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 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
500 slots = __kvm_memslots(kvm, i);
501 kvm_for_each_memslot(slot, slots) {
502 unsigned long hva_start, hva_end;
504 hva_start = max(range->start, slot->userspace_addr);
505 hva_end = min(range->end, slot->userspace_addr +
506 (slot->npages << PAGE_SHIFT));
507 if (hva_start >= hva_end)
511 * To optimize for the likely case where the address
512 * range is covered by zero or one memslots, don't
513 * bother making these conditional (to avoid writes on
514 * the second or later invocation of the handler).
516 gfn_range.pte = range->pte;
517 gfn_range.may_block = range->may_block;
520 * {gfn(page) | page intersects with [hva_start, hva_end)} =
521 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
523 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
524 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
525 gfn_range.slot = slot;
530 if (!IS_KVM_NULL_FN(range->on_lock))
531 range->on_lock(kvm, range->start, range->end);
532 if (IS_KVM_NULL_FN(range->handler))
535 ret |= range->handler(kvm, &gfn_range);
539 if (range->flush_on_ret && (ret || kvm->tlbs_dirty))
540 kvm_flush_remote_tlbs(kvm);
545 srcu_read_unlock(&kvm->srcu, idx);
547 /* The notifiers are averse to booleans. :-( */
551 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
555 hva_handler_t handler)
557 struct kvm *kvm = mmu_notifier_to_kvm(mn);
558 const struct kvm_hva_range range = {
563 .on_lock = (void *)kvm_null_fn,
564 .flush_on_ret = true,
568 return __kvm_handle_hva_range(kvm, &range);
571 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
574 hva_handler_t handler)
576 struct kvm *kvm = mmu_notifier_to_kvm(mn);
577 const struct kvm_hva_range range = {
582 .on_lock = (void *)kvm_null_fn,
583 .flush_on_ret = false,
587 return __kvm_handle_hva_range(kvm, &range);
589 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
590 struct mm_struct *mm,
591 unsigned long address,
594 struct kvm *kvm = mmu_notifier_to_kvm(mn);
596 trace_kvm_set_spte_hva(address);
599 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
600 * If mmu_notifier_count is zero, then no in-progress invalidations,
601 * including this one, found a relevant memslot at start(); rechecking
602 * memslots here is unnecessary. Note, a false positive (count elevated
603 * by a different invalidation) is sub-optimal but functionally ok.
605 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
606 if (!READ_ONCE(kvm->mmu_notifier_count))
609 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
612 static void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
616 * The count increase must become visible at unlock time as no
617 * spte can be established without taking the mmu_lock and
618 * count is also read inside the mmu_lock critical section.
620 kvm->mmu_notifier_count++;
621 if (likely(kvm->mmu_notifier_count == 1)) {
622 kvm->mmu_notifier_range_start = start;
623 kvm->mmu_notifier_range_end = end;
626 * Fully tracking multiple concurrent ranges has dimishing
627 * returns. Keep things simple and just find the minimal range
628 * which includes the current and new ranges. As there won't be
629 * enough information to subtract a range after its invalidate
630 * completes, any ranges invalidated concurrently will
631 * accumulate and persist until all outstanding invalidates
634 kvm->mmu_notifier_range_start =
635 min(kvm->mmu_notifier_range_start, start);
636 kvm->mmu_notifier_range_end =
637 max(kvm->mmu_notifier_range_end, end);
641 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
642 const struct mmu_notifier_range *range)
644 struct kvm *kvm = mmu_notifier_to_kvm(mn);
645 const struct kvm_hva_range hva_range = {
646 .start = range->start,
649 .handler = kvm_unmap_gfn_range,
650 .on_lock = kvm_inc_notifier_count,
651 .flush_on_ret = true,
652 .may_block = mmu_notifier_range_blockable(range),
655 trace_kvm_unmap_hva_range(range->start, range->end);
658 * Prevent memslot modification between range_start() and range_end()
659 * so that conditionally locking provides the same result in both
660 * functions. Without that guarantee, the mmu_notifier_count
661 * adjustments will be imbalanced.
663 * Pairs with the decrement in range_end().
665 spin_lock(&kvm->mn_invalidate_lock);
666 kvm->mn_active_invalidate_count++;
667 spin_unlock(&kvm->mn_invalidate_lock);
669 __kvm_handle_hva_range(kvm, &hva_range);
674 static void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
678 * This sequence increase will notify the kvm page fault that
679 * the page that is going to be mapped in the spte could have
682 kvm->mmu_notifier_seq++;
685 * The above sequence increase must be visible before the
686 * below count decrease, which is ensured by the smp_wmb above
687 * in conjunction with the smp_rmb in mmu_notifier_retry().
689 kvm->mmu_notifier_count--;
692 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
693 const struct mmu_notifier_range *range)
695 struct kvm *kvm = mmu_notifier_to_kvm(mn);
696 const struct kvm_hva_range hva_range = {
697 .start = range->start,
700 .handler = (void *)kvm_null_fn,
701 .on_lock = kvm_dec_notifier_count,
702 .flush_on_ret = false,
703 .may_block = mmu_notifier_range_blockable(range),
707 __kvm_handle_hva_range(kvm, &hva_range);
709 /* Pairs with the increment in range_start(). */
710 spin_lock(&kvm->mn_invalidate_lock);
711 wake = (--kvm->mn_active_invalidate_count == 0);
712 spin_unlock(&kvm->mn_invalidate_lock);
715 * There can only be one waiter, since the wait happens under
719 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
721 BUG_ON(kvm->mmu_notifier_count < 0);
724 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
725 struct mm_struct *mm,
729 trace_kvm_age_hva(start, end);
731 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
734 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
735 struct mm_struct *mm,
739 trace_kvm_age_hva(start, end);
742 * Even though we do not flush TLB, this will still adversely
743 * affect performance on pre-Haswell Intel EPT, where there is
744 * no EPT Access Bit to clear so that we have to tear down EPT
745 * tables instead. If we find this unacceptable, we can always
746 * add a parameter to kvm_age_hva so that it effectively doesn't
747 * do anything on clear_young.
749 * Also note that currently we never issue secondary TLB flushes
750 * from clear_young, leaving this job up to the regular system
751 * cadence. If we find this inaccurate, we might come up with a
752 * more sophisticated heuristic later.
754 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
757 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
758 struct mm_struct *mm,
759 unsigned long address)
761 trace_kvm_test_age_hva(address);
763 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
767 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
768 struct mm_struct *mm)
770 struct kvm *kvm = mmu_notifier_to_kvm(mn);
773 idx = srcu_read_lock(&kvm->srcu);
774 kvm_arch_flush_shadow_all(kvm);
775 srcu_read_unlock(&kvm->srcu, idx);
778 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
779 .invalidate_range = kvm_mmu_notifier_invalidate_range,
780 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
781 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
782 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
783 .clear_young = kvm_mmu_notifier_clear_young,
784 .test_young = kvm_mmu_notifier_test_young,
785 .change_pte = kvm_mmu_notifier_change_pte,
786 .release = kvm_mmu_notifier_release,
789 static int kvm_init_mmu_notifier(struct kvm *kvm)
791 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
792 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
795 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
797 static int kvm_init_mmu_notifier(struct kvm *kvm)
802 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
804 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
805 static int kvm_pm_notifier_call(struct notifier_block *bl,
809 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
811 return kvm_arch_pm_notifier(kvm, state);
814 static void kvm_init_pm_notifier(struct kvm *kvm)
816 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
817 /* Suspend KVM before we suspend ftrace, RCU, etc. */
818 kvm->pm_notifier.priority = INT_MAX;
819 register_pm_notifier(&kvm->pm_notifier);
822 static void kvm_destroy_pm_notifier(struct kvm *kvm)
824 unregister_pm_notifier(&kvm->pm_notifier);
826 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
827 static void kvm_init_pm_notifier(struct kvm *kvm)
831 static void kvm_destroy_pm_notifier(struct kvm *kvm)
834 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
836 static struct kvm_memslots *kvm_alloc_memslots(void)
839 struct kvm_memslots *slots;
841 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
845 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
846 slots->id_to_index[i] = -1;
851 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
853 if (!memslot->dirty_bitmap)
856 kvfree(memslot->dirty_bitmap);
857 memslot->dirty_bitmap = NULL;
860 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
862 kvm_destroy_dirty_bitmap(slot);
864 kvm_arch_free_memslot(kvm, slot);
870 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
872 struct kvm_memory_slot *memslot;
877 kvm_for_each_memslot(memslot, slots)
878 kvm_free_memslot(kvm, memslot);
883 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
885 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
886 case KVM_STATS_TYPE_INSTANT:
888 case KVM_STATS_TYPE_CUMULATIVE:
889 case KVM_STATS_TYPE_PEAK:
896 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
899 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
900 kvm_vcpu_stats_header.num_desc;
902 if (!kvm->debugfs_dentry)
905 debugfs_remove_recursive(kvm->debugfs_dentry);
907 if (kvm->debugfs_stat_data) {
908 for (i = 0; i < kvm_debugfs_num_entries; i++)
909 kfree(kvm->debugfs_stat_data[i]);
910 kfree(kvm->debugfs_stat_data);
914 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
916 char dir_name[ITOA_MAX_LEN * 2];
917 struct kvm_stat_data *stat_data;
918 const struct _kvm_stats_desc *pdesc;
920 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
921 kvm_vcpu_stats_header.num_desc;
923 if (!debugfs_initialized())
926 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
927 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
929 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
930 sizeof(*kvm->debugfs_stat_data),
932 if (!kvm->debugfs_stat_data)
935 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
936 pdesc = &kvm_vm_stats_desc[i];
937 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
941 stat_data->kvm = kvm;
942 stat_data->desc = pdesc;
943 stat_data->kind = KVM_STAT_VM;
944 kvm->debugfs_stat_data[i] = stat_data;
945 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
946 kvm->debugfs_dentry, stat_data,
950 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
951 pdesc = &kvm_vcpu_stats_desc[i];
952 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
956 stat_data->kvm = kvm;
957 stat_data->desc = pdesc;
958 stat_data->kind = KVM_STAT_VCPU;
959 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
960 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
961 kvm->debugfs_dentry, stat_data,
968 * Called after the VM is otherwise initialized, but just before adding it to
971 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
977 * Called just after removing the VM from the vm_list, but before doing any
980 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
984 static struct kvm *kvm_create_vm(unsigned long type)
986 struct kvm *kvm = kvm_arch_alloc_vm();
991 return ERR_PTR(-ENOMEM);
993 KVM_MMU_LOCK_INIT(kvm);
995 kvm->mm = current->mm;
996 kvm_eventfd_init(kvm);
997 mutex_init(&kvm->lock);
998 mutex_init(&kvm->irq_lock);
999 mutex_init(&kvm->slots_lock);
1000 mutex_init(&kvm->slots_arch_lock);
1001 spin_lock_init(&kvm->mn_invalidate_lock);
1002 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1004 INIT_LIST_HEAD(&kvm->devices);
1006 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1008 if (init_srcu_struct(&kvm->srcu))
1009 goto out_err_no_srcu;
1010 if (init_srcu_struct(&kvm->irq_srcu))
1011 goto out_err_no_irq_srcu;
1013 refcount_set(&kvm->users_count, 1);
1014 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1015 struct kvm_memslots *slots = kvm_alloc_memslots();
1018 goto out_err_no_arch_destroy_vm;
1019 /* Generations must be different for each address space. */
1020 slots->generation = i;
1021 rcu_assign_pointer(kvm->memslots[i], slots);
1024 for (i = 0; i < KVM_NR_BUSES; i++) {
1025 rcu_assign_pointer(kvm->buses[i],
1026 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1028 goto out_err_no_arch_destroy_vm;
1031 kvm->max_halt_poll_ns = halt_poll_ns;
1033 r = kvm_arch_init_vm(kvm, type);
1035 goto out_err_no_arch_destroy_vm;
1037 r = hardware_enable_all();
1039 goto out_err_no_disable;
1041 #ifdef CONFIG_HAVE_KVM_IRQFD
1042 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1045 r = kvm_init_mmu_notifier(kvm);
1047 goto out_err_no_mmu_notifier;
1049 r = kvm_arch_post_init_vm(kvm);
1053 mutex_lock(&kvm_lock);
1054 list_add(&kvm->vm_list, &vm_list);
1055 mutex_unlock(&kvm_lock);
1057 preempt_notifier_inc();
1058 kvm_init_pm_notifier(kvm);
1063 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1064 if (kvm->mmu_notifier.ops)
1065 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1067 out_err_no_mmu_notifier:
1068 hardware_disable_all();
1070 kvm_arch_destroy_vm(kvm);
1071 out_err_no_arch_destroy_vm:
1072 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1073 for (i = 0; i < KVM_NR_BUSES; i++)
1074 kfree(kvm_get_bus(kvm, i));
1075 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1076 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1077 cleanup_srcu_struct(&kvm->irq_srcu);
1078 out_err_no_irq_srcu:
1079 cleanup_srcu_struct(&kvm->srcu);
1081 kvm_arch_free_vm(kvm);
1082 mmdrop(current->mm);
1086 static void kvm_destroy_devices(struct kvm *kvm)
1088 struct kvm_device *dev, *tmp;
1091 * We do not need to take the kvm->lock here, because nobody else
1092 * has a reference to the struct kvm at this point and therefore
1093 * cannot access the devices list anyhow.
