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 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
160 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 unsigned long start, unsigned long end)
165 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
168 * The metadata used by is_zone_device_page() to determine whether or
169 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
170 * the device has been pinned, e.g. by get_user_pages(). WARN if the
171 * page_count() is zero to help detect bad usage of this helper.
173 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
176 return is_zone_device_page(pfn_to_page(pfn));
179 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
182 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
183 * perspective they are "normal" pages, albeit with slightly different
187 return PageReserved(pfn_to_page(pfn)) &&
189 !kvm_is_zone_device_pfn(pfn);
195 * Switches to specified vcpu, until a matching vcpu_put()
197 void vcpu_load(struct kvm_vcpu *vcpu)
201 __this_cpu_write(kvm_running_vcpu, vcpu);
202 preempt_notifier_register(&vcpu->preempt_notifier);
203 kvm_arch_vcpu_load(vcpu, cpu);
206 EXPORT_SYMBOL_GPL(vcpu_load);
208 void vcpu_put(struct kvm_vcpu *vcpu)
211 kvm_arch_vcpu_put(vcpu);
212 preempt_notifier_unregister(&vcpu->preempt_notifier);
213 __this_cpu_write(kvm_running_vcpu, NULL);
216 EXPORT_SYMBOL_GPL(vcpu_put);
218 /* TODO: merge with kvm_arch_vcpu_should_kick */
219 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
221 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
224 * We need to wait for the VCPU to reenable interrupts and get out of
225 * READING_SHADOW_PAGE_TABLES mode.
227 if (req & KVM_REQUEST_WAIT)
228 return mode != OUTSIDE_GUEST_MODE;
231 * Need to kick a running VCPU, but otherwise there is nothing to do.
233 return mode == IN_GUEST_MODE;
236 static void ack_flush(void *_completed)
240 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
242 if (cpumask_empty(cpus))
245 smp_call_function_many(cpus, ack_flush, NULL, wait);
249 static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu,
250 unsigned int req, struct cpumask *tmp,
255 kvm_make_request(req, vcpu);
257 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
261 * Note, the vCPU could get migrated to a different pCPU at any point
262 * after kvm_request_needs_ipi(), which could result in sending an IPI
263 * to the previous pCPU. But, that's OK because the purpose of the IPI
264 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
265 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
266 * after this point is also OK, as the requirement is only that KVM wait
267 * for vCPUs that were reading SPTEs _before_ any changes were
268 * finalized. See kvm_vcpu_kick() for more details on handling requests.
270 if (kvm_request_needs_ipi(vcpu, req)) {
271 cpu = READ_ONCE(vcpu->cpu);
272 if (cpu != -1 && cpu != current_cpu)
273 __cpumask_set_cpu(cpu, tmp);
277 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
278 unsigned long *vcpu_bitmap)
280 struct kvm_vcpu *vcpu;
281 struct cpumask *cpus;
287 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
290 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
291 vcpu = kvm_get_vcpu(kvm, i);
294 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
297 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
303 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
304 struct kvm_vcpu *except)
306 struct kvm_vcpu *vcpu;
307 struct cpumask *cpus;
314 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
317 kvm_for_each_vcpu(i, vcpu, kvm) {
320 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
323 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
329 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
331 return kvm_make_all_cpus_request_except(kvm, req, NULL);
333 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
335 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
336 void kvm_flush_remote_tlbs(struct kvm *kvm)
338 ++kvm->stat.generic.remote_tlb_flush_requests;
341 * We want to publish modifications to the page tables before reading
342 * mode. Pairs with a memory barrier in arch-specific code.
343 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
344 * and smp_mb in walk_shadow_page_lockless_begin/end.
345 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
347 * There is already an smp_mb__after_atomic() before
348 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
351 if (!kvm_arch_flush_remote_tlb(kvm)
352 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
353 ++kvm->stat.generic.remote_tlb_flush;
355 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
358 void kvm_reload_remote_mmus(struct kvm *kvm)
360 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
363 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
364 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
367 gfp_flags |= mc->gfp_zero;
370 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
372 return (void *)__get_free_page(gfp_flags);
375 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
379 if (mc->nobjs >= min)
381 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
382 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
384 return mc->nobjs >= min ? 0 : -ENOMEM;
385 mc->objects[mc->nobjs++] = obj;
390 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
395 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
399 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
401 free_page((unsigned long)mc->objects[--mc->nobjs]);
405 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
409 if (WARN_ON(!mc->nobjs))
410 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
412 p = mc->objects[--mc->nobjs];
418 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
420 mutex_init(&vcpu->mutex);
425 #ifndef __KVM_HAVE_ARCH_WQP
426 rcuwait_init(&vcpu->wait);
428 kvm_async_pf_vcpu_init(vcpu);
431 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
433 kvm_vcpu_set_in_spin_loop(vcpu, false);
434 kvm_vcpu_set_dy_eligible(vcpu, false);
435 vcpu->preempted = false;
437 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
438 vcpu->last_used_slot = NULL;
441 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
443 kvm_dirty_ring_free(&vcpu->dirty_ring);
444 kvm_arch_vcpu_destroy(vcpu);
447 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
448 * the vcpu->pid pointer, and at destruction time all file descriptors
451 put_pid(rcu_dereference_protected(vcpu->pid, 1));
453 free_page((unsigned long)vcpu->run);
454 kmem_cache_free(kvm_vcpu_cache, vcpu);
457 void kvm_destroy_vcpus(struct kvm *kvm)
460 struct kvm_vcpu *vcpu;
462 kvm_for_each_vcpu(i, vcpu, kvm) {
463 kvm_vcpu_destroy(vcpu);
464 xa_erase(&kvm->vcpu_array, i);
467 atomic_set(&kvm->online_vcpus, 0);
469 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
471 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
472 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
474 return container_of(mn, struct kvm, mmu_notifier);
477 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
478 struct mm_struct *mm,
479 unsigned long start, unsigned long end)
481 struct kvm *kvm = mmu_notifier_to_kvm(mn);
484 idx = srcu_read_lock(&kvm->srcu);
485 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
486 srcu_read_unlock(&kvm->srcu, idx);
489 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
491 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
494 struct kvm_hva_range {
498 hva_handler_t handler;
499 on_lock_fn_t on_lock;
505 * Use a dedicated stub instead of NULL to indicate that there is no callback
506 * function/handler. The compiler technically can't guarantee that a real
507 * function will have a non-zero address, and so it will generate code to
508 * check for !NULL, whereas comparing against a stub will be elided at compile
509 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
511 static void kvm_null_fn(void)
515 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
517 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
518 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
519 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
521 node = interval_tree_iter_next(node, start, last)) \
523 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
524 const struct kvm_hva_range *range)
526 bool ret = false, locked = false;
527 struct kvm_gfn_range gfn_range;
528 struct kvm_memory_slot *slot;
529 struct kvm_memslots *slots;
532 if (WARN_ON_ONCE(range->end <= range->start))
535 /* A null handler is allowed if and only if on_lock() is provided. */
536 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
537 IS_KVM_NULL_FN(range->handler)))
540 idx = srcu_read_lock(&kvm->srcu);
542 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
543 struct interval_tree_node *node;
545 slots = __kvm_memslots(kvm, i);
546 kvm_for_each_memslot_in_hva_range(node, slots,
547 range->start, range->end - 1) {
548 unsigned long hva_start, hva_end;
550 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
551 hva_start = max(range->start, slot->userspace_addr);
552 hva_end = min(range->end, slot->userspace_addr +
553 (slot->npages << PAGE_SHIFT));
556 * To optimize for the likely case where the address
557 * range is covered by zero or one memslots, don't
558 * bother making these conditional (to avoid writes on
559 * the second or later invocation of the handler).
561 gfn_range.pte = range->pte;
562 gfn_range.may_block = range->may_block;
565 * {gfn(page) | page intersects with [hva_start, hva_end)} =
566 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
568 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
569 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
570 gfn_range.slot = slot;
575 if (!IS_KVM_NULL_FN(range->on_lock))
576 range->on_lock(kvm, range->start, range->end);
577 if (IS_KVM_NULL_FN(range->handler))
580 ret |= range->handler(kvm, &gfn_range);
584 if (range->flush_on_ret && ret)
585 kvm_flush_remote_tlbs(kvm);
590 srcu_read_unlock(&kvm->srcu, idx);
592 /* The notifiers are averse to booleans. :-( */
596 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
600 hva_handler_t handler)
602 struct kvm *kvm = mmu_notifier_to_kvm(mn);
603 const struct kvm_hva_range range = {
608 .on_lock = (void *)kvm_null_fn,
609 .flush_on_ret = true,
613 return __kvm_handle_hva_range(kvm, &range);
616 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
619 hva_handler_t handler)
621 struct kvm *kvm = mmu_notifier_to_kvm(mn);
622 const struct kvm_hva_range range = {
627 .on_lock = (void *)kvm_null_fn,
628 .flush_on_ret = false,
632 return __kvm_handle_hva_range(kvm, &range);
634 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
635 struct mm_struct *mm,
636 unsigned long address,
639 struct kvm *kvm = mmu_notifier_to_kvm(mn);
641 trace_kvm_set_spte_hva(address);
644 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
645 * If mmu_notifier_count is zero, then no in-progress invalidations,
646 * including this one, found a relevant memslot at start(); rechecking
647 * memslots here is unnecessary. Note, a false positive (count elevated
648 * by a different invalidation) is sub-optimal but functionally ok.
650 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
651 if (!READ_ONCE(kvm->mmu_notifier_count))
654 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
657 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
661 * The count increase must become visible at unlock time as no
662 * spte can be established without taking the mmu_lock and
663 * count is also read inside the mmu_lock critical section.
665 kvm->mmu_notifier_count++;
666 if (likely(kvm->mmu_notifier_count == 1)) {
667 kvm->mmu_notifier_range_start = start;
668 kvm->mmu_notifier_range_end = end;
671 * Fully tracking multiple concurrent ranges has dimishing
672 * returns. Keep things simple and just find the minimal range
673 * which includes the current and new ranges. As there won't be
674 * enough information to subtract a range after its invalidate
675 * completes, any ranges invalidated concurrently will
676 * accumulate and persist until all outstanding invalidates
679 kvm->mmu_notifier_range_start =
680 min(kvm->mmu_notifier_range_start, start);
681 kvm->mmu_notifier_range_end =
682 max(kvm->mmu_notifier_range_end, end);
686 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
687 const struct mmu_notifier_range *range)
689 struct kvm *kvm = mmu_notifier_to_kvm(mn);
690 const struct kvm_hva_range hva_range = {
691 .start = range->start,
694 .handler = kvm_unmap_gfn_range,
695 .on_lock = kvm_inc_notifier_count,
696 .flush_on_ret = true,
697 .may_block = mmu_notifier_range_blockable(range),
700 trace_kvm_unmap_hva_range(range->start, range->end);
703 * Prevent memslot modification between range_start() and range_end()
704 * so that conditionally locking provides the same result in both
705 * functions. Without that guarantee, the mmu_notifier_count
706 * adjustments will be imbalanced.
708 * Pairs with the decrement in range_end().
710 spin_lock(&kvm->mn_invalidate_lock);
711 kvm->mn_active_invalidate_count++;
712 spin_unlock(&kvm->mn_invalidate_lock);
714 __kvm_handle_hva_range(kvm, &hva_range);
719 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
723 * This sequence increase will notify the kvm page fault that
724 * the page that is going to be mapped in the spte could have
727 kvm->mmu_notifier_seq++;
730 * The above sequence increase must be visible before the
731 * below count decrease, which is ensured by the smp_wmb above
732 * in conjunction with the smp_rmb in mmu_notifier_retry().
734 kvm->mmu_notifier_count--;
737 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
738 const struct mmu_notifier_range *range)
740 struct kvm *kvm = mmu_notifier_to_kvm(mn);
741 const struct kvm_hva_range hva_range = {
742 .start = range->start,
745 .handler = (void *)kvm_null_fn,
746 .on_lock = kvm_dec_notifier_count,
747 .flush_on_ret = false,
748 .may_block = mmu_notifier_range_blockable(range),
752 __kvm_handle_hva_range(kvm, &hva_range);
754 /* Pairs with the increment in range_start(). */
755 spin_lock(&kvm->mn_invalidate_lock);
756 wake = (--kvm->mn_active_invalidate_count == 0);
757 spin_unlock(&kvm->mn_invalidate_lock);
760 * There can only be one waiter, since the wait happens under
764 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
766 BUG_ON(kvm->mmu_notifier_count < 0);
769 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
770 struct mm_struct *mm,
774 trace_kvm_age_hva(start, end);
776 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
779 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
780 struct mm_struct *mm,
784 trace_kvm_age_hva(start, end);
787 * Even though we do not flush TLB, this will still adversely
788 * affect performance on pre-Haswell Intel EPT, where there is
789 * no EPT Access Bit to clear so that we have to tear down EPT
790 * tables instead. If we find this unacceptable, we can always
791 * add a parameter to kvm_age_hva so that it effectively doesn't
792 * do anything on clear_young.
794 * Also note that currently we never issue secondary TLB flushes
795 * from clear_young, leaving this job up to the regular system
796 * cadence. If we find this inaccurate, we might come up with a
797 * more sophisticated heuristic later.
