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>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 #include <linux/kvm_dirty_ring.h>
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
99 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
102 DEFINE_MUTEX(kvm_lock);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
106 static cpumask_var_t cpus_hardware_enabled;
107 static int kvm_usage_count;
108 static atomic_t hardware_enable_failed;
110 static struct kmem_cache *kvm_vcpu_cache;
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
115 struct dentry *kvm_debugfs_dir;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
118 static const struct file_operations stat_fops_per_vm;
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
159 unsigned long start, unsigned long end)
163 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
174 return is_zone_device_page(pfn_to_page(pfn));
177 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
180 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
181 * perspective they are "normal" pages, albeit with slightly different
185 return PageReserved(pfn_to_page(pfn)) &&
187 !kvm_is_zone_device_pfn(pfn);
192 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
194 struct page *page = pfn_to_page(pfn);
196 if (!PageTransCompoundMap(page))
199 return is_transparent_hugepage(compound_head(page));
203 * Switches to specified vcpu, until a matching vcpu_put()
205 void vcpu_load(struct kvm_vcpu *vcpu)
209 __this_cpu_write(kvm_running_vcpu, vcpu);
210 preempt_notifier_register(&vcpu->preempt_notifier);
211 kvm_arch_vcpu_load(vcpu, cpu);
214 EXPORT_SYMBOL_GPL(vcpu_load);
216 void vcpu_put(struct kvm_vcpu *vcpu)
219 kvm_arch_vcpu_put(vcpu);
220 preempt_notifier_unregister(&vcpu->preempt_notifier);
221 __this_cpu_write(kvm_running_vcpu, NULL);
224 EXPORT_SYMBOL_GPL(vcpu_put);
226 /* TODO: merge with kvm_arch_vcpu_should_kick */
227 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
229 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
232 * We need to wait for the VCPU to reenable interrupts and get out of
233 * READING_SHADOW_PAGE_TABLES mode.
235 if (req & KVM_REQUEST_WAIT)
236 return mode != OUTSIDE_GUEST_MODE;
239 * Need to kick a running VCPU, but otherwise there is nothing to do.
241 return mode == IN_GUEST_MODE;
244 static void ack_flush(void *_completed)
248 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
251 cpus = cpu_online_mask;
253 if (cpumask_empty(cpus))
256 smp_call_function_many(cpus, ack_flush, NULL, wait);
260 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
261 struct kvm_vcpu *except,
262 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
265 struct kvm_vcpu *vcpu;
270 kvm_for_each_vcpu(i, vcpu, kvm) {
271 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
275 kvm_make_request(req, vcpu);
278 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 if (tmp != NULL && cpu != -1 && cpu != me &&
282 kvm_request_needs_ipi(vcpu, req))
283 __cpumask_set_cpu(cpu, tmp);
286 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
292 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
293 struct kvm_vcpu *except)
298 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
300 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
302 free_cpumask_var(cpus);
306 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
308 return kvm_make_all_cpus_request_except(kvm, req, NULL);
310 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
312 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
313 void kvm_flush_remote_tlbs(struct kvm *kvm)
316 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
317 * kvm_make_all_cpus_request.
319 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
321 ++kvm->stat.generic.remote_tlb_flush_requests;
323 * We want to publish modifications to the page tables before reading
324 * mode. Pairs with a memory barrier in arch-specific code.
325 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
326 * and smp_mb in walk_shadow_page_lockless_begin/end.
327 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
329 * There is already an smp_mb__after_atomic() before
330 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
333 if (!kvm_arch_flush_remote_tlb(kvm)
334 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
335 ++kvm->stat.generic.remote_tlb_flush;
336 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
338 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
341 void kvm_reload_remote_mmus(struct kvm *kvm)
343 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
346 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
347 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
350 gfp_flags |= mc->gfp_zero;
353 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
355 return (void *)__get_free_page(gfp_flags);
358 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
362 if (mc->nobjs >= min)
364 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
365 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
367 return mc->nobjs >= min ? 0 : -ENOMEM;
368 mc->objects[mc->nobjs++] = obj;
373 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
378 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
382 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
384 free_page((unsigned long)mc->objects[--mc->nobjs]);
388 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
392 if (WARN_ON(!mc->nobjs))
393 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
395 p = mc->objects[--mc->nobjs];
401 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
403 mutex_init(&vcpu->mutex);
408 rcuwait_init(&vcpu->wait);
409 kvm_async_pf_vcpu_init(vcpu);
412 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
414 kvm_vcpu_set_in_spin_loop(vcpu, false);
415 kvm_vcpu_set_dy_eligible(vcpu, false);
416 vcpu->preempted = false;
418 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
419 vcpu->last_used_slot = 0;
422 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
424 kvm_dirty_ring_free(&vcpu->dirty_ring);
425 kvm_arch_vcpu_destroy(vcpu);
428 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
429 * the vcpu->pid pointer, and at destruction time all file descriptors
432 put_pid(rcu_dereference_protected(vcpu->pid, 1));
434 free_page((unsigned long)vcpu->run);
435 kmem_cache_free(kvm_vcpu_cache, vcpu);
437 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
439 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
440 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
442 return container_of(mn, struct kvm, mmu_notifier);
445 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
446 struct mm_struct *mm,
447 unsigned long start, unsigned long end)
449 struct kvm *kvm = mmu_notifier_to_kvm(mn);
452 idx = srcu_read_lock(&kvm->srcu);
453 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
454 srcu_read_unlock(&kvm->srcu, idx);
457 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
459 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
462 struct kvm_hva_range {
466 hva_handler_t handler;
467 on_lock_fn_t on_lock;
473 * Use a dedicated stub instead of NULL to indicate that there is no callback
474 * function/handler. The compiler technically can't guarantee that a real
475 * function will have a non-zero address, and so it will generate code to
476 * check for !NULL, whereas comparing against a stub will be elided at compile
477 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
479 static void kvm_null_fn(void)
483 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
485 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
486 const struct kvm_hva_range *range)
488 bool ret = false, locked = false;
489 struct kvm_gfn_range gfn_range;
490 struct kvm_memory_slot *slot;
491 struct kvm_memslots *slots;
494 /* A null handler is allowed if and only if on_lock() is provided. */
495 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
496 IS_KVM_NULL_FN(range->handler)))
499 idx = srcu_read_lock(&kvm->srcu);
501 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
502 slots = __kvm_memslots(kvm, i);
503 kvm_for_each_memslot(slot, slots) {
504 unsigned long hva_start, hva_end;
506 hva_start = max(range->start, slot->userspace_addr);
507 hva_end = min(range->end, slot->userspace_addr +
508 (slot->npages << PAGE_SHIFT));
509 if (hva_start >= hva_end)
513 * To optimize for the likely case where the address
514 * range is covered by zero or one memslots, don't
515 * bother making these conditional (to avoid writes on
516 * the second or later invocation of the handler).
518 gfn_range.pte = range->pte;
519 gfn_range.may_block = range->may_block;
522 * {gfn(page) | page intersects with [hva_start, hva_end)} =
523 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
525 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
526 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
527 gfn_range.slot = slot;
532 if (!IS_KVM_NULL_FN(range->on_lock))
533 range->on_lock(kvm, range->start, range->end);
534 if (IS_KVM_NULL_FN(range->handler))
537 ret |= range->handler(kvm, &gfn_range);
541 if (range->flush_on_ret && (ret || kvm->tlbs_dirty))
542 kvm_flush_remote_tlbs(kvm);
547 srcu_read_unlock(&kvm->srcu, idx);
549 /* The notifiers are averse to booleans. :-( */
553 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
557 hva_handler_t handler)
559 struct kvm *kvm = mmu_notifier_to_kvm(mn);
560 const struct kvm_hva_range range = {
565 .on_lock = (void *)kvm_null_fn,
566 .flush_on_ret = true,
570 return __kvm_handle_hva_range(kvm, &range);
573 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
576 hva_handler_t handler)
578 struct kvm *kvm = mmu_notifier_to_kvm(mn);
579 const struct kvm_hva_range range = {
584 .on_lock = (void *)kvm_null_fn,
585 .flush_on_ret = false,
589 return __kvm_handle_hva_range(kvm, &range);
591 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
592 struct mm_struct *mm,
593 unsigned long address,
596 struct kvm *kvm = mmu_notifier_to_kvm(mn);
598 trace_kvm_set_spte_hva(address);
601 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
602 * If mmu_notifier_count is zero, then no in-progress invalidations,
603 * including this one, found a relevant memslot at start(); rechecking
604 * memslots here is unnecessary. Note, a false positive (count elevated
605 * by a different invalidation) is sub-optimal but functionally ok.
607 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
608 if (!READ_ONCE(kvm->mmu_notifier_count))
611 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
614 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
618 * The count increase must become visible at unlock time as no
619 * spte can be established without taking the mmu_lock and
620 * count is also read inside the mmu_lock critical section.
622 kvm->mmu_notifier_count++;
623 if (likely(kvm->mmu_notifier_count == 1)) {
624 kvm->mmu_notifier_range_start = start;
625 kvm->mmu_notifier_range_end = end;
628 * Fully tracking multiple concurrent ranges has dimishing
629 * returns. Keep things simple and just find the minimal range
630 * which includes the current and new ranges. As there won't be
631 * enough information to subtract a range after its invalidate
632 * completes, any ranges invalidated concurrently will
633 * accumulate and persist until all outstanding invalidates
636 kvm->mmu_notifier_range_start =
637 min(kvm->mmu_notifier_range_start, start);
638 kvm->mmu_notifier_range_end =
639 max(kvm->mmu_notifier_range_end, end);
643 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
644 const struct mmu_notifier_range *range)
646 struct kvm *kvm = mmu_notifier_to_kvm(mn);
647 const struct kvm_hva_range hva_range = {
648 .start = range->start,
651 .handler = kvm_unmap_gfn_range,
652 .on_lock = kvm_inc_notifier_count,
653 .flush_on_ret = true,
654 .may_block = mmu_notifier_range_blockable(range),
657 trace_kvm_unmap_hva_range(range->start, range->end);
660 * Prevent memslot modification between range_start() and range_end()
661 * so that conditionally locking provides the same result in both
662 * functions. Without that guarantee, the mmu_notifier_count
663 * adjustments will be imbalanced.
665 * Pairs with the decrement in range_end().
667 spin_lock(&kvm->mn_invalidate_lock);
668 kvm->mn_active_invalidate_count++;
669 spin_unlock(&kvm->mn_invalidate_lock);
671 __kvm_handle_hva_range(kvm, &hva_range);
676 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
680 * This sequence increase will notify the kvm page fault that
681 * the page that is going to be mapped in the spte could have
684 kvm->mmu_notifier_seq++;
687 * The above sequence increase must be visible before the
688 * below count decrease, which is ensured by the smp_wmb above
689 * in conjunction with the smp_rmb in mmu_notifier_retry().
691 kvm->mmu_notifier_count--;
694 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
695 const struct mmu_notifier_range *range)
697 struct kvm *kvm = mmu_notifier_to_kvm(mn);
698 const struct kvm_hva_range hva_range = {
699 .start = range->start,
702 .handler = (void *)kvm_null_fn,
703 .on_lock = kvm_dec_notifier_count,
704 .flush_on_ret = false,
705 .may_block = mmu_notifier_range_blockable(range),
709 __kvm_handle_hva_range(kvm, &hva_range);
711 /* Pairs with the increment in range_start(). */
712 spin_lock(&kvm->mn_invalidate_lock);
713 wake = (--kvm->mn_active_invalidate_count == 0);
714 spin_unlock(&kvm->mn_invalidate_lock);
717 * There can only be one waiter, since the wait happens under
721 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
723 BUG_ON(kvm->mmu_notifier_count < 0);
726 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
727 struct mm_struct *mm,
731 trace_kvm_age_hva(start, end);
733 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
736 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
737 struct mm_struct *mm,
741 trace_kvm_age_hva(start, end);
744 * Even though we do not flush TLB, this will still adversely
745 * affect performance on pre-Haswell Intel EPT, where there is
746 * no EPT Access Bit to clear so that we have to tear down EPT
747 * tables instead. If we find this unacceptable, we can always
748 * add a parameter to kvm_age_hva so that it effectively doesn't
749 * do anything on clear_young.
751 * Also note that currently we never issue secondary TLB flushes
752 * from clear_young, leaving this job up to the regular system
753 * cadence. If we find this inaccurate, we might come up with a
754 * more sophisticated heuristic later.