1095 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1096 list_del(&dev->vm_node);
1097 dev->ops->destroy(dev);
1101 static void kvm_destroy_vm(struct kvm *kvm)
1104 struct mm_struct *mm = kvm->mm;
1106 kvm_destroy_pm_notifier(kvm);
1107 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1108 kvm_destroy_vm_debugfs(kvm);
1109 kvm_arch_sync_events(kvm);
1110 mutex_lock(&kvm_lock);
1111 list_del(&kvm->vm_list);
1112 mutex_unlock(&kvm_lock);
1113 kvm_arch_pre_destroy_vm(kvm);
1115 kvm_free_irq_routing(kvm);
1116 for (i = 0; i < KVM_NR_BUSES; i++) {
1117 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1120 kvm_io_bus_destroy(bus);
1121 kvm->buses[i] = NULL;
1123 kvm_coalesced_mmio_free(kvm);
1124 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1125 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1127 * At this point, pending calls to invalidate_range_start()
1128 * have completed but no more MMU notifiers will run, so
1129 * mn_active_invalidate_count may remain unbalanced.
1130 * No threads can be waiting in install_new_memslots as the
1131 * last reference on KVM has been dropped, but freeing
1132 * memslots would deadlock without this manual intervention.
1134 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1135 kvm->mn_active_invalidate_count = 0;
1137 kvm_arch_flush_shadow_all(kvm);
1139 kvm_arch_destroy_vm(kvm);
1140 kvm_destroy_devices(kvm);
1141 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1142 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1143 cleanup_srcu_struct(&kvm->irq_srcu);
1144 cleanup_srcu_struct(&kvm->srcu);
1145 kvm_arch_free_vm(kvm);
1146 preempt_notifier_dec();
1147 hardware_disable_all();
1151 void kvm_get_kvm(struct kvm *kvm)
1153 refcount_inc(&kvm->users_count);
1155 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1158 * Make sure the vm is not during destruction, which is a safe version of
1159 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1161 bool kvm_get_kvm_safe(struct kvm *kvm)
1163 return refcount_inc_not_zero(&kvm->users_count);
1165 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1167 void kvm_put_kvm(struct kvm *kvm)
1169 if (refcount_dec_and_test(&kvm->users_count))
1170 kvm_destroy_vm(kvm);
1172 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1175 * Used to put a reference that was taken on behalf of an object associated
1176 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1177 * of the new file descriptor fails and the reference cannot be transferred to
1178 * its final owner. In such cases, the caller is still actively using @kvm and
1179 * will fail miserably if the refcount unexpectedly hits zero.
1181 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1183 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1185 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1187 static int kvm_vm_release(struct inode *inode, struct file *filp)
1189 struct kvm *kvm = filp->private_data;
1191 kvm_irqfd_release(kvm);
1198 * Allocation size is twice as large as the actual dirty bitmap size.
1199 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1201 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1203 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1205 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1206 if (!memslot->dirty_bitmap)
1213 * Delete a memslot by decrementing the number of used slots and shifting all
1214 * other entries in the array forward one spot.
1216 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1217 struct kvm_memory_slot *memslot)
1219 struct kvm_memory_slot *mslots = slots->memslots;
1222 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1225 slots->used_slots--;
1227 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
1228 atomic_set(&slots->lru_slot, 0);
1230 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1231 mslots[i] = mslots[i + 1];
1232 slots->id_to_index[mslots[i].id] = i;
1234 mslots[i] = *memslot;
1235 slots->id_to_index[memslot->id] = -1;
1239 * "Insert" a new memslot by incrementing the number of used slots. Returns
1240 * the new slot's initial index into the memslots array.
1242 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1244 return slots->used_slots++;
1248 * Move a changed memslot backwards in the array by shifting existing slots
1249 * with a higher GFN toward the front of the array. Note, the changed memslot
1250 * itself is not preserved in the array, i.e. not swapped at this time, only
1251 * its new index into the array is tracked. Returns the changed memslot's
1252 * current index into the memslots array.
1254 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1255 struct kvm_memory_slot *memslot)
1257 struct kvm_memory_slot *mslots = slots->memslots;
1260 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1261 WARN_ON_ONCE(!slots->used_slots))
1265 * Move the target memslot backward in the array by shifting existing
1266 * memslots with a higher GFN (than the target memslot) towards the
1267 * front of the array.
1269 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1270 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1273 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1275 /* Shift the next memslot forward one and update its index. */
1276 mslots[i] = mslots[i + 1];
1277 slots->id_to_index[mslots[i].id] = i;
1283 * Move a changed memslot forwards in the array by shifting existing slots with
1284 * a lower GFN toward the back of the array. Note, the changed memslot itself
1285 * is not preserved in the array, i.e. not swapped at this time, only its new
1286 * index into the array is tracked. Returns the changed memslot's final index
1287 * into the memslots array.
1289 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1290 struct kvm_memory_slot *memslot,
1293 struct kvm_memory_slot *mslots = slots->memslots;
1296 for (i = start; i > 0; i--) {
1297 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1300 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1302 /* Shift the next memslot back one and update its index. */
1303 mslots[i] = mslots[i - 1];
1304 slots->id_to_index[mslots[i].id] = i;
1310 * Re-sort memslots based on their GFN to account for an added, deleted, or
1311 * moved memslot. Sorting memslots by GFN allows using a binary search during
1314 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1315 * at memslots[0] has the highest GFN.
1317 * The sorting algorithm takes advantage of having initially sorted memslots
1318 * and knowing the position of the changed memslot. Sorting is also optimized
1319 * by not swapping the updated memslot and instead only shifting other memslots
1320 * and tracking the new index for the update memslot. Only once its final
1321 * index is known is the updated memslot copied into its position in the array.
1323 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1324 * the end of the array.
1326 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1327 * end of the array and then it forward to its correct location.
1329 * - When moving a memslot, the algorithm first moves the updated memslot
1330 * backward to handle the scenario where the memslot's GFN was changed to a
1331 * lower value. update_memslots() then falls through and runs the same flow
1332 * as creating a memslot to move the memslot forward to handle the scenario
1333 * where its GFN was changed to a higher value.
1335 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1336 * historical reasons. Originally, invalid memslots where denoted by having
1337 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1338 * to the end of the array. The current algorithm uses dedicated logic to
1339 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1341 * The other historical motiviation for highest->lowest was to improve the
1342 * performance of memslot lookup. KVM originally used a linear search starting
1343 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1344 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1345 * single memslot above the 4gb boundary. As the largest memslot is also the
1346 * most likely to be referenced, sorting it to the front of the array was
1347 * advantageous. The current binary search starts from the middle of the array
1348 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1350 static void update_memslots(struct kvm_memslots *slots,
1351 struct kvm_memory_slot *memslot,
1352 enum kvm_mr_change change)
1356 if (change == KVM_MR_DELETE) {
1357 kvm_memslot_delete(slots, memslot);
1359 if (change == KVM_MR_CREATE)
1360 i = kvm_memslot_insert_back(slots);
1362 i = kvm_memslot_move_backward(slots, memslot);
1363 i = kvm_memslot_move_forward(slots, memslot, i);
1366 * Copy the memslot to its new position in memslots and update
1367 * its index accordingly.
1369 slots->memslots[i] = *memslot;
1370 slots->id_to_index[memslot->id] = i;
1374 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1376 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1378 #ifdef __KVM_HAVE_READONLY_MEM
1379 valid_flags |= KVM_MEM_READONLY;
1382 if (mem->flags & ~valid_flags)
1388 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1389 int as_id, struct kvm_memslots *slots)
1391 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1392 u64 gen = old_memslots->generation;
1394 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1395 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1398 * Do not store the new memslots while there are invalidations in
1399 * progress, otherwise the locking in invalidate_range_start and
1400 * invalidate_range_end will be unbalanced.
1402 spin_lock(&kvm->mn_invalidate_lock);
1403 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1404 while (kvm->mn_active_invalidate_count) {
1405 set_current_state(TASK_UNINTERRUPTIBLE);
1406 spin_unlock(&kvm->mn_invalidate_lock);
1408 spin_lock(&kvm->mn_invalidate_lock);
1410 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1411 rcu_assign_pointer(kvm->memslots[as_id], slots);
1412 spin_unlock(&kvm->mn_invalidate_lock);
1415 * Acquired in kvm_set_memslot. Must be released before synchronize
1416 * SRCU below in order to avoid deadlock with another thread
1417 * acquiring the slots_arch_lock in an srcu critical section.
1419 mutex_unlock(&kvm->slots_arch_lock);
1421 synchronize_srcu_expedited(&kvm->srcu);
1424 * Increment the new memslot generation a second time, dropping the
1425 * update in-progress flag and incrementing the generation based on
1426 * the number of address spaces. This provides a unique and easily
1427 * identifiable generation number while the memslots are in flux.
1429 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1432 * Generations must be unique even across address spaces. We do not need
1433 * a global counter for that, instead the generation space is evenly split
1434 * across address spaces. For example, with two address spaces, address
1435 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1436 * use generations 1, 3, 5, ...
1438 gen += KVM_ADDRESS_SPACE_NUM;
1440 kvm_arch_memslots_updated(kvm, gen);
1442 slots->generation = gen;
1444 return old_memslots;
1447 static size_t kvm_memslots_size(int slots)
1449 return sizeof(struct kvm_memslots) +
1450 (sizeof(struct kvm_memory_slot) * slots);
1453 static void kvm_copy_memslots(struct kvm_memslots *to,
1454 struct kvm_memslots *from)
1456 memcpy(to, from, kvm_memslots_size(from->used_slots));
1460 * Note, at a minimum, the current number of used slots must be allocated, even
1461 * when deleting a memslot, as we need a complete duplicate of the memslots for
1462 * use when invalidating a memslot prior to deleting/moving the memslot.
1464 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1465 enum kvm_mr_change change)
1467 struct kvm_memslots *slots;
1470 if (change == KVM_MR_CREATE)
1471 new_size = kvm_memslots_size(old->used_slots + 1);
1473 new_size = kvm_memslots_size(old->used_slots);
1475 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1477 kvm_copy_memslots(slots, old);
1482 static int kvm_set_memslot(struct kvm *kvm,
1483 const struct kvm_userspace_memory_region *mem,
1484 struct kvm_memory_slot *old,
1485 struct kvm_memory_slot *new, int as_id,
1486 enum kvm_mr_change change)
1488 struct kvm_memory_slot *slot;
1489 struct kvm_memslots *slots;
1493 * Released in install_new_memslots.
1495 * Must be held from before the current memslots are copied until
1496 * after the new memslots are installed with rcu_assign_pointer,
1497 * then released before the synchronize srcu in install_new_memslots.
1499 * When modifying memslots outside of the slots_lock, must be held
1500 * before reading the pointer to the current memslots until after all
1501 * changes to those memslots are complete.
1503 * These rules ensure that installing new memslots does not lose
1504 * changes made to the previous memslots.
1506 mutex_lock(&kvm->slots_arch_lock);
1508 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1510 mutex_unlock(&kvm->slots_arch_lock);
1514 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1516 * Note, the INVALID flag needs to be in the appropriate entry
1517 * in the freshly allocated memslots, not in @old or @new.
