2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
69 /* Architectures should define their poll value according to the halt latency */
70 static unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
71 module_param(halt_poll_ns
, uint
, S_IRUGO
| S_IWUSR
);
73 /* Default doubles per-vcpu halt_poll_ns. */
74 static unsigned int halt_poll_ns_grow
= 2;
75 module_param(halt_poll_ns_grow
, uint
, S_IRUGO
| S_IWUSR
);
77 /* Default resets per-vcpu halt_poll_ns . */
78 static unsigned int halt_poll_ns_shrink
;
79 module_param(halt_poll_ns_shrink
, uint
, S_IRUGO
| S_IWUSR
);
84 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
87 DEFINE_SPINLOCK(kvm_lock
);
88 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
91 static cpumask_var_t cpus_hardware_enabled
;
92 static int kvm_usage_count
;
93 static atomic_t hardware_enable_failed
;
95 struct kmem_cache
*kvm_vcpu_cache
;
96 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
98 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
100 struct dentry
*kvm_debugfs_dir
;
101 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
103 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
105 #ifdef CONFIG_KVM_COMPAT
106 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
109 static int hardware_enable_all(void);
110 static void hardware_disable_all(void);
112 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
114 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
);
115 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
117 __visible
bool kvm_rebooting
;
118 EXPORT_SYMBOL_GPL(kvm_rebooting
);
120 static bool largepages_enabled
= true;
122 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
125 return PageReserved(pfn_to_page(pfn
));
131 * Switches to specified vcpu, until a matching vcpu_put()
133 int vcpu_load(struct kvm_vcpu
*vcpu
)
137 if (mutex_lock_killable(&vcpu
->mutex
))
140 preempt_notifier_register(&vcpu
->preempt_notifier
);
141 kvm_arch_vcpu_load(vcpu
, cpu
);
146 void vcpu_put(struct kvm_vcpu
*vcpu
)
149 kvm_arch_vcpu_put(vcpu
);
150 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
152 mutex_unlock(&vcpu
->mutex
);
155 static void ack_flush(void *_completed
)
159 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
164 struct kvm_vcpu
*vcpu
;
166 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
169 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
170 kvm_make_request(req
, vcpu
);
173 /* Set ->requests bit before we read ->mode. */
174 smp_mb__after_atomic();
176 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
177 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
178 cpumask_set_cpu(cpu
, cpus
);
180 if (unlikely(cpus
== NULL
))
181 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
182 else if (!cpumask_empty(cpus
))
183 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
187 free_cpumask_var(cpus
);
191 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
192 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
194 long dirty_count
= kvm
->tlbs_dirty
;
197 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
198 ++kvm
->stat
.remote_tlb_flush
;
199 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
201 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
204 void kvm_reload_remote_mmus(struct kvm
*kvm
)
206 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
209 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
214 mutex_init(&vcpu
->mutex
);
219 init_swait_queue_head(&vcpu
->wq
);
220 kvm_async_pf_vcpu_init(vcpu
);
223 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
225 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
230 vcpu
->run
= page_address(page
);
232 kvm_vcpu_set_in_spin_loop(vcpu
, false);
233 kvm_vcpu_set_dy_eligible(vcpu
, false);
234 vcpu
->preempted
= false;
236 r
= kvm_arch_vcpu_init(vcpu
);
242 free_page((unsigned long)vcpu
->run
);
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
248 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
251 kvm_arch_vcpu_uninit(vcpu
);
252 free_page((unsigned long)vcpu
->run
);
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
259 return container_of(mn
, struct kvm
, mmu_notifier
);
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
263 struct mm_struct
*mm
,
264 unsigned long address
)
266 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
267 int need_tlb_flush
, idx
;
270 * When ->invalidate_page runs, the linux pte has been zapped
271 * already but the page is still allocated until
272 * ->invalidate_page returns. So if we increase the sequence
273 * here the kvm page fault will notice if the spte can't be
274 * established because the page is going to be freed. If
275 * instead the kvm page fault establishes the spte before
276 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * The sequence increase only need to be seen at spin_unlock
280 * time, and not at spin_lock time.
282 * Increasing the sequence after the spin_unlock would be
283 * unsafe because the kvm page fault could then establish the
284 * pte after kvm_unmap_hva returned, without noticing the page
285 * is going to be freed.
287 idx
= srcu_read_lock(&kvm
->srcu
);
288 spin_lock(&kvm
->mmu_lock
);
290 kvm
->mmu_notifier_seq
++;
291 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
292 /* we've to flush the tlb before the pages can be freed */
294 kvm_flush_remote_tlbs(kvm
);
296 spin_unlock(&kvm
->mmu_lock
);
298 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
300 srcu_read_unlock(&kvm
->srcu
, idx
);
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
304 struct mm_struct
*mm
,
305 unsigned long address
,
308 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
311 idx
= srcu_read_lock(&kvm
->srcu
);
312 spin_lock(&kvm
->mmu_lock
);
313 kvm
->mmu_notifier_seq
++;
314 kvm_set_spte_hva(kvm
, address
, pte
);
315 spin_unlock(&kvm
->mmu_lock
);
316 srcu_read_unlock(&kvm
->srcu
, idx
);
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
320 struct mm_struct
*mm
,
324 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
325 int need_tlb_flush
= 0, idx
;
327 idx
= srcu_read_lock(&kvm
->srcu
);
328 spin_lock(&kvm
->mmu_lock
);
330 * The count increase must become visible at unlock time as no
331 * spte can be established without taking the mmu_lock and
332 * count is also read inside the mmu_lock critical section.
334 kvm
->mmu_notifier_count
++;
335 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
336 need_tlb_flush
|= kvm
->tlbs_dirty
;
337 /* we've to flush the tlb before the pages can be freed */
339 kvm_flush_remote_tlbs(kvm
);
341 spin_unlock(&kvm
->mmu_lock
);
342 srcu_read_unlock(&kvm
->srcu
, idx
);
345 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
346 struct mm_struct
*mm
,
350 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
352 spin_lock(&kvm
->mmu_lock
);
354 * This sequence increase will notify the kvm page fault that
355 * the page that is going to be mapped in the spte could have
358 kvm
->mmu_notifier_seq
++;
361 * The above sequence increase must be visible before the
362 * below count decrease, which is ensured by the smp_wmb above
363 * in conjunction with the smp_rmb in mmu_notifier_retry().
365 kvm
->mmu_notifier_count
--;
366 spin_unlock(&kvm
->mmu_lock
);
368 BUG_ON(kvm
->mmu_notifier_count
< 0);
371 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
372 struct mm_struct
*mm
,
376 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
379 idx
= srcu_read_lock(&kvm
->srcu
);
380 spin_lock(&kvm
->mmu_lock
);
382 young
= kvm_age_hva(kvm
, start
, end
);
384 kvm_flush_remote_tlbs(kvm
);
386 spin_unlock(&kvm
->mmu_lock
);
387 srcu_read_unlock(&kvm
->srcu
, idx
);
392 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
393 struct mm_struct
*mm
,
397 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
400 idx
= srcu_read_lock(&kvm
->srcu
);
401 spin_lock(&kvm
->mmu_lock
);
403 * Even though we do not flush TLB, this will still adversely
404 * affect performance on pre-Haswell Intel EPT, where there is
405 * no EPT Access Bit to clear so that we have to tear down EPT
406 * tables instead. If we find this unacceptable, we can always
407 * add a parameter to kvm_age_hva so that it effectively doesn't
408 * do anything on clear_young.
410 * Also note that currently we never issue secondary TLB flushes
411 * from clear_young, leaving this job up to the regular system
412 * cadence. If we find this inaccurate, we might come up with a
413 * more sophisticated heuristic later.