799 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
802 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
803 struct mm_struct *mm,
804 unsigned long address)
806 trace_kvm_test_age_hva(address);
808 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
812 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
813 struct mm_struct *mm)
815 struct kvm *kvm = mmu_notifier_to_kvm(mn);
818 idx = srcu_read_lock(&kvm->srcu);
819 kvm_arch_flush_shadow_all(kvm);
820 srcu_read_unlock(&kvm->srcu, idx);
823 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
824 .invalidate_range = kvm_mmu_notifier_invalidate_range,
825 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
826 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
827 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
828 .clear_young = kvm_mmu_notifier_clear_young,
829 .test_young = kvm_mmu_notifier_test_young,
830 .change_pte = kvm_mmu_notifier_change_pte,
831 .release = kvm_mmu_notifier_release,
834 static int kvm_init_mmu_notifier(struct kvm *kvm)
836 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
837 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
840 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
842 static int kvm_init_mmu_notifier(struct kvm *kvm)
847 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
849 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
850 static int kvm_pm_notifier_call(struct notifier_block *bl,
854 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
856 return kvm_arch_pm_notifier(kvm, state);
859 static void kvm_init_pm_notifier(struct kvm *kvm)
861 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
862 /* Suspend KVM before we suspend ftrace, RCU, etc. */
863 kvm->pm_notifier.priority = INT_MAX;
864 register_pm_notifier(&kvm->pm_notifier);
867 static void kvm_destroy_pm_notifier(struct kvm *kvm)
869 unregister_pm_notifier(&kvm->pm_notifier);
871 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
872 static void kvm_init_pm_notifier(struct kvm *kvm)
876 static void kvm_destroy_pm_notifier(struct kvm *kvm)
879 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
881 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
883 if (!memslot->dirty_bitmap)
886 kvfree(memslot->dirty_bitmap);
887 memslot->dirty_bitmap = NULL;
890 /* This does not remove the slot from struct kvm_memslots data structures */
891 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
893 kvm_destroy_dirty_bitmap(slot);
895 kvm_arch_free_memslot(kvm, slot);
900 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
902 struct hlist_node *idnode;
903 struct kvm_memory_slot *memslot;
907 * The same memslot objects live in both active and inactive sets,
908 * arbitrarily free using index '1' so the second invocation of this
909 * function isn't operating over a structure with dangling pointers
910 * (even though this function isn't actually touching them).
912 if (!slots->node_idx)
915 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
916 kvm_free_memslot(kvm, memslot);
919 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
921 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
922 case KVM_STATS_TYPE_INSTANT:
924 case KVM_STATS_TYPE_CUMULATIVE:
925 case KVM_STATS_TYPE_PEAK:
932 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
935 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
936 kvm_vcpu_stats_header.num_desc;
938 if (!kvm->debugfs_dentry)
941 debugfs_remove_recursive(kvm->debugfs_dentry);
943 if (kvm->debugfs_stat_data) {
944 for (i = 0; i < kvm_debugfs_num_entries; i++)
945 kfree(kvm->debugfs_stat_data[i]);
946 kfree(kvm->debugfs_stat_data);
950 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
952 static DEFINE_MUTEX(kvm_debugfs_lock);
954 char dir_name[ITOA_MAX_LEN * 2];
955 struct kvm_stat_data *stat_data;
956 const struct _kvm_stats_desc *pdesc;
958 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
959 kvm_vcpu_stats_header.num_desc;
961 if (!debugfs_initialized())
964 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
965 mutex_lock(&kvm_debugfs_lock);
966 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
968 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
970 mutex_unlock(&kvm_debugfs_lock);
973 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
974 mutex_unlock(&kvm_debugfs_lock);
978 kvm->debugfs_dentry = dent;
979 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
980 sizeof(*kvm->debugfs_stat_data),
982 if (!kvm->debugfs_stat_data)
985 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
986 pdesc = &kvm_vm_stats_desc[i];
987 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
991 stat_data->kvm = kvm;
992 stat_data->desc = pdesc;
993 stat_data->kind = KVM_STAT_VM;
994 kvm->debugfs_stat_data[i] = stat_data;
995 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
996 kvm->debugfs_dentry, stat_data,
1000 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1001 pdesc = &kvm_vcpu_stats_desc[i];
1002 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1006 stat_data->kvm = kvm;
1007 stat_data->desc = pdesc;
1008 stat_data->kind = KVM_STAT_VCPU;
1009 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1010 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1011 kvm->debugfs_dentry, stat_data,
1015 ret = kvm_arch_create_vm_debugfs(kvm);
1017 kvm_destroy_vm_debugfs(kvm);
1025 * Called after the VM is otherwise initialized, but just before adding it to
1028 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1034 * Called just after removing the VM from the vm_list, but before doing any
1035 * other destruction.
1037 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1042 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1043 * be setup already, so we can create arch-specific debugfs entries under it.
1044 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1045 * a per-arch destroy interface is not needed.
1047 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1052 static struct kvm *kvm_create_vm(unsigned long type)
1054 struct kvm *kvm = kvm_arch_alloc_vm();
1055 struct kvm_memslots *slots;
1060 return ERR_PTR(-ENOMEM);
1062 KVM_MMU_LOCK_INIT(kvm);
1063 mmgrab(current->mm);
1064 kvm->mm = current->mm;
1065 kvm_eventfd_init(kvm);
1066 mutex_init(&kvm->lock);
1067 mutex_init(&kvm->irq_lock);
1068 mutex_init(&kvm->slots_lock);
1069 mutex_init(&kvm->slots_arch_lock);
1070 spin_lock_init(&kvm->mn_invalidate_lock);
1071 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1072 xa_init(&kvm->vcpu_array);
1074 INIT_LIST_HEAD(&kvm->devices);
1076 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1078 if (init_srcu_struct(&kvm->srcu))
1079 goto out_err_no_srcu;
1080 if (init_srcu_struct(&kvm->irq_srcu))
1081 goto out_err_no_irq_srcu;
1083 refcount_set(&kvm->users_count, 1);
1084 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1085 for (j = 0; j < 2; j++) {
1086 slots = &kvm->__memslots[i][j];
1088 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1089 slots->hva_tree = RB_ROOT_CACHED;
1090 slots->gfn_tree = RB_ROOT;
1091 hash_init(slots->id_hash);
1092 slots->node_idx = j;
1094 /* Generations must be different for each address space. */
1095 slots->generation = i;
1098 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1101 for (i = 0; i < KVM_NR_BUSES; i++) {
1102 rcu_assign_pointer(kvm->buses[i],
1103 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1105 goto out_err_no_arch_destroy_vm;
1108 kvm->max_halt_poll_ns = halt_poll_ns;
1110 r = kvm_arch_init_vm(kvm, type);
1112 goto out_err_no_arch_destroy_vm;
1114 r = hardware_enable_all();
1116 goto out_err_no_disable;
1118 #ifdef CONFIG_HAVE_KVM_IRQFD
1119 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1122 r = kvm_init_mmu_notifier(kvm);
1124 goto out_err_no_mmu_notifier;
1126 r = kvm_arch_post_init_vm(kvm);
1130 mutex_lock(&kvm_lock);
1131 list_add(&kvm->vm_list, &vm_list);
1132 mutex_unlock(&kvm_lock);
1134 preempt_notifier_inc();
1135 kvm_init_pm_notifier(kvm);
1140 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1141 if (kvm->mmu_notifier.ops)
1142 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1144 out_err_no_mmu_notifier:
1145 hardware_disable_all();
1147 kvm_arch_destroy_vm(kvm);
1148 out_err_no_arch_destroy_vm:
1149 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1150 for (i = 0; i < KVM_NR_BUSES; i++)
1151 kfree(kvm_get_bus(kvm, i));
1152 cleanup_srcu_struct(&kvm->irq_srcu);
1153 out_err_no_irq_srcu:
1154 cleanup_srcu_struct(&kvm->srcu);
1156 kvm_arch_free_vm(kvm);
1157 mmdrop(current->mm);
1161 static void kvm_destroy_devices(struct kvm *kvm)
1163 struct kvm_device *dev, *tmp;
1166 * We do not need to take the kvm->lock here, because nobody else
1167 * has a reference to the struct kvm at this point and therefore
1168 * cannot access the devices list anyhow.
1170 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1171 list_del(&dev->vm_node);
1172 dev->ops->destroy(dev);
1176 static void kvm_destroy_vm(struct kvm *kvm)
1179 struct mm_struct *mm = kvm->mm;
1181 kvm_destroy_pm_notifier(kvm);
1182 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1183 kvm_destroy_vm_debugfs(kvm);
1184 kvm_arch_sync_events(kvm);
1185 mutex_lock(&kvm_lock);
1186 list_del(&kvm->vm_list);
1187 mutex_unlock(&kvm_lock);
1188 kvm_arch_pre_destroy_vm(kvm);
1190 kvm_free_irq_routing(kvm);
1191 for (i = 0; i < KVM_NR_BUSES; i++) {
1192 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1195 kvm_io_bus_destroy(bus);
1196 kvm->buses[i] = NULL;
1198 kvm_coalesced_mmio_free(kvm);
1199 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1200 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1202 * At this point, pending calls to invalidate_range_start()
1203 * have completed but no more MMU notifiers will run, so
1204 * mn_active_invalidate_count may remain unbalanced.
1205 * No threads can be waiting in install_new_memslots as the
1206 * last reference on KVM has been dropped, but freeing
1207 * memslots would deadlock without this manual intervention.
1209 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1210 kvm->mn_active_invalidate_count = 0;
1212 kvm_arch_flush_shadow_all(kvm);
1214 kvm_arch_destroy_vm(kvm);
1215 kvm_destroy_devices(kvm);
1216 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1217 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1218 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1220 cleanup_srcu_struct(&kvm->irq_srcu);
1221 cleanup_srcu_struct(&kvm->srcu);
1222 kvm_arch_free_vm(kvm);
1223 preempt_notifier_dec();
1224 hardware_disable_all();
1228 void kvm_get_kvm(struct kvm *kvm)
1230 refcount_inc(&kvm->users_count);
1232 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1235 * Make sure the vm is not during destruction, which is a safe version of
1236 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1238 bool kvm_get_kvm_safe(struct kvm *kvm)
1240 return refcount_inc_not_zero(&kvm->users_count);
1242 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1244 void kvm_put_kvm(struct kvm *kvm)
1246 if (refcount_dec_and_test(&kvm->users_count))
1247 kvm_destroy_vm(kvm);
1249 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1252 * Used to put a reference that was taken on behalf of an object associated
1253 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1254 * of the new file descriptor fails and the reference cannot be transferred to
1255 * its final owner. In such cases, the caller is still actively using @kvm and
1256 * will fail miserably if the refcount unexpectedly hits zero.
1258 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1260 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1262 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1264 static int kvm_vm_release(struct inode *inode, struct file *filp)
1266 struct kvm *kvm = filp->private_data;
1268 kvm_irqfd_release(kvm);
1275 * Allocation size is twice as large as the actual dirty bitmap size.
1276 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1278 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1280 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1282 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1283 if (!memslot->dirty_bitmap)
1289 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1291 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1292 int node_idx_inactive = active->node_idx ^ 1;
1294 return &kvm->__memslots[as_id][node_idx_inactive];
1298 * Helper to get the address space ID when one of memslot pointers may be NULL.
1299 * This also serves as a sanity that at least one of the pointers is non-NULL,
1300 * and that their address space IDs don't diverge.
1302 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1303 struct kvm_memory_slot *b)
1305 if (WARN_ON_ONCE(!a && !b))
1313 WARN_ON_ONCE(a->as_id != b->as_id);
1317 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1318 struct kvm_memory_slot *slot)
1320 struct rb_root *gfn_tree = &slots->gfn_tree;
1321 struct rb_node **node, *parent;
1322 int idx = slots->node_idx;
1325 for (node = &gfn_tree->rb_node; *node; ) {
1326 struct kvm_memory_slot *tmp;
1328 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1330 if (slot->base_gfn < tmp->base_gfn)
1331 node = &(*node)->rb_left;
1332 else if (slot->base_gfn > tmp->base_gfn)
1333 node = &(*node)->rb_right;
1338 rb_link_node(&slot->gfn_node[idx], parent, node);
1339 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1342 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1343 struct kvm_memory_slot *slot)
1345 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1348 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1349 struct kvm_memory_slot *old,
1350 struct kvm_memory_slot *new)
1352 int idx = slots->node_idx;
1354 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1356 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1361 * Replace @old with @new in the inactive memslots.
1363 * With NULL @old this simply adds @new.
1364 * With NULL @new this simply removes @old.
1366 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1369 static void kvm_replace_memslot(struct kvm *kvm,
1370 struct kvm_memory_slot *old,
1371 struct kvm_memory_slot *new)
1373 int as_id = kvm_memslots_get_as_id(old, new);
1374 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1375 int idx = slots->node_idx;
1378 hash_del(&old->id_node[idx]);
1379 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1381 if ((long)old == atomic_long_read(&slots->last_used_slot))
1382 atomic_long_set(&slots->last_used_slot, (long)new);
1385 kvm_erase_gfn_node(slots, old);
1391 * Initialize @new's hva range. Do this even when replacing an @old
1392 * slot, kvm_copy_memslot() deliberately does not touch node data.
1394 new->hva_node[idx].start = new->userspace_addr;
1395 new->hva_node[idx].last = new->userspace_addr +
1396 (new->npages << PAGE_SHIFT) - 1;
1399 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1400 * hva_node needs to be swapped with remove+insert even though hva can't
1401 * change when replacing an existing slot.
1403 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1404 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1407 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1408 * switch the node in the gfn tree instead of removing the old and
1409 * inserting the new as two separate operations. Replacement is a
1410 * single O(1) operation versus two O(log(n)) operations for
1413 if (old && old->base_gfn == new->base_gfn) {
1414 kvm_replace_gfn_node(slots, old, new);
1417 kvm_erase_gfn_node(slots, old);
1418 kvm_insert_gfn_node(slots, new);
1422 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1424 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1426 #ifdef __KVM_HAVE_READONLY_MEM
1427 valid_flags |= KVM_MEM_READONLY;
1430 if (mem->flags & ~valid_flags)
1436 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1438 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1440 /* Grab the generation from the activate memslots. */
1441 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1443 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1444 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1447 * Do not store the new memslots while there are invalidations in
1448 * progress, otherwise the locking in invalidate_range_start and
1449 * invalidate_range_end will be unbalanced.
