_bitmap
)
1623 dirty_bitmap
= memslot
->dirty_bitmap
;
1625 n
= ALIGN(log
->num_pages
, BITS_PER_LONG
) / 8;
1627 if (log
->first_page
> memslot
->npages
||
1628 log
->num_pages
> memslot
->npages
- log
->first_page
||
1629 (log
->num_pages
< memslot
->npages
- log
->first_page
&& (log
->num_pages
& 63)))
1632 kvm_arch_sync_dirty_log(kvm
, memslot
);
1635 dirty_bitmap_buffer
= kvm_second_dirty_bitmap(memslot
);
1636 if (copy_from_user(dirty_bitmap_buffer
, log
->dirty_bitmap
, n
))
1639 spin_lock(&kvm
->mmu_lock
);
1640 for (offset
= log
->first_page
, i
= offset
/ BITS_PER_LONG
,
1641 n
= DIV_ROUND_UP(log
->num_pages
, BITS_PER_LONG
); n
--;
1642 i
++, offset
+= BITS_PER_LONG
) {
1643 unsigned long mask
= *dirty_bitmap_buffer
++;
1644 atomic_long_t
*p
= (atomic_long_t
*) &dirty_bitmap
[i
];
1648 mask
&= atomic_long_fetch_andnot(mask
, p
);
1651 * mask contains the bits that really have been cleared. This
1652 * never includes any bits beyond the length of the memslot (if
1653 * the length is not aligned to 64 pages), therefore it is not
1654 * a problem if userspace sets them in log->dirty_bitmap.
1658 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1662 spin_unlock(&kvm
->mmu_lock
);
1665 kvm_arch_flush_remote_tlbs_memslot(kvm
, memslot
);
1670 static int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
,
1671 struct kvm_clear_dirty_log
*log
)
1675 mutex_lock(&kvm
->slots_lock
);
1677 r
= kvm_clear_dirty_log_protect(kvm
, log
);
1679 mutex_unlock(&kvm
->slots_lock
);
1682 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1684 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1686 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1688 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1690 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1692 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1694 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot
);
1696 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1698 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1700 return kvm_is_visible_memslot(memslot
);
1702 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1704 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1706 struct kvm_memory_slot
*memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1708 return kvm_is_visible_memslot(memslot
);
1710 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn
);
1712 unsigned long kvm_host_page_size(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1714 struct vm_area_struct
*vma
;
1715 unsigned long addr
, size
;
1719 addr
= kvm_vcpu_gfn_to_hva_prot(vcpu
, gfn
, NULL
);
1720 if (kvm_is_error_hva(addr
))
1723 mmap_read_lock(current
->mm
);
1724 vma
= find_vma(current
->mm
, addr
);
1728 size
= vma_kernel_pagesize(vma
);
1731 mmap_read_unlock(current
->mm
);
1736 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1738 return slot
->flags
& KVM_MEM_READONLY
;
1741 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1742 gfn_t
*nr_pages
, bool write
)
1744 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1745 return KVM_HVA_ERR_BAD
;
1747 if (memslot_is_readonly(slot
) && write
)
1748 return KVM_HVA_ERR_RO_BAD
;
1751 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1753 return __gfn_to_hva_memslot(slot
, gfn
);
1756 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1759 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1762 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1765 return gfn_to_hva_many(slot
, gfn
, NULL
);
1767 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1769 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1771 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1773 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1775 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1777 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1779 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1782 * Return the hva of a @gfn and the R/W attribute if possible.
1784 * @slot: the kvm_memory_slot which contains @gfn
1785 * @gfn: the gfn to be translated
1786 * @writable: used to return the read/write attribute of the @slot if the hva
1787 * is valid and @writable is not NULL
1789 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1790 gfn_t gfn
, bool *writable
)
1792 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1794 if (!kvm_is_error_hva(hva
) && writable
)
1795 *writable
= !memslot_is_readonly(slot
);
1800 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1802 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1804 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1807 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1809 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1811 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1814 static inline int check_user_page_hwpoison(unsigned long addr
)
1816 int rc
, flags
= FOLL_HWPOISON
| FOLL_WRITE
;
1818 rc
= get_user_pages(addr
, 1, flags
, NULL
, NULL
);
1819 return rc
== -EHWPOISON
;
1823 * The fast path to get the writable pfn which will be stored in @pfn,
1824 * true indicates success, otherwise false is returned. It's also the
1825 * only part that runs if we can in atomic context.
1827 static bool hva_to_pfn_fast(unsigned long addr
, bool write_fault
,
1828 bool *writable
, kvm_pfn_t
*pfn
)
1830 struct page
*page
[1];
1833 * Fast pin a writable pfn only if it is a write fault request
1834 * or the caller allows to map a writable pfn for a read fault
1837 if (!(write_fault
|| writable
))
1840 if (get_user_page_fast_only(addr
, FOLL_WRITE
, page
)) {
1841 *pfn
= page_to_pfn(page
[0]);
1852 * The slow path to get the pfn of the specified host virtual address,
1853 * 1 indicates success, -errno is returned if error is detected.
1855 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1856 bool *writable
, kvm_pfn_t
*pfn
)
1858 unsigned int flags
= FOLL_HWPOISON
;
1865 *writable
= write_fault
;
1868 flags
|= FOLL_WRITE
;
1870 flags
|= FOLL_NOWAIT
;
1872 npages
= get_user_pages_unlocked(addr
, 1, &page
, flags
);
1876 /* map read fault as writable if possible */
1877 if (unlikely(!write_fault
) && writable
) {
1880 if (get_user_page_fast_only(addr
, FOLL_WRITE
, &wpage
)) {
1886 *pfn
= page_to_pfn(page
);
1890 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1892 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1895 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1901 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1902 unsigned long addr
, bool *async
,
1903 bool write_fault
, bool *writable
,
1909 r
= follow_pfn(vma
, addr
, &pfn
);
1912 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1913 * not call the fault handler, so do it here.
1915 bool unlocked
= false;
1916 r
= fixup_user_fault(current
->mm
, addr
,
1917 (write_fault
? FAULT_FLAG_WRITE
: 0),
1924 r
= follow_pfn(vma
, addr
, &pfn
);
1934 * Get a reference here because callers of *hva_to_pfn* and
1935 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1936 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1937 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1938 * simply do nothing for reserved pfns.
1940 * Whoever called remap_pfn_range is also going to call e.g.
1941 * unmap_mapping_range before the underlying pages are freed,
1942 * causing a call to our MMU notifier.
1951 * Pin guest page in memory and return its pfn.
1952 * @addr: host virtual address which maps memory to the guest
1953 * @atomic: whether this function can sleep
1954 * @async: whether this function need to wait IO complete if the
1955 * host page is not in the memory
1956 * @write_fault: whether we should get a writable host page
1957 * @writable: whether it allows to map a writable host page for !@write_fault
1959 * The function will map a writable host page for these two cases:
1960 * 1): @write_fault = true
1961 * 2): @write_fault = false && @writable, @writable will tell the caller
1962 * whether the mapping is writable.
