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 * Used to put a reference that was taken on behalf of an object associated
843 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
844 * of the new file descriptor fails and the reference cannot be transferred to
845 * its final owner. In such cases, the caller is still actively using @kvm and
846 * will fail miserably if the refcount unexpectedly hits zero.
848 void kvm_put_kvm_no_destroy(struct kvm *kvm)
850 WARN_ON(refcount_dec_and_test(&kvm->users_count));
852 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
854 static int kvm_vm_release(struct inode *inode, struct file *filp)
856 struct kvm *kvm = filp->private_data;
858 kvm_irqfd_release(kvm);
865 * Allocation size is twice as large as the actual dirty bitmap size.
866 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
868 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
870 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
872 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
873 if (!memslot->dirty_bitmap)
880 * Insert memslot and re-sort memslots based on their GFN,
881 * so binary search could be used to lookup GFN.
882 * Sorting algorithm takes advantage of having initially
883 * sorted array and known changed memslot position.
885 static void update_memslots(struct kvm_memslots *slots,
886 struct kvm_memory_slot *new,
887 enum kvm_mr_change change)
890 int i = slots->id_to_index[id];
891 struct kvm_memory_slot *mslots = slots->memslots;
893 WARN_ON(mslots[i].id != id);
897 WARN_ON(mslots[i].npages || !new->npages);
901 WARN_ON(new->npages || !mslots[i].npages);
907 while (i < KVM_MEM_SLOTS_NUM - 1 &&
908 new->base_gfn <= mslots[i + 1].base_gfn) {
909 if (!mslots[i + 1].npages)
911 mslots[i] = mslots[i + 1];
912 slots->id_to_index[mslots[i].id] = i;
917 * The ">=" is needed when creating a slot with base_gfn == 0,
918 * so that it moves before all those with base_gfn == npages == 0.
920 * On the other hand, if new->npages is zero, the above loop has
921 * already left i pointing to the beginning of the empty part of
922 * mslots, and the ">=" would move the hole backwards in this
923 * case---which is wrong. So skip the loop when deleting a slot.
927 new->base_gfn >= mslots[i - 1].base_gfn) {
928 mslots[i] = mslots[i - 1];
929 slots->id_to_index[mslots[i].id] = i;
933 WARN_ON_ONCE(i != slots->used_slots);
936 slots->id_to_index[mslots[i].id] = i;
939 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
941 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
943 #ifdef __KVM_HAVE_READONLY_MEM
944 valid_flags |= KVM_MEM_READONLY;
947 if (mem->flags & ~valid_flags)
953 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
954 int as_id, struct kvm_memslots *slots)
956 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
957 u64 gen = old_memslots->generation;
959 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
960 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
962 rcu_assign_pointer(kvm->memslots[as_id], slots);
963 synchronize_srcu_expedited(&kvm->srcu);
966 * Increment the new memslot generation a second time, dropping the
967 * update in-progress flag and incrementing then generation based on
968 * the number of address spaces. This provides a unique and easily
969 * identifiable generation number while the memslots are in flux.
971 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
974 * Generations must be unique even across address spaces. We do not need
975 * a global counter for that, instead the generation space is evenly split
976 * across address spaces. For example, with two address spaces, address
977 * space 0 will use generations 0, 2, 4, ... while address space 1 will
978 * use generations 1, 3, 5, ...
980 gen += KVM_ADDRESS_SPACE_NUM;
982 kvm_arch_memslots_updated(kvm, gen);
984 slots->generation = gen;
990 * Allocate some memory and give it an address in the guest physical address
993 * Discontiguous memory is allowed, mostly for framebuffers.
995 * Must be called holding kvm->slots_lock for write.
997 int __kvm_set_memory_region(struct kvm *kvm,
998 const struct kvm_userspace_memory_region *mem)
1002 unsigned long npages;
1003 struct kvm_memory_slot *slot;
1004 struct kvm_memory_slot old, new;
1005 struct kvm_memslots *slots = NULL, *old_memslots;
1007 enum kvm_mr_change change;
1009 r = check_memory_region_flags(mem);
1014 as_id = mem->slot >> 16;
1015 id = (u16)mem->slot;
1017 /* General sanity checks */
1018 if (mem->memory_size & (PAGE_SIZE - 1))
1020 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1022 /* We can read the guest memory with __xxx_user() later on. */
1023 if ((id < KVM_USER_MEM_SLOTS) &&
1024 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1025 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1028 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1030 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1033 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1034 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1035 npages = mem->memory_size >> PAGE_SHIFT;
1037 if (npages > KVM_MEM_MAX_NR_PAGES)
1043 new.base_gfn = base_gfn;
1044 new.npages = npages;
1045 new.flags = mem->flags;
1049 change = KVM_MR_CREATE;
1050 else { /* Modify an existing slot. */
1051 if ((mem->userspace_addr != old.userspace_addr) ||
1052 (npages != old.npages) ||
1053 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1056 if (base_gfn != old.base_gfn)
1057 change = KVM_MR_MOVE;
1058 else if (new.flags != old.flags)
1059 change = KVM_MR_FLAGS_ONLY;
1060 else { /* Nothing to change. */
1069 change = KVM_MR_DELETE;
1074 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1075 /* Check for overlaps */
1077 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1080 if (!((base_gfn + npages <= slot->base_gfn) ||
1081 (base_gfn >= slot->base_gfn + slot->npages)))
1086 /* Free page dirty bitmap if unneeded */
1087 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1088 new.dirty_bitmap = NULL;
1091 if (change == KVM_MR_CREATE) {
1092 new.userspace_addr = mem->userspace_addr;
1094 if (kvm_arch_create_memslot(kvm, &new, npages))
1098 /* Allocate page dirty bitmap if needed */
1099 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1100 if (kvm_create_dirty_bitmap(&new) < 0)
1104 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1107 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1109 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1110 slot = id_to_memslot(slots, id);
1111 slot->flags |= KVM_MEMSLOT_INVALID;
1113 old_memslots = install_new_memslots(kvm, as_id, slots);
1115 /* From this point no new shadow pages pointing to a deleted,
1116 * or moved, memslot will be created.
