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>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
75 module_param(halt_poll_ns, uint, 0644);
76 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow = 2;
80 module_param(halt_poll_ns_grow, uint, 0644);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 /* The start value to grow halt_poll_ns from */
84 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
85 module_param(halt_poll_ns_grow_start, uint, 0644);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
88 /* Default resets per-vcpu halt_poll_ns . */
89 unsigned int halt_poll_ns_shrink;
90 module_param(halt_poll_ns_shrink, uint, 0644);
91 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
96 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
99 DEFINE_MUTEX(kvm_lock);
100 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
103 static cpumask_var_t cpus_hardware_enabled;
104 static int kvm_usage_count;
105 static atomic_t hardware_enable_failed;
107 struct kmem_cache *kvm_vcpu_cache;
108 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
110 static __read_mostly struct preempt_ops kvm_preempt_ops;
112 struct dentry *kvm_debugfs_dir;
113 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
115 static int kvm_debugfs_num_entries;
116 static const struct file_operations *stat_fops_per_vm[];
118 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 * For architectures that don't implement a compat infrastructure,
127 * adopt a double line of defense:
128 * - Prevent a compat task from opening /dev/kvm
129 * - If the open has been done by a 64bit task, and the KVM fd
130 * passed to a compat task, let the ioctls fail.
132 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
133 unsigned long arg) { return -EINVAL; }
135 static int kvm_no_compat_open(struct inode *inode, struct file *file)
137 return is_compat_task() ? -ENODEV : 0;
139 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
140 .open = kvm_no_compat_open
142 static int hardware_enable_all(void);
143 static void hardware_disable_all(void);
145 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
147 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 static bool largepages_enabled = true;
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
160 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 unsigned long start, unsigned long end, bool blockable)
166 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
169 * The metadata used by is_zone_device_page() to determine whether or
170 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
171 * the device has been pinned, e.g. by get_user_pages(). WARN if the
172 * page_count() is zero to help detect bad usage of this helper.
174 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
177 return is_zone_device_page(pfn_to_page(pfn));
180 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
183 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
184 * perspective they are "normal" pages, albeit with slightly different
188 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)
200 preempt_notifier_register(&vcpu->preempt_notifier);
201 kvm_arch_vcpu_load(vcpu, cpu);
204 EXPORT_SYMBOL_GPL(vcpu_load);
206 void vcpu_put(struct kvm_vcpu *vcpu)
209 kvm_arch_vcpu_put(vcpu);
210 preempt_notifier_unregister(&vcpu->preempt_notifier);
213 EXPORT_SYMBOL_GPL(vcpu_put);
215 /* TODO: merge with kvm_arch_vcpu_should_kick */
216 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
218 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
221 * We need to wait for the VCPU to reenable interrupts and get out of
222 * READING_SHADOW_PAGE_TABLES mode.
224 if (req & KVM_REQUEST_WAIT)
225 return mode != OUTSIDE_GUEST_MODE;
228 * Need to kick a running VCPU, but otherwise there is nothing to do.
230 return mode == IN_GUEST_MODE;
233 static void ack_flush(void *_completed)
237 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
240 cpus = cpu_online_mask;
242 if (cpumask_empty(cpus))
245 smp_call_function_many(cpus, ack_flush, NULL, wait);
249 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
250 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
253 struct kvm_vcpu *vcpu;
258 kvm_for_each_vcpu(i, vcpu, kvm) {
259 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
262 kvm_make_request(req, vcpu);
265 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
268 if (tmp != NULL && cpu != -1 && cpu != me &&
269 kvm_request_needs_ipi(vcpu, req))
270 __cpumask_set_cpu(cpu, tmp);
273 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
279 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
284 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
286 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
288 free_cpumask_var(cpus);
292 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
293 void kvm_flush_remote_tlbs(struct kvm *kvm)
296 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
297 * kvm_make_all_cpus_request.
299 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
302 * We want to publish modifications to the page tables before reading
303 * mode. Pairs with a memory barrier in arch-specific code.
304 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
305 * and smp_mb in walk_shadow_page_lockless_begin/end.
306 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
308 * There is already an smp_mb__after_atomic() before
309 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
312 if (!kvm_arch_flush_remote_tlb(kvm)
313 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
314 ++kvm->stat.remote_tlb_flush;
315 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
317 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
320 void kvm_reload_remote_mmus(struct kvm *kvm)
322 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
325 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
330 mutex_init(&vcpu->mutex);
335 init_swait_queue_head(&vcpu->wq);
336 kvm_async_pf_vcpu_init(vcpu);
339 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
341 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
346 vcpu->run = page_address(page);
348 kvm_vcpu_set_in_spin_loop(vcpu, false);
349 kvm_vcpu_set_dy_eligible(vcpu, false);
350 vcpu->preempted = false;
353 r = kvm_arch_vcpu_init(vcpu);
359 free_page((unsigned long)vcpu->run);
363 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
365 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
368 * no need for rcu_read_lock as VCPU_RUN is the only place that
369 * will change the vcpu->pid pointer and on uninit all file
370 * descriptors are already gone.
372 put_pid(rcu_dereference_protected(vcpu->pid, 1));
373 kvm_arch_vcpu_uninit(vcpu);
374 free_page((unsigned long)vcpu->run);
376 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
378 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
379 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
381 return container_of(mn, struct kvm, mmu_notifier);
384 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
385 struct mm_struct *mm,
386 unsigned long address,
389 struct kvm *kvm = mmu_notifier_to_kvm(mn);
392 idx = srcu_read_lock(&kvm->srcu);
393 spin_lock(&kvm->mmu_lock);
394 kvm->mmu_notifier_seq++;
396 if (kvm_set_spte_hva(kvm, address, pte))
397 kvm_flush_remote_tlbs(kvm);
399 spin_unlock(&kvm->mmu_lock);
400 srcu_read_unlock(&kvm->srcu, idx);
403 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
404 const struct mmu_notifier_range *range)
406 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 int need_tlb_flush = 0, idx;
410 idx = srcu_read_lock(&kvm->srcu);
411 spin_lock(&kvm->mmu_lock);
413 * The count increase must become visible at unlock time as no
414 * spte can be established without taking the mmu_lock and
415 * count is also read inside the mmu_lock critical section.
417 kvm->mmu_notifier_count++;
418 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
419 need_tlb_flush |= kvm->tlbs_dirty;
420 /* we've to flush the tlb before the pages can be freed */
422 kvm_flush_remote_tlbs(kvm);
424 spin_unlock(&kvm->mmu_lock);
426 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
428 mmu_notifier_range_blockable(range));
430 srcu_read_unlock(&kvm->srcu, idx);
435 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
436 const struct mmu_notifier_range *range)
438 struct kvm *kvm = mmu_notifier_to_kvm(mn);
440 spin_lock(&kvm->mmu_lock);
442 * This sequence increase will notify the kvm page fault that
443 * the page that is going to be mapped in the spte could have
446 kvm->mmu_notifier_seq++;
449 * The above sequence increase must be visible before the
450 * below count decrease, which is ensured by the smp_wmb above
451 * in conjunction with the smp_rmb in mmu_notifier_retry().
453 kvm->mmu_notifier_count--;
454 spin_unlock(&kvm->mmu_lock);
456 BUG_ON(kvm->mmu_notifier_count < 0);
459 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
460 struct mm_struct *mm,
464 struct kvm *kvm = mmu_notifier_to_kvm(mn);
467 idx = srcu_read_lock(&kvm->srcu);
468 spin_lock(&kvm->mmu_lock);
470 young = kvm_age_hva(kvm, start, end);
472 kvm_flush_remote_tlbs(kvm);
474 spin_unlock(&kvm->mmu_lock);
475 srcu_read_unlock(&kvm->srcu, idx);
480 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
481 struct mm_struct *mm,
485 struct kvm *kvm = mmu_notifier_to_kvm(mn);
488 idx = srcu_read_lock(&kvm->srcu);
489 spin_lock(&kvm->mmu_lock);
491 * Even though we do not flush TLB, this will still adversely
492 * affect performance on pre-Haswell Intel EPT, where there is
493 * no EPT Access Bit to clear so that we have to tear down EPT
494 * tables instead. If we find this unacceptable, we can always
495 * add a parameter to kvm_age_hva so that it effectively doesn't
496 * do anything on clear_young.
498 * Also note that currently we never issue secondary TLB flushes
499 * from clear_young, leaving this job up to the regular system
500 * cadence. If we find this inaccurate, we might come up with a
501 * more sophisticated heuristic later.
503 young = kvm_age_hva(kvm, start, end);
504 spin_unlock(&kvm->mmu_lock);
505 srcu_read_unlock(&kvm->srcu, idx);
510 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
511 struct mm_struct *mm,
512 unsigned long address)
514 struct kvm *kvm = mmu_notifier_to_kvm(mn);
517 idx = srcu_read_lock(&kvm->srcu);
518 spin_lock(&kvm->mmu_lock);
519 young = kvm_test_age_hva(kvm, address);
520 spin_unlock(&kvm->mmu_lock);
521 srcu_read_unlock(&kvm->srcu, idx);
526 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
527 struct mm_struct *mm)
529 struct kvm *kvm = mmu_notifier_to_kvm(mn);
532 idx = srcu_read_lock(&kvm->srcu);
533 kvm_arch_flush_shadow_all(kvm);
534 srcu_read_unlock(&kvm->srcu, idx);
537 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
538 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
539 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
540 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
541 .clear_young = kvm_mmu_notifier_clear_young,
542 .test_young = kvm_mmu_notifier_test_young,
543 .change_pte = kvm_mmu_notifier_change_pte,
544 .release = kvm_mmu_notifier_release,
547 static int kvm_init_mmu_notifier(struct kvm *kvm)
549 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
550 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
553 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
555 static int kvm_init_mmu_notifier(struct kvm *kvm)
560 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
562 static struct kvm_memslots *kvm_alloc_memslots(void)
565 struct kvm_memslots *slots;
567 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
571 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
572 slots->id_to_index[i] = slots->memslots[i].id = i;
577 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
579 if (!memslot->dirty_bitmap)
582 kvfree(memslot->dirty_bitmap);
583 memslot->dirty_bitmap = NULL;
587 * Free any memory in @free but not in @dont.