1519 slot = id_to_memslot(slots, old->id);
1520 slot->flags |= KVM_MEMSLOT_INVALID;
1523 * We can re-use the memory from the old memslots.
1524 * It will be overwritten with a copy of the new memslots
1525 * after reacquiring the slots_arch_lock below.
1527 slots = install_new_memslots(kvm, as_id, slots);
1529 /* From this point no new shadow pages pointing to a deleted,
1530 * or moved, memslot will be created.
1532 * validation of sp->gfn happens in:
1533 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1534 * - kvm_is_visible_gfn (mmu_check_root)
1536 kvm_arch_flush_shadow_memslot(kvm, slot);
1538 /* Released in install_new_memslots. */
1539 mutex_lock(&kvm->slots_arch_lock);
1542 * The arch-specific fields of the memslots could have changed
1543 * between releasing the slots_arch_lock in
1544 * install_new_memslots and here, so get a fresh copy of the
1547 kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id));
1550 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1554 update_memslots(slots, new, change);
1555 slots = install_new_memslots(kvm, as_id, slots);
1557 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1563 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1564 slot = id_to_memslot(slots, old->id);
1565 slot->flags &= ~KVM_MEMSLOT_INVALID;
1566 slots = install_new_memslots(kvm, as_id, slots);
1568 mutex_unlock(&kvm->slots_arch_lock);
1574 static int kvm_delete_memslot(struct kvm *kvm,
1575 const struct kvm_userspace_memory_region *mem,
1576 struct kvm_memory_slot *old, int as_id)
1578 struct kvm_memory_slot new;
1584 memset(&new, 0, sizeof(new));
1587 * This is only for debugging purpose; it should never be referenced
1588 * for a removed memslot.
1592 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1596 kvm_free_memslot(kvm, old);
1601 * Allocate some memory and give it an address in the guest physical address
1604 * Discontiguous memory is allowed, mostly for framebuffers.
1606 * Must be called holding kvm->slots_lock for write.
1608 int __kvm_set_memory_region(struct kvm *kvm,
1609 const struct kvm_userspace_memory_region *mem)
1611 struct kvm_memory_slot old, new;
1612 struct kvm_memory_slot *tmp;
1613 enum kvm_mr_change change;
1617 r = check_memory_region_flags(mem);
1621 as_id = mem->slot >> 16;
1622 id = (u16)mem->slot;
1624 /* General sanity checks */
1625 if (mem->memory_size & (PAGE_SIZE - 1))
1627 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1629 /* We can read the guest memory with __xxx_user() later on. */
1630 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1631 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1632 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1635 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1637 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1641 * Make a full copy of the old memslot, the pointer will become stale
1642 * when the memslots are re-sorted by update_memslots(), and the old
1643 * memslot needs to be referenced after calling update_memslots(), e.g.
1644 * to free its resources and for arch specific behavior.
1646 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1651 memset(&old, 0, sizeof(old));
1655 if (!mem->memory_size)
1656 return kvm_delete_memslot(kvm, mem, &old, as_id);
1660 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1661 new.npages = mem->memory_size >> PAGE_SHIFT;
1662 new.flags = mem->flags;
1663 new.userspace_addr = mem->userspace_addr;
1665 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1669 change = KVM_MR_CREATE;
1670 new.dirty_bitmap = NULL;
1671 memset(&new.arch, 0, sizeof(new.arch));
1672 } else { /* Modify an existing slot. */
1673 if ((new.userspace_addr != old.userspace_addr) ||
1674 (new.npages != old.npages) ||
1675 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1678 if (new.base_gfn != old.base_gfn)
1679 change = KVM_MR_MOVE;
1680 else if (new.flags != old.flags)
1681 change = KVM_MR_FLAGS_ONLY;
1682 else /* Nothing to change. */
1685 /* Copy dirty_bitmap and arch from the current memslot. */
1686 new.dirty_bitmap = old.dirty_bitmap;
1687 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1690 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1691 /* Check for overlaps */
1692 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1695 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1696 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1701 /* Allocate/free page dirty bitmap as needed */
1702 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1703 new.dirty_bitmap = NULL;
1704 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1705 r = kvm_alloc_dirty_bitmap(&new);
1709 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1710 bitmap_set(new.dirty_bitmap, 0, new.npages);
1713 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1717 if (old.dirty_bitmap && !new.dirty_bitmap)
1718 kvm_destroy_dirty_bitmap(&old);
1722 if (new.dirty_bitmap && !old.dirty_bitmap)
1723 kvm_destroy_dirty_bitmap(&new);
1726 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1728 int kvm_set_memory_region(struct kvm *kvm,
1729 const struct kvm_userspace_memory_region *mem)
1733 mutex_lock(&kvm->slots_lock);
1734 r = __kvm_set_memory_region(kvm, mem);
1735 mutex_unlock(&kvm->slots_lock);
1738 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1740 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1741 struct kvm_userspace_memory_region *mem)
1743 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1746 return kvm_set_memory_region(kvm, mem);
1749 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1751 * kvm_get_dirty_log - get a snapshot of dirty pages
1752 * @kvm: pointer to kvm instance
1753 * @log: slot id and address to which we copy the log
1754 * @is_dirty: set to '1' if any dirty pages were found
1755 * @memslot: set to the associated memslot, always valid on success
1757 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1758 int *is_dirty, struct kvm_memory_slot **memslot)
1760 struct kvm_memslots *slots;
1763 unsigned long any = 0;
1765 /* Dirty ring tracking is exclusive to dirty log tracking */
1766 if (kvm->dirty_ring_size)
1772 as_id = log->slot >> 16;
1773 id = (u16)log->slot;
1774 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1777 slots = __kvm_memslots(kvm, as_id);
1778 *memslot = id_to_memslot(slots, id);
1779 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1782 kvm_arch_sync_dirty_log(kvm, *memslot);
1784 n = kvm_dirty_bitmap_bytes(*memslot);
1786 for (i = 0; !any && i < n/sizeof(long); ++i)
1787 any = (*memslot)->dirty_bitmap[i];
1789 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1796 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1798 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1800 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1801 * and reenable dirty page tracking for the corresponding pages.
1802 * @kvm: pointer to kvm instance
1803 * @log: slot id and address to which we copy the log
1805 * We need to keep it in mind that VCPU threads can write to the bitmap
1806 * concurrently. So, to avoid losing track of dirty pages we keep the
1809 * 1. Take a snapshot of the bit and clear it if needed.
1810 * 2. Write protect the corresponding page.
1811 * 3. Copy the snapshot to the userspace.
1812 * 4. Upon return caller flushes TLB's if needed.
1814 * Between 2 and 4, the guest may write to the page using the remaining TLB
1815 * entry. This is not a problem because the page is reported dirty using
1816 * the snapshot taken before and step 4 ensures that writes done after
1817 * exiting to userspace will be logged for the next call.
1820 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1822 struct kvm_memslots *slots;
1823 struct kvm_memory_slot *memslot;
1826 unsigned long *dirty_bitmap;
1827 unsigned long *dirty_bitmap_buffer;
1830 /* Dirty ring tracking is exclusive to dirty log tracking */
1831 if (kvm->dirty_ring_size)
1834 as_id = log->slot >> 16;
1835 id = (u16)log->slot;
1836 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1839 slots = __kvm_memslots(kvm, as_id);
1840 memslot = id_to_memslot(slots, id);
1841 if (!memslot || !memslot->dirty_bitmap)
1844 dirty_bitmap = memslot->dirty_bitmap;
1846 kvm_arch_sync_dirty_log(kvm, memslot);
1848 n = kvm_dirty_bitmap_bytes(memslot);
1850 if (kvm->manual_dirty_log_protect) {
1852 * Unlike kvm_get_dirty_log, we always return false in *flush,
1853 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1854 * is some code duplication between this function and
1855 * kvm_get_dirty_log, but hopefully all architecture
1856 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1857 * can be eliminated.
1859 dirty_bitmap_buffer = dirty_bitmap;
1861 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1862 memset(dirty_bitmap_buffer, 0, n);
1865 for (i = 0; i < n / sizeof(long); i++) {
1869 if (!dirty_bitmap[i])
1873 mask = xchg(&dirty_bitmap[i], 0);
1874 dirty_bitmap_buffer[i] = mask;
1876 offset = i * BITS_PER_LONG;
1877 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1880 KVM_MMU_UNLOCK(kvm);
1884 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1886 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1893 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1894 * @kvm: kvm instance
1895 * @log: slot id and address to which we copy the log
1897 * Steps 1-4 below provide general overview of dirty page logging. See
1898 * kvm_get_dirty_log_protect() function description for additional details.
1900 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1901 * always flush the TLB (step 4) even if previous step failed and the dirty
1902 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1903 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1904 * writes will be marked dirty for next log read.
1906 * 1. Take a snapshot of the bit and clear it if needed.
1907 * 2. Write protect the corresponding page.
1908 * 3. Copy the snapshot to the userspace.
1909 * 4. Flush TLB's if needed.
1911 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1912 struct kvm_dirty_log *log)
1916 mutex_lock(&kvm->slots_lock);
1918 r = kvm_get_dirty_log_protect(kvm, log);
1920 mutex_unlock(&kvm->slots_lock);
1925 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1926 * and reenable dirty page tracking for the corresponding pages.
1927 * @kvm: pointer to kvm instance
1928 * @log: slot id and address from which to fetch the bitmap of dirty pages
1930 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1931 struct kvm_clear_dirty_log *log)
1933 struct kvm_memslots *slots;
1934 struct kvm_memory_slot *memslot;
1938 unsigned long *dirty_bitmap;
1939 unsigned long *dirty_bitmap_buffer;
1942 /* Dirty ring tracking is exclusive to dirty log tracking */
1943 if (kvm->dirty_ring_size)
1946 as_id = log->slot >> 16;
1947 id = (u16)log->slot;
1948 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1951 if (log->first_page & 63)
1954 slots = __kvm_memslots(kvm, as_id);
1955 memslot = id_to_memslot(slots, id);
1956 if (!memslot || !memslot->dirty_bitmap)
1959 dirty_bitmap = memslot->dirty_bitmap;
1961 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1963 if (log->first_page > memslot->npages ||
1964 log->num_pages > memslot->npages - log->first_page ||
1965 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1968 kvm_arch_sync_dirty_log(kvm, memslot);
1971 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1972 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1976 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1977 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1978 i++, offset += BITS_PER_LONG) {
1979 unsigned long mask = *dirty_bitmap_buffer++;
1980 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1984 mask &= atomic_long_fetch_andnot(mask, p);
1987 * mask contains the bits that really have been cleared. This
1988 * never includes any bits beyond the length of the memslot (if
1989 * the length is not aligned to 64 pages), therefore it is not
1990 * a problem if userspace sets them in log->dirty_bitmap.
1994 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1998 KVM_MMU_UNLOCK(kvm);
2001 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2006 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2007 struct kvm_clear_dirty_log *log)
2011 mutex_lock(&kvm->slots_lock);
2013 r = kvm_clear_dirty_log_protect(kvm, log);
2015 mutex_unlock(&kvm->slots_lock);
2018 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2020 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2022 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2024 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2026 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2028 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
2030 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
2032 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2034 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2036 return kvm_is_visible_memslot(memslot);
2038 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2040 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2042 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2044 return kvm_is_visible_memslot(memslot);
2046 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2048 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2050 struct vm_area_struct *vma;
2051 unsigned long addr, size;
2055 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2056 if (kvm_is_error_hva(addr))
2059 mmap_read_lock(current->mm);
2060 vma = find_vma(current->mm, addr);
2064 size = vma_kernel_pagesize(vma);
2067 mmap_read_unlock(current->mm);
2072 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
2074 return slot->flags & KVM_MEM_READONLY;
2077 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2078 gfn_t *nr_pages, bool write)
2080 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2081 return KVM_HVA_ERR_BAD;
2083 if (memslot_is_readonly(slot) && write)
2084 return KVM_HVA_ERR_RO_BAD;
2087 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2089 return __gfn_to_hva_memslot(slot, gfn);
2092 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2095 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2098 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2101 return gfn_to_hva_many(slot, gfn, NULL);
2103 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2105 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2107 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2109 EXPORT_SYMBOL_GPL(gfn_to_hva);
2111 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2113 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2115 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2118 * Return the hva of a @gfn and the R/W attribute if possible.
