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);
322 * We want to publish modifications to the page tables before reading
323 * mode. Pairs with a memory barrier in arch-specific code.
324 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
325 * and smp_mb in walk_shadow_page_lockless_begin/end.
326 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
328 * There is already an smp_mb__after_atomic() before
329 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
332 if (!kvm_arch_flush_remote_tlb(kvm)
333 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
334 ++kvm->stat.generic.remote_tlb_flush;
335 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
337 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
340 void kvm_reload_remote_mmus(struct kvm *kvm)
342 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
345 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
346 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
349 gfp_flags |= mc->gfp_zero;
352 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
354 return (void *)__get_free_page(gfp_flags);
357 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
361 if (mc->nobjs >= min)
363 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
364 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
366 return mc->nobjs >= min ? 0 : -ENOMEM;
367 mc->objects[mc->nobjs++] = obj;
372 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
377 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
381 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
383 free_page((unsigned long)mc->objects[--mc->nobjs]);
387 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
391 if (WARN_ON(!mc->nobjs))
392 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
394 p = mc->objects[--mc->nobjs];
400 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
402 mutex_init(&vcpu->mutex);
407 rcuwait_init(&vcpu->wait);
408 kvm_async_pf_vcpu_init(vcpu);
411 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
413 kvm_vcpu_set_in_spin_loop(vcpu, false);
414 kvm_vcpu_set_dy_eligible(vcpu, false);
415 vcpu->preempted = false;
417 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
420 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
422 kvm_dirty_ring_free(&vcpu->dirty_ring);
423 kvm_arch_vcpu_destroy(vcpu);
426 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
427 * the vcpu->pid pointer, and at destruction time all file descriptors
430 put_pid(rcu_dereference_protected(vcpu->pid, 1));
432 free_page((unsigned long)vcpu->run);
433 kmem_cache_free(kvm_vcpu_cache, vcpu);
435 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
437 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
438 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
440 return container_of(mn, struct kvm, mmu_notifier);
443 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
444 struct mm_struct *mm,
445 unsigned long start, unsigned long end)
447 struct kvm *kvm = mmu_notifier_to_kvm(mn);
450 idx = srcu_read_lock(&kvm->srcu);
451 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
452 srcu_read_unlock(&kvm->srcu, idx);
455 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
457 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
460 struct kvm_hva_range {
464 hva_handler_t handler;
465 on_lock_fn_t on_lock;
471 * Use a dedicated stub instead of NULL to indicate that there is no callback
472 * function/handler. The compiler technically can't guarantee that a real
473 * function will have a non-zero address, and so it will generate code to
474 * check for !NULL, whereas comparing against a stub will be elided at compile
475 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
477 static void kvm_null_fn(void)
481 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
483 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
484 const struct kvm_hva_range *range)
486 bool ret = false, locked = false;
487 struct kvm_gfn_range gfn_range;
488 struct kvm_memory_slot *slot;
489 struct kvm_memslots *slots;
492 /* A null handler is allowed if and only if on_lock() is provided. */
493 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
494 IS_KVM_NULL_FN(range->handler)))
497 idx = srcu_read_lock(&kvm->srcu);
499 /* The on_lock() path does not yet support lock elision. */
500 if (!IS_KVM_NULL_FN(range->on_lock)) {
504 range->on_lock(kvm, range->start, range->end);
506 if (IS_KVM_NULL_FN(range->handler))
510 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
511 slots = __kvm_memslots(kvm, i);
512 kvm_for_each_memslot(slot, slots) {
513 unsigned long hva_start, hva_end;
515 hva_start = max(range->start, slot->userspace_addr);
516 hva_end = min(range->end, slot->userspace_addr +
517 (slot->npages << PAGE_SHIFT));
518 if (hva_start >= hva_end)
522 * To optimize for the likely case where the address
523 * range is covered by zero or one memslots, don't
524 * bother making these conditional (to avoid writes on
525 * the second or later invocation of the handler).
527 gfn_range.pte = range->pte;
528 gfn_range.may_block = range->may_block;
531 * {gfn(page) | page intersects with [hva_start, hva_end)} =
532 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
534 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
535 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
536 gfn_range.slot = slot;
542 ret |= range->handler(kvm, &gfn_range);
546 if (range->flush_on_ret && (ret || kvm->tlbs_dirty))
547 kvm_flush_remote_tlbs(kvm);
553 srcu_read_unlock(&kvm->srcu, idx);
555 /* The notifiers are averse to booleans. :-( */
559 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
563 hva_handler_t handler)
565 struct kvm *kvm = mmu_notifier_to_kvm(mn);
566 const struct kvm_hva_range range = {
571 .on_lock = (void *)kvm_null_fn,
572 .flush_on_ret = true,
576 return __kvm_handle_hva_range(kvm, &range);
579 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
582 hva_handler_t handler)
584 struct kvm *kvm = mmu_notifier_to_kvm(mn);
585 const struct kvm_hva_range range = {
590 .on_lock = (void *)kvm_null_fn,
591 .flush_on_ret = false,
595 return __kvm_handle_hva_range(kvm, &range);
597 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
598 struct mm_struct *mm,
599 unsigned long address,
602 struct kvm *kvm = mmu_notifier_to_kvm(mn);
604 trace_kvm_set_spte_hva(address);
607 * .change_pte() must be surrounded by .invalidate_range_{start,end}(),
608 * and so always runs with an elevated notifier count. This obviates
609 * the need to bump the sequence count.
611 WARN_ON_ONCE(!kvm->mmu_notifier_count);
613 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
616 static void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
620 * The count increase must become visible at unlock time as no
621 * spte can be established without taking the mmu_lock and
622 * count is also read inside the mmu_lock critical section.
624 kvm->mmu_notifier_count++;
625 if (likely(kvm->mmu_notifier_count == 1)) {
626 kvm->mmu_notifier_range_start = start;
627 kvm->mmu_notifier_range_end = end;
630 * Fully tracking multiple concurrent ranges has dimishing
631 * returns. Keep things simple and just find the minimal range
632 * which includes the current and new ranges. As there won't be
633 * enough information to subtract a range after its invalidate
634 * completes, any ranges invalidated concurrently will
635 * accumulate and persist until all outstanding invalidates
638 kvm->mmu_notifier_range_start =
639 min(kvm->mmu_notifier_range_start, start);
640 kvm->mmu_notifier_range_end =
641 max(kvm->mmu_notifier_range_end, end);
645 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
646 const struct mmu_notifier_range *range)
648 struct kvm *kvm = mmu_notifier_to_kvm(mn);
649 const struct kvm_hva_range hva_range = {
650 .start = range->start,
653 .handler = kvm_unmap_gfn_range,
654 .on_lock = kvm_inc_notifier_count,
655 .flush_on_ret = true,
656 .may_block = mmu_notifier_range_blockable(range),
659 trace_kvm_unmap_hva_range(range->start, range->end);
661 __kvm_handle_hva_range(kvm, &hva_range);
666 static void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
670 * This sequence increase will notify the kvm page fault that
671 * the page that is going to be mapped in the spte could have
674 kvm->mmu_notifier_seq++;
677 * The above sequence increase must be visible before the
678 * below count decrease, which is ensured by the smp_wmb above
679 * in conjunction with the smp_rmb in mmu_notifier_retry().
681 kvm->mmu_notifier_count--;
684 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
685 const struct mmu_notifier_range *range)
687 struct kvm *kvm = mmu_notifier_to_kvm(mn);
688 const struct kvm_hva_range hva_range = {
689 .start = range->start,
692 .handler = (void *)kvm_null_fn,
693 .on_lock = kvm_dec_notifier_count,
694 .flush_on_ret = false,
695 .may_block = mmu_notifier_range_blockable(range),
698 __kvm_handle_hva_range(kvm, &hva_range);
700 BUG_ON(kvm->mmu_notifier_count < 0);
703 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
704 struct mm_struct *mm,
708 trace_kvm_age_hva(start, end);
710 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
713 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
714 struct mm_struct *mm,
718 trace_kvm_age_hva(start, end);
721 * Even though we do not flush TLB, this will still adversely
722 * affect performance on pre-Haswell Intel EPT, where there is
723 * no EPT Access Bit to clear so that we have to tear down EPT
724 * tables instead. If we find this unacceptable, we can always
725 * add a parameter to kvm_age_hva so that it effectively doesn't
726 * do anything on clear_young.
728 * Also note that currently we never issue secondary TLB flushes
729 * from clear_young, leaving this job up to the regular system
730 * cadence. If we find this inaccurate, we might come up with a
731 * more sophisticated heuristic later.
733 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
736 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
737 struct mm_struct *mm,
738 unsigned long address)
740 trace_kvm_test_age_hva(address);
742 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
746 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
747 struct mm_struct *mm)
749 struct kvm *kvm = mmu_notifier_to_kvm(mn);
752 idx = srcu_read_lock(&kvm->srcu);
753 kvm_arch_flush_shadow_all(kvm);
754 srcu_read_unlock(&kvm->srcu, idx);
757 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
758 .invalidate_range = kvm_mmu_notifier_invalidate_range,
759 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
760 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
761 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
762 .clear_young = kvm_mmu_notifier_clear_young,
763 .test_young = kvm_mmu_notifier_test_young,
764 .change_pte = kvm_mmu_notifier_change_pte,
765 .release = kvm_mmu_notifier_release,
768 static int kvm_init_mmu_notifier(struct kvm *kvm)
770 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
771 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
774 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
776 static int kvm_init_mmu_notifier(struct kvm *kvm)
781 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
783 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
784 static int kvm_pm_notifier_call(struct notifier_block *bl,
788 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
790 return kvm_arch_pm_notifier(kvm, state);
793 static void kvm_init_pm_notifier(struct kvm *kvm)
795 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
796 /* Suspend KVM before we suspend ftrace, RCU, etc. */
797 kvm->pm_notifier.priority = INT_MAX;
798 register_pm_notifier(&kvm->pm_notifier);
801 static void kvm_destroy_pm_notifier(struct kvm *kvm)
803 unregister_pm_notifier(&kvm->pm_notifier);
805 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
806 static void kvm_init_pm_notifier(struct kvm *kvm)
810 static void kvm_destroy_pm_notifier(struct kvm *kvm)
813 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
815 static struct kvm_memslots *kvm_alloc_memslots(void)
818 struct kvm_memslots *slots;
820 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
824 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
825 slots->id_to_index[i] = -1;
830 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
832 if (!memslot->dirty_bitmap)
835 kvfree(memslot->dirty_bitmap);
836 memslot->dirty_bitmap = NULL;
839 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
841 kvm_destroy_dirty_bitmap(slot);
843 kvm_arch_free_memslot(kvm, slot);
849 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
851 struct kvm_memory_slot *memslot;
856 kvm_for_each_memslot(memslot, slots)
857 kvm_free_memslot(kvm, memslot);
862 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
864 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
865 case KVM_STATS_TYPE_INSTANT:
867 case KVM_STATS_TYPE_CUMULATIVE:
868 case KVM_STATS_TYPE_PEAK:
875 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
878 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
879 kvm_vcpu_stats_header.num_desc;
881 if (!kvm->debugfs_dentry)
884 debugfs_remove_recursive(kvm->debugfs_dentry);
886 if (kvm->debugfs_stat_data) {
887 for (i = 0; i < kvm_debugfs_num_entries; i++)
888 kfree(kvm->debugfs_stat_data[i]);
889 kfree(kvm->debugfs_stat_data);
893 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
895 static DEFINE_MUTEX(kvm_debugfs_lock);
897 char dir_name[ITOA_MAX_LEN * 2];
898 struct kvm_stat_data *stat_data;
899 const struct _kvm_stats_desc *pdesc;
901 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
902 kvm_vcpu_stats_header.num_desc;
904 if (!debugfs_initialized())
907 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
908 mutex_lock(&kvm_debugfs_lock);
909 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
911 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
913 mutex_unlock(&kvm_debugfs_lock);
916 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
917 mutex_unlock(&kvm_debugfs_lock);
921 kvm->debugfs_dentry = dent;
922 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
923 sizeof(*kvm->debugfs_stat_data),
925 if (!kvm->debugfs_stat_data)
928 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
929 pdesc = &kvm_vm_stats_desc[i];
930 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
934 stat_data->kvm = kvm;
935 stat_data->desc = pdesc;
936 stat_data->kind = KVM_STAT_VM;
937 kvm->debugfs_stat_data[i] = stat_data;
938 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
939 kvm->debugfs_dentry, stat_data,
943 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
944 pdesc = &kvm_vcpu_stats_desc[i];
945 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
949 stat_data->kvm = kvm;
950 stat_data->desc = pdesc;
951 stat_data->kind = KVM_STAT_VCPU;
952 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
953 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
954 kvm->debugfs_dentry, stat_data,
961 * Called after the VM is otherwise initialized, but just before adding it to
964 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
970 * Called just after removing the VM from the vm_list, but before doing any
973 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
977 static struct kvm *kvm_create_vm(unsigned long type)
979 struct kvm *kvm = kvm_arch_alloc_vm();
984 return ERR_PTR(-ENOMEM);
986 KVM_MMU_LOCK_INIT(kvm);
988 kvm->mm = current->mm;
989 kvm_eventfd_init(kvm);
990 mutex_init(&kvm->lock);
991 mutex_init(&kvm->irq_lock);
992 mutex_init(&kvm->slots_lock);
993 mutex_init(&kvm->slots_arch_lock);
994 INIT_LIST_HEAD(&kvm->devices);
996 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
998 if (init_srcu_struct(&kvm->srcu))
999 goto out_err_no_srcu;
1000 if (init_srcu_struct(&kvm->irq_srcu))
1001 goto out_err_no_irq_srcu;
1003 refcount_set(&kvm->users_count, 1);
1004 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1005 struct kvm_memslots *slots = kvm_alloc_memslots();
1008 goto out_err_no_arch_destroy_vm;
1009 /* Generations must be different for each address space. */
1010 slots->generation = i;
1011 rcu_assign_pointer(kvm->memslots[i], slots);
1014 for (i = 0; i < KVM_NR_BUSES; i++) {
1015 rcu_assign_pointer(kvm->buses[i],
1016 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1018 goto out_err_no_arch_destroy_vm;
1021 kvm->max_halt_poll_ns = halt_poll_ns;
1023 r = kvm_arch_init_vm(kvm, type);
1025 goto out_err_no_arch_destroy_vm;
1027 r = hardware_enable_all();
1029 goto out_err_no_disable;
1031 #ifdef CONFIG_HAVE_KVM_IRQFD
1032 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1035 r = kvm_init_mmu_notifier(kvm);
1037 goto out_err_no_mmu_notifier;
1039 r = kvm_arch_post_init_vm(kvm);
1043 mutex_lock(&kvm_lock);
1044 list_add(&kvm->vm_list, &vm_list);
1045 mutex_unlock(&kvm_lock);
1047 preempt_notifier_inc();
1048 kvm_init_pm_notifier(kvm);
1053 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1054 if (kvm->mmu_notifier.ops)
1055 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1057 out_err_no_mmu_notifier:
1058 hardware_disable_all();
1060 kvm_arch_destroy_vm(kvm);
1061 out_err_no_arch_destroy_vm:
1062 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1063 for (i = 0; i < KVM_NR_BUSES; i++)
1064 kfree(kvm_get_bus(kvm, i));
1065 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1066 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1067 cleanup_srcu_struct(&kvm->irq_srcu);
1068 out_err_no_irq_srcu:
1069 cleanup_srcu_struct(&kvm->srcu);
1071 kvm_arch_free_vm(kvm);
1072 mmdrop(current->mm);
1076 static void kvm_destroy_devices(struct kvm *kvm)
1078 struct kvm_device *dev, *tmp;
1081 * We do not need to take the kvm->lock here, because nobody else
1082 * has a reference to the struct kvm at this point and therefore
1083 * cannot access the devices list anyhow.