589 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
590 struct kvm_memory_slot *dont)
592 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
593 kvm_destroy_dirty_bitmap(free);
595 kvm_arch_free_memslot(kvm, free, dont);
600 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
602 struct kvm_memory_slot *memslot;
607 kvm_for_each_memslot(memslot, slots)
608 kvm_free_memslot(kvm, memslot, NULL);
613 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
617 if (!kvm->debugfs_dentry)
620 debugfs_remove_recursive(kvm->debugfs_dentry);
622 if (kvm->debugfs_stat_data) {
623 for (i = 0; i < kvm_debugfs_num_entries; i++)
624 kfree(kvm->debugfs_stat_data[i]);
625 kfree(kvm->debugfs_stat_data);
629 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
631 char dir_name[ITOA_MAX_LEN * 2];
632 struct kvm_stat_data *stat_data;
633 struct kvm_stats_debugfs_item *p;
635 if (!debugfs_initialized())
638 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
639 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
641 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
642 sizeof(*kvm->debugfs_stat_data),
644 if (!kvm->debugfs_stat_data)
647 for (p = debugfs_entries; p->name; p++) {
648 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
652 stat_data->kvm = kvm;
653 stat_data->offset = p->offset;
654 stat_data->mode = p->mode ? p->mode : 0644;
655 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
656 debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry,
657 stat_data, stat_fops_per_vm[p->kind]);
663 * Called after the VM is otherwise initialized, but just before adding it to
666 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
672 * Called just after removing the VM from the vm_list, but before doing any
675 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
679 static struct kvm *kvm_create_vm(unsigned long type)
681 struct kvm *kvm = kvm_arch_alloc_vm();
686 return ERR_PTR(-ENOMEM);
688 spin_lock_init(&kvm->mmu_lock);
690 kvm->mm = current->mm;
691 kvm_eventfd_init(kvm);
692 mutex_init(&kvm->lock);
693 mutex_init(&kvm->irq_lock);
694 mutex_init(&kvm->slots_lock);
695 INIT_LIST_HEAD(&kvm->devices);
697 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
699 if (init_srcu_struct(&kvm->srcu))
700 goto out_err_no_srcu;
701 if (init_srcu_struct(&kvm->irq_srcu))
702 goto out_err_no_irq_srcu;
704 refcount_set(&kvm->users_count, 1);
705 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
706 struct kvm_memslots *slots = kvm_alloc_memslots();
709 goto out_err_no_arch_destroy_vm;
710 /* Generations must be different for each address space. */
711 slots->generation = i;
712 rcu_assign_pointer(kvm->memslots[i], slots);
715 for (i = 0; i < KVM_NR_BUSES; i++) {
716 rcu_assign_pointer(kvm->buses[i],
717 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
719 goto out_err_no_arch_destroy_vm;
722 r = kvm_arch_init_vm(kvm, type);
724 goto out_err_no_arch_destroy_vm;
726 r = hardware_enable_all();
728 goto out_err_no_disable;
730 #ifdef CONFIG_HAVE_KVM_IRQFD
731 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
734 r = kvm_init_mmu_notifier(kvm);
736 goto out_err_no_mmu_notifier;
738 r = kvm_arch_post_init_vm(kvm);
742 mutex_lock(&kvm_lock);
743 list_add(&kvm->vm_list, &vm_list);
744 mutex_unlock(&kvm_lock);
746 preempt_notifier_inc();
751 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
752 if (kvm->mmu_notifier.ops)
753 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
755 out_err_no_mmu_notifier:
756 hardware_disable_all();
758 kvm_arch_destroy_vm(kvm);
759 out_err_no_arch_destroy_vm:
760 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
761 for (i = 0; i < KVM_NR_BUSES; i++)
762 kfree(kvm_get_bus(kvm, i));
763 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
764 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
765 cleanup_srcu_struct(&kvm->irq_srcu);
767 cleanup_srcu_struct(&kvm->srcu);
769 kvm_arch_free_vm(kvm);
774 static void kvm_destroy_devices(struct kvm *kvm)
776 struct kvm_device *dev, *tmp;
779 * We do not need to take the kvm->lock here, because nobody else
780 * has a reference to the struct kvm at this point and therefore
781 * cannot access the devices list anyhow.
783 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
784 list_del(&dev->vm_node);
785 dev->ops->destroy(dev);
789 static void kvm_destroy_vm(struct kvm *kvm)
792 struct mm_struct *mm = kvm->mm;
794 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
795 kvm_destroy_vm_debugfs(kvm);
796 kvm_arch_sync_events(kvm);
797 mutex_lock(&kvm_lock);
798 list_del(&kvm->vm_list);
799 mutex_unlock(&kvm_lock);
800 kvm_arch_pre_destroy_vm(kvm);
802 kvm_free_irq_routing(kvm);
803 for (i = 0; i < KVM_NR_BUSES; i++) {
804 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
807 kvm_io_bus_destroy(bus);
808 kvm->buses[i] = NULL;
810 kvm_coalesced_mmio_free(kvm);
811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
812 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
814 kvm_arch_flush_shadow_all(kvm);
816 kvm_arch_destroy_vm(kvm);
817 kvm_destroy_devices(kvm);
818 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
819 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
820 cleanup_srcu_struct(&kvm->irq_srcu);
821 cleanup_srcu_struct(&kvm->srcu);
822 kvm_arch_free_vm(kvm);
823 preempt_notifier_dec();
824 hardware_disable_all();
828 void kvm_get_kvm(struct kvm *kvm)
830 refcount_inc(&kvm->users_count);
832 EXPORT_SYMBOL_GPL(kvm_get_kvm);
834 void kvm_put_kvm(struct kvm *kvm)
836 if (refcount_dec_and_test(&kvm->users_count))
839 EXPORT_SYMBOL_GPL(kvm_put_kvm);
842 static int kvm_vm_release(struct inode *inode, struct file *filp)
844 struct kvm *kvm = filp->private_data;
846 kvm_irqfd_release(kvm);
853 * Allocation size is twice as large as the actual dirty bitmap size.
854 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
856 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
858 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
860 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
861 if (!memslot->dirty_bitmap)
868 * Insert memslot and re-sort memslots based on their GFN,
869 * so binary search could be used to lookup GFN.
870 * Sorting algorithm takes advantage of having initially
871 * sorted array and known changed memslot position.
873 static void update_memslots(struct kvm_memslots *slots,
874 struct kvm_memory_slot *new,
875 enum kvm_mr_change change)
878 int i = slots->id_to_index[id];
879 struct kvm_memory_slot *mslots = slots->memslots;
881 WARN_ON(mslots[i].id != id);
885 WARN_ON(mslots[i].npages || !new->npages);
889 WARN_ON(new->npages || !mslots[i].npages);
895 while (i < KVM_MEM_SLOTS_NUM - 1 &&
896 new->base_gfn <= mslots[i + 1].base_gfn) {
897 if (!mslots[i + 1].npages)
899 mslots[i] = mslots[i + 1];
900 slots->id_to_index[mslots[i].id] = i;
905 * The ">=" is needed when creating a slot with base_gfn == 0,
906 * so that it moves before all those with base_gfn == npages == 0.
908 * On the other hand, if new->npages is zero, the above loop has
909 * already left i pointing to the beginning of the empty part of
910 * mslots, and the ">=" would move the hole backwards in this
911 * case---which is wrong. So skip the loop when deleting a slot.
915 new->base_gfn >= mslots[i - 1].base_gfn) {
916 mslots[i] = mslots[i - 1];
917 slots->id_to_index[mslots[i].id] = i;
921 WARN_ON_ONCE(i != slots->used_slots);
924 slots->id_to_index[mslots[i].id] = i;
927 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
929 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
931 #ifdef __KVM_HAVE_READONLY_MEM
932 valid_flags |= KVM_MEM_READONLY;
935 if (mem->flags & ~valid_flags)
941 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
942 int as_id, struct kvm_memslots *slots)
944 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
945 u64 gen = old_memslots->generation;
947 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
948 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
950 rcu_assign_pointer(kvm->memslots[as_id], slots);
951 synchronize_srcu_expedited(&kvm->srcu);
954 * Increment the new memslot generation a second time, dropping the
955 * update in-progress flag and incrementing then generation based on
956 * the number of address spaces. This provides a unique and easily
957 * identifiable generation number while the memslots are in flux.
959 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
962 * Generations must be unique even across address spaces. We do not need
963 * a global counter for that, instead the generation space is evenly split
964 * across address spaces. For example, with two address spaces, address
965 * space 0 will use generations 0, 2, 4, ... while address space 1 will
966 * use generations 1, 3, 5, ...
968 gen += KVM_ADDRESS_SPACE_NUM;
970 kvm_arch_memslots_updated(kvm, gen);
972 slots->generation = gen;
978 * Allocate some memory and give it an address in the guest physical address
981 * Discontiguous memory is allowed, mostly for framebuffers.
983 * Must be called holding kvm->slots_lock for write.
985 int __kvm_set_memory_region(struct kvm *kvm,
986 const struct kvm_userspace_memory_region *mem)
990 unsigned long npages;
991 struct kvm_memory_slot *slot;
992 struct kvm_memory_slot old, new;
993 struct kvm_memslots *slots = NULL, *old_memslots;
995 enum kvm_mr_change change;
997 r = check_memory_region_flags(mem);
1002 as_id = mem->slot >> 16;
1003 id = (u16)mem->slot;
1005 /* General sanity checks */
1006 if (mem->memory_size & (PAGE_SIZE - 1))
1008 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1010 /* We can read the guest memory with __xxx_user() later on. */
1011 if ((id < KVM_USER_MEM_SLOTS) &&
1012 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1013 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1016 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1018 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1021 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1022 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1023 npages = mem->memory_size >> PAGE_SHIFT;
1025 if (npages > KVM_MEM_MAX_NR_PAGES)
1031 new.base_gfn = base_gfn;
1032 new.npages = npages;
1033 new.flags = mem->flags;
1037 change = KVM_MR_CREATE;
1038 else { /* Modify an existing slot. */
1039 if ((mem->userspace_addr != old.userspace_addr) ||
1040 (npages != old.npages) ||
1041 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1044 if (base_gfn != old.base_gfn)
1045 change = KVM_MR_MOVE;
1046 else if (new.flags != old.flags)
1047 change = KVM_MR_FLAGS_ONLY;
1048 else { /* Nothing to change. */
1057 change = KVM_MR_DELETE;
1062 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1063 /* Check for overlaps */
1065 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1068 if (!((base_gfn + npages <= slot->base_gfn) ||
1069 (base_gfn >= slot->base_gfn + slot->npages)))
1074 /* Free page dirty bitmap if unneeded */
1075 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1076 new.dirty_bitmap = NULL;
1079 if (change == KVM_MR_CREATE) {
1080 new.userspace_addr = mem->userspace_addr;
1082 if (kvm_arch_create_memslot(kvm, &new, npages))
1086 /* Allocate page dirty bitmap if needed */
1087 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1088 if (kvm_create_dirty_bitmap(&new) < 0)
1092 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1095 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1097 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1098 slot = id_to_memslot(slots, id);
1099 slot->flags |= KVM_MEMSLOT_INVALID;
1101 old_memslots = install_new_memslots(kvm, as_id, slots);
1103 /* From this point no new shadow pages pointing to a deleted,
1104 * or moved, memslot will be created.