756 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
759 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
760 struct mm_struct *mm,
761 unsigned long address)
763 trace_kvm_test_age_hva(address);
765 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
769 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
770 struct mm_struct *mm)
772 struct kvm *kvm = mmu_notifier_to_kvm(mn);
775 idx = srcu_read_lock(&kvm->srcu);
776 kvm_arch_flush_shadow_all(kvm);
777 srcu_read_unlock(&kvm->srcu, idx);
780 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
781 .invalidate_range = kvm_mmu_notifier_invalidate_range,
782 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
783 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
784 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
785 .clear_young = kvm_mmu_notifier_clear_young,
786 .test_young = kvm_mmu_notifier_test_young,
787 .change_pte = kvm_mmu_notifier_change_pte,
788 .release = kvm_mmu_notifier_release,
791 static int kvm_init_mmu_notifier(struct kvm *kvm)
793 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
794 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
797 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
799 static int kvm_init_mmu_notifier(struct kvm *kvm)
804 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
806 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
807 static int kvm_pm_notifier_call(struct notifier_block *bl,
811 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
813 return kvm_arch_pm_notifier(kvm, state);
816 static void kvm_init_pm_notifier(struct kvm *kvm)
818 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
819 /* Suspend KVM before we suspend ftrace, RCU, etc. */
820 kvm->pm_notifier.priority = INT_MAX;
821 register_pm_notifier(&kvm->pm_notifier);
824 static void kvm_destroy_pm_notifier(struct kvm *kvm)
826 unregister_pm_notifier(&kvm->pm_notifier);
828 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
829 static void kvm_init_pm_notifier(struct kvm *kvm)
833 static void kvm_destroy_pm_notifier(struct kvm *kvm)
836 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
838 static struct kvm_memslots *kvm_alloc_memslots(void)
841 struct kvm_memslots *slots;
843 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
847 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
848 slots->id_to_index[i] = -1;
853 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
855 if (!memslot->dirty_bitmap)
858 kvfree(memslot->dirty_bitmap);
859 memslot->dirty_bitmap = NULL;
862 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
864 kvm_destroy_dirty_bitmap(slot);
866 kvm_arch_free_memslot(kvm, slot);
872 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
874 struct kvm_memory_slot *memslot;
879 kvm_for_each_memslot(memslot, slots)
880 kvm_free_memslot(kvm, memslot);
885 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
887 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
888 case KVM_STATS_TYPE_INSTANT:
890 case KVM_STATS_TYPE_CUMULATIVE:
891 case KVM_STATS_TYPE_PEAK:
898 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
901 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
902 kvm_vcpu_stats_header.num_desc;
904 if (!kvm->debugfs_dentry)
907 debugfs_remove_recursive(kvm->debugfs_dentry);
909 if (kvm->debugfs_stat_data) {
910 for (i = 0; i < kvm_debugfs_num_entries; i++)
911 kfree(kvm->debugfs_stat_data[i]);
912 kfree(kvm->debugfs_stat_data);
916 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
918 static DEFINE_MUTEX(kvm_debugfs_lock);
920 char dir_name[ITOA_MAX_LEN * 2];
921 struct kvm_stat_data *stat_data;
922 const struct _kvm_stats_desc *pdesc;
924 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
925 kvm_vcpu_stats_header.num_desc;
927 if (!debugfs_initialized())
930 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
931 mutex_lock(&kvm_debugfs_lock);
932 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
934 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
936 mutex_unlock(&kvm_debugfs_lock);
939 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
940 mutex_unlock(&kvm_debugfs_lock);
944 kvm->debugfs_dentry = dent;
945 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
946 sizeof(*kvm->debugfs_stat_data),
948 if (!kvm->debugfs_stat_data)
951 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
952 pdesc = &kvm_vm_stats_desc[i];
953 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
957 stat_data->kvm = kvm;
958 stat_data->desc = pdesc;
959 stat_data->kind = KVM_STAT_VM;
960 kvm->debugfs_stat_data[i] = stat_data;
961 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
962 kvm->debugfs_dentry, stat_data,
966 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
967 pdesc = &kvm_vcpu_stats_desc[i];
968 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
972 stat_data->kvm = kvm;
973 stat_data->desc = pdesc;
974 stat_data->kind = KVM_STAT_VCPU;
975 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
976 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
977 kvm->debugfs_dentry, stat_data,
981 ret = kvm_arch_create_vm_debugfs(kvm);
983 kvm_destroy_vm_debugfs(kvm);
991 * Called after the VM is otherwise initialized, but just before adding it to
994 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1000 * Called just after removing the VM from the vm_list, but before doing any
1001 * other destruction.
1003 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1008 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1009 * be setup already, so we can create arch-specific debugfs entries under it.
1010 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1011 * a per-arch destroy interface is not needed.
1013 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1018 static struct kvm *kvm_create_vm(unsigned long type)
1020 struct kvm *kvm = kvm_arch_alloc_vm();
1025 return ERR_PTR(-ENOMEM);
1027 KVM_MMU_LOCK_INIT(kvm);
1028 mmgrab(current->mm);
1029 kvm->mm = current->mm;
1030 kvm_eventfd_init(kvm);
1031 mutex_init(&kvm->lock);
1032 mutex_init(&kvm->irq_lock);
1033 mutex_init(&kvm->slots_lock);
1034 mutex_init(&kvm->slots_arch_lock);
1035 spin_lock_init(&kvm->mn_invalidate_lock);
1036 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1038 INIT_LIST_HEAD(&kvm->devices);
1040 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1042 if (init_srcu_struct(&kvm->srcu))
1043 goto out_err_no_srcu;
1044 if (init_srcu_struct(&kvm->irq_srcu))
1045 goto out_err_no_irq_srcu;
1047 refcount_set(&kvm->users_count, 1);
1048 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1049 struct kvm_memslots *slots = kvm_alloc_memslots();
1052 goto out_err_no_arch_destroy_vm;
1053 /* Generations must be different for each address space. */
1054 slots->generation = i;
1055 rcu_assign_pointer(kvm->memslots[i], slots);
1058 for (i = 0; i < KVM_NR_BUSES; i++) {
1059 rcu_assign_pointer(kvm->buses[i],
1060 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1062 goto out_err_no_arch_destroy_vm;
1065 kvm->max_halt_poll_ns = halt_poll_ns;
1067 r = kvm_arch_init_vm(kvm, type);
1069 goto out_err_no_arch_destroy_vm;
1071 r = hardware_enable_all();
1073 goto out_err_no_disable;
1075 #ifdef CONFIG_HAVE_KVM_IRQFD
1076 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1079 r = kvm_init_mmu_notifier(kvm);
1081 goto out_err_no_mmu_notifier;
1083 r = kvm_arch_post_init_vm(kvm);
1087 mutex_lock(&kvm_lock);
1088 list_add(&kvm->vm_list, &vm_list);
1089 mutex_unlock(&kvm_lock);
1091 preempt_notifier_inc();
1092 kvm_init_pm_notifier(kvm);
1097 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1098 if (kvm->mmu_notifier.ops)
1099 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1101 out_err_no_mmu_notifier:
1102 hardware_disable_all();
1104 kvm_arch_destroy_vm(kvm);
1105 out_err_no_arch_destroy_vm:
1106 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1107 for (i = 0; i < KVM_NR_BUSES; i++)
1108 kfree(kvm_get_bus(kvm, i));
1109 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1110 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1111 cleanup_srcu_struct(&kvm->irq_srcu);
1112 out_err_no_irq_srcu:
1113 cleanup_srcu_struct(&kvm->srcu);
1115 kvm_arch_free_vm(kvm);
1116 mmdrop(current->mm);
1120 static void kvm_destroy_devices(struct kvm *kvm)
1122 struct kvm_device *dev, *tmp;
1125 * We do not need to take the kvm->lock here, because nobody else
1126 * has a reference to the struct kvm at this point and therefore
1127 * cannot access the devices list anyhow.
1129 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1130 list_del(&dev->vm_node);
1131 dev->ops->destroy(dev);
1135 static void kvm_destroy_vm(struct kvm *kvm)
1138 struct mm_struct *mm = kvm->mm;
1140 kvm_destroy_pm_notifier(kvm);
1141 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1142 kvm_destroy_vm_debugfs(kvm);
1143 kvm_arch_sync_events(kvm);
1144 mutex_lock(&kvm_lock);
1145 list_del(&kvm->vm_list);
1146 mutex_unlock(&kvm_lock);
1147 kvm_arch_pre_destroy_vm(kvm);
1149 kvm_free_irq_routing(kvm);
1150 for (i = 0; i < KVM_NR_BUSES; i++) {
1151 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1154 kvm_io_bus_destroy(bus);
1155 kvm->buses[i] = NULL;
1157 kvm_coalesced_mmio_free(kvm);
1158 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1159 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1161 * At this point, pending calls to invalidate_range_start()
1162 * have completed but no more MMU notifiers will run, so
1163 * mn_active_invalidate_count may remain unbalanced.
1164 * No threads can be waiting in install_new_memslots as the
1165 * last reference on KVM has been dropped, but freeing
1166 * memslots would deadlock without this manual intervention.
1168 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1169 kvm->mn_active_invalidate_count = 0;
1171 kvm_arch_flush_shadow_all(kvm);
1173 kvm_arch_destroy_vm(kvm);
1174 kvm_destroy_devices(kvm);
1175 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1176 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1177 cleanup_srcu_struct(&kvm->irq_srcu);
1178 cleanup_srcu_struct(&kvm->srcu);
1179 kvm_arch_free_vm(kvm);
1180 preempt_notifier_dec();
1181 hardware_disable_all();
1185 void kvm_get_kvm(struct kvm *kvm)
1187 refcount_inc(&kvm->users_count);
1189 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1192 * Make sure the vm is not during destruction, which is a safe version of
1193 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1195 bool kvm_get_kvm_safe(struct kvm *kvm)
1197 return refcount_inc_not_zero(&kvm->users_count);
1199 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1201 void kvm_put_kvm(struct kvm *kvm)
1203 if (refcount_dec_and_test(&kvm->users_count))
1204 kvm_destroy_vm(kvm);
1206 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1209 * Used to put a reference that was taken on behalf of an object associated
1210 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1211 * of the new file descriptor fails and the reference cannot be transferred to
1212 * its final owner. In such cases, the caller is still actively using @kvm and
1213 * will fail miserably if the refcount unexpectedly hits zero.
1215 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1217 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1219 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1221 static int kvm_vm_release(struct inode *inode, struct file *filp)
1223 struct kvm *kvm = filp->private_data;
1225 kvm_irqfd_release(kvm);
1232 * Allocation size is twice as large as the actual dirty bitmap size.
1233 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1235 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1237 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1239 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1240 if (!memslot->dirty_bitmap)
1247 * Delete a memslot by decrementing the number of used slots and shifting all
1248 * other entries in the array forward one spot.
1250 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1251 struct kvm_memory_slot *memslot)
1253 struct kvm_memory_slot *mslots = slots->memslots;
1256 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1259 slots->used_slots--;
1261 if (atomic_read(&slots->last_used_slot) >= slots->used_slots)
1262 atomic_set(&slots->last_used_slot, 0);
1264 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1265 mslots[i] = mslots[i + 1];
1266 slots->id_to_index[mslots[i].id] = i;
1268 mslots[i] = *memslot;
1269 slots->id_to_index[memslot->id] = -1;
1273 * "Insert" a new memslot by incrementing the number of used slots. Returns
1274 * the new slot's initial index into the memslots array.
1276 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1278 return slots->used_slots++;
1282 * Move a changed memslot backwards in the array by shifting existing slots
1283 * with a higher GFN toward the front of the array. Note, the changed memslot
1284 * itself is not preserved in the array, i.e. not swapped at this time, only
1285 * its new index into the array is tracked. Returns the changed memslot's
1286 * current index into the memslots array.
1288 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1289 struct kvm_memory_slot *memslot)
1291 struct kvm_memory_slot *mslots = slots->memslots;
1294 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1295 WARN_ON_ONCE(!slots->used_slots))
1299 * Move the target memslot backward in the array by shifting existing
1300 * memslots with a higher GFN (than the target memslot) towards the
1301 * front of the array.
1303 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1304 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1307 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1309 /* Shift the next memslot forward one and update its index. */
1310 mslots[i] = mslots[i + 1];
1311 slots->id_to_index[mslots[i].id] = i;
1317 * Move a changed memslot forwards in the array by shifting existing slots with
1318 * a lower GFN toward the back of the array. Note, the changed memslot itself
1319 * is not preserved in the array, i.e. not swapped at this time, only its new
1320 * index into the array is tracked. Returns the changed memslot's final index
1321 * into the memslots array.
1323 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1324 struct kvm_memory_slot *memslot,
1327 struct kvm_memory_slot *mslots = slots->memslots;
1330 for (i = start; i > 0; i--) {
1331 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1334 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1336 /* Shift the next memslot back one and update its index. */
1337 mslots[i] = mslots[i - 1];
1338 slots->id_to_index[mslots[i].id] = i;
1344 * Re-sort memslots based on their GFN to account for an added, deleted, or
1345 * moved memslot. Sorting memslots by GFN allows using a binary search during
1348 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1349 * at memslots[0] has the highest GFN.
1351 * The sorting algorithm takes advantage of having initially sorted memslots
1352 * and knowing the position of the changed memslot. Sorting is also optimized
1353 * by not swapping the updated memslot and instead only shifting other memslots
1354 * and tracking the new index for the update memslot. Only once its final
1355 * index is known is the updated memslot copied into its position in the array.
1357 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1358 * the end of the array.
1360 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1361 * end of the array and then it forward to its correct location.
1363 * - When moving a memslot, the algorithm first moves the updated memslot
1364 * backward to handle the scenario where the memslot's GFN was changed to a
1365 * lower value. update_memslots() then falls through and runs the same flow
1366 * as creating a memslot to move the memslot forward to handle the scenario
1367 * where its GFN was changed to a higher value.