1085 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1086 list_del(&dev->vm_node);
1087 dev->ops->destroy(dev);
1091 static void kvm_destroy_vm(struct kvm *kvm)
1094 struct mm_struct *mm = kvm->mm;
1096 kvm_destroy_pm_notifier(kvm);
1097 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1098 kvm_destroy_vm_debugfs(kvm);
1099 kvm_arch_sync_events(kvm);
1100 mutex_lock(&kvm_lock);
1101 list_del(&kvm->vm_list);
1102 mutex_unlock(&kvm_lock);
1103 kvm_arch_pre_destroy_vm(kvm);
1105 kvm_free_irq_routing(kvm);
1106 for (i = 0; i < KVM_NR_BUSES; i++) {
1107 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1110 kvm_io_bus_destroy(bus);
1111 kvm->buses[i] = NULL;
1113 kvm_coalesced_mmio_free(kvm);
1114 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1115 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1117 kvm_arch_flush_shadow_all(kvm);
1119 kvm_arch_destroy_vm(kvm);
1120 kvm_destroy_devices(kvm);
1121 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1122 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1123 cleanup_srcu_struct(&kvm->irq_srcu);
1124 cleanup_srcu_struct(&kvm->srcu);
1125 kvm_arch_free_vm(kvm);
1126 preempt_notifier_dec();
1127 hardware_disable_all();
1131 void kvm_get_kvm(struct kvm *kvm)
1133 refcount_inc(&kvm->users_count);
1135 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1137 void kvm_put_kvm(struct kvm *kvm)
1139 if (refcount_dec_and_test(&kvm->users_count))
1140 kvm_destroy_vm(kvm);
1142 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1145 * Used to put a reference that was taken on behalf of an object associated
1146 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1147 * of the new file descriptor fails and the reference cannot be transferred to
1148 * its final owner. In such cases, the caller is still actively using @kvm and
1149 * will fail miserably if the refcount unexpectedly hits zero.
1151 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1153 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1155 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1157 static int kvm_vm_release(struct inode *inode, struct file *filp)
1159 struct kvm *kvm = filp->private_data;
1161 kvm_irqfd_release(kvm);
1168 * Allocation size is twice as large as the actual dirty bitmap size.
1169 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1171 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1173 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1175 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1176 if (!memslot->dirty_bitmap)
1183 * Delete a memslot by decrementing the number of used slots and shifting all
1184 * other entries in the array forward one spot.
1186 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1187 struct kvm_memory_slot *memslot)
1189 struct kvm_memory_slot *mslots = slots->memslots;
1192 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1195 slots->used_slots--;
1197 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
1198 atomic_set(&slots->lru_slot, 0);
1200 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1201 mslots[i] = mslots[i + 1];
1202 slots->id_to_index[mslots[i].id] = i;
1204 mslots[i] = *memslot;
1205 slots->id_to_index[memslot->id] = -1;
1209 * "Insert" a new memslot by incrementing the number of used slots. Returns
1210 * the new slot's initial index into the memslots array.
1212 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1214 return slots->used_slots++;
1218 * Move a changed memslot backwards in the array by shifting existing slots
1219 * with a higher GFN toward the front of the array. Note, the changed memslot
1220 * itself is not preserved in the array, i.e. not swapped at this time, only
1221 * its new index into the array is tracked. Returns the changed memslot's
1222 * current index into the memslots array.
1224 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1225 struct kvm_memory_slot *memslot)
1227 struct kvm_memory_slot *mslots = slots->memslots;
1230 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1231 WARN_ON_ONCE(!slots->used_slots))
1235 * Move the target memslot backward in the array by shifting existing
1236 * memslots with a higher GFN (than the target memslot) towards the
1237 * front of the array.
1239 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1240 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1243 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1245 /* Shift the next memslot forward one and update its index. */
1246 mslots[i] = mslots[i + 1];
1247 slots->id_to_index[mslots[i].id] = i;
1253 * Move a changed memslot forwards in the array by shifting existing slots with
1254 * a lower GFN toward the back of the array. Note, the changed memslot itself
1255 * is not preserved in the array, i.e. not swapped at this time, only its new
1256 * index into the array is tracked. Returns the changed memslot's final index
1257 * into the memslots array.
1259 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1260 struct kvm_memory_slot *memslot,
1263 struct kvm_memory_slot *mslots = slots->memslots;
1266 for (i = start; i > 0; i--) {
1267 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1270 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1272 /* Shift the next memslot back one and update its index. */
1273 mslots[i] = mslots[i - 1];
1274 slots->id_to_index[mslots[i].id] = i;
1280 * Re-sort memslots based on their GFN to account for an added, deleted, or
1281 * moved memslot. Sorting memslots by GFN allows using a binary search during
1284 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1285 * at memslots[0] has the highest GFN.
1287 * The sorting algorithm takes advantage of having initially sorted memslots
1288 * and knowing the position of the changed memslot. Sorting is also optimized
1289 * by not swapping the updated memslot and instead only shifting other memslots
1290 * and tracking the new index for the update memslot. Only once its final
1291 * index is known is the updated memslot copied into its position in the array.
1293 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1294 * the end of the array.
1296 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1297 * end of the array and then it forward to its correct location.
1299 * - When moving a memslot, the algorithm first moves the updated memslot
1300 * backward to handle the scenario where the memslot's GFN was changed to a
1301 * lower value. update_memslots() then falls through and runs the same flow
1302 * as creating a memslot to move the memslot forward to handle the scenario
1303 * where its GFN was changed to a higher value.
1305 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1306 * historical reasons. Originally, invalid memslots where denoted by having
1307 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1308 * to the end of the array. The current algorithm uses dedicated logic to
1309 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1311 * The other historical motiviation for highest->lowest was to improve the
1312 * performance of memslot lookup. KVM originally used a linear search starting
1313 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1314 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1315 * single memslot above the 4gb boundary. As the largest memslot is also the
1316 * most likely to be referenced, sorting it to the front of the array was
1317 * advantageous. The current binary search starts from the middle of the array
1318 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1320 static void update_memslots(struct kvm_memslots *slots,
1321 struct kvm_memory_slot *memslot,
1322 enum kvm_mr_change change)
1326 if (change == KVM_MR_DELETE) {
1327 kvm_memslot_delete(slots, memslot);
1329 if (change == KVM_MR_CREATE)
1330 i = kvm_memslot_insert_back(slots);
1332 i = kvm_memslot_move_backward(slots, memslot);
1333 i = kvm_memslot_move_forward(slots, memslot, i);
1336 * Copy the memslot to its new position in memslots and update
1337 * its index accordingly.
1339 slots->memslots[i] = *memslot;
1340 slots->id_to_index[memslot->id] = i;
1344 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1346 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1348 #ifdef __KVM_HAVE_READONLY_MEM
1349 valid_flags |= KVM_MEM_READONLY;
1352 if (mem->flags & ~valid_flags)
1358 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1359 int as_id, struct kvm_memslots *slots)
1361 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1362 u64 gen = old_memslots->generation;
1364 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1365 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1367 rcu_assign_pointer(kvm->memslots[as_id], slots);
1370 * Acquired in kvm_set_memslot. Must be released before synchronize
1371 * SRCU below in order to avoid deadlock with another thread
1372 * acquiring the slots_arch_lock in an srcu critical section.
1374 mutex_unlock(&kvm->slots_arch_lock);
1376 synchronize_srcu_expedited(&kvm->srcu);
1379 * Increment the new memslot generation a second time, dropping the
1380 * update in-progress flag and incrementing the generation based on
1381 * the number of address spaces. This provides a unique and easily
1382 * identifiable generation number while the memslots are in flux.
1384 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1387 * Generations must be unique even across address spaces. We do not need
1388 * a global counter for that, instead the generation space is evenly split
1389 * across address spaces. For example, with two address spaces, address
1390 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1391 * use generations 1, 3, 5, ...
1393 gen += KVM_ADDRESS_SPACE_NUM;
1395 kvm_arch_memslots_updated(kvm, gen);
1397 slots->generation = gen;
1399 return old_memslots;
1402 static size_t kvm_memslots_size(int slots)
1404 return sizeof(struct kvm_memslots) +
1405 (sizeof(struct kvm_memory_slot) * slots);
1408 static void kvm_copy_memslots(struct kvm_memslots *to,
1409 struct kvm_memslots *from)
1411 memcpy(to, from, kvm_memslots_size(from->used_slots));
1415 * Note, at a minimum, the current number of used slots must be allocated, even
1416 * when deleting a memslot, as we need a complete duplicate of the memslots for
1417 * use when invalidating a memslot prior to deleting/moving the memslot.
1419 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1420 enum kvm_mr_change change)
1422 struct kvm_memslots *slots;
1425 if (change == KVM_MR_CREATE)
1426 new_size = kvm_memslots_size(old->used_slots + 1);
1428 new_size = kvm_memslots_size(old->used_slots);
1430 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1432 kvm_copy_memslots(slots, old);
1437 static int kvm_set_memslot(struct kvm *kvm,
1438 const struct kvm_userspace_memory_region *mem,
1439 struct kvm_memory_slot *old,
1440 struct kvm_memory_slot *new, int as_id,
1441 enum kvm_mr_change change)
1443 struct kvm_memory_slot *slot;
1444 struct kvm_memslots *slots;
1448 * Released in install_new_memslots.