415 young
= kvm_age_hva(kvm
, start
, end
);
416 spin_unlock(&kvm
->mmu_lock
);
417 srcu_read_unlock(&kvm
->srcu
, idx
);
422 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
423 struct mm_struct
*mm
,
424 unsigned long address
)
426 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
429 idx
= srcu_read_lock(&kvm
->srcu
);
430 spin_lock(&kvm
->mmu_lock
);
431 young
= kvm_test_age_hva(kvm
, address
);
432 spin_unlock(&kvm
->mmu_lock
);
433 srcu_read_unlock(&kvm
->srcu
, idx
);
438 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
439 struct mm_struct
*mm
)
441 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
444 idx
= srcu_read_lock(&kvm
->srcu
);
445 kvm_arch_flush_shadow_all(kvm
);
446 srcu_read_unlock(&kvm
->srcu
, idx
);
449 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
450 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
451 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
452 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
453 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
454 .clear_young
= kvm_mmu_notifier_clear_young
,
455 .test_young
= kvm_mmu_notifier_test_young
,
456 .change_pte
= kvm_mmu_notifier_change_pte
,
457 .release
= kvm_mmu_notifier_release
,
460 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
462 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
463 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
466 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
468 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
473 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
475 static struct kvm_memslots
*kvm_alloc_memslots(void)
478 struct kvm_memslots
*slots
;
480 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
485 * Init kvm generation close to the maximum to easily test the
486 * code of handling generation number wrap-around.
488 slots
->generation
= -150;
489 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
490 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
495 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
497 if (!memslot
->dirty_bitmap
)
500 kvfree(memslot
->dirty_bitmap
);
501 memslot
->dirty_bitmap
= NULL
;
505 * Free any memory in @free but not in @dont.
507 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
508 struct kvm_memory_slot
*dont
)
510 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
511 kvm_destroy_dirty_bitmap(free
);
513 kvm_arch_free_memslot(kvm
, free
, dont
);
518 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
520 struct kvm_memory_slot
*memslot
;
525 kvm_for_each_memslot(memslot
, slots
)
526 kvm_free_memslot(kvm
, memslot
, NULL
);
531 static struct kvm
*kvm_create_vm(unsigned long type
)
534 struct kvm
*kvm
= kvm_arch_alloc_vm();
537 return ERR_PTR(-ENOMEM
);
539 spin_lock_init(&kvm
->mmu_lock
);
540 atomic_inc(¤t
->mm
->mm_count
);
541 kvm
->mm
= current
->mm
;
542 kvm_eventfd_init(kvm
);
543 mutex_init(&kvm
->lock
);
544 mutex_init(&kvm
->irq_lock
);
545 mutex_init(&kvm
->slots_lock
);
546 atomic_set(&kvm
->users_count
, 1);
547 INIT_LIST_HEAD(&kvm
->devices
);
549 r
= kvm_arch_init_vm(kvm
, type
);
551 goto out_err_no_disable
;
553 r
= hardware_enable_all();
555 goto out_err_no_disable
;
557 #ifdef CONFIG_HAVE_KVM_IRQFD
558 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
561 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
564 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
565 kvm
->memslots
[i
] = kvm_alloc_memslots();
566 if (!kvm
->memslots
[i
])
567 goto out_err_no_srcu
;
570 if (init_srcu_struct(&kvm
->srcu
))
571 goto out_err_no_srcu
;
572 if (init_srcu_struct(&kvm
->irq_srcu
))
573 goto out_err_no_irq_srcu
;
574 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
575 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
581 r
= kvm_init_mmu_notifier(kvm
);
585 spin_lock(&kvm_lock
);
586 list_add(&kvm
->vm_list
, &vm_list
);
587 spin_unlock(&kvm_lock
);
589 preempt_notifier_inc();
594 cleanup_srcu_struct(&kvm
->irq_srcu
);
596 cleanup_srcu_struct(&kvm
->srcu
);
598 hardware_disable_all();
600 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
601 kfree(kvm
->buses
[i
]);
602 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
603 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
604 kvm_arch_free_vm(kvm
);
610 * Avoid using vmalloc for a small buffer.
611 * Should not be used when the size is statically known.
613 void *kvm_kvzalloc(unsigned long size
)
615 if (size
> PAGE_SIZE
)
616 return vzalloc(size
);
618 return kzalloc(size
, GFP_KERNEL
);
621 static void kvm_destroy_devices(struct kvm
*kvm
)
623 struct kvm_device
*dev
, *tmp
;
625 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
626 list_del(&dev
->vm_node
);
627 dev
->ops
->destroy(dev
);
631 static void kvm_destroy_vm(struct kvm
*kvm
)
634 struct mm_struct
*mm
= kvm
->mm
;
636 kvm_arch_sync_events(kvm
);
637 spin_lock(&kvm_lock
);
638 list_del(&kvm
->vm_list
);
639 spin_unlock(&kvm_lock
);
640 kvm_free_irq_routing(kvm
);
641 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
642 kvm_io_bus_destroy(kvm
->buses
[i
]);
643 kvm_coalesced_mmio_free(kvm
);
644 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
645 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
647 kvm_arch_flush_shadow_all(kvm
);
649 kvm_arch_destroy_vm(kvm
);
650 kvm_destroy_devices(kvm
);
651 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
652 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
653 cleanup_srcu_struct(&kvm
->irq_srcu
);
654 cleanup_srcu_struct(&kvm
->srcu
);
655 kvm_arch_free_vm(kvm
);
656 preempt_notifier_dec();
657 hardware_disable_all();
661 void kvm_get_kvm(struct kvm
*kvm
)
663 atomic_inc(&kvm
->users_count
);
665 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
667 void kvm_put_kvm(struct kvm
*kvm
)
669 if (atomic_dec_and_test(&kvm
->users_count
))
672 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
675 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
677 struct kvm
*kvm
= filp
->private_data
;
679 kvm_irqfd_release(kvm
);
686 * Allocation size is twice as large as the actual dirty bitmap size.
687 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
689 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
691 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
693 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
694 if (!memslot
->dirty_bitmap
)
701 * Insert memslot and re-sort memslots based on their GFN,
702 * so binary search could be used to lookup GFN.
703 * Sorting algorithm takes advantage of having initially
704 * sorted array and known changed memslot position.
706 static void update_memslots(struct kvm_memslots
*slots
,
707 struct kvm_memory_slot
*new)
710 int i
= slots
->id_to_index
[id
];
711 struct kvm_memory_slot
*mslots
= slots
->memslots
;
713 WARN_ON(mslots
[i
].id
!= id
);
715 WARN_ON(!mslots
[i
].npages
);
716 if (mslots
[i
].npages
)
719 if (!mslots
[i
].npages
)
723 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
724 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
725 if (!mslots
[i
+ 1].npages
)
727 mslots
[i
] = mslots
[i
+ 1];
728 slots
->id_to_index
[mslots
[i
].id
] = i
;
733 * The ">=" is needed when creating a slot with base_gfn == 0,
734 * so that it moves before all those with base_gfn == npages == 0.
736 * On the other hand, if new->npages is zero, the above loop has
737 * already left i pointing to the beginning of the empty part of
738 * mslots, and the ">=" would move the hole backwards in this
739 * case---which is wrong. So skip the loop when deleting a slot.
743 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
744 mslots
[i
] = mslots
[i
- 1];
745 slots
->id_to_index
[mslots
[i
].id
] = i
;
749 WARN_ON_ONCE(i
!= slots
->used_slots
);
752 slots
->id_to_index
[mslots
[i
].id
] = i
;
755 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
757 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
759 #ifdef __KVM_HAVE_READONLY_MEM
760 valid_flags
|= KVM_MEM_READONLY
;
763 if (mem
->flags
& ~valid_flags
)
769 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
770 int as_id
, struct kvm_memslots
*slots
)
772 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
775 * Set the low bit in the generation, which disables SPTE caching
776 * until the end of synchronize_srcu_expedited.
778 WARN_ON(old_memslots
->generation
& 1);
779 slots
->generation
= old_memslots
->generation
+ 1;
781 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
782 synchronize_srcu_expedited(&kvm
->srcu
);
785 * Increment the new memslot generation a second time. This prevents
786 * vm exits that race with memslot updates from caching a memslot
787 * generation that will (potentially) be valid forever.
791 kvm_arch_memslots_updated(kvm
, slots
);
797 * Allocate some memory and give it an address in the guest physical address
800 * Discontiguous memory is allowed, mostly for framebuffers.