1451 spin_lock(&kvm->mn_invalidate_lock);
1452 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1453 while (kvm->mn_active_invalidate_count) {
1454 set_current_state(TASK_UNINTERRUPTIBLE);
1455 spin_unlock(&kvm->mn_invalidate_lock);
1457 spin_lock(&kvm->mn_invalidate_lock);
1459 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1460 rcu_assign_pointer(kvm->memslots[as_id], slots);
1461 spin_unlock(&kvm->mn_invalidate_lock);
1464 * Acquired in kvm_set_memslot. Must be released before synchronize
1465 * SRCU below in order to avoid deadlock with another thread
1466 * acquiring the slots_arch_lock in an srcu critical section.
1468 mutex_unlock(&kvm->slots_arch_lock);
1470 synchronize_srcu_expedited(&kvm->srcu);
1473 * Increment the new memslot generation a second time, dropping the
1474 * update in-progress flag and incrementing the generation based on
1475 * the number of address spaces. This provides a unique and easily
1476 * identifiable generation number while the memslots are in flux.
1478 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1481 * Generations must be unique even across address spaces. We do not need
1482 * a global counter for that, instead the generation space is evenly split
1483 * across address spaces. For example, with two address spaces, address
1484 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1485 * use generations 1, 3, 5, ...
1487 gen += KVM_ADDRESS_SPACE_NUM;
1489 kvm_arch_memslots_updated(kvm, gen);
1491 slots->generation = gen;
1494 static int kvm_prepare_memory_region(struct kvm *kvm,
1495 const struct kvm_memory_slot *old,
1496 struct kvm_memory_slot *new,
1497 enum kvm_mr_change change)
1502 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1503 * will be freed on "commit". If logging is enabled in both old and
1504 * new, reuse the existing bitmap. If logging is enabled only in the
1505 * new and KVM isn't using a ring buffer, allocate and initialize a
1508 if (change != KVM_MR_DELETE) {
1509 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1510 new->dirty_bitmap = NULL;
1511 else if (old && old->dirty_bitmap)
1512 new->dirty_bitmap = old->dirty_bitmap;
1513 else if (!kvm->dirty_ring_size) {
1514 r = kvm_alloc_dirty_bitmap(new);
1518 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1519 bitmap_set(new->dirty_bitmap, 0, new->npages);
1523 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1525 /* Free the bitmap on failure if it was allocated above. */
1526 if (r && new && new->dirty_bitmap && old && !old->dirty_bitmap)
1527 kvm_destroy_dirty_bitmap(new);
1532 static void kvm_commit_memory_region(struct kvm *kvm,
1533 struct kvm_memory_slot *old,
1534 const struct kvm_memory_slot *new,
1535 enum kvm_mr_change change)
1538 * Update the total number of memslot pages before calling the arch
1539 * hook so that architectures can consume the result directly.
1541 if (change == KVM_MR_DELETE)
1542 kvm->nr_memslot_pages -= old->npages;
1543 else if (change == KVM_MR_CREATE)
1544 kvm->nr_memslot_pages += new->npages;
1546 kvm_arch_commit_memory_region(kvm, old, new, change);
1550 /* Nothing more to do. */
1553 /* Free the old memslot and all its metadata. */
1554 kvm_free_memslot(kvm, old);
1557 case KVM_MR_FLAGS_ONLY:
1559 * Free the dirty bitmap as needed; the below check encompasses
1560 * both the flags and whether a ring buffer is being used)
1562 if (old->dirty_bitmap && !new->dirty_bitmap)
1563 kvm_destroy_dirty_bitmap(old);
1566 * The final quirk. Free the detached, old slot, but only its
1567 * memory, not any metadata. Metadata, including arch specific
1568 * data, may be reused by @new.
1578 * Activate @new, which must be installed in the inactive slots by the caller,
1579 * by swapping the active slots and then propagating @new to @old once @old is
1580 * unreachable and can be safely modified.
1582 * With NULL @old this simply adds @new to @active (while swapping the sets).
1583 * With NULL @new this simply removes @old from @active and frees it
1584 * (while also swapping the sets).
1586 static void kvm_activate_memslot(struct kvm *kvm,
1587 struct kvm_memory_slot *old,
1588 struct kvm_memory_slot *new)
1590 int as_id = kvm_memslots_get_as_id(old, new);
1592 kvm_swap_active_memslots(kvm, as_id);
1594 /* Propagate the new memslot to the now inactive memslots. */
1595 kvm_replace_memslot(kvm, old, new);
1598 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1599 const struct kvm_memory_slot *src)
1601 dest->base_gfn = src->base_gfn;
1602 dest->npages = src->npages;
1603 dest->dirty_bitmap = src->dirty_bitmap;
1604 dest->arch = src->arch;
1605 dest->userspace_addr = src->userspace_addr;
1606 dest->flags = src->flags;
1608 dest->as_id = src->as_id;
1611 static void kvm_invalidate_memslot(struct kvm *kvm,
1612 struct kvm_memory_slot *old,
1613 struct kvm_memory_slot *invalid_slot)
1616 * Mark the current slot INVALID. As with all memslot modifications,
1617 * this must be done on an unreachable slot to avoid modifying the
1618 * current slot in the active tree.
1620 kvm_copy_memslot(invalid_slot, old);
1621 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1622 kvm_replace_memslot(kvm, old, invalid_slot);
1625 * Activate the slot that is now marked INVALID, but don't propagate
1626 * the slot to the now inactive slots. The slot is either going to be
1627 * deleted or recreated as a new slot.
1629 kvm_swap_active_memslots(kvm, old->as_id);
1632 * From this point no new shadow pages pointing to a deleted, or moved,
1633 * memslot will be created. Validation of sp->gfn happens in:
1634 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1635 * - kvm_is_visible_gfn (mmu_check_root)
1637 kvm_arch_flush_shadow_memslot(kvm, old);
1639 /* Was released by kvm_swap_active_memslots, reacquire. */
1640 mutex_lock(&kvm->slots_arch_lock);
1643 * Copy the arch-specific field of the newly-installed slot back to the
1644 * old slot as the arch data could have changed between releasing
1645 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1646 * above. Writers are required to retrieve memslots *after* acquiring
1647 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1649 old->arch = invalid_slot->arch;
1652 static void kvm_create_memslot(struct kvm *kvm,
1653 struct kvm_memory_slot *new)
1655 /* Add the new memslot to the inactive set and activate. */
1656 kvm_replace_memslot(kvm, NULL, new);
1657 kvm_activate_memslot(kvm, NULL, new);
1660 static void kvm_delete_memslot(struct kvm *kvm,
1661 struct kvm_memory_slot *old,
1662 struct kvm_memory_slot *invalid_slot)
1665 * Remove the old memslot (in the inactive memslots) by passing NULL as
1666 * the "new" slot, and for the invalid version in the active slots.
1668 kvm_replace_memslot(kvm, old, NULL);
1669 kvm_activate_memslot(kvm, invalid_slot, NULL);
1672 static void kvm_move_memslot(struct kvm *kvm,
1673 struct kvm_memory_slot *old,
1674 struct kvm_memory_slot *new,
1675 struct kvm_memory_slot *invalid_slot)
1678 * Replace the old memslot in the inactive slots, and then swap slots
1679 * and replace the current INVALID with the new as well.
1681 kvm_replace_memslot(kvm, old, new);
1682 kvm_activate_memslot(kvm, invalid_slot, new);
1685 static void kvm_update_flags_memslot(struct kvm *kvm,
1686 struct kvm_memory_slot *old,
1687 struct kvm_memory_slot *new)
1690 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1691 * an intermediate step. Instead, the old memslot is simply replaced
1692 * with a new, updated copy in both memslot sets.
1694 kvm_replace_memslot(kvm, old, new);
1695 kvm_activate_memslot(kvm, old, new);
1698 static int kvm_set_memslot(struct kvm *kvm,
1699 struct kvm_memory_slot *old,
1700 struct kvm_memory_slot *new,
1701 enum kvm_mr_change change)
1703 struct kvm_memory_slot *invalid_slot;
1707 * Released in kvm_swap_active_memslots.
1709 * Must be held from before the current memslots are copied until
1710 * after the new memslots are installed with rcu_assign_pointer,
1711 * then released before the synchronize srcu in kvm_swap_active_memslots.
1713 * When modifying memslots outside of the slots_lock, must be held
1714 * before reading the pointer to the current memslots until after all
1715 * changes to those memslots are complete.
1717 * These rules ensure that installing new memslots does not lose
1718 * changes made to the previous memslots.
1720 mutex_lock(&kvm->slots_arch_lock);
1723 * Invalidate the old slot if it's being deleted or moved. This is
1724 * done prior to actually deleting/moving the memslot to allow vCPUs to
1725 * continue running by ensuring there are no mappings or shadow pages
1726 * for the memslot when it is deleted/moved. Without pre-invalidation
1727 * (and without a lock), a window would exist between effecting the
1728 * delete/move and committing the changes in arch code where KVM or a
1729 * guest could access a non-existent memslot.
1731 * Modifications are done on a temporary, unreachable slot. The old
1732 * slot needs to be preserved in case a later step fails and the
1733 * invalidation needs to be reverted.
1735 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1736 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1737 if (!invalid_slot) {
1738 mutex_unlock(&kvm->slots_arch_lock);
1741 kvm_invalidate_memslot(kvm, old, invalid_slot);
1744 r = kvm_prepare_memory_region(kvm, old, new, change);
1747 * For DELETE/MOVE, revert the above INVALID change. No
1748 * modifications required since the original slot was preserved
1749 * in the inactive slots. Changing the active memslots also
1750 * release slots_arch_lock.
1752 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1753 kvm_activate_memslot(kvm, invalid_slot, old);
1754 kfree(invalid_slot);
1756 mutex_unlock(&kvm->slots_arch_lock);
1762 * For DELETE and MOVE, the working slot is now active as the INVALID
1763 * version of the old slot. MOVE is particularly special as it reuses
1764 * the old slot and returns a copy of the old slot (in working_slot).
1765 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1766 * old slot is detached but otherwise preserved.
1768 if (change == KVM_MR_CREATE)
1769 kvm_create_memslot(kvm, new);
1770 else if (change == KVM_MR_DELETE)
1771 kvm_delete_memslot(kvm, old, invalid_slot);
1772 else if (change == KVM_MR_MOVE)
1773 kvm_move_memslot(kvm, old, new, invalid_slot);
1774 else if (change == KVM_MR_FLAGS_ONLY)
1775 kvm_update_flags_memslot(kvm, old, new);
1779 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1780 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1781 kfree(invalid_slot);
1784 * No need to refresh new->arch, changes after dropping slots_arch_lock
1785 * will directly hit the final, active memsot. Architectures are
1786 * responsible for knowing that new->arch may be stale.
1788 kvm_commit_memory_region(kvm, old, new, change);
1793 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1794 gfn_t start, gfn_t end)
1796 struct kvm_memslot_iter iter;
1798 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1799 if (iter.slot->id != id)
1807 * Allocate some memory and give it an address in the guest physical address
1810 * Discontiguous memory is allowed, mostly for framebuffers.
1812 * Must be called holding kvm->slots_lock for write.
1814 int __kvm_set_memory_region(struct kvm *kvm,
1815 const struct kvm_userspace_memory_region *mem)
1817 struct kvm_memory_slot *old, *new;
1818 struct kvm_memslots *slots;
1819 enum kvm_mr_change change;
1820 unsigned long npages;
1825 r = check_memory_region_flags(mem);
1829 as_id = mem->slot >> 16;
1830 id = (u16)mem->slot;
1832 /* General sanity checks */
1833 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1834 (mem->memory_size != (unsigned long)mem->memory_size))
1836 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1838 /* We can read the guest memory with __xxx_user() later on. */
1839 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1840 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1841 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1844 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1846 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1848 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1851 slots = __kvm_memslots(kvm, as_id);
1854 * Note, the old memslot (and the pointer itself!) may be invalidated
1855 * and/or destroyed by kvm_set_memslot().
1857 old = id_to_memslot(slots, id);
1859 if (!mem->memory_size) {
1860 if (!old || !old->npages)
1863 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1866 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1869 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1870 npages = (mem->memory_size >> PAGE_SHIFT);
1872 if (!old || !old->npages) {
1873 change = KVM_MR_CREATE;
1876 * To simplify KVM internals, the total number of pages across
1877 * all memslots must fit in an unsigned long.
1879 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1881 } else { /* Modify an existing slot. */
1882 if ((mem->userspace_addr != old->userspace_addr) ||
1883 (npages != old->npages) ||
1884 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1887 if (base_gfn != old->base_gfn)
1888 change = KVM_MR_MOVE;
1889 else if (mem->flags != old->flags)
1890 change = KVM_MR_FLAGS_ONLY;
1891 else /* Nothing to change. */
1895 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1896 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1899 /* Allocate a slot that will persist in the memslot. */
1900 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1906 new->base_gfn = base_gfn;
1907 new->npages = npages;
1908 new->flags = mem->flags;
1909 new->userspace_addr = mem->userspace_addr;
1911 r = kvm_set_memslot(kvm, old, new, change);
1916 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1918 int kvm_set_memory_region(struct kvm *kvm,
1919 const struct kvm_userspace_memory_region *mem)
1923 mutex_lock(&kvm->slots_lock);
1924 r = __kvm_set_memory_region(kvm, mem);
1925 mutex_unlock(&kvm->slots_lock);
1928 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1930 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1931 struct kvm_userspace_memory_region *mem)
1933 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1936 return kvm_set_memory_region(kvm, mem);
1939 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1941 * kvm_get_dirty_log - get a snapshot of dirty pages
1942 * @kvm: pointer to kvm instance
1943 * @log: slot id and address to which we copy the log
1944 * @is_dirty: set to '1' if any dirty pages were found
1945 * @memslot: set to the associated memslot, always valid on success
1947 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1948 int *is_dirty, struct kvm_memory_slot **memslot)
1950 struct kvm_memslots *slots;
1953 unsigned long any = 0;
1955 /* Dirty ring tracking is exclusive to dirty log tracking */
1956 if (kvm->dirty_ring_size)
1962 as_id = log->slot >> 16;
1963 id = (u16)log->slot;
1964 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1967 slots = __kvm_memslots(kvm, as_id);
1968 *memslot = id_to_memslot(slots, id);
1969 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1972 kvm_arch_sync_dirty_log(kvm, *memslot);
1974 n = kvm_dirty_bitmap_bytes(*memslot);
1976 for (i = 0; !any && i < n/sizeof(long); ++i)
1977 any = (*memslot)->dirty_bitmap[i];
1979 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1986 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1988 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1990 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1991 * and reenable dirty page tracking for the corresponding pages.