1964 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1965 bool write_fault
, bool *writable
)
1967 struct vm_area_struct
*vma
;
1971 /* we can do it either atomically or asynchronously, not both */
1972 BUG_ON(atomic
&& async
);
1974 if (hva_to_pfn_fast(addr
, write_fault
, writable
, &pfn
))
1978 return KVM_PFN_ERR_FAULT
;
1980 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1984 mmap_read_lock(current
->mm
);
1985 if (npages
== -EHWPOISON
||
1986 (!async
&& check_user_page_hwpoison(addr
))) {
1987 pfn
= KVM_PFN_ERR_HWPOISON
;
1992 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1995 pfn
= KVM_PFN_ERR_FAULT
;
1996 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1997 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, writable
, &pfn
);
2001 pfn
= KVM_PFN_ERR_FAULT
;
2003 if (async
&& vma_is_valid(vma
, write_fault
))
2005 pfn
= KVM_PFN_ERR_FAULT
;
2008 mmap_read_unlock(current
->mm
);
2012 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2013 bool atomic
, bool *async
, bool write_fault
,
2016 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
2018 if (addr
== KVM_HVA_ERR_RO_BAD
) {
2021 return KVM_PFN_ERR_RO_FAULT
;
2024 if (kvm_is_error_hva(addr
)) {
2027 return KVM_PFN_NOSLOT
;
2030 /* Do not map writable pfn in the readonly memslot. */
2031 if (writable
&& memslot_is_readonly(slot
)) {
2036 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
2039 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
2041 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
2044 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
2045 write_fault
, writable
);
2047 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
2049 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
2051 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
2053 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
2055 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
2057 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
2059 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
2061 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2063 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2065 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
2067 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
2069 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
2071 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
2073 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2075 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
2077 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
2079 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2080 struct page
**pages
, int nr_pages
)
2085 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
2086 if (kvm_is_error_hva(addr
))
2089 if (entry
< nr_pages
)
2092 return get_user_pages_fast_only(addr
, nr_pages
, FOLL_WRITE
, pages
);
2094 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
2096 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
2098 if (is_error_noslot_pfn(pfn
))
2099 return KVM_ERR_PTR_BAD_PAGE
;
2101 if (kvm_is_reserved_pfn(pfn
)) {
2103 return KVM_ERR_PTR_BAD_PAGE
;
2106 return pfn_to_page(pfn
);
2109 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
2113 pfn
= gfn_to_pfn(kvm
, gfn
);
2115 return kvm_pfn_to_page(pfn
);
2117 EXPORT_SYMBOL_GPL(gfn_to_page
);
2119 void kvm_release_pfn(kvm_pfn_t pfn
, bool dirty
, struct gfn_to_pfn_cache
*cache
)
2125 cache
->pfn
= cache
->gfn
= 0;
2128 kvm_release_pfn_dirty(pfn
);
2130 kvm_release_pfn_clean(pfn
);
2133 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2134 struct gfn_to_pfn_cache
*cache
, u64 gen
)
2136 kvm_release_pfn(cache
->pfn
, cache
->dirty
, cache
);
2138 cache
->pfn
= gfn_to_pfn_memslot(slot
, gfn
);
2140 cache
->dirty
= false;
2141 cache
->generation
= gen
;
2144 static int __kvm_map_gfn(struct kvm_memslots
*slots
, gfn_t gfn
,
2145 struct kvm_host_map
*map
,
2146 struct gfn_to_pfn_cache
*cache
,
2151 struct page
*page
= KVM_UNMAPPED_PAGE
;
2152 struct kvm_memory_slot
*slot
= __gfn_to_memslot(slots
, gfn
);
2153 u64 gen
= slots
->generation
;
2159 if (!cache
->pfn
|| cache
->gfn
!= gfn
||
2160 cache
->generation
!= gen
) {
2163 kvm_cache_gfn_to_pfn(slot
, gfn
, cache
, gen
);
2169 pfn
= gfn_to_pfn_memslot(slot
, gfn
);
2171 if (is_error_noslot_pfn(pfn
))
2174 if (pfn_valid(pfn
)) {
2175 page
= pfn_to_page(pfn
);
2177 hva
= kmap_atomic(page
);
2180 #ifdef CONFIG_HAS_IOMEM
2181 } else if (!atomic
) {
2182 hva
= memremap(pfn_to_hpa(pfn
), PAGE_SIZE
, MEMREMAP_WB
);
2199 int kvm_map_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
,
2200 struct gfn_to_pfn_cache
*cache
, bool atomic
)
2202 return __kvm_map_gfn(kvm_memslots(vcpu
->kvm
), gfn
, map
,
2205 EXPORT_SYMBOL_GPL(kvm_map_gfn
);
2207 int kvm_vcpu_map(struct kvm_vcpu
*vcpu
, gfn_t gfn
, struct kvm_host_map
*map
)
2209 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu
), gfn
, map
,
2212 EXPORT_SYMBOL_GPL(kvm_vcpu_map
);
2214 static void __kvm_unmap_gfn(struct kvm
*kvm
,
2215 struct kvm_memory_slot
*memslot
,
2216 struct kvm_host_map
*map
,
2217 struct gfn_to_pfn_cache
*cache
,
2218 bool dirty
, bool atomic
)
2226 if (map
->page
!= KVM_UNMAPPED_PAGE
) {
2228 kunmap_atomic(map
->hva
);
2232 #ifdef CONFIG_HAS_IOMEM
2236 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2240 mark_page_dirty_in_slot(kvm
, memslot
, map
->gfn
);
2243 cache
->dirty
|= dirty
;
2245 kvm_release_pfn(map
->pfn
, dirty
, NULL
);
2251 int kvm_unmap_gfn(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
,
2252 struct gfn_to_pfn_cache
*cache
, bool dirty
, bool atomic
)
2254 __kvm_unmap_gfn(vcpu
->kvm
, gfn_to_memslot(vcpu
->kvm
, map
->gfn
), map
,
2255 cache
, dirty
, atomic
);
2258 EXPORT_SYMBOL_GPL(kvm_unmap_gfn
);
2260 void kvm_vcpu_unmap(struct kvm_vcpu
*vcpu
, struct kvm_host_map
*map
, bool dirty
)
2262 __kvm_unmap_gfn(vcpu
->kvm
, kvm_vcpu_gfn_to_memslot(vcpu
, map
->gfn
),
2263 map
, NULL
, dirty
, false);
2265 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap
);
2267 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2271 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
2273 return kvm_pfn_to_page(pfn
);
2275 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
2277 void kvm_release_page_clean(struct page
*page
)
2279 WARN_ON(is_error_page(page
));
2281 kvm_release_pfn_clean(page_to_pfn(page
));
2283 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
2285 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
2287 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
2288 put_page(pfn_to_page(pfn
));
2290 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
2292 void kvm_release_page_dirty(struct page
*page
)
2294 WARN_ON(is_error_page(page
));
2296 kvm_release_pfn_dirty(page_to_pfn(page
));
2298 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
2300 void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
2302 kvm_set_pfn_dirty(pfn
);
2303 kvm_release_pfn_clean(pfn
);
2305 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty
);
2307 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
2309 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
2310 SetPageDirty(pfn_to_page(pfn
));
2312 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
2314 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
2316 if (!kvm_is_reserved_pfn(pfn
) && !kvm_is_zone_device_pfn(pfn
))
2317 mark_page_accessed(pfn_to_page(pfn
));
2319 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
2321 void kvm_get_pfn(kvm_pfn_t pfn
)
2323 if (!kvm_is_reserved_pfn(pfn
))
2324 get_page(pfn_to_page(pfn
));
2326 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
2328 static int next_segment(unsigned long len
, int offset
)
2330 if (len
> PAGE_SIZE
- offset
)
2331 return PAGE_SIZE
- offset
;
2336 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2337 void *data
, int offset
, int len
)
2342 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
2343 if (kvm_is_error_hva(addr
))
2345 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
2351 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
2354 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2356 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
2358 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
2360 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
2361 int offset
, int len
)
2363 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2365 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
2367 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
2369 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
2371 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2373 int offset
= offset_in_page(gpa
);
2376 while ((seg
= next_segment(len
, offset
)) != 0) {
2377 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
2387 EXPORT_SYMBOL_GPL(kvm_read_guest
);
2389 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
2391 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2393 int offset
= offset_in_page(gpa
);
2396 while ((seg
= next_segment(len
, offset
)) != 0) {
2397 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2407 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
2409 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
2410 void *data
, int offset
, unsigned long len
)
2415 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
2416 if (kvm_is_error_hva(addr
))
2418 pagefault_disable();
2419 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
2426 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2427 void *data
, unsigned long len
)
2429 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2430 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2431 int offset
= offset_in_page(gpa
);
2433 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
2435 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
2437 static int __kvm_write_guest_page(struct kvm
*kvm
,
2438 struct kvm_memory_slot
*memslot
, gfn_t gfn
,
2439 const void *data
, int offset
, int len
)
2444 addr
= gfn_to_hva_memslot(memslot
, gfn
);
2445 if (kvm_is_error_hva(addr
))
2447 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
2450 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
2454 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
2455 const void *data
, int offset
, int len
)
2457 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
2459 return __kvm_write_guest_page(kvm
, slot
, gfn
, data
, offset
, len
);
2461 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
2463 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
2464 const void *data
, int offset
, int len
)
2466 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2468 return __kvm_write_guest_page(vcpu
->kvm
, slot
, gfn
, data
, offset
, len
);
2470 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
2472 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
2475 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2477 int offset
= offset_in_page(gpa
);
2480 while ((seg
= next_segment(len
, offset
)) != 0) {
2481 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
2491 EXPORT_SYMBOL_GPL(kvm_write_guest
);
2493 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
2496 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2498 int offset
= offset_in_page(gpa
);
2501 while ((seg
= next_segment(len
, offset
)) != 0) {
2502 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
2512 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
2514 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots
*slots
,
2515 struct gfn_to_hva_cache
*ghc
,
2516 gpa_t gpa
, unsigned long len
)
2518 int offset
= offset_in_page(gpa
);
2519 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
2520 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
2521 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
2522 gfn_t nr_pages_avail
;
2524 /* Update ghc->generation before performing any error checks. */
2525 ghc
->generation
= slots
->generation
;
2527 if (start_gfn
> end_gfn
) {
2528 ghc
->hva
= KVM_HVA_ERR_BAD
;
2533 * If the requested region crosses two memslots, we still
2534 * verify that the entire region is valid here.