1369 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1370 * historical reasons. Originally, invalid memslots where denoted by having
1371 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1372 * to the end of the array. The current algorithm uses dedicated logic to
1373 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1375 * The other historical motiviation for highest->lowest was to improve the
1376 * performance of memslot lookup. KVM originally used a linear search starting
1377 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1378 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1379 * single memslot above the 4gb boundary. As the largest memslot is also the
1380 * most likely to be referenced, sorting it to the front of the array was
1381 * advantageous. The current binary search starts from the middle of the array
1382 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1384 static void update_memslots(struct kvm_memslots *slots,
1385 struct kvm_memory_slot *memslot,
1386 enum kvm_mr_change change)
1390 if (change == KVM_MR_DELETE) {
1391 kvm_memslot_delete(slots, memslot);
1393 if (change == KVM_MR_CREATE)
1394 i = kvm_memslot_insert_back(slots);
1396 i = kvm_memslot_move_backward(slots, memslot);
1397 i = kvm_memslot_move_forward(slots, memslot, i);
1400 * Copy the memslot to its new position in memslots and update
1401 * its index accordingly.
1403 slots->memslots[i] = *memslot;
1404 slots->id_to_index[memslot->id] = i;
1408 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1410 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1412 #ifdef __KVM_HAVE_READONLY_MEM
1413 valid_flags |= KVM_MEM_READONLY;
1416 if (mem->flags & ~valid_flags)
1422 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1423 int as_id, struct kvm_memslots *slots)
1425 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1426 u64 gen = old_memslots->generation;
1428 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1429 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1432 * Do not store the new memslots while there are invalidations in
1433 * progress, otherwise the locking in invalidate_range_start and
1434 * invalidate_range_end will be unbalanced.
1436 spin_lock(&kvm->mn_invalidate_lock);
1437 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1438 while (kvm->mn_active_invalidate_count) {
1439 set_current_state(TASK_UNINTERRUPTIBLE);
1440 spin_unlock(&kvm->mn_invalidate_lock);
1442 spin_lock(&kvm->mn_invalidate_lock);
1444 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1445 rcu_assign_pointer(kvm->memslots[as_id], slots);
1446 spin_unlock(&kvm->mn_invalidate_lock);
1449 * Acquired in kvm_set_memslot. Must be released before synchronize
1450 * SRCU below in order to avoid deadlock with another thread
1451 * acquiring the slots_arch_lock in an srcu critical section.
1453 mutex_unlock(&kvm->slots_arch_lock);
1455 synchronize_srcu_expedited(&kvm->srcu);
1458 * Increment the new memslot generation a second time, dropping the
1459 * update in-progress flag and incrementing the generation based on
1460 * the number of address spaces. This provides a unique and easily
1461 * identifiable generation number while the memslots are in flux.
1463 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1466 * Generations must be unique even across address spaces. We do not need
1467 * a global counter for that, instead the generation space is evenly split
1468 * across address spaces. For example, with two address spaces, address
1469 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1470 * use generations 1, 3, 5, ...
1472 gen += KVM_ADDRESS_SPACE_NUM;
1474 kvm_arch_memslots_updated(kvm, gen);
1476 slots->generation = gen;
1478 return old_memslots;
1481 static size_t kvm_memslots_size(int slots)
1483 return sizeof(struct kvm_memslots) +
1484 (sizeof(struct kvm_memory_slot) * slots);
1487 static void kvm_copy_memslots(struct kvm_memslots *to,
1488 struct kvm_memslots *from)
1490 memcpy(to, from, kvm_memslots_size(from->used_slots));
1494 * Note, at a minimum, the current number of used slots must be allocated, even
1495 * when deleting a memslot, as we need a complete duplicate of the memslots for
1496 * use when invalidating a memslot prior to deleting/moving the memslot.
1498 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1499 enum kvm_mr_change change)
1501 struct kvm_memslots *slots;
1504 if (change == KVM_MR_CREATE)
1505 new_size = kvm_memslots_size(old->used_slots + 1);
1507 new_size = kvm_memslots_size(old->used_slots);
1509 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1511 kvm_copy_memslots(slots, old);
1516 static int kvm_set_memslot(struct kvm *kvm,
1517 const struct kvm_userspace_memory_region *mem,
1518 struct kvm_memory_slot *old,
1519 struct kvm_memory_slot *new, int as_id,
1520 enum kvm_mr_change change)
1522 struct kvm_memory_slot *slot;
1523 struct kvm_memslots *slots;
1527 * Released in install_new_memslots.
1529 * Must be held from before the current memslots are copied until
1530 * after the new memslots are installed with rcu_assign_pointer,
1531 * then released before the synchronize srcu in install_new_memslots.
1533 * When modifying memslots outside of the slots_lock, must be held
1534 * before reading the pointer to the current memslots until after all
1535 * changes to those memslots are complete.
1537 * These rules ensure that installing new memslots does not lose
1538 * changes made to the previous memslots.
1540 mutex_lock(&kvm->slots_arch_lock);
1542 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1544 mutex_unlock(&kvm->slots_arch_lock);
1548 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1550 * Note, the INVALID flag needs to be in the appropriate entry
1551 * in the freshly allocated memslots, not in @old or @new.
1553 slot = id_to_memslot(slots, old->id);
1554 slot->flags |= KVM_MEMSLOT_INVALID;
1557 * We can re-use the memory from the old memslots.
1558 * It will be overwritten with a copy of the new memslots
1559 * after reacquiring the slots_arch_lock below.
1561 slots = install_new_memslots(kvm, as_id, slots);
1563 /* From this point no new shadow pages pointing to a deleted,
1564 * or moved, memslot will be created.
1566 * validation of sp->gfn happens in:
1567 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1568 * - kvm_is_visible_gfn (mmu_check_root)
1570 kvm_arch_flush_shadow_memslot(kvm, slot);
1572 /* Released in install_new_memslots. */
1573 mutex_lock(&kvm->slots_arch_lock);
1576 * The arch-specific fields of the memslots could have changed
1577 * between releasing the slots_arch_lock in
1578 * install_new_memslots and here, so get a fresh copy of the
1581 kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id));
1584 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1588 update_memslots(slots, new, change);
1589 slots = install_new_memslots(kvm, as_id, slots);
1591 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1597 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1598 slot = id_to_memslot(slots, old->id);
1599 slot->flags &= ~KVM_MEMSLOT_INVALID;
1600 slots = install_new_memslots(kvm, as_id, slots);
1602 mutex_unlock(&kvm->slots_arch_lock);
1608 static int kvm_delete_memslot(struct kvm *kvm,
1609 const struct kvm_userspace_memory_region *mem,
1610 struct kvm_memory_slot *old, int as_id)
1612 struct kvm_memory_slot new;
1618 memset(&new, 0, sizeof(new));
1621 * This is only for debugging purpose; it should never be referenced
1622 * for a removed memslot.
1626 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1630 kvm_free_memslot(kvm, old);
1635 * Allocate some memory and give it an address in the guest physical address
1638 * Discontiguous memory is allowed, mostly for framebuffers.
1640 * Must be called holding kvm->slots_lock for write.
1642 int __kvm_set_memory_region(struct kvm *kvm,
1643 const struct kvm_userspace_memory_region *mem)
1645 struct kvm_memory_slot old, new;
1646 struct kvm_memory_slot *tmp;
1647 enum kvm_mr_change change;
1651 r = check_memory_region_flags(mem);
1655 as_id = mem->slot >> 16;
1656 id = (u16)mem->slot;
1658 /* General sanity checks */
1659 if (mem->memory_size & (PAGE_SIZE - 1))
1661 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1663 /* We can read the guest memory with __xxx_user() later on. */
1664 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1665 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1666 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1669 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1671 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1675 * Make a full copy of the old memslot, the pointer will become stale
1676 * when the memslots are re-sorted by update_memslots(), and the old
1677 * memslot needs to be referenced after calling update_memslots(), e.g.
1678 * to free its resources and for arch specific behavior.
1680 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1685 memset(&old, 0, sizeof(old));
1689 if (!mem->memory_size)
1690 return kvm_delete_memslot(kvm, mem, &old, as_id);
1694 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1695 new.npages = mem->memory_size >> PAGE_SHIFT;
1696 new.flags = mem->flags;
1697 new.userspace_addr = mem->userspace_addr;
1699 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1703 change = KVM_MR_CREATE;
1704 new.dirty_bitmap = NULL;
1705 memset(&new.arch, 0, sizeof(new.arch));
1706 } else { /* Modify an existing slot. */
1707 if ((new.userspace_addr != old.userspace_addr) ||
1708 (new.npages != old.npages) ||
1709 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1712 if (new.base_gfn != old.base_gfn)
1713 change = KVM_MR_MOVE;
1714 else if (new.flags != old.flags)
1715 change = KVM_MR_FLAGS_ONLY;
1716 else /* Nothing to change. */
1719 /* Copy dirty_bitmap and arch from the current memslot. */
1720 new.dirty_bitmap = old.dirty_bitmap;
1721 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1724 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1725 /* Check for overlaps */
1726 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1729 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1730 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1735 /* Allocate/free page dirty bitmap as needed */
1736 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1737 new.dirty_bitmap = NULL;
1738 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1739 r = kvm_alloc_dirty_bitmap(&new);
1743 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1744 bitmap_set(new.dirty_bitmap, 0, new.npages);
1747 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1751 if (old.dirty_bitmap && !new.dirty_bitmap)
1752 kvm_destroy_dirty_bitmap(&old);
1756 if (new.dirty_bitmap && !old.dirty_bitmap)
1757 kvm_destroy_dirty_bitmap(&new);
1760 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1762 int kvm_set_memory_region(struct kvm *kvm,
1763 const struct kvm_userspace_memory_region *mem)
1767 mutex_lock(&kvm->slots_lock);
1768 r = __kvm_set_memory_region(kvm, mem);
1769 mutex_unlock(&kvm->slots_lock);
1772 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1774 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1775 struct kvm_userspace_memory_region *mem)
1777 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1780 return kvm_set_memory_region(kvm, mem);
1783 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1785 * kvm_get_dirty_log - get a snapshot of dirty pages
1786 * @kvm: pointer to kvm instance
1787 * @log: slot id and address to which we copy the log
1788 * @is_dirty: set to '1' if any dirty pages were found
1789 * @memslot: set to the associated memslot, always valid on success
1791 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1792 int *is_dirty, struct kvm_memory_slot **memslot)
1794 struct kvm_memslots *slots;
1797 unsigned long any = 0;
1799 /* Dirty ring tracking is exclusive to dirty log tracking */
1800 if (kvm->dirty_ring_size)
1806 as_id = log->slot >> 16;
1807 id = (u16)log->slot;
1808 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1811 slots = __kvm_memslots(kvm, as_id);
1812 *memslot = id_to_memslot(slots, id);
1813 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1816 kvm_arch_sync_dirty_log(kvm, *memslot);
1818 n = kvm_dirty_bitmap_bytes(*memslot);
1820 for (i = 0; !any && i < n/sizeof(long); ++i)
1821 any = (*memslot)->dirty_bitmap[i];
1823 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1830 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1832 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1834 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1835 * and reenable dirty page tracking for the corresponding pages.
1836 * @kvm: pointer to kvm instance
1837 * @log: slot id and address to which we copy the log
1839 * We need to keep it in mind that VCPU threads can write to the bitmap
1840 * concurrently. So, to avoid losing track of dirty pages we keep the
1843 * 1. Take a snapshot of the bit and clear it if needed.
1844 * 2. Write protect the corresponding page.
1845 * 3. Copy the snapshot to the userspace.
1846 * 4. Upon return caller flushes TLB's if needed.
1848 * Between 2 and 4, the guest may write to the page using the remaining TLB
1849 * entry. This is not a problem because the page is reported dirty using
1850 * the snapshot taken before and step 4 ensures that writes done after
1851 * exiting to userspace will be logged for the next call.
1854 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1856 struct kvm_memslots *slots;
1857 struct kvm_memory_slot *memslot;
1860 unsigned long *dirty_bitmap;
1861 unsigned long *dirty_bitmap_buffer;
1864 /* Dirty ring tracking is exclusive to dirty log tracking */
1865 if (kvm->dirty_ring_size)
1868 as_id = log->slot >> 16;
1869 id = (u16)log->slot;
1870 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1873 slots = __kvm_memslots(kvm, as_id);
1874 memslot = id_to_memslot(slots, id);
1875 if (!memslot || !memslot->dirty_bitmap)
1878 dirty_bitmap = memslot->dirty_bitmap;
1880 kvm_arch_sync_dirty_log(kvm, memslot);
1882 n = kvm_dirty_bitmap_bytes(memslot);
1884 if (kvm->manual_dirty_log_protect) {
1886 * Unlike kvm_get_dirty_log, we always return false in *flush,
1887 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1888 * is some code duplication between this function and
1889 * kvm_get_dirty_log, but hopefully all architecture
1890 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1891 * can be eliminated.
1893 dirty_bitmap_buffer = dirty_bitmap;
1895 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1896 memset(dirty_bitmap_buffer, 0, n);
1899 for (i = 0; i < n / sizeof(long); i++) {
1903 if (!dirty_bitmap[i])
1907 mask = xchg(&dirty_bitmap[i], 0);
1908 dirty_bitmap_buffer[i] = mask;
1910 offset = i * BITS_PER_LONG;
1911 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1914 KVM_MMU_UNLOCK(kvm);
1918 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1920 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1927 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1928 * @kvm: kvm instance
1929 * @log: slot id and address to which we copy the log
1931 * Steps 1-4 below provide general overview of dirty page logging. See
1932 * kvm_get_dirty_log_protect() function description for additional details.