2120 * @slot: the kvm_memory_slot which contains @gfn
2121 * @gfn: the gfn to be translated
2122 * @writable: used to return the read/write attribute of the @slot if the hva
2123 * is valid and @writable is not NULL
2125 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2126 gfn_t gfn, bool *writable)
2128 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2130 if (!kvm_is_error_hva(hva) && writable)
2131 *writable = !memslot_is_readonly(slot);
2136 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2138 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2140 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2143 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2145 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2147 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2150 static inline int check_user_page_hwpoison(unsigned long addr)
2152 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2154 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2155 return rc == -EHWPOISON;
2159 * The fast path to get the writable pfn which will be stored in @pfn,
2160 * true indicates success, otherwise false is returned. It's also the
2161 * only part that runs if we can in atomic context.
2163 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2164 bool *writable, kvm_pfn_t *pfn)
2166 struct page *page[1];
2169 * Fast pin a writable pfn only if it is a write fault request
2170 * or the caller allows to map a writable pfn for a read fault
2173 if (!(write_fault || writable))
2176 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2177 *pfn = page_to_pfn(page[0]);
2188 * The slow path to get the pfn of the specified host virtual address,
2189 * 1 indicates success, -errno is returned if error is detected.
2191 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2192 bool *writable, kvm_pfn_t *pfn)
2194 unsigned int flags = FOLL_HWPOISON;
2201 *writable = write_fault;
2204 flags |= FOLL_WRITE;
2206 flags |= FOLL_NOWAIT;
2208 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2212 /* map read fault as writable if possible */
2213 if (unlikely(!write_fault) && writable) {
2216 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2222 *pfn = page_to_pfn(page);
2226 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2228 if (unlikely(!(vma->vm_flags & VM_READ)))
2231 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2237 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2239 if (kvm_is_reserved_pfn(pfn))
2241 return get_page_unless_zero(pfn_to_page(pfn));
2244 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2245 unsigned long addr, bool *async,
2246 bool write_fault, bool *writable,
2254 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2257 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2258 * not call the fault handler, so do it here.
2260 bool unlocked = false;
2261 r = fixup_user_fault(current->mm, addr,
2262 (write_fault ? FAULT_FLAG_WRITE : 0),
2269 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2274 if (write_fault && !pte_write(*ptep)) {
2275 pfn = KVM_PFN_ERR_RO_FAULT;
2280 *writable = pte_write(*ptep);
2281 pfn = pte_pfn(*ptep);
2284 * Get a reference here because callers of *hva_to_pfn* and
2285 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2286 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2287 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2288 * simply do nothing for reserved pfns.
2290 * Whoever called remap_pfn_range is also going to call e.g.
2291 * unmap_mapping_range before the underlying pages are freed,
2292 * causing a call to our MMU notifier.
2294 * Certain IO or PFNMAP mappings can be backed with valid
2295 * struct pages, but be allocated without refcounting e.g.,
2296 * tail pages of non-compound higher order allocations, which
2297 * would then underflow the refcount when the caller does the
2298 * required put_page. Don't allow those pages here.
2300 if (!kvm_try_get_pfn(pfn))
2304 pte_unmap_unlock(ptep, ptl);
2311 * Pin guest page in memory and return its pfn.
2312 * @addr: host virtual address which maps memory to the guest
2313 * @atomic: whether this function can sleep
2314 * @async: whether this function need to wait IO complete if the
2315 * host page is not in the memory
2316 * @write_fault: whether we should get a writable host page
2317 * @writable: whether it allows to map a writable host page for !@write_fault
2319 * The function will map a writable host page for these two cases:
2320 * 1): @write_fault = true
2321 * 2): @write_fault = false && @writable, @writable will tell the caller
2322 * whether the mapping is writable.
2324 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2325 bool write_fault, bool *writable)
2327 struct vm_area_struct *vma;
2331 /* we can do it either atomically or asynchronously, not both */
2332 BUG_ON(atomic && async);
2334 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2338 return KVM_PFN_ERR_FAULT;
2340 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2344 mmap_read_lock(current->mm);
2345 if (npages == -EHWPOISON ||
2346 (!async && check_user_page_hwpoison(addr))) {
2347 pfn = KVM_PFN_ERR_HWPOISON;
2352 vma = vma_lookup(current->mm, addr);
2355 pfn = KVM_PFN_ERR_FAULT;
2356 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2357 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2361 pfn = KVM_PFN_ERR_FAULT;
2363 if (async && vma_is_valid(vma, write_fault))
2365 pfn = KVM_PFN_ERR_FAULT;
2368 mmap_read_unlock(current->mm);
2372 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2373 bool atomic, bool *async, bool write_fault,
2374 bool *writable, hva_t *hva)
2376 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2381 if (addr == KVM_HVA_ERR_RO_BAD) {
2384 return KVM_PFN_ERR_RO_FAULT;
2387 if (kvm_is_error_hva(addr)) {
2390 return KVM_PFN_NOSLOT;
2393 /* Do not map writable pfn in the readonly memslot. */
2394 if (writable && memslot_is_readonly(slot)) {
2399 return hva_to_pfn(addr, atomic, async, write_fault,
2402 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2404 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2407 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2408 write_fault, writable, NULL);
2410 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2412 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2414 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2416 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2418 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2420 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2422 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2424 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2426 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2428 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2430 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2432 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2434 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2436 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2438 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2440 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2442 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2443 struct page **pages, int nr_pages)
2448 addr = gfn_to_hva_many(slot, gfn, &entry);
2449 if (kvm_is_error_hva(addr))
2452 if (entry < nr_pages)
2455 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2457 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2459 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2461 if (is_error_noslot_pfn(pfn))
2462 return KVM_ERR_PTR_BAD_PAGE;
2464 if (kvm_is_reserved_pfn(pfn)) {
2466 return KVM_ERR_PTR_BAD_PAGE;
2469 return pfn_to_page(pfn);
2472 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2476 pfn = gfn_to_pfn(kvm, gfn);
2478 return kvm_pfn_to_page(pfn);
2480 EXPORT_SYMBOL_GPL(gfn_to_page);
2482 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2488 cache->pfn = cache->gfn = 0;
2491 kvm_release_pfn_dirty(pfn);
2493 kvm_release_pfn_clean(pfn);
2496 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2497 struct gfn_to_pfn_cache *cache, u64 gen)
2499 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2501 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2503 cache->dirty = false;
2504 cache->generation = gen;
2507 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2508 struct kvm_host_map *map,
2509 struct gfn_to_pfn_cache *cache,
2514 struct page *page = KVM_UNMAPPED_PAGE;
2515 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2516 u64 gen = slots->generation;
2522 if (!cache->pfn || cache->gfn != gfn ||
2523 cache->generation != gen) {
2526 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2532 pfn = gfn_to_pfn_memslot(slot, gfn);
2534 if (is_error_noslot_pfn(pfn))
2537 if (pfn_valid(pfn)) {
2538 page = pfn_to_page(pfn);
2540 hva = kmap_atomic(page);
2543 #ifdef CONFIG_HAS_IOMEM
2544 } else if (!atomic) {
2545 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2562 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2563 struct gfn_to_pfn_cache *cache, bool atomic)
2565 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2568 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2570 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2572 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2575 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2577 static void __kvm_unmap_gfn(struct kvm *kvm,
2578 struct kvm_memory_slot *memslot,
2579 struct kvm_host_map *map,
2580 struct gfn_to_pfn_cache *cache,
2581 bool dirty, bool atomic)
2589 if (map->page != KVM_UNMAPPED_PAGE) {
2591 kunmap_atomic(map->hva);
2595 #ifdef CONFIG_HAS_IOMEM
2599 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2603 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2606 cache->dirty |= dirty;
2608 kvm_release_pfn(map->pfn, dirty, NULL);
2614 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2615 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2617 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2618 cache, dirty, atomic);
2621 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2623 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2625 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2626 map, NULL, dirty, false);
2628 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2630 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2634 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2636 return kvm_pfn_to_page(pfn);
2638 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2640 void kvm_release_page_clean(struct page *page)
2642 WARN_ON(is_error_page(page));
2644 kvm_release_pfn_clean(page_to_pfn(page));
2646 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2648 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2650 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2651 put_page(pfn_to_page(pfn));
2653 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2655 void kvm_release_page_dirty(struct page *page)
2657 WARN_ON(is_error_page(page));
2659 kvm_release_pfn_dirty(page_to_pfn(page));
2661 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2663 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2665 kvm_set_pfn_dirty(pfn);
2666 kvm_release_pfn_clean(pfn);
2668 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2670 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2672 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2673 SetPageDirty(pfn_to_page(pfn));
2675 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2677 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2679 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2680 mark_page_accessed(pfn_to_page(pfn));
2682 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2684 void kvm_get_pfn(kvm_pfn_t pfn)
2686 if (!kvm_is_reserved_pfn(pfn))
2687 get_page(pfn_to_page(pfn));
2689 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2691 static int next_segment(unsigned long len, int offset)
2693 if (len > PAGE_SIZE - offset)
2694 return PAGE_SIZE - offset;
2699 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2700 void *data, int offset, int len)
2705 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2706 if (kvm_is_error_hva(addr))
2708 r = __copy_from_user(data, (void __user *)addr + offset, len);
2714 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2717 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2719 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2721 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2723 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2724 int offset, int len)
2726 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2728 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2730 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2732 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2734 gfn_t gfn = gpa >> PAGE_SHIFT;
2736 int offset = offset_in_page(gpa);
2739 while ((seg = next_segment(len, offset)) != 0) {
2740 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2750 EXPORT_SYMBOL_GPL(kvm_read_guest);
2752 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2754 gfn_t gfn = gpa >> PAGE_SHIFT;
2756 int offset = offset_in_page(gpa);
2759 while ((seg = next_segment(len, offset)) != 0) {
2760 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2770 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2772 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2773 void *data, int offset, unsigned long len)
2778 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2779 if (kvm_is_error_hva(addr))
2781 pagefault_disable();
2782 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2789 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2790 void *data, unsigned long len)
2792 gfn_t gfn = gpa >> PAGE_SHIFT;
2793 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2794 int offset = offset_in_page(gpa);
2796 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2798 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2800 static int __kvm_write_guest_page(struct kvm *kvm,
2801 struct kvm_memory_slot *memslot, gfn_t gfn,
2802 const void *data, int offset, int len)
2807 addr = gfn_to_hva_memslot(memslot, gfn);
2808 if (kvm_is_error_hva(addr))
2810 r = __copy_to_user((void __user *)addr + offset, data, len);
2813 mark_page_dirty_in_slot(kvm, memslot, gfn);
2817 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2818 const void *data, int offset, int len)
2820 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2822 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2824 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2826 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2827 const void *data, int offset, int len)
2829 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2831 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2833 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2835 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2838 gfn_t gfn = gpa >> PAGE_SHIFT;
2840 int offset = offset_in_page(gpa);
2843 while ((seg = next_segment(len, offset)) != 0) {
2844 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2854 EXPORT_SYMBOL_GPL(kvm_write_guest);
2856 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2859 gfn_t gfn = gpa >> PAGE_SHIFT;
2861 int offset = offset_in_page(gpa);
2864 while ((seg = next_segment(len, offset)) != 0) {
2865 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2875 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2877 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2878 struct gfn_to_hva_cache *ghc,
2879 gpa_t gpa, unsigned long len)
2881 int offset = offset_in_page(gpa);
2882 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2883 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2884 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2885 gfn_t nr_pages_avail;
2887 /* Update ghc->generation before performing any error checks. */
2888 ghc->generation = slots->generation;
2890 if (start_gfn > end_gfn) {
2891 ghc->hva = KVM_HVA_ERR_BAD;
2896 * If the requested region crosses two memslots, we still
2897 * verify that the entire region is valid here.