1450 * Must be held from before the current memslots are copied until
1451 * after the new memslots are installed with rcu_assign_pointer,
1452 * then released before the synchronize srcu in install_new_memslots.
1454 * When modifying memslots outside of the slots_lock, must be held
1455 * before reading the pointer to the current memslots until after all
1456 * changes to those memslots are complete.
1458 * These rules ensure that installing new memslots does not lose
1459 * changes made to the previous memslots.
1461 mutex_lock(&kvm->slots_arch_lock);
1463 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1465 mutex_unlock(&kvm->slots_arch_lock);
1469 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1471 * Note, the INVALID flag needs to be in the appropriate entry
1472 * in the freshly allocated memslots, not in @old or @new.
1474 slot = id_to_memslot(slots, old->id);
1475 slot->flags |= KVM_MEMSLOT_INVALID;
1478 * We can re-use the memory from the old memslots.
1479 * It will be overwritten with a copy of the new memslots
1480 * after reacquiring the slots_arch_lock below.
1482 slots = install_new_memslots(kvm, as_id, slots);
1484 /* From this point no new shadow pages pointing to a deleted,
1485 * or moved, memslot will be created.
1487 * validation of sp->gfn happens in:
1488 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1489 * - kvm_is_visible_gfn (mmu_check_root)
1491 kvm_arch_flush_shadow_memslot(kvm, slot);
1493 /* Released in install_new_memslots. */
1494 mutex_lock(&kvm->slots_arch_lock);
1497 * The arch-specific fields of the memslots could have changed
1498 * between releasing the slots_arch_lock in
1499 * install_new_memslots and here, so get a fresh copy of the
1502 kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id));
1505 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1509 update_memslots(slots, new, change);
1510 slots = install_new_memslots(kvm, as_id, slots);
1512 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1518 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1519 slot = id_to_memslot(slots, old->id);
1520 slot->flags &= ~KVM_MEMSLOT_INVALID;
1521 slots = install_new_memslots(kvm, as_id, slots);
1523 mutex_unlock(&kvm->slots_arch_lock);
1529 static int kvm_delete_memslot(struct kvm *kvm,
1530 const struct kvm_userspace_memory_region *mem,
1531 struct kvm_memory_slot *old, int as_id)
1533 struct kvm_memory_slot new;
1539 memset(&new, 0, sizeof(new));
1542 * This is only for debugging purpose; it should never be referenced
1543 * for a removed memslot.
1547 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1551 kvm_free_memslot(kvm, old);
1556 * Allocate some memory and give it an address in the guest physical address
1559 * Discontiguous memory is allowed, mostly for framebuffers.
1561 * Must be called holding kvm->slots_lock for write.
1563 int __kvm_set_memory_region(struct kvm *kvm,
1564 const struct kvm_userspace_memory_region *mem)
1566 struct kvm_memory_slot old, new;
1567 struct kvm_memory_slot *tmp;
1568 enum kvm_mr_change change;
1572 r = check_memory_region_flags(mem);
1576 as_id = mem->slot >> 16;
1577 id = (u16)mem->slot;
1579 /* General sanity checks */
1580 if (mem->memory_size & (PAGE_SIZE - 1))
1582 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1584 /* We can read the guest memory with __xxx_user() later on. */
1585 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1586 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1587 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1590 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1592 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1596 * Make a full copy of the old memslot, the pointer will become stale
1597 * when the memslots are re-sorted by update_memslots(), and the old
1598 * memslot needs to be referenced after calling update_memslots(), e.g.
1599 * to free its resources and for arch specific behavior.
1601 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1606 memset(&old, 0, sizeof(old));
1610 if (!mem->memory_size)
1611 return kvm_delete_memslot(kvm, mem, &old, as_id);
1615 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1616 new.npages = mem->memory_size >> PAGE_SHIFT;
1617 new.flags = mem->flags;
1618 new.userspace_addr = mem->userspace_addr;
1620 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1624 change = KVM_MR_CREATE;
1625 new.dirty_bitmap = NULL;
1626 memset(&new.arch, 0, sizeof(new.arch));
1627 } else { /* Modify an existing slot. */
1628 if ((new.userspace_addr != old.userspace_addr) ||
1629 (new.npages != old.npages) ||
1630 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1633 if (new.base_gfn != old.base_gfn)
1634 change = KVM_MR_MOVE;
1635 else if (new.flags != old.flags)
1636 change = KVM_MR_FLAGS_ONLY;
1637 else /* Nothing to change. */
1640 /* Copy dirty_bitmap and arch from the current memslot. */
1641 new.dirty_bitmap = old.dirty_bitmap;
1642 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1645 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1646 /* Check for overlaps */
1647 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1650 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1651 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1656 /* Allocate/free page dirty bitmap as needed */
1657 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1658 new.dirty_bitmap = NULL;
1659 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1660 r = kvm_alloc_dirty_bitmap(&new);
1664 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1665 bitmap_set(new.dirty_bitmap, 0, new.npages);
1668 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1672 if (old.dirty_bitmap && !new.dirty_bitmap)
1673 kvm_destroy_dirty_bitmap(&old);
1677 if (new.dirty_bitmap && !old.dirty_bitmap)
1678 kvm_destroy_dirty_bitmap(&new);
1681 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1683 int kvm_set_memory_region(struct kvm *kvm,
1684 const struct kvm_userspace_memory_region *mem)
1688 mutex_lock(&kvm->slots_lock);
1689 r = __kvm_set_memory_region(kvm, mem);
1690 mutex_unlock(&kvm->slots_lock);
1693 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1695 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1696 struct kvm_userspace_memory_region *mem)
1698 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1701 return kvm_set_memory_region(kvm, mem);
1704 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1706 * kvm_get_dirty_log - get a snapshot of dirty pages
1707 * @kvm: pointer to kvm instance
1708 * @log: slot id and address to which we copy the log
1709 * @is_dirty: set to '1' if any dirty pages were found
1710 * @memslot: set to the associated memslot, always valid on success
1712 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1713 int *is_dirty, struct kvm_memory_slot **memslot)
1715 struct kvm_memslots *slots;
1718 unsigned long any = 0;
1720 /* Dirty ring tracking is exclusive to dirty log tracking */
1721 if (kvm->dirty_ring_size)
1727 as_id = log->slot >> 16;
1728 id = (u16)log->slot;
1729 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1732 slots = __kvm_memslots(kvm, as_id);
1733 *memslot = id_to_memslot(slots, id);
1734 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1737 kvm_arch_sync_dirty_log(kvm, *memslot);
1739 n = kvm_dirty_bitmap_bytes(*memslot);
1741 for (i = 0; !any && i < n/sizeof(long); ++i)
1742 any = (*memslot)->dirty_bitmap[i];
1744 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1751 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1753 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1755 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1756 * and reenable dirty page tracking for the corresponding pages.
1757 * @kvm: pointer to kvm instance
1758 * @log: slot id and address to which we copy the log
1760 * We need to keep it in mind that VCPU threads can write to the bitmap
1761 * concurrently. So, to avoid losing track of dirty pages we keep the
1764 * 1. Take a snapshot of the bit and clear it if needed.
1765 * 2. Write protect the corresponding page.
1766 * 3. Copy the snapshot to the userspace.
1767 * 4. Upon return caller flushes TLB's if needed.
1769 * Between 2 and 4, the guest may write to the page using the remaining TLB
1770 * entry. This is not a problem because the page is reported dirty using
1771 * the snapshot taken before and step 4 ensures that writes done after
1772 * exiting to userspace will be logged for the next call.
1775 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1777 struct kvm_memslots *slots;
1778 struct kvm_memory_slot *memslot;
1781 unsigned long *dirty_bitmap;
1782 unsigned long *dirty_bitmap_buffer;
1785 /* Dirty ring tracking is exclusive to dirty log tracking */
1786 if (kvm->dirty_ring_size)
1789 as_id = log->slot >> 16;
1790 id = (u16)log->slot;
1791 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1794 slots = __kvm_memslots(kvm, as_id);
1795 memslot = id_to_memslot(slots, id);
1796 if (!memslot || !memslot->dirty_bitmap)
1799 dirty_bitmap = memslot->dirty_bitmap;
1801 kvm_arch_sync_dirty_log(kvm, memslot);
1803 n = kvm_dirty_bitmap_bytes(memslot);
1805 if (kvm->manual_dirty_log_protect) {
1807 * Unlike kvm_get_dirty_log, we always return false in *flush,
1808 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1809 * is some code duplication between this function and
1810 * kvm_get_dirty_log, but hopefully all architecture
1811 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1812 * can be eliminated.
1814 dirty_bitmap_buffer = dirty_bitmap;
1816 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1817 memset(dirty_bitmap_buffer, 0, n);
1820 for (i = 0; i < n / sizeof(long); i++) {
1824 if (!dirty_bitmap[i])
1828 mask = xchg(&dirty_bitmap[i], 0);
1829 dirty_bitmap_buffer[i] = mask;
1831 offset = i * BITS_PER_LONG;
1832 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1835 KVM_MMU_UNLOCK(kvm);
1839 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1841 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1848 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1849 * @kvm: kvm instance
1850 * @log: slot id and address to which we copy the log
1852 * Steps 1-4 below provide general overview of dirty page logging. See
1853 * kvm_get_dirty_log_protect() function description for additional details.
1855 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1856 * always flush the TLB (step 4) even if previous step failed and the dirty
1857 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1858 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1859 * writes will be marked dirty for next log read.
1861 * 1. Take a snapshot of the bit and clear it if needed.
1862 * 2. Write protect the corresponding page.
1863 * 3. Copy the snapshot to the userspace.
1864 * 4. Flush TLB's if needed.
1866 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1867 struct kvm_dirty_log *log)
1871 mutex_lock(&kvm->slots_lock);
1873 r = kvm_get_dirty_log_protect(kvm, log);
1875 mutex_unlock(&kvm->slots_lock);
1880 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1881 * and reenable dirty page tracking for the corresponding pages.
1882 * @kvm: pointer to kvm instance
1883 * @log: slot id and address from which to fetch the bitmap of dirty pages
1885 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1886 struct kvm_clear_dirty_log *log)
1888 struct kvm_memslots *slots;
1889 struct kvm_memory_slot *memslot;
1893 unsigned long *dirty_bitmap;
1894 unsigned long *dirty_bitmap_buffer;
1897 /* Dirty ring tracking is exclusive to dirty log tracking */
1898 if (kvm->dirty_ring_size)
1901 as_id = log->slot >> 16;
1902 id = (u16)log->slot;
1903 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1906 if (log->first_page & 63)
1909 slots = __kvm_memslots(kvm, as_id);
1910 memslot = id_to_memslot(slots, id);
1911 if (!memslot || !memslot->dirty_bitmap)
1914 dirty_bitmap = memslot->dirty_bitmap;
1916 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1918 if (log->first_page > memslot->npages ||
1919 log->num_pages > memslot->npages - log->first_page ||
1920 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1923 kvm_arch_sync_dirty_log(kvm, memslot);
1926 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1927 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1931 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1932 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1933 i++, offset += BITS_PER_LONG) {
1934 unsigned long mask = *dirty_bitmap_buffer++;
1935 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1939 mask &= atomic_long_fetch_andnot(mask, p);
1942 * mask contains the bits that really have been cleared. This
1943 * never includes any bits beyond the length of the memslot (if
1944 * the length is not aligned to 64 pages), therefore it is not
1945 * a problem if userspace sets them in log->dirty_bitmap.
1949 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1953 KVM_MMU_UNLOCK(kvm);
1956 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1961 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1962 struct kvm_clear_dirty_log *log)
1966 mutex_lock(&kvm->slots_lock);
1968 r = kvm_clear_dirty_log_protect(kvm, log);
1970 mutex_unlock(&kvm->slots_lock);
1973 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1975 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1977 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1979 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1981 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1983 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1985 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1987 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1989 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1991 return kvm_is_visible_memslot(memslot);
1993 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1995 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1997 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1999 return kvm_is_visible_memslot(memslot);
2001 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2003 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2005 struct vm_area_struct *vma;
2006 unsigned long addr, size;
2010 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2011 if (kvm_is_error_hva(addr))
2014 mmap_read_lock(current->mm);
2015 vma = find_vma(current->mm, addr);
2019 size = vma_kernel_pagesize(vma);
2022 mmap_read_unlock(current->mm);
2027 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
2029 return slot->flags & KVM_MEM_READONLY;
2032 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2033 gfn_t *nr_pages, bool write)
2035 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2036 return KVM_HVA_ERR_BAD;
2038 if (memslot_is_readonly(slot) && write)
2039 return KVM_HVA_ERR_RO_BAD;
2042 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2044 return __gfn_to_hva_memslot(slot, gfn);
2047 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2050 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2053 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2056 return gfn_to_hva_many(slot, gfn, NULL);
2058 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2060 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2062 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2064 EXPORT_SYMBOL_GPL(gfn_to_hva);
2066 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2068 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2070 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2073 * Return the hva of a @gfn and the R/W attribute if possible.