1934 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1935 * always flush the TLB (step 4) even if previous step failed and the dirty
1936 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1937 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1938 * writes will be marked dirty for next log read.
1940 * 1. Take a snapshot of the bit and clear it if needed.
1941 * 2. Write protect the corresponding page.
1942 * 3. Copy the snapshot to the userspace.
1943 * 4. Flush TLB's if needed.
1945 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1946 struct kvm_dirty_log *log)
1950 mutex_lock(&kvm->slots_lock);
1952 r = kvm_get_dirty_log_protect(kvm, log);
1954 mutex_unlock(&kvm->slots_lock);
1959 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1960 * and reenable dirty page tracking for the corresponding pages.
1961 * @kvm: pointer to kvm instance
1962 * @log: slot id and address from which to fetch the bitmap of dirty pages
1964 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1965 struct kvm_clear_dirty_log *log)
1967 struct kvm_memslots *slots;
1968 struct kvm_memory_slot *memslot;
1972 unsigned long *dirty_bitmap;
1973 unsigned long *dirty_bitmap_buffer;
1976 /* Dirty ring tracking is exclusive to dirty log tracking */
1977 if (kvm->dirty_ring_size)
1980 as_id = log->slot >> 16;
1981 id = (u16)log->slot;
1982 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1985 if (log->first_page & 63)
1988 slots = __kvm_memslots(kvm, as_id);
1989 memslot = id_to_memslot(slots, id);
1990 if (!memslot || !memslot->dirty_bitmap)
1993 dirty_bitmap = memslot->dirty_bitmap;
1995 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1997 if (log->first_page > memslot->npages ||
1998 log->num_pages > memslot->npages - log->first_page ||
1999 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2002 kvm_arch_sync_dirty_log(kvm, memslot);
2005 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2006 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2010 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2011 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2012 i++, offset += BITS_PER_LONG) {
2013 unsigned long mask = *dirty_bitmap_buffer++;
2014 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2018 mask &= atomic_long_fetch_andnot(mask, p);
2021 * mask contains the bits that really have been cleared. This
2022 * never includes any bits beyond the length of the memslot (if
2023 * the length is not aligned to 64 pages), therefore it is not
2024 * a problem if userspace sets them in log->dirty_bitmap.
2028 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2032 KVM_MMU_UNLOCK(kvm);
2035 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2040 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2041 struct kvm_clear_dirty_log *log)
2045 mutex_lock(&kvm->slots_lock);
2047 r = kvm_clear_dirty_log_protect(kvm, log);
2049 mutex_unlock(&kvm->slots_lock);
2052 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2054 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2056 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2058 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2060 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2062 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2063 struct kvm_memory_slot *slot;
2066 slot = try_get_memslot(slots, vcpu->last_used_slot, gfn);
2071 * Fall back to searching all memslots. We purposely use
2072 * search_memslots() instead of __gfn_to_memslot() to avoid
2073 * thrashing the VM-wide last_used_index in kvm_memslots.
2075 slot = search_memslots(slots, gfn, &slot_index);
2077 vcpu->last_used_slot = slot_index;
2083 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
2085 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2087 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2089 return kvm_is_visible_memslot(memslot);
2091 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2093 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2095 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2097 return kvm_is_visible_memslot(memslot);
2099 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2101 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2103 struct vm_area_struct *vma;
2104 unsigned long addr, size;
2108 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2109 if (kvm_is_error_hva(addr))
2112 mmap_read_lock(current->mm);
2113 vma = find_vma(current->mm, addr);
2117 size = vma_kernel_pagesize(vma);
2120 mmap_read_unlock(current->mm);
2125 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
2127 return slot->flags & KVM_MEM_READONLY;
2130 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2131 gfn_t *nr_pages, bool write)
2133 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2134 return KVM_HVA_ERR_BAD;
2136 if (memslot_is_readonly(slot) && write)
2137 return KVM_HVA_ERR_RO_BAD;
2140 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2142 return __gfn_to_hva_memslot(slot, gfn);
2145 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2148 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2151 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2154 return gfn_to_hva_many(slot, gfn, NULL);
2156 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2158 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2160 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2162 EXPORT_SYMBOL_GPL(gfn_to_hva);
2164 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2166 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2168 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2171 * Return the hva of a @gfn and the R/W attribute if possible.
2173 * @slot: the kvm_memory_slot which contains @gfn
2174 * @gfn: the gfn to be translated
2175 * @writable: used to return the read/write attribute of the @slot if the hva
2176 * is valid and @writable is not NULL
2178 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2179 gfn_t gfn, bool *writable)
2181 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2183 if (!kvm_is_error_hva(hva) && writable)
2184 *writable = !memslot_is_readonly(slot);
2189 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2191 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2193 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2196 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2198 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2200 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2203 static inline int check_user_page_hwpoison(unsigned long addr)
2205 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2207 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2208 return rc == -EHWPOISON;
2212 * The fast path to get the writable pfn which will be stored in @pfn,
2213 * true indicates success, otherwise false is returned. It's also the
2214 * only part that runs if we can in atomic context.
2216 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2217 bool *writable, kvm_pfn_t *pfn)
2219 struct page *page[1];
2222 * Fast pin a writable pfn only if it is a write fault request
2223 * or the caller allows to map a writable pfn for a read fault
2226 if (!(write_fault || writable))
2229 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2230 *pfn = page_to_pfn(page[0]);
2241 * The slow path to get the pfn of the specified host virtual address,
2242 * 1 indicates success, -errno is returned if error is detected.
2244 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2245 bool *writable, kvm_pfn_t *pfn)
2247 unsigned int flags = FOLL_HWPOISON;
2254 *writable = write_fault;
2257 flags |= FOLL_WRITE;
2259 flags |= FOLL_NOWAIT;
2261 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2265 /* map read fault as writable if possible */
2266 if (unlikely(!write_fault) && writable) {
2269 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2275 *pfn = page_to_pfn(page);
2279 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2281 if (unlikely(!(vma->vm_flags & VM_READ)))
2284 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2290 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2292 if (kvm_is_reserved_pfn(pfn))
2294 return get_page_unless_zero(pfn_to_page(pfn));
2297 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2298 unsigned long addr, bool *async,
2299 bool write_fault, bool *writable,
2307 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2310 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2311 * not call the fault handler, so do it here.
2313 bool unlocked = false;
2314 r = fixup_user_fault(current->mm, addr,
2315 (write_fault ? FAULT_FLAG_WRITE : 0),
2322 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2327 if (write_fault && !pte_write(*ptep)) {
2328 pfn = KVM_PFN_ERR_RO_FAULT;
2333 *writable = pte_write(*ptep);
2334 pfn = pte_pfn(*ptep);
2337 * Get a reference here because callers of *hva_to_pfn* and
2338 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2339 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2340 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2341 * simply do nothing for reserved pfns.
2343 * Whoever called remap_pfn_range is also going to call e.g.
2344 * unmap_mapping_range before the underlying pages are freed,
2345 * causing a call to our MMU notifier.
2347 * Certain IO or PFNMAP mappings can be backed with valid
2348 * struct pages, but be allocated without refcounting e.g.,
2349 * tail pages of non-compound higher order allocations, which
2350 * would then underflow the refcount when the caller does the
2351 * required put_page. Don't allow those pages here.
2353 if (!kvm_try_get_pfn(pfn))
2357 pte_unmap_unlock(ptep, ptl);
2364 * Pin guest page in memory and return its pfn.
2365 * @addr: host virtual address which maps memory to the guest
2366 * @atomic: whether this function can sleep
2367 * @async: whether this function need to wait IO complete if the
2368 * host page is not in the memory
2369 * @write_fault: whether we should get a writable host page
2370 * @writable: whether it allows to map a writable host page for !@write_fault
2372 * The function will map a writable host page for these two cases:
2373 * 1): @write_fault = true
2374 * 2): @write_fault = false && @writable, @writable will tell the caller
2375 * whether the mapping is writable.
2377 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2378 bool write_fault, bool *writable)
2380 struct vm_area_struct *vma;
2384 /* we can do it either atomically or asynchronously, not both */
2385 BUG_ON(atomic && async);
2387 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2391 return KVM_PFN_ERR_FAULT;
2393 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2397 mmap_read_lock(current->mm);
2398 if (npages == -EHWPOISON ||
2399 (!async && check_user_page_hwpoison(addr))) {
2400 pfn = KVM_PFN_ERR_HWPOISON;
2405 vma = vma_lookup(current->mm, addr);
2408 pfn = KVM_PFN_ERR_FAULT;
2409 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2410 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2414 pfn = KVM_PFN_ERR_FAULT;
2416 if (async && vma_is_valid(vma, write_fault))
2418 pfn = KVM_PFN_ERR_FAULT;
2421 mmap_read_unlock(current->mm);
2425 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2426 bool atomic, bool *async, bool write_fault,
2427 bool *writable, hva_t *hva)
2429 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2434 if (addr == KVM_HVA_ERR_RO_BAD) {
2437 return KVM_PFN_ERR_RO_FAULT;
2440 if (kvm_is_error_hva(addr)) {
2443 return KVM_PFN_NOSLOT;
2446 /* Do not map writable pfn in the readonly memslot. */
2447 if (writable && memslot_is_readonly(slot)) {
2452 return hva_to_pfn(addr, atomic, async, write_fault,
2455 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2457 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2460 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2461 write_fault, writable, NULL);
2463 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2465 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2467 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2469 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2471 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2473 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2475 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2477 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2479 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2481 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2483 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2485 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2487 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2489 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2491 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2493 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2495 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2496 struct page **pages, int nr_pages)
2501 addr = gfn_to_hva_many(slot, gfn, &entry);
2502 if (kvm_is_error_hva(addr))
2505 if (entry < nr_pages)
2508 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2510 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2512 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2514 if (is_error_noslot_pfn(pfn))
2515 return KVM_ERR_PTR_BAD_PAGE;
2517 if (kvm_is_reserved_pfn(pfn)) {
2519 return KVM_ERR_PTR_BAD_PAGE;
2522 return pfn_to_page(pfn);
2525 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2529 pfn = gfn_to_pfn(kvm, gfn);
2531 return kvm_pfn_to_page(pfn);
2533 EXPORT_SYMBOL_GPL(gfn_to_page);
2535 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2541 cache->pfn = cache->gfn = 0;
2544 kvm_release_pfn_dirty(pfn);
2546 kvm_release_pfn_clean(pfn);
2549 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2550 struct gfn_to_pfn_cache *cache, u64 gen)
2552 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2554 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2556 cache->dirty = false;
2557 cache->generation = gen;
2560 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2561 struct kvm_host_map *map,
2562 struct gfn_to_pfn_cache *cache,
2567 struct page *page = KVM_UNMAPPED_PAGE;
2568 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2569 u64 gen = slots->generation;
2575 if (!cache->pfn || cache->gfn != gfn ||
2576 cache->generation != gen) {
2579 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2585 pfn = gfn_to_pfn_memslot(slot, gfn);
2587 if (is_error_noslot_pfn(pfn))
2590 if (pfn_valid(pfn)) {
2591 page = pfn_to_page(pfn);
2593 hva = kmap_atomic(page);
2596 #ifdef CONFIG_HAS_IOMEM
2597 } else if (!atomic) {
2598 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2615 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2616 struct gfn_to_pfn_cache *cache, bool atomic)
2618 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2621 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2623 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2625 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2628 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2630 static void __kvm_unmap_gfn(struct kvm *kvm,
2631 struct kvm_memory_slot *memslot,
2632 struct kvm_host_map *map,
2633 struct gfn_to_pfn_cache *cache,
2634 bool dirty, bool atomic)
2642 if (map->page != KVM_UNMAPPED_PAGE) {
2644 kunmap_atomic(map->hva);
2648 #ifdef CONFIG_HAS_IOMEM
2652 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2656 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2659 cache->dirty |= dirty;
2661 kvm_release_pfn(map->pfn, dirty, NULL);
2667 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2668 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2670 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2671 cache, dirty, atomic);
2674 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2676 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2678 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2679 map, NULL, dirty, false);
2681 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2683 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2687 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2689 return kvm_pfn_to_page(pfn);
2691 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2693 void kvm_release_page_clean(struct page *page)
2695 WARN_ON(is_error_page(page));
2697 kvm_release_pfn_clean(page_to_pfn(page));
2699 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2701 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2703 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2704 put_page(pfn_to_page(pfn));
2706 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2708 void kvm_release_page_dirty(struct page *page)
2710 WARN_ON(is_error_page(page));
2712 kvm_release_pfn_dirty(page_to_pfn(page));
2714 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2716 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2718 kvm_set_pfn_dirty(pfn);
2719 kvm_release_pfn_clean(pfn);
2721 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2723 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2725 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2726 SetPageDirty(pfn_to_page(pfn));
2728 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2730 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2732 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2733 mark_page_accessed(pfn_to_page(pfn));
2735 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2737 void kvm_get_pfn(kvm_pfn_t pfn)
2739 if (!kvm_is_reserved_pfn(pfn))
2740 get_page(pfn_to_page(pfn));
2742 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2744 static int next_segment(unsigned long len, int offset)
2746 if (len > PAGE_SIZE - offset)
2747 return PAGE_SIZE - offset;
2752 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2753 void *data, int offset, int len)
2758 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2759 if (kvm_is_error_hva(addr))
2761 r = __copy_from_user(data, (void __user *)addr + offset, len);
2767 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2770 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2772 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2774 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2776 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2777 int offset, int len)
2779 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2781 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2783 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2785 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2787 gfn_t gfn = gpa >> PAGE_SHIFT;
2789 int offset = offset_in_page(gpa);
2792 while ((seg = next_segment(len, offset)) != 0) {
2793 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2803 EXPORT_SYMBOL_GPL(kvm_read_guest);
2805 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2807 gfn_t gfn = gpa >> PAGE_SHIFT;
2809 int offset = offset_in_page(gpa);
2812 while ((seg = next_segment(len, offset)) != 0) {
2813 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2823 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2825 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2826 void *data, int offset, unsigned long len)
2831 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2832 if (kvm_is_error_hva(addr))
2834 pagefault_disable();
2835 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2842 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2843 void *data, unsigned long len)
2845 gfn_t gfn = gpa >> PAGE_SHIFT;
2846 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2847 int offset = offset_in_page(gpa);
2849 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2851 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2853 static int __kvm_write_guest_page(struct kvm *kvm,
2854 struct kvm_memory_slot *memslot, gfn_t gfn,
2855 const void *data, int offset, int len)
2860 addr = gfn_to_hva_memslot(memslot, gfn);
2861 if (kvm_is_error_hva(addr))
2863 r = __copy_to_user((void __user *)addr + offset, data, len);
2866 mark_page_dirty_in_slot(kvm, memslot, gfn);
2870 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2871 const void *data, int offset, int len)
2873 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2875 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2877 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2879 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2880 const void *data, int offset, int len)
2882 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2884 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2886 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2888 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2891 gfn_t gfn = gpa >> PAGE_SHIFT;
2893 int offset = offset_in_page(gpa);
2896 while ((seg = next_segment(len, offset)) != 0) {
2897 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2907 EXPORT_SYMBOL_GPL(kvm_write_guest);
2909 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2912 gfn_t gfn = gpa >> PAGE_SHIFT;
2914 int offset = offset_in_page(gpa);
2917 while ((seg = next_segment(len, offset)) != 0) {
2918 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2928 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2930 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2931 struct gfn_to_hva_cache *ghc,
2932 gpa_t gpa, unsigned long len)
2934 int offset = offset_in_page(gpa);
2935 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2936 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2937 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2938 gfn_t nr_pages_avail;
2940 /* Update ghc->generation before performing any error checks. */
2941 ghc->generation = slots->generation;
2943 if (start_gfn > end_gfn) {
2944 ghc->hva = KVM_HVA_ERR_BAD;
2949 * If the requested region crosses two memslots, we still
2950 * verify that the entire region is valid here.