2899 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2900 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2901 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2903 if (kvm_is_error_hva(ghc->hva))
2907 /* Use the slow path for cross page reads and writes. */
2908 if (nr_pages_needed == 1)
2911 ghc->memslot = NULL;
2918 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2919 gpa_t gpa, unsigned long len)
2921 struct kvm_memslots *slots = kvm_memslots(kvm);
2922 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2924 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2926 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2927 void *data, unsigned int offset,
2930 struct kvm_memslots *slots = kvm_memslots(kvm);
2932 gpa_t gpa = ghc->gpa + offset;
2934 BUG_ON(len + offset > ghc->len);
2936 if (slots->generation != ghc->generation) {
2937 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2941 if (kvm_is_error_hva(ghc->hva))
2944 if (unlikely(!ghc->memslot))
2945 return kvm_write_guest(kvm, gpa, data, len);
2947 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2950 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2954 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2956 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2957 void *data, unsigned long len)
2959 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2961 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2963 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2964 void *data, unsigned int offset,
2967 struct kvm_memslots *slots = kvm_memslots(kvm);
2969 gpa_t gpa = ghc->gpa + offset;
2971 BUG_ON(len + offset > ghc->len);
2973 if (slots->generation != ghc->generation) {
2974 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2978 if (kvm_is_error_hva(ghc->hva))
2981 if (unlikely(!ghc->memslot))
2982 return kvm_read_guest(kvm, gpa, data, len);
2984 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2990 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2992 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2993 void *data, unsigned long len)
2995 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2997 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2999 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3001 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3002 gfn_t gfn = gpa >> PAGE_SHIFT;
3004 int offset = offset_in_page(gpa);
3007 while ((seg = next_segment(len, offset)) != 0) {
3008 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3017 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3019 void mark_page_dirty_in_slot(struct kvm *kvm,
3020 struct kvm_memory_slot *memslot,
3023 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3024 unsigned long rel_gfn = gfn - memslot->base_gfn;
3025 u32 slot = (memslot->as_id << 16) | memslot->id;
3027 if (kvm->dirty_ring_size)
3028 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
3031 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3034 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3036 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3038 struct kvm_memory_slot *memslot;
3040 memslot = gfn_to_memslot(kvm, gfn);
3041 mark_page_dirty_in_slot(kvm, memslot, gfn);
3043 EXPORT_SYMBOL_GPL(mark_page_dirty);
3045 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3047 struct kvm_memory_slot *memslot;
3049 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3050 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3052 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3054 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3056 if (!vcpu->sigset_active)
3060 * This does a lockless modification of ->real_blocked, which is fine
3061 * because, only current can change ->real_blocked and all readers of
3062 * ->real_blocked don't care as long ->real_blocked is always a subset
3065 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3068 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3070 if (!vcpu->sigset_active)
3073 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3074 sigemptyset(¤t->real_blocked);
3077 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3079 unsigned int old, val, grow, grow_start;
3081 old = val = vcpu->halt_poll_ns;
3082 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3083 grow = READ_ONCE(halt_poll_ns_grow);
3088 if (val < grow_start)
3091 if (val > vcpu->kvm->max_halt_poll_ns)
3092 val = vcpu->kvm->max_halt_poll_ns;
3094 vcpu->halt_poll_ns = val;
3096 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3099 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3101 unsigned int old, val, shrink;
3103 old = val = vcpu->halt_poll_ns;
3104 shrink = READ_ONCE(halt_poll_ns_shrink);
3110 vcpu->halt_poll_ns = val;
3111 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3114 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3117 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3119 if (kvm_arch_vcpu_runnable(vcpu)) {
3120 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3123 if (kvm_cpu_has_pending_timer(vcpu))
3125 if (signal_pending(current))
3127 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3132 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3137 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
3140 vcpu->stat.generic.halt_poll_fail_ns += poll_ns;
3142 vcpu->stat.generic.halt_poll_success_ns += poll_ns;
3146 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3148 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
3150 ktime_t start, cur, poll_end;
3151 bool waited = false;
3154 kvm_arch_vcpu_blocking(vcpu);
3156 start = cur = poll_end = ktime_get();
3157 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
3158 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3160 ++vcpu->stat.generic.halt_attempted_poll;
3163 * This sets KVM_REQ_UNHALT if an interrupt
3166 if (kvm_vcpu_check_block(vcpu) < 0) {
3167 ++vcpu->stat.generic.halt_successful_poll;
3168 if (!vcpu_valid_wakeup(vcpu))
3169 ++vcpu->stat.generic.halt_poll_invalid;
3173 poll_end = cur = ktime_get();
3174 } while (kvm_vcpu_can_poll(cur, stop));
3177 prepare_to_rcuwait(&vcpu->wait);
3179 set_current_state(TASK_INTERRUPTIBLE);
3181 if (kvm_vcpu_check_block(vcpu) < 0)
3187 finish_rcuwait(&vcpu->wait);
3190 kvm_arch_vcpu_unblocking(vcpu);
3191 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3193 update_halt_poll_stats(
3194 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3196 if (!kvm_arch_no_poll(vcpu)) {
3197 if (!vcpu_valid_wakeup(vcpu)) {
3198 shrink_halt_poll_ns(vcpu);
3199 } else if (vcpu->kvm->max_halt_poll_ns) {
3200 if (block_ns <= vcpu->halt_poll_ns)
3202 /* we had a long block, shrink polling */
3203 else if (vcpu->halt_poll_ns &&
3204 block_ns > vcpu->kvm->max_halt_poll_ns)
3205 shrink_halt_poll_ns(vcpu);
3206 /* we had a short halt and our poll time is too small */
3207 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3208 block_ns < vcpu->kvm->max_halt_poll_ns)
3209 grow_halt_poll_ns(vcpu);
3211 vcpu->halt_poll_ns = 0;
3215 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3216 kvm_arch_vcpu_block_finish(vcpu);
3218 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3220 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3222 struct rcuwait *waitp;
3224 waitp = kvm_arch_vcpu_get_wait(vcpu);
3225 if (rcuwait_wake_up(waitp)) {
3226 WRITE_ONCE(vcpu->ready, true);
3227 ++vcpu->stat.generic.halt_wakeup;
3233 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3237 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3239 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3242 int cpu = vcpu->cpu;
3244 if (kvm_vcpu_wake_up(vcpu))
3248 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3249 if (kvm_arch_vcpu_should_kick(vcpu))
3250 smp_send_reschedule(cpu);
3253 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3254 #endif /* !CONFIG_S390 */
3256 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3259 struct task_struct *task = NULL;
3263 pid = rcu_dereference(target->pid);
3265 task = get_pid_task(pid, PIDTYPE_PID);
3269 ret = yield_to(task, 1);
3270 put_task_struct(task);
3274 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3277 * Helper that checks whether a VCPU is eligible for directed yield.
3278 * Most eligible candidate to yield is decided by following heuristics:
3280 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3281 * (preempted lock holder), indicated by @in_spin_loop.
3282 * Set at the beginning and cleared at the end of interception/PLE handler.
3284 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3285 * chance last time (mostly it has become eligible now since we have probably
3286 * yielded to lockholder in last iteration. This is done by toggling
3287 * @dy_eligible each time a VCPU checked for eligibility.)
3289 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3290 * to preempted lock-holder could result in wrong VCPU selection and CPU
3291 * burning. Giving priority for a potential lock-holder increases lock
3294 * Since algorithm is based on heuristics, accessing another VCPU data without
3295 * locking does not harm. It may result in trying to yield to same VCPU, fail
3296 * and continue with next VCPU and so on.
3298 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3300 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3303 eligible = !vcpu->spin_loop.in_spin_loop ||
3304 vcpu->spin_loop.dy_eligible;
3306 if (vcpu->spin_loop.in_spin_loop)
3307 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3316 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3317 * a vcpu_load/vcpu_put pair. However, for most architectures
3318 * kvm_arch_vcpu_runnable does not require vcpu_load.
3320 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3322 return kvm_arch_vcpu_runnable(vcpu);
3325 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3327 if (kvm_arch_dy_runnable(vcpu))
3330 #ifdef CONFIG_KVM_ASYNC_PF
3331 if (!list_empty_careful(&vcpu->async_pf.done))
3338 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3343 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3345 struct kvm *kvm = me->kvm;
3346 struct kvm_vcpu *vcpu;
3347 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3353 kvm_vcpu_set_in_spin_loop(me, true);
3355 * We boost the priority of a VCPU that is runnable but not
3356 * currently running, because it got preempted by something
3357 * else and called schedule in __vcpu_run. Hopefully that
3358 * VCPU is holding the lock that we need and will release it.
3359 * We approximate round-robin by starting at the last boosted VCPU.
3361 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3362 kvm_for_each_vcpu(i, vcpu, kvm) {
3363 if (!pass && i <= last_boosted_vcpu) {
3364 i = last_boosted_vcpu;
3366 } else if (pass && i > last_boosted_vcpu)
3368 if (!READ_ONCE(vcpu->ready))
3372 if (rcuwait_active(&vcpu->wait) &&
3373 !vcpu_dy_runnable(vcpu))
3375 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3376 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3377 !kvm_arch_vcpu_in_kernel(vcpu))
3379 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3382 yielded = kvm_vcpu_yield_to(vcpu);
3384 kvm->last_boosted_vcpu = i;
3386 } else if (yielded < 0) {
3393 kvm_vcpu_set_in_spin_loop(me, false);
3395 /* Ensure vcpu is not eligible during next spinloop */
3396 kvm_vcpu_set_dy_eligible(me, false);
3398 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3400 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3402 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3403 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3404 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3405 kvm->dirty_ring_size / PAGE_SIZE);
3411 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3413 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3416 if (vmf->pgoff == 0)
3417 page = virt_to_page(vcpu->run);
3419 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3420 page = virt_to_page(vcpu->arch.pio_data);
3422 #ifdef CONFIG_KVM_MMIO
3423 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3424 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3426 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3427 page = kvm_dirty_ring_get_page(
3429 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3431 return kvm_arch_vcpu_fault(vcpu, vmf);
3437 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3438 .fault = kvm_vcpu_fault,
3441 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3443 struct kvm_vcpu *vcpu = file->private_data;
3444 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3446 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3447 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3448 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3451 vma->vm_ops = &kvm_vcpu_vm_ops;
3455 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3457 struct kvm_vcpu *vcpu = filp->private_data;
3459 kvm_put_kvm(vcpu->kvm);
3463 static struct file_operations kvm_vcpu_fops = {
3464 .release = kvm_vcpu_release,
3465 .unlocked_ioctl = kvm_vcpu_ioctl,
3466 .mmap = kvm_vcpu_mmap,
3467 .llseek = noop_llseek,
3468 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3472 * Allocates an inode for the vcpu.
3474 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3476 char name[8 + 1 + ITOA_MAX_LEN + 1];
3478 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3479 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3482 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3484 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3485 struct dentry *debugfs_dentry;
3486 char dir_name[ITOA_MAX_LEN * 2];
3488 if (!debugfs_initialized())
3491 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3492 debugfs_dentry = debugfs_create_dir(dir_name,
3493 vcpu->kvm->debugfs_dentry);
3495 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3500 * Creates some virtual cpus. Good luck creating more than one.
3502 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3505 struct kvm_vcpu *vcpu;
3508 if (id >= KVM_MAX_VCPU_ID)
3511 mutex_lock(&kvm->lock);
3512 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3513 mutex_unlock(&kvm->lock);
3517 kvm->created_vcpus++;
3518 mutex_unlock(&kvm->lock);
3520 r = kvm_arch_vcpu_precreate(kvm, id);
3522 goto vcpu_decrement;
3524 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3527 goto vcpu_decrement;
3530 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3531 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3536 vcpu->run = page_address(page);
3538 kvm_vcpu_init(vcpu, kvm, id);
3540 r = kvm_arch_vcpu_create(vcpu);
3542 goto vcpu_free_run_page;
3544 if (kvm->dirty_ring_size) {
3545 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3546 id, kvm->dirty_ring_size);
3548 goto arch_vcpu_destroy;
3551 mutex_lock(&kvm->lock);
3552 if (kvm_get_vcpu_by_id(kvm, id)) {
3554 goto unlock_vcpu_destroy;
3557 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3558 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3560 /* Fill the stats id string for the vcpu */
3561 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3562 task_pid_nr(current), id);
3564 /* Now it's all set up, let userspace reach it */
3566 r = create_vcpu_fd(vcpu);
3568 kvm_put_kvm_no_destroy(kvm);
3569 goto unlock_vcpu_destroy;
3572 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3575 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3576 * before kvm->online_vcpu's incremented value.