1992 * @kvm: pointer to kvm instance
1993 * @log: slot id and address to which we copy the log
1995 * We need to keep it in mind that VCPU threads can write to the bitmap
1996 * concurrently. So, to avoid losing track of dirty pages we keep the
1999 * 1. Take a snapshot of the bit and clear it if needed.
2000 * 2. Write protect the corresponding page.
2001 * 3. Copy the snapshot to the userspace.
2002 * 4. Upon return caller flushes TLB's if needed.
2004 * Between 2 and 4, the guest may write to the page using the remaining TLB
2005 * entry. This is not a problem because the page is reported dirty using
2006 * the snapshot taken before and step 4 ensures that writes done after
2007 * exiting to userspace will be logged for the next call.
2010 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2012 struct kvm_memslots *slots;
2013 struct kvm_memory_slot *memslot;
2016 unsigned long *dirty_bitmap;
2017 unsigned long *dirty_bitmap_buffer;
2020 /* Dirty ring tracking is exclusive to dirty log tracking */
2021 if (kvm->dirty_ring_size)
2024 as_id = log->slot >> 16;
2025 id = (u16)log->slot;
2026 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2029 slots = __kvm_memslots(kvm, as_id);
2030 memslot = id_to_memslot(slots, id);
2031 if (!memslot || !memslot->dirty_bitmap)
2034 dirty_bitmap = memslot->dirty_bitmap;
2036 kvm_arch_sync_dirty_log(kvm, memslot);
2038 n = kvm_dirty_bitmap_bytes(memslot);
2040 if (kvm->manual_dirty_log_protect) {
2042 * Unlike kvm_get_dirty_log, we always return false in *flush,
2043 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2044 * is some code duplication between this function and
2045 * kvm_get_dirty_log, but hopefully all architecture
2046 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2047 * can be eliminated.
2049 dirty_bitmap_buffer = dirty_bitmap;
2051 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2052 memset(dirty_bitmap_buffer, 0, n);
2055 for (i = 0; i < n / sizeof(long); i++) {
2059 if (!dirty_bitmap[i])
2063 mask = xchg(&dirty_bitmap[i], 0);
2064 dirty_bitmap_buffer[i] = mask;
2066 offset = i * BITS_PER_LONG;
2067 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2070 KVM_MMU_UNLOCK(kvm);
2074 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2076 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2083 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2084 * @kvm: kvm instance
2085 * @log: slot id and address to which we copy the log
2087 * Steps 1-4 below provide general overview of dirty page logging. See
2088 * kvm_get_dirty_log_protect() function description for additional details.
2090 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2091 * always flush the TLB (step 4) even if previous step failed and the dirty
2092 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2093 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2094 * writes will be marked dirty for next log read.
2096 * 1. Take a snapshot of the bit and clear it if needed.
2097 * 2. Write protect the corresponding page.
2098 * 3. Copy the snapshot to the userspace.
2099 * 4. Flush TLB's if needed.
2101 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2102 struct kvm_dirty_log *log)
2106 mutex_lock(&kvm->slots_lock);
2108 r = kvm_get_dirty_log_protect(kvm, log);
2110 mutex_unlock(&kvm->slots_lock);
2115 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2116 * and reenable dirty page tracking for the corresponding pages.
2117 * @kvm: pointer to kvm instance
2118 * @log: slot id and address from which to fetch the bitmap of dirty pages
2120 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2121 struct kvm_clear_dirty_log *log)
2123 struct kvm_memslots *slots;
2124 struct kvm_memory_slot *memslot;
2128 unsigned long *dirty_bitmap;
2129 unsigned long *dirty_bitmap_buffer;
2132 /* Dirty ring tracking is exclusive to dirty log tracking */
2133 if (kvm->dirty_ring_size)
2136 as_id = log->slot >> 16;
2137 id = (u16)log->slot;
2138 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2141 if (log->first_page & 63)
2144 slots = __kvm_memslots(kvm, as_id);
2145 memslot = id_to_memslot(slots, id);
2146 if (!memslot || !memslot->dirty_bitmap)
2149 dirty_bitmap = memslot->dirty_bitmap;
2151 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2153 if (log->first_page > memslot->npages ||
2154 log->num_pages > memslot->npages - log->first_page ||
2155 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2158 kvm_arch_sync_dirty_log(kvm, memslot);
2161 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2162 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2166 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2167 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2168 i++, offset += BITS_PER_LONG) {
2169 unsigned long mask = *dirty_bitmap_buffer++;
2170 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2174 mask &= atomic_long_fetch_andnot(mask, p);
2177 * mask contains the bits that really have been cleared. This
2178 * never includes any bits beyond the length of the memslot (if
2179 * the length is not aligned to 64 pages), therefore it is not
2180 * a problem if userspace sets them in log->dirty_bitmap.
2184 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2188 KVM_MMU_UNLOCK(kvm);
2191 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2196 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2197 struct kvm_clear_dirty_log *log)
2201 mutex_lock(&kvm->slots_lock);
2203 r = kvm_clear_dirty_log_protect(kvm, log);
2205 mutex_unlock(&kvm->slots_lock);
2208 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2210 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2212 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2214 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2216 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2218 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2219 u64 gen = slots->generation;
2220 struct kvm_memory_slot *slot;
2223 * This also protects against using a memslot from a different address space,
2224 * since different address spaces have different generation numbers.
2226 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2227 vcpu->last_used_slot = NULL;
2228 vcpu->last_used_slot_gen = gen;
2231 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2236 * Fall back to searching all memslots. We purposely use
2237 * search_memslots() instead of __gfn_to_memslot() to avoid
2238 * thrashing the VM-wide last_used_slot in kvm_memslots.
2240 slot = search_memslots(slots, gfn, false);
2242 vcpu->last_used_slot = slot;
2248 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
2250 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2252 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2254 return kvm_is_visible_memslot(memslot);
2256 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2258 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2260 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2262 return kvm_is_visible_memslot(memslot);
2264 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2266 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2268 struct vm_area_struct *vma;
2269 unsigned long addr, size;
2273 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2274 if (kvm_is_error_hva(addr))
2277 mmap_read_lock(current->mm);
2278 vma = find_vma(current->mm, addr);
2282 size = vma_kernel_pagesize(vma);
2285 mmap_read_unlock(current->mm);
2290 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2292 return slot->flags & KVM_MEM_READONLY;
2295 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2296 gfn_t *nr_pages, bool write)
2298 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2299 return KVM_HVA_ERR_BAD;
2301 if (memslot_is_readonly(slot) && write)
2302 return KVM_HVA_ERR_RO_BAD;
2305 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2307 return __gfn_to_hva_memslot(slot, gfn);
2310 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2313 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2316 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2319 return gfn_to_hva_many(slot, gfn, NULL);
2321 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2323 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2325 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2327 EXPORT_SYMBOL_GPL(gfn_to_hva);
2329 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2331 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2333 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2336 * Return the hva of a @gfn and the R/W attribute if possible.
2338 * @slot: the kvm_memory_slot which contains @gfn
2339 * @gfn: the gfn to be translated
2340 * @writable: used to return the read/write attribute of the @slot if the hva
2341 * is valid and @writable is not NULL
2343 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2344 gfn_t gfn, bool *writable)
2346 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2348 if (!kvm_is_error_hva(hva) && writable)
2349 *writable = !memslot_is_readonly(slot);
2354 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2356 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2358 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2361 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2363 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2365 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2368 static inline int check_user_page_hwpoison(unsigned long addr)
2370 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2372 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2373 return rc == -EHWPOISON;
2377 * The fast path to get the writable pfn which will be stored in @pfn,
2378 * true indicates success, otherwise false is returned. It's also the
2379 * only part that runs if we can in atomic context.
2381 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2382 bool *writable, kvm_pfn_t *pfn)
2384 struct page *page[1];
2387 * Fast pin a writable pfn only if it is a write fault request
2388 * or the caller allows to map a writable pfn for a read fault
2391 if (!(write_fault || writable))
2394 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2395 *pfn = page_to_pfn(page[0]);
2406 * The slow path to get the pfn of the specified host virtual address,
2407 * 1 indicates success, -errno is returned if error is detected.
2409 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2410 bool *writable, kvm_pfn_t *pfn)
2412 unsigned int flags = FOLL_HWPOISON;
2419 *writable = write_fault;
2422 flags |= FOLL_WRITE;
2424 flags |= FOLL_NOWAIT;
2426 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2430 /* map read fault as writable if possible */
2431 if (unlikely(!write_fault) && writable) {
2434 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2440 *pfn = page_to_pfn(page);
2444 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2446 if (unlikely(!(vma->vm_flags & VM_READ)))
2449 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2455 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2457 if (kvm_is_reserved_pfn(pfn))
2459 return get_page_unless_zero(pfn_to_page(pfn));
2462 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2463 unsigned long addr, bool *async,
2464 bool write_fault, bool *writable,
2472 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2475 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2476 * not call the fault handler, so do it here.
2478 bool unlocked = false;
2479 r = fixup_user_fault(current->mm, addr,
2480 (write_fault ? FAULT_FLAG_WRITE : 0),
2487 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2492 if (write_fault && !pte_write(*ptep)) {
2493 pfn = KVM_PFN_ERR_RO_FAULT;
2498 *writable = pte_write(*ptep);
2499 pfn = pte_pfn(*ptep);
2502 * Get a reference here because callers of *hva_to_pfn* and
2503 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2504 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2505 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2506 * simply do nothing for reserved pfns.
2508 * Whoever called remap_pfn_range is also going to call e.g.
2509 * unmap_mapping_range before the underlying pages are freed,
2510 * causing a call to our MMU notifier.
2512 * Certain IO or PFNMAP mappings can be backed with valid
2513 * struct pages, but be allocated without refcounting e.g.,
2514 * tail pages of non-compound higher order allocations, which
2515 * would then underflow the refcount when the caller does the
2516 * required put_page. Don't allow those pages here.
2518 if (!kvm_try_get_pfn(pfn))
2522 pte_unmap_unlock(ptep, ptl);
2529 * Pin guest page in memory and return its pfn.
2530 * @addr: host virtual address which maps memory to the guest
2531 * @atomic: whether this function can sleep
2532 * @async: whether this function need to wait IO complete if the
2533 * host page is not in the memory
2534 * @write_fault: whether we should get a writable host page
2535 * @writable: whether it allows to map a writable host page for !@write_fault
2537 * The function will map a writable host page for these two cases:
2538 * 1): @write_fault = true
2539 * 2): @write_fault = false && @writable, @writable will tell the caller
2540 * whether the mapping is writable.