2952 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2953 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2954 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2956 if (kvm_is_error_hva(ghc->hva))
2960 /* Use the slow path for cross page reads and writes. */
2961 if (nr_pages_needed == 1)
2964 ghc->memslot = NULL;
2971 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2972 gpa_t gpa, unsigned long len)
2974 struct kvm_memslots *slots = kvm_memslots(kvm);
2975 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2977 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2979 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2980 void *data, unsigned int offset,
2983 struct kvm_memslots *slots = kvm_memslots(kvm);
2985 gpa_t gpa = ghc->gpa + offset;
2987 BUG_ON(len + offset > ghc->len);
2989 if (slots->generation != ghc->generation) {
2990 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2994 if (kvm_is_error_hva(ghc->hva))
2997 if (unlikely(!ghc->memslot))
2998 return kvm_write_guest(kvm, gpa, data, len);
3000 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3003 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3007 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3009 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3010 void *data, unsigned long len)
3012 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3014 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3016 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3017 void *data, unsigned int offset,
3020 struct kvm_memslots *slots = kvm_memslots(kvm);
3022 gpa_t gpa = ghc->gpa + offset;
3024 BUG_ON(len + offset > ghc->len);
3026 if (slots->generation != ghc->generation) {
3027 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3031 if (kvm_is_error_hva(ghc->hva))
3034 if (unlikely(!ghc->memslot))
3035 return kvm_read_guest(kvm, gpa, data, len);
3037 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3043 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3045 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3046 void *data, unsigned long len)
3048 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3050 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3052 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3054 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3055 gfn_t gfn = gpa >> PAGE_SHIFT;
3057 int offset = offset_in_page(gpa);
3060 while ((seg = next_segment(len, offset)) != 0) {
3061 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3070 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3072 void mark_page_dirty_in_slot(struct kvm *kvm,
3073 struct kvm_memory_slot *memslot,
3076 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3077 unsigned long rel_gfn = gfn - memslot->base_gfn;
3078 u32 slot = (memslot->as_id << 16) | memslot->id;
3080 if (kvm->dirty_ring_size)
3081 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
3084 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3087 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3089 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3091 struct kvm_memory_slot *memslot;
3093 memslot = gfn_to_memslot(kvm, gfn);
3094 mark_page_dirty_in_slot(kvm, memslot, gfn);
3096 EXPORT_SYMBOL_GPL(mark_page_dirty);
3098 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3100 struct kvm_memory_slot *memslot;
3102 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3103 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3105 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3107 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3109 if (!vcpu->sigset_active)
3113 * This does a lockless modification of ->real_blocked, which is fine
3114 * because, only current can change ->real_blocked and all readers of
3115 * ->real_blocked don't care as long ->real_blocked is always a subset
3118 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3121 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3123 if (!vcpu->sigset_active)
3126 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3127 sigemptyset(¤t->real_blocked);
3130 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3132 unsigned int old, val, grow, grow_start;
3134 old = val = vcpu->halt_poll_ns;
3135 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3136 grow = READ_ONCE(halt_poll_ns_grow);
3141 if (val < grow_start)
3144 if (val > vcpu->kvm->max_halt_poll_ns)
3145 val = vcpu->kvm->max_halt_poll_ns;
3147 vcpu->halt_poll_ns = val;
3149 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3152 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3154 unsigned int old, val, shrink;
3156 old = val = vcpu->halt_poll_ns;
3157 shrink = READ_ONCE(halt_poll_ns_shrink);
3163 vcpu->halt_poll_ns = val;
3164 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3167 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3170 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3172 if (kvm_arch_vcpu_runnable(vcpu)) {
3173 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3176 if (kvm_cpu_has_pending_timer(vcpu))
3178 if (signal_pending(current))
3180 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3185 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3190 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
3193 vcpu->stat.generic.halt_poll_fail_ns += poll_ns;
3195 vcpu->stat.generic.halt_poll_success_ns += poll_ns;
3199 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3201 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
3203 ktime_t start, cur, poll_end;
3204 bool waited = false;
3207 kvm_arch_vcpu_blocking(vcpu);
3209 start = cur = poll_end = ktime_get();
3210 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
3211 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3213 ++vcpu->stat.generic.halt_attempted_poll;
3216 * This sets KVM_REQ_UNHALT if an interrupt
3219 if (kvm_vcpu_check_block(vcpu) < 0) {
3220 ++vcpu->stat.generic.halt_successful_poll;
3221 if (!vcpu_valid_wakeup(vcpu))
3222 ++vcpu->stat.generic.halt_poll_invalid;
3224 KVM_STATS_LOG_HIST_UPDATE(
3225 vcpu->stat.generic.halt_poll_success_hist,
3226 ktime_to_ns(ktime_get()) -
3227 ktime_to_ns(start));
3231 poll_end = cur = ktime_get();
3232 } while (kvm_vcpu_can_poll(cur, stop));
3234 KVM_STATS_LOG_HIST_UPDATE(
3235 vcpu->stat.generic.halt_poll_fail_hist,
3236 ktime_to_ns(ktime_get()) - ktime_to_ns(start));
3240 prepare_to_rcuwait(&vcpu->wait);
3242 set_current_state(TASK_INTERRUPTIBLE);
3244 if (kvm_vcpu_check_block(vcpu) < 0)
3250 finish_rcuwait(&vcpu->wait);
3253 vcpu->stat.generic.halt_wait_ns +=
3254 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3255 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3256 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3259 kvm_arch_vcpu_unblocking(vcpu);
3260 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3262 update_halt_poll_stats(
3263 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3265 if (!kvm_arch_no_poll(vcpu)) {
3266 if (!vcpu_valid_wakeup(vcpu)) {
3267 shrink_halt_poll_ns(vcpu);
3268 } else if (vcpu->kvm->max_halt_poll_ns) {
3269 if (block_ns <= vcpu->halt_poll_ns)
3271 /* we had a long block, shrink polling */
3272 else if (vcpu->halt_poll_ns &&
3273 block_ns > vcpu->kvm->max_halt_poll_ns)
3274 shrink_halt_poll_ns(vcpu);
3275 /* we had a short halt and our poll time is too small */
3276 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3277 block_ns < vcpu->kvm->max_halt_poll_ns)
3278 grow_halt_poll_ns(vcpu);
3280 vcpu->halt_poll_ns = 0;
3284 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3285 kvm_arch_vcpu_block_finish(vcpu);
3287 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3289 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3291 struct rcuwait *waitp;
3293 waitp = kvm_arch_vcpu_get_wait(vcpu);
3294 if (rcuwait_wake_up(waitp)) {
3295 WRITE_ONCE(vcpu->ready, true);
3296 ++vcpu->stat.generic.halt_wakeup;
3302 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3306 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3308 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3311 int cpu = vcpu->cpu;
3313 if (kvm_vcpu_wake_up(vcpu))
3317 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3318 if (kvm_arch_vcpu_should_kick(vcpu))
3319 smp_send_reschedule(cpu);
3322 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3323 #endif /* !CONFIG_S390 */
3325 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3328 struct task_struct *task = NULL;
3332 pid = rcu_dereference(target->pid);
3334 task = get_pid_task(pid, PIDTYPE_PID);
3338 ret = yield_to(task, 1);
3339 put_task_struct(task);
3343 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3346 * Helper that checks whether a VCPU is eligible for directed yield.
3347 * Most eligible candidate to yield is decided by following heuristics:
3349 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3350 * (preempted lock holder), indicated by @in_spin_loop.
3351 * Set at the beginning and cleared at the end of interception/PLE handler.
3353 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3354 * chance last time (mostly it has become eligible now since we have probably
3355 * yielded to lockholder in last iteration. This is done by toggling
3356 * @dy_eligible each time a VCPU checked for eligibility.)
3358 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3359 * to preempted lock-holder could result in wrong VCPU selection and CPU
3360 * burning. Giving priority for a potential lock-holder increases lock
3363 * Since algorithm is based on heuristics, accessing another VCPU data without
3364 * locking does not harm. It may result in trying to yield to same VCPU, fail
3365 * and continue with next VCPU and so on.
3367 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3369 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3372 eligible = !vcpu->spin_loop.in_spin_loop ||
3373 vcpu->spin_loop.dy_eligible;
3375 if (vcpu->spin_loop.in_spin_loop)
3376 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3385 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3386 * a vcpu_load/vcpu_put pair. However, for most architectures
3387 * kvm_arch_vcpu_runnable does not require vcpu_load.
3389 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3391 return kvm_arch_vcpu_runnable(vcpu);
3394 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3396 if (kvm_arch_dy_runnable(vcpu))
3399 #ifdef CONFIG_KVM_ASYNC_PF
3400 if (!list_empty_careful(&vcpu->async_pf.done))
3407 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3412 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3414 struct kvm *kvm = me->kvm;
3415 struct kvm_vcpu *vcpu;
3416 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3422 kvm_vcpu_set_in_spin_loop(me, true);
3424 * We boost the priority of a VCPU that is runnable but not
3425 * currently running, because it got preempted by something
3426 * else and called schedule in __vcpu_run. Hopefully that
3427 * VCPU is holding the lock that we need and will release it.
3428 * We approximate round-robin by starting at the last boosted VCPU.
3430 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3431 kvm_for_each_vcpu(i, vcpu, kvm) {
3432 if (!pass && i <= last_boosted_vcpu) {
3433 i = last_boosted_vcpu;
3435 } else if (pass && i > last_boosted_vcpu)
3437 if (!READ_ONCE(vcpu->ready))
3441 if (rcuwait_active(&vcpu->wait) &&
3442 !vcpu_dy_runnable(vcpu))
3444 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3445 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3446 !kvm_arch_vcpu_in_kernel(vcpu))
3448 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3451 yielded = kvm_vcpu_yield_to(vcpu);
3453 kvm->last_boosted_vcpu = i;
3455 } else if (yielded < 0) {
3462 kvm_vcpu_set_in_spin_loop(me, false);
3464 /* Ensure vcpu is not eligible during next spinloop */
3465 kvm_vcpu_set_dy_eligible(me, false);
3467 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3469 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3471 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3472 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3473 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3474 kvm->dirty_ring_size / PAGE_SIZE);
3480 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3482 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3485 if (vmf->pgoff == 0)
3486 page = virt_to_page(vcpu->run);
3488 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3489 page = virt_to_page(vcpu->arch.pio_data);
3491 #ifdef CONFIG_KVM_MMIO
3492 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3493 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3495 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3496 page = kvm_dirty_ring_get_page(
3498 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3500 return kvm_arch_vcpu_fault(vcpu, vmf);
3506 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3507 .fault = kvm_vcpu_fault,
3510 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3512 struct kvm_vcpu *vcpu = file->private_data;
3513 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3515 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3516 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3517 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3520 vma->vm_ops = &kvm_vcpu_vm_ops;
3524 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3526 struct kvm_vcpu *vcpu = filp->private_data;
3528 kvm_put_kvm(vcpu->kvm);
3532 static struct file_operations kvm_vcpu_fops = {
3533 .release = kvm_vcpu_release,
3534 .unlocked_ioctl = kvm_vcpu_ioctl,
3535 .mmap = kvm_vcpu_mmap,
3536 .llseek = noop_llseek,
3537 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3541 * Allocates an inode for the vcpu.