2075 * @slot: the kvm_memory_slot which contains @gfn
2076 * @gfn: the gfn to be translated
2077 * @writable: used to return the read/write attribute of the @slot if the hva
2078 * is valid and @writable is not NULL
2080 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2081 gfn_t gfn, bool *writable)
2083 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2085 if (!kvm_is_error_hva(hva) && writable)
2086 *writable = !memslot_is_readonly(slot);
2091 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2093 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2095 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2098 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2100 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2102 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2105 static inline int check_user_page_hwpoison(unsigned long addr)
2107 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2109 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2110 return rc == -EHWPOISON;
2114 * The fast path to get the writable pfn which will be stored in @pfn,
2115 * true indicates success, otherwise false is returned. It's also the
2116 * only part that runs if we can in atomic context.
2118 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2119 bool *writable, kvm_pfn_t *pfn)
2121 struct page *page[1];
2124 * Fast pin a writable pfn only if it is a write fault request
2125 * or the caller allows to map a writable pfn for a read fault
2128 if (!(write_fault || writable))
2131 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2132 *pfn = page_to_pfn(page[0]);
2143 * The slow path to get the pfn of the specified host virtual address,
2144 * 1 indicates success, -errno is returned if error is detected.
2146 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2147 bool *writable, kvm_pfn_t *pfn)
2149 unsigned int flags = FOLL_HWPOISON;
2156 *writable = write_fault;
2159 flags |= FOLL_WRITE;
2161 flags |= FOLL_NOWAIT;
2163 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2167 /* map read fault as writable if possible */
2168 if (unlikely(!write_fault) && writable) {
2171 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2177 *pfn = page_to_pfn(page);
2181 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2183 if (unlikely(!(vma->vm_flags & VM_READ)))
2186 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2192 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2194 if (kvm_is_reserved_pfn(pfn))
2196 return get_page_unless_zero(pfn_to_page(pfn));
2199 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2200 unsigned long addr, bool *async,
2201 bool write_fault, bool *writable,
2209 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2212 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2213 * not call the fault handler, so do it here.
2215 bool unlocked = false;
2216 r = fixup_user_fault(current->mm, addr,
2217 (write_fault ? FAULT_FLAG_WRITE : 0),
2224 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2229 if (write_fault && !pte_write(*ptep)) {
2230 pfn = KVM_PFN_ERR_RO_FAULT;
2235 *writable = pte_write(*ptep);
2236 pfn = pte_pfn(*ptep);
2239 * Get a reference here because callers of *hva_to_pfn* and
2240 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2241 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2242 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2243 * simply do nothing for reserved pfns.
2245 * Whoever called remap_pfn_range is also going to call e.g.
2246 * unmap_mapping_range before the underlying pages are freed,
2247 * causing a call to our MMU notifier.
2249 * Certain IO or PFNMAP mappings can be backed with valid
2250 * struct pages, but be allocated without refcounting e.g.,
2251 * tail pages of non-compound higher order allocations, which
2252 * would then underflow the refcount when the caller does the
2253 * required put_page. Don't allow those pages here.
2255 if (!kvm_try_get_pfn(pfn))
2259 pte_unmap_unlock(ptep, ptl);
2266 * Pin guest page in memory and return its pfn.
2267 * @addr: host virtual address which maps memory to the guest
2268 * @atomic: whether this function can sleep
2269 * @async: whether this function need to wait IO complete if the
2270 * host page is not in the memory
2271 * @write_fault: whether we should get a writable host page
2272 * @writable: whether it allows to map a writable host page for !@write_fault
2274 * The function will map a writable host page for these two cases:
2275 * 1): @write_fault = true
2276 * 2): @write_fault = false && @writable, @writable will tell the caller
2277 * whether the mapping is writable.
2279 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2280 bool write_fault, bool *writable)
2282 struct vm_area_struct *vma;
2286 /* we can do it either atomically or asynchronously, not both */
2287 BUG_ON(atomic && async);
2289 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2293 return KVM_PFN_ERR_FAULT;
2295 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2299 mmap_read_lock(current->mm);
2300 if (npages == -EHWPOISON ||
2301 (!async && check_user_page_hwpoison(addr))) {
2302 pfn = KVM_PFN_ERR_HWPOISON;
2307 vma = vma_lookup(current->mm, addr);
2310 pfn = KVM_PFN_ERR_FAULT;
2311 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2312 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2316 pfn = KVM_PFN_ERR_FAULT;
2318 if (async && vma_is_valid(vma, write_fault))
2320 pfn = KVM_PFN_ERR_FAULT;
2323 mmap_read_unlock(current->mm);
2327 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2328 bool atomic, bool *async, bool write_fault,
2329 bool *writable, hva_t *hva)
2331 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2336 if (addr == KVM_HVA_ERR_RO_BAD) {
2339 return KVM_PFN_ERR_RO_FAULT;
2342 if (kvm_is_error_hva(addr)) {
2345 return KVM_PFN_NOSLOT;
2348 /* Do not map writable pfn in the readonly memslot. */
2349 if (writable && memslot_is_readonly(slot)) {
2354 return hva_to_pfn(addr, atomic, async, write_fault,
2357 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2359 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2362 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2363 write_fault, writable, NULL);
2365 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2367 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2369 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2371 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2373 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2375 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2377 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2379 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2381 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2383 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2385 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2387 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2389 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2391 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2393 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2395 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2397 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2398 struct page **pages, int nr_pages)
2403 addr = gfn_to_hva_many(slot, gfn, &entry);
2404 if (kvm_is_error_hva(addr))
2407 if (entry < nr_pages)
2410 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2412 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2414 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2416 if (is_error_noslot_pfn(pfn))
2417 return KVM_ERR_PTR_BAD_PAGE;
2419 if (kvm_is_reserved_pfn(pfn)) {
2421 return KVM_ERR_PTR_BAD_PAGE;
2424 return pfn_to_page(pfn);
2427 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2431 pfn = gfn_to_pfn(kvm, gfn);
2433 return kvm_pfn_to_page(pfn);
2435 EXPORT_SYMBOL_GPL(gfn_to_page);
2437 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2443 cache->pfn = cache->gfn = 0;
2446 kvm_release_pfn_dirty(pfn);
2448 kvm_release_pfn_clean(pfn);
2451 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2452 struct gfn_to_pfn_cache *cache, u64 gen)
2454 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2456 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2458 cache->dirty = false;
2459 cache->generation = gen;
2462 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2463 struct kvm_host_map *map,
2464 struct gfn_to_pfn_cache *cache,
2469 struct page *page = KVM_UNMAPPED_PAGE;
2470 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2471 u64 gen = slots->generation;
2477 if (!cache->pfn || cache->gfn != gfn ||
2478 cache->generation != gen) {
2481 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2487 pfn = gfn_to_pfn_memslot(slot, gfn);
2489 if (is_error_noslot_pfn(pfn))
2492 if (pfn_valid(pfn)) {
2493 page = pfn_to_page(pfn);
2495 hva = kmap_atomic(page);
2498 #ifdef CONFIG_HAS_IOMEM
2499 } else if (!atomic) {
2500 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2517 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2518 struct gfn_to_pfn_cache *cache, bool atomic)
2520 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2523 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2525 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2527 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2530 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2532 static void __kvm_unmap_gfn(struct kvm *kvm,
2533 struct kvm_memory_slot *memslot,
2534 struct kvm_host_map *map,
2535 struct gfn_to_pfn_cache *cache,
2536 bool dirty, bool atomic)
2544 if (map->page != KVM_UNMAPPED_PAGE) {
2546 kunmap_atomic(map->hva);
2550 #ifdef CONFIG_HAS_IOMEM
2554 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2558 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2561 cache->dirty |= dirty;
2563 kvm_release_pfn(map->pfn, dirty, NULL);
2569 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2570 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2572 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2573 cache, dirty, atomic);
2576 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2578 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2580 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2581 map, NULL, dirty, false);
2583 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2585 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2589 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2591 return kvm_pfn_to_page(pfn);
2593 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2595 void kvm_release_page_clean(struct page *page)
2597 WARN_ON(is_error_page(page));
2599 kvm_release_pfn_clean(page_to_pfn(page));
2601 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2603 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2605 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2606 put_page(pfn_to_page(pfn));
2608 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2610 void kvm_release_page_dirty(struct page *page)
2612 WARN_ON(is_error_page(page));
2614 kvm_release_pfn_dirty(page_to_pfn(page));
2616 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2618 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2620 kvm_set_pfn_dirty(pfn);
2621 kvm_release_pfn_clean(pfn);
2623 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2625 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2627 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2628 SetPageDirty(pfn_to_page(pfn));
2630 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2632 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2634 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2635 mark_page_accessed(pfn_to_page(pfn));
2637 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2639 void kvm_get_pfn(kvm_pfn_t pfn)
2641 if (!kvm_is_reserved_pfn(pfn))
2642 get_page(pfn_to_page(pfn));
2644 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2646 static int next_segment(unsigned long len, int offset)
2648 if (len > PAGE_SIZE - offset)
2649 return PAGE_SIZE - offset;
2654 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2655 void *data, int offset, int len)
2660 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2661 if (kvm_is_error_hva(addr))
2663 r = __copy_from_user(data, (void __user *)addr + offset, len);
2669 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2672 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2674 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2676 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2678 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2679 int offset, int len)
2681 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2683 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2685 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2687 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2689 gfn_t gfn = gpa >> PAGE_SHIFT;
2691 int offset = offset_in_page(gpa);
2694 while ((seg = next_segment(len, offset)) != 0) {
2695 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2705 EXPORT_SYMBOL_GPL(kvm_read_guest);
2707 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2709 gfn_t gfn = gpa >> PAGE_SHIFT;
2711 int offset = offset_in_page(gpa);
2714 while ((seg = next_segment(len, offset)) != 0) {
2715 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2725 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2727 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2728 void *data, int offset, unsigned long len)
2733 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2734 if (kvm_is_error_hva(addr))
2736 pagefault_disable();
2737 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2744 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2745 void *data, unsigned long len)
2747 gfn_t gfn = gpa >> PAGE_SHIFT;
2748 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2749 int offset = offset_in_page(gpa);
2751 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2753 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2755 static int __kvm_write_guest_page(struct kvm *kvm,
2756 struct kvm_memory_slot *memslot, gfn_t gfn,
2757 const void *data, int offset, int len)
2762 addr = gfn_to_hva_memslot(memslot, gfn);
2763 if (kvm_is_error_hva(addr))
2765 r = __copy_to_user((void __user *)addr + offset, data, len);
2768 mark_page_dirty_in_slot(kvm, memslot, gfn);
2772 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2773 const void *data, int offset, int len)
2775 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2777 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2779 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2781 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2782 const void *data, int offset, int len)
2784 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2786 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2788 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2790 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2793 gfn_t gfn = gpa >> PAGE_SHIFT;
2795 int offset = offset_in_page(gpa);
2798 while ((seg = next_segment(len, offset)) != 0) {
2799 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2809 EXPORT_SYMBOL_GPL(kvm_write_guest);
2811 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2814 gfn_t gfn = gpa >> PAGE_SHIFT;
2816 int offset = offset_in_page(gpa);
2819 while ((seg = next_segment(len, offset)) != 0) {
2820 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2830 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2832 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2833 struct gfn_to_hva_cache *ghc,
2834 gpa_t gpa, unsigned long len)
2836 int offset = offset_in_page(gpa);
2837 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2838 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2839 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2840 gfn_t nr_pages_avail;
2842 /* Update ghc->generation before performing any error checks. */
2843 ghc->generation = slots->generation;
2845 if (start_gfn > end_gfn) {
2846 ghc->hva = KVM_HVA_ERR_BAD;
2851 * If the requested region crosses two memslots, we still
2852 * verify that the entire region is valid here.