2542 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2543 bool write_fault, bool *writable)
2545 struct vm_area_struct *vma;
2549 /* we can do it either atomically or asynchronously, not both */
2550 BUG_ON(atomic && async);
2552 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2556 return KVM_PFN_ERR_FAULT;
2558 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2562 mmap_read_lock(current->mm);
2563 if (npages == -EHWPOISON ||
2564 (!async && check_user_page_hwpoison(addr))) {
2565 pfn = KVM_PFN_ERR_HWPOISON;
2570 vma = vma_lookup(current->mm, addr);
2573 pfn = KVM_PFN_ERR_FAULT;
2574 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2575 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2579 pfn = KVM_PFN_ERR_FAULT;
2581 if (async && vma_is_valid(vma, write_fault))
2583 pfn = KVM_PFN_ERR_FAULT;
2586 mmap_read_unlock(current->mm);
2590 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2591 bool atomic, bool *async, bool write_fault,
2592 bool *writable, hva_t *hva)
2594 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2599 if (addr == KVM_HVA_ERR_RO_BAD) {
2602 return KVM_PFN_ERR_RO_FAULT;
2605 if (kvm_is_error_hva(addr)) {
2608 return KVM_PFN_NOSLOT;
2611 /* Do not map writable pfn in the readonly memslot. */
2612 if (writable && memslot_is_readonly(slot)) {
2617 return hva_to_pfn(addr, atomic, async, write_fault,
2620 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2622 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2625 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2626 write_fault, writable, NULL);
2628 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2630 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2632 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2634 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2636 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2638 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2640 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2642 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2644 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2646 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2648 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2650 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2652 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2654 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2656 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2658 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2660 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2661 struct page **pages, int nr_pages)
2666 addr = gfn_to_hva_many(slot, gfn, &entry);
2667 if (kvm_is_error_hva(addr))
2670 if (entry < nr_pages)
2673 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2675 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2677 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2679 if (is_error_noslot_pfn(pfn))
2680 return KVM_ERR_PTR_BAD_PAGE;
2682 if (kvm_is_reserved_pfn(pfn)) {
2684 return KVM_ERR_PTR_BAD_PAGE;
2687 return pfn_to_page(pfn);
2690 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2694 pfn = gfn_to_pfn(kvm, gfn);
2696 return kvm_pfn_to_page(pfn);
2698 EXPORT_SYMBOL_GPL(gfn_to_page);
2700 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2706 kvm_release_pfn_dirty(pfn);
2708 kvm_release_pfn_clean(pfn);
2711 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2715 struct page *page = KVM_UNMAPPED_PAGE;
2720 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2721 if (is_error_noslot_pfn(pfn))
2724 if (pfn_valid(pfn)) {
2725 page = pfn_to_page(pfn);
2727 #ifdef CONFIG_HAS_IOMEM
2729 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2743 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2745 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2753 if (map->page != KVM_UNMAPPED_PAGE)
2755 #ifdef CONFIG_HAS_IOMEM
2761 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2763 kvm_release_pfn(map->pfn, dirty);
2768 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2770 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2774 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2776 return kvm_pfn_to_page(pfn);
2778 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2780 void kvm_release_page_clean(struct page *page)
2782 WARN_ON(is_error_page(page));
2784 kvm_release_pfn_clean(page_to_pfn(page));
2786 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2788 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2790 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2791 put_page(pfn_to_page(pfn));
2793 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2795 void kvm_release_page_dirty(struct page *page)
2797 WARN_ON(is_error_page(page));
2799 kvm_release_pfn_dirty(page_to_pfn(page));
2801 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2803 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2805 kvm_set_pfn_dirty(pfn);
2806 kvm_release_pfn_clean(pfn);
2808 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2810 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2812 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2813 SetPageDirty(pfn_to_page(pfn));
2815 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2817 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2819 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2820 mark_page_accessed(pfn_to_page(pfn));
2822 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2824 static int next_segment(unsigned long len, int offset)
2826 if (len > PAGE_SIZE - offset)
2827 return PAGE_SIZE - offset;
2832 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2833 void *data, int offset, int len)
2838 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2839 if (kvm_is_error_hva(addr))
2841 r = __copy_from_user(data, (void __user *)addr + offset, len);
2847 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2850 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2852 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2854 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2856 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2857 int offset, int len)
2859 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2861 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2863 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2865 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2867 gfn_t gfn = gpa >> PAGE_SHIFT;
2869 int offset = offset_in_page(gpa);
2872 while ((seg = next_segment(len, offset)) != 0) {
2873 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2883 EXPORT_SYMBOL_GPL(kvm_read_guest);
2885 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2887 gfn_t gfn = gpa >> PAGE_SHIFT;
2889 int offset = offset_in_page(gpa);
2892 while ((seg = next_segment(len, offset)) != 0) {
2893 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2903 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2905 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2906 void *data, int offset, unsigned long len)
2911 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2912 if (kvm_is_error_hva(addr))
2914 pagefault_disable();
2915 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2922 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2923 void *data, unsigned long len)
2925 gfn_t gfn = gpa >> PAGE_SHIFT;
2926 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2927 int offset = offset_in_page(gpa);
2929 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2931 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2933 static int __kvm_write_guest_page(struct kvm *kvm,
2934 struct kvm_memory_slot *memslot, gfn_t gfn,
2935 const void *data, int offset, int len)
2940 addr = gfn_to_hva_memslot(memslot, gfn);
2941 if (kvm_is_error_hva(addr))
2943 r = __copy_to_user((void __user *)addr + offset, data, len);
2946 mark_page_dirty_in_slot(kvm, memslot, gfn);
2950 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2951 const void *data, int offset, int len)
2953 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2955 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2957 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2959 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2960 const void *data, int offset, int len)
2962 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2964 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2966 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2968 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2971 gfn_t gfn = gpa >> PAGE_SHIFT;
2973 int offset = offset_in_page(gpa);
2976 while ((seg = next_segment(len, offset)) != 0) {
2977 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2987 EXPORT_SYMBOL_GPL(kvm_write_guest);
2989 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2992 gfn_t gfn = gpa >> PAGE_SHIFT;
2994 int offset = offset_in_page(gpa);
2997 while ((seg = next_segment(len, offset)) != 0) {
2998 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3008 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3010 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3011 struct gfn_to_hva_cache *ghc,
3012 gpa_t gpa, unsigned long len)
3014 int offset = offset_in_page(gpa);
3015 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3016 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3017 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3018 gfn_t nr_pages_avail;
3020 /* Update ghc->generation before performing any error checks. */
3021 ghc->generation = slots->generation;
3023 if (start_gfn > end_gfn) {
3024 ghc->hva = KVM_HVA_ERR_BAD;
3029 * If the requested region crosses two memslots, we still
3030 * verify that the entire region is valid here.
3032 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3033 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3034 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3036 if (kvm_is_error_hva(ghc->hva))
3040 /* Use the slow path for cross page reads and writes. */
3041 if (nr_pages_needed == 1)
3044 ghc->memslot = NULL;
3051 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3052 gpa_t gpa, unsigned long len)
3054 struct kvm_memslots *slots = kvm_memslots(kvm);
3055 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3057 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3059 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3060 void *data, unsigned int offset,
3063 struct kvm_memslots *slots = kvm_memslots(kvm);
3065 gpa_t gpa = ghc->gpa + offset;
3067 if (WARN_ON_ONCE(len + offset > ghc->len))
3070 if (slots->generation != ghc->generation) {
3071 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3075 if (kvm_is_error_hva(ghc->hva))
3078 if (unlikely(!ghc->memslot))
3079 return kvm_write_guest(kvm, gpa, data, len);
3081 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3084 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3088 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3090 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3091 void *data, unsigned long len)
3093 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3095 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3097 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3098 void *data, unsigned int offset,
3101 struct kvm_memslots *slots = kvm_memslots(kvm);
3103 gpa_t gpa = ghc->gpa + offset;
3105 if (WARN_ON_ONCE(len + offset > ghc->len))
3108 if (slots->generation != ghc->generation) {
3109 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3113 if (kvm_is_error_hva(ghc->hva))
3116 if (unlikely(!ghc->memslot))
3117 return kvm_read_guest(kvm, gpa, data, len);
3119 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3125 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3127 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3128 void *data, unsigned long len)
3130 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3132 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3134 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3136 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3137 gfn_t gfn = gpa >> PAGE_SHIFT;
3139 int offset = offset_in_page(gpa);
3142 while ((seg = next_segment(len, offset)) != 0) {
3143 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3152 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3154 void mark_page_dirty_in_slot(struct kvm *kvm,
3155 const struct kvm_memory_slot *memslot,
3158 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3160 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3163 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3164 unsigned long rel_gfn = gfn - memslot->base_gfn;
3165 u32 slot = (memslot->as_id << 16) | memslot->id;
3167 if (kvm->dirty_ring_size)
3168 kvm_dirty_ring_push(&vcpu->dirty_ring,
3171 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3174 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3176 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3178 struct kvm_memory_slot *memslot;
3180 memslot = gfn_to_memslot(kvm, gfn);
3181 mark_page_dirty_in_slot(kvm, memslot, gfn);
3183 EXPORT_SYMBOL_GPL(mark_page_dirty);
3185 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3187 struct kvm_memory_slot *memslot;
3189 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3190 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3192 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3194 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3196 if (!vcpu->sigset_active)
3200 * This does a lockless modification of ->real_blocked, which is fine
3201 * because, only current can change ->real_blocked and all readers of
3202 * ->real_blocked don't care as long ->real_blocked is always a subset
3205 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3208 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3210 if (!vcpu->sigset_active)
3213 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3214 sigemptyset(¤t->real_blocked);
3217 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3219 unsigned int old, val, grow, grow_start;
3221 old = val = vcpu->halt_poll_ns;
3222 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3223 grow = READ_ONCE(halt_poll_ns_grow);
3228 if (val < grow_start)
3231 if (val > vcpu->kvm->max_halt_poll_ns)
3232 val = vcpu->kvm->max_halt_poll_ns;
3234 vcpu->halt_poll_ns = val;
3236 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3239 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3241 unsigned int old, val, shrink, grow_start;
3243 old = val = vcpu->halt_poll_ns;
3244 shrink = READ_ONCE(halt_poll_ns_shrink);
3245 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3251 if (val < grow_start)
3254 vcpu->halt_poll_ns = val;
3255 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3258 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3261 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3263 if (kvm_arch_vcpu_runnable(vcpu)) {
3264 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3267 if (kvm_cpu_has_pending_timer(vcpu))
3269 if (signal_pending(current))
3271 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3276 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3281 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3282 * pending. This is mostly used when halting a vCPU, but may also be used
3283 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3285 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3287 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3288 bool waited = false;
3290 vcpu->stat.generic.blocking = 1;
3292 kvm_arch_vcpu_blocking(vcpu);
3294 prepare_to_rcuwait(wait);
3296 set_current_state(TASK_INTERRUPTIBLE);
3298 if (kvm_vcpu_check_block(vcpu) < 0)
3304 finish_rcuwait(wait);
3306 kvm_arch_vcpu_unblocking(vcpu);
3308 vcpu->stat.generic.blocking = 0;
3313 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3314 ktime_t end, bool success)
3316 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3317 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3319 ++vcpu->stat.generic.halt_attempted_poll;
3322 ++vcpu->stat.generic.halt_successful_poll;
3324 if (!vcpu_valid_wakeup(vcpu))
3325 ++vcpu->stat.generic.halt_poll_invalid;
3327 stats->halt_poll_success_ns += poll_ns;
3328 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3330 stats->halt_poll_fail_ns += poll_ns;
3331 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3336 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3337 * polling is enabled, busy wait for a short time before blocking to avoid the
3338 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3341 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3343 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3344 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3345 ktime_t start, cur, poll_end;
3346 bool waited = false;
3349 start = cur = poll_end = ktime_get();
3351 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3355 * This sets KVM_REQ_UNHALT if an interrupt
3358 if (kvm_vcpu_check_block(vcpu) < 0)
3361 poll_end = cur = ktime_get();
3362 } while (kvm_vcpu_can_poll(cur, stop));
3365 waited = kvm_vcpu_block(vcpu);
3369 vcpu->stat.generic.halt_wait_ns +=
3370 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3371 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3372 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3375 /* The total time the vCPU was "halted", including polling time. */
3376 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3379 * Note, halt-polling is considered successful so long as the vCPU was
3380 * never actually scheduled out, i.e. even if the wake event arrived
3381 * after of the halt-polling loop itself, but before the full wait.
3384 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3386 if (halt_poll_allowed) {
3387 if (!vcpu_valid_wakeup(vcpu)) {
3388 shrink_halt_poll_ns(vcpu);
3389 } else if (vcpu->kvm->max_halt_poll_ns) {
3390 if (halt_ns <= vcpu->halt_poll_ns)
3392 /* we had a long block, shrink polling */
3393 else if (vcpu->halt_poll_ns &&
3394 halt_ns > vcpu->kvm->max_halt_poll_ns)
3395 shrink_halt_poll_ns(vcpu);
3396 /* we had a short halt and our poll time is too small */
3397 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3398 halt_ns < vcpu->kvm->max_halt_poll_ns)
3399 grow_halt_poll_ns(vcpu);
3401 vcpu->halt_poll_ns = 0;
3405 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3407 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3409 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3411 if (__kvm_vcpu_wake_up(vcpu)) {
3412 WRITE_ONCE(vcpu->ready, true);
3413 ++vcpu->stat.generic.halt_wakeup;
3419 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3423 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3425 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3429 if (kvm_vcpu_wake_up(vcpu))
3434 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3435 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3436 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3437 * within the vCPU thread itself.
3439 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3440 if (vcpu->mode == IN_GUEST_MODE)
3441 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3446 * Note, the vCPU could get migrated to a different pCPU at any point
3447 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3448 * IPI to the previous pCPU. But, that's ok because the purpose of the
3449 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3450 * vCPU also requires it to leave IN_GUEST_MODE.
3452 if (kvm_arch_vcpu_should_kick(vcpu)) {
3453 cpu = READ_ONCE(vcpu->cpu);
3454 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3455 smp_send_reschedule(cpu);
3460 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3461 #endif /* !CONFIG_S390 */
3463 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3466 struct task_struct *task = NULL;
3470 pid = rcu_dereference(target->pid);
3472 task = get_pid_task(pid, PIDTYPE_PID);
3476 ret = yield_to(task, 1);
3477 put_task_struct(task);
3481 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3484 * Helper that checks whether a VCPU is eligible for directed yield.
3485 * Most eligible candidate to yield is decided by following heuristics:
3487 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3488 * (preempted lock holder), indicated by @in_spin_loop.
3489 * Set at the beginning and cleared at the end of interception/PLE handler.
3491 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3492 * chance last time (mostly it has become eligible now since we have probably
3493 * yielded to lockholder in last iteration. This is done by toggling
3494 * @dy_eligible each time a VCPU checked for eligibility.)
3496 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3497 * to preempted lock-holder could result in wrong VCPU selection and CPU
3498 * burning. Giving priority for a potential lock-holder increases lock
3501 * Since algorithm is based on heuristics, accessing another VCPU data without
3502 * locking does not harm. It may result in trying to yield to same VCPU, fail
3503 * and continue with next VCPU and so on.
3505 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3507 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3510 eligible = !vcpu->spin_loop.in_spin_loop ||
3511 vcpu->spin_loop.dy_eligible;
3513 if (vcpu->spin_loop.in_spin_loop)
3514 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3523 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3524 * a vcpu_load/vcpu_put pair. However, for most architectures
3525 * kvm_arch_vcpu_runnable does not require vcpu_load.
3527 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3529 return kvm_arch_vcpu_runnable(vcpu);
3532 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3534 if (kvm_arch_dy_runnable(vcpu))
3537 #ifdef CONFIG_KVM_ASYNC_PF
3538 if (!list_empty_careful(&vcpu->async_pf.done))
3545 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3550 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3552 struct kvm *kvm = me->kvm;
3553 struct kvm_vcpu *vcpu;
3554 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3560 kvm_vcpu_set_in_spin_loop(me, true);
3562 * We boost the priority of a VCPU that is runnable but not
3563 * currently running, because it got preempted by something
3564 * else and called schedule in __vcpu_run. Hopefully that
3565 * VCPU is holding the lock that we need and will release it.