2536 for ( ; start_gfn
<= end_gfn
; start_gfn
+= nr_pages_avail
) {
2537 ghc
->memslot
= __gfn_to_memslot(slots
, start_gfn
);
2538 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
2540 if (kvm_is_error_hva(ghc
->hva
))
2544 /* Use the slow path for cross page reads and writes. */
2545 if (nr_pages_needed
== 1)
2548 ghc
->memslot
= NULL
;
2555 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2556 gpa_t gpa
, unsigned long len
)
2558 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2559 return __kvm_gfn_to_hva_cache_init(slots
, ghc
, gpa
, len
);
2561 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
2563 int kvm_write_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2564 void *data
, unsigned int offset
,
2567 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2569 gpa_t gpa
= ghc
->gpa
+ offset
;
2571 BUG_ON(len
+ offset
> ghc
->len
);
2573 if (slots
->generation
!= ghc
->generation
) {
2574 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
2578 if (kvm_is_error_hva(ghc
->hva
))
2581 if (unlikely(!ghc
->memslot
))
2582 return kvm_write_guest(kvm
, gpa
, data
, len
);
2584 r
= __copy_to_user((void __user
*)ghc
->hva
+ offset
, data
, len
);
2587 mark_page_dirty_in_slot(kvm
, ghc
->memslot
, gpa
>> PAGE_SHIFT
);
2591 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached
);
2593 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2594 void *data
, unsigned long len
)
2596 return kvm_write_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2598 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
2600 int kvm_read_guest_offset_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2601 void *data
, unsigned int offset
,
2604 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
2606 gpa_t gpa
= ghc
->gpa
+ offset
;
2608 BUG_ON(len
+ offset
> ghc
->len
);
2610 if (slots
->generation
!= ghc
->generation
) {
2611 if (__kvm_gfn_to_hva_cache_init(slots
, ghc
, ghc
->gpa
, ghc
->len
))
2615 if (kvm_is_error_hva(ghc
->hva
))
2618 if (unlikely(!ghc
->memslot
))
2619 return kvm_read_guest(kvm
, gpa
, data
, len
);
2621 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
+ offset
, len
);
2627 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached
);
2629 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
2630 void *data
, unsigned long len
)
2632 return kvm_read_guest_offset_cached(kvm
, ghc
, data
, 0, len
);
2634 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2636 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2638 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2639 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2641 int offset
= offset_in_page(gpa
);
2644 while ((seg
= next_segment(len
, offset
)) != 0) {
2645 ret
= kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2654 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2656 void mark_page_dirty_in_slot(struct kvm
*kvm
,
2657 struct kvm_memory_slot
*memslot
,
2660 if (memslot
&& kvm_slot_dirty_track_enabled(memslot
)) {
2661 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2662 u32 slot
= (memslot
->as_id
<< 16) | memslot
->id
;
2664 if (kvm
->dirty_ring_size
)
2665 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm
),
2668 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2671 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot
);
2673 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2675 struct kvm_memory_slot
*memslot
;
2677 memslot
= gfn_to_memslot(kvm
, gfn
);
2678 mark_page_dirty_in_slot(kvm
, memslot
, gfn
);
2680 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2682 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2684 struct kvm_memory_slot
*memslot
;
2686 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2687 mark_page_dirty_in_slot(vcpu
->kvm
, memslot
, gfn
);
2689 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2691 void kvm_sigset_activate(struct kvm_vcpu
*vcpu
)
2693 if (!vcpu
->sigset_active
)
2697 * This does a lockless modification of ->real_blocked, which is fine
2698 * because, only current can change ->real_blocked and all readers of
2699 * ->real_blocked don't care as long ->real_blocked is always a subset
2702 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, ¤t
->real_blocked
);
2705 void kvm_sigset_deactivate(struct kvm_vcpu
*vcpu
)
2707 if (!vcpu
->sigset_active
)
2710 sigprocmask(SIG_SETMASK
, ¤t
->real_blocked
, NULL
);
2711 sigemptyset(¤t
->real_blocked
);
2714 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2716 unsigned int old
, val
, grow
, grow_start
;
2718 old
= val
= vcpu
->halt_poll_ns
;
2719 grow_start
= READ_ONCE(halt_poll_ns_grow_start
);
2720 grow
= READ_ONCE(halt_poll_ns_grow
);
2725 if (val
< grow_start
)
2728 if (val
> halt_poll_ns
)
2731 vcpu
->halt_poll_ns
= val
;
2733 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2736 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2738 unsigned int old
, val
, shrink
;
2740 old
= val
= vcpu
->halt_poll_ns
;
2741 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2747 vcpu
->halt_poll_ns
= val
;
2748 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2751 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2754 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2756 if (kvm_arch_vcpu_runnable(vcpu
)) {
2757 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2760 if (kvm_cpu_has_pending_timer(vcpu
))
2762 if (signal_pending(current
))
2767 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
2772 update_halt_poll_stats(struct kvm_vcpu
*vcpu
, u64 poll_ns
, bool waited
)
2775 vcpu
->stat
.halt_poll_fail_ns
+= poll_ns
;
2777 vcpu
->stat
.halt_poll_success_ns
+= poll_ns
;
2781 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2783 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2785 ktime_t start
, cur
, poll_end
;
2786 bool waited
= false;
2789 kvm_arch_vcpu_blocking(vcpu
);
2791 start
= cur
= poll_end
= ktime_get();
2792 if (vcpu
->halt_poll_ns
&& !kvm_arch_no_poll(vcpu
)) {
2793 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2795 ++vcpu
->stat
.halt_attempted_poll
;
2798 * This sets KVM_REQ_UNHALT if an interrupt
2801 if (kvm_vcpu_check_block(vcpu
) < 0) {
2802 ++vcpu
->stat
.halt_successful_poll
;
2803 if (!vcpu_valid_wakeup(vcpu
))
2804 ++vcpu
->stat
.halt_poll_invalid
;
2807 poll_end
= cur
= ktime_get();
2808 } while (single_task_running() && ktime_before(cur
, stop
));
2811 prepare_to_rcuwait(&vcpu
->wait
);
2813 set_current_state(TASK_INTERRUPTIBLE
);
2815 if (kvm_vcpu_check_block(vcpu
) < 0)
2821 finish_rcuwait(&vcpu
->wait
);
2824 kvm_arch_vcpu_unblocking(vcpu
);
2825 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2827 update_halt_poll_stats(
2828 vcpu
, ktime_to_ns(ktime_sub(poll_end
, start
)), waited
);
2830 if (!kvm_arch_no_poll(vcpu
)) {
2831 if (!vcpu_valid_wakeup(vcpu
)) {
2832 shrink_halt_poll_ns(vcpu
);
2833 } else if (vcpu
->kvm
->max_halt_poll_ns
) {
2834 if (block_ns
<= vcpu
->halt_poll_ns
)
2836 /* we had a long block, shrink polling */
2837 else if (vcpu
->halt_poll_ns
&&
2838 block_ns
> vcpu
->kvm
->max_halt_poll_ns
)
2839 shrink_halt_poll_ns(vcpu
);
2840 /* we had a short halt and our poll time is too small */
2841 else if (vcpu
->halt_poll_ns
< vcpu
->kvm
->max_halt_poll_ns
&&
2842 block_ns
< vcpu
->kvm
->max_halt_poll_ns
)
2843 grow_halt_poll_ns(vcpu
);
2845 vcpu
->halt_poll_ns
= 0;
2849 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2850 kvm_arch_vcpu_block_finish(vcpu
);
2852 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2854 bool kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2856 struct rcuwait
*waitp
;
2858 waitp
= kvm_arch_vcpu_get_wait(vcpu
);
2859 if (rcuwait_wake_up(waitp
)) {
2860 WRITE_ONCE(vcpu
->ready
, true);
2861 ++vcpu
->stat
.halt_wakeup
;
2867 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2871 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2873 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2876 int cpu
= vcpu
->cpu
;
2878 if (kvm_vcpu_wake_up(vcpu
))
2882 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2883 if (kvm_arch_vcpu_should_kick(vcpu
))
2884 smp_send_reschedule(cpu
);
2887 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2888 #endif /* !CONFIG_S390 */
2890 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2893 struct task_struct
*task
= NULL
;
2897 pid
= rcu_dereference(target
->pid
);
2899 task
= get_pid_task(pid
, PIDTYPE_PID
);
2903 ret
= yield_to(task
, 1);
2904 put_task_struct(task
);
2908 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2911 * Helper that checks whether a VCPU is eligible for directed yield.
2912 * Most eligible candidate to yield is decided by following heuristics:
2914 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2915 * (preempted lock holder), indicated by @in_spin_loop.
2916 * Set at the beginning and cleared at the end of interception/PLE handler.
2918 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2919 * chance last time (mostly it has become eligible now since we have probably
2920 * yielded to lockholder in last iteration. This is done by toggling
2921 * @dy_eligible each time a VCPU checked for eligibility.)
2923 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2924 * to preempted lock-holder could result in wrong VCPU selection and CPU
2925 * burning. Giving priority for a potential lock-holder increases lock
2928 * Since algorithm is based on heuristics, accessing another VCPU data without
2929 * locking does not harm. It may result in trying to yield to same VCPU, fail
2930 * and continue with next VCPU and so on.
2932 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2934 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2937 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2938 vcpu
->spin_loop
.dy_eligible
;
2940 if (vcpu
->spin_loop
.in_spin_loop
)
2941 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2950 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2951 * a vcpu_load/vcpu_put pair. However, for most architectures
2952 * kvm_arch_vcpu_runnable does not require vcpu_load.