3579 atomic_inc(&kvm->online_vcpus);
3581 mutex_unlock(&kvm->lock);
3582 kvm_arch_vcpu_postcreate(vcpu);
3583 kvm_create_vcpu_debugfs(vcpu);
3586 unlock_vcpu_destroy:
3587 mutex_unlock(&kvm->lock);
3588 kvm_dirty_ring_free(&vcpu->dirty_ring);
3590 kvm_arch_vcpu_destroy(vcpu);
3592 free_page((unsigned long)vcpu->run);
3594 kmem_cache_free(kvm_vcpu_cache, vcpu);
3596 mutex_lock(&kvm->lock);
3597 kvm->created_vcpus--;
3598 mutex_unlock(&kvm->lock);
3602 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3605 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3606 vcpu->sigset_active = 1;
3607 vcpu->sigset = *sigset;
3609 vcpu->sigset_active = 0;
3613 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3614 size_t size, loff_t *offset)
3616 struct kvm_vcpu *vcpu = file->private_data;
3618 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3619 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3620 sizeof(vcpu->stat), user_buffer, size, offset);
3623 static const struct file_operations kvm_vcpu_stats_fops = {
3624 .read = kvm_vcpu_stats_read,
3625 .llseek = noop_llseek,
3628 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3632 char name[15 + ITOA_MAX_LEN + 1];
3634 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3636 fd = get_unused_fd_flags(O_CLOEXEC);
3640 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3643 return PTR_ERR(file);
3645 file->f_mode |= FMODE_PREAD;
3646 fd_install(fd, file);
3651 static long kvm_vcpu_ioctl(struct file *filp,
3652 unsigned int ioctl, unsigned long arg)
3654 struct kvm_vcpu *vcpu = filp->private_data;
3655 void __user *argp = (void __user *)arg;
3657 struct kvm_fpu *fpu = NULL;
3658 struct kvm_sregs *kvm_sregs = NULL;
3660 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3663 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3667 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3668 * execution; mutex_lock() would break them.
3670 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3671 if (r != -ENOIOCTLCMD)
3674 if (mutex_lock_killable(&vcpu->mutex))
3682 oldpid = rcu_access_pointer(vcpu->pid);
3683 if (unlikely(oldpid != task_pid(current))) {
3684 /* The thread running this VCPU changed. */
3687 r = kvm_arch_vcpu_run_pid_change(vcpu);
3691 newpid = get_task_pid(current, PIDTYPE_PID);
3692 rcu_assign_pointer(vcpu->pid, newpid);
3697 r = kvm_arch_vcpu_ioctl_run(vcpu);
3698 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3701 case KVM_GET_REGS: {
3702 struct kvm_regs *kvm_regs;
3705 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3708 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3712 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3719 case KVM_SET_REGS: {
3720 struct kvm_regs *kvm_regs;
3722 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3723 if (IS_ERR(kvm_regs)) {
3724 r = PTR_ERR(kvm_regs);
3727 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3731 case KVM_GET_SREGS: {
3732 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3733 GFP_KERNEL_ACCOUNT);
3737 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3741 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3746 case KVM_SET_SREGS: {
3747 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3748 if (IS_ERR(kvm_sregs)) {
3749 r = PTR_ERR(kvm_sregs);
3753 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3756 case KVM_GET_MP_STATE: {
3757 struct kvm_mp_state mp_state;
3759 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3763 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3768 case KVM_SET_MP_STATE: {
3769 struct kvm_mp_state mp_state;
3772 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3774 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3777 case KVM_TRANSLATE: {
3778 struct kvm_translation tr;
3781 if (copy_from_user(&tr, argp, sizeof(tr)))
3783 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3787 if (copy_to_user(argp, &tr, sizeof(tr)))
3792 case KVM_SET_GUEST_DEBUG: {
3793 struct kvm_guest_debug dbg;
3796 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3798 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3801 case KVM_SET_SIGNAL_MASK: {
3802 struct kvm_signal_mask __user *sigmask_arg = argp;
3803 struct kvm_signal_mask kvm_sigmask;
3804 sigset_t sigset, *p;
3809 if (copy_from_user(&kvm_sigmask, argp,
3810 sizeof(kvm_sigmask)))
3813 if (kvm_sigmask.len != sizeof(sigset))
3816 if (copy_from_user(&sigset, sigmask_arg->sigset,
3821 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3825 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3829 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3833 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3839 fpu = memdup_user(argp, sizeof(*fpu));
3845 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3848 case KVM_GET_STATS_FD: {
3849 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
3853 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3856 mutex_unlock(&vcpu->mutex);
3862 #ifdef CONFIG_KVM_COMPAT
3863 static long kvm_vcpu_compat_ioctl(struct file *filp,
3864 unsigned int ioctl, unsigned long arg)
3866 struct kvm_vcpu *vcpu = filp->private_data;
3867 void __user *argp = compat_ptr(arg);
3870 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3874 case KVM_SET_SIGNAL_MASK: {
3875 struct kvm_signal_mask __user *sigmask_arg = argp;
3876 struct kvm_signal_mask kvm_sigmask;
3881 if (copy_from_user(&kvm_sigmask, argp,
3882 sizeof(kvm_sigmask)))
3885 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3888 if (get_compat_sigset(&sigset,
3889 (compat_sigset_t __user *)sigmask_arg->sigset))
3891 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3893 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3897 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3905 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3907 struct kvm_device *dev = filp->private_data;
3910 return dev->ops->mmap(dev, vma);
3915 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3916 int (*accessor)(struct kvm_device *dev,
3917 struct kvm_device_attr *attr),
3920 struct kvm_device_attr attr;
3925 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3928 return accessor(dev, &attr);
3931 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3934 struct kvm_device *dev = filp->private_data;
3936 if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged)
3940 case KVM_SET_DEVICE_ATTR:
3941 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3942 case KVM_GET_DEVICE_ATTR:
3943 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3944 case KVM_HAS_DEVICE_ATTR:
3945 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3947 if (dev->ops->ioctl)
3948 return dev->ops->ioctl(dev, ioctl, arg);
3954 static int kvm_device_release(struct inode *inode, struct file *filp)
3956 struct kvm_device *dev = filp->private_data;
3957 struct kvm *kvm = dev->kvm;
3959 if (dev->ops->release) {
3960 mutex_lock(&kvm->lock);
3961 list_del(&dev->vm_node);
3962 dev->ops->release(dev);
3963 mutex_unlock(&kvm->lock);
3970 static const struct file_operations kvm_device_fops = {
3971 .unlocked_ioctl = kvm_device_ioctl,
3972 .release = kvm_device_release,
3973 KVM_COMPAT(kvm_device_ioctl),
3974 .mmap = kvm_device_mmap,
3977 struct kvm_device *kvm_device_from_filp(struct file *filp)
3979 if (filp->f_op != &kvm_device_fops)
3982 return filp->private_data;
3985 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3986 #ifdef CONFIG_KVM_MPIC
3987 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3988 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3992 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3994 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3997 if (kvm_device_ops_table[type] != NULL)
4000 kvm_device_ops_table[type] = ops;
4004 void kvm_unregister_device_ops(u32 type)
4006 if (kvm_device_ops_table[type] != NULL)
4007 kvm_device_ops_table[type] = NULL;
4010 static int kvm_ioctl_create_device(struct kvm *kvm,
4011 struct kvm_create_device *cd)
4013 const struct kvm_device_ops *ops = NULL;
4014 struct kvm_device *dev;
4015 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4019 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4022 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4023 ops = kvm_device_ops_table[type];
4030 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4037 mutex_lock(&kvm->lock);
4038 ret = ops->create(dev, type);
4040 mutex_unlock(&kvm->lock);
4044 list_add(&dev->vm_node, &kvm->devices);
4045 mutex_unlock(&kvm->lock);
4051 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4053 kvm_put_kvm_no_destroy(kvm);
4054 mutex_lock(&kvm->lock);
4055 list_del(&dev->vm_node);
4056 mutex_unlock(&kvm->lock);
4065 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4068 case KVM_CAP_USER_MEMORY:
4069 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4070 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4071 case KVM_CAP_INTERNAL_ERROR_DATA:
4072 #ifdef CONFIG_HAVE_KVM_MSI
4073 case KVM_CAP_SIGNAL_MSI:
4075 #ifdef CONFIG_HAVE_KVM_IRQFD
4077 case KVM_CAP_IRQFD_RESAMPLE:
4079 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4080 case KVM_CAP_CHECK_EXTENSION_VM:
4081 case KVM_CAP_ENABLE_CAP_VM:
4082 case KVM_CAP_HALT_POLL:
4084 #ifdef CONFIG_KVM_MMIO
4085 case KVM_CAP_COALESCED_MMIO:
4086 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4087 case KVM_CAP_COALESCED_PIO:
4090 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4091 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4092 return KVM_DIRTY_LOG_MANUAL_CAPS;
4094 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4095 case KVM_CAP_IRQ_ROUTING:
4096 return KVM_MAX_IRQ_ROUTES;
4098 #if KVM_ADDRESS_SPACE_NUM > 1
4099 case KVM_CAP_MULTI_ADDRESS_SPACE:
4100 return KVM_ADDRESS_SPACE_NUM;
4102 case KVM_CAP_NR_MEMSLOTS:
4103 return KVM_USER_MEM_SLOTS;
4104 case KVM_CAP_DIRTY_LOG_RING:
4105 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4106 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4110 case KVM_CAP_BINARY_STATS_FD:
4115 return kvm_vm_ioctl_check_extension(kvm, arg);
4118 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4122 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4125 /* the size should be power of 2 */
4126 if (!size || (size & (size - 1)))
4129 /* Should be bigger to keep the reserved entries, or a page */
4130 if (size < kvm_dirty_ring_get_rsvd_entries() *
4131 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4134 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4135 sizeof(struct kvm_dirty_gfn))
4138 /* We only allow it to set once */
4139 if (kvm->dirty_ring_size)
4142 mutex_lock(&kvm->lock);
4144 if (kvm->created_vcpus) {
4145 /* We don't allow to change this value after vcpu created */
4148 kvm->dirty_ring_size = size;
4152 mutex_unlock(&kvm->lock);
4156 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4159 struct kvm_vcpu *vcpu;
4162 if (!kvm->dirty_ring_size)
4165 mutex_lock(&kvm->slots_lock);
4167 kvm_for_each_vcpu(i, vcpu, kvm)
4168 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4170 mutex_unlock(&kvm->slots_lock);
4173 kvm_flush_remote_tlbs(kvm);
4178 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4179 struct kvm_enable_cap *cap)
4184 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4185 struct kvm_enable_cap *cap)
4188 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4189 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4190 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4192 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4193 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4195 if (cap->flags || (cap->args[0] & ~allowed_options))
4197 kvm->manual_dirty_log_protect = cap->args[0];
4201 case KVM_CAP_HALT_POLL: {
4202 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4205 kvm->max_halt_poll_ns = cap->args[0];
4208 case KVM_CAP_DIRTY_LOG_RING:
4209 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4211 return kvm_vm_ioctl_enable_cap(kvm, cap);
4215 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4216 size_t size, loff_t *offset)
4218 struct kvm *kvm = file->private_data;
4220 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4221 &kvm_vm_stats_desc[0], &kvm->stat,
4222 sizeof(kvm->stat), user_buffer, size, offset);
4225 static const struct file_operations kvm_vm_stats_fops = {
4226 .read = kvm_vm_stats_read,
4227 .llseek = noop_llseek,
4230 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4235 fd = get_unused_fd_flags(O_CLOEXEC);
4239 file = anon_inode_getfile("kvm-vm-stats",
4240 &kvm_vm_stats_fops, kvm, O_RDONLY);
4243 return PTR_ERR(file);
4245 file->f_mode |= FMODE_PREAD;
4246 fd_install(fd, file);
4251 static long kvm_vm_ioctl(struct file *filp,
4252 unsigned int ioctl, unsigned long arg)
4254 struct kvm *kvm = filp->private_data;
4255 void __user *argp = (void __user *)arg;
4258 if (kvm->mm != current->mm || kvm->vm_bugged)
4261 case KVM_CREATE_VCPU:
4262 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4264 case KVM_ENABLE_CAP: {
4265 struct kvm_enable_cap cap;
4268 if (copy_from_user(&cap, argp, sizeof(cap)))
4270 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4273 case KVM_SET_USER_MEMORY_REGION: {
4274 struct kvm_userspace_memory_region kvm_userspace_mem;
4277 if (copy_from_user(&kvm_userspace_mem, argp,
4278 sizeof(kvm_userspace_mem)))
4281 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4284 case KVM_GET_DIRTY_LOG: {
4285 struct kvm_dirty_log log;
4288 if (copy_from_user(&log, argp, sizeof(log)))
4290 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4293 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4294 case KVM_CLEAR_DIRTY_LOG: {
4295 struct kvm_clear_dirty_log log;
4298 if (copy_from_user(&log, argp, sizeof(log)))