1106 * validation of sp->gfn happens in:
1107 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1108 * - kvm_is_visible_gfn (mmu_check_roots)
1110 kvm_arch_flush_shadow_memslot(kvm, slot);
1113 * We can re-use the old_memslots from above, the only difference
1114 * from the currently installed memslots is the invalid flag. This
1115 * will get overwritten by update_memslots anyway.
1117 slots = old_memslots;
1120 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1124 /* actual memory is freed via old in kvm_free_memslot below */
1125 if (change == KVM_MR_DELETE) {
1126 new.dirty_bitmap = NULL;
1127 memset(&new.arch, 0, sizeof(new.arch));
1130 update_memslots(slots, &new, change);
1131 old_memslots = install_new_memslots(kvm, as_id, slots);
1133 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1135 kvm_free_memslot(kvm, &old, &new);
1136 kvfree(old_memslots);
1142 kvm_free_memslot(kvm, &new, &old);
1146 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1148 int kvm_set_memory_region(struct kvm *kvm,
1149 const struct kvm_userspace_memory_region *mem)
1153 mutex_lock(&kvm->slots_lock);
1154 r = __kvm_set_memory_region(kvm, mem);
1155 mutex_unlock(&kvm->slots_lock);
1158 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1160 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1161 struct kvm_userspace_memory_region *mem)
1163 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1166 return kvm_set_memory_region(kvm, mem);
1169 int kvm_get_dirty_log(struct kvm *kvm,
1170 struct kvm_dirty_log *log, int *is_dirty)
1172 struct kvm_memslots *slots;
1173 struct kvm_memory_slot *memslot;
1176 unsigned long any = 0;
1178 as_id = log->slot >> 16;
1179 id = (u16)log->slot;
1180 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1183 slots = __kvm_memslots(kvm, as_id);
1184 memslot = id_to_memslot(slots, id);
1185 if (!memslot->dirty_bitmap)
1188 n = kvm_dirty_bitmap_bytes(memslot);
1190 for (i = 0; !any && i < n/sizeof(long); ++i)
1191 any = memslot->dirty_bitmap[i];
1193 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1200 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1202 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1204 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1205 * and reenable dirty page tracking for the corresponding pages.
1206 * @kvm: pointer to kvm instance
1207 * @log: slot id and address to which we copy the log
1208 * @flush: true if TLB flush is needed by caller
1210 * We need to keep it in mind that VCPU threads can write to the bitmap
1211 * concurrently. So, to avoid losing track of dirty pages we keep the
1214 * 1. Take a snapshot of the bit and clear it if needed.
1215 * 2. Write protect the corresponding page.
1216 * 3. Copy the snapshot to the userspace.
1217 * 4. Upon return caller flushes TLB's if needed.
1219 * Between 2 and 4, the guest may write to the page using the remaining TLB
1220 * entry. This is not a problem because the page is reported dirty using
1221 * the snapshot taken before and step 4 ensures that writes done after
1222 * exiting to userspace will be logged for the next call.
1225 int kvm_get_dirty_log_protect(struct kvm *kvm,
1226 struct kvm_dirty_log *log, bool *flush)
1228 struct kvm_memslots *slots;
1229 struct kvm_memory_slot *memslot;
1232 unsigned long *dirty_bitmap;
1233 unsigned long *dirty_bitmap_buffer;
1235 as_id = log->slot >> 16;
1236 id = (u16)log->slot;
1237 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1240 slots = __kvm_memslots(kvm, as_id);
1241 memslot = id_to_memslot(slots, id);
1243 dirty_bitmap = memslot->dirty_bitmap;
1247 n = kvm_dirty_bitmap_bytes(memslot);
1249 if (kvm->manual_dirty_log_protect) {
1251 * Unlike kvm_get_dirty_log, we always return false in *flush,
1252 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1253 * is some code duplication between this function and
1254 * kvm_get_dirty_log, but hopefully all architecture
1255 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1256 * can be eliminated.
1258 dirty_bitmap_buffer = dirty_bitmap;
1260 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1261 memset(dirty_bitmap_buffer, 0, n);
1263 spin_lock(&kvm->mmu_lock);
1264 for (i = 0; i < n / sizeof(long); i++) {
1268 if (!dirty_bitmap[i])
1272 mask = xchg(&dirty_bitmap[i], 0);
1273 dirty_bitmap_buffer[i] = mask;
1275 offset = i * BITS_PER_LONG;
1276 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1279 spin_unlock(&kvm->mmu_lock);
1282 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1286 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1289 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1290 * and reenable dirty page tracking for the corresponding pages.
1291 * @kvm: pointer to kvm instance
1292 * @log: slot id and address from which to fetch the bitmap of dirty pages
1293 * @flush: true if TLB flush is needed by caller
1295 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1296 struct kvm_clear_dirty_log *log, bool *flush)
1298 struct kvm_memslots *slots;
1299 struct kvm_memory_slot *memslot;
1303 unsigned long *dirty_bitmap;
1304 unsigned long *dirty_bitmap_buffer;
1306 as_id = log->slot >> 16;
1307 id = (u16)log->slot;
1308 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1311 if (log->first_page & 63)
1314 slots = __kvm_memslots(kvm, as_id);
1315 memslot = id_to_memslot(slots, id);
1317 dirty_bitmap = memslot->dirty_bitmap;
1321 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1323 if (log->first_page > memslot->npages ||
1324 log->num_pages > memslot->npages - log->first_page ||
1325 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1329 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1330 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1333 spin_lock(&kvm->mmu_lock);
1334 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1335 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1336 i++, offset += BITS_PER_LONG) {
1337 unsigned long mask = *dirty_bitmap_buffer++;
1338 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1342 mask &= atomic_long_fetch_andnot(mask, p);
1345 * mask contains the bits that really have been cleared. This
1346 * never includes any bits beyond the length of the memslot (if
1347 * the length is not aligned to 64 pages), therefore it is not
1348 * a problem if userspace sets them in log->dirty_bitmap.
1352 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1356 spin_unlock(&kvm->mmu_lock);
1360 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1363 bool kvm_largepages_enabled(void)
1365 return largepages_enabled;
1368 void kvm_disable_largepages(void)
1370 largepages_enabled = false;
1372 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1374 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1376 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1378 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1380 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1382 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1385 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1387 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1389 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1390 memslot->flags & KVM_MEMSLOT_INVALID)
1395 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1397 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1399 struct vm_area_struct *vma;
1400 unsigned long addr, size;
1404 addr = gfn_to_hva(kvm, gfn);
1405 if (kvm_is_error_hva(addr))
1408 down_read(¤t->mm->mmap_sem);
1409 vma = find_vma(current->mm, addr);
1413 size = vma_kernel_pagesize(vma);
1416 up_read(¤t->mm->mmap_sem);
1421 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1423 return slot->flags & KVM_MEM_READONLY;
1426 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1427 gfn_t *nr_pages, bool write)
1429 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1430 return KVM_HVA_ERR_BAD;
1432 if (memslot_is_readonly(slot) && write)
1433 return KVM_HVA_ERR_RO_BAD;
1436 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1438 return __gfn_to_hva_memslot(slot, gfn);
1441 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1444 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1447 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1450 return gfn_to_hva_many(slot, gfn, NULL);
1452 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1454 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1456 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1458 EXPORT_SYMBOL_GPL(gfn_to_hva);
1460 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1462 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1464 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1467 * Return the hva of a @gfn and the R/W attribute if possible.
1469 * @slot: the kvm_memory_slot which contains @gfn
1470 * @gfn: the gfn to be translated
1471 * @writable: used to return the read/write attribute of the @slot if the hva
1472 * is valid and @writable is not NULL
1474 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1475 gfn_t gfn, bool *writable)
1477 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1479 if (!kvm_is_error_hva(hva) && writable)
1480 *writable = !memslot_is_readonly(slot);
1485 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1487 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1489 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1492 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1494 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1496 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1499 static inline int check_user_page_hwpoison(unsigned long addr)
1501 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1503 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1504 return rc == -EHWPOISON;
1508 * The fast path to get the writable pfn which will be stored in @pfn,
1509 * true indicates success, otherwise false is returned. It's also the
1510 * only part that runs if we can are in atomic context.
1512 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1513 bool *writable, kvm_pfn_t *pfn)
1515 struct page *page[1];
1519 * Fast pin a writable pfn only if it is a write fault request
1520 * or the caller allows to map a writable pfn for a read fault
1523 if (!(write_fault || writable))
1526 npages = __get_user_pages_fast(addr, 1, 1, page);
1528 *pfn = page_to_pfn(page[0]);
1539 * The slow path to get the pfn of the specified host virtual address,
1540 * 1 indicates success, -errno is returned if error is detected.
1542 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1543 bool *writable, kvm_pfn_t *pfn)
1545 unsigned int flags = FOLL_HWPOISON;
1552 *writable = write_fault;
1555 flags |= FOLL_WRITE;
1557 flags |= FOLL_NOWAIT;
1559 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1563 /* map read fault as writable if possible */
1564 if (unlikely(!write_fault) && writable) {
1567 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1573 *pfn = page_to_pfn(page);
1577 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1579 if (unlikely(!(vma->vm_flags & VM_READ)))
1582 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1588 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1589 unsigned long addr, bool *async,
1590 bool write_fault, bool *writable,
1596 r = follow_pfn(vma, addr, &pfn);
1599 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1600 * not call the fault handler, so do it here.