2954 bool __weak
kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
2956 return kvm_arch_vcpu_runnable(vcpu
);
2959 static bool vcpu_dy_runnable(struct kvm_vcpu
*vcpu
)
2961 if (kvm_arch_dy_runnable(vcpu
))
2964 #ifdef CONFIG_KVM_ASYNC_PF
2965 if (!list_empty_careful(&vcpu
->async_pf
.done
))
2972 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
, bool yield_to_kernel_mode
)
2974 struct kvm
*kvm
= me
->kvm
;
2975 struct kvm_vcpu
*vcpu
;
2976 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2982 kvm_vcpu_set_in_spin_loop(me
, true);
2984 * We boost the priority of a VCPU that is runnable but not
2985 * currently running, because it got preempted by something
2986 * else and called schedule in __vcpu_run. Hopefully that
2987 * VCPU is holding the lock that we need and will release it.
2988 * We approximate round-robin by starting at the last boosted VCPU.
2990 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2991 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2992 if (!pass
&& i
<= last_boosted_vcpu
) {
2993 i
= last_boosted_vcpu
;
2995 } else if (pass
&& i
> last_boosted_vcpu
)
2997 if (!READ_ONCE(vcpu
->ready
))
3001 if (rcuwait_active(&vcpu
->wait
) &&
3002 !vcpu_dy_runnable(vcpu
))
3004 if (READ_ONCE(vcpu
->preempted
) && yield_to_kernel_mode
&&
3005 !kvm_arch_vcpu_in_kernel(vcpu
))
3007 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
3010 yielded
= kvm_vcpu_yield_to(vcpu
);
3012 kvm
->last_boosted_vcpu
= i
;
3014 } else if (yielded
< 0) {
3021 kvm_vcpu_set_in_spin_loop(me
, false);
3023 /* Ensure vcpu is not eligible during next spinloop */
3024 kvm_vcpu_set_dy_eligible(me
, false);
3026 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
3028 static bool kvm_page_in_dirty_ring(struct kvm
*kvm
, unsigned long pgoff
)
3030 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3031 return (pgoff
>= KVM_DIRTY_LOG_PAGE_OFFSET
) &&
3032 (pgoff
< KVM_DIRTY_LOG_PAGE_OFFSET
+
3033 kvm
->dirty_ring_size
/ PAGE_SIZE
);
3039 static vm_fault_t
kvm_vcpu_fault(struct vm_fault
*vmf
)
3041 struct kvm_vcpu
*vcpu
= vmf
->vma
->vm_file
->private_data
;
3044 if (vmf
->pgoff
== 0)
3045 page
= virt_to_page(vcpu
->run
);
3047 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
3048 page
= virt_to_page(vcpu
->arch
.pio_data
);
3050 #ifdef CONFIG_KVM_MMIO
3051 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
3052 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
3054 else if (kvm_page_in_dirty_ring(vcpu
->kvm
, vmf
->pgoff
))
3055 page
= kvm_dirty_ring_get_page(
3057 vmf
->pgoff
- KVM_DIRTY_LOG_PAGE_OFFSET
);
3059 return kvm_arch_vcpu_fault(vcpu
, vmf
);
3065 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
3066 .fault
= kvm_vcpu_fault
,
3069 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
3071 struct kvm_vcpu
*vcpu
= file
->private_data
;
3072 unsigned long pages
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
3074 if ((kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
) ||
3075 kvm_page_in_dirty_ring(vcpu
->kvm
, vma
->vm_pgoff
+ pages
- 1)) &&
3076 ((vma
->vm_flags
& VM_EXEC
) || !(vma
->vm_flags
& VM_SHARED
)))
3079 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
3083 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
3085 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3087 kvm_put_kvm(vcpu
->kvm
);
3091 static struct file_operations kvm_vcpu_fops
= {
3092 .release
= kvm_vcpu_release
,
3093 .unlocked_ioctl
= kvm_vcpu_ioctl
,
3094 .mmap
= kvm_vcpu_mmap
,
3095 .llseek
= noop_llseek
,
3096 KVM_COMPAT(kvm_vcpu_compat_ioctl
),
3100 * Allocates an inode for the vcpu.
3102 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
3104 char name
[8 + 1 + ITOA_MAX_LEN
+ 1];
3106 snprintf(name
, sizeof(name
), "kvm-vcpu:%d", vcpu
->vcpu_id
);
3107 return anon_inode_getfd(name
, &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
3110 static void kvm_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
3112 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3113 struct dentry
*debugfs_dentry
;
3114 char dir_name
[ITOA_MAX_LEN
* 2];
3116 if (!debugfs_initialized())
3119 snprintf(dir_name
, sizeof(dir_name
), "vcpu%d", vcpu
->vcpu_id
);
3120 debugfs_dentry
= debugfs_create_dir(dir_name
,
3121 vcpu
->kvm
->debugfs_dentry
);
3123 kvm_arch_create_vcpu_debugfs(vcpu
, debugfs_dentry
);
3128 * Creates some virtual cpus. Good luck creating more than one.
3130 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
3133 struct kvm_vcpu
*vcpu
;
3136 if (id
>= KVM_MAX_VCPU_ID
)
3139 mutex_lock(&kvm
->lock
);
3140 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
3141 mutex_unlock(&kvm
->lock
);
3145 kvm
->created_vcpus
++;
3146 mutex_unlock(&kvm
->lock
);
3148 r
= kvm_arch_vcpu_precreate(kvm
, id
);
3150 goto vcpu_decrement
;
3152 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
3155 goto vcpu_decrement
;
3158 BUILD_BUG_ON(sizeof(struct kvm_run
) > PAGE_SIZE
);
3159 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
3164 vcpu
->run
= page_address(page
);
3166 kvm_vcpu_init(vcpu
, kvm
, id
);
3168 r
= kvm_arch_vcpu_create(vcpu
);
3170 goto vcpu_free_run_page
;
3172 if (kvm
->dirty_ring_size
) {
3173 r
= kvm_dirty_ring_alloc(&vcpu
->dirty_ring
,
3174 id
, kvm
->dirty_ring_size
);
3176 goto arch_vcpu_destroy
;
3179 mutex_lock(&kvm
->lock
);
3180 if (kvm_get_vcpu_by_id(kvm
, id
)) {
3182 goto unlock_vcpu_destroy
;
3185 vcpu
->vcpu_idx
= atomic_read(&kvm
->online_vcpus
);
3186 BUG_ON(kvm
->vcpus
[vcpu
->vcpu_idx
]);
3188 /* Now it's all set up, let userspace reach it */
3190 r
= create_vcpu_fd(vcpu
);
3192 kvm_put_kvm_no_destroy(kvm
);
3193 goto unlock_vcpu_destroy
;
3196 kvm
->vcpus
[vcpu
->vcpu_idx
] = vcpu
;
3199 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3200 * before kvm->online_vcpu's incremented value.
3203 atomic_inc(&kvm
->online_vcpus
);
3205 mutex_unlock(&kvm
->lock
);
3206 kvm_arch_vcpu_postcreate(vcpu
);
3207 kvm_create_vcpu_debugfs(vcpu
);
3210 unlock_vcpu_destroy
:
3211 mutex_unlock(&kvm
->lock
);
3212 kvm_dirty_ring_free(&vcpu
->dirty_ring
);
3214 kvm_arch_vcpu_destroy(vcpu
);
3216 free_page((unsigned long)vcpu
->run
);
3218 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
3220 mutex_lock(&kvm
->lock
);
3221 kvm
->created_vcpus
--;
3222 mutex_unlock(&kvm
->lock
);
3226 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
3229 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
3230 vcpu
->sigset_active
= 1;
3231 vcpu
->sigset
= *sigset
;
3233 vcpu
->sigset_active
= 0;
3237 static long kvm_vcpu_ioctl(struct file
*filp
,
3238 unsigned int ioctl
, unsigned long arg
)
3240 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3241 void __user
*argp
= (void __user
*)arg
;
3243 struct kvm_fpu
*fpu
= NULL
;
3244 struct kvm_sregs
*kvm_sregs
= NULL
;
3246 if (vcpu
->kvm
->mm
!= current
->mm
)
3249 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
3253 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3254 * execution; mutex_lock() would break them.