802 * Must be called holding kvm->slots_lock for write.
804 int __kvm_set_memory_region(struct kvm
*kvm
,
805 const struct kvm_userspace_memory_region
*mem
)
809 unsigned long npages
;
810 struct kvm_memory_slot
*slot
;
811 struct kvm_memory_slot old
, new;
812 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
814 enum kvm_mr_change change
;
816 r
= check_memory_region_flags(mem
);
821 as_id
= mem
->slot
>> 16;
824 /* General sanity checks */
825 if (mem
->memory_size
& (PAGE_SIZE
- 1))
827 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
829 /* We can read the guest memory with __xxx_user() later on. */
830 if ((id
< KVM_USER_MEM_SLOTS
) &&
831 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
832 !access_ok(VERIFY_WRITE
,
833 (void __user
*)(unsigned long)mem
->userspace_addr
,
836 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
838 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
841 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
842 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
843 npages
= mem
->memory_size
>> PAGE_SHIFT
;
845 if (npages
> KVM_MEM_MAX_NR_PAGES
)
851 new.base_gfn
= base_gfn
;
853 new.flags
= mem
->flags
;
857 change
= KVM_MR_CREATE
;
858 else { /* Modify an existing slot. */
859 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
860 (npages
!= old
.npages
) ||
861 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
864 if (base_gfn
!= old
.base_gfn
)
865 change
= KVM_MR_MOVE
;
866 else if (new.flags
!= old
.flags
)
867 change
= KVM_MR_FLAGS_ONLY
;
868 else { /* Nothing to change. */
877 change
= KVM_MR_DELETE
;
882 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
883 /* Check for overlaps */
885 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
886 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
889 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
890 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
895 /* Free page dirty bitmap if unneeded */
896 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
897 new.dirty_bitmap
= NULL
;
900 if (change
== KVM_MR_CREATE
) {
901 new.userspace_addr
= mem
->userspace_addr
;
903 if (kvm_arch_create_memslot(kvm
, &new, npages
))
907 /* Allocate page dirty bitmap if needed */
908 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
909 if (kvm_create_dirty_bitmap(&new) < 0)
913 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
916 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
918 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
919 slot
= id_to_memslot(slots
, id
);
920 slot
->flags
|= KVM_MEMSLOT_INVALID
;
922 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
924 /* slot was deleted or moved, clear iommu mapping */
925 kvm_iommu_unmap_pages(kvm
, &old
);
926 /* From this point no new shadow pages pointing to a deleted,
927 * or moved, memslot will be created.
929 * validation of sp->gfn happens in:
930 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
931 * - kvm_is_visible_gfn (mmu_check_roots)
933 kvm_arch_flush_shadow_memslot(kvm
, slot
);
936 * We can re-use the old_memslots from above, the only difference
937 * from the currently installed memslots is the invalid flag. This
938 * will get overwritten by update_memslots anyway.
940 slots
= old_memslots
;
943 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
947 /* actual memory is freed via old in kvm_free_memslot below */
948 if (change
== KVM_MR_DELETE
) {
949 new.dirty_bitmap
= NULL
;
950 memset(&new.arch
, 0, sizeof(new.arch
));
953 update_memslots(slots
, &new);
954 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
956 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
958 kvm_free_memslot(kvm
, &old
, &new);
959 kvfree(old_memslots
);
962 * IOMMU mapping: New slots need to be mapped. Old slots need to be
963 * un-mapped and re-mapped if their base changes. Since base change
964 * unmapping is handled above with slot deletion, mapping alone is
965 * needed here. Anything else the iommu might care about for existing
966 * slots (size changes, userspace addr changes and read-only flag
967 * changes) is disallowed above, so any other attribute changes getting
968 * here can be skipped.
970 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
971 r
= kvm_iommu_map_pages(kvm
, &new);
980 kvm_free_memslot(kvm
, &new, &old
);
984 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
986 int kvm_set_memory_region(struct kvm
*kvm
,
987 const struct kvm_userspace_memory_region
*mem
)
991 mutex_lock(&kvm
->slots_lock
);
992 r
= __kvm_set_memory_region(kvm
, mem
);
993 mutex_unlock(&kvm
->slots_lock
);
996 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
998 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
999 struct kvm_userspace_memory_region
*mem
)
1001 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1004 return kvm_set_memory_region(kvm
, mem
);
1007 int kvm_get_dirty_log(struct kvm
*kvm
,
1008 struct kvm_dirty_log
*log
, int *is_dirty
)
1010 struct kvm_memslots
*slots
;
1011 struct kvm_memory_slot
*memslot
;
1012 int r
, i
, as_id
, id
;
1014 unsigned long any
= 0;
1017 as_id
= log
->slot
>> 16;
1018 id
= (u16
)log
->slot
;
1019 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1022 slots
= __kvm_memslots(kvm
, as_id
);
1023 memslot
= id_to_memslot(slots
, id
);
1025 if (!memslot
->dirty_bitmap
)
1028 n
= kvm_dirty_bitmap_bytes(memslot
);
1030 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1031 any
= memslot
->dirty_bitmap
[i
];
1034 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1044 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1046 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1048 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1049 * are dirty write protect them for next write.
1050 * @kvm: pointer to kvm instance
1051 * @log: slot id and address to which we copy the log
1052 * @is_dirty: flag set if any page is dirty
1054 * We need to keep it in mind that VCPU threads can write to the bitmap
1055 * concurrently. So, to avoid losing track of dirty pages we keep the
1058 * 1. Take a snapshot of the bit and clear it if needed.
1059 * 2. Write protect the corresponding page.
1060 * 3. Copy the snapshot to the userspace.
1061 * 4. Upon return caller flushes TLB's if needed.
1063 * Between 2 and 4, the guest may write to the page using the remaining TLB
1064 * entry. This is not a problem because the page is reported dirty using
1065 * the snapshot taken before and step 4 ensures that writes done after
1066 * exiting to userspace will be logged for the next call.
1069 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1070 struct kvm_dirty_log
*log
, bool *is_dirty
)
1072 struct kvm_memslots
*slots
;
1073 struct kvm_memory_slot
*memslot
;
1074 int r
, i
, as_id
, id
;
1076 unsigned long *dirty_bitmap
;
1077 unsigned long *dirty_bitmap_buffer
;
1080 as_id
= log
->slot
>> 16;
1081 id
= (u16
)log
->slot
;
1082 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1085 slots
= __kvm_memslots(kvm
, as_id
);
1086 memslot
= id_to_memslot(slots
, id
);
1088 dirty_bitmap
= memslot
->dirty_bitmap
;
1093 n
= kvm_dirty_bitmap_bytes(memslot
);
1095 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1096 memset(dirty_bitmap_buffer
, 0, n
);
1098 spin_lock(&kvm
->mmu_lock
);
1100 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1104 if (!dirty_bitmap
[i
])
1109 mask
= xchg(&dirty_bitmap
[i
], 0);
1110 dirty_bitmap_buffer
[i
] = mask
;
1113 offset
= i
* BITS_PER_LONG
;
1114 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1119 spin_unlock(&kvm
->mmu_lock
);
1122 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1129 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1132 bool kvm_largepages_enabled(void)
1134 return largepages_enabled
;
1137 void kvm_disable_largepages(void)
1139 largepages_enabled
= false;
1141 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1143 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1145 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1147 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1149 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1151 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1154 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1156 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1158 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1159 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1164 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1166 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1168 struct vm_area_struct
*vma
;
1169 unsigned long addr
, size
;
1173 addr
= gfn_to_hva(kvm
, gfn
);
1174 if (kvm_is_error_hva(addr
))
1177 down_read(¤t
->mm
->mmap_sem
);
1178 vma
= find_vma(current
->mm
, addr
);
1182 size
= vma_kernel_pagesize(vma
);
1185 up_read(¤t
->mm
->mmap_sem
);
1190 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1192 return slot
->flags
& KVM_MEM_READONLY
;
1195 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1196 gfn_t
*nr_pages
, bool write
)
1198 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1199 return KVM_HVA_ERR_BAD
;
1201 if (memslot_is_readonly(slot
) && write
)
1202 return KVM_HVA_ERR_RO_BAD
;
1205 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1207 return __gfn_to_hva_memslot(slot
, gfn
);
1210 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1213 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1216 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1219 return gfn_to_hva_many(slot
, gfn
, NULL
);
1221 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1223 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1225 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1227 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1229 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1231 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1233 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1236 * If writable is set to false, the hva returned by this function is only
1237 * allowed to be read.