1602 bool unlocked = false;
1603 r = fixup_user_fault(current, current->mm, addr,
1604 (write_fault ? FAULT_FLAG_WRITE : 0),
1611 r = follow_pfn(vma, addr, &pfn);
1621 * Get a reference here because callers of *hva_to_pfn* and
1622 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1623 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1624 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1625 * simply do nothing for reserved pfns.
1627 * Whoever called remap_pfn_range is also going to call e.g.
1628 * unmap_mapping_range before the underlying pages are freed,
1629 * causing a call to our MMU notifier.
1638 * Pin guest page in memory and return its pfn.
1639 * @addr: host virtual address which maps memory to the guest
1640 * @atomic: whether this function can sleep
1641 * @async: whether this function need to wait IO complete if the
1642 * host page is not in the memory
1643 * @write_fault: whether we should get a writable host page
1644 * @writable: whether it allows to map a writable host page for !@write_fault
1646 * The function will map a writable host page for these two cases:
1647 * 1): @write_fault = true
1648 * 2): @write_fault = false && @writable, @writable will tell the caller
1649 * whether the mapping is writable.
1651 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1652 bool write_fault, bool *writable)
1654 struct vm_area_struct *vma;
1658 /* we can do it either atomically or asynchronously, not both */
1659 BUG_ON(atomic && async);
1661 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1665 return KVM_PFN_ERR_FAULT;
1667 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1671 down_read(¤t->mm->mmap_sem);
1672 if (npages == -EHWPOISON ||
1673 (!async && check_user_page_hwpoison(addr))) {
1674 pfn = KVM_PFN_ERR_HWPOISON;
1679 vma = find_vma_intersection(current->mm, addr, addr + 1);
1682 pfn = KVM_PFN_ERR_FAULT;
1683 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1684 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1688 pfn = KVM_PFN_ERR_FAULT;
1690 if (async && vma_is_valid(vma, write_fault))
1692 pfn = KVM_PFN_ERR_FAULT;
1695 up_read(¤t->mm->mmap_sem);
1699 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1700 bool atomic, bool *async, bool write_fault,
1703 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1705 if (addr == KVM_HVA_ERR_RO_BAD) {
1708 return KVM_PFN_ERR_RO_FAULT;
1711 if (kvm_is_error_hva(addr)) {
1714 return KVM_PFN_NOSLOT;
1717 /* Do not map writable pfn in the readonly memslot. */
1718 if (writable && memslot_is_readonly(slot)) {
1723 return hva_to_pfn(addr, atomic, async, write_fault,
1726 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1728 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1731 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1732 write_fault, writable);
1734 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1736 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1738 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1740 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1742 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1744 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1746 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1748 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1750 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1752 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1754 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1756 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1758 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1760 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1762 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1764 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1766 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1768 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1770 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1772 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1773 struct page **pages, int nr_pages)
1778 addr = gfn_to_hva_many(slot, gfn, &entry);
1779 if (kvm_is_error_hva(addr))
1782 if (entry < nr_pages)
1785 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1787 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1789 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1791 if (is_error_noslot_pfn(pfn))
1792 return KVM_ERR_PTR_BAD_PAGE;
1794 if (kvm_is_reserved_pfn(pfn)) {
1796 return KVM_ERR_PTR_BAD_PAGE;
1799 return pfn_to_page(pfn);
1802 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1806 pfn = gfn_to_pfn(kvm, gfn);
1808 return kvm_pfn_to_page(pfn);
1810 EXPORT_SYMBOL_GPL(gfn_to_page);
1812 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1813 struct kvm_host_map *map)
1817 struct page *page = KVM_UNMAPPED_PAGE;
1822 pfn = gfn_to_pfn_memslot(slot, gfn);
1823 if (is_error_noslot_pfn(pfn))
1826 if (pfn_valid(pfn)) {
1827 page = pfn_to_page(pfn);
1829 #ifdef CONFIG_HAS_IOMEM
1831 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1846 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1848 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1850 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1852 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1861 if (map->page != KVM_UNMAPPED_PAGE)
1863 #ifdef CONFIG_HAS_IOMEM
1869 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1870 kvm_release_pfn_dirty(map->pfn);
1872 kvm_release_pfn_clean(map->pfn);
1878 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1880 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1884 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1886 return kvm_pfn_to_page(pfn);
1888 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1890 void kvm_release_page_clean(struct page *page)
1892 WARN_ON(is_error_page(page));
1894 kvm_release_pfn_clean(page_to_pfn(page));
1896 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1898 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1900 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1901 put_page(pfn_to_page(pfn));
1903 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1905 void kvm_release_page_dirty(struct page *page)
1907 WARN_ON(is_error_page(page));
1909 kvm_release_pfn_dirty(page_to_pfn(page));
1911 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1913 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1915 kvm_set_pfn_dirty(pfn);
1916 kvm_release_pfn_clean(pfn);
1918 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1920 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1922 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) {
1923 struct page *page = pfn_to_page(pfn);
1928 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1930 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1932 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1933 mark_page_accessed(pfn_to_page(pfn));
1935 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1937 void kvm_get_pfn(kvm_pfn_t pfn)
1939 if (!kvm_is_reserved_pfn(pfn))
1940 get_page(pfn_to_page(pfn));
1942 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1944 static int next_segment(unsigned long len, int offset)
1946 if (len > PAGE_SIZE - offset)
1947 return PAGE_SIZE - offset;
1952 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1953 void *data, int offset, int len)
1958 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1959 if (kvm_is_error_hva(addr))
1961 r = __copy_from_user(data, (void __user *)addr + offset, len);
1967 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1970 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1972 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1974 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1976 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1977 int offset, int len)
1979 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1981 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1983 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1985 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1987 gfn_t gfn = gpa >> PAGE_SHIFT;
1989 int offset = offset_in_page(gpa);
1992 while ((seg = next_segment(len, offset)) != 0) {
1993 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2003 EXPORT_SYMBOL_GPL(kvm_read_guest);
2005 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2007 gfn_t gfn = gpa >> PAGE_SHIFT;
2009 int offset = offset_in_page(gpa);
2012 while ((seg = next_segment(len, offset)) != 0) {
2013 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2023 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2025 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2026 void *data, int offset, unsigned long len)
2031 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2032 if (kvm_is_error_hva(addr))
2034 pagefault_disable();
2035 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2042 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
2045 gfn_t gfn = gpa >> PAGE_SHIFT;
2046 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2047 int offset = offset_in_page(gpa);
2049 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2051 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
2053 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2054 void *data, unsigned long len)
2056 gfn_t gfn = gpa >> PAGE_SHIFT;
2057 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2058 int offset = offset_in_page(gpa);
2060 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2062 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2064 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2065 const void *data, int offset, int len)
2070 addr = gfn_to_hva_memslot(memslot, gfn);
2071 if (kvm_is_error_hva(addr))
2073 r = __copy_to_user((void __user *)addr + offset, data, len);
2076 mark_page_dirty_in_slot(memslot, gfn);
2080 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2081 const void *data, int offset, int len)
2083 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2085 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2087 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2089 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2090 const void *data, int offset, int len)
2092 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2094 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2096 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2098 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2101 gfn_t gfn = gpa >> PAGE_SHIFT;
2103 int offset = offset_in_page(gpa);
2106 while ((seg = next_segment(len, offset)) != 0) {
2107 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2117 EXPORT_SYMBOL_GPL(kvm_write_guest);
2119 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2122 gfn_t gfn = gpa >> PAGE_SHIFT;
2124 int offset = offset_in_page(gpa);
2127 while ((seg = next_segment(len, offset)) != 0) {
2128 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2138 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2140 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2141 struct gfn_to_hva_cache *ghc,
2142 gpa_t gpa, unsigned long len)
2144 int offset = offset_in_page(gpa);
2145 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2146 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2147 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2148 gfn_t nr_pages_avail;
2149 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2152 ghc->generation = slots->generation;
2154 ghc->hva = KVM_HVA_ERR_BAD;
2157 * If the requested region crosses two memslots, we still
2158 * verify that the entire region is valid here.