3566 * We approximate round-robin by starting at the last boosted VCPU.
3568 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3569 kvm_for_each_vcpu(i, vcpu, kvm) {
3570 if (!pass && i <= last_boosted_vcpu) {
3571 i = last_boosted_vcpu;
3573 } else if (pass && i > last_boosted_vcpu)
3575 if (!READ_ONCE(vcpu->ready))
3579 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3581 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3582 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3583 !kvm_arch_vcpu_in_kernel(vcpu))
3585 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3588 yielded = kvm_vcpu_yield_to(vcpu);
3590 kvm->last_boosted_vcpu = i;
3592 } else if (yielded < 0) {
3599 kvm_vcpu_set_in_spin_loop(me, false);
3601 /* Ensure vcpu is not eligible during next spinloop */
3602 kvm_vcpu_set_dy_eligible(me, false);
3604 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3606 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3608 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3609 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3610 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3611 kvm->dirty_ring_size / PAGE_SIZE);
3617 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3619 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3622 if (vmf->pgoff == 0)
3623 page = virt_to_page(vcpu->run);
3625 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3626 page = virt_to_page(vcpu->arch.pio_data);
3628 #ifdef CONFIG_KVM_MMIO
3629 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3630 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3632 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3633 page = kvm_dirty_ring_get_page(
3635 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3637 return kvm_arch_vcpu_fault(vcpu, vmf);
3643 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3644 .fault = kvm_vcpu_fault,
3647 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3649 struct kvm_vcpu *vcpu = file->private_data;
3650 unsigned long pages = vma_pages(vma);
3652 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3653 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3654 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3657 vma->vm_ops = &kvm_vcpu_vm_ops;
3661 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3663 struct kvm_vcpu *vcpu = filp->private_data;
3665 kvm_put_kvm(vcpu->kvm);
3669 static struct file_operations kvm_vcpu_fops = {
3670 .release = kvm_vcpu_release,
3671 .unlocked_ioctl = kvm_vcpu_ioctl,
3672 .mmap = kvm_vcpu_mmap,
3673 .llseek = noop_llseek,
3674 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3678 * Allocates an inode for the vcpu.
3680 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3682 char name[8 + 1 + ITOA_MAX_LEN + 1];
3684 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3685 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3688 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3690 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3691 struct dentry *debugfs_dentry;
3692 char dir_name[ITOA_MAX_LEN * 2];
3694 if (!debugfs_initialized())
3697 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3698 debugfs_dentry = debugfs_create_dir(dir_name,
3699 vcpu->kvm->debugfs_dentry);
3701 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3706 * Creates some virtual cpus. Good luck creating more than one.
3708 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3711 struct kvm_vcpu *vcpu;
3714 if (id >= KVM_MAX_VCPU_IDS)
3717 mutex_lock(&kvm->lock);
3718 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3719 mutex_unlock(&kvm->lock);
3723 kvm->created_vcpus++;
3724 mutex_unlock(&kvm->lock);
3726 r = kvm_arch_vcpu_precreate(kvm, id);
3728 goto vcpu_decrement;
3730 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3733 goto vcpu_decrement;
3736 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3737 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3742 vcpu->run = page_address(page);
3744 kvm_vcpu_init(vcpu, kvm, id);
3746 r = kvm_arch_vcpu_create(vcpu);
3748 goto vcpu_free_run_page;
3750 if (kvm->dirty_ring_size) {
3751 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3752 id, kvm->dirty_ring_size);
3754 goto arch_vcpu_destroy;
3757 mutex_lock(&kvm->lock);
3758 if (kvm_get_vcpu_by_id(kvm, id)) {
3760 goto unlock_vcpu_destroy;
3763 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3764 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3765 BUG_ON(r == -EBUSY);
3767 goto unlock_vcpu_destroy;
3769 /* Fill the stats id string for the vcpu */
3770 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3771 task_pid_nr(current), id);
3773 /* Now it's all set up, let userspace reach it */
3775 r = create_vcpu_fd(vcpu);
3777 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3778 kvm_put_kvm_no_destroy(kvm);
3779 goto unlock_vcpu_destroy;
3783 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3784 * pointer before kvm->online_vcpu's incremented value.
3787 atomic_inc(&kvm->online_vcpus);
3789 mutex_unlock(&kvm->lock);
3790 kvm_arch_vcpu_postcreate(vcpu);
3791 kvm_create_vcpu_debugfs(vcpu);
3794 unlock_vcpu_destroy:
3795 mutex_unlock(&kvm->lock);
3796 kvm_dirty_ring_free(&vcpu->dirty_ring);
3798 kvm_arch_vcpu_destroy(vcpu);
3800 free_page((unsigned long)vcpu->run);
3802 kmem_cache_free(kvm_vcpu_cache, vcpu);
3804 mutex_lock(&kvm->lock);
3805 kvm->created_vcpus--;
3806 mutex_unlock(&kvm->lock);
3810 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3813 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3814 vcpu->sigset_active = 1;
3815 vcpu->sigset = *sigset;
3817 vcpu->sigset_active = 0;
3821 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3822 size_t size, loff_t *offset)
3824 struct kvm_vcpu *vcpu = file->private_data;
3826 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3827 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3828 sizeof(vcpu->stat), user_buffer, size, offset);
3831 static const struct file_operations kvm_vcpu_stats_fops = {
3832 .read = kvm_vcpu_stats_read,
3833 .llseek = noop_llseek,
3836 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3840 char name[15 + ITOA_MAX_LEN + 1];
3842 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3844 fd = get_unused_fd_flags(O_CLOEXEC);
3848 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3851 return PTR_ERR(file);
3853 file->f_mode |= FMODE_PREAD;
3854 fd_install(fd, file);
3859 static long kvm_vcpu_ioctl(struct file *filp,
3860 unsigned int ioctl, unsigned long arg)
3862 struct kvm_vcpu *vcpu = filp->private_data;
3863 void __user *argp = (void __user *)arg;
3865 struct kvm_fpu *fpu = NULL;
3866 struct kvm_sregs *kvm_sregs = NULL;
3868 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3871 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3875 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3876 * execution; mutex_lock() would break them.
3878 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3879 if (r != -ENOIOCTLCMD)
3882 if (mutex_lock_killable(&vcpu->mutex))
3890 oldpid = rcu_access_pointer(vcpu->pid);
3891 if (unlikely(oldpid != task_pid(current))) {
3892 /* The thread running this VCPU changed. */
3895 r = kvm_arch_vcpu_run_pid_change(vcpu);
3899 newpid = get_task_pid(current, PIDTYPE_PID);
3900 rcu_assign_pointer(vcpu->pid, newpid);
3905 r = kvm_arch_vcpu_ioctl_run(vcpu);
3906 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3909 case KVM_GET_REGS: {
3910 struct kvm_regs *kvm_regs;
3913 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3916 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3920 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3927 case KVM_SET_REGS: {
3928 struct kvm_regs *kvm_regs;
3930 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3931 if (IS_ERR(kvm_regs)) {
3932 r = PTR_ERR(kvm_regs);
3935 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3939 case KVM_GET_SREGS: {
3940 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3941 GFP_KERNEL_ACCOUNT);
3945 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3949 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3954 case KVM_SET_SREGS: {
3955 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3956 if (IS_ERR(kvm_sregs)) {
3957 r = PTR_ERR(kvm_sregs);
3961 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3964 case KVM_GET_MP_STATE: {
3965 struct kvm_mp_state mp_state;
3967 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3971 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3976 case KVM_SET_MP_STATE: {
3977 struct kvm_mp_state mp_state;
3980 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3982 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3985 case KVM_TRANSLATE: {
3986 struct kvm_translation tr;
3989 if (copy_from_user(&tr, argp, sizeof(tr)))
3991 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3995 if (copy_to_user(argp, &tr, sizeof(tr)))
4000 case KVM_SET_GUEST_DEBUG: {
4001 struct kvm_guest_debug dbg;
4004 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4006 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4009 case KVM_SET_SIGNAL_MASK: {
4010 struct kvm_signal_mask __user *sigmask_arg = argp;
4011 struct kvm_signal_mask kvm_sigmask;
4012 sigset_t sigset, *p;
4017 if (copy_from_user(&kvm_sigmask, argp,
4018 sizeof(kvm_sigmask)))
4021 if (kvm_sigmask.len != sizeof(sigset))
4024 if (copy_from_user(&sigset, sigmask_arg->sigset,
4029 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4033 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4037 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4041 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4047 fpu = memdup_user(argp, sizeof(*fpu));
4053 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4056 case KVM_GET_STATS_FD: {
4057 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4061 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4064 mutex_unlock(&vcpu->mutex);
4070 #ifdef CONFIG_KVM_COMPAT
4071 static long kvm_vcpu_compat_ioctl(struct file *filp,
4072 unsigned int ioctl, unsigned long arg)
4074 struct kvm_vcpu *vcpu = filp->private_data;
4075 void __user *argp = compat_ptr(arg);
4078 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4082 case KVM_SET_SIGNAL_MASK: {
4083 struct kvm_signal_mask __user *sigmask_arg = argp;
4084 struct kvm_signal_mask kvm_sigmask;
4089 if (copy_from_user(&kvm_sigmask, argp,
4090 sizeof(kvm_sigmask)))
4093 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4096 if (get_compat_sigset(&sigset,
4097 (compat_sigset_t __user *)sigmask_arg->sigset))
4099 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4101 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4105 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4113 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4115 struct kvm_device *dev = filp->private_data;
4118 return dev->ops->mmap(dev, vma);
4123 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4124 int (*accessor)(struct kvm_device *dev,
4125 struct kvm_device_attr *attr),
4128 struct kvm_device_attr attr;
4133 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4136 return accessor(dev, &attr);
4139 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4142 struct kvm_device *dev = filp->private_data;
4144 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4148 case KVM_SET_DEVICE_ATTR:
4149 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4150 case KVM_GET_DEVICE_ATTR:
4151 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4152 case KVM_HAS_DEVICE_ATTR:
4153 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4155 if (dev->ops->ioctl)
4156 return dev->ops->ioctl(dev, ioctl, arg);
4162 static int kvm_device_release(struct inode *inode, struct file *filp)
4164 struct kvm_device *dev = filp->private_data;
4165 struct kvm *kvm = dev->kvm;
4167 if (dev->ops->release) {
4168 mutex_lock(&kvm->lock);
4169 list_del(&dev->vm_node);
4170 dev->ops->release(dev);
4171 mutex_unlock(&kvm->lock);
4178 static const struct file_operations kvm_device_fops = {
4179 .unlocked_ioctl = kvm_device_ioctl,
4180 .release = kvm_device_release,
4181 KVM_COMPAT(kvm_device_ioctl),
4182 .mmap = kvm_device_mmap,
4185 struct kvm_device *kvm_device_from_filp(struct file *filp)
4187 if (filp->f_op != &kvm_device_fops)
4190 return filp->private_data;
4193 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4194 #ifdef CONFIG_KVM_MPIC
4195 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4196 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4200 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4202 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4205 if (kvm_device_ops_table[type] != NULL)
4208 kvm_device_ops_table[type] = ops;
4212 void kvm_unregister_device_ops(u32 type)
4214 if (kvm_device_ops_table[type] != NULL)
4215 kvm_device_ops_table[type] = NULL;
4218 static int kvm_ioctl_create_device(struct kvm *kvm,
4219 struct kvm_create_device *cd)
4221 const struct kvm_device_ops *ops = NULL;
4222 struct kvm_device *dev;
4223 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4227 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4230 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4231 ops = kvm_device_ops_table[type];
4238 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4245 mutex_lock(&kvm->lock);
4246 ret = ops->create(dev, type);
4248 mutex_unlock(&kvm->lock);
4252 list_add(&dev->vm_node, &kvm->devices);
4253 mutex_unlock(&kvm->lock);
4259 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4261 kvm_put_kvm_no_destroy(kvm);
4262 mutex_lock(&kvm->lock);
4263 list_del(&dev->vm_node);
4264 mutex_unlock(&kvm->lock);
4273 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4276 case KVM_CAP_USER_MEMORY:
4277 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4278 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4279 case KVM_CAP_INTERNAL_ERROR_DATA:
4280 #ifdef CONFIG_HAVE_KVM_MSI
4281 case KVM_CAP_SIGNAL_MSI:
4283 #ifdef CONFIG_HAVE_KVM_IRQFD
4285 case KVM_CAP_IRQFD_RESAMPLE:
4287 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4288 case KVM_CAP_CHECK_EXTENSION_VM:
4289 case KVM_CAP_ENABLE_CAP_VM:
4290 case KVM_CAP_HALT_POLL:
4292 #ifdef CONFIG_KVM_MMIO
4293 case KVM_CAP_COALESCED_MMIO:
4294 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4295 case KVM_CAP_COALESCED_PIO:
4298 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4299 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4300 return KVM_DIRTY_LOG_MANUAL_CAPS;