2854 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2855 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2856 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2858 if (kvm_is_error_hva(ghc->hva))
2862 /* Use the slow path for cross page reads and writes. */
2863 if (nr_pages_needed == 1)
2866 ghc->memslot = NULL;
2873 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2874 gpa_t gpa, unsigned long len)
2876 struct kvm_memslots *slots = kvm_memslots(kvm);
2877 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2879 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2881 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2882 void *data, unsigned int offset,
2885 struct kvm_memslots *slots = kvm_memslots(kvm);
2887 gpa_t gpa = ghc->gpa + offset;
2889 BUG_ON(len + offset > ghc->len);
2891 if (slots->generation != ghc->generation) {
2892 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2896 if (kvm_is_error_hva(ghc->hva))
2899 if (unlikely(!ghc->memslot))
2900 return kvm_write_guest(kvm, gpa, data, len);
2902 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2905 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2909 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2911 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2912 void *data, unsigned long len)
2914 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2916 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2918 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2919 void *data, unsigned int offset,
2922 struct kvm_memslots *slots = kvm_memslots(kvm);
2924 gpa_t gpa = ghc->gpa + offset;
2926 BUG_ON(len + offset > ghc->len);
2928 if (slots->generation != ghc->generation) {
2929 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2933 if (kvm_is_error_hva(ghc->hva))
2936 if (unlikely(!ghc->memslot))
2937 return kvm_read_guest(kvm, gpa, data, len);
2939 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2945 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2947 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2948 void *data, unsigned long len)
2950 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2952 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2954 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2956 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2957 gfn_t gfn = gpa >> PAGE_SHIFT;
2959 int offset = offset_in_page(gpa);
2962 while ((seg = next_segment(len, offset)) != 0) {
2963 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2972 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2974 void mark_page_dirty_in_slot(struct kvm *kvm,
2975 struct kvm_memory_slot *memslot,
2978 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
2979 unsigned long rel_gfn = gfn - memslot->base_gfn;
2980 u32 slot = (memslot->as_id << 16) | memslot->id;
2982 if (kvm->dirty_ring_size)
2983 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
2986 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2989 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
2991 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2993 struct kvm_memory_slot *memslot;
2995 memslot = gfn_to_memslot(kvm, gfn);
2996 mark_page_dirty_in_slot(kvm, memslot, gfn);
2998 EXPORT_SYMBOL_GPL(mark_page_dirty);
3000 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3002 struct kvm_memory_slot *memslot;
3004 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3005 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3007 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3009 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3011 if (!vcpu->sigset_active)
3015 * This does a lockless modification of ->real_blocked, which is fine
3016 * because, only current can change ->real_blocked and all readers of
3017 * ->real_blocked don't care as long ->real_blocked is always a subset
3020 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3023 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3025 if (!vcpu->sigset_active)
3028 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3029 sigemptyset(¤t->real_blocked);
3032 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3034 unsigned int old, val, grow, grow_start;
3036 old = val = vcpu->halt_poll_ns;
3037 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3038 grow = READ_ONCE(halt_poll_ns_grow);
3043 if (val < grow_start)
3046 if (val > vcpu->kvm->max_halt_poll_ns)
3047 val = vcpu->kvm->max_halt_poll_ns;
3049 vcpu->halt_poll_ns = val;
3051 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3054 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3056 unsigned int old, val, shrink;
3058 old = val = vcpu->halt_poll_ns;
3059 shrink = READ_ONCE(halt_poll_ns_shrink);
3065 vcpu->halt_poll_ns = val;
3066 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3069 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3072 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3074 if (kvm_arch_vcpu_runnable(vcpu)) {
3075 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3078 if (kvm_cpu_has_pending_timer(vcpu))
3080 if (signal_pending(current))
3082 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3087 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3092 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
3095 vcpu->stat.generic.halt_poll_fail_ns += poll_ns;
3097 vcpu->stat.generic.halt_poll_success_ns += poll_ns;
3101 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3103 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
3105 ktime_t start, cur, poll_end;
3106 bool waited = false;
3109 kvm_arch_vcpu_blocking(vcpu);
3111 start = cur = poll_end = ktime_get();
3112 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
3113 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3115 ++vcpu->stat.generic.halt_attempted_poll;
3118 * This sets KVM_REQ_UNHALT if an interrupt
3121 if (kvm_vcpu_check_block(vcpu) < 0) {
3122 ++vcpu->stat.generic.halt_successful_poll;
3123 if (!vcpu_valid_wakeup(vcpu))
3124 ++vcpu->stat.generic.halt_poll_invalid;
3128 poll_end = cur = ktime_get();
3129 } while (kvm_vcpu_can_poll(cur, stop));
3132 prepare_to_rcuwait(&vcpu->wait);
3134 set_current_state(TASK_INTERRUPTIBLE);
3136 if (kvm_vcpu_check_block(vcpu) < 0)
3142 finish_rcuwait(&vcpu->wait);
3145 kvm_arch_vcpu_unblocking(vcpu);
3146 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3148 update_halt_poll_stats(
3149 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3151 if (!kvm_arch_no_poll(vcpu)) {
3152 if (!vcpu_valid_wakeup(vcpu)) {
3153 shrink_halt_poll_ns(vcpu);
3154 } else if (vcpu->kvm->max_halt_poll_ns) {
3155 if (block_ns <= vcpu->halt_poll_ns)
3157 /* we had a long block, shrink polling */
3158 else if (vcpu->halt_poll_ns &&
3159 block_ns > vcpu->kvm->max_halt_poll_ns)
3160 shrink_halt_poll_ns(vcpu);
3161 /* we had a short halt and our poll time is too small */
3162 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3163 block_ns < vcpu->kvm->max_halt_poll_ns)
3164 grow_halt_poll_ns(vcpu);
3166 vcpu->halt_poll_ns = 0;
3170 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3171 kvm_arch_vcpu_block_finish(vcpu);
3173 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3175 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3177 struct rcuwait *waitp;
3179 waitp = kvm_arch_vcpu_get_wait(vcpu);
3180 if (rcuwait_wake_up(waitp)) {
3181 WRITE_ONCE(vcpu->ready, true);
3182 ++vcpu->stat.generic.halt_wakeup;
3188 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3192 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3194 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3197 int cpu = vcpu->cpu;
3199 if (kvm_vcpu_wake_up(vcpu))
3203 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3204 if (kvm_arch_vcpu_should_kick(vcpu))
3205 smp_send_reschedule(cpu);
3208 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3209 #endif /* !CONFIG_S390 */
3211 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3214 struct task_struct *task = NULL;
3218 pid = rcu_dereference(target->pid);
3220 task = get_pid_task(pid, PIDTYPE_PID);
3224 ret = yield_to(task, 1);
3225 put_task_struct(task);
3229 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3232 * Helper that checks whether a VCPU is eligible for directed yield.
3233 * Most eligible candidate to yield is decided by following heuristics:
3235 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3236 * (preempted lock holder), indicated by @in_spin_loop.
3237 * Set at the beginning and cleared at the end of interception/PLE handler.
3239 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3240 * chance last time (mostly it has become eligible now since we have probably
3241 * yielded to lockholder in last iteration. This is done by toggling
3242 * @dy_eligible each time a VCPU checked for eligibility.)
3244 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3245 * to preempted lock-holder could result in wrong VCPU selection and CPU
3246 * burning. Giving priority for a potential lock-holder increases lock
3249 * Since algorithm is based on heuristics, accessing another VCPU data without
3250 * locking does not harm. It may result in trying to yield to same VCPU, fail
3251 * and continue with next VCPU and so on.
3253 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3255 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3258 eligible = !vcpu->spin_loop.in_spin_loop ||
3259 vcpu->spin_loop.dy_eligible;
3261 if (vcpu->spin_loop.in_spin_loop)
3262 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3271 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3272 * a vcpu_load/vcpu_put pair. However, for most architectures
3273 * kvm_arch_vcpu_runnable does not require vcpu_load.
3275 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3277 return kvm_arch_vcpu_runnable(vcpu);
3280 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3282 if (kvm_arch_dy_runnable(vcpu))
3285 #ifdef CONFIG_KVM_ASYNC_PF
3286 if (!list_empty_careful(&vcpu->async_pf.done))
3293 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3298 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3300 struct kvm *kvm = me->kvm;
3301 struct kvm_vcpu *vcpu;
3302 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3308 kvm_vcpu_set_in_spin_loop(me, true);
3310 * We boost the priority of a VCPU that is runnable but not
3311 * currently running, because it got preempted by something
3312 * else and called schedule in __vcpu_run. Hopefully that
3313 * VCPU is holding the lock that we need and will release it.
3314 * We approximate round-robin by starting at the last boosted VCPU.
3316 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3317 kvm_for_each_vcpu(i, vcpu, kvm) {
3318 if (!pass && i <= last_boosted_vcpu) {
3319 i = last_boosted_vcpu;
3321 } else if (pass && i > last_boosted_vcpu)
3323 if (!READ_ONCE(vcpu->ready))
3327 if (rcuwait_active(&vcpu->wait) &&
3328 !vcpu_dy_runnable(vcpu))
3330 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3331 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3332 !kvm_arch_vcpu_in_kernel(vcpu))
3334 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3337 yielded = kvm_vcpu_yield_to(vcpu);
3339 kvm->last_boosted_vcpu = i;
3341 } else if (yielded < 0) {
3348 kvm_vcpu_set_in_spin_loop(me, false);
3350 /* Ensure vcpu is not eligible during next spinloop */
3351 kvm_vcpu_set_dy_eligible(me, false);
3353 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3355 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3357 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3358 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3359 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3360 kvm->dirty_ring_size / PAGE_SIZE);
3366 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3368 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3371 if (vmf->pgoff == 0)
3372 page = virt_to_page(vcpu->run);
3374 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3375 page = virt_to_page(vcpu->arch.pio_data);
3377 #ifdef CONFIG_KVM_MMIO
3378 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3379 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3381 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3382 page = kvm_dirty_ring_get_page(
3384 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3386 return kvm_arch_vcpu_fault(vcpu, vmf);
3392 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3393 .fault = kvm_vcpu_fault,
3396 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3398 struct kvm_vcpu *vcpu = file->private_data;
3399 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3401 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3402 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3403 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3406 vma->vm_ops = &kvm_vcpu_vm_ops;
3410 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3412 struct kvm_vcpu *vcpu = filp->private_data;
3414 kvm_put_kvm(vcpu->kvm);
3418 static struct file_operations kvm_vcpu_fops = {
3419 .release = kvm_vcpu_release,
3420 .unlocked_ioctl = kvm_vcpu_ioctl,
3421 .mmap = kvm_vcpu_mmap,
3422 .llseek = noop_llseek,
3423 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3427 * Allocates an inode for the vcpu.
3429 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3431 char name[8 + 1 + ITOA_MAX_LEN + 1];
3433 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3434 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3437 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3439 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3440 struct dentry *debugfs_dentry;
3441 char dir_name[ITOA_MAX_LEN * 2];
3443 if (!debugfs_initialized())
3446 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3447 debugfs_dentry = debugfs_create_dir(dir_name,
3448 vcpu->kvm->debugfs_dentry);
3450 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3455 * Creates some virtual cpus. Good luck creating more than one.
3457 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3460 struct kvm_vcpu *vcpu;
3463 if (id >= KVM_MAX_VCPU_ID)
3466 mutex_lock(&kvm->lock);
3467 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3468 mutex_unlock(&kvm->lock);
3472 kvm->created_vcpus++;
3473 mutex_unlock(&kvm->lock);
3475 r = kvm_arch_vcpu_precreate(kvm, id);
3477 goto vcpu_decrement;
3479 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3482 goto vcpu_decrement;
3485 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3486 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3491 vcpu->run = page_address(page);
3493 kvm_vcpu_init(vcpu, kvm, id);
3495 r = kvm_arch_vcpu_create(vcpu);
3497 goto vcpu_free_run_page;
3499 if (kvm->dirty_ring_size) {
3500 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3501 id, kvm->dirty_ring_size);
3503 goto arch_vcpu_destroy;
3506 mutex_lock(&kvm->lock);
3507 if (kvm_get_vcpu_by_id(kvm, id)) {
3509 goto unlock_vcpu_destroy;
3512 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3513 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3515 /* Fill the stats id string for the vcpu */
3516 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3517 task_pid_nr(current), id);
3519 /* Now it's all set up, let userspace reach it */
3521 r = create_vcpu_fd(vcpu);
3523 kvm_put_kvm_no_destroy(kvm);
3524 goto unlock_vcpu_destroy;
3527 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3530 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3531 * before kvm->online_vcpu's incremented value.