2160 while (!r && start_gfn <= end_gfn) {
2161 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2162 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2164 if (kvm_is_error_hva(ghc->hva))
2166 start_gfn += nr_pages_avail;
2169 /* Use the slow path for cross page reads and writes. */
2170 if (!r && nr_pages_needed == 1)
2173 ghc->memslot = NULL;
2178 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2179 gpa_t gpa, unsigned long len)
2181 struct kvm_memslots *slots = kvm_memslots(kvm);
2182 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2184 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2186 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2187 void *data, unsigned int offset,
2190 struct kvm_memslots *slots = kvm_memslots(kvm);
2192 gpa_t gpa = ghc->gpa + offset;
2194 BUG_ON(len + offset > ghc->len);
2196 if (slots->generation != ghc->generation)
2197 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2199 if (unlikely(!ghc->memslot))
2200 return kvm_write_guest(kvm, gpa, data, len);
2202 if (kvm_is_error_hva(ghc->hva))
2205 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2208 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2212 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2214 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2215 void *data, unsigned long len)
2217 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2219 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2221 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2222 void *data, unsigned long len)
2224 struct kvm_memslots *slots = kvm_memslots(kvm);
2227 BUG_ON(len > ghc->len);
2229 if (slots->generation != ghc->generation)
2230 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2232 if (unlikely(!ghc->memslot))
2233 return kvm_read_guest(kvm, ghc->gpa, data, len);
2235 if (kvm_is_error_hva(ghc->hva))
2238 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2244 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2246 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2248 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2250 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2252 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2254 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2256 gfn_t gfn = gpa >> PAGE_SHIFT;
2258 int offset = offset_in_page(gpa);
2261 while ((seg = next_segment(len, offset)) != 0) {
2262 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2271 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2273 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2276 if (memslot && memslot->dirty_bitmap) {
2277 unsigned long rel_gfn = gfn - memslot->base_gfn;
2279 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2283 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2285 struct kvm_memory_slot *memslot;
2287 memslot = gfn_to_memslot(kvm, gfn);
2288 mark_page_dirty_in_slot(memslot, gfn);
2290 EXPORT_SYMBOL_GPL(mark_page_dirty);
2292 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2294 struct kvm_memory_slot *memslot;
2296 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2297 mark_page_dirty_in_slot(memslot, gfn);
2299 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2301 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2303 if (!vcpu->sigset_active)
2307 * This does a lockless modification of ->real_blocked, which is fine
2308 * because, only current can change ->real_blocked and all readers of
2309 * ->real_blocked don't care as long ->real_blocked is always a subset
2312 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2315 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2317 if (!vcpu->sigset_active)
2320 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2321 sigemptyset(¤t->real_blocked);
2324 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2326 unsigned int old, val, grow, grow_start;
2328 old = val = vcpu->halt_poll_ns;
2329 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2330 grow = READ_ONCE(halt_poll_ns_grow);
2335 if (val < grow_start)
2338 if (val > halt_poll_ns)
2341 vcpu->halt_poll_ns = val;
2343 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2346 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2348 unsigned int old, val, shrink;
2350 old = val = vcpu->halt_poll_ns;
2351 shrink = READ_ONCE(halt_poll_ns_shrink);
2357 vcpu->halt_poll_ns = val;
2358 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2361 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2364 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2366 if (kvm_arch_vcpu_runnable(vcpu)) {
2367 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2370 if (kvm_cpu_has_pending_timer(vcpu))
2372 if (signal_pending(current))
2377 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2382 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2384 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2387 DECLARE_SWAITQUEUE(wait);
2388 bool waited = false;
2391 kvm_arch_vcpu_blocking(vcpu);
2393 start = cur = ktime_get();
2394 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2395 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2397 ++vcpu->stat.halt_attempted_poll;
2400 * This sets KVM_REQ_UNHALT if an interrupt
2403 if (kvm_vcpu_check_block(vcpu) < 0) {
2404 ++vcpu->stat.halt_successful_poll;
2405 if (!vcpu_valid_wakeup(vcpu))
2406 ++vcpu->stat.halt_poll_invalid;
2410 } while (single_task_running() && ktime_before(cur, stop));
2414 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2416 if (kvm_vcpu_check_block(vcpu) < 0)
2423 finish_swait(&vcpu->wq, &wait);
2426 kvm_arch_vcpu_unblocking(vcpu);
2427 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2429 if (!kvm_arch_no_poll(vcpu)) {
2430 if (!vcpu_valid_wakeup(vcpu)) {
2431 shrink_halt_poll_ns(vcpu);
2432 } else if (halt_poll_ns) {
2433 if (block_ns <= vcpu->halt_poll_ns)
2435 /* we had a long block, shrink polling */
2436 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2437 shrink_halt_poll_ns(vcpu);
2438 /* we had a short halt and our poll time is too small */
2439 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2440 block_ns < halt_poll_ns)
2441 grow_halt_poll_ns(vcpu);
2443 vcpu->halt_poll_ns = 0;
2447 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2448 kvm_arch_vcpu_block_finish(vcpu);
2450 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2452 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2454 struct swait_queue_head *wqp;
2456 wqp = kvm_arch_vcpu_wq(vcpu);
2457 if (swq_has_sleeper(wqp)) {
2459 WRITE_ONCE(vcpu->ready, true);
2460 ++vcpu->stat.halt_wakeup;
2466 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2470 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2472 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2475 int cpu = vcpu->cpu;
2477 if (kvm_vcpu_wake_up(vcpu))
2481 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2482 if (kvm_arch_vcpu_should_kick(vcpu))
2483 smp_send_reschedule(cpu);
2486 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2487 #endif /* !CONFIG_S390 */
2489 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2492 struct task_struct *task = NULL;
2496 pid = rcu_dereference(target->pid);
2498 task = get_pid_task(pid, PIDTYPE_PID);
2502 ret = yield_to(task, 1);
2503 put_task_struct(task);
2507 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2510 * Helper that checks whether a VCPU is eligible for directed yield.
2511 * Most eligible candidate to yield is decided by following heuristics:
2513 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2514 * (preempted lock holder), indicated by @in_spin_loop.
2515 * Set at the beiginning and cleared at the end of interception/PLE handler.
2517 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2518 * chance last time (mostly it has become eligible now since we have probably
2519 * yielded to lockholder in last iteration. This is done by toggling
2520 * @dy_eligible each time a VCPU checked for eligibility.)
2522 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2523 * to preempted lock-holder could result in wrong VCPU selection and CPU
2524 * burning. Giving priority for a potential lock-holder increases lock
2527 * Since algorithm is based on heuristics, accessing another VCPU data without
2528 * locking does not harm. It may result in trying to yield to same VCPU, fail
2529 * and continue with next VCPU and so on.
2531 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2533 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2536 eligible = !vcpu->spin_loop.in_spin_loop ||
2537 vcpu->spin_loop.dy_eligible;
2539 if (vcpu->spin_loop.in_spin_loop)
2540 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2549 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2550 * a vcpu_load/vcpu_put pair. However, for most architectures
2551 * kvm_arch_vcpu_runnable does not require vcpu_load.
2553 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2555 return kvm_arch_vcpu_runnable(vcpu);
2558 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2560 if (kvm_arch_dy_runnable(vcpu))
2563 #ifdef CONFIG_KVM_ASYNC_PF
2564 if (!list_empty_careful(&vcpu->async_pf.done))
2571 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2573 struct kvm *kvm = me->kvm;
2574 struct kvm_vcpu *vcpu;
2575 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2581 kvm_vcpu_set_in_spin_loop(me, true);
2583 * We boost the priority of a VCPU that is runnable but not
2584 * currently running, because it got preempted by something
2585 * else and called schedule in __vcpu_run. Hopefully that
2586 * VCPU is holding the lock that we need and will release it.
2587 * We approximate round-robin by starting at the last boosted VCPU.
2589 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2590 kvm_for_each_vcpu(i, vcpu, kvm) {
2591 if (!pass && i <= last_boosted_vcpu) {
2592 i = last_boosted_vcpu;
2594 } else if (pass && i > last_boosted_vcpu)
2596 if (!READ_ONCE(vcpu->ready))
2600 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2602 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2603 !kvm_arch_vcpu_in_kernel(vcpu))
2605 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2608 yielded = kvm_vcpu_yield_to(vcpu);
2610 kvm->last_boosted_vcpu = i;
2612 } else if (yielded < 0) {
2619 kvm_vcpu_set_in_spin_loop(me, false);
2621 /* Ensure vcpu is not eligible during next spinloop */
2622 kvm_vcpu_set_dy_eligible(me, false);
2624 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2626 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2628 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2631 if (vmf->pgoff == 0)
2632 page = virt_to_page(vcpu->run);
2634 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2635 page = virt_to_page(vcpu->arch.pio_data);
2637 #ifdef CONFIG_KVM_MMIO
2638 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2639 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2642 return kvm_arch_vcpu_fault(vcpu, vmf);
2648 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2649 .fault = kvm_vcpu_fault,
2652 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2654 vma->vm_ops = &kvm_vcpu_vm_ops;
2658 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2660 struct kvm_vcpu *vcpu = filp->private_data;
2662 debugfs_remove_recursive(vcpu->debugfs_dentry);
2663 kvm_put_kvm(vcpu->kvm);
2667 static struct file_operations kvm_vcpu_fops = {
2668 .release = kvm_vcpu_release,
2669 .unlocked_ioctl = kvm_vcpu_ioctl,
2670 .mmap = kvm_vcpu_mmap,
2671 .llseek = noop_llseek,
2672 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2676 * Allocates an inode for the vcpu.
2678 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2680 char name[8 + 1 + ITOA_MAX_LEN + 1];
2682 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2683 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2686 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2688 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2689 char dir_name[ITOA_MAX_LEN * 2];
2691 if (!debugfs_initialized())
2694 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2695 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2696 vcpu->kvm->debugfs_dentry);
2698 kvm_arch_create_vcpu_debugfs(vcpu);
2703 * Creates some virtual cpus. Good luck creating more than one.
2705 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2708 struct kvm_vcpu *vcpu;
2710 if (id >= KVM_MAX_VCPU_ID)
2713 mutex_lock(&kvm->lock);
2714 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2715 mutex_unlock(&kvm->lock);
2719 kvm->created_vcpus++;
2720 mutex_unlock(&kvm->lock);
2722 vcpu = kvm_arch_vcpu_create(kvm, id);
2725 goto vcpu_decrement;
2728 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2730 r = kvm_arch_vcpu_setup(vcpu);
2734 kvm_create_vcpu_debugfs(vcpu);
2736 mutex_lock(&kvm->lock);
2737 if (kvm_get_vcpu_by_id(kvm, id)) {
2739 goto unlock_vcpu_destroy;
2742 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2744 /* Now it's all set up, let userspace reach it */
2746 r = create_vcpu_fd(vcpu);
2749 goto unlock_vcpu_destroy;
2752 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2755 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2756 * before kvm->online_vcpu's incremented value.
2759 atomic_inc(&kvm->online_vcpus);
2761 mutex_unlock(&kvm->lock);
2762 kvm_arch_vcpu_postcreate(vcpu);
2765 unlock_vcpu_destroy:
2766 mutex_unlock(&kvm->lock);
2767 debugfs_remove_recursive(vcpu->debugfs_dentry);
2769 kvm_arch_vcpu_destroy(vcpu);
2771 mutex_lock(&kvm->lock);
2772 kvm->created_vcpus--;
2773 mutex_unlock(&kvm->lock);
2777 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2780 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2781 vcpu->sigset_active = 1;
2782 vcpu->sigset = *sigset;
2784 vcpu->sigset_active = 0;
2788 static long kvm_vcpu_ioctl(struct file *filp,
2789 unsigned int ioctl, unsigned long arg)
2791 struct kvm_vcpu *vcpu = filp->private_data;
2792 void __user *argp = (void __user *)arg;
2794 struct kvm_fpu *fpu = NULL;
2795 struct kvm_sregs *kvm_sregs = NULL;
2797 if (vcpu->kvm->mm != current->mm)
2800 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2804 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2805 * execution; mutex_lock() would break them.