1239 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1240 gfn_t gfn
, bool *writable
)
1242 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1244 if (!kvm_is_error_hva(hva
) && writable
)
1245 *writable
= !memslot_is_readonly(slot
);
1250 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1252 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1254 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1257 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1259 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1261 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1264 static int get_user_page_nowait(unsigned long start
, int write
,
1267 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1270 flags
|= FOLL_WRITE
;
1272 return __get_user_pages(current
, current
->mm
, start
, 1, flags
, page
,
1276 static inline int check_user_page_hwpoison(unsigned long addr
)
1278 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1280 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1281 flags
, NULL
, NULL
, NULL
);
1282 return rc
== -EHWPOISON
;
1286 * The atomic path to get the writable pfn which will be stored in @pfn,
1287 * true indicates success, otherwise false is returned.
1289 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1290 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1292 struct page
*page
[1];
1295 if (!(async
|| atomic
))
1299 * Fast pin a writable pfn only if it is a write fault request
1300 * or the caller allows to map a writable pfn for a read fault
1303 if (!(write_fault
|| writable
))
1306 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1308 *pfn
= page_to_pfn(page
[0]);
1319 * The slow path to get the pfn of the specified host virtual address,
1320 * 1 indicates success, -errno is returned if error is detected.
1322 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1323 bool *writable
, kvm_pfn_t
*pfn
)
1325 struct page
*page
[1];
1331 *writable
= write_fault
;
1334 down_read(¤t
->mm
->mmap_sem
);
1335 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1336 up_read(¤t
->mm
->mmap_sem
);
1338 npages
= __get_user_pages_unlocked(current
, current
->mm
, addr
, 1,
1339 write_fault
, 0, page
,
1340 FOLL_TOUCH
|FOLL_HWPOISON
);
1344 /* map read fault as writable if possible */
1345 if (unlikely(!write_fault
) && writable
) {
1346 struct page
*wpage
[1];
1348 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1357 *pfn
= page_to_pfn(page
[0]);
1361 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1363 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1366 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1373 * Pin guest page in memory and return its pfn.
1374 * @addr: host virtual address which maps memory to the guest
1375 * @atomic: whether this function can sleep
1376 * @async: whether this function need to wait IO complete if the
1377 * host page is not in the memory
1378 * @write_fault: whether we should get a writable host page
1379 * @writable: whether it allows to map a writable host page for !@write_fault
1381 * The function will map a writable host page for these two cases:
1382 * 1): @write_fault = true
1383 * 2): @write_fault = false && @writable, @writable will tell the caller
1384 * whether the mapping is writable.
1386 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1387 bool write_fault
, bool *writable
)
1389 struct vm_area_struct
*vma
;
1393 /* we can do it either atomically or asynchronously, not both */
1394 BUG_ON(atomic
&& async
);
1396 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1400 return KVM_PFN_ERR_FAULT
;
1402 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1406 down_read(¤t
->mm
->mmap_sem
);
1407 if (npages
== -EHWPOISON
||
1408 (!async
&& check_user_page_hwpoison(addr
))) {
1409 pfn
= KVM_PFN_ERR_HWPOISON
;
1413 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1416 pfn
= KVM_PFN_ERR_FAULT
;
1417 else if ((vma
->vm_flags
& VM_PFNMAP
)) {
1418 pfn
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) +
1420 BUG_ON(!kvm_is_reserved_pfn(pfn
));
1422 if (async
&& vma_is_valid(vma
, write_fault
))
1424 pfn
= KVM_PFN_ERR_FAULT
;
1427 up_read(¤t
->mm
->mmap_sem
);
1431 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1432 bool atomic
, bool *async
, bool write_fault
,
1435 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1437 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1440 return KVM_PFN_ERR_RO_FAULT
;
1443 if (kvm_is_error_hva(addr
)) {
1446 return KVM_PFN_NOSLOT
;
1449 /* Do not map writable pfn in the readonly memslot. */
1450 if (writable
&& memslot_is_readonly(slot
)) {
1455 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1458 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1460 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1463 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1464 write_fault
, writable
);
1466 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1468 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1470 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1472 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1474 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1476 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1478 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1480 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1482 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1484 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1486 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1488 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1490 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1492 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1494 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1496 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1498 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1500 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1502 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1504 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1505 struct page
**pages
, int nr_pages
)
1510 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1511 if (kvm_is_error_hva(addr
))
1514 if (entry
< nr_pages
)
1517 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1519 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1521 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1523 if (is_error_noslot_pfn(pfn
))
1524 return KVM_ERR_PTR_BAD_PAGE
;
1526 if (kvm_is_reserved_pfn(pfn
)) {
1528 return KVM_ERR_PTR_BAD_PAGE
;
1531 return pfn_to_page(pfn
);
1534 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1538 pfn
= gfn_to_pfn(kvm
, gfn
);
1540 return kvm_pfn_to_page(pfn
);
1542 EXPORT_SYMBOL_GPL(gfn_to_page
);
1544 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1548 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1550 return kvm_pfn_to_page(pfn
);
1552 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1554 void kvm_release_page_clean(struct page
*page
)
1556 WARN_ON(is_error_page(page
));
1558 kvm_release_pfn_clean(page_to_pfn(page
));
1560 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1562 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1564 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1565 put_page(pfn_to_page(pfn
));
1567 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1569 void kvm_release_page_dirty(struct page
*page
)
1571 WARN_ON(is_error_page(page
));
1573 kvm_release_pfn_dirty(page_to_pfn(page
));
1575 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1577 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1579 kvm_set_pfn_dirty(pfn
);
1580 kvm_release_pfn_clean(pfn
);
1583 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1585 if (!kvm_is_reserved_pfn(pfn
)) {
1586 struct page
*page
= pfn_to_page(pfn
);
1588 if (!PageReserved(page
))
1592 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1594 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1596 if (!kvm_is_reserved_pfn(pfn
))
1597 mark_page_accessed(pfn_to_page(pfn
));
1599 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1601 void kvm_get_pfn(kvm_pfn_t pfn
)
1603 if (!kvm_is_reserved_pfn(pfn
))
1604 get_page(pfn_to_page(pfn
));
1606 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1608 static int next_segment(unsigned long len
, int offset
)
1610 if (len
> PAGE_SIZE
- offset
)
1611 return PAGE_SIZE
- offset
;
1616 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1617 void *data
, int offset
, int len
)
1622 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1623 if (kvm_is_error_hva(addr
))
1625 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1631 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1634 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1636 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1638 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1640 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1641 int offset
, int len
)
1643 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1645 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1647 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1649 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1651 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1653 int offset
= offset_in_page(gpa
);
1656 while ((seg
= next_segment(len
, offset
)) != 0) {
1657 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1667 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1669 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1671 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1673 int offset
= offset_in_page(gpa
);
1676 while ((seg
= next_segment(len
, offset
)) != 0) {
1677 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1687 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1689 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1690 void *data
, int offset
, unsigned long len
)
1695 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1696 if (kvm_is_error_hva(addr
))
1698 pagefault_disable();
1699 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1706 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1709 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1710 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1711 int offset
= offset_in_page(gpa
);
1713 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1715 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1717 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1718 void *data
, unsigned long len
)
1720 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1721 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1722 int offset
= offset_in_page(gpa
);
1724 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1726 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1728 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1729 const void *data
, int offset
, int len
)
1734 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1735 if (kvm_is_error_hva(addr
))
1737 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1740 mark_page_dirty_in_slot(memslot
, gfn
);
1744 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1745 const void *data
, int offset
, int len
)
1747 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1749 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1751 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1753 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1754 const void *data
, int offset
, int len
)
1756 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1758 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1760 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1762 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1765 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1767 int offset
= offset_in_page(gpa
);
1770 while ((seg
= next_segment(len
, offset
)) != 0) {
1771 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1781 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1783 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1786 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1788 int offset
= offset_in_page(gpa
);
1791 while ((seg
= next_segment(len
, offset
)) != 0) {
1792 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1802 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1804 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1805 gpa_t gpa
, unsigned long len
)
1807 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1808 int offset
= offset_in_page(gpa
);
1809 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1810 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1811 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1812 gfn_t nr_pages_avail
;
1815 ghc
->generation
= slots
->generation
;
1817 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1818 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1819 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1823 * If the requested region crosses two memslots, we still
1824 * verify that the entire region is valid here.