3534 atomic_inc(&kvm->online_vcpus);
3536 mutex_unlock(&kvm->lock);
3537 kvm_arch_vcpu_postcreate(vcpu);
3538 kvm_create_vcpu_debugfs(vcpu);
3541 unlock_vcpu_destroy:
3542 mutex_unlock(&kvm->lock);
3543 kvm_dirty_ring_free(&vcpu->dirty_ring);
3545 kvm_arch_vcpu_destroy(vcpu);
3547 free_page((unsigned long)vcpu->run);
3549 kmem_cache_free(kvm_vcpu_cache, vcpu);
3551 mutex_lock(&kvm->lock);
3552 kvm->created_vcpus--;
3553 mutex_unlock(&kvm->lock);
3557 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3560 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3561 vcpu->sigset_active = 1;
3562 vcpu->sigset = *sigset;
3564 vcpu->sigset_active = 0;
3568 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3569 size_t size, loff_t *offset)
3571 struct kvm_vcpu *vcpu = file->private_data;
3573 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3574 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3575 sizeof(vcpu->stat), user_buffer, size, offset);
3578 static const struct file_operations kvm_vcpu_stats_fops = {
3579 .read = kvm_vcpu_stats_read,
3580 .llseek = noop_llseek,
3583 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3587 char name[15 + ITOA_MAX_LEN + 1];
3589 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3591 fd = get_unused_fd_flags(O_CLOEXEC);
3595 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3598 return PTR_ERR(file);
3600 file->f_mode |= FMODE_PREAD;
3601 fd_install(fd, file);
3606 static long kvm_vcpu_ioctl(struct file *filp,
3607 unsigned int ioctl, unsigned long arg)
3609 struct kvm_vcpu *vcpu = filp->private_data;
3610 void __user *argp = (void __user *)arg;
3612 struct kvm_fpu *fpu = NULL;
3613 struct kvm_sregs *kvm_sregs = NULL;
3615 if (vcpu->kvm->mm != current->mm)
3618 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3622 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3623 * execution; mutex_lock() would break them.
3625 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3626 if (r != -ENOIOCTLCMD)
3629 if (mutex_lock_killable(&vcpu->mutex))
3637 oldpid = rcu_access_pointer(vcpu->pid);
3638 if (unlikely(oldpid != task_pid(current))) {
3639 /* The thread running this VCPU changed. */
3642 r = kvm_arch_vcpu_run_pid_change(vcpu);
3646 newpid = get_task_pid(current, PIDTYPE_PID);
3647 rcu_assign_pointer(vcpu->pid, newpid);
3652 r = kvm_arch_vcpu_ioctl_run(vcpu);
3653 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3656 case KVM_GET_REGS: {
3657 struct kvm_regs *kvm_regs;
3660 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3663 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3667 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3674 case KVM_SET_REGS: {
3675 struct kvm_regs *kvm_regs;
3677 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3678 if (IS_ERR(kvm_regs)) {
3679 r = PTR_ERR(kvm_regs);
3682 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3686 case KVM_GET_SREGS: {
3687 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3688 GFP_KERNEL_ACCOUNT);
3692 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3696 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3701 case KVM_SET_SREGS: {
3702 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3703 if (IS_ERR(kvm_sregs)) {
3704 r = PTR_ERR(kvm_sregs);
3708 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3711 case KVM_GET_MP_STATE: {
3712 struct kvm_mp_state mp_state;
3714 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3718 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3723 case KVM_SET_MP_STATE: {
3724 struct kvm_mp_state mp_state;
3727 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3729 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3732 case KVM_TRANSLATE: {
3733 struct kvm_translation tr;
3736 if (copy_from_user(&tr, argp, sizeof(tr)))
3738 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3742 if (copy_to_user(argp, &tr, sizeof(tr)))
3747 case KVM_SET_GUEST_DEBUG: {
3748 struct kvm_guest_debug dbg;
3751 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3753 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3756 case KVM_SET_SIGNAL_MASK: {
3757 struct kvm_signal_mask __user *sigmask_arg = argp;
3758 struct kvm_signal_mask kvm_sigmask;
3759 sigset_t sigset, *p;
3764 if (copy_from_user(&kvm_sigmask, argp,
3765 sizeof(kvm_sigmask)))
3768 if (kvm_sigmask.len != sizeof(sigset))
3771 if (copy_from_user(&sigset, sigmask_arg->sigset,
3776 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3780 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3784 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3788 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3794 fpu = memdup_user(argp, sizeof(*fpu));
3800 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3803 case KVM_GET_STATS_FD: {
3804 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
3808 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3811 mutex_unlock(&vcpu->mutex);
3817 #ifdef CONFIG_KVM_COMPAT
3818 static long kvm_vcpu_compat_ioctl(struct file *filp,
3819 unsigned int ioctl, unsigned long arg)
3821 struct kvm_vcpu *vcpu = filp->private_data;
3822 void __user *argp = compat_ptr(arg);
3825 if (vcpu->kvm->mm != current->mm)
3829 case KVM_SET_SIGNAL_MASK: {
3830 struct kvm_signal_mask __user *sigmask_arg = argp;
3831 struct kvm_signal_mask kvm_sigmask;
3836 if (copy_from_user(&kvm_sigmask, argp,
3837 sizeof(kvm_sigmask)))
3840 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3843 if (get_compat_sigset(&sigset,
3844 (compat_sigset_t __user *)sigmask_arg->sigset))
3846 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3848 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3852 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3860 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3862 struct kvm_device *dev = filp->private_data;
3865 return dev->ops->mmap(dev, vma);
3870 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3871 int (*accessor)(struct kvm_device *dev,
3872 struct kvm_device_attr *attr),
3875 struct kvm_device_attr attr;
3880 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3883 return accessor(dev, &attr);
3886 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3889 struct kvm_device *dev = filp->private_data;
3891 if (dev->kvm->mm != current->mm)
3895 case KVM_SET_DEVICE_ATTR:
3896 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3897 case KVM_GET_DEVICE_ATTR:
3898 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3899 case KVM_HAS_DEVICE_ATTR:
3900 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3902 if (dev->ops->ioctl)
3903 return dev->ops->ioctl(dev, ioctl, arg);
3909 static int kvm_device_release(struct inode *inode, struct file *filp)
3911 struct kvm_device *dev = filp->private_data;
3912 struct kvm *kvm = dev->kvm;
3914 if (dev->ops->release) {
3915 mutex_lock(&kvm->lock);
3916 list_del(&dev->vm_node);
3917 dev->ops->release(dev);
3918 mutex_unlock(&kvm->lock);
3925 static const struct file_operations kvm_device_fops = {
3926 .unlocked_ioctl = kvm_device_ioctl,
3927 .release = kvm_device_release,
3928 KVM_COMPAT(kvm_device_ioctl),
3929 .mmap = kvm_device_mmap,
3932 struct kvm_device *kvm_device_from_filp(struct file *filp)
3934 if (filp->f_op != &kvm_device_fops)
3937 return filp->private_data;
3940 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3941 #ifdef CONFIG_KVM_MPIC
3942 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3943 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3947 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3949 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3952 if (kvm_device_ops_table[type] != NULL)
3955 kvm_device_ops_table[type] = ops;
3959 void kvm_unregister_device_ops(u32 type)
3961 if (kvm_device_ops_table[type] != NULL)
3962 kvm_device_ops_table[type] = NULL;
3965 static int kvm_ioctl_create_device(struct kvm *kvm,
3966 struct kvm_create_device *cd)
3968 const struct kvm_device_ops *ops = NULL;
3969 struct kvm_device *dev;
3970 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3974 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3977 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3978 ops = kvm_device_ops_table[type];
3985 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3992 mutex_lock(&kvm->lock);
3993 ret = ops->create(dev, type);
3995 mutex_unlock(&kvm->lock);
3999 list_add(&dev->vm_node, &kvm->devices);
4000 mutex_unlock(&kvm->lock);
4006 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4008 kvm_put_kvm_no_destroy(kvm);
4009 mutex_lock(&kvm->lock);
4010 list_del(&dev->vm_node);
4011 mutex_unlock(&kvm->lock);
4020 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4023 case KVM_CAP_USER_MEMORY:
4024 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4025 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4026 case KVM_CAP_INTERNAL_ERROR_DATA:
4027 #ifdef CONFIG_HAVE_KVM_MSI
4028 case KVM_CAP_SIGNAL_MSI:
4030 #ifdef CONFIG_HAVE_KVM_IRQFD
4032 case KVM_CAP_IRQFD_RESAMPLE:
4034 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4035 case KVM_CAP_CHECK_EXTENSION_VM:
4036 case KVM_CAP_ENABLE_CAP_VM:
4037 case KVM_CAP_HALT_POLL:
4039 #ifdef CONFIG_KVM_MMIO
4040 case KVM_CAP_COALESCED_MMIO:
4041 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4042 case KVM_CAP_COALESCED_PIO:
4045 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4046 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4047 return KVM_DIRTY_LOG_MANUAL_CAPS;
4049 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4050 case KVM_CAP_IRQ_ROUTING:
4051 return KVM_MAX_IRQ_ROUTES;
4053 #if KVM_ADDRESS_SPACE_NUM > 1
4054 case KVM_CAP_MULTI_ADDRESS_SPACE:
4055 return KVM_ADDRESS_SPACE_NUM;
4057 case KVM_CAP_NR_MEMSLOTS:
4058 return KVM_USER_MEM_SLOTS;
4059 case KVM_CAP_DIRTY_LOG_RING:
4060 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4061 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4065 case KVM_CAP_BINARY_STATS_FD:
4070 return kvm_vm_ioctl_check_extension(kvm, arg);
4073 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4077 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4080 /* the size should be power of 2 */
4081 if (!size || (size & (size - 1)))
4084 /* Should be bigger to keep the reserved entries, or a page */
4085 if (size < kvm_dirty_ring_get_rsvd_entries() *
4086 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4089 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4090 sizeof(struct kvm_dirty_gfn))
4093 /* We only allow it to set once */
4094 if (kvm->dirty_ring_size)
4097 mutex_lock(&kvm->lock);
4099 if (kvm->created_vcpus) {
4100 /* We don't allow to change this value after vcpu created */
4103 kvm->dirty_ring_size = size;
4107 mutex_unlock(&kvm->lock);
4111 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4114 struct kvm_vcpu *vcpu;
4117 if (!kvm->dirty_ring_size)
4120 mutex_lock(&kvm->slots_lock);
4122 kvm_for_each_vcpu(i, vcpu, kvm)
4123 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4125 mutex_unlock(&kvm->slots_lock);
4128 kvm_flush_remote_tlbs(kvm);
4133 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4134 struct kvm_enable_cap *cap)
4139 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4140 struct kvm_enable_cap *cap)
4143 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4144 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4145 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4147 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4148 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4150 if (cap->flags || (cap->args[0] & ~allowed_options))
4152 kvm->manual_dirty_log_protect = cap->args[0];
4156 case KVM_CAP_HALT_POLL: {
4157 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4160 kvm->max_halt_poll_ns = cap->args[0];
4163 case KVM_CAP_DIRTY_LOG_RING:
4164 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4166 return kvm_vm_ioctl_enable_cap(kvm, cap);
4170 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4171 size_t size, loff_t *offset)
4173 struct kvm *kvm = file->private_data;
4175 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4176 &kvm_vm_stats_desc[0], &kvm->stat,
4177 sizeof(kvm->stat), user_buffer, size, offset);
4180 static const struct file_operations kvm_vm_stats_fops = {
4181 .read = kvm_vm_stats_read,
4182 .llseek = noop_llseek,
4185 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4190 fd = get_unused_fd_flags(O_CLOEXEC);
4194 file = anon_inode_getfile("kvm-vm-stats",
4195 &kvm_vm_stats_fops, kvm, O_RDONLY);
4198 return PTR_ERR(file);
4200 file->f_mode |= FMODE_PREAD;
4201 fd_install(fd, file);
4206 static long kvm_vm_ioctl(struct file *filp,
4207 unsigned int ioctl, unsigned long arg)
4209 struct kvm *kvm = filp->private_data;
4210 void __user *argp = (void __user *)arg;
4213 if (kvm->mm != current->mm)
4216 case KVM_CREATE_VCPU:
4217 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4219 case KVM_ENABLE_CAP: {
4220 struct kvm_enable_cap cap;
4223 if (copy_from_user(&cap, argp, sizeof(cap)))
4225 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4228 case KVM_SET_USER_MEMORY_REGION: {
4229 struct kvm_userspace_memory_region kvm_userspace_mem;
4232 if (copy_from_user(&kvm_userspace_mem, argp,
4233 sizeof(kvm_userspace_mem)))
4236 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4239 case KVM_GET_DIRTY_LOG: {
4240 struct kvm_dirty_log log;
4243 if (copy_from_user(&log, argp, sizeof(log)))
4245 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4248 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4249 case KVM_CLEAR_DIRTY_LOG: {
4250 struct kvm_clear_dirty_log log;
4253 if (copy_from_user(&log, argp, sizeof(log)))