2807 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2808 if (r != -ENOIOCTLCMD)
2811 if (mutex_lock_killable(&vcpu->mutex))
2819 oldpid = rcu_access_pointer(vcpu->pid);
2820 if (unlikely(oldpid != task_pid(current))) {
2821 /* The thread running this VCPU changed. */
2824 r = kvm_arch_vcpu_run_pid_change(vcpu);
2828 newpid = get_task_pid(current, PIDTYPE_PID);
2829 rcu_assign_pointer(vcpu->pid, newpid);
2834 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2835 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2838 case KVM_GET_REGS: {
2839 struct kvm_regs *kvm_regs;
2842 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2845 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2849 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2856 case KVM_SET_REGS: {
2857 struct kvm_regs *kvm_regs;
2860 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2861 if (IS_ERR(kvm_regs)) {
2862 r = PTR_ERR(kvm_regs);
2865 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2869 case KVM_GET_SREGS: {
2870 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2871 GFP_KERNEL_ACCOUNT);
2875 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2879 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2884 case KVM_SET_SREGS: {
2885 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2886 if (IS_ERR(kvm_sregs)) {
2887 r = PTR_ERR(kvm_sregs);
2891 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2894 case KVM_GET_MP_STATE: {
2895 struct kvm_mp_state mp_state;
2897 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2901 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2906 case KVM_SET_MP_STATE: {
2907 struct kvm_mp_state mp_state;
2910 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2912 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2915 case KVM_TRANSLATE: {
2916 struct kvm_translation tr;
2919 if (copy_from_user(&tr, argp, sizeof(tr)))
2921 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2925 if (copy_to_user(argp, &tr, sizeof(tr)))
2930 case KVM_SET_GUEST_DEBUG: {
2931 struct kvm_guest_debug dbg;
2934 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2936 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2939 case KVM_SET_SIGNAL_MASK: {
2940 struct kvm_signal_mask __user *sigmask_arg = argp;
2941 struct kvm_signal_mask kvm_sigmask;
2942 sigset_t sigset, *p;
2947 if (copy_from_user(&kvm_sigmask, argp,
2948 sizeof(kvm_sigmask)))
2951 if (kvm_sigmask.len != sizeof(sigset))
2954 if (copy_from_user(&sigset, sigmask_arg->sigset,
2959 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2963 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2967 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2971 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2977 fpu = memdup_user(argp, sizeof(*fpu));
2983 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2987 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2990 mutex_unlock(&vcpu->mutex);
2996 #ifdef CONFIG_KVM_COMPAT
2997 static long kvm_vcpu_compat_ioctl(struct file *filp,
2998 unsigned int ioctl, unsigned long arg)
3000 struct kvm_vcpu *vcpu = filp->private_data;
3001 void __user *argp = compat_ptr(arg);
3004 if (vcpu->kvm->mm != current->mm)
3008 case KVM_SET_SIGNAL_MASK: {
3009 struct kvm_signal_mask __user *sigmask_arg = argp;
3010 struct kvm_signal_mask kvm_sigmask;
3015 if (copy_from_user(&kvm_sigmask, argp,
3016 sizeof(kvm_sigmask)))
3019 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3022 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3024 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3026 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3030 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3038 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3040 struct kvm_device *dev = filp->private_data;
3043 return dev->ops->mmap(dev, vma);
3048 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3049 int (*accessor)(struct kvm_device *dev,
3050 struct kvm_device_attr *attr),
3053 struct kvm_device_attr attr;
3058 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3061 return accessor(dev, &attr);
3064 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3067 struct kvm_device *dev = filp->private_data;
3069 if (dev->kvm->mm != current->mm)
3073 case KVM_SET_DEVICE_ATTR:
3074 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3075 case KVM_GET_DEVICE_ATTR:
3076 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3077 case KVM_HAS_DEVICE_ATTR:
3078 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3080 if (dev->ops->ioctl)
3081 return dev->ops->ioctl(dev, ioctl, arg);
3087 static int kvm_device_release(struct inode *inode, struct file *filp)
3089 struct kvm_device *dev = filp->private_data;
3090 struct kvm *kvm = dev->kvm;
3092 if (dev->ops->release) {
3093 mutex_lock(&kvm->lock);
3094 list_del(&dev->vm_node);
3095 dev->ops->release(dev);
3096 mutex_unlock(&kvm->lock);
3103 static const struct file_operations kvm_device_fops = {
3104 .unlocked_ioctl = kvm_device_ioctl,
3105 .release = kvm_device_release,
3106 KVM_COMPAT(kvm_device_ioctl),
3107 .mmap = kvm_device_mmap,
3110 struct kvm_device *kvm_device_from_filp(struct file *filp)
3112 if (filp->f_op != &kvm_device_fops)
3115 return filp->private_data;
3118 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3119 #ifdef CONFIG_KVM_MPIC
3120 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3121 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3125 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3127 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3130 if (kvm_device_ops_table[type] != NULL)
3133 kvm_device_ops_table[type] = ops;
3137 void kvm_unregister_device_ops(u32 type)
3139 if (kvm_device_ops_table[type] != NULL)
3140 kvm_device_ops_table[type] = NULL;
3143 static int kvm_ioctl_create_device(struct kvm *kvm,
3144 struct kvm_create_device *cd)
3146 struct kvm_device_ops *ops = NULL;
3147 struct kvm_device *dev;
3148 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3152 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3155 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3156 ops = kvm_device_ops_table[type];
3163 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3170 mutex_lock(&kvm->lock);
3171 ret = ops->create(dev, type);
3173 mutex_unlock(&kvm->lock);
3177 list_add(&dev->vm_node, &kvm->devices);
3178 mutex_unlock(&kvm->lock);
3184 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3187 mutex_lock(&kvm->lock);
3188 list_del(&dev->vm_node);
3189 mutex_unlock(&kvm->lock);
3198 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3201 case KVM_CAP_USER_MEMORY:
3202 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3203 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3204 case KVM_CAP_INTERNAL_ERROR_DATA:
3205 #ifdef CONFIG_HAVE_KVM_MSI
3206 case KVM_CAP_SIGNAL_MSI:
3208 #ifdef CONFIG_HAVE_KVM_IRQFD
3210 case KVM_CAP_IRQFD_RESAMPLE:
3212 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3213 case KVM_CAP_CHECK_EXTENSION_VM:
3214 case KVM_CAP_ENABLE_CAP_VM:
3215 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3216 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3219 #ifdef CONFIG_KVM_MMIO
3220 case KVM_CAP_COALESCED_MMIO:
3221 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3222 case KVM_CAP_COALESCED_PIO:
3225 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3226 case KVM_CAP_IRQ_ROUTING:
3227 return KVM_MAX_IRQ_ROUTES;
3229 #if KVM_ADDRESS_SPACE_NUM > 1
3230 case KVM_CAP_MULTI_ADDRESS_SPACE:
3231 return KVM_ADDRESS_SPACE_NUM;
3233 case KVM_CAP_NR_MEMSLOTS:
3234 return KVM_USER_MEM_SLOTS;
3238 return kvm_vm_ioctl_check_extension(kvm, arg);
3241 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3242 struct kvm_enable_cap *cap)
3247 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3248 struct kvm_enable_cap *cap)
3251 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3252 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3253 if (cap->flags || (cap->args[0] & ~1))
3255 kvm->manual_dirty_log_protect = cap->args[0];
3259 return kvm_vm_ioctl_enable_cap(kvm, cap);
3263 static long kvm_vm_ioctl(struct file *filp,
3264 unsigned int ioctl, unsigned long arg)
3266 struct kvm *kvm = filp->private_data;
3267 void __user *argp = (void __user *)arg;
3270 if (kvm->mm != current->mm)
3273 case KVM_CREATE_VCPU:
3274 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3276 case KVM_ENABLE_CAP: {
3277 struct kvm_enable_cap cap;
3280 if (copy_from_user(&cap, argp, sizeof(cap)))
3282 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3285 case KVM_SET_USER_MEMORY_REGION: {
3286 struct kvm_userspace_memory_region kvm_userspace_mem;
3289 if (copy_from_user(&kvm_userspace_mem, argp,
3290 sizeof(kvm_userspace_mem)))
3293 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3296 case KVM_GET_DIRTY_LOG: {
3297 struct kvm_dirty_log log;
3300 if (copy_from_user(&log, argp, sizeof(log)))
3302 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3305 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3306 case KVM_CLEAR_DIRTY_LOG: {
3307 struct kvm_clear_dirty_log log;
3310 if (copy_from_user(&log, argp, sizeof(log)))
3312 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3316 #ifdef CONFIG_KVM_MMIO
3317 case KVM_REGISTER_COALESCED_MMIO: {
3318 struct kvm_coalesced_mmio_zone zone;
3321 if (copy_from_user(&zone, argp, sizeof(zone)))
3323 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3326 case KVM_UNREGISTER_COALESCED_MMIO: {
3327 struct kvm_coalesced_mmio_zone zone;
3330 if (copy_from_user(&zone, argp, sizeof(zone)))
3332 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3337 struct kvm_irqfd data;
3340 if (copy_from_user(&data, argp, sizeof(data)))
3342 r = kvm_irqfd(kvm, &data);
3345 case KVM_IOEVENTFD: {
3346 struct kvm_ioeventfd data;
3349 if (copy_from_user(&data, argp, sizeof(data)))
3351 r = kvm_ioeventfd(kvm, &data);
3354 #ifdef CONFIG_HAVE_KVM_MSI
3355 case KVM_SIGNAL_MSI: {
3359 if (copy_from_user(&msi, argp, sizeof(msi)))
3361 r = kvm_send_userspace_msi(kvm, &msi);
3365 #ifdef __KVM_HAVE_IRQ_LINE
3366 case KVM_IRQ_LINE_STATUS:
3367 case KVM_IRQ_LINE: {
3368 struct kvm_irq_level irq_event;
3371 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3374 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3375 ioctl == KVM_IRQ_LINE_STATUS);
3380 if (ioctl == KVM_IRQ_LINE_STATUS) {
3381 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3389 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3390 case KVM_SET_GSI_ROUTING: {
3391 struct kvm_irq_routing routing;
3392 struct kvm_irq_routing __user *urouting;
3393 struct kvm_irq_routing_entry *entries = NULL;
3396 if (copy_from_user(&routing, argp, sizeof(routing)))
3399 if (!kvm_arch_can_set_irq_routing(kvm))
3401 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3407 entries = vmalloc(array_size(sizeof(*entries),
3413 if (copy_from_user(entries, urouting->entries,
3414 routing.nr * sizeof(*entries)))
3415 goto out_free_irq_routing;
3417 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3419 out_free_irq_routing:
3423 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3424 case KVM_CREATE_DEVICE: {
3425 struct kvm_create_device cd;
3428 if (copy_from_user(&cd, argp, sizeof(cd)))
3431 r = kvm_ioctl_create_device(kvm, &cd);
3436 if (copy_to_user(argp, &cd, sizeof(cd)))
3442 case KVM_CHECK_EXTENSION:
3443 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3446 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3452 #ifdef CONFIG_KVM_COMPAT
3453 struct compat_kvm_dirty_log {
3457 compat_uptr_t dirty_bitmap; /* one bit per page */
3462 static long kvm_vm_compat_ioctl(struct file *filp,
3463 unsigned int ioctl, unsigned long arg)
3465 struct kvm *kvm = filp->private_data;
3468 if (kvm->mm != current->mm)
3471 case KVM_GET_DIRTY_LOG: {
3472 struct compat_kvm_dirty_log compat_log;
3473 struct kvm_dirty_log log;
3475 if (copy_from_user(&compat_log, (void __user *)arg,
3476 sizeof(compat_log)))
3478 log.slot = compat_log.slot;
3479 log.padding1 = compat_log.padding1;
3480 log.padding2 = compat_log.padding2;
3481 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3483 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3487 r = kvm_vm_ioctl(filp, ioctl, arg);
3493 static struct file_operations kvm_vm_fops = {
3494 .release = kvm_vm_release,
3495 .unlocked_ioctl = kvm_vm_ioctl,
3496 .llseek = noop_llseek,
3497 KVM_COMPAT(kvm_vm_compat_ioctl),
3500 static int kvm_dev_ioctl_create_vm(unsigned long type)
3506 kvm = kvm_create_vm(type);
3508 return PTR_ERR(kvm);
3509 #ifdef CONFIG_KVM_MMIO
3510 r = kvm_coalesced_mmio_init(kvm);
3514 r = get_unused_fd_flags(O_CLOEXEC);
3518 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3526 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3527 * already set, with ->release() being kvm_vm_release(). In error
3528 * cases it will be called by the final fput(file) and will take
3529 * care of doing kvm_put_kvm(kvm).