1826 while (start_gfn
<= end_gfn
) {
1827 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1828 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1830 if (kvm_is_error_hva(ghc
->hva
))
1832 start_gfn
+= nr_pages_avail
;
1834 /* Use the slow path for cross page reads and writes. */
1835 ghc
->memslot
= NULL
;
1839 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1841 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1842 void *data
, unsigned long len
)
1844 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1847 BUG_ON(len
> ghc
->len
);
1849 if (slots
->generation
!= ghc
->generation
)
1850 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1852 if (unlikely(!ghc
->memslot
))
1853 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1855 if (kvm_is_error_hva(ghc
->hva
))
1858 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1861 mark_page_dirty_in_slot(ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1865 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1867 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1868 void *data
, unsigned long len
)
1870 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1873 BUG_ON(len
> ghc
->len
);
1875 if (slots
->generation
!= ghc
->generation
)
1876 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1878 if (unlikely(!ghc
->memslot
))
1879 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
1881 if (kvm_is_error_hva(ghc
->hva
))
1884 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
1890 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
1892 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
1894 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1896 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
1898 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
1900 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
1902 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1904 int offset
= offset_in_page(gpa
);
1907 while ((seg
= next_segment(len
, offset
)) != 0) {
1908 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
1917 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
1919 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
1922 if (memslot
&& memslot
->dirty_bitmap
) {
1923 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
1925 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
1929 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
1931 struct kvm_memory_slot
*memslot
;
1933 memslot
= gfn_to_memslot(kvm
, gfn
);
1934 mark_page_dirty_in_slot(memslot
, gfn
);
1936 EXPORT_SYMBOL_GPL(mark_page_dirty
);
1938 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1940 struct kvm_memory_slot
*memslot
;
1942 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1943 mark_page_dirty_in_slot(memslot
, gfn
);
1945 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
1947 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
1949 unsigned int old
, val
, grow
;
1951 old
= val
= vcpu
->halt_poll_ns
;
1952 grow
= READ_ONCE(halt_poll_ns_grow
);
1954 if (val
== 0 && grow
)
1959 if (val
> halt_poll_ns
)
1962 vcpu
->halt_poll_ns
= val
;
1963 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
1966 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
1968 unsigned int old
, val
, shrink
;
1970 old
= val
= vcpu
->halt_poll_ns
;
1971 shrink
= READ_ONCE(halt_poll_ns_shrink
);
1977 vcpu
->halt_poll_ns
= val
;
1978 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
1981 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
1983 if (kvm_arch_vcpu_runnable(vcpu
)) {
1984 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
1987 if (kvm_cpu_has_pending_timer(vcpu
))
1989 if (signal_pending(current
))
1996 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1998 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2001 DECLARE_SWAITQUEUE(wait
);
2002 bool waited
= false;
2005 start
= cur
= ktime_get();
2006 if (vcpu
->halt_poll_ns
) {
2007 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2009 ++vcpu
->stat
.halt_attempted_poll
;
2012 * This sets KVM_REQ_UNHALT if an interrupt
2015 if (kvm_vcpu_check_block(vcpu
) < 0) {
2016 ++vcpu
->stat
.halt_successful_poll
;
2020 } while (single_task_running() && ktime_before(cur
, stop
));
2023 kvm_arch_vcpu_blocking(vcpu
);
2026 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2028 if (kvm_vcpu_check_block(vcpu
) < 0)
2035 finish_swait(&vcpu
->wq
, &wait
);
2038 kvm_arch_vcpu_unblocking(vcpu
);
2040 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2043 if (block_ns
<= vcpu
->halt_poll_ns
)
2045 /* we had a long block, shrink polling */
2046 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2047 shrink_halt_poll_ns(vcpu
);
2048 /* we had a short halt and our poll time is too small */
2049 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2050 block_ns
< halt_poll_ns
)
2051 grow_halt_poll_ns(vcpu
);
2053 vcpu
->halt_poll_ns
= 0;
2055 trace_kvm_vcpu_wakeup(block_ns
, waited
);
2057 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2061 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2063 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2066 int cpu
= vcpu
->cpu
;
2067 struct swait_queue_head
*wqp
;
2069 wqp
= kvm_arch_vcpu_wq(vcpu
);
2070 if (swait_active(wqp
)) {
2072 ++vcpu
->stat
.halt_wakeup
;
2076 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2077 if (kvm_arch_vcpu_should_kick(vcpu
))
2078 smp_send_reschedule(cpu
);
2081 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2082 #endif /* !CONFIG_S390 */
2084 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2087 struct task_struct
*task
= NULL
;
2091 pid
= rcu_dereference(target
->pid
);
2093 task
= get_pid_task(pid
, PIDTYPE_PID
);
2097 ret
= yield_to(task
, 1);
2098 put_task_struct(task
);
2102 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2105 * Helper that checks whether a VCPU is eligible for directed yield.
2106 * Most eligible candidate to yield is decided by following heuristics:
2108 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2109 * (preempted lock holder), indicated by @in_spin_loop.
2110 * Set at the beiginning and cleared at the end of interception/PLE handler.
2112 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2113 * chance last time (mostly it has become eligible now since we have probably
2114 * yielded to lockholder in last iteration. This is done by toggling
2115 * @dy_eligible each time a VCPU checked for eligibility.)
2117 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2118 * to preempted lock-holder could result in wrong VCPU selection and CPU
2119 * burning. Giving priority for a potential lock-holder increases lock
2122 * Since algorithm is based on heuristics, accessing another VCPU data without
2123 * locking does not harm. It may result in trying to yield to same VCPU, fail
2124 * and continue with next VCPU and so on.
2126 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2128 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2131 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2132 vcpu
->spin_loop
.dy_eligible
;
2134 if (vcpu
->spin_loop
.in_spin_loop
)
2135 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2143 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2145 struct kvm
*kvm
= me
->kvm
;
2146 struct kvm_vcpu
*vcpu
;
2147 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2153 kvm_vcpu_set_in_spin_loop(me
, true);
2155 * We boost the priority of a VCPU that is runnable but not
2156 * currently running, because it got preempted by something
2157 * else and called schedule in __vcpu_run. Hopefully that
2158 * VCPU is holding the lock that we need and will release it.
2159 * We approximate round-robin by starting at the last boosted VCPU.
2161 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2162 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2163 if (!pass
&& i
<= last_boosted_vcpu
) {
2164 i
= last_boosted_vcpu
;
2166 } else if (pass
&& i
> last_boosted_vcpu
)
2168 if (!ACCESS_ONCE(vcpu
->preempted
))
2172 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2174 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2177 yielded
= kvm_vcpu_yield_to(vcpu
);
2179 kvm
->last_boosted_vcpu
= i
;
2181 } else if (yielded
< 0) {
2188 kvm_vcpu_set_in_spin_loop(me
, false);
2190 /* Ensure vcpu is not eligible during next spinloop */
2191 kvm_vcpu_set_dy_eligible(me
, false);
2193 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2195 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2197 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2200 if (vmf
->pgoff
== 0)
2201 page
= virt_to_page(vcpu
->run
);
2203 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2204 page
= virt_to_page(vcpu
->arch
.pio_data
);
2206 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2207 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2208 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2211 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2217 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2218 .fault
= kvm_vcpu_fault
,
2221 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2223 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2227 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2229 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2231 kvm_put_kvm(vcpu
->kvm
);
2235 static struct file_operations kvm_vcpu_fops
= {
2236 .release
= kvm_vcpu_release
,
2237 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2238 #ifdef CONFIG_KVM_COMPAT
2239 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2241 .mmap
= kvm_vcpu_mmap
,
2242 .llseek
= noop_llseek
,
2246 * Allocates an inode for the vcpu.