4255 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4259 #ifdef CONFIG_KVM_MMIO
4260 case KVM_REGISTER_COALESCED_MMIO: {
4261 struct kvm_coalesced_mmio_zone zone;
4264 if (copy_from_user(&zone, argp, sizeof(zone)))
4266 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4269 case KVM_UNREGISTER_COALESCED_MMIO: {
4270 struct kvm_coalesced_mmio_zone zone;
4273 if (copy_from_user(&zone, argp, sizeof(zone)))
4275 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4280 struct kvm_irqfd data;
4283 if (copy_from_user(&data, argp, sizeof(data)))
4285 r = kvm_irqfd(kvm, &data);
4288 case KVM_IOEVENTFD: {
4289 struct kvm_ioeventfd data;
4292 if (copy_from_user(&data, argp, sizeof(data)))
4294 r = kvm_ioeventfd(kvm, &data);
4297 #ifdef CONFIG_HAVE_KVM_MSI
4298 case KVM_SIGNAL_MSI: {
4302 if (copy_from_user(&msi, argp, sizeof(msi)))
4304 r = kvm_send_userspace_msi(kvm, &msi);
4308 #ifdef __KVM_HAVE_IRQ_LINE
4309 case KVM_IRQ_LINE_STATUS:
4310 case KVM_IRQ_LINE: {
4311 struct kvm_irq_level irq_event;
4314 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4317 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4318 ioctl == KVM_IRQ_LINE_STATUS);
4323 if (ioctl == KVM_IRQ_LINE_STATUS) {
4324 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4332 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4333 case KVM_SET_GSI_ROUTING: {
4334 struct kvm_irq_routing routing;
4335 struct kvm_irq_routing __user *urouting;
4336 struct kvm_irq_routing_entry *entries = NULL;
4339 if (copy_from_user(&routing, argp, sizeof(routing)))
4342 if (!kvm_arch_can_set_irq_routing(kvm))
4344 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4350 entries = vmemdup_user(urouting->entries,
4351 array_size(sizeof(*entries),
4353 if (IS_ERR(entries)) {
4354 r = PTR_ERR(entries);
4358 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4363 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4364 case KVM_CREATE_DEVICE: {
4365 struct kvm_create_device cd;
4368 if (copy_from_user(&cd, argp, sizeof(cd)))
4371 r = kvm_ioctl_create_device(kvm, &cd);
4376 if (copy_to_user(argp, &cd, sizeof(cd)))
4382 case KVM_CHECK_EXTENSION:
4383 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4385 case KVM_RESET_DIRTY_RINGS:
4386 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4388 case KVM_GET_STATS_FD:
4389 r = kvm_vm_ioctl_get_stats_fd(kvm);
4392 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4398 #ifdef CONFIG_KVM_COMPAT
4399 struct compat_kvm_dirty_log {
4403 compat_uptr_t dirty_bitmap; /* one bit per page */
4408 struct compat_kvm_clear_dirty_log {
4413 compat_uptr_t dirty_bitmap; /* one bit per page */
4418 static long kvm_vm_compat_ioctl(struct file *filp,
4419 unsigned int ioctl, unsigned long arg)
4421 struct kvm *kvm = filp->private_data;
4424 if (kvm->mm != current->mm)
4427 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4428 case KVM_CLEAR_DIRTY_LOG: {
4429 struct compat_kvm_clear_dirty_log compat_log;
4430 struct kvm_clear_dirty_log log;
4432 if (copy_from_user(&compat_log, (void __user *)arg,
4433 sizeof(compat_log)))
4435 log.slot = compat_log.slot;
4436 log.num_pages = compat_log.num_pages;
4437 log.first_page = compat_log.first_page;
4438 log.padding2 = compat_log.padding2;
4439 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4441 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4445 case KVM_GET_DIRTY_LOG: {
4446 struct compat_kvm_dirty_log compat_log;
4447 struct kvm_dirty_log log;
4449 if (copy_from_user(&compat_log, (void __user *)arg,
4450 sizeof(compat_log)))
4452 log.slot = compat_log.slot;
4453 log.padding1 = compat_log.padding1;
4454 log.padding2 = compat_log.padding2;
4455 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4457 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4461 r = kvm_vm_ioctl(filp, ioctl, arg);
4467 static struct file_operations kvm_vm_fops = {
4468 .release = kvm_vm_release,
4469 .unlocked_ioctl = kvm_vm_ioctl,
4470 .llseek = noop_llseek,
4471 KVM_COMPAT(kvm_vm_compat_ioctl),
4474 bool file_is_kvm(struct file *file)
4476 return file && file->f_op == &kvm_vm_fops;
4478 EXPORT_SYMBOL_GPL(file_is_kvm);
4480 static int kvm_dev_ioctl_create_vm(unsigned long type)
4486 kvm = kvm_create_vm(type);
4488 return PTR_ERR(kvm);
4489 #ifdef CONFIG_KVM_MMIO
4490 r = kvm_coalesced_mmio_init(kvm);
4494 r = get_unused_fd_flags(O_CLOEXEC);
4498 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4499 "kvm-%d", task_pid_nr(current));
4501 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4509 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4510 * already set, with ->release() being kvm_vm_release(). In error
4511 * cases it will be called by the final fput(file) and will take
4512 * care of doing kvm_put_kvm(kvm).
4514 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4519 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4521 fd_install(r, file);
4529 static long kvm_dev_ioctl(struct file *filp,
4530 unsigned int ioctl, unsigned long arg)
4535 case KVM_GET_API_VERSION:
4538 r = KVM_API_VERSION;
4541 r = kvm_dev_ioctl_create_vm(arg);
4543 case KVM_CHECK_EXTENSION:
4544 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4546 case KVM_GET_VCPU_MMAP_SIZE:
4549 r = PAGE_SIZE; /* struct kvm_run */
4551 r += PAGE_SIZE; /* pio data page */
4553 #ifdef CONFIG_KVM_MMIO
4554 r += PAGE_SIZE; /* coalesced mmio ring page */
4557 case KVM_TRACE_ENABLE:
4558 case KVM_TRACE_PAUSE:
4559 case KVM_TRACE_DISABLE:
4563 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4569 static struct file_operations kvm_chardev_ops = {
4570 .unlocked_ioctl = kvm_dev_ioctl,
4571 .llseek = noop_llseek,
4572 KVM_COMPAT(kvm_dev_ioctl),
4575 static struct miscdevice kvm_dev = {
4581 static void hardware_enable_nolock(void *junk)
4583 int cpu = raw_smp_processor_id();
4586 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4589 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4591 r = kvm_arch_hardware_enable();
4594 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4595 atomic_inc(&hardware_enable_failed);
4596 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4600 static int kvm_starting_cpu(unsigned int cpu)
4602 raw_spin_lock(&kvm_count_lock);
4603 if (kvm_usage_count)
4604 hardware_enable_nolock(NULL);
4605 raw_spin_unlock(&kvm_count_lock);
4609 static void hardware_disable_nolock(void *junk)
4611 int cpu = raw_smp_processor_id();
4613 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4615 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4616 kvm_arch_hardware_disable();
4619 static int kvm_dying_cpu(unsigned int cpu)
4621 raw_spin_lock(&kvm_count_lock);
4622 if (kvm_usage_count)
4623 hardware_disable_nolock(NULL);
4624 raw_spin_unlock(&kvm_count_lock);
4628 static void hardware_disable_all_nolock(void)
4630 BUG_ON(!kvm_usage_count);
4633 if (!kvm_usage_count)
4634 on_each_cpu(hardware_disable_nolock, NULL, 1);
4637 static void hardware_disable_all(void)
4639 raw_spin_lock(&kvm_count_lock);
4640 hardware_disable_all_nolock();
4641 raw_spin_unlock(&kvm_count_lock);
4644 static int hardware_enable_all(void)
4648 raw_spin_lock(&kvm_count_lock);
4651 if (kvm_usage_count == 1) {
4652 atomic_set(&hardware_enable_failed, 0);
4653 on_each_cpu(hardware_enable_nolock, NULL, 1);
4655 if (atomic_read(&hardware_enable_failed)) {
4656 hardware_disable_all_nolock();
4661 raw_spin_unlock(&kvm_count_lock);
4666 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4670 * Some (well, at least mine) BIOSes hang on reboot if
4673 * And Intel TXT required VMX off for all cpu when system shutdown.
4675 pr_info("kvm: exiting hardware virtualization\n");
4676 kvm_rebooting = true;
4677 on_each_cpu(hardware_disable_nolock, NULL, 1);
4681 static struct notifier_block kvm_reboot_notifier = {
4682 .notifier_call = kvm_reboot,
4686 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4690 for (i = 0; i < bus->dev_count; i++) {
4691 struct kvm_io_device *pos = bus->range[i].dev;
4693 kvm_iodevice_destructor(pos);
4698 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4699 const struct kvm_io_range *r2)
4701 gpa_t addr1 = r1->addr;
4702 gpa_t addr2 = r2->addr;
4707 /* If r2->len == 0, match the exact address. If r2->len != 0,
4708 * accept any overlapping write. Any order is acceptable for
4709 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4710 * we process all of them.
4723 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4725 return kvm_io_bus_cmp(p1, p2);
4728 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4729 gpa_t addr, int len)
4731 struct kvm_io_range *range, key;
4734 key = (struct kvm_io_range) {
4739 range = bsearch(&key, bus->range, bus->dev_count,
4740 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4744 off = range - bus->range;
4746 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4752 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4753 struct kvm_io_range *range, const void *val)
4757 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4761 while (idx < bus->dev_count &&
4762 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4763 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4772 /* kvm_io_bus_write - called under kvm->slots_lock */
4773 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4774 int len, const void *val)
4776 struct kvm_io_bus *bus;
4777 struct kvm_io_range range;
4780 range = (struct kvm_io_range) {
4785 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4788 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4789 return r < 0 ? r : 0;
4791 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4793 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4794 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4795 gpa_t addr, int len, const void *val, long cookie)
4797 struct kvm_io_bus *bus;
4798 struct kvm_io_range range;
4800 range = (struct kvm_io_range) {
4805 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4809 /* First try the device referenced by cookie. */
4810 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4811 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4812 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4817 * cookie contained garbage; fall back to search and return the
4818 * correct cookie value.
4820 return __kvm_io_bus_write(vcpu, bus, &range, val);
4823 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4824 struct kvm_io_range *range, void *val)
4828 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4832 while (idx < bus->dev_count &&
4833 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4834 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4843 /* kvm_io_bus_read - called under kvm->slots_lock */
4844 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4847 struct kvm_io_bus *bus;
4848 struct kvm_io_range range;
4851 range = (struct kvm_io_range) {
4856 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4859 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4860 return r < 0 ? r : 0;
4863 /* Caller must hold slots_lock. */
4864 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4865 int len, struct kvm_io_device *dev)
4868 struct kvm_io_bus *new_bus, *bus;
4869 struct kvm_io_range range;
4871 bus = kvm_get_bus(kvm, bus_idx);
4875 /* exclude ioeventfd which is limited by maximum fd */
4876 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4879 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4880 GFP_KERNEL_ACCOUNT);
4884 range = (struct kvm_io_range) {
4890 for (i = 0; i < bus->dev_count; i++)
4891 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4894 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4895 new_bus->dev_count++;
4896 new_bus->range[i] = range;
4897 memcpy(new_bus->range + i + 1, bus->range + i,
4898 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4899 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4900 synchronize_srcu_expedited(&kvm->srcu);
4906 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4907 struct kvm_io_device *dev)
4910 struct kvm_io_bus *new_bus, *bus;
4912 lockdep_assert_held(&kvm->slots_lock);
4914 bus = kvm_get_bus(kvm, bus_idx);
4918 for (i = 0; i < bus->dev_count; i++) {
4919 if (bus->range[i].dev == dev) {
4924 if (i == bus->dev_count)
4927 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4928 GFP_KERNEL_ACCOUNT);
4930 memcpy(new_bus, bus, struct_size(bus, range, i));
4931 new_bus->dev_count--;
4932 memcpy(new_bus->range + i, bus->range + i + 1,
4933 flex_array_size(new_bus, range, new_bus->dev_count - i));
4936 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4937 synchronize_srcu_expedited(&kvm->srcu);
4939 /* Destroy the old bus _after_ installing the (null) bus. */
4941 pr_err("kvm: failed to shrink bus, removing it completely\n");
4942 for (j = 0; j < bus->dev_count; j++) {
4945 kvm_iodevice_destructor(bus->range[j].dev);
4950 return new_bus ? 0 : -ENOMEM;
4953 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4956 struct kvm_io_bus *bus;
4957 int dev_idx, srcu_idx;
4958 struct kvm_io_device *iodev = NULL;
4960 srcu_idx = srcu_read_lock(&kvm->srcu);
4962 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4966 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4970 iodev = bus->range[dev_idx].dev;
4973 srcu_read_unlock(&kvm->srcu, srcu_idx);
4977 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4979 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4980 int (*get)(void *, u64 *), int (*set)(void *, u64),
4983 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4986 /* The debugfs files are a reference to the kvm struct which
4987 * is still valid when kvm_destroy_vm is called.