3531 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3536 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3538 fd_install(r, file);
3546 static long kvm_dev_ioctl(struct file *filp,
3547 unsigned int ioctl, unsigned long arg)
3552 case KVM_GET_API_VERSION:
3555 r = KVM_API_VERSION;
3558 r = kvm_dev_ioctl_create_vm(arg);
3560 case KVM_CHECK_EXTENSION:
3561 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3563 case KVM_GET_VCPU_MMAP_SIZE:
3566 r = PAGE_SIZE; /* struct kvm_run */
3568 r += PAGE_SIZE; /* pio data page */
3570 #ifdef CONFIG_KVM_MMIO
3571 r += PAGE_SIZE; /* coalesced mmio ring page */
3574 case KVM_TRACE_ENABLE:
3575 case KVM_TRACE_PAUSE:
3576 case KVM_TRACE_DISABLE:
3580 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3586 static struct file_operations kvm_chardev_ops = {
3587 .unlocked_ioctl = kvm_dev_ioctl,
3588 .llseek = noop_llseek,
3589 KVM_COMPAT(kvm_dev_ioctl),
3592 static struct miscdevice kvm_dev = {
3598 static void hardware_enable_nolock(void *junk)
3600 int cpu = raw_smp_processor_id();
3603 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3606 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3608 r = kvm_arch_hardware_enable();
3611 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3612 atomic_inc(&hardware_enable_failed);
3613 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3617 static int kvm_starting_cpu(unsigned int cpu)
3619 raw_spin_lock(&kvm_count_lock);
3620 if (kvm_usage_count)
3621 hardware_enable_nolock(NULL);
3622 raw_spin_unlock(&kvm_count_lock);
3626 static void hardware_disable_nolock(void *junk)
3628 int cpu = raw_smp_processor_id();
3630 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3632 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3633 kvm_arch_hardware_disable();
3636 static int kvm_dying_cpu(unsigned int cpu)
3638 raw_spin_lock(&kvm_count_lock);
3639 if (kvm_usage_count)
3640 hardware_disable_nolock(NULL);
3641 raw_spin_unlock(&kvm_count_lock);
3645 static void hardware_disable_all_nolock(void)
3647 BUG_ON(!kvm_usage_count);
3650 if (!kvm_usage_count)
3651 on_each_cpu(hardware_disable_nolock, NULL, 1);
3654 static void hardware_disable_all(void)
3656 raw_spin_lock(&kvm_count_lock);
3657 hardware_disable_all_nolock();
3658 raw_spin_unlock(&kvm_count_lock);
3661 static int hardware_enable_all(void)
3665 raw_spin_lock(&kvm_count_lock);
3668 if (kvm_usage_count == 1) {
3669 atomic_set(&hardware_enable_failed, 0);
3670 on_each_cpu(hardware_enable_nolock, NULL, 1);
3672 if (atomic_read(&hardware_enable_failed)) {
3673 hardware_disable_all_nolock();
3678 raw_spin_unlock(&kvm_count_lock);
3683 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3687 * Some (well, at least mine) BIOSes hang on reboot if
3690 * And Intel TXT required VMX off for all cpu when system shutdown.
3692 pr_info("kvm: exiting hardware virtualization\n");
3693 kvm_rebooting = true;
3694 on_each_cpu(hardware_disable_nolock, NULL, 1);
3698 static struct notifier_block kvm_reboot_notifier = {
3699 .notifier_call = kvm_reboot,
3703 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3707 for (i = 0; i < bus->dev_count; i++) {
3708 struct kvm_io_device *pos = bus->range[i].dev;
3710 kvm_iodevice_destructor(pos);
3715 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3716 const struct kvm_io_range *r2)
3718 gpa_t addr1 = r1->addr;
3719 gpa_t addr2 = r2->addr;
3724 /* If r2->len == 0, match the exact address. If r2->len != 0,
3725 * accept any overlapping write. Any order is acceptable for
3726 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3727 * we process all of them.
3740 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3742 return kvm_io_bus_cmp(p1, p2);
3745 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3746 gpa_t addr, int len)
3748 struct kvm_io_range *range, key;
3751 key = (struct kvm_io_range) {
3756 range = bsearch(&key, bus->range, bus->dev_count,
3757 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3761 off = range - bus->range;
3763 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3769 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3770 struct kvm_io_range *range, const void *val)
3774 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3778 while (idx < bus->dev_count &&
3779 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3780 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3789 /* kvm_io_bus_write - called under kvm->slots_lock */
3790 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3791 int len, const void *val)
3793 struct kvm_io_bus *bus;
3794 struct kvm_io_range range;
3797 range = (struct kvm_io_range) {
3802 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3805 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3806 return r < 0 ? r : 0;
3808 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3810 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3811 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3812 gpa_t addr, int len, const void *val, long cookie)
3814 struct kvm_io_bus *bus;
3815 struct kvm_io_range range;
3817 range = (struct kvm_io_range) {
3822 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3826 /* First try the device referenced by cookie. */
3827 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3828 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3829 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3834 * cookie contained garbage; fall back to search and return the
3835 * correct cookie value.
3837 return __kvm_io_bus_write(vcpu, bus, &range, val);
3840 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3841 struct kvm_io_range *range, void *val)
3845 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3849 while (idx < bus->dev_count &&
3850 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3851 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3860 /* kvm_io_bus_read - called under kvm->slots_lock */
3861 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3864 struct kvm_io_bus *bus;
3865 struct kvm_io_range range;
3868 range = (struct kvm_io_range) {
3873 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3876 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3877 return r < 0 ? r : 0;
3880 /* Caller must hold slots_lock. */
3881 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3882 int len, struct kvm_io_device *dev)
3885 struct kvm_io_bus *new_bus, *bus;
3886 struct kvm_io_range range;
3888 bus = kvm_get_bus(kvm, bus_idx);
3892 /* exclude ioeventfd which is limited by maximum fd */
3893 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3896 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3897 GFP_KERNEL_ACCOUNT);
3901 range = (struct kvm_io_range) {
3907 for (i = 0; i < bus->dev_count; i++)
3908 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3911 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3912 new_bus->dev_count++;
3913 new_bus->range[i] = range;
3914 memcpy(new_bus->range + i + 1, bus->range + i,
3915 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3916 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3917 synchronize_srcu_expedited(&kvm->srcu);
3923 /* Caller must hold slots_lock. */
3924 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3925 struct kvm_io_device *dev)
3928 struct kvm_io_bus *new_bus, *bus;
3930 bus = kvm_get_bus(kvm, bus_idx);
3934 for (i = 0; i < bus->dev_count; i++)
3935 if (bus->range[i].dev == dev) {
3939 if (i == bus->dev_count)
3942 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3943 GFP_KERNEL_ACCOUNT);
3945 pr_err("kvm: failed to shrink bus, removing it completely\n");
3949 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3950 new_bus->dev_count--;
3951 memcpy(new_bus->range + i, bus->range + i + 1,
3952 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3955 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3956 synchronize_srcu_expedited(&kvm->srcu);
3961 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3964 struct kvm_io_bus *bus;
3965 int dev_idx, srcu_idx;
3966 struct kvm_io_device *iodev = NULL;
3968 srcu_idx = srcu_read_lock(&kvm->srcu);
3970 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3974 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3978 iodev = bus->range[dev_idx].dev;
3981 srcu_read_unlock(&kvm->srcu, srcu_idx);
3985 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3987 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3988 int (*get)(void *, u64 *), int (*set)(void *, u64),
3991 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3994 /* The debugfs files are a reference to the kvm struct which
3995 * is still valid when kvm_destroy_vm is called.
3996 * To avoid the race between open and the removal of the debugfs
3997 * directory we test against the users count.