2248 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2250 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2254 * Creates some virtual cpus. Good luck creating more than one.
2256 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2259 struct kvm_vcpu
*vcpu
;
2261 if (id
>= KVM_MAX_VCPUS
)
2264 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2266 return PTR_ERR(vcpu
);
2268 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2270 r
= kvm_arch_vcpu_setup(vcpu
);
2274 mutex_lock(&kvm
->lock
);
2275 if (!kvm_vcpu_compatible(vcpu
)) {
2277 goto unlock_vcpu_destroy
;
2279 if (atomic_read(&kvm
->online_vcpus
) == KVM_MAX_VCPUS
) {
2281 goto unlock_vcpu_destroy
;
2283 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2285 goto unlock_vcpu_destroy
;
2288 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2290 /* Now it's all set up, let userspace reach it */
2292 r
= create_vcpu_fd(vcpu
);
2295 goto unlock_vcpu_destroy
;
2298 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2301 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2302 * before kvm->online_vcpu's incremented value.
2305 atomic_inc(&kvm
->online_vcpus
);
2307 mutex_unlock(&kvm
->lock
);
2308 kvm_arch_vcpu_postcreate(vcpu
);
2311 unlock_vcpu_destroy
:
2312 mutex_unlock(&kvm
->lock
);
2314 kvm_arch_vcpu_destroy(vcpu
);
2318 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2321 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2322 vcpu
->sigset_active
= 1;
2323 vcpu
->sigset
= *sigset
;
2325 vcpu
->sigset_active
= 0;
2329 static long kvm_vcpu_ioctl(struct file
*filp
,
2330 unsigned int ioctl
, unsigned long arg
)
2332 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2333 void __user
*argp
= (void __user
*)arg
;
2335 struct kvm_fpu
*fpu
= NULL
;
2336 struct kvm_sregs
*kvm_sregs
= NULL
;
2338 if (vcpu
->kvm
->mm
!= current
->mm
)
2341 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2344 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2346 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2347 * so vcpu_load() would break it.
2349 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2350 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2354 r
= vcpu_load(vcpu
);
2362 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2363 /* The thread running this VCPU changed. */
2364 struct pid
*oldpid
= vcpu
->pid
;
2365 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2367 rcu_assign_pointer(vcpu
->pid
, newpid
);
2372 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2373 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2375 case KVM_GET_REGS
: {
2376 struct kvm_regs
*kvm_regs
;
2379 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2382 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2386 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2393 case KVM_SET_REGS
: {
2394 struct kvm_regs
*kvm_regs
;
2397 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2398 if (IS_ERR(kvm_regs
)) {
2399 r
= PTR_ERR(kvm_regs
);
2402 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2406 case KVM_GET_SREGS
: {
2407 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2411 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2415 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2420 case KVM_SET_SREGS
: {
2421 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2422 if (IS_ERR(kvm_sregs
)) {
2423 r
= PTR_ERR(kvm_sregs
);
2427 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2430 case KVM_GET_MP_STATE
: {
2431 struct kvm_mp_state mp_state
;
2433 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2437 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2442 case KVM_SET_MP_STATE
: {
2443 struct kvm_mp_state mp_state
;
2446 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2448 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2451 case KVM_TRANSLATE
: {
2452 struct kvm_translation tr
;
2455 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2457 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2461 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2466 case KVM_SET_GUEST_DEBUG
: {
2467 struct kvm_guest_debug dbg
;
2470 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2472 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2475 case KVM_SET_SIGNAL_MASK
: {
2476 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2477 struct kvm_signal_mask kvm_sigmask
;
2478 sigset_t sigset
, *p
;
2483 if (copy_from_user(&kvm_sigmask
, argp
,
2484 sizeof(kvm_sigmask
)))
2487 if (kvm_sigmask
.len
!= sizeof(sigset
))
2490 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2495 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2499 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2503 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2507 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2513 fpu
= memdup_user(argp
, sizeof(*fpu
));
2519 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2523 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2532 #ifdef CONFIG_KVM_COMPAT
2533 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2534 unsigned int ioctl
, unsigned long arg
)
2536 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2537 void __user
*argp
= compat_ptr(arg
);
2540 if (vcpu
->kvm
->mm
!= current
->mm
)
2544 case KVM_SET_SIGNAL_MASK
: {
2545 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2546 struct kvm_signal_mask kvm_sigmask
;
2547 compat_sigset_t csigset
;
2552 if (copy_from_user(&kvm_sigmask
, argp
,
2553 sizeof(kvm_sigmask
)))
2556 if (kvm_sigmask
.len
!= sizeof(csigset
))
2559 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2562 sigset_from_compat(&sigset
, &csigset
);
2563 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2565 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2569 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2577 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2578 int (*accessor
)(struct kvm_device
*dev
,
2579 struct kvm_device_attr
*attr
),
2582 struct kvm_device_attr attr
;
2587 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2590 return accessor(dev
, &attr
);
2593 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2596 struct kvm_device
*dev
= filp
->private_data
;
2599 case KVM_SET_DEVICE_ATTR
:
2600 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2601 case KVM_GET_DEVICE_ATTR
:
2602 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2603 case KVM_HAS_DEVICE_ATTR
:
2604 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2606 if (dev
->ops
->ioctl
)
2607 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2613 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2615 struct kvm_device
*dev
= filp
->private_data
;
2616 struct kvm
*kvm
= dev
->kvm
;
2622 static const struct file_operations kvm_device_fops
= {
2623 .unlocked_ioctl
= kvm_device_ioctl
,
2624 #ifdef CONFIG_KVM_COMPAT
2625 .compat_ioctl
= kvm_device_ioctl
,
2627 .release
= kvm_device_release
,
2630 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2632 if (filp
->f_op
!= &kvm_device_fops
)
2635 return filp
->private_data
;
2638 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2639 #ifdef CONFIG_KVM_MPIC
2640 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2641 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2644 #ifdef CONFIG_KVM_XICS
2645 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2649 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2651 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2654 if (kvm_device_ops_table
[type
] != NULL
)
2657 kvm_device_ops_table
[type
] = ops
;
2661 void kvm_unregister_device_ops(u32 type
)
2663 if (kvm_device_ops_table
[type
] != NULL
)
2664 kvm_device_ops_table
[type
] = NULL
;
2667 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2668 struct kvm_create_device
*cd
)
2670 struct kvm_device_ops
*ops
= NULL
;
2671 struct kvm_device
*dev
;
2672 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2675 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2678 ops
= kvm_device_ops_table
[cd
->type
];
2685 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2692 ret
= ops
->create(dev
, cd
->type
);
2698 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2704 list_add(&dev
->vm_node
, &kvm
->devices
);
2710 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2713 case KVM_CAP_USER_MEMORY
:
2714 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2715 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2716 case KVM_CAP_INTERNAL_ERROR_DATA
:
2717 #ifdef CONFIG_HAVE_KVM_MSI
2718 case KVM_CAP_SIGNAL_MSI
:
2720 #ifdef CONFIG_HAVE_KVM_IRQFD