3256 r
= kvm_arch_vcpu_async_ioctl(filp
, ioctl
, arg
);
3257 if (r
!= -ENOIOCTLCMD
)
3260 if (mutex_lock_killable(&vcpu
->mutex
))
3268 oldpid
= rcu_access_pointer(vcpu
->pid
);
3269 if (unlikely(oldpid
!= task_pid(current
))) {
3270 /* The thread running this VCPU changed. */
3273 r
= kvm_arch_vcpu_run_pid_change(vcpu
);
3277 newpid
= get_task_pid(current
, PIDTYPE_PID
);
3278 rcu_assign_pointer(vcpu
->pid
, newpid
);
3283 r
= kvm_arch_vcpu_ioctl_run(vcpu
);
3284 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
3287 case KVM_GET_REGS
: {
3288 struct kvm_regs
*kvm_regs
;
3291 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL_ACCOUNT
);
3294 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
3298 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
3305 case KVM_SET_REGS
: {
3306 struct kvm_regs
*kvm_regs
;
3308 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
3309 if (IS_ERR(kvm_regs
)) {
3310 r
= PTR_ERR(kvm_regs
);
3313 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
3317 case KVM_GET_SREGS
: {
3318 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
),
3319 GFP_KERNEL_ACCOUNT
);
3323 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
3327 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
3332 case KVM_SET_SREGS
: {
3333 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
3334 if (IS_ERR(kvm_sregs
)) {
3335 r
= PTR_ERR(kvm_sregs
);
3339 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
3342 case KVM_GET_MP_STATE
: {
3343 struct kvm_mp_state mp_state
;
3345 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
3349 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
3354 case KVM_SET_MP_STATE
: {
3355 struct kvm_mp_state mp_state
;
3358 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
3360 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
3363 case KVM_TRANSLATE
: {
3364 struct kvm_translation tr
;
3367 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
3369 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
3373 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
3378 case KVM_SET_GUEST_DEBUG
: {
3379 struct kvm_guest_debug dbg
;
3382 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
3384 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
3387 case KVM_SET_SIGNAL_MASK
: {
3388 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
3389 struct kvm_signal_mask kvm_sigmask
;
3390 sigset_t sigset
, *p
;
3395 if (copy_from_user(&kvm_sigmask
, argp
,
3396 sizeof(kvm_sigmask
)))
3399 if (kvm_sigmask
.len
!= sizeof(sigset
))
3402 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
3407 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
3411 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL_ACCOUNT
);
3415 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
3419 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
3425 fpu
= memdup_user(argp
, sizeof(*fpu
));
3431 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
3435 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
3438 mutex_unlock(&vcpu
->mutex
);
3444 #ifdef CONFIG_KVM_COMPAT
3445 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
3446 unsigned int ioctl
, unsigned long arg
)
3448 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3449 void __user
*argp
= compat_ptr(arg
);
3452 if (vcpu
->kvm
->mm
!= current
->mm
)
3456 case KVM_SET_SIGNAL_MASK
: {
3457 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
3458 struct kvm_signal_mask kvm_sigmask
;
3463 if (copy_from_user(&kvm_sigmask
, argp
,
3464 sizeof(kvm_sigmask
)))
3467 if (kvm_sigmask
.len
!= sizeof(compat_sigset_t
))
3470 if (get_compat_sigset(&sigset
,
3471 (compat_sigset_t __user
*)sigmask_arg
->sigset
))
3473 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
3475 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
3479 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
3487 static int kvm_device_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
3489 struct kvm_device
*dev
= filp
->private_data
;
3492 return dev
->ops
->mmap(dev
, vma
);
3497 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
3498 int (*accessor
)(struct kvm_device
*dev
,
3499 struct kvm_device_attr
*attr
),
3502 struct kvm_device_attr attr
;
3507 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
3510 return accessor(dev
, &attr
);
3513 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
3516 struct kvm_device
*dev
= filp
->private_data
;
3518 if (dev
->kvm
->mm
!= current
->mm
)
3522 case KVM_SET_DEVICE_ATTR
:
3523 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
3524 case KVM_GET_DEVICE_ATTR
:
3525 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
3526 case KVM_HAS_DEVICE_ATTR
:
3527 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
3529 if (dev
->ops
->ioctl
)
3530 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
3536 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
3538 struct kvm_device
*dev
= filp
->private_data
;
3539 struct kvm
*kvm
= dev
->kvm
;
3541 if (dev
->ops
->release
) {
3542 mutex_lock(&kvm
->lock
);
3543 list_del(&dev
->vm_node
);
3544 dev
->ops
->release(dev
);
3545 mutex_unlock(&kvm
->lock
);
3552 static const struct file_operations kvm_device_fops
= {
3553 .unlocked_ioctl
= kvm_device_ioctl
,
3554 .release
= kvm_device_release
,
3555 KVM_COMPAT(kvm_device_ioctl
),
3556 .mmap
= kvm_device_mmap
,
3559 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
3561 if (filp
->f_op
!= &kvm_device_fops
)
3564 return filp
->private_data
;
3567 static const struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
3568 #ifdef CONFIG_KVM_MPIC
3569 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
3570 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
3574 int kvm_register_device_ops(const struct kvm_device_ops
*ops
, u32 type
)
3576 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
3579 if (kvm_device_ops_table
[type
] != NULL
)
3582 kvm_device_ops_table
[type
] = ops
;
3586 void kvm_unregister_device_ops(u32 type
)
3588 if (kvm_device_ops_table
[type
] != NULL
)
3589 kvm_device_ops_table
[type
] = NULL
;
3592 static int kvm_ioctl_create_device(struct kvm
*kvm
,
3593 struct kvm_create_device
*cd
)
3595 const struct kvm_device_ops
*ops
= NULL
;
3596 struct kvm_device
*dev
;
3597 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
3601 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
3604 type
= array_index_nospec(cd
->type
, ARRAY_SIZE(kvm_device_ops_table
));
3605 ops
= kvm_device_ops_table
[type
];
3612 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL_ACCOUNT
);
3619 mutex_lock(&kvm
->lock
);
3620 ret
= ops
->create(dev
, type
);
3622 mutex_unlock(&kvm
->lock
);
3626 list_add(&dev
->vm_node
, &kvm
->devices
);
3627 mutex_unlock(&kvm
->lock
);
3633 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
3635 kvm_put_kvm_no_destroy(kvm
);
3636 mutex_lock(&kvm
->lock
);
3637 list_del(&dev
->vm_node
);
3638 mutex_unlock(&kvm
->lock
);
3647 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
3650 case KVM_CAP_USER_MEMORY
:
3651 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
3652 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
3653 case KVM_CAP_INTERNAL_ERROR_DATA
:
3654 #ifdef CONFIG_HAVE_KVM_MSI
3655 case KVM_CAP_SIGNAL_MSI
:
3657 #ifdef CONFIG_HAVE_KVM_IRQFD
3659 case KVM_CAP_IRQFD_RESAMPLE
:
3661 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
3662 case KVM_CAP_CHECK_EXTENSION_VM
:
3663 case KVM_CAP_ENABLE_CAP_VM
:
3664 case KVM_CAP_HALT_POLL
:
3666 #ifdef CONFIG_KVM_MMIO
3667 case KVM_CAP_COALESCED_MMIO
:
3668 return KVM_COALESCED_MMIO_PAGE_OFFSET
;
3669 case KVM_CAP_COALESCED_PIO
:
3672 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3673 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
:
3674 return KVM_DIRTY_LOG_MANUAL_CAPS
;
3676 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3677 case KVM_CAP_IRQ_ROUTING
:
3678 return KVM_MAX_IRQ_ROUTES
;
3680 #if KVM_ADDRESS_SPACE_NUM > 1
3681 case KVM_CAP_MULTI_ADDRESS_SPACE
:
3682 return KVM_ADDRESS_SPACE_NUM
;
3684 case KVM_CAP_NR_MEMSLOTS
:
3685 return KVM_USER_MEM_SLOTS
;
3686 case KVM_CAP_DIRTY_LOG_RING
:
3687 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3688 return KVM_DIRTY_RING_MAX_ENTRIES
* sizeof(struct kvm_dirty_gfn
);
3695 return kvm_vm_ioctl_check_extension(kvm
, arg
);
3698 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm
*kvm
, u32 size
)
3702 if (!KVM_DIRTY_LOG_PAGE_OFFSET
)
3705 /* the size should be power of 2 */
3706 if (!size
|| (size
& (size
- 1)))
3709 /* Should be bigger to keep the reserved entries, or a page */
3710 if (size
< kvm_dirty_ring_get_rsvd_entries() *
3711 sizeof(struct kvm_dirty_gfn
) || size
< PAGE_SIZE
)
3714 if (size
> KVM_DIRTY_RING_MAX_ENTRIES
*
3715 sizeof(struct kvm_dirty_gfn
))
3718 /* We only allow it to set once */
3719 if (kvm
->dirty_ring_size
)
3722 mutex_lock(&kvm
->lock
);
3724 if (kvm
->created_vcpus
) {
3725 /* We don't allow to change this value after vcpu created */
3728 kvm
->dirty_ring_size
= size
;
3732 mutex_unlock(&kvm
->lock
);
3736 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm
*kvm
)
3739 struct kvm_vcpu
*vcpu
;
3742 if (!kvm
->dirty_ring_size
)
3745 mutex_lock(&kvm