3543 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3545 char name[8 + 1 + ITOA_MAX_LEN + 1];
3547 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3548 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3551 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3553 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3554 struct dentry *debugfs_dentry;
3555 char dir_name[ITOA_MAX_LEN * 2];
3557 if (!debugfs_initialized())
3560 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3561 debugfs_dentry = debugfs_create_dir(dir_name,
3562 vcpu->kvm->debugfs_dentry);
3564 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3569 * Creates some virtual cpus. Good luck creating more than one.
3571 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3574 struct kvm_vcpu *vcpu;
3577 if (id >= KVM_MAX_VCPU_ID)
3580 mutex_lock(&kvm->lock);
3581 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3582 mutex_unlock(&kvm->lock);
3586 kvm->created_vcpus++;
3587 mutex_unlock(&kvm->lock);
3589 r = kvm_arch_vcpu_precreate(kvm, id);
3591 goto vcpu_decrement;
3593 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3596 goto vcpu_decrement;
3599 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3600 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3605 vcpu->run = page_address(page);
3607 kvm_vcpu_init(vcpu, kvm, id);
3609 r = kvm_arch_vcpu_create(vcpu);
3611 goto vcpu_free_run_page;
3613 if (kvm->dirty_ring_size) {
3614 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3615 id, kvm->dirty_ring_size);
3617 goto arch_vcpu_destroy;
3620 mutex_lock(&kvm->lock);
3621 if (kvm_get_vcpu_by_id(kvm, id)) {
3623 goto unlock_vcpu_destroy;
3626 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3627 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3629 /* Fill the stats id string for the vcpu */
3630 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3631 task_pid_nr(current), id);
3633 /* Now it's all set up, let userspace reach it */
3635 r = create_vcpu_fd(vcpu);
3637 kvm_put_kvm_no_destroy(kvm);
3638 goto unlock_vcpu_destroy;
3641 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3644 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3645 * before kvm->online_vcpu's incremented value.
3648 atomic_inc(&kvm->online_vcpus);
3650 mutex_unlock(&kvm->lock);
3651 kvm_arch_vcpu_postcreate(vcpu);
3652 kvm_create_vcpu_debugfs(vcpu);
3655 unlock_vcpu_destroy:
3656 mutex_unlock(&kvm->lock);
3657 kvm_dirty_ring_free(&vcpu->dirty_ring);
3659 kvm_arch_vcpu_destroy(vcpu);
3661 free_page((unsigned long)vcpu->run);
3663 kmem_cache_free(kvm_vcpu_cache, vcpu);
3665 mutex_lock(&kvm->lock);
3666 kvm->created_vcpus--;
3667 mutex_unlock(&kvm->lock);
3671 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3674 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3675 vcpu->sigset_active = 1;
3676 vcpu->sigset = *sigset;
3678 vcpu->sigset_active = 0;
3682 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3683 size_t size, loff_t *offset)
3685 struct kvm_vcpu *vcpu = file->private_data;
3687 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3688 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3689 sizeof(vcpu->stat), user_buffer, size, offset);
3692 static const struct file_operations kvm_vcpu_stats_fops = {
3693 .read = kvm_vcpu_stats_read,
3694 .llseek = noop_llseek,
3697 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3701 char name[15 + ITOA_MAX_LEN + 1];
3703 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3705 fd = get_unused_fd_flags(O_CLOEXEC);
3709 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3712 return PTR_ERR(file);
3714 file->f_mode |= FMODE_PREAD;
3715 fd_install(fd, file);
3720 static long kvm_vcpu_ioctl(struct file *filp,
3721 unsigned int ioctl, unsigned long arg)
3723 struct kvm_vcpu *vcpu = filp->private_data;
3724 void __user *argp = (void __user *)arg;
3726 struct kvm_fpu *fpu = NULL;
3727 struct kvm_sregs *kvm_sregs = NULL;
3729 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3732 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3736 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3737 * execution; mutex_lock() would break them.
3739 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3740 if (r != -ENOIOCTLCMD)
3743 if (mutex_lock_killable(&vcpu->mutex))
3751 oldpid = rcu_access_pointer(vcpu->pid);
3752 if (unlikely(oldpid != task_pid(current))) {
3753 /* The thread running this VCPU changed. */
3756 r = kvm_arch_vcpu_run_pid_change(vcpu);
3760 newpid = get_task_pid(current, PIDTYPE_PID);
3761 rcu_assign_pointer(vcpu->pid, newpid);
3766 r = kvm_arch_vcpu_ioctl_run(vcpu);
3767 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3770 case KVM_GET_REGS: {
3771 struct kvm_regs *kvm_regs;
3774 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3777 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3781 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3788 case KVM_SET_REGS: {
3789 struct kvm_regs *kvm_regs;
3791 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3792 if (IS_ERR(kvm_regs)) {
3793 r = PTR_ERR(kvm_regs);
3796 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3800 case KVM_GET_SREGS: {
3801 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3802 GFP_KERNEL_ACCOUNT);
3806 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3810 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3815 case KVM_SET_SREGS: {
3816 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3817 if (IS_ERR(kvm_sregs)) {
3818 r = PTR_ERR(kvm_sregs);
3822 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3825 case KVM_GET_MP_STATE: {
3826 struct kvm_mp_state mp_state;
3828 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3832 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3837 case KVM_SET_MP_STATE: {
3838 struct kvm_mp_state mp_state;
3841 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3843 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3846 case KVM_TRANSLATE: {
3847 struct kvm_translation tr;
3850 if (copy_from_user(&tr, argp, sizeof(tr)))
3852 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3856 if (copy_to_user(argp, &tr, sizeof(tr)))
3861 case KVM_SET_GUEST_DEBUG: {
3862 struct kvm_guest_debug dbg;
3865 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3867 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3870 case KVM_SET_SIGNAL_MASK: {
3871 struct kvm_signal_mask __user *sigmask_arg = argp;
3872 struct kvm_signal_mask kvm_sigmask;
3873 sigset_t sigset, *p;
3878 if (copy_from_user(&kvm_sigmask, argp,
3879 sizeof(kvm_sigmask)))
3882 if (kvm_sigmask.len != sizeof(sigset))
3885 if (copy_from_user(&sigset, sigmask_arg->sigset,
3890 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3894 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3898 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3902 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3908 fpu = memdup_user(argp, sizeof(*fpu));
3914 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3917 case KVM_GET_STATS_FD: {
3918 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
3922 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3925 mutex_unlock(&vcpu->mutex);
3931 #ifdef CONFIG_KVM_COMPAT
3932 static long kvm_vcpu_compat_ioctl(struct file *filp,
3933 unsigned int ioctl, unsigned long arg)
3935 struct kvm_vcpu *vcpu = filp->private_data;
3936 void __user *argp = compat_ptr(arg);
3939 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3943 case KVM_SET_SIGNAL_MASK: {
3944 struct kvm_signal_mask __user *sigmask_arg = argp;
3945 struct kvm_signal_mask kvm_sigmask;
3950 if (copy_from_user(&kvm_sigmask, argp,
3951 sizeof(kvm_sigmask)))
3954 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3957 if (get_compat_sigset(&sigset,
3958 (compat_sigset_t __user *)sigmask_arg->sigset))
3960 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3962 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3966 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3974 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3976 struct kvm_device *dev = filp->private_data;
3979 return dev->ops->mmap(dev, vma);
3984 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3985 int (*accessor)(struct kvm_device *dev,
3986 struct kvm_device_attr *attr),
3989 struct kvm_device_attr attr;
3994 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3997 return accessor(dev, &attr);
4000 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4003 struct kvm_device *dev = filp->private_data;
4005 if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged)
4009 case KVM_SET_DEVICE_ATTR:
4010 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4011 case KVM_GET_DEVICE_ATTR:
4012 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4013 case KVM_HAS_DEVICE_ATTR:
4014 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4016 if (dev->ops->ioctl)
4017 return dev->ops->ioctl(dev, ioctl, arg);
4023 static int kvm_device_release(struct inode *inode, struct file *filp)
4025 struct kvm_device *dev = filp->private_data;
4026 struct kvm *kvm = dev->kvm;
4028 if (dev->ops->release) {
4029 mutex_lock(&kvm->lock);
4030 list_del(&dev->vm_node);
4031 dev->ops->release(dev);
4032 mutex_unlock(&kvm->lock);
4039 static const struct file_operations kvm_device_fops = {
4040 .unlocked_ioctl = kvm_device_ioctl,
4041 .release = kvm_device_release,
4042 KVM_COMPAT(kvm_device_ioctl),
4043 .mmap = kvm_device_mmap,
4046 struct kvm_device *kvm_device_from_filp(struct file *filp)
4048 if (filp->f_op != &kvm_device_fops)
4051 return filp->private_data;
4054 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4055 #ifdef CONFIG_KVM_MPIC
4056 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4057 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4061 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4063 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4066 if (kvm_device_ops_table[type] != NULL)
4069 kvm_device_ops_table[type] = ops;
4073 void kvm_unregister_device_ops(u32 type)
4075 if (kvm_device_ops_table[type] != NULL)
4076 kvm_device_ops_table[type] = NULL;
4079 static int kvm_ioctl_create_device(struct kvm *kvm,
4080 struct kvm_create_device *cd)
4082 const struct kvm_device_ops *ops = NULL;
4083 struct kvm_device *dev;
4084 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4088 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4091 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4092 ops = kvm_device_ops_table[type];
4099 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4106 mutex_lock(&kvm->lock);
4107 ret = ops->create(dev, type);
4109 mutex_unlock(&kvm->lock);
4113 list_add(&dev->vm_node, &kvm->devices);
4114 mutex_unlock(&kvm->lock);
4120 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4122 kvm_put_kvm_no_destroy(kvm);
4123 mutex_lock(&kvm->lock);
4124 list_del(&dev->vm_node);
4125 mutex_unlock(&kvm->lock);
4134 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4137 case KVM_CAP_USER_MEMORY:
4138 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4139 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4140 case KVM_CAP_INTERNAL_ERROR_DATA:
4141 #ifdef CONFIG_HAVE_KVM_MSI
4142 case KVM_CAP_SIGNAL_MSI:
4144 #ifdef CONFIG_HAVE_KVM_IRQFD
4146 case KVM_CAP_IRQFD_RESAMPLE:
4148 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4149 case KVM_CAP_CHECK_EXTENSION_VM:
4150 case KVM_CAP_ENABLE_CAP_VM:
4151 case KVM_CAP_HALT_POLL:
4153 #ifdef CONFIG_KVM_MMIO
4154 case KVM_CAP_COALESCED_MMIO:
4155 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4156 case KVM_CAP_COALESCED_PIO:
4159 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4160 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4161 return KVM_DIRTY_LOG_MANUAL_CAPS;
4163 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4164 case KVM_CAP_IRQ_ROUTING:
4165 return KVM_MAX_IRQ_ROUTES;
4167 #if KVM_ADDRESS_SPACE_NUM > 1
4168 case KVM_CAP_MULTI_ADDRESS_SPACE:
4169 return KVM_ADDRESS_SPACE_NUM;
4171 case KVM_CAP_NR_MEMSLOTS:
4172 return KVM_USER_MEM_SLOTS;
4173 case KVM_CAP_DIRTY_LOG_RING:
4174 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4175 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4179 case KVM_CAP_BINARY_STATS_FD:
4184 return kvm_vm_ioctl_check_extension(kvm, arg);
4187 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4191 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4194 /* the size should be power of 2 */
4195 if (!size || (size & (size - 1)))
4198 /* Should be bigger to keep the reserved entries, or a page */
4199 if (size < kvm_dirty_ring_get_rsvd_entries() *
4200 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4203 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4204 sizeof(struct kvm_dirty_gfn))
4207 /* We only allow it to set once */
4208 if (kvm->dirty_ring_size)
4211 mutex_lock(&kvm->lock);
4213 if (kvm->created_vcpus) {
4214 /* We don't allow to change this value after vcpu created */
4217 kvm->dirty_ring_size = size;
4221 mutex_unlock(&kvm->lock);
4225 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4228 struct kvm_vcpu *vcpu;