2722 case KVM_CAP_IRQFD_RESAMPLE
:
2724 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2725 case KVM_CAP_CHECK_EXTENSION_VM
:
2727 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2728 case KVM_CAP_IRQ_ROUTING
:
2729 return KVM_MAX_IRQ_ROUTES
;
2731 #if KVM_ADDRESS_SPACE_NUM > 1
2732 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2733 return KVM_ADDRESS_SPACE_NUM
;
2738 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2741 static long kvm_vm_ioctl(struct file
*filp
,
2742 unsigned int ioctl
, unsigned long arg
)
2744 struct kvm
*kvm
= filp
->private_data
;
2745 void __user
*argp
= (void __user
*)arg
;
2748 if (kvm
->mm
!= current
->mm
)
2751 case KVM_CREATE_VCPU
:
2752 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2754 case KVM_SET_USER_MEMORY_REGION
: {
2755 struct kvm_userspace_memory_region kvm_userspace_mem
;
2758 if (copy_from_user(&kvm_userspace_mem
, argp
,
2759 sizeof(kvm_userspace_mem
)))
2762 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2765 case KVM_GET_DIRTY_LOG
: {
2766 struct kvm_dirty_log log
;
2769 if (copy_from_user(&log
, argp
, sizeof(log
)))
2771 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2774 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2775 case KVM_REGISTER_COALESCED_MMIO
: {
2776 struct kvm_coalesced_mmio_zone zone
;
2779 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2781 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2784 case KVM_UNREGISTER_COALESCED_MMIO
: {
2785 struct kvm_coalesced_mmio_zone zone
;
2788 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2790 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2795 struct kvm_irqfd data
;
2798 if (copy_from_user(&data
, argp
, sizeof(data
)))
2800 r
= kvm_irqfd(kvm
, &data
);
2803 case KVM_IOEVENTFD
: {
2804 struct kvm_ioeventfd data
;
2807 if (copy_from_user(&data
, argp
, sizeof(data
)))
2809 r
= kvm_ioeventfd(kvm
, &data
);
2812 #ifdef CONFIG_HAVE_KVM_MSI
2813 case KVM_SIGNAL_MSI
: {
2817 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
2819 r
= kvm_send_userspace_msi(kvm
, &msi
);
2823 #ifdef __KVM_HAVE_IRQ_LINE
2824 case KVM_IRQ_LINE_STATUS
:
2825 case KVM_IRQ_LINE
: {
2826 struct kvm_irq_level irq_event
;
2829 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
2832 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2833 ioctl
== KVM_IRQ_LINE_STATUS
);
2838 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2839 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
2847 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2848 case KVM_SET_GSI_ROUTING
: {
2849 struct kvm_irq_routing routing
;
2850 struct kvm_irq_routing __user
*urouting
;
2851 struct kvm_irq_routing_entry
*entries
;
2854 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
2857 if (routing
.nr
>= KVM_MAX_IRQ_ROUTES
)
2862 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
2867 if (copy_from_user(entries
, urouting
->entries
,
2868 routing
.nr
* sizeof(*entries
)))
2869 goto out_free_irq_routing
;
2870 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
2872 out_free_irq_routing
:
2876 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2877 case KVM_CREATE_DEVICE
: {
2878 struct kvm_create_device cd
;
2881 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
2884 r
= kvm_ioctl_create_device(kvm
, &cd
);
2889 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
2895 case KVM_CHECK_EXTENSION
:
2896 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
2899 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
2905 #ifdef CONFIG_KVM_COMPAT
2906 struct compat_kvm_dirty_log
{
2910 compat_uptr_t dirty_bitmap
; /* one bit per page */
2915 static long kvm_vm_compat_ioctl(struct file
*filp
,
2916 unsigned int ioctl
, unsigned long arg
)
2918 struct kvm
*kvm
= filp
->private_data
;
2921 if (kvm
->mm
!= current
->mm
)
2924 case KVM_GET_DIRTY_LOG
: {
2925 struct compat_kvm_dirty_log compat_log
;
2926 struct kvm_dirty_log log
;
2929 if (copy_from_user(&compat_log
, (void __user
*)arg
,
2930 sizeof(compat_log
)))
2932 log
.slot
= compat_log
.slot
;
2933 log
.padding1
= compat_log
.padding1
;
2934 log
.padding2
= compat_log
.padding2
;
2935 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
2937 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2941 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
2949 static struct file_operations kvm_vm_fops
= {
2950 .release
= kvm_vm_release
,
2951 .unlocked_ioctl
= kvm_vm_ioctl
,
2952 #ifdef CONFIG_KVM_COMPAT
2953 .compat_ioctl
= kvm_vm_compat_ioctl
,
2955 .llseek
= noop_llseek
,
2958 static int kvm_dev_ioctl_create_vm(unsigned long type
)
2963 kvm
= kvm_create_vm(type
);
2965 return PTR_ERR(kvm
);
2966 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2967 r
= kvm_coalesced_mmio_init(kvm
);
2973 r
= anon_inode_getfd("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
| O_CLOEXEC
);
2980 static long kvm_dev_ioctl(struct file
*filp
,
2981 unsigned int ioctl
, unsigned long arg
)
2986 case KVM_GET_API_VERSION
:
2989 r
= KVM_API_VERSION
;
2992 r
= kvm_dev_ioctl_create_vm(arg
);
2994 case KVM_CHECK_EXTENSION
:
2995 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
2997 case KVM_GET_VCPU_MMAP_SIZE
:
3000 r
= PAGE_SIZE
; /* struct kvm_run */
3002 r
+= PAGE_SIZE
; /* pio data page */
3004 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3005 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3008 case KVM_TRACE_ENABLE
:
3009 case KVM_TRACE_PAUSE
:
3010 case KVM_TRACE_DISABLE
:
3014 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3020 static struct file_operations kvm_chardev_ops
= {
3021 .unlocked_ioctl
= kvm_dev_ioctl
,
3022 .compat_ioctl
= kvm_dev_ioctl
,
3023 .llseek
= noop_llseek
,
3026 static struct miscdevice kvm_dev
= {
3032 static void hardware_enable_nolock(void *junk
)
3034 int cpu
= raw_smp_processor_id();
3037 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3040 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3042 r
= kvm_arch_hardware_enable();
3045 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3046 atomic_inc(&hardware_enable_failed
);
3047 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3051 static void hardware_enable(void)
3053 raw_spin_lock(&kvm_count_lock
);
3054 if (kvm_usage_count
)
3055 hardware_enable_nolock(NULL
);
3056 raw_spin_unlock(&kvm_count_lock
);
3059 static void hardware_disable_nolock(void *junk
)
3061 int cpu
= raw_smp_processor_id();
3063 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3065 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3066 kvm_arch_hardware_disable();
3069 static void hardware_disable(void)
3071 raw_spin_lock(&kvm_count_lock
);
3072 if (kvm_usage_count
)
3073 hardware_disable_nolock(NULL
);
3074 raw_spin_unlock(&kvm_count_lock
);
3077 static void hardware_disable_all_nolock(void)
3079 BUG_ON(!kvm_usage_count
);
3082 if (!kvm_usage_count
)
3083 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3086 static void hardware_disable_all(void)
3088 raw_spin_lock(&kvm_count_lock
);
3089 hardware_disable_all_nolock();
3090 raw_spin_unlock(&kvm_count_lock
);
3093 static int hardware_enable_all(void)
3097 raw_spin_lock(&kvm_count_lock
);
3100 if (kvm_usage_count
== 1) {
3101 atomic_set(&hardware_enable_failed
, 0);
3102 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3104 if (atomic_read(&hardware_enable_failed
)) {
3105 hardware_disable_all_nolock();
3110 raw_spin_unlock(&kvm_count_lock
);
3115 static int kvm_cpu_hotplug(struct notifier_block
*notifier
, unsigned long val
,
3118 val
&= ~CPU_TASKS_FROZEN
;
3130 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3134 * Some (well, at least mine) BIOSes hang on reboot if
3137 * And Intel TXT required VMX off for all cpu when system shutdown.
3139 pr_info("kvm: exiting hardware virtualization\n");
3140 kvm_rebooting
= true;
3141 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3145 static struct notifier_block kvm_reboot_notifier
= {
3146 .notifier_call
= kvm_reboot
,
3150 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3154 for (i
= 0; i
< bus
->dev_count
; i
++) {
3155 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3157 kvm_iodevice_destructor(pos
);
3162 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3163 const struct kvm_io_range
*r2
)
3165 gpa_t addr1
= r1
->addr
;
3166 gpa_t addr2
= r2
->addr
;
3171 /* If r2->len == 0, match the exact address. If r2->len != 0,
3172 * accept any overlapping write. Any order is acceptable for
3173 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3174 * we process all of them.