4300 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4304 #ifdef CONFIG_KVM_MMIO
4305 case KVM_REGISTER_COALESCED_MMIO: {
4306 struct kvm_coalesced_mmio_zone zone;
4309 if (copy_from_user(&zone, argp, sizeof(zone)))
4311 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4314 case KVM_UNREGISTER_COALESCED_MMIO: {
4315 struct kvm_coalesced_mmio_zone zone;
4318 if (copy_from_user(&zone, argp, sizeof(zone)))
4320 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4325 struct kvm_irqfd data;
4328 if (copy_from_user(&data, argp, sizeof(data)))
4330 r = kvm_irqfd(kvm, &data);
4333 case KVM_IOEVENTFD: {
4334 struct kvm_ioeventfd data;
4337 if (copy_from_user(&data, argp, sizeof(data)))
4339 r = kvm_ioeventfd(kvm, &data);
4342 #ifdef CONFIG_HAVE_KVM_MSI
4343 case KVM_SIGNAL_MSI: {
4347 if (copy_from_user(&msi, argp, sizeof(msi)))
4349 r = kvm_send_userspace_msi(kvm, &msi);
4353 #ifdef __KVM_HAVE_IRQ_LINE
4354 case KVM_IRQ_LINE_STATUS:
4355 case KVM_IRQ_LINE: {
4356 struct kvm_irq_level irq_event;
4359 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4362 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4363 ioctl == KVM_IRQ_LINE_STATUS);
4368 if (ioctl == KVM_IRQ_LINE_STATUS) {
4369 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4377 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4378 case KVM_SET_GSI_ROUTING: {
4379 struct kvm_irq_routing routing;
4380 struct kvm_irq_routing __user *urouting;
4381 struct kvm_irq_routing_entry *entries = NULL;
4384 if (copy_from_user(&routing, argp, sizeof(routing)))
4387 if (!kvm_arch_can_set_irq_routing(kvm))
4389 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4395 entries = vmemdup_user(urouting->entries,
4396 array_size(sizeof(*entries),
4398 if (IS_ERR(entries)) {
4399 r = PTR_ERR(entries);
4403 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4408 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4409 case KVM_CREATE_DEVICE: {
4410 struct kvm_create_device cd;
4413 if (copy_from_user(&cd, argp, sizeof(cd)))
4416 r = kvm_ioctl_create_device(kvm, &cd);
4421 if (copy_to_user(argp, &cd, sizeof(cd)))
4427 case KVM_CHECK_EXTENSION:
4428 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4430 case KVM_RESET_DIRTY_RINGS:
4431 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4433 case KVM_GET_STATS_FD:
4434 r = kvm_vm_ioctl_get_stats_fd(kvm);
4437 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4443 #ifdef CONFIG_KVM_COMPAT
4444 struct compat_kvm_dirty_log {
4448 compat_uptr_t dirty_bitmap; /* one bit per page */
4453 struct compat_kvm_clear_dirty_log {
4458 compat_uptr_t dirty_bitmap; /* one bit per page */
4463 static long kvm_vm_compat_ioctl(struct file *filp,
4464 unsigned int ioctl, unsigned long arg)
4466 struct kvm *kvm = filp->private_data;
4469 if (kvm->mm != current->mm || kvm->vm_bugged)
4472 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4473 case KVM_CLEAR_DIRTY_LOG: {
4474 struct compat_kvm_clear_dirty_log compat_log;
4475 struct kvm_clear_dirty_log log;
4477 if (copy_from_user(&compat_log, (void __user *)arg,
4478 sizeof(compat_log)))
4480 log.slot = compat_log.slot;
4481 log.num_pages = compat_log.num_pages;
4482 log.first_page = compat_log.first_page;
4483 log.padding2 = compat_log.padding2;
4484 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4486 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4490 case KVM_GET_DIRTY_LOG: {
4491 struct compat_kvm_dirty_log compat_log;
4492 struct kvm_dirty_log log;
4494 if (copy_from_user(&compat_log, (void __user *)arg,
4495 sizeof(compat_log)))
4497 log.slot = compat_log.slot;
4498 log.padding1 = compat_log.padding1;
4499 log.padding2 = compat_log.padding2;
4500 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4502 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4506 r = kvm_vm_ioctl(filp, ioctl, arg);
4512 static struct file_operations kvm_vm_fops = {
4513 .release = kvm_vm_release,
4514 .unlocked_ioctl = kvm_vm_ioctl,
4515 .llseek = noop_llseek,
4516 KVM_COMPAT(kvm_vm_compat_ioctl),
4519 bool file_is_kvm(struct file *file)
4521 return file && file->f_op == &kvm_vm_fops;
4523 EXPORT_SYMBOL_GPL(file_is_kvm);
4525 static int kvm_dev_ioctl_create_vm(unsigned long type)
4531 kvm = kvm_create_vm(type);
4533 return PTR_ERR(kvm);
4534 #ifdef CONFIG_KVM_MMIO
4535 r = kvm_coalesced_mmio_init(kvm);
4539 r = get_unused_fd_flags(O_CLOEXEC);
4543 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4544 "kvm-%d", task_pid_nr(current));
4546 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4554 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4555 * already set, with ->release() being kvm_vm_release(). In error
4556 * cases it will be called by the final fput(file) and will take
4557 * care of doing kvm_put_kvm(kvm).
4559 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4564 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4566 fd_install(r, file);
4574 static long kvm_dev_ioctl(struct file *filp,
4575 unsigned int ioctl, unsigned long arg)
4580 case KVM_GET_API_VERSION:
4583 r = KVM_API_VERSION;
4586 r = kvm_dev_ioctl_create_vm(arg);
4588 case KVM_CHECK_EXTENSION:
4589 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4591 case KVM_GET_VCPU_MMAP_SIZE:
4594 r = PAGE_SIZE; /* struct kvm_run */
4596 r += PAGE_SIZE; /* pio data page */
4598 #ifdef CONFIG_KVM_MMIO
4599 r += PAGE_SIZE; /* coalesced mmio ring page */
4602 case KVM_TRACE_ENABLE:
4603 case KVM_TRACE_PAUSE:
4604 case KVM_TRACE_DISABLE:
4608 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4614 static struct file_operations kvm_chardev_ops = {
4615 .unlocked_ioctl = kvm_dev_ioctl,
4616 .llseek = noop_llseek,
4617 KVM_COMPAT(kvm_dev_ioctl),
4620 static struct miscdevice kvm_dev = {
4626 static void hardware_enable_nolock(void *junk)
4628 int cpu = raw_smp_processor_id();
4631 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4634 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4636 r = kvm_arch_hardware_enable();
4639 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4640 atomic_inc(&hardware_enable_failed);
4641 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4645 static int kvm_starting_cpu(unsigned int cpu)
4647 raw_spin_lock(&kvm_count_lock);
4648 if (kvm_usage_count)
4649 hardware_enable_nolock(NULL);
4650 raw_spin_unlock(&kvm_count_lock);
4654 static void hardware_disable_nolock(void *junk)
4656 int cpu = raw_smp_processor_id();
4658 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4660 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4661 kvm_arch_hardware_disable();
4664 static int kvm_dying_cpu(unsigned int cpu)
4666 raw_spin_lock(&kvm_count_lock);
4667 if (kvm_usage_count)
4668 hardware_disable_nolock(NULL);
4669 raw_spin_unlock(&kvm_count_lock);
4673 static void hardware_disable_all_nolock(void)
4675 BUG_ON(!kvm_usage_count);
4678 if (!kvm_usage_count)
4679 on_each_cpu(hardware_disable_nolock, NULL, 1);
4682 static void hardware_disable_all(void)
4684 raw_spin_lock(&kvm_count_lock);
4685 hardware_disable_all_nolock();
4686 raw_spin_unlock(&kvm_count_lock);
4689 static int hardware_enable_all(void)
4693 raw_spin_lock(&kvm_count_lock);
4696 if (kvm_usage_count == 1) {
4697 atomic_set(&hardware_enable_failed, 0);
4698 on_each_cpu(hardware_enable_nolock, NULL, 1);
4700 if (atomic_read(&hardware_enable_failed)) {
4701 hardware_disable_all_nolock();
4706 raw_spin_unlock(&kvm_count_lock);
4711 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4715 * Some (well, at least mine) BIOSes hang on reboot if
4718 * And Intel TXT required VMX off for all cpu when system shutdown.
4720 pr_info("kvm: exiting hardware virtualization\n");
4721 kvm_rebooting = true;
4722 on_each_cpu(hardware_disable_nolock, NULL, 1);
4726 static struct notifier_block kvm_reboot_notifier = {
4727 .notifier_call = kvm_reboot,
4731 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4735 for (i = 0; i < bus->dev_count; i++) {
4736 struct kvm_io_device *pos = bus->range[i].dev;
4738 kvm_iodevice_destructor(pos);
4743 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4744 const struct kvm_io_range *r2)
4746 gpa_t addr1 = r1->addr;
4747 gpa_t addr2 = r2->addr;
4752 /* If r2->len == 0, match the exact address. If r2->len != 0,
4753 * accept any overlapping write. Any order is acceptable for
4754 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4755 * we process all of them.
4768 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4770 return kvm_io_bus_cmp(p1, p2);
4773 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4774 gpa_t addr, int len)
4776 struct kvm_io_range *range, key;
4779 key = (struct kvm_io_range) {
4784 range = bsearch(&key, bus->range, bus->dev_count,
4785 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4789 off = range - bus->range;
4791 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4797 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4798 struct kvm_io_range *range, const void *val)
4802 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4806 while (idx < bus->dev_count &&
4807 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4808 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4817 /* kvm_io_bus_write - called under kvm->slots_lock */
4818 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4819 int len, const void *val)
4821 struct kvm_io_bus *bus;
4822 struct kvm_io_range range;
4825 range = (struct kvm_io_range) {
4830 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4833 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4834 return r < 0 ? r : 0;
4836 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4838 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4839 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4840 gpa_t addr, int len, const void *val, long cookie)
4842 struct kvm_io_bus *bus;
4843 struct kvm_io_range range;
4845 range = (struct kvm_io_range) {
4850 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4854 /* First try the device referenced by cookie. */
4855 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4856 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4857 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4862 * cookie contained garbage; fall back to search and return the
4863 * correct cookie value.
4865 return __kvm_io_bus_write(vcpu, bus, &range, val);
4868 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4869 struct kvm_io_range *range, void *val)
4873 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4877 while (idx < bus->dev_count &&
4878 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4879 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4888 /* kvm_io_bus_read - called under kvm->slots_lock */
4889 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4892 struct kvm_io_bus *bus;
4893 struct kvm_io_range range;
4896 range = (struct kvm_io_range) {
4901 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4904 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4905 return r < 0 ? r : 0;
4908 /* Caller must hold slots_lock. */
4909 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4910 int len, struct kvm_io_device *dev)
4913 struct kvm_io_bus *new_bus, *bus;
4914 struct kvm_io_range range;
4916 bus = kvm_get_bus(kvm, bus_idx);
4920 /* exclude ioeventfd which is limited by maximum fd */
4921 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4924 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4925 GFP_KERNEL_ACCOUNT);
4929 range = (struct kvm_io_range) {
4935 for (i = 0; i < bus->dev_count; i++)
4936 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4939 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4940 new_bus->dev_count++;
4941 new_bus->range[i] = range;
4942 memcpy(new_bus->range + i + 1, bus->range + i,
4943 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4944 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4945 synchronize_srcu_expedited(&kvm->srcu);
4951 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4952 struct kvm_io_device *dev)
4955 struct kvm_io_bus *new_bus, *bus;
4957 lockdep_assert_held(&kvm->slots_lock);
4959 bus = kvm_get_bus(kvm, bus_idx);
4963 for (i = 0; i < bus->dev_count; i++) {
4964 if (bus->range[i].dev == dev) {
4969 if (i == bus->dev_count)
4972 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4973 GFP_KERNEL_ACCOUNT);
4975 memcpy(new_bus, bus, struct_size(bus, range, i));
4976 new_bus->dev_count--;
4977 memcpy(new_bus->range + i, bus->range + i + 1,
4978 flex_array_size(new_bus, range, new_bus->dev_count - i));
4981 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4982 synchronize_srcu_expedited(&kvm->srcu);
4984 /* Destroy the old bus _after_ installing the (null) bus. */
4986 pr_err("kvm: failed to shrink bus, removing it completely\n");
4987 for (j = 0; j < bus->dev_count; j++) {
4990 kvm_iodevice_destructor(bus->range[j].dev);
4995 return new_bus ? 0 : -ENOMEM;
4998 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5001 struct kvm_io_bus *bus;
5002 int dev_idx, srcu_idx;
5003 struct kvm_io_device *iodev = NULL;
5005 srcu_idx = srcu_read_lock(&kvm->srcu);
5007 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5011 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5015 iodev = bus->range[dev_idx].dev;
5018 srcu_read_unlock(&kvm->srcu, srcu_idx);
5022 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5024 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5025 int (*get)(void *, u64 *), int (*set)(void *, u64),
5028 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5032 * The debugfs files are a reference to the kvm struct which
5033 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5034 * avoids the race between open and the removal of the debugfs directory.