4302 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4303 case KVM_CAP_IRQ_ROUTING:
4304 return KVM_MAX_IRQ_ROUTES;
4306 #if KVM_ADDRESS_SPACE_NUM > 1
4307 case KVM_CAP_MULTI_ADDRESS_SPACE:
4308 return KVM_ADDRESS_SPACE_NUM;
4310 case KVM_CAP_NR_MEMSLOTS:
4311 return KVM_USER_MEM_SLOTS;
4312 case KVM_CAP_DIRTY_LOG_RING:
4313 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4314 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4318 case KVM_CAP_BINARY_STATS_FD:
4323 return kvm_vm_ioctl_check_extension(kvm, arg);
4326 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4330 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4333 /* the size should be power of 2 */
4334 if (!size || (size & (size - 1)))
4337 /* Should be bigger to keep the reserved entries, or a page */
4338 if (size < kvm_dirty_ring_get_rsvd_entries() *
4339 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4342 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4343 sizeof(struct kvm_dirty_gfn))
4346 /* We only allow it to set once */
4347 if (kvm->dirty_ring_size)
4350 mutex_lock(&kvm->lock);
4352 if (kvm->created_vcpus) {
4353 /* We don't allow to change this value after vcpu created */
4356 kvm->dirty_ring_size = size;
4360 mutex_unlock(&kvm->lock);
4364 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4367 struct kvm_vcpu *vcpu;
4370 if (!kvm->dirty_ring_size)
4373 mutex_lock(&kvm->slots_lock);
4375 kvm_for_each_vcpu(i, vcpu, kvm)
4376 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4378 mutex_unlock(&kvm->slots_lock);
4381 kvm_flush_remote_tlbs(kvm);
4386 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4387 struct kvm_enable_cap *cap)
4392 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4393 struct kvm_enable_cap *cap)
4396 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4397 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4398 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4400 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4401 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4403 if (cap->flags || (cap->args[0] & ~allowed_options))
4405 kvm->manual_dirty_log_protect = cap->args[0];
4409 case KVM_CAP_HALT_POLL: {
4410 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4413 kvm->max_halt_poll_ns = cap->args[0];
4416 case KVM_CAP_DIRTY_LOG_RING:
4417 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4419 return kvm_vm_ioctl_enable_cap(kvm, cap);
4423 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4424 size_t size, loff_t *offset)
4426 struct kvm *kvm = file->private_data;
4428 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4429 &kvm_vm_stats_desc[0], &kvm->stat,
4430 sizeof(kvm->stat), user_buffer, size, offset);
4433 static const struct file_operations kvm_vm_stats_fops = {
4434 .read = kvm_vm_stats_read,
4435 .llseek = noop_llseek,
4438 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4443 fd = get_unused_fd_flags(O_CLOEXEC);
4447 file = anon_inode_getfile("kvm-vm-stats",
4448 &kvm_vm_stats_fops, kvm, O_RDONLY);
4451 return PTR_ERR(file);
4453 file->f_mode |= FMODE_PREAD;
4454 fd_install(fd, file);
4459 static long kvm_vm_ioctl(struct file *filp,
4460 unsigned int ioctl, unsigned long arg)
4462 struct kvm *kvm = filp->private_data;
4463 void __user *argp = (void __user *)arg;
4466 if (kvm->mm != current->mm || kvm->vm_dead)
4469 case KVM_CREATE_VCPU:
4470 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4472 case KVM_ENABLE_CAP: {
4473 struct kvm_enable_cap cap;
4476 if (copy_from_user(&cap, argp, sizeof(cap)))
4478 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4481 case KVM_SET_USER_MEMORY_REGION: {
4482 struct kvm_userspace_memory_region kvm_userspace_mem;
4485 if (copy_from_user(&kvm_userspace_mem, argp,
4486 sizeof(kvm_userspace_mem)))
4489 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4492 case KVM_GET_DIRTY_LOG: {
4493 struct kvm_dirty_log log;
4496 if (copy_from_user(&log, argp, sizeof(log)))
4498 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4501 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4502 case KVM_CLEAR_DIRTY_LOG: {
4503 struct kvm_clear_dirty_log log;
4506 if (copy_from_user(&log, argp, sizeof(log)))
4508 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4512 #ifdef CONFIG_KVM_MMIO
4513 case KVM_REGISTER_COALESCED_MMIO: {
4514 struct kvm_coalesced_mmio_zone zone;
4517 if (copy_from_user(&zone, argp, sizeof(zone)))
4519 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4522 case KVM_UNREGISTER_COALESCED_MMIO: {
4523 struct kvm_coalesced_mmio_zone zone;
4526 if (copy_from_user(&zone, argp, sizeof(zone)))
4528 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4533 struct kvm_irqfd data;
4536 if (copy_from_user(&data, argp, sizeof(data)))
4538 r = kvm_irqfd(kvm, &data);
4541 case KVM_IOEVENTFD: {
4542 struct kvm_ioeventfd data;
4545 if (copy_from_user(&data, argp, sizeof(data)))
4547 r = kvm_ioeventfd(kvm, &data);
4550 #ifdef CONFIG_HAVE_KVM_MSI
4551 case KVM_SIGNAL_MSI: {
4555 if (copy_from_user(&msi, argp, sizeof(msi)))
4557 r = kvm_send_userspace_msi(kvm, &msi);
4561 #ifdef __KVM_HAVE_IRQ_LINE
4562 case KVM_IRQ_LINE_STATUS:
4563 case KVM_IRQ_LINE: {
4564 struct kvm_irq_level irq_event;
4567 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4570 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4571 ioctl == KVM_IRQ_LINE_STATUS);
4576 if (ioctl == KVM_IRQ_LINE_STATUS) {
4577 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4585 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4586 case KVM_SET_GSI_ROUTING: {
4587 struct kvm_irq_routing routing;
4588 struct kvm_irq_routing __user *urouting;
4589 struct kvm_irq_routing_entry *entries = NULL;
4592 if (copy_from_user(&routing, argp, sizeof(routing)))
4595 if (!kvm_arch_can_set_irq_routing(kvm))
4597 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4603 entries = vmemdup_user(urouting->entries,
4604 array_size(sizeof(*entries),
4606 if (IS_ERR(entries)) {
4607 r = PTR_ERR(entries);
4611 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4616 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4617 case KVM_CREATE_DEVICE: {
4618 struct kvm_create_device cd;
4621 if (copy_from_user(&cd, argp, sizeof(cd)))
4624 r = kvm_ioctl_create_device(kvm, &cd);
4629 if (copy_to_user(argp, &cd, sizeof(cd)))
4635 case KVM_CHECK_EXTENSION:
4636 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4638 case KVM_RESET_DIRTY_RINGS:
4639 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4641 case KVM_GET_STATS_FD:
4642 r = kvm_vm_ioctl_get_stats_fd(kvm);
4645 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4651 #ifdef CONFIG_KVM_COMPAT
4652 struct compat_kvm_dirty_log {
4656 compat_uptr_t dirty_bitmap; /* one bit per page */
4661 struct compat_kvm_clear_dirty_log {
4666 compat_uptr_t dirty_bitmap; /* one bit per page */
4671 static long kvm_vm_compat_ioctl(struct file *filp,
4672 unsigned int ioctl, unsigned long arg)
4674 struct kvm *kvm = filp->private_data;
4677 if (kvm->mm != current->mm || kvm->vm_dead)
4680 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4681 case KVM_CLEAR_DIRTY_LOG: {
4682 struct compat_kvm_clear_dirty_log compat_log;
4683 struct kvm_clear_dirty_log log;
4685 if (copy_from_user(&compat_log, (void __user *)arg,
4686 sizeof(compat_log)))
4688 log.slot = compat_log.slot;
4689 log.num_pages = compat_log.num_pages;
4690 log.first_page = compat_log.first_page;
4691 log.padding2 = compat_log.padding2;
4692 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4694 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4698 case KVM_GET_DIRTY_LOG: {
4699 struct compat_kvm_dirty_log compat_log;
4700 struct kvm_dirty_log log;
4702 if (copy_from_user(&compat_log, (void __user *)arg,
4703 sizeof(compat_log)))
4705 log.slot = compat_log.slot;
4706 log.padding1 = compat_log.padding1;
4707 log.padding2 = compat_log.padding2;
4708 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4710 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4714 r = kvm_vm_ioctl(filp, ioctl, arg);
4720 static struct file_operations kvm_vm_fops = {
4721 .release = kvm_vm_release,
4722 .unlocked_ioctl = kvm_vm_ioctl,
4723 .llseek = noop_llseek,
4724 KVM_COMPAT(kvm_vm_compat_ioctl),
4727 bool file_is_kvm(struct file *file)
4729 return file && file->f_op == &kvm_vm_fops;
4731 EXPORT_SYMBOL_GPL(file_is_kvm);
4733 static int kvm_dev_ioctl_create_vm(unsigned long type)
4739 kvm = kvm_create_vm(type);
4741 return PTR_ERR(kvm);
4742 #ifdef CONFIG_KVM_MMIO
4743 r = kvm_coalesced_mmio_init(kvm);
4747 r = get_unused_fd_flags(O_CLOEXEC);
4751 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4752 "kvm-%d", task_pid_nr(current));
4754 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4762 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4763 * already set, with ->release() being kvm_vm_release(). In error
4764 * cases it will be called by the final fput(file) and will take
4765 * care of doing kvm_put_kvm(kvm).
4767 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4772 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4774 fd_install(r, file);
4782 static long kvm_dev_ioctl(struct file *filp,
4783 unsigned int ioctl, unsigned long arg)
4788 case KVM_GET_API_VERSION:
4791 r = KVM_API_VERSION;
4794 r = kvm_dev_ioctl_create_vm(arg);
4796 case KVM_CHECK_EXTENSION:
4797 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4799 case KVM_GET_VCPU_MMAP_SIZE:
4802 r = PAGE_SIZE; /* struct kvm_run */
4804 r += PAGE_SIZE; /* pio data page */
4806 #ifdef CONFIG_KVM_MMIO
4807 r += PAGE_SIZE; /* coalesced mmio ring page */
4810 case KVM_TRACE_ENABLE:
4811 case KVM_TRACE_PAUSE:
4812 case KVM_TRACE_DISABLE:
4816 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4822 static struct file_operations kvm_chardev_ops = {
4823 .unlocked_ioctl = kvm_dev_ioctl,
4824 .llseek = noop_llseek,
4825 KVM_COMPAT(kvm_dev_ioctl),
4828 static struct miscdevice kvm_dev = {
4834 static void hardware_enable_nolock(void *junk)
4836 int cpu = raw_smp_processor_id();
4839 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4842 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4844 r = kvm_arch_hardware_enable();
4847 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4848 atomic_inc(&hardware_enable_failed);
4849 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4853 static int kvm_starting_cpu(unsigned int cpu)
4855 raw_spin_lock(&kvm_count_lock);
4856 if (kvm_usage_count)
4857 hardware_enable_nolock(NULL);
4858 raw_spin_unlock(&kvm_count_lock);
4862 static void hardware_disable_nolock(void *junk)
4864 int cpu = raw_smp_processor_id();
4866 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4868 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4869 kvm_arch_hardware_disable();
4872 static int kvm_dying_cpu(unsigned int cpu)
4874 raw_spin_lock(&kvm_count_lock);
4875 if (kvm_usage_count)
4876 hardware_disable_nolock(NULL);
4877 raw_spin_unlock(&kvm_count_lock);
4881 static void hardware_disable_all_nolock(void)
4883 BUG_ON(!kvm_usage_count);
4886 if (!kvm_usage_count)
4887 on_each_cpu(hardware_disable_nolock, NULL, 1);
4890 static void hardware_disable_all(void)
4892 raw_spin_lock(&kvm_count_lock);
4893 hardware_disable_all_nolock();
4894 raw_spin_unlock(&kvm_count_lock);
4897 static int hardware_enable_all(void)
4901 raw_spin_lock(&kvm_count_lock);
4904 if (kvm_usage_count == 1) {
4905 atomic_set(&hardware_enable_failed, 0);
4906 on_each_cpu(hardware_enable_nolock, NULL, 1);
4908 if (atomic_read(&hardware_enable_failed)) {
4909 hardware_disable_all_nolock();
4914 raw_spin_unlock(&kvm_count_lock);
4919 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4923 * Some (well, at least mine) BIOSes hang on reboot if
4926 * And Intel TXT required VMX off for all cpu when system shutdown.
4928 pr_info("kvm: exiting hardware virtualization\n");
4929 kvm_rebooting = true;
4930 on_each_cpu(hardware_disable_nolock, NULL, 1);
4934 static struct notifier_block kvm_reboot_notifier = {
4935 .notifier_call = kvm_reboot,
4939 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4943 for (i = 0; i < bus->dev_count; i++) {
4944 struct kvm_io_device *pos = bus->range[i].dev;
4946 kvm_iodevice_destructor(pos);
4951 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4952 const struct kvm_io_range *r2)
4954 gpa_t addr1 = r1->addr;
4955 gpa_t addr2 = r2->addr;
4960 /* If r2->len == 0, match the exact address. If r2->len != 0,
4961 * accept any overlapping write. Any order is acceptable for
4962 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4963 * we process all of them.
4976 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4978 return kvm_io_bus_cmp(p1, p2);
4981 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4982 gpa_t addr, int len)
4984 struct kvm_io_range *range, key;
4987 key = (struct kvm_io_range) {
4992 range = bsearch(&key, bus->range, bus->dev_count,
4993 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4997 off = range - bus->range;
4999 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5005 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5006 struct kvm_io_range *range, const void *val)
5010 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5014 while (idx < bus->dev_count &&
5015 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5016 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5025 /* kvm_io_bus_write - called under kvm->slots_lock */
5026 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5027 int len, const void *val)
5029 struct kvm_io_bus *bus;
5030 struct kvm_io_range range;
5033 range = (struct kvm_io_range) {
5038 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5041 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5042 return r < 0 ? r : 0;
5044 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5046 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5047 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5048 gpa_t addr, int len, const void *val, long cookie)
5050 struct kvm_io_bus *bus;
5051 struct kvm_io_range range;
5053 range = (struct kvm_io_range) {
5058 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5062 /* First try the device referenced by cookie. */
5063 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5064 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5065 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5070 * cookie contained garbage; fall back to search and return the
5071 * correct cookie value.