3187 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3189 return kvm_io_bus_cmp(p1
, p2
);
3192 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3193 gpa_t addr
, int len
)
3195 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3201 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3202 kvm_io_bus_sort_cmp
, NULL
);
3207 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3208 gpa_t addr
, int len
)
3210 struct kvm_io_range
*range
, key
;
3213 key
= (struct kvm_io_range
) {
3218 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3219 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3223 off
= range
- bus
->range
;
3225 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3231 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3232 struct kvm_io_range
*range
, const void *val
)
3236 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3240 while (idx
< bus
->dev_count
&&
3241 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3242 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3251 /* kvm_io_bus_write - called under kvm->slots_lock */
3252 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3253 int len
, const void *val
)
3255 struct kvm_io_bus
*bus
;
3256 struct kvm_io_range range
;
3259 range
= (struct kvm_io_range
) {
3264 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3265 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3266 return r
< 0 ? r
: 0;
3269 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3270 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3271 gpa_t addr
, int len
, const void *val
, long cookie
)
3273 struct kvm_io_bus
*bus
;
3274 struct kvm_io_range range
;
3276 range
= (struct kvm_io_range
) {
3281 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3283 /* First try the device referenced by cookie. */
3284 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3285 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3286 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3291 * cookie contained garbage; fall back to search and return the
3292 * correct cookie value.
3294 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3297 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3298 struct kvm_io_range
*range
, void *val
)
3302 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3306 while (idx
< bus
->dev_count
&&
3307 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3308 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3316 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3318 /* kvm_io_bus_read - called under kvm->slots_lock */
3319 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3322 struct kvm_io_bus
*bus
;
3323 struct kvm_io_range range
;
3326 range
= (struct kvm_io_range
) {
3331 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3332 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3333 return r
< 0 ? r
: 0;
3337 /* Caller must hold slots_lock. */
3338 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3339 int len
, struct kvm_io_device
*dev
)
3341 struct kvm_io_bus
*new_bus
, *bus
;
3343 bus
= kvm
->buses
[bus_idx
];
3344 /* exclude ioeventfd which is limited by maximum fd */
3345 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3348 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3349 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3352 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3353 sizeof(struct kvm_io_range
)));
3354 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3355 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3356 synchronize_srcu_expedited(&kvm
->srcu
);
3362 /* Caller must hold slots_lock. */
3363 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3364 struct kvm_io_device
*dev
)
3367 struct kvm_io_bus
*new_bus
, *bus
;
3369 bus
= kvm
->buses
[bus_idx
];
3371 for (i
= 0; i
< bus
->dev_count
; i
++)
3372 if (bus
->range
[i
].dev
== dev
) {
3380 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3381 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3385 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3386 new_bus
->dev_count
--;
3387 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3388 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3390 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3391 synchronize_srcu_expedited(&kvm
->srcu
);
3396 static struct notifier_block kvm_cpu_notifier
= {
3397 .notifier_call
= kvm_cpu_hotplug
,
3400 static int vm_stat_get(void *_offset
, u64
*val
)
3402 unsigned offset
= (long)_offset
;
3406 spin_lock(&kvm_lock
);
3407 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3408 *val
+= *(u32
*)((void *)kvm
+ offset
);
3409 spin_unlock(&kvm_lock
);
3413 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3415 static int vcpu_stat_get(void *_offset
, u64
*val
)
3417 unsigned offset
= (long)_offset
;
3419 struct kvm_vcpu
*vcpu
;
3423 spin_lock(&kvm_lock
);
3424 list_for_each_entry(kvm
, &vm_list
, vm_list
)
3425 kvm_for_each_vcpu(i
, vcpu
, kvm
)
3426 *val
+= *(u32
*)((void *)vcpu
+ offset
);
3428 spin_unlock(&kvm_lock
);
3432 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3434 static const struct file_operations
*stat_fops
[] = {
3435 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3436 [KVM_STAT_VM
] = &vm_stat_fops
,
3439 static int kvm_init_debug(void)
3442 struct kvm_stats_debugfs_item
*p
;
3444 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3445 if (kvm_debugfs_dir
== NULL
)
3448 for (p
= debugfs_entries
; p
->name
; ++p
) {
3449 if (!debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3450 (void *)(long)p
->offset
,
3451 stat_fops
[p
->kind
]))
3458 debugfs_remove_recursive(kvm_debugfs_dir
);
3463 static int kvm_suspend(void)
3465 if (kvm_usage_count
)
3466 hardware_disable_nolock(NULL
);
3470 static void kvm_resume(void)
3472 if (kvm_usage_count
) {
3473 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3474 hardware_enable_nolock(NULL
);
3478 static struct syscore_ops kvm_syscore_ops
= {
3479 .suspend
= kvm_suspend
,
3480 .resume
= kvm_resume
,
3484 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3486 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3489 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3491 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3493 if (vcpu
->preempted
)
3494 vcpu
->preempted
= false;
3496 kvm_arch_sched_in(vcpu
, cpu
);
3498 kvm_arch_vcpu_load(vcpu
, cpu
);
3501 static void kvm_sched_out(struct preempt_notifier
*pn
,
3502 struct task_struct
*next
)
3504 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3506 if (current
->state
== TASK_RUNNING
)
3507 vcpu
->preempted
= true;
3508 kvm_arch_vcpu_put(vcpu
);
3511 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3512 struct module
*module
)
3517 r
= kvm_arch_init(opaque
);
3522 * kvm_arch_init makes sure there's at most one caller
3523 * for architectures that support multiple implementations,
3524 * like intel and amd on x86.
3525 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3526 * conflicts in case kvm is already setup for another implementation.
3528 r
= kvm_irqfd_init();
3532 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3537 r
= kvm_arch_hardware_setup();
3541 for_each_online_cpu(cpu
) {
3542 smp_call_function_single(cpu
,
3543 kvm_arch_check_processor_compat
,
3549 r
= register_cpu_notifier(&kvm_cpu_notifier
);
3552 register_reboot_notifier(&kvm_reboot_notifier
);
3554 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3556 vcpu_align
= __alignof__(struct kvm_vcpu
);
3557 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3559 if (!kvm_vcpu_cache
) {
3564 r
= kvm_async_pf_init();
3568 kvm_chardev_ops
.owner
= module
;
3569 kvm_vm_fops
.owner
= module
;
3570 kvm_vcpu_fops
.owner
= module
;
3572 r
= misc_register(&kvm_dev
);
3574 pr_err("kvm: misc device register failed\n");
3578 register_syscore_ops(&kvm_syscore_ops
);
3580 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3581 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3583 r
= kvm_init_debug();
3585 pr_err("kvm: create debugfs files failed\n");
3589 r
= kvm_vfio_ops_init();
3595 unregister_syscore_ops(&kvm_syscore_ops
);
3596 misc_deregister(&kvm_dev
);
3598 kvm_async_pf_deinit();
3600 kmem_cache_destroy(kvm_vcpu_cache
);
3602 unregister_reboot_notifier(&kvm_reboot_notifier
);
3603 unregister_cpu_notifier(&kvm_cpu_notifier
);
3606 kvm_arch_hardware_unsetup();
3608 free_cpumask_var(cpus_hardware_enabled
);
3616 EXPORT_SYMBOL_GPL(kvm_init
);
3620 debugfs_remove_recursive(kvm_debugfs_dir
);
3621 misc_deregister(&kvm_dev
);
3622 kmem_cache_destroy(kvm_vcpu_cache
);
3623 kvm_async_pf_deinit();
3624 unregister_syscore_ops(&kvm_syscore_ops
);
3625 unregister_reboot_notifier(&kvm_reboot_notifier
);
3626 unregister_cpu_notifier(&kvm_cpu_notifier
);
3627 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3628 kvm_arch_hardware_unsetup();
3631 free_cpumask_var(cpus_hardware_enabled
);
3632 kvm_vfio_ops_exit();
3634 EXPORT_SYMBOL_GPL(kvm_exit
);