->slots_lock
);
3747 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3748 cleared
+= kvm_dirty_ring_reset(vcpu
->kvm
, &vcpu
->dirty_ring
);
3750 mutex_unlock(&kvm
->slots_lock
);
3753 kvm_flush_remote_tlbs(kvm
);
3758 int __attribute__((weak
)) kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
3759 struct kvm_enable_cap
*cap
)
3764 static int kvm_vm_ioctl_enable_cap_generic(struct kvm
*kvm
,
3765 struct kvm_enable_cap
*cap
)
3768 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3769 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
: {
3770 u64 allowed_options
= KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
;
3772 if (cap
->args
[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE
)
3773 allowed_options
= KVM_DIRTY_LOG_MANUAL_CAPS
;
3775 if (cap
->flags
|| (cap
->args
[0] & ~allowed_options
))
3777 kvm
->manual_dirty_log_protect
= cap
->args
[0];
3781 case KVM_CAP_HALT_POLL
: {
3782 if (cap
->flags
|| cap
->args
[0] != (unsigned int)cap
->args
[0])
3785 kvm
->max_halt_poll_ns
= cap
->args
[0];
3788 case KVM_CAP_DIRTY_LOG_RING
:
3789 return kvm_vm_ioctl_enable_dirty_log_ring(kvm
, cap
->args
[0]);
3791 return kvm_vm_ioctl_enable_cap(kvm
, cap
);
3795 static long kvm_vm_ioctl(struct file
*filp
,
3796 unsigned int ioctl
, unsigned long arg
)
3798 struct kvm
*kvm
= filp
->private_data
;
3799 void __user
*argp
= (void __user
*)arg
;
3802 if (kvm
->mm
!= current
->mm
)
3805 case KVM_CREATE_VCPU
:
3806 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
3808 case KVM_ENABLE_CAP
: {
3809 struct kvm_enable_cap cap
;
3812 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3814 r
= kvm_vm_ioctl_enable_cap_generic(kvm
, &cap
);
3817 case KVM_SET_USER_MEMORY_REGION
: {
3818 struct kvm_userspace_memory_region kvm_userspace_mem
;
3821 if (copy_from_user(&kvm_userspace_mem
, argp
,
3822 sizeof(kvm_userspace_mem
)))
3825 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
3828 case KVM_GET_DIRTY_LOG
: {
3829 struct kvm_dirty_log log
;
3832 if (copy_from_user(&log
, argp
, sizeof(log
)))
3834 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3837 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3838 case KVM_CLEAR_DIRTY_LOG
: {
3839 struct kvm_clear_dirty_log log
;
3842 if (copy_from_user(&log
, argp
, sizeof(log
)))
3844 r
= kvm_vm_ioctl_clear_dirty_log(kvm
, &log
);
3848 #ifdef CONFIG_KVM_MMIO
3849 case KVM_REGISTER_COALESCED_MMIO
: {
3850 struct kvm_coalesced_mmio_zone zone
;
3853 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3855 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
3858 case KVM_UNREGISTER_COALESCED_MMIO
: {
3859 struct kvm_coalesced_mmio_zone zone
;
3862 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
3864 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
3869 struct kvm_irqfd data
;
3872 if (copy_from_user(&data
, argp
, sizeof(data
)))
3874 r
= kvm_irqfd(kvm
, &data
);
3877 case KVM_IOEVENTFD
: {
3878 struct kvm_ioeventfd data
;
3881 if (copy_from_user(&data
, argp
, sizeof(data
)))
3883 r
= kvm_ioeventfd(kvm
, &data
);
3886 #ifdef CONFIG_HAVE_KVM_MSI
3887 case KVM_SIGNAL_MSI
: {
3891 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
3893 r
= kvm_send_userspace_msi(kvm
, &msi
);
3897 #ifdef __KVM_HAVE_IRQ_LINE
3898 case KVM_IRQ_LINE_STATUS
:
3899 case KVM_IRQ_LINE
: {
3900 struct kvm_irq_level irq_event
;
3903 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
3906 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
3907 ioctl
== KVM_IRQ_LINE_STATUS
);
3912 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
3913 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
3921 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3922 case KVM_SET_GSI_ROUTING
: {
3923 struct kvm_irq_routing routing
;
3924 struct kvm_irq_routing __user
*urouting
;
3925 struct kvm_irq_routing_entry
*entries
= NULL
;
3928 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3931 if (!kvm_arch_can_set_irq_routing(kvm
))
3933 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3939 entries
= vmemdup_user(urouting
->entries
,
3940 array_size(sizeof(*entries
),
3942 if (IS_ERR(entries
)) {
3943 r
= PTR_ERR(entries
);
3947 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3952 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3953 case KVM_CREATE_DEVICE
: {
3954 struct kvm_create_device cd
;
3957 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3960 r
= kvm_ioctl_create_device(kvm
, &cd
);
3965 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3971 case KVM_CHECK_EXTENSION
:
3972 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3974 case KVM_RESET_DIRTY_RINGS
:
3975 r
= kvm_vm_ioctl_reset_dirty_pages(kvm
);
3978 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3984 #ifdef CONFIG_KVM_COMPAT
3985 struct compat_kvm_dirty_log
{
3989 compat_uptr_t dirty_bitmap
; /* one bit per page */
3994 static long kvm_vm_compat_ioctl(struct file
*filp
,
3995 unsigned int ioctl
, unsigned long arg
)
3997 struct kvm
*kvm
= filp
->private_data
;
4000 if (kvm
->mm
!= current
->mm
)
4003 case KVM_GET_DIRTY_LOG
: {
4004 struct compat_kvm_dirty_log compat_log
;
4005 struct kvm_dirty_log log
;
4007 if (copy_from_user(&compat_log
, (void __user
*)arg
,
4008 sizeof(compat_log
)))
4010 log
.slot
= compat_log
.slot
;
4011 log
.padding1
= compat_log
.padding1
;
4012 log
.padding2
= compat_log
.padding2
;
4013 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
4015 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
4019 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
4025 static struct file_operations kvm_vm_fops
= {
4026 .release
= kvm_vm_release
,
4027 .unlocked_ioctl
= kvm_vm_ioctl
,
4028 .llseek
= noop_llseek
,
4029 KVM_COMPAT(kvm_vm_compat_ioctl
),
4032 static int kvm_dev_ioctl_create_vm(unsigned long type
)
4038 kvm
= kvm_create_vm(type
);
4040 return PTR_ERR(kvm
);
4041 #ifdef CONFIG_KVM_MMIO
4042 r
= kvm_coalesced_mmio_init(kvm
);
4046 r
= get_unused_fd_flags(O_CLOEXEC
);
4050 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
4058 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4059 * already set, with ->release() being kvm_vm_release(). In error
4060 * cases it will be called by the final fput(file) and will take
4061 * care of doing kvm_put_kvm(kvm).
4063 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
4068 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM
, kvm
);
4070 fd_install(r
, file
);
4078 static long kvm_dev_ioctl(struct file
*filp
,
4079 unsigned int ioctl
, unsigned long arg
)
4084 case KVM_GET_API_VERSION
:
4087 r
= KVM_API_VERSION
;
4090 r
= kvm_dev_ioctl_create_vm(arg
);
4092 case KVM_CHECK_EXTENSION
:
4093 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
4095 case KVM_GET_VCPU_MMAP_SIZE
:
4098 r
= PAGE_SIZE
; /* struct kvm_run */
4100 r
+= PAGE_SIZE
; /* pio data page */
4102 #ifdef CONFIG_KVM_MMIO
4103 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
4106 case KVM_TRACE_ENABLE
:
4107 case KVM_TRACE_PAUSE
:
4108 case KVM_TRACE_DISABLE
:
4112 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
4118 static struct file_operations kvm_chardev_ops
= {
4119 .unlocked_ioctl
= kvm_dev_ioctl
,
4120 .llseek
= noop_llseek
,
4121 KVM_COMPAT(kvm_dev_ioctl
),
4124 static struct miscdevice kvm_dev
= {
4130 static void hardware_enable_nolock(void *junk
)
4132 int cpu
= raw_smp_processor_id();
4135 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
4138 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
4140 r
= kvm_arch_hardware_enable();
4143 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
4144 atomic_inc(&hardware_enable_failed
);
4145 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
4149 static int kvm_starting_cpu(unsigned int cpu
)
4151 raw_spin_lock(&kvm_count_lock
);
4152 if (kvm_usage_count
)
4153 hardware_enable_nolock(NULL
);
4154 raw_spin_unlock(&kvm_count_lock
);
4158 static void hardware_disable_nolock(void *junk
)
4160 int cpu
= raw_smp_processor_id();
4162 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
4164 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
4165 kvm_arch_hardware_disable();
4168 static int kvm_dying_cpu(unsigned int cpu
)
4170 raw_spin_lock(&kvm_count_lock
);
4171 if (kvm_usage_count
)
4172 hardware_disable_nolock(NULL
);
4173 raw_spin_unlock(&kvm_count_lock
);
4177 static void hardware_disable_all_nolock(void)
4179 BUG_ON(!kvm_usage_count
);
4182 if (!kvm_usage_count
)
4183 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4186 static void hardware_disable_all(void)
4188 raw_spin_lock(&kvm_count_lock
);
4189 hardware_disable_all_nolock();
4190 raw_spin_unlock(&kvm_count_lock
);
4193 static int hardware_enable_all(void)
4197 raw_spin_lock(&kvm_count_lock
);
4200 if (kvm_usage_count
== 1) {
4201 atomic_set(&hardware_enable_failed
, 0);
4202 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
4204 if (atomic_read(&hardware_enable_failed
)) {
4205 hardware_disable_all_nolock();
4210 raw_spin_unlock(&kvm_count_lock
);
4215 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
4219 * Some (well, at least mine) BIOSes hang on reboot if
4222 * And Intel TXT required VMX off for all cpu when system shutdown.