4988 * To avoid the race between open and the removal of the debugfs
4989 * directory we test against the users count.
4991 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4994 if (simple_attr_open(inode, file, get,
4995 kvm_stats_debugfs_mode(stat_data->desc) & 0222
4998 kvm_put_kvm(stat_data->kvm);
5005 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5007 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5010 simple_attr_release(inode, file);
5011 kvm_put_kvm(stat_data->kvm);
5016 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5018 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5023 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5025 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5030 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5033 struct kvm_vcpu *vcpu;
5037 kvm_for_each_vcpu(i, vcpu, kvm)
5038 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5043 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5046 struct kvm_vcpu *vcpu;
5048 kvm_for_each_vcpu(i, vcpu, kvm)
5049 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5054 static int kvm_stat_data_get(void *data, u64 *val)
5057 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5059 switch (stat_data->kind) {
5061 r = kvm_get_stat_per_vm(stat_data->kvm,
5062 stat_data->desc->desc.offset, val);
5065 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5066 stat_data->desc->desc.offset, val);
5073 static int kvm_stat_data_clear(void *data, u64 val)
5076 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5081 switch (stat_data->kind) {
5083 r = kvm_clear_stat_per_vm(stat_data->kvm,
5084 stat_data->desc->desc.offset);
5087 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5088 stat_data->desc->desc.offset);
5095 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5097 __simple_attr_check_format("%llu\n", 0ull);
5098 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5099 kvm_stat_data_clear, "%llu\n");
5102 static const struct file_operations stat_fops_per_vm = {
5103 .owner = THIS_MODULE,
5104 .open = kvm_stat_data_open,
5105 .release = kvm_debugfs_release,
5106 .read = simple_attr_read,
5107 .write = simple_attr_write,
5108 .llseek = no_llseek,
5111 static int vm_stat_get(void *_offset, u64 *val)
5113 unsigned offset = (long)_offset;
5118 mutex_lock(&kvm_lock);
5119 list_for_each_entry(kvm, &vm_list, vm_list) {
5120 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5123 mutex_unlock(&kvm_lock);
5127 static int vm_stat_clear(void *_offset, u64 val)
5129 unsigned offset = (long)_offset;
5135 mutex_lock(&kvm_lock);
5136 list_for_each_entry(kvm, &vm_list, vm_list) {
5137 kvm_clear_stat_per_vm(kvm, offset);
5139 mutex_unlock(&kvm_lock);
5144 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5145 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5147 static int vcpu_stat_get(void *_offset, u64 *val)
5149 unsigned offset = (long)_offset;
5154 mutex_lock(&kvm_lock);
5155 list_for_each_entry(kvm, &vm_list, vm_list) {
5156 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5159 mutex_unlock(&kvm_lock);
5163 static int vcpu_stat_clear(void *_offset, u64 val)
5165 unsigned offset = (long)_offset;
5171 mutex_lock(&kvm_lock);
5172 list_for_each_entry(kvm, &vm_list, vm_list) {
5173 kvm_clear_stat_per_vcpu(kvm, offset);
5175 mutex_unlock(&kvm_lock);
5180 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5182 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5184 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5186 struct kobj_uevent_env *env;
5187 unsigned long long created, active;
5189 if (!kvm_dev.this_device || !kvm)
5192 mutex_lock(&kvm_lock);
5193 if (type == KVM_EVENT_CREATE_VM) {
5194 kvm_createvm_count++;
5196 } else if (type == KVM_EVENT_DESTROY_VM) {
5199 created = kvm_createvm_count;
5200 active = kvm_active_vms;
5201 mutex_unlock(&kvm_lock);
5203 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5207 add_uevent_var(env, "CREATED=%llu", created);
5208 add_uevent_var(env, "COUNT=%llu", active);
5210 if (type == KVM_EVENT_CREATE_VM) {
5211 add_uevent_var(env, "EVENT=create");
5212 kvm->userspace_pid = task_pid_nr(current);
5213 } else if (type == KVM_EVENT_DESTROY_VM) {
5214 add_uevent_var(env, "EVENT=destroy");
5216 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5218 if (kvm->debugfs_dentry) {
5219 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5222 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5224 add_uevent_var(env, "STATS_PATH=%s", tmp);
5228 /* no need for checks, since we are adding at most only 5 keys */
5229 env->envp[env->envp_idx++] = NULL;
5230 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5234 static void kvm_init_debug(void)
5236 const struct file_operations *fops;
5237 const struct _kvm_stats_desc *pdesc;
5240 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5242 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5243 pdesc = &kvm_vm_stats_desc[i];
5244 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5245 fops = &vm_stat_fops;
5247 fops = &vm_stat_readonly_fops;
5248 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5250 (void *)(long)pdesc->desc.offset, fops);
5253 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5254 pdesc = &kvm_vcpu_stats_desc[i];
5255 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5256 fops = &vcpu_stat_fops;
5258 fops = &vcpu_stat_readonly_fops;
5259 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5261 (void *)(long)pdesc->desc.offset, fops);
5265 static int kvm_suspend(void)
5267 if (kvm_usage_count)
5268 hardware_disable_nolock(NULL);
5272 static void kvm_resume(void)
5274 if (kvm_usage_count) {
5275 #ifdef CONFIG_LOCKDEP
5276 WARN_ON(lockdep_is_held(&kvm_count_lock));
5278 hardware_enable_nolock(NULL);
5282 static struct syscore_ops kvm_syscore_ops = {
5283 .suspend = kvm_suspend,
5284 .resume = kvm_resume,
5288 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5290 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5293 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5295 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5297 WRITE_ONCE(vcpu->preempted, false);
5298 WRITE_ONCE(vcpu->ready, false);
5300 __this_cpu_write(kvm_running_vcpu, vcpu);
5301 kvm_arch_sched_in(vcpu, cpu);
5302 kvm_arch_vcpu_load(vcpu, cpu);
5305 static void kvm_sched_out(struct preempt_notifier *pn,
5306 struct task_struct *next)
5308 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5310 if (current->on_rq) {
5311 WRITE_ONCE(vcpu->preempted, true);
5312 WRITE_ONCE(vcpu->ready, true);
5314 kvm_arch_vcpu_put(vcpu);
5315 __this_cpu_write(kvm_running_vcpu, NULL);
5319 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5321 * We can disable preemption locally around accessing the per-CPU variable,
5322 * and use the resolved vcpu pointer after enabling preemption again,
5323 * because even if the current thread is migrated to another CPU, reading
5324 * the per-CPU value later will give us the same value as we update the
5325 * per-CPU variable in the preempt notifier handlers.
5327 struct kvm_vcpu *kvm_get_running_vcpu(void)
5329 struct kvm_vcpu *vcpu;
5332 vcpu = __this_cpu_read(kvm_running_vcpu);
5337 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5340 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5342 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5344 return &kvm_running_vcpu;
5347 struct kvm_cpu_compat_check {
5352 static void check_processor_compat(void *data)
5354 struct kvm_cpu_compat_check *c = data;
5356 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5359 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5360 struct module *module)
5362 struct kvm_cpu_compat_check c;
5366 r = kvm_arch_init(opaque);
5371 * kvm_arch_init makes sure there's at most one caller
5372 * for architectures that support multiple implementations,
5373 * like intel and amd on x86.
5374 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5375 * conflicts in case kvm is already setup for another implementation.
5377 r = kvm_irqfd_init();
5381 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5386 r = kvm_arch_hardware_setup(opaque);
5392 for_each_online_cpu(cpu) {
5393 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5398 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5399 kvm_starting_cpu, kvm_dying_cpu);
5402 register_reboot_notifier(&kvm_reboot_notifier);
5404 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5406 vcpu_align = __alignof__(struct kvm_vcpu);
5408 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5410 offsetof(struct kvm_vcpu, arch),
5411 offsetofend(struct kvm_vcpu, stats_id)
5412 - offsetof(struct kvm_vcpu, arch),
5414 if (!kvm_vcpu_cache) {
5419 r = kvm_async_pf_init();
5423 kvm_chardev_ops.owner = module;
5424 kvm_vm_fops.owner = module;
5425 kvm_vcpu_fops.owner = module;
5427 r = misc_register(&kvm_dev);
5429 pr_err("kvm: misc device register failed\n");
5433 register_syscore_ops(&kvm_syscore_ops);
5435 kvm_preempt_ops.sched_in = kvm_sched_in;
5436 kvm_preempt_ops.sched_out = kvm_sched_out;
5440 r = kvm_vfio_ops_init();
5446 kvm_async_pf_deinit();
5448 kmem_cache_destroy(kvm_vcpu_cache);
5450 unregister_reboot_notifier(&kvm_reboot_notifier);
5451 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5453 kvm_arch_hardware_unsetup();
5455 free_cpumask_var(cpus_hardware_enabled);
5463 EXPORT_SYMBOL_GPL(kvm_init);
5467 debugfs_remove_recursive(kvm_debugfs_dir);
5468 misc_deregister(&kvm_dev);
5469 kmem_cache_destroy(kvm_vcpu_cache);
5470 kvm_async_pf_deinit();
5471 unregister_syscore_ops(&kvm_syscore_ops);
5472 unregister_reboot_notifier(&kvm_reboot_notifier);
5473 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5474 on_each_cpu(hardware_disable_nolock, NULL, 1);
5475 kvm_arch_hardware_unsetup();
5478 free_cpumask_var(cpus_hardware_enabled);
5479 kvm_vfio_ops_exit();
5481 EXPORT_SYMBOL_GPL(kvm_exit);
5483 struct kvm_vm_worker_thread_context {
5485 struct task_struct *parent;
5486 struct completion init_done;
5487 kvm_vm_thread_fn_t thread_fn;
5492 static int kvm_vm_worker_thread(void *context)
5495 * The init_context is allocated on the stack of the parent thread, so
5496 * we have to locally copy anything that is needed beyond initialization
5498 struct kvm_vm_worker_thread_context *init_context = context;
5499 struct kvm *kvm = init_context->kvm;
5500 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5501 uintptr_t data = init_context->data;
5504 err = kthread_park(current);
5505 /* kthread_park(current) is never supposed to return an error */
5510 err = cgroup_attach_task_all(init_context->parent, current);
5512 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5517 set_user_nice(current, task_nice(init_context->parent));
5520 init_context->err = err;
5521 complete(&init_context->init_done);
5522 init_context = NULL;
5527 /* Wait to be woken up by the spawner before proceeding. */
5530 if (!kthread_should_stop())
5531 err = thread_fn(kvm, data);
5536 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5537 uintptr_t data, const char *name,
5538 struct task_struct **thread_ptr)
5540 struct kvm_vm_worker_thread_context init_context = {};
5541 struct task_struct *thread;
5544 init_context.kvm = kvm;
5545 init_context.parent = current;
5546 init_context.thread_fn = thread_fn;
5547 init_context.data = data;
5548 init_completion(&init_context.init_done);
5550 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5551 "%s-%d", name, task_pid_nr(current));
5553 return PTR_ERR(thread);
5555 /* kthread_run is never supposed to return NULL */
5556 WARN_ON(thread == NULL);
5558 wait_for_completion(&init_context.init_done);
5560 if (!init_context.err)
5561 *thread_ptr = thread;
5563 return init_context.err;