5036 if (!kvm_get_kvm_safe(stat_data->kvm))
5039 if (simple_attr_open(inode, file, get,
5040 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5043 kvm_put_kvm(stat_data->kvm);
5050 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5052 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5055 simple_attr_release(inode, file);
5056 kvm_put_kvm(stat_data->kvm);
5061 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5063 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5068 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5070 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5075 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5078 struct kvm_vcpu *vcpu;
5082 kvm_for_each_vcpu(i, vcpu, kvm)
5083 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5088 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5091 struct kvm_vcpu *vcpu;
5093 kvm_for_each_vcpu(i, vcpu, kvm)
5094 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5099 static int kvm_stat_data_get(void *data, u64 *val)
5102 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5104 switch (stat_data->kind) {
5106 r = kvm_get_stat_per_vm(stat_data->kvm,
5107 stat_data->desc->desc.offset, val);
5110 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5111 stat_data->desc->desc.offset, val);
5118 static int kvm_stat_data_clear(void *data, u64 val)
5121 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5126 switch (stat_data->kind) {
5128 r = kvm_clear_stat_per_vm(stat_data->kvm,
5129 stat_data->desc->desc.offset);
5132 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5133 stat_data->desc->desc.offset);
5140 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5142 __simple_attr_check_format("%llu\n", 0ull);
5143 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5144 kvm_stat_data_clear, "%llu\n");
5147 static const struct file_operations stat_fops_per_vm = {
5148 .owner = THIS_MODULE,
5149 .open = kvm_stat_data_open,
5150 .release = kvm_debugfs_release,
5151 .read = simple_attr_read,
5152 .write = simple_attr_write,
5153 .llseek = no_llseek,
5156 static int vm_stat_get(void *_offset, u64 *val)
5158 unsigned offset = (long)_offset;
5163 mutex_lock(&kvm_lock);
5164 list_for_each_entry(kvm, &vm_list, vm_list) {
5165 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5168 mutex_unlock(&kvm_lock);
5172 static int vm_stat_clear(void *_offset, u64 val)
5174 unsigned offset = (long)_offset;
5180 mutex_lock(&kvm_lock);
5181 list_for_each_entry(kvm, &vm_list, vm_list) {
5182 kvm_clear_stat_per_vm(kvm, offset);
5184 mutex_unlock(&kvm_lock);
5189 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5190 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5192 static int vcpu_stat_get(void *_offset, u64 *val)
5194 unsigned offset = (long)_offset;
5199 mutex_lock(&kvm_lock);
5200 list_for_each_entry(kvm, &vm_list, vm_list) {
5201 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5204 mutex_unlock(&kvm_lock);
5208 static int vcpu_stat_clear(void *_offset, u64 val)
5210 unsigned offset = (long)_offset;
5216 mutex_lock(&kvm_lock);
5217 list_for_each_entry(kvm, &vm_list, vm_list) {
5218 kvm_clear_stat_per_vcpu(kvm, offset);
5220 mutex_unlock(&kvm_lock);
5225 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5227 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5229 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5231 struct kobj_uevent_env *env;
5232 unsigned long long created, active;
5234 if (!kvm_dev.this_device || !kvm)
5237 mutex_lock(&kvm_lock);
5238 if (type == KVM_EVENT_CREATE_VM) {
5239 kvm_createvm_count++;
5241 } else if (type == KVM_EVENT_DESTROY_VM) {
5244 created = kvm_createvm_count;
5245 active = kvm_active_vms;
5246 mutex_unlock(&kvm_lock);
5248 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5252 add_uevent_var(env, "CREATED=%llu", created);
5253 add_uevent_var(env, "COUNT=%llu", active);
5255 if (type == KVM_EVENT_CREATE_VM) {
5256 add_uevent_var(env, "EVENT=create");
5257 kvm->userspace_pid = task_pid_nr(current);
5258 } else if (type == KVM_EVENT_DESTROY_VM) {
5259 add_uevent_var(env, "EVENT=destroy");
5261 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5263 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
5264 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5267 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5269 add_uevent_var(env, "STATS_PATH=%s", tmp);
5273 /* no need for checks, since we are adding at most only 5 keys */
5274 env->envp[env->envp_idx++] = NULL;
5275 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5279 static void kvm_init_debug(void)
5281 const struct file_operations *fops;
5282 const struct _kvm_stats_desc *pdesc;
5285 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5287 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5288 pdesc = &kvm_vm_stats_desc[i];
5289 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5290 fops = &vm_stat_fops;
5292 fops = &vm_stat_readonly_fops;
5293 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5295 (void *)(long)pdesc->desc.offset, fops);
5298 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5299 pdesc = &kvm_vcpu_stats_desc[i];
5300 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5301 fops = &vcpu_stat_fops;
5303 fops = &vcpu_stat_readonly_fops;
5304 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5306 (void *)(long)pdesc->desc.offset, fops);
5310 static int kvm_suspend(void)
5312 if (kvm_usage_count)
5313 hardware_disable_nolock(NULL);
5317 static void kvm_resume(void)
5319 if (kvm_usage_count) {
5320 #ifdef CONFIG_LOCKDEP
5321 WARN_ON(lockdep_is_held(&kvm_count_lock));
5323 hardware_enable_nolock(NULL);
5327 static struct syscore_ops kvm_syscore_ops = {
5328 .suspend = kvm_suspend,
5329 .resume = kvm_resume,
5333 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5335 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5338 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5340 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5342 WRITE_ONCE(vcpu->preempted, false);
5343 WRITE_ONCE(vcpu->ready, false);
5345 __this_cpu_write(kvm_running_vcpu, vcpu);
5346 kvm_arch_sched_in(vcpu, cpu);
5347 kvm_arch_vcpu_load(vcpu, cpu);
5350 static void kvm_sched_out(struct preempt_notifier *pn,
5351 struct task_struct *next)
5353 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5355 if (current->on_rq) {
5356 WRITE_ONCE(vcpu->preempted, true);
5357 WRITE_ONCE(vcpu->ready, true);
5359 kvm_arch_vcpu_put(vcpu);
5360 __this_cpu_write(kvm_running_vcpu, NULL);
5364 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5366 * We can disable preemption locally around accessing the per-CPU variable,
5367 * and use the resolved vcpu pointer after enabling preemption again,
5368 * because even if the current thread is migrated to another CPU, reading
5369 * the per-CPU value later will give us the same value as we update the
5370 * per-CPU variable in the preempt notifier handlers.
5372 struct kvm_vcpu *kvm_get_running_vcpu(void)
5374 struct kvm_vcpu *vcpu;
5377 vcpu = __this_cpu_read(kvm_running_vcpu);
5382 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5385 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5387 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5389 return &kvm_running_vcpu;
5392 struct kvm_cpu_compat_check {
5397 static void check_processor_compat(void *data)
5399 struct kvm_cpu_compat_check *c = data;
5401 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5404 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5405 struct module *module)
5407 struct kvm_cpu_compat_check c;
5411 r = kvm_arch_init(opaque);
5416 * kvm_arch_init makes sure there's at most one caller
5417 * for architectures that support multiple implementations,
5418 * like intel and amd on x86.
5419 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5420 * conflicts in case kvm is already setup for another implementation.
5422 r = kvm_irqfd_init();
5426 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5431 r = kvm_arch_hardware_setup(opaque);
5437 for_each_online_cpu(cpu) {
5438 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5443 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5444 kvm_starting_cpu, kvm_dying_cpu);
5447 register_reboot_notifier(&kvm_reboot_notifier);
5449 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5451 vcpu_align = __alignof__(struct kvm_vcpu);
5453 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5455 offsetof(struct kvm_vcpu, arch),
5456 offsetofend(struct kvm_vcpu, stats_id)
5457 - offsetof(struct kvm_vcpu, arch),
5459 if (!kvm_vcpu_cache) {
5464 r = kvm_async_pf_init();
5468 kvm_chardev_ops.owner = module;
5469 kvm_vm_fops.owner = module;
5470 kvm_vcpu_fops.owner = module;
5472 r = misc_register(&kvm_dev);
5474 pr_err("kvm: misc device register failed\n");
5478 register_syscore_ops(&kvm_syscore_ops);
5480 kvm_preempt_ops.sched_in = kvm_sched_in;
5481 kvm_preempt_ops.sched_out = kvm_sched_out;
5485 r = kvm_vfio_ops_init();
5491 kvm_async_pf_deinit();
5493 kmem_cache_destroy(kvm_vcpu_cache);
5495 unregister_reboot_notifier(&kvm_reboot_notifier);
5496 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5498 kvm_arch_hardware_unsetup();
5500 free_cpumask_var(cpus_hardware_enabled);
5508 EXPORT_SYMBOL_GPL(kvm_init);
5512 debugfs_remove_recursive(kvm_debugfs_dir);
5513 misc_deregister(&kvm_dev);
5514 kmem_cache_destroy(kvm_vcpu_cache);
5515 kvm_async_pf_deinit();
5516 unregister_syscore_ops(&kvm_syscore_ops);
5517 unregister_reboot_notifier(&kvm_reboot_notifier);
5518 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5519 on_each_cpu(hardware_disable_nolock, NULL, 1);
5520 kvm_arch_hardware_unsetup();
5523 free_cpumask_var(cpus_hardware_enabled);
5524 kvm_vfio_ops_exit();
5526 EXPORT_SYMBOL_GPL(kvm_exit);
5528 struct kvm_vm_worker_thread_context {
5530 struct task_struct *parent;
5531 struct completion init_done;
5532 kvm_vm_thread_fn_t thread_fn;
5537 static int kvm_vm_worker_thread(void *context)
5540 * The init_context is allocated on the stack of the parent thread, so
5541 * we have to locally copy anything that is needed beyond initialization
5543 struct kvm_vm_worker_thread_context *init_context = context;
5544 struct kvm *kvm = init_context->kvm;
5545 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5546 uintptr_t data = init_context->data;
5549 err = kthread_park(current);
5550 /* kthread_park(current) is never supposed to return an error */
5555 err = cgroup_attach_task_all(init_context->parent, current);
5557 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5562 set_user_nice(current, task_nice(init_context->parent));
5565 init_context->err = err;
5566 complete(&init_context->init_done);
5567 init_context = NULL;
5572 /* Wait to be woken up by the spawner before proceeding. */
5575 if (!kthread_should_stop())
5576 err = thread_fn(kvm, data);
5581 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5582 uintptr_t data, const char *name,
5583 struct task_struct **thread_ptr)
5585 struct kvm_vm_worker_thread_context init_context = {};
5586 struct task_struct *thread;
5589 init_context.kvm = kvm;
5590 init_context.parent = current;
5591 init_context.thread_fn = thread_fn;
5592 init_context.data = data;
5593 init_completion(&init_context.init_done);
5595 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5596 "%s-%d", name, task_pid_nr(current));
5598 return PTR_ERR(thread);
5600 /* kthread_run is never supposed to return NULL */
5601 WARN_ON(thread == NULL);
5603 wait_for_completion(&init_context.init_done);
5605 if (!init_context.err)
5606 *thread_ptr = thread;
5608 return init_context.err;