4231 if (!kvm->dirty_ring_size)
4234 mutex_lock(&kvm->slots_lock);
4236 kvm_for_each_vcpu(i, vcpu, kvm)
4237 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4239 mutex_unlock(&kvm->slots_lock);
4242 kvm_flush_remote_tlbs(kvm);
4247 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4248 struct kvm_enable_cap *cap)
4253 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4254 struct kvm_enable_cap *cap)
4257 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4258 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4259 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4261 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4262 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4264 if (cap->flags || (cap->args[0] & ~allowed_options))
4266 kvm->manual_dirty_log_protect = cap->args[0];
4270 case KVM_CAP_HALT_POLL: {
4271 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4274 kvm->max_halt_poll_ns = cap->args[0];
4277 case KVM_CAP_DIRTY_LOG_RING:
4278 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4280 return kvm_vm_ioctl_enable_cap(kvm, cap);
4284 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4285 size_t size, loff_t *offset)
4287 struct kvm *kvm = file->private_data;
4289 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4290 &kvm_vm_stats_desc[0], &kvm->stat,
4291 sizeof(kvm->stat), user_buffer, size, offset);
4294 static const struct file_operations kvm_vm_stats_fops = {
4295 .read = kvm_vm_stats_read,
4296 .llseek = noop_llseek,
4299 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4304 fd = get_unused_fd_flags(O_CLOEXEC);
4308 file = anon_inode_getfile("kvm-vm-stats",
4309 &kvm_vm_stats_fops, kvm, O_RDONLY);
4312 return PTR_ERR(file);
4314 file->f_mode |= FMODE_PREAD;
4315 fd_install(fd, file);
4320 static long kvm_vm_ioctl(struct file *filp,
4321 unsigned int ioctl, unsigned long arg)
4323 struct kvm *kvm = filp->private_data;
4324 void __user *argp = (void __user *)arg;
4327 if (kvm->mm != current->mm || kvm->vm_bugged)
4330 case KVM_CREATE_VCPU:
4331 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4333 case KVM_ENABLE_CAP: {
4334 struct kvm_enable_cap cap;
4337 if (copy_from_user(&cap, argp, sizeof(cap)))
4339 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4342 case KVM_SET_USER_MEMORY_REGION: {
4343 struct kvm_userspace_memory_region kvm_userspace_mem;
4346 if (copy_from_user(&kvm_userspace_mem, argp,
4347 sizeof(kvm_userspace_mem)))
4350 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4353 case KVM_GET_DIRTY_LOG: {
4354 struct kvm_dirty_log log;
4357 if (copy_from_user(&log, argp, sizeof(log)))
4359 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4362 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4363 case KVM_CLEAR_DIRTY_LOG: {
4364 struct kvm_clear_dirty_log log;
4367 if (copy_from_user(&log, argp, sizeof(log)))
4369 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4373 #ifdef CONFIG_KVM_MMIO
4374 case KVM_REGISTER_COALESCED_MMIO: {
4375 struct kvm_coalesced_mmio_zone zone;
4378 if (copy_from_user(&zone, argp, sizeof(zone)))
4380 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4383 case KVM_UNREGISTER_COALESCED_MMIO: {
4384 struct kvm_coalesced_mmio_zone zone;
4387 if (copy_from_user(&zone, argp, sizeof(zone)))
4389 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4394 struct kvm_irqfd data;
4397 if (copy_from_user(&data, argp, sizeof(data)))
4399 r = kvm_irqfd(kvm, &data);
4402 case KVM_IOEVENTFD: {
4403 struct kvm_ioeventfd data;
4406 if (copy_from_user(&data, argp, sizeof(data)))
4408 r = kvm_ioeventfd(kvm, &data);
4411 #ifdef CONFIG_HAVE_KVM_MSI
4412 case KVM_SIGNAL_MSI: {
4416 if (copy_from_user(&msi, argp, sizeof(msi)))
4418 r = kvm_send_userspace_msi(kvm, &msi);
4422 #ifdef __KVM_HAVE_IRQ_LINE
4423 case KVM_IRQ_LINE_STATUS:
4424 case KVM_IRQ_LINE: {
4425 struct kvm_irq_level irq_event;
4428 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4431 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4432 ioctl == KVM_IRQ_LINE_STATUS);
4437 if (ioctl == KVM_IRQ_LINE_STATUS) {
4438 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4446 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4447 case KVM_SET_GSI_ROUTING: {
4448 struct kvm_irq_routing routing;
4449 struct kvm_irq_routing __user *urouting;
4450 struct kvm_irq_routing_entry *entries = NULL;
4453 if (copy_from_user(&routing, argp, sizeof(routing)))
4456 if (!kvm_arch_can_set_irq_routing(kvm))
4458 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4464 entries = vmemdup_user(urouting->entries,
4465 array_size(sizeof(*entries),
4467 if (IS_ERR(entries)) {
4468 r = PTR_ERR(entries);
4472 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4477 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4478 case KVM_CREATE_DEVICE: {
4479 struct kvm_create_device cd;
4482 if (copy_from_user(&cd, argp, sizeof(cd)))
4485 r = kvm_ioctl_create_device(kvm, &cd);
4490 if (copy_to_user(argp, &cd, sizeof(cd)))
4496 case KVM_CHECK_EXTENSION:
4497 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4499 case KVM_RESET_DIRTY_RINGS:
4500 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4502 case KVM_GET_STATS_FD:
4503 r = kvm_vm_ioctl_get_stats_fd(kvm);
4506 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4512 #ifdef CONFIG_KVM_COMPAT
4513 struct compat_kvm_dirty_log {
4517 compat_uptr_t dirty_bitmap; /* one bit per page */
4522 struct compat_kvm_clear_dirty_log {
4527 compat_uptr_t dirty_bitmap; /* one bit per page */
4532 static long kvm_vm_compat_ioctl(struct file *filp,
4533 unsigned int ioctl, unsigned long arg)
4535 struct kvm *kvm = filp->private_data;
4538 if (kvm->mm != current->mm || kvm->vm_bugged)
4541 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4542 case KVM_CLEAR_DIRTY_LOG: {
4543 struct compat_kvm_clear_dirty_log compat_log;
4544 struct kvm_clear_dirty_log log;
4546 if (copy_from_user(&compat_log, (void __user *)arg,
4547 sizeof(compat_log)))
4549 log.slot = compat_log.slot;
4550 log.num_pages = compat_log.num_pages;
4551 log.first_page = compat_log.first_page;
4552 log.padding2 = compat_log.padding2;
4553 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4555 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4559 case KVM_GET_DIRTY_LOG: {
4560 struct compat_kvm_dirty_log compat_log;
4561 struct kvm_dirty_log log;
4563 if (copy_from_user(&compat_log, (void __user *)arg,
4564 sizeof(compat_log)))
4566 log.slot = compat_log.slot;
4567 log.padding1 = compat_log.padding1;
4568 log.padding2 = compat_log.padding2;
4569 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4571 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4575 r = kvm_vm_ioctl(filp, ioctl, arg);
4581 static struct file_operations kvm_vm_fops = {
4582 .release = kvm_vm_release,
4583 .unlocked_ioctl = kvm_vm_ioctl,
4584 .llseek = noop_llseek,
4585 KVM_COMPAT(kvm_vm_compat_ioctl),
4588 bool file_is_kvm(struct file *file)
4590 return file && file->f_op == &kvm_vm_fops;
4592 EXPORT_SYMBOL_GPL(file_is_kvm);
4594 static int kvm_dev_ioctl_create_vm(unsigned long type)
4600 kvm = kvm_create_vm(type);
4602 return PTR_ERR(kvm);
4603 #ifdef CONFIG_KVM_MMIO
4604 r = kvm_coalesced_mmio_init(kvm);
4608 r = get_unused_fd_flags(O_CLOEXEC);
4612 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4613 "kvm-%d", task_pid_nr(current));
4615 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4623 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4624 * already set, with ->release() being kvm_vm_release(). In error
4625 * cases it will be called by the final fput(file) and will take
4626 * care of doing kvm_put_kvm(kvm).
4628 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4633 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4635 fd_install(r, file);
4643 static long kvm_dev_ioctl(struct file *filp,
4644 unsigned int ioctl, unsigned long arg)
4649 case KVM_GET_API_VERSION:
4652 r = KVM_API_VERSION;
4655 r = kvm_dev_ioctl_create_vm(arg);
4657 case KVM_CHECK_EXTENSION:
4658 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4660 case KVM_GET_VCPU_MMAP_SIZE:
4663 r = PAGE_SIZE; /* struct kvm_run */
4665 r += PAGE_SIZE; /* pio data page */
4667 #ifdef CONFIG_KVM_MMIO
4668 r += PAGE_SIZE; /* coalesced mmio ring page */
4671 case KVM_TRACE_ENABLE:
4672 case KVM_TRACE_PAUSE:
4673 case KVM_TRACE_DISABLE:
4677 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4683 static struct file_operations kvm_chardev_ops = {
4684 .unlocked_ioctl = kvm_dev_ioctl,
4685 .llseek = noop_llseek,
4686 KVM_COMPAT(kvm_dev_ioctl),
4689 static struct miscdevice kvm_dev = {
4695 static void hardware_enable_nolock(void *junk)
4697 int cpu = raw_smp_processor_id();
4700 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4703 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4705 r = kvm_arch_hardware_enable();
4708 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4709 atomic_inc(&hardware_enable_failed);
4710 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4714 static int kvm_starting_cpu(unsigned int cpu)
4716 raw_spin_lock(&kvm_count_lock);
4717 if (kvm_usage_count)
4718 hardware_enable_nolock(NULL);
4719 raw_spin_unlock(&kvm_count_lock);
4723 static void hardware_disable_nolock(void *junk)
4725 int cpu = raw_smp_processor_id();
4727 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4729 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4730 kvm_arch_hardware_disable();
4733 static int kvm_dying_cpu(unsigned int cpu)
4735 raw_spin_lock(&kvm_count_lock);
4736 if (kvm_usage_count)
4737 hardware_disable_nolock(NULL);
4738 raw_spin_unlock(&kvm_count_lock);
4742 static void hardware_disable_all_nolock(void)
4744 BUG_ON(!kvm_usage_count);
4747 if (!kvm_usage_count)
4748 on_each_cpu(hardware_disable_nolock, NULL, 1);
4751 static void hardware_disable_all(void)
4753 raw_spin_lock(&kvm_count_lock);
4754 hardware_disable_all_nolock();
4755 raw_spin_unlock(&kvm_count_lock);
4758 static int hardware_enable_all(void)
4762 raw_spin_lock(&kvm_count_lock);
4765 if (kvm_usage_count == 1) {
4766 atomic_set(&hardware_enable_failed, 0);
4767 on_each_cpu(hardware_enable_nolock, NULL, 1);
4769 if (atomic_read(&hardware_enable_failed)) {
4770 hardware_disable_all_nolock();
4775 raw_spin_unlock(&kvm_count_lock);
4780 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4784 * Some (well, at least mine) BIOSes hang on reboot if
4787 * And Intel TXT required VMX off for all cpu when system shutdown.
4789 pr_info("kvm: exiting hardware virtualization\n");
4790 kvm_rebooting = true;
4791 on_each_cpu(hardware_disable_nolock, NULL, 1);
4795 static struct notifier_block kvm_reboot_notifier = {
4796 .notifier_call = kvm_reboot,
4800 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4804 for (i = 0; i < bus->dev_count; i++) {
4805 struct kvm_io_device *pos = bus->range[i].dev;
4807 kvm_iodevice_destructor(pos);
4812 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4813 const struct kvm_io_range *r2)
4815 gpa_t addr1 = r1->addr;
4816 gpa_t addr2 = r2->addr;
4821 /* If r2->len == 0, match the exact address. If r2->len != 0,
4822 * accept any overlapping write. Any order is acceptable for
4823 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4824 * we process all of them.
4837 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4839 return kvm_io_bus_cmp(p1, p2);
4842 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4843 gpa_t addr, int len)
4845 struct kvm_io_range *range, key;
4848 key = (struct kvm_io_range) {
4853 range = bsearch(&key, bus->range, bus->dev_count,
4854 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4858 off = range - bus->range;
4860 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4866 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4867 struct kvm_io_range *range, const void *val)
4871 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4875 while (idx < bus->dev_count &&
4876 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4877 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4886 /* kvm_io_bus_write - called under kvm->slots_lock */
4887 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4888 int len, const void *val)
4890 struct kvm_io_bus *bus;
4891 struct kvm_io_range range;
4894 range = (struct kvm_io_range) {
4899 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4902 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4903 return r < 0 ? r : 0;
4905 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4907 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4908 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4909 gpa_t addr, int len, const void *val, long cookie)
4911 struct kvm_io_bus *bus;
4912 struct kvm_io_range range;
4914 range = (struct kvm_io_range) {
4919 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4923 /* First try the device referenced by cookie. */
4924 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4925 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4926 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4931 * cookie contained garbage; fall back to search and return the
4932 * correct cookie value.