4224 pr_info("kvm: exiting hardware virtualization\n");
4225 kvm_rebooting
= true;
4226 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
4230 static struct notifier_block kvm_reboot_notifier
= {
4231 .notifier_call
= kvm_reboot
,
4235 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
4239 for (i
= 0; i
< bus
->dev_count
; i
++) {
4240 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
4242 kvm_iodevice_destructor(pos
);
4247 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
4248 const struct kvm_io_range
*r2
)
4250 gpa_t addr1
= r1
->addr
;
4251 gpa_t addr2
= r2
->addr
;
4256 /* If r2->len == 0, match the exact address. If r2->len != 0,
4257 * accept any overlapping write. Any order is acceptable for
4258 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4259 * we process all of them.
4272 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
4274 return kvm_io_bus_cmp(p1
, p2
);
4277 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
4278 gpa_t addr
, int len
)
4280 struct kvm_io_range
*range
, key
;
4283 key
= (struct kvm_io_range
) {
4288 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
4289 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
4293 off
= range
- bus
->range
;
4295 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
4301 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
4302 struct kvm_io_range
*range
, const void *val
)
4306 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
4310 while (idx
< bus
->dev_count
&&
4311 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
4312 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
4321 /* kvm_io_bus_write - called under kvm->slots_lock */
4322 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
4323 int len
, const void *val
)
4325 struct kvm_io_bus
*bus
;
4326 struct kvm_io_range range
;
4329 range
= (struct kvm_io_range
) {
4334 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4337 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
4338 return r
< 0 ? r
: 0;
4340 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
4342 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4343 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
4344 gpa_t addr
, int len
, const void *val
, long cookie
)
4346 struct kvm_io_bus
*bus
;
4347 struct kvm_io_range range
;
4349 range
= (struct kvm_io_range
) {
4354 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4358 /* First try the device referenced by cookie. */
4359 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
4360 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
4361 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
4366 * cookie contained garbage; fall back to search and return the
4367 * correct cookie value.
4369 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
4372 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
4373 struct kvm_io_range
*range
, void *val
)
4377 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
4381 while (idx
< bus
->dev_count
&&
4382 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
4383 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
4392 /* kvm_io_bus_read - called under kvm->slots_lock */
4393 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
4396 struct kvm_io_bus
*bus
;
4397 struct kvm_io_range range
;
4400 range
= (struct kvm_io_range
) {
4405 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
4408 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
4409 return r
< 0 ? r
: 0;
4412 /* Caller must hold slots_lock. */
4413 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
4414 int len
, struct kvm_io_device
*dev
)
4417 struct kvm_io_bus
*new_bus
, *bus
;
4418 struct kvm_io_range range
;
4420 bus
= kvm_get_bus(kvm
, bus_idx
);
4424 /* exclude ioeventfd which is limited by maximum fd */
4425 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
4428 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
+ 1),
4429 GFP_KERNEL_ACCOUNT
);
4433 range
= (struct kvm_io_range
) {
4439 for (i
= 0; i
< bus
->dev_count
; i
++)
4440 if (kvm_io_bus_cmp(&bus
->range
[i
], &range
) > 0)
4443 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
4444 new_bus
->dev_count
++;
4445 new_bus
->range
[i
] = range
;
4446 memcpy(new_bus
->range
+ i
+ 1, bus
->range
+ i
,
4447 (bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
4448 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
4449 synchronize_srcu_expedited(&kvm
->srcu
);
4455 /* Caller must hold slots_lock. */
4456 void kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
4457 struct kvm_io_device
*dev
)
4460 struct kvm_io_bus
*new_bus
, *bus
;
4462 bus
= kvm_get_bus(kvm
, bus_idx
);
4466 for (i
= 0; i
< bus
->dev_count
; i
++)
4467 if (bus
->range
[i
].dev
== dev
) {
4471 if (i
== bus
->dev_count
)
4474 new_bus
= kmalloc(struct_size(bus
, range
, bus
->dev_count
- 1),
4475 GFP_KERNEL_ACCOUNT
);
4477 memcpy(new_bus
, bus
, struct_size(bus
, range
, i
));
4478 new_bus
->dev_count
--;
4479 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
4480 flex_array_size(new_bus
, range
, new_bus
->dev_count
- i
));
4482 pr_err("kvm: failed to shrink bus, removing it completely\n");
4483 for (j
= 0; j
< bus
->dev_count
; j
++) {
4486 kvm_iodevice_destructor(bus
->range
[j
].dev
);
4490 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
4491 synchronize_srcu_expedited(&kvm
->srcu
);
4496 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
4499 struct kvm_io_bus
*bus
;
4500 int dev_idx
, srcu_idx
;
4501 struct kvm_io_device
*iodev
= NULL
;
4503 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
4505 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
4509 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
4513 iodev
= bus
->range
[dev_idx
].dev
;
4516 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
4520 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
4522 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
4523 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
4526 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
4529 /* The debugfs files are a reference to the kvm struct which
4530 * is still valid when kvm_destroy_vm is called.
4531 * To avoid the race between open and the removal of the debugfs
4532 * directory we test against the users count.
4534 if (!refcount_inc_not_zero(&stat_data
->kvm
->users_count
))
4537 if (simple_attr_open(inode
, file
, get
,
4538 KVM_DBGFS_GET_MODE(stat_data
->dbgfs_item
) & 0222
4541 kvm_put_kvm(stat_data
->kvm
);
4548 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
4550 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
4553 simple_attr_release(inode
, file
);
4554 kvm_put_kvm(stat_data
->kvm
);
4559 static int kvm_get_stat_per_vm(struct kvm
*kvm
, size_t offset
, u64
*val
)
4561 *val
= *(ulong
*)((void *)kvm
+ offset
);
4566 static int kvm_clear_stat_per_vm(struct kvm
*kvm
, size_t offset
)
4568 *(ulong
*)((void *)kvm
+ offset
) = 0;
4573 static int kvm_get_stat_per_vcpu(struct kvm
*kvm
, size_t offset
, u64
*val
)
4576 struct kvm_vcpu
*vcpu
;
4580 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4581 *val
+= *(u64
*)((void *)vcpu
+ offset
);
4586 static int kvm_clear_stat_per_vcpu(struct kvm
*kvm
, size_t offset
)
4589 struct kvm_vcpu
*vcpu
;
4591 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4592 *(u64
*)((void *)vcpu
+ offset
) = 0;
4597 static int kvm_stat_data_get(void *data
, u64
*val
)
4600 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
4602 switch (stat_data
->dbgfs_item
->kind
) {
4604 r
= kvm_get_stat_per_vm(stat_data
->kvm
,
4605 stat_data
->dbgfs_item
->offset
, val
);
4608 r
= kvm_get_stat_per_vcpu(stat_data
->kvm
,
4609 stat_data
->dbgfs_item
->offset
, val
);
4616 static int kvm_stat_data_clear(void *data
, u64 val
)
4619 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
4624 switch (stat_data
->dbgfs_item
->kind
) {
4626 r
= kvm_clear_stat_per_vm(stat_data
->kvm
,
4627 stat_data
->dbgfs_item
->offset
);
4630 r
= kvm_clear_stat_per_vcpu(stat_data
->kvm
,
4631 stat_data
->dbgfs_item
->offset
);
4638 static int kvm_stat_data_open(struct inode
*inode
, struct file
*file
)
4640 __simple_attr_check_format("%llu\n", 0ull);
4641 return kvm_debugfs_open(inode
, file
, kvm_stat_data_get
,
4642 kvm_stat_data_clear
, "%llu\n");
4645 static const struct file_operations stat_fops_per_vm
= {
4646 .owner
= THIS_MODULE
,
4647 .open
= kvm_stat_data_open
,
4648 .release
= kvm_debugfs_release
,
4649 .read
= simple_attr_read
,
4650 .write
= simple_attr_write
,
4651 .llseek
= no_llseek
,
4654 static int vm_stat_get(void *_offset
, u64
*val
)
4656 unsigned offset
= (long)_offset
;
4661 mutex_lock(&kvm_lock
);
4662 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4663 kvm_get_stat_per_vm(kvm
, offset
, &tmp_val
);
4666 mutex_unlock(&kvm_lock
);
4670 static int vm_stat_clear(void *_offset
, u64 val
)
4672 unsigned offset
= (long)_offset
;
4678 mutex_lock(&kvm_lock
);
4679 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4680 kvm_clear_stat_per_vm(kvm
, offset
);
4682 mutex_unlock(&kvm_lock