5073 return __kvm_io_bus_write(vcpu, bus, &range, val);
5076 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5077 struct kvm_io_range *range, void *val)
5081 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5085 while (idx < bus->dev_count &&
5086 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5087 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5096 /* kvm_io_bus_read - called under kvm->slots_lock */
5097 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5100 struct kvm_io_bus *bus;
5101 struct kvm_io_range range;
5104 range = (struct kvm_io_range) {
5109 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5112 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5113 return r < 0 ? r : 0;
5116 /* Caller must hold slots_lock. */
5117 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5118 int len, struct kvm_io_device *dev)
5121 struct kvm_io_bus *new_bus, *bus;
5122 struct kvm_io_range range;
5124 bus = kvm_get_bus(kvm, bus_idx);
5128 /* exclude ioeventfd which is limited by maximum fd */
5129 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5132 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5133 GFP_KERNEL_ACCOUNT);
5137 range = (struct kvm_io_range) {
5143 for (i = 0; i < bus->dev_count; i++)
5144 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5147 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5148 new_bus->dev_count++;
5149 new_bus->range[i] = range;
5150 memcpy(new_bus->range + i + 1, bus->range + i,
5151 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5152 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5153 synchronize_srcu_expedited(&kvm->srcu);
5159 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5160 struct kvm_io_device *dev)
5163 struct kvm_io_bus *new_bus, *bus;
5165 lockdep_assert_held(&kvm->slots_lock);
5167 bus = kvm_get_bus(kvm, bus_idx);
5171 for (i = 0; i < bus->dev_count; i++) {
5172 if (bus->range[i].dev == dev) {
5177 if (i == bus->dev_count)
5180 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5181 GFP_KERNEL_ACCOUNT);
5183 memcpy(new_bus, bus, struct_size(bus, range, i));
5184 new_bus->dev_count--;
5185 memcpy(new_bus->range + i, bus->range + i + 1,
5186 flex_array_size(new_bus, range, new_bus->dev_count - i));
5189 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5190 synchronize_srcu_expedited(&kvm->srcu);
5192 /* Destroy the old bus _after_ installing the (null) bus. */
5194 pr_err("kvm: failed to shrink bus, removing it completely\n");
5195 for (j = 0; j < bus->dev_count; j++) {
5198 kvm_iodevice_destructor(bus->range[j].dev);
5203 return new_bus ? 0 : -ENOMEM;
5206 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5209 struct kvm_io_bus *bus;
5210 int dev_idx, srcu_idx;
5211 struct kvm_io_device *iodev = NULL;
5213 srcu_idx = srcu_read_lock(&kvm->srcu);
5215 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5219 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5223 iodev = bus->range[dev_idx].dev;
5226 srcu_read_unlock(&kvm->srcu, srcu_idx);
5230 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5232 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5233 int (*get)(void *, u64 *), int (*set)(void *, u64),
5236 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5240 * The debugfs files are a reference to the kvm struct which
5241 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5242 * avoids the race between open and the removal of the debugfs directory.
5244 if (!kvm_get_kvm_safe(stat_data->kvm))
5247 if (simple_attr_open(inode, file, get,
5248 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5251 kvm_put_kvm(stat_data->kvm);
5258 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5260 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5263 simple_attr_release(inode, file);
5264 kvm_put_kvm(stat_data->kvm);
5269 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5271 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5276 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5278 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5283 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5286 struct kvm_vcpu *vcpu;
5290 kvm_for_each_vcpu(i, vcpu, kvm)
5291 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5296 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5299 struct kvm_vcpu *vcpu;
5301 kvm_for_each_vcpu(i, vcpu, kvm)
5302 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5307 static int kvm_stat_data_get(void *data, u64 *val)
5310 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5312 switch (stat_data->kind) {
5314 r = kvm_get_stat_per_vm(stat_data->kvm,
5315 stat_data->desc->desc.offset, val);
5318 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5319 stat_data->desc->desc.offset, val);
5326 static int kvm_stat_data_clear(void *data, u64 val)
5329 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5334 switch (stat_data->kind) {
5336 r = kvm_clear_stat_per_vm(stat_data->kvm,
5337 stat_data->desc->desc.offset);
5340 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5341 stat_data->desc->desc.offset);
5348 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5350 __simple_attr_check_format("%llu\n", 0ull);
5351 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5352 kvm_stat_data_clear, "%llu\n");
5355 static const struct file_operations stat_fops_per_vm = {
5356 .owner = THIS_MODULE,
5357 .open = kvm_stat_data_open,
5358 .release = kvm_debugfs_release,
5359 .read = simple_attr_read,
5360 .write = simple_attr_write,
5361 .llseek = no_llseek,
5364 static int vm_stat_get(void *_offset, u64 *val)
5366 unsigned offset = (long)_offset;
5371 mutex_lock(&kvm_lock);
5372 list_for_each_entry(kvm, &vm_list, vm_list) {
5373 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5376 mutex_unlock(&kvm_lock);
5380 static int vm_stat_clear(void *_offset, u64 val)
5382 unsigned offset = (long)_offset;
5388 mutex_lock(&kvm_lock);
5389 list_for_each_entry(kvm, &vm_list, vm_list) {
5390 kvm_clear_stat_per_vm(kvm, offset);
5392 mutex_unlock(&kvm_lock);
5397 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5398 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5400 static int vcpu_stat_get(void *_offset, u64 *val)
5402 unsigned offset = (long)_offset;
5407 mutex_lock(&kvm_lock);
5408 list_for_each_entry(kvm, &vm_list, vm_list) {
5409 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5412 mutex_unlock(&kvm_lock);
5416 static int vcpu_stat_clear(void *_offset, u64 val)
5418 unsigned offset = (long)_offset;
5424 mutex_lock(&kvm_lock);
5425 list_for_each_entry(kvm, &vm_list, vm_list) {
5426 kvm_clear_stat_per_vcpu(kvm, offset);
5428 mutex_unlock(&kvm_lock);
5433 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5435 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5437 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5439 struct kobj_uevent_env *env;
5440 unsigned long long created, active;
5442 if (!kvm_dev.this_device || !kvm)
5445 mutex_lock(&kvm_lock);
5446 if (type == KVM_EVENT_CREATE_VM) {
5447 kvm_createvm_count++;
5449 } else if (type == KVM_EVENT_DESTROY_VM) {
5452 created = kvm_createvm_count;
5453 active = kvm_active_vms;
5454 mutex_unlock(&kvm_lock);
5456 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5460 add_uevent_var(env, "CREATED=%llu", created);
5461 add_uevent_var(env, "COUNT=%llu", active);
5463 if (type == KVM_EVENT_CREATE_VM) {
5464 add_uevent_var(env, "EVENT=create");
5465 kvm->userspace_pid = task_pid_nr(current);
5466 } else if (type == KVM_EVENT_DESTROY_VM) {
5467 add_uevent_var(env, "EVENT=destroy");
5469 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5471 if (kvm->debugfs_dentry) {
5472 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5475 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5477 add_uevent_var(env, "STATS_PATH=%s", tmp);
5481 /* no need for checks, since we are adding at most only 5 keys */
5482 env->envp[env->envp_idx++] = NULL;
5483 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5487 static void kvm_init_debug(void)
5489 const struct file_operations *fops;
5490 const struct _kvm_stats_desc *pdesc;
5493 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5495 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5496 pdesc = &kvm_vm_stats_desc[i];
5497 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5498 fops = &vm_stat_fops;
5500 fops = &vm_stat_readonly_fops;
5501 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5503 (void *)(long)pdesc->desc.offset, fops);
5506 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5507 pdesc = &kvm_vcpu_stats_desc[i];
5508 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5509 fops = &vcpu_stat_fops;
5511 fops = &vcpu_stat_readonly_fops;
5512 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5514 (void *)(long)pdesc->desc.offset, fops);
5518 static int kvm_suspend(void)
5520 if (kvm_usage_count)
5521 hardware_disable_nolock(NULL);
5525 static void kvm_resume(void)
5527 if (kvm_usage_count) {
5528 #ifdef CONFIG_LOCKDEP
5529 WARN_ON(lockdep_is_held(&kvm_count_lock));
5531 hardware_enable_nolock(NULL);
5535 static struct syscore_ops kvm_syscore_ops = {
5536 .suspend = kvm_suspend,
5537 .resume = kvm_resume,
5541 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5543 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5546 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5548 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5550 WRITE_ONCE(vcpu->preempted, false);
5551 WRITE_ONCE(vcpu->ready, false);
5553 __this_cpu_write(kvm_running_vcpu, vcpu);
5554 kvm_arch_sched_in(vcpu, cpu);
5555 kvm_arch_vcpu_load(vcpu, cpu);
5558 static void kvm_sched_out(struct preempt_notifier *pn,
5559 struct task_struct *next)
5561 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5563 if (current->on_rq) {
5564 WRITE_ONCE(vcpu->preempted, true);
5565 WRITE_ONCE(vcpu->ready, true);
5567 kvm_arch_vcpu_put(vcpu);
5568 __this_cpu_write(kvm_running_vcpu, NULL);
5572 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5574 * We can disable preemption locally around accessing the per-CPU variable,
5575 * and use the resolved vcpu pointer after enabling preemption again,
5576 * because even if the current thread is migrated to another CPU, reading
5577 * the per-CPU value later will give us the same value as we update the
5578 * per-CPU variable in the preempt notifier handlers.
5580 struct kvm_vcpu *kvm_get_running_vcpu(void)
5582 struct kvm_vcpu *vcpu;
5585 vcpu = __this_cpu_read(kvm_running_vcpu);
5590 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5593 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5595 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5597 return &kvm_running_vcpu;
5600 struct kvm_cpu_compat_check {
5605 static void check_processor_compat(void *data)
5607 struct kvm_cpu_compat_check *c = data;
5609 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5612 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5613 struct module *module)
5615 struct kvm_cpu_compat_check c;
5619 r = kvm_arch_init(opaque);
5624 * kvm_arch_init makes sure there's at most one caller
5625 * for architectures that support multiple implementations,
5626 * like intel and amd on x86.
5627 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5628 * conflicts in case kvm is already setup for another implementation.
5630 r = kvm_irqfd_init();
5634 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5639 r = kvm_arch_hardware_setup(opaque);
5645 for_each_online_cpu(cpu) {
5646 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5651 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5652 kvm_starting_cpu, kvm_dying_cpu);
5655 register_reboot_notifier(&kvm_reboot_notifier);
5657 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5659 vcpu_align = __alignof__(struct kvm_vcpu);
5661 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5663 offsetof(struct kvm_vcpu, arch),
5664 offsetofend(struct kvm_vcpu, stats_id)
5665 - offsetof(struct kvm_vcpu, arch),
5667 if (!kvm_vcpu_cache) {
5672 for_each_possible_cpu(cpu) {
5673 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5674 GFP_KERNEL, cpu_to_node(cpu))) {
5680 r = kvm_async_pf_init();
5684 kvm_chardev_ops.owner = module;
5685 kvm_vm_fops.owner = module;
5686 kvm_vcpu_fops.owner = module;
5688 r = misc_register(&kvm_dev);
5690 pr_err("kvm: misc device register failed\n");
5694 register_syscore_ops(&kvm_syscore_ops);
5696 kvm_preempt_ops.sched_in = kvm_sched_in;
5697 kvm_preempt_ops.sched_out = kvm_sched_out;
5701 r = kvm_vfio_ops_init();
5707 kvm_async_pf_deinit();
5709 for_each_possible_cpu(cpu)
5710 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5712 kmem_cache_destroy(kvm_vcpu_cache);
5714 unregister_reboot_notifier(&kvm_reboot_notifier);
5715 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5717 kvm_arch_hardware_unsetup();
5719 free_cpumask_var(cpus_hardware_enabled);
5727 EXPORT_SYMBOL_GPL(kvm_init);
5733 debugfs_remove_recursive(kvm_debugfs_dir);
5734 misc_deregister(&kvm_dev);
5735 for_each_possible_cpu(cpu)
5736 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5737 kmem_cache_destroy(kvm_vcpu_cache);
5738 kvm_async_pf_deinit();
5739 unregister_syscore_ops(&kvm_syscore_ops);
5740 unregister_reboot_notifier(&kvm_reboot_notifier);
5741 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5742 on_each_cpu(hardware_disable_nolock, NULL, 1);
5743 kvm_arch_hardware_unsetup();
5746 free_cpumask_var(cpus_hardware_enabled);
5747 kvm_vfio_ops_exit();
5749 EXPORT_SYMBOL_GPL(kvm_exit);
5751 struct kvm_vm_worker_thread_context {
5753 struct task_struct *parent;
5754 struct completion init_done;
5755 kvm_vm_thread_fn_t thread_fn;
5760 static int kvm_vm_worker_thread(void *context)
5763 * The init_context is allocated on the stack of the parent thread, so
5764 * we have to locally copy anything that is needed beyond initialization
5766 struct kvm_vm_worker_thread_context *init_context = context;
5767 struct kvm *kvm = init_context->kvm;
5768 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5769 uintptr_t data = init_context->data;
5772 err = kthread_park(current);
5773 /* kthread_park(current) is never supposed to return an error */
5778 err = cgroup_attach_task_all(init_context->parent, current);
5780 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5785 set_user_nice(current, task_nice(init_context->parent));
5788 init_context->err = err;
5789 complete(&init_context->init_done);
5790 init_context = NULL;
5795 /* Wait to be woken up by the spawner before proceeding. */
5798 if (!kthread_should_stop())
5799 err = thread_fn(kvm, data);
5804 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5805 uintptr_t data, const char *name,
5806 struct task_struct **thread_ptr)
5808 struct kvm_vm_worker_thread_context init_context = {};
5809 struct task_struct *thread;
5812 init_context.kvm = kvm;
5813 init_context.parent = current;
5814 init_context.thread_fn = thread_fn;
5815 init_context.data = data;
5816 init_completion(&init_context.init_done);
5818 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5819 "%s-%d", name, task_pid_nr(current));
5821 return PTR_ERR(thread);
5823 /* kthread_run is never supposed to return NULL */
5824 WARN_ON(thread == NULL);
5826 wait_for_completion(&init_context.init_done);
5828 if (!init_context.err)
5829 *thread_ptr = thread;
5831 return init_context.err;