4934 return __kvm_io_bus_write(vcpu, bus, &range, val);
4937 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4938 struct kvm_io_range *range, void *val)
4942 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4946 while (idx < bus->dev_count &&
4947 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4948 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4957 /* kvm_io_bus_read - called under kvm->slots_lock */
4958 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4961 struct kvm_io_bus *bus;
4962 struct kvm_io_range range;
4965 range = (struct kvm_io_range) {
4970 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4973 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4974 return r < 0 ? r : 0;
4977 /* Caller must hold slots_lock. */
4978 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4979 int len, struct kvm_io_device *dev)
4982 struct kvm_io_bus *new_bus, *bus;
4983 struct kvm_io_range range;
4985 bus = kvm_get_bus(kvm, bus_idx);
4989 /* exclude ioeventfd which is limited by maximum fd */
4990 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4993 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4994 GFP_KERNEL_ACCOUNT);
4998 range = (struct kvm_io_range) {
5004 for (i = 0; i < bus->dev_count; i++)
5005 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5008 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5009 new_bus->dev_count++;
5010 new_bus->range[i] = range;
5011 memcpy(new_bus->range + i + 1, bus->range + i,
5012 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5013 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5014 synchronize_srcu_expedited(&kvm->srcu);
5020 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5021 struct kvm_io_device *dev)
5024 struct kvm_io_bus *new_bus, *bus;
5026 lockdep_assert_held(&kvm->slots_lock);
5028 bus = kvm_get_bus(kvm, bus_idx);
5032 for (i = 0; i < bus->dev_count; i++) {
5033 if (bus->range[i].dev == dev) {
5038 if (i == bus->dev_count)
5041 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5042 GFP_KERNEL_ACCOUNT);
5044 memcpy(new_bus, bus, struct_size(bus, range, i));
5045 new_bus->dev_count--;
5046 memcpy(new_bus->range + i, bus->range + i + 1,
5047 flex_array_size(new_bus, range, new_bus->dev_count - i));
5050 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5051 synchronize_srcu_expedited(&kvm->srcu);
5053 /* Destroy the old bus _after_ installing the (null) bus. */
5055 pr_err("kvm: failed to shrink bus, removing it completely\n");
5056 for (j = 0; j < bus->dev_count; j++) {
5059 kvm_iodevice_destructor(bus->range[j].dev);
5064 return new_bus ? 0 : -ENOMEM;
5067 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5070 struct kvm_io_bus *bus;
5071 int dev_idx, srcu_idx;
5072 struct kvm_io_device *iodev = NULL;
5074 srcu_idx = srcu_read_lock(&kvm->srcu);
5076 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5080 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5084 iodev = bus->range[dev_idx].dev;
5087 srcu_read_unlock(&kvm->srcu, srcu_idx);
5091 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5093 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5094 int (*get)(void *, u64 *), int (*set)(void *, u64),
5097 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5101 * The debugfs files are a reference to the kvm struct which
5102 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5103 * avoids the race between open and the removal of the debugfs directory.
5105 if (!kvm_get_kvm_safe(stat_data->kvm))
5108 if (simple_attr_open(inode, file, get,
5109 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5112 kvm_put_kvm(stat_data->kvm);
5119 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5121 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5124 simple_attr_release(inode, file);
5125 kvm_put_kvm(stat_data->kvm);
5130 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5132 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5137 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5139 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5144 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5147 struct kvm_vcpu *vcpu;
5151 kvm_for_each_vcpu(i, vcpu, kvm)
5152 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5157 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5160 struct kvm_vcpu *vcpu;
5162 kvm_for_each_vcpu(i, vcpu, kvm)
5163 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5168 static int kvm_stat_data_get(void *data, u64 *val)
5171 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5173 switch (stat_data->kind) {
5175 r = kvm_get_stat_per_vm(stat_data->kvm,
5176 stat_data->desc->desc.offset, val);
5179 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5180 stat_data->desc->desc.offset, val);
5187 static int kvm_stat_data_clear(void *data, u64 val)
5190 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5195 switch (stat_data->kind) {
5197 r = kvm_clear_stat_per_vm(stat_data->kvm,
5198 stat_data->desc->desc.offset);
5201 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5202 stat_data->desc->desc.offset);
5209 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5211 __simple_attr_check_format("%llu\n", 0ull);
5212 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5213 kvm_stat_data_clear, "%llu\n");
5216 static const struct file_operations stat_fops_per_vm = {
5217 .owner = THIS_MODULE,
5218 .open = kvm_stat_data_open,
5219 .release = kvm_debugfs_release,
5220 .read = simple_attr_read,
5221 .write = simple_attr_write,
5222 .llseek = no_llseek,
5225 static int vm_stat_get(void *_offset, u64 *val)
5227 unsigned offset = (long)_offset;
5232 mutex_lock(&kvm_lock);
5233 list_for_each_entry(kvm, &vm_list, vm_list) {
5234 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5237 mutex_unlock(&kvm_lock);
5241 static int vm_stat_clear(void *_offset, u64 val)
5243 unsigned offset = (long)_offset;
5249 mutex_lock(&kvm_lock);
5250 list_for_each_entry(kvm, &vm_list, vm_list) {
5251 kvm_clear_stat_per_vm(kvm, offset);
5253 mutex_unlock(&kvm_lock);
5258 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5259 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5261 static int vcpu_stat_get(void *_offset, u64 *val)
5263 unsigned offset = (long)_offset;
5268 mutex_lock(&kvm_lock);
5269 list_for_each_entry(kvm, &vm_list, vm_list) {
5270 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5273 mutex_unlock(&kvm_lock);
5277 static int vcpu_stat_clear(void *_offset, u64 val)
5279 unsigned offset = (long)_offset;
5285 mutex_lock(&kvm_lock);
5286 list_for_each_entry(kvm, &vm_list, vm_list) {
5287 kvm_clear_stat_per_vcpu(kvm, offset);
5289 mutex_unlock(&kvm_lock);
5294 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5296 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5298 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5300 struct kobj_uevent_env *env;
5301 unsigned long long created, active;
5303 if (!kvm_dev.this_device || !kvm)
5306 mutex_lock(&kvm_lock);
5307 if (type == KVM_EVENT_CREATE_VM) {
5308 kvm_createvm_count++;
5310 } else if (type == KVM_EVENT_DESTROY_VM) {
5313 created = kvm_createvm_count;
5314 active = kvm_active_vms;
5315 mutex_unlock(&kvm_lock);
5317 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5321 add_uevent_var(env, "CREATED=%llu", created);
5322 add_uevent_var(env, "COUNT=%llu", active);
5324 if (type == KVM_EVENT_CREATE_VM) {
5325 add_uevent_var(env, "EVENT=create");
5326 kvm->userspace_pid = task_pid_nr(current);
5327 } else if (type == KVM_EVENT_DESTROY_VM) {
5328 add_uevent_var(env, "EVENT=destroy");
5330 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5332 if (kvm->debugfs_dentry) {
5333 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5336 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5338 add_uevent_var(env, "STATS_PATH=%s", tmp);
5342 /* no need for checks, since we are adding at most only 5 keys */
5343 env->envp[env->envp_idx++] = NULL;
5344 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5348 static void kvm_init_debug(void)
5350 const struct file_operations *fops;
5351 const struct _kvm_stats_desc *pdesc;
5354 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5356 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5357 pdesc = &kvm_vm_stats_desc[i];
5358 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5359 fops = &vm_stat_fops;
5361 fops = &vm_stat_readonly_fops;
5362 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5364 (void *)(long)pdesc->desc.offset, fops);
5367 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5368 pdesc = &kvm_vcpu_stats_desc[i];
5369 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5370 fops = &vcpu_stat_fops;
5372 fops = &vcpu_stat_readonly_fops;
5373 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5375 (void *)(long)pdesc->desc.offset, fops);
5379 static int kvm_suspend(void)
5381 if (kvm_usage_count)
5382 hardware_disable_nolock(NULL);
5386 static void kvm_resume(void)
5388 if (kvm_usage_count) {
5389 #ifdef CONFIG_LOCKDEP
5390 WARN_ON(lockdep_is_held(&kvm_count_lock));
5392 hardware_enable_nolock(NULL);
5396 static struct syscore_ops kvm_syscore_ops = {
5397 .suspend = kvm_suspend,
5398 .resume = kvm_resume,
5402 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5404 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5407 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5409 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5411 WRITE_ONCE(vcpu->preempted, false);
5412 WRITE_ONCE(vcpu->ready, false);
5414 __this_cpu_write(kvm_running_vcpu, vcpu);
5415 kvm_arch_sched_in(vcpu, cpu);
5416 kvm_arch_vcpu_load(vcpu, cpu);
5419 static void kvm_sched_out(struct preempt_notifier *pn,
5420 struct task_struct *next)
5422 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5424 if (current->on_rq) {
5425 WRITE_ONCE(vcpu->preempted, true);
5426 WRITE_ONCE(vcpu->ready, true);
5428 kvm_arch_vcpu_put(vcpu);
5429 __this_cpu_write(kvm_running_vcpu, NULL);
5433 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5435 * We can disable preemption locally around accessing the per-CPU variable,
5436 * and use the resolved vcpu pointer after enabling preemption again,
5437 * because even if the current thread is migrated to another CPU, reading
5438 * the per-CPU value later will give us the same value as we update the
5439 * per-CPU variable in the preempt notifier handlers.
5441 struct kvm_vcpu *kvm_get_running_vcpu(void)
5443 struct kvm_vcpu *vcpu;
5446 vcpu = __this_cpu_read(kvm_running_vcpu);
5451 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5454 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5456 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5458 return &kvm_running_vcpu;
5461 struct kvm_cpu_compat_check {
5466 static void check_processor_compat(void *data)
5468 struct kvm_cpu_compat_check *c = data;
5470 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5473 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5474 struct module *module)
5476 struct kvm_cpu_compat_check c;
5480 r = kvm_arch_init(opaque);
5485 * kvm_arch_init makes sure there's at most one caller
5486 * for architectures that support multiple implementations,
5487 * like intel and amd on x86.
5488 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5489 * conflicts in case kvm is already setup for another implementation.
5491 r = kvm_irqfd_init();
5495 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5500 r = kvm_arch_hardware_setup(opaque);
5506 for_each_online_cpu(cpu) {
5507 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5512 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5513 kvm_starting_cpu, kvm_dying_cpu);
5516 register_reboot_notifier(&kvm_reboot_notifier);
5518 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5520 vcpu_align = __alignof__(struct kvm_vcpu);
5522 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5524 offsetof(struct kvm_vcpu, arch),
5525 offsetofend(struct kvm_vcpu, stats_id)
5526 - offsetof(struct kvm_vcpu, arch),
5528 if (!kvm_vcpu_cache) {
5533 r = kvm_async_pf_init();
5537 kvm_chardev_ops.owner = module;
5538 kvm_vm_fops.owner = module;
5539 kvm_vcpu_fops.owner = module;
5541 r = misc_register(&kvm_dev);
5543 pr_err("kvm: misc device register failed\n");
5547 register_syscore_ops(&kvm_syscore_ops);
5549 kvm_preempt_ops.sched_in = kvm_sched_in;
5550 kvm_preempt_ops.sched_out = kvm_sched_out;
5554 r = kvm_vfio_ops_init();
5560 kvm_async_pf_deinit();
5562 kmem_cache_destroy(kvm_vcpu_cache);
5564 unregister_reboot_notifier(&kvm_reboot_notifier);
5565 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5567 kvm_arch_hardware_unsetup();
5569 free_cpumask_var(cpus_hardware_enabled);
5577 EXPORT_SYMBOL_GPL(kvm_init);
5581 debugfs_remove_recursive(kvm_debugfs_dir);
5582 misc_deregister(&kvm_dev);
5583 kmem_cache_destroy(kvm_vcpu_cache);
5584 kvm_async_pf_deinit();
5585 unregister_syscore_ops(&kvm_syscore_ops);
5586 unregister_reboot_notifier(&kvm_reboot_notifier);
5587 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5588 on_each_cpu(hardware_disable_nolock, NULL, 1);
5589 kvm_arch_hardware_unsetup();
5592 free_cpumask_var(cpus_hardware_enabled);
5593 kvm_vfio_ops_exit();
5595 EXPORT_SYMBOL_GPL(kvm_exit);
5597 struct kvm_vm_worker_thread_context {
5599 struct task_struct *parent;
5600 struct completion init_done;
5601 kvm_vm_thread_fn_t thread_fn;
5606 static int kvm_vm_worker_thread(void *context)
5609 * The init_context is allocated on the stack of the parent thread, so
5610 * we have to locally copy anything that is needed beyond initialization
5612 struct kvm_vm_worker_thread_context *init_context = context;
5613 struct kvm *kvm = init_context->kvm;
5614 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5615 uintptr_t data = init_context->data;
5618 err = kthread_park(current);
5619 /* kthread_park(current) is never supposed to return an error */
5624 err = cgroup_attach_task_all(init_context->parent, current);
5626 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5631 set_user_nice(current, task_nice(init_context->parent));
5634 init_context->err = err;
5635 complete(&init_context->init_done);
5636 init_context = NULL;
5641 /* Wait to be woken up by the spawner before proceeding. */
5644 if (!kthread_should_stop())
5645 err = thread_fn(kvm, data);
5650 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5651 uintptr_t data, const char *name,
5652 struct task_struct **thread_ptr)
5654 struct kvm_vm_worker_thread_context init_context = {};
5655 struct task_struct *thread;
5658 init_context.kvm = kvm;
5659 init_context.parent = current;
5660 init_context.thread_fn = thread_fn;
5661 init_context.data = data;
5662 init_completion(&init_context.init_done);
5664 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5665 "%s-%d", name, task_pid_nr(current));
5667 return PTR_ERR(thread);
5669 /* kthread_run is never supposed to return NULL */
5670 WARN_ON(thread == NULL);
5672 wait_for_completion(&init_context.init_done);
5674 if (!init_context.err)
5675 *thread_ptr = thread;
5677 return init_context.err;