3999 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4002 if (simple_attr_open(inode, file, get,
4003 stat_data->mode & S_IWUGO ? set : NULL,
4005 kvm_put_kvm(stat_data->kvm);
4012 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4014 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4017 simple_attr_release(inode, file);
4018 kvm_put_kvm(stat_data->kvm);
4023 static int vm_stat_get_per_vm(void *data, u64 *val)
4025 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4027 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
4032 static int vm_stat_clear_per_vm(void *data, u64 val)
4034 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4039 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
4044 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
4046 __simple_attr_check_format("%llu\n", 0ull);
4047 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
4048 vm_stat_clear_per_vm, "%llu\n");
4051 static const struct file_operations vm_stat_get_per_vm_fops = {
4052 .owner = THIS_MODULE,
4053 .open = vm_stat_get_per_vm_open,
4054 .release = kvm_debugfs_release,
4055 .read = simple_attr_read,
4056 .write = simple_attr_write,
4057 .llseek = no_llseek,
4060 static int vcpu_stat_get_per_vm(void *data, u64 *val)
4063 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4064 struct kvm_vcpu *vcpu;
4068 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4069 *val += *(u64 *)((void *)vcpu + stat_data->offset);
4074 static int vcpu_stat_clear_per_vm(void *data, u64 val)
4077 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4078 struct kvm_vcpu *vcpu;
4083 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4084 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4089 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4091 __simple_attr_check_format("%llu\n", 0ull);
4092 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4093 vcpu_stat_clear_per_vm, "%llu\n");
4096 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4097 .owner = THIS_MODULE,
4098 .open = vcpu_stat_get_per_vm_open,
4099 .release = kvm_debugfs_release,
4100 .read = simple_attr_read,
4101 .write = simple_attr_write,
4102 .llseek = no_llseek,
4105 static const struct file_operations *stat_fops_per_vm[] = {
4106 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4107 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4110 static int vm_stat_get(void *_offset, u64 *val)
4112 unsigned offset = (long)_offset;
4114 struct kvm_stat_data stat_tmp = {.offset = offset};
4118 mutex_lock(&kvm_lock);
4119 list_for_each_entry(kvm, &vm_list, vm_list) {
4121 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4124 mutex_unlock(&kvm_lock);
4128 static int vm_stat_clear(void *_offset, u64 val)
4130 unsigned offset = (long)_offset;
4132 struct kvm_stat_data stat_tmp = {.offset = offset};
4137 mutex_lock(&kvm_lock);
4138 list_for_each_entry(kvm, &vm_list, vm_list) {
4140 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4142 mutex_unlock(&kvm_lock);
4147 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4149 static int vcpu_stat_get(void *_offset, u64 *val)
4151 unsigned offset = (long)_offset;
4153 struct kvm_stat_data stat_tmp = {.offset = offset};
4157 mutex_lock(&kvm_lock);
4158 list_for_each_entry(kvm, &vm_list, vm_list) {
4160 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4163 mutex_unlock(&kvm_lock);
4167 static int vcpu_stat_clear(void *_offset, u64 val)
4169 unsigned offset = (long)_offset;
4171 struct kvm_stat_data stat_tmp = {.offset = offset};
4176 mutex_lock(&kvm_lock);
4177 list_for_each_entry(kvm, &vm_list, vm_list) {
4179 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4181 mutex_unlock(&kvm_lock);
4186 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4189 static const struct file_operations *stat_fops[] = {
4190 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4191 [KVM_STAT_VM] = &vm_stat_fops,
4194 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4196 struct kobj_uevent_env *env;
4197 unsigned long long created, active;
4199 if (!kvm_dev.this_device || !kvm)
4202 mutex_lock(&kvm_lock);
4203 if (type == KVM_EVENT_CREATE_VM) {
4204 kvm_createvm_count++;
4206 } else if (type == KVM_EVENT_DESTROY_VM) {
4209 created = kvm_createvm_count;
4210 active = kvm_active_vms;
4211 mutex_unlock(&kvm_lock);
4213 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4217 add_uevent_var(env, "CREATED=%llu", created);
4218 add_uevent_var(env, "COUNT=%llu", active);
4220 if (type == KVM_EVENT_CREATE_VM) {
4221 add_uevent_var(env, "EVENT=create");
4222 kvm->userspace_pid = task_pid_nr(current);
4223 } else if (type == KVM_EVENT_DESTROY_VM) {
4224 add_uevent_var(env, "EVENT=destroy");
4226 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4228 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4229 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4232 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4234 add_uevent_var(env, "STATS_PATH=%s", tmp);
4238 /* no need for checks, since we are adding at most only 5 keys */
4239 env->envp[env->envp_idx++] = NULL;
4240 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4244 static void kvm_init_debug(void)
4246 struct kvm_stats_debugfs_item *p;
4248 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4250 kvm_debugfs_num_entries = 0;
4251 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4252 int mode = p->mode ? p->mode : 0644;
4253 debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4254 (void *)(long)p->offset,
4255 stat_fops[p->kind]);
4259 static int kvm_suspend(void)
4261 if (kvm_usage_count)
4262 hardware_disable_nolock(NULL);
4266 static void kvm_resume(void)
4268 if (kvm_usage_count) {
4269 #ifdef CONFIG_LOCKDEP
4270 WARN_ON(lockdep_is_held(&kvm_count_lock));
4272 hardware_enable_nolock(NULL);
4276 static struct syscore_ops kvm_syscore_ops = {
4277 .suspend = kvm_suspend,
4278 .resume = kvm_resume,
4282 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4284 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4287 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4289 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4291 WRITE_ONCE(vcpu->preempted, false);
4292 WRITE_ONCE(vcpu->ready, false);
4294 kvm_arch_sched_in(vcpu, cpu);
4296 kvm_arch_vcpu_load(vcpu, cpu);
4299 static void kvm_sched_out(struct preempt_notifier *pn,
4300 struct task_struct *next)
4302 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4304 if (current->state == TASK_RUNNING) {
4305 WRITE_ONCE(vcpu->preempted, true);
4306 WRITE_ONCE(vcpu->ready, true);
4308 kvm_arch_vcpu_put(vcpu);
4311 static void check_processor_compat(void *rtn)
4313 *(int *)rtn = kvm_arch_check_processor_compat();
4316 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4317 struct module *module)
4322 r = kvm_arch_init(opaque);
4327 * kvm_arch_init makes sure there's at most one caller
4328 * for architectures that support multiple implementations,
4329 * like intel and amd on x86.
4330 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4331 * conflicts in case kvm is already setup for another implementation.
4333 r = kvm_irqfd_init();
4337 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4342 r = kvm_arch_hardware_setup();
4346 for_each_online_cpu(cpu) {
4347 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4352 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4353 kvm_starting_cpu, kvm_dying_cpu);
4356 register_reboot_notifier(&kvm_reboot_notifier);
4358 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4360 vcpu_align = __alignof__(struct kvm_vcpu);
4362 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4364 offsetof(struct kvm_vcpu, arch),
4365 sizeof_field(struct kvm_vcpu, arch),
4367 if (!kvm_vcpu_cache) {
4372 r = kvm_async_pf_init();
4376 kvm_chardev_ops.owner = module;
4377 kvm_vm_fops.owner = module;
4378 kvm_vcpu_fops.owner = module;
4380 r = misc_register(&kvm_dev);
4382 pr_err("kvm: misc device register failed\n");
4386 register_syscore_ops(&kvm_syscore_ops);
4388 kvm_preempt_ops.sched_in = kvm_sched_in;
4389 kvm_preempt_ops.sched_out = kvm_sched_out;
4393 r = kvm_vfio_ops_init();
4399 kvm_async_pf_deinit();
4401 kmem_cache_destroy(kvm_vcpu_cache);
4403 unregister_reboot_notifier(&kvm_reboot_notifier);
4404 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4407 kvm_arch_hardware_unsetup();
4409 free_cpumask_var(cpus_hardware_enabled);
4417 EXPORT_SYMBOL_GPL(kvm_init);
4421 debugfs_remove_recursive(kvm_debugfs_dir);
4422 misc_deregister(&kvm_dev);
4423 kmem_cache_destroy(kvm_vcpu_cache);
4424 kvm_async_pf_deinit();
4425 unregister_syscore_ops(&kvm_syscore_ops);
4426 unregister_reboot_notifier(&kvm_reboot_notifier);
4427 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4428 on_each_cpu(hardware_disable_nolock, NULL, 1);
4429 kvm_arch_hardware_unsetup();
4432 free_cpumask_var(cpus_hardware_enabled);
4433 kvm_vfio_ops_exit();
4435 EXPORT_SYMBOL_GPL(kvm_exit);
4437 struct kvm_vm_worker_thread_context {
4439 struct task_struct *parent;
4440 struct completion init_done;
4441 kvm_vm_thread_fn_t thread_fn;
4446 static int kvm_vm_worker_thread(void *context)
4449 * The init_context is allocated on the stack of the parent thread, so
4450 * we have to locally copy anything that is needed beyond initialization
4452 struct kvm_vm_worker_thread_context *init_context = context;
4453 struct kvm *kvm = init_context->kvm;
4454 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4455 uintptr_t data = init_context->data;
4458 err = kthread_park(current);
4459 /* kthread_park(current) is never supposed to return an error */
4464 err = cgroup_attach_task_all(init_context->parent, current);
4466 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4471 set_user_nice(current, task_nice(init_context->parent));
4474 init_context->err = err;
4475 complete(&init_context->init_done);
4476 init_context = NULL;
4481 /* Wait to be woken up by the spawner before proceeding. */
4484 if (!kthread_should_stop())
4485 err = thread_fn(kvm, data);
4490 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4491 uintptr_t data, const char *name,
4492 struct task_struct **thread_ptr)
4494 struct kvm_vm_worker_thread_context init_context = {};
4495 struct task_struct *thread;
4498 init_context.kvm = kvm;
4499 init_context.parent = current;
4500 init_context.thread_fn = thread_fn;
4501 init_context.data = data;
4502 init_completion(&init_context.init_done);
4504 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4505 "%s-%d", name, task_pid_nr(current));
4507 return PTR_ERR(thread);
4509 /* kthread_run is never supposed to return NULL */
4510 WARN_ON(thread == NULL);
4512 wait_for_completion(&init_context.init_done);
4514 if (!init_context.err)
4515 *thread_ptr = thread;
4517 return init_context.err;