);
4687 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, vm_stat_clear
, "%llu\n");
4689 static int vcpu_stat_get(void *_offset
, u64
*val
)
4691 unsigned offset
= (long)_offset
;
4696 mutex_lock(&kvm_lock
);
4697 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4698 kvm_get_stat_per_vcpu(kvm
, offset
, &tmp_val
);
4701 mutex_unlock(&kvm_lock
);
4705 static int vcpu_stat_clear(void *_offset
, u64 val
)
4707 unsigned offset
= (long)_offset
;
4713 mutex_lock(&kvm_lock
);
4714 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
4715 kvm_clear_stat_per_vcpu(kvm
, offset
);
4717 mutex_unlock(&kvm_lock
);
4722 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, vcpu_stat_clear
,
4725 static const struct file_operations
*stat_fops
[] = {
4726 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
4727 [KVM_STAT_VM
] = &vm_stat_fops
,
4730 static void kvm_uevent_notify_change(unsigned int type
, struct kvm
*kvm
)
4732 struct kobj_uevent_env
*env
;
4733 unsigned long long created
, active
;
4735 if (!kvm_dev
.this_device
|| !kvm
)
4738 mutex_lock(&kvm_lock
);
4739 if (type
== KVM_EVENT_CREATE_VM
) {
4740 kvm_createvm_count
++;
4742 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4745 created
= kvm_createvm_count
;
4746 active
= kvm_active_vms
;
4747 mutex_unlock(&kvm_lock
);
4749 env
= kzalloc(sizeof(*env
), GFP_KERNEL_ACCOUNT
);
4753 add_uevent_var(env
, "CREATED=%llu", created
);
4754 add_uevent_var(env
, "COUNT=%llu", active
);
4756 if (type
== KVM_EVENT_CREATE_VM
) {
4757 add_uevent_var(env
, "EVENT=create");
4758 kvm
->userspace_pid
= task_pid_nr(current
);
4759 } else if (type
== KVM_EVENT_DESTROY_VM
) {
4760 add_uevent_var(env
, "EVENT=destroy");
4762 add_uevent_var(env
, "PID=%d", kvm
->userspace_pid
);
4764 if (!IS_ERR_OR_NULL(kvm
->debugfs_dentry
)) {
4765 char *tmp
, *p
= kmalloc(PATH_MAX
, GFP_KERNEL_ACCOUNT
);
4768 tmp
= dentry_path_raw(kvm
->debugfs_dentry
, p
, PATH_MAX
);
4770 add_uevent_var(env
, "STATS_PATH=%s", tmp
);
4774 /* no need for checks, since we are adding at most only 5 keys */
4775 env
->envp
[env
->envp_idx
++] = NULL
;
4776 kobject_uevent_env(&kvm_dev
.this_device
->kobj
, KOBJ_CHANGE
, env
->envp
);
4780 static void kvm_init_debug(void)
4782 struct kvm_stats_debugfs_item
*p
;
4784 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
4786 kvm_debugfs_num_entries
= 0;
4787 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
4788 debugfs_create_file(p
->name
, KVM_DBGFS_GET_MODE(p
),
4789 kvm_debugfs_dir
, (void *)(long)p
->offset
,
4790 stat_fops
[p
->kind
]);
4794 static int kvm_suspend(void)
4796 if (kvm_usage_count
)
4797 hardware_disable_nolock(NULL
);
4801 static void kvm_resume(void)
4803 if (kvm_usage_count
) {
4804 #ifdef CONFIG_LOCKDEP
4805 WARN_ON(lockdep_is_held(&kvm_count_lock
));
4807 hardware_enable_nolock(NULL
);
4811 static struct syscore_ops kvm_syscore_ops
= {
4812 .suspend
= kvm_suspend
,
4813 .resume
= kvm_resume
,
4817 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
4819 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
4822 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
4824 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4826 WRITE_ONCE(vcpu
->preempted
, false);
4827 WRITE_ONCE(vcpu
->ready
, false);
4829 __this_cpu_write(kvm_running_vcpu
, vcpu
);
4830 kvm_arch_sched_in(vcpu
, cpu
);
4831 kvm_arch_vcpu_load(vcpu
, cpu
);
4834 static void kvm_sched_out(struct preempt_notifier
*pn
,
4835 struct task_struct
*next
)
4837 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
4839 if (current
->state
== TASK_RUNNING
) {
4840 WRITE_ONCE(vcpu
->preempted
, true);
4841 WRITE_ONCE(vcpu
->ready
, true);
4843 kvm_arch_vcpu_put(vcpu
);
4844 __this_cpu_write(kvm_running_vcpu
, NULL
);
4848 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4850 * We can disable preemption locally around accessing the per-CPU variable,
4851 * and use the resolved vcpu pointer after enabling preemption again,
4852 * because even if the current thread is migrated to another CPU, reading
4853 * the per-CPU value later will give us the same value as we update the
4854 * per-CPU variable in the preempt notifier handlers.
4856 struct kvm_vcpu
*kvm_get_running_vcpu(void)
4858 struct kvm_vcpu
*vcpu
;
4861 vcpu
= __this_cpu_read(kvm_running_vcpu
);
4866 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu
);
4869 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4871 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
4873 return &kvm_running_vcpu
;
4876 struct kvm_cpu_compat_check
{
4881 static void check_processor_compat(void *data
)
4883 struct kvm_cpu_compat_check
*c
= data
;
4885 *c
->ret
= kvm_arch_check_processor_compat(c
->opaque
);
4888 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
4889 struct module
*module
)
4891 struct kvm_cpu_compat_check c
;
4895 r
= kvm_arch_init(opaque
);
4900 * kvm_arch_init makes sure there's at most one caller
4901 * for architectures that support multiple implementations,
4902 * like intel and amd on x86.
4903 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4904 * conflicts in case kvm is already setup for another implementation.
4906 r
= kvm_irqfd_init();
4910 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
4915 r
= kvm_arch_hardware_setup(opaque
);
4921 for_each_online_cpu(cpu
) {
4922 smp_call_function_single(cpu
, check_processor_compat
, &c
, 1);
4927 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "kvm/cpu:starting",
4928 kvm_starting_cpu
, kvm_dying_cpu
);
4931 register_reboot_notifier(&kvm_reboot_notifier
);
4933 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4935 vcpu_align
= __alignof__(struct kvm_vcpu
);
4937 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size
, vcpu_align
,
4939 offsetof(struct kvm_vcpu
, arch
),
4940 sizeof_field(struct kvm_vcpu
, arch
),
4942 if (!kvm_vcpu_cache
) {
4947 r
= kvm_async_pf_init();
4951 kvm_chardev_ops
.owner
= module
;
4952 kvm_vm_fops
.owner
= module
;
4953 kvm_vcpu_fops
.owner
= module
;
4955 r
= misc_register(&kvm_dev
);
4957 pr_err("kvm: misc device register failed\n");
4961 register_syscore_ops(&kvm_syscore_ops
);
4963 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
4964 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
4968 r
= kvm_vfio_ops_init();
4974 kvm_async_pf_deinit();
4976 kmem_cache_destroy(kvm_vcpu_cache
);
4978 unregister_reboot_notifier(&kvm_reboot_notifier
);
4979 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
4981 kvm_arch_hardware_unsetup();
4983 free_cpumask_var(cpus_hardware_enabled
);
4991 EXPORT_SYMBOL_GPL(kvm_init
);
4995 debugfs_remove_recursive(kvm_debugfs_dir
);
4996 misc_deregister(&kvm_dev
);
4997 kmem_cache_destroy(kvm_vcpu_cache
);
4998 kvm_async_pf_deinit();
4999 unregister_syscore_ops(&kvm_syscore_ops
);
5000 unregister_reboot_notifier(&kvm_reboot_notifier
);
5001 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
5002 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
5003 kvm_arch_hardware_unsetup();
5006 free_cpumask_var(cpus_hardware_enabled
);
5007 kvm_vfio_ops_exit();
5009 EXPORT_SYMBOL_GPL(kvm_exit
);
5011 struct kvm_vm_worker_thread_context
{
5013 struct task_struct
*parent
;
5014 struct completion init_done
;
5015 kvm_vm_thread_fn_t thread_fn
;
5020 static int kvm_vm_worker_thread(void *context
)
5023 * The init_context is allocated on the stack of the parent thread, so
5024 * we have to locally copy anything that is needed beyond initialization
5026 struct kvm_vm_worker_thread_context
*init_context
= context
;
5027 struct kvm
*kvm
= init_context
->kvm
;
5028 kvm_vm_thread_fn_t thread_fn
= init_context
->thread_fn
;
5029 uintptr_t data
= init_context
->data
;
5032 err
= kthread_park(current
);
5033 /* kthread_park(current) is never supposed to return an error */
5038 err
= cgroup_attach_task_all(init_context
->parent
, current
);
5040 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5045 set_user_nice(current
, task_nice(init_context
->parent
));
5048 init_context
->err
= err
;
5049 complete(&init_context
->init_done
);
5050 init_context
= NULL
;
5055 /* Wait to be woken up by the spawner before proceeding. */
5058 if (!kthread_should_stop())
5059 err
= thread_fn(kvm
, data
);
5064 int kvm_vm_create_worker_thread(struct kvm
*kvm
, kvm_vm_thread_fn_t thread_fn
,
5065 uintptr_t data
, const char *name
,
5066 struct task_struct
**thread_ptr
)
5068 struct kvm_vm_worker_thread_context init_context
= {};
5069 struct task_struct
*thread
;
5072 init_context
.kvm
= kvm
;
5073 init_context
.parent
= current
;
5074 init_context
.thread_fn
= thread_fn
;
5075 init_context
.data
= data
;
5076 init_completion(&init_context
.init_done
);
5078 thread
= kthread_run(kvm_vm_worker_thread
, &init_context
,
5079 "%s-%d", name
, task_pid_nr(current
));
5081 return PTR_ERR(thread
);
5083 /* kthread_run is never supposed to return NULL */
5084 WARN_ON(thread
== NULL
);
5086 wait_for_completion(&init_context
.init_done
);
5088 if (!init_context
.err
)
5089 *thread_ptr
= thread
;
5091 return init_context
.err
;