1118 * validation of sp->gfn happens in:
1119 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1120 * - kvm_is_visible_gfn (mmu_check_roots)
1122 kvm_arch_flush_shadow_memslot(kvm, slot);
1125 * We can re-use the old_memslots from above, the only difference
1126 * from the currently installed memslots is the invalid flag. This
1127 * will get overwritten by update_memslots anyway.
1129 slots = old_memslots;
1132 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1136 /* actual memory is freed via old in kvm_free_memslot below */
1137 if (change == KVM_MR_DELETE) {
1138 new.dirty_bitmap = NULL;
1139 memset(&new.arch, 0, sizeof(new.arch));
1142 update_memslots(slots, &new, change);
1143 old_memslots = install_new_memslots(kvm, as_id, slots);
1145 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1147 kvm_free_memslot(kvm, &old, &new);
1148 kvfree(old_memslots);
1154 kvm_free_memslot(kvm, &new, &old);
1158 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1160 int kvm_set_memory_region(struct kvm *kvm,
1161 const struct kvm_userspace_memory_region *mem)
1165 mutex_lock(&kvm->slots_lock);
1166 r = __kvm_set_memory_region(kvm, mem);
1167 mutex_unlock(&kvm->slots_lock);
1170 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1172 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1173 struct kvm_userspace_memory_region *mem)
1175 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1178 return kvm_set_memory_region(kvm, mem);
1181 int kvm_get_dirty_log(struct kvm *kvm,
1182 struct kvm_dirty_log *log, int *is_dirty)
1184 struct kvm_memslots *slots;
1185 struct kvm_memory_slot *memslot;
1188 unsigned long any = 0;
1190 as_id = log->slot >> 16;
1191 id = (u16)log->slot;
1192 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1195 slots = __kvm_memslots(kvm, as_id);
1196 memslot = id_to_memslot(slots, id);
1197 if (!memslot->dirty_bitmap)
1200 n = kvm_dirty_bitmap_bytes(memslot);
1202 for (i = 0; !any && i < n/sizeof(long); ++i)
1203 any = memslot->dirty_bitmap[i];
1205 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1212 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1214 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1216 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1217 * and reenable dirty page tracking for the corresponding pages.
1218 * @kvm: pointer to kvm instance
1219 * @log: slot id and address to which we copy the log
1220 * @flush: true if TLB flush is needed by caller
1222 * We need to keep it in mind that VCPU threads can write to the bitmap
1223 * concurrently. So, to avoid losing track of dirty pages we keep the
1226 * 1. Take a snapshot of the bit and clear it if needed.
1227 * 2. Write protect the corresponding page.
1228 * 3. Copy the snapshot to the userspace.
1229 * 4. Upon return caller flushes TLB's if needed.
1231 * Between 2 and 4, the guest may write to the page using the remaining TLB
1232 * entry. This is not a problem because the page is reported dirty using
1233 * the snapshot taken before and step 4 ensures that writes done after
1234 * exiting to userspace will be logged for the next call.
1237 int kvm_get_dirty_log_protect(struct kvm *kvm,
1238 struct kvm_dirty_log *log, bool *flush)
1240 struct kvm_memslots *slots;
1241 struct kvm_memory_slot *memslot;
1244 unsigned long *dirty_bitmap;
1245 unsigned long *dirty_bitmap_buffer;
1247 as_id = log->slot >> 16;
1248 id = (u16)log->slot;
1249 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1252 slots = __kvm_memslots(kvm, as_id);
1253 memslot = id_to_memslot(slots, id);
1255 dirty_bitmap = memslot->dirty_bitmap;
1259 n = kvm_dirty_bitmap_bytes(memslot);
1261 if (kvm->manual_dirty_log_protect) {
1263 * Unlike kvm_get_dirty_log, we always return false in *flush,
1264 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1265 * is some code duplication between this function and
1266 * kvm_get_dirty_log, but hopefully all architecture
1267 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1268 * can be eliminated.
1270 dirty_bitmap_buffer = dirty_bitmap;
1272 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1273 memset(dirty_bitmap_buffer, 0, n);
1275 spin_lock(&kvm->mmu_lock);
1276 for (i = 0; i < n / sizeof(long); i++) {
1280 if (!dirty_bitmap[i])
1284 mask = xchg(&dirty_bitmap[i], 0);
1285 dirty_bitmap_buffer[i] = mask;
1287 offset = i * BITS_PER_LONG;
1288 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1291 spin_unlock(&kvm->mmu_lock);
1294 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1298 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1301 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1302 * and reenable dirty page tracking for the corresponding pages.
1303 * @kvm: pointer to kvm instance
1304 * @log: slot id and address from which to fetch the bitmap of dirty pages
1305 * @flush: true if TLB flush is needed by caller
1307 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1308 struct kvm_clear_dirty_log *log, bool *flush)
1310 struct kvm_memslots *slots;
1311 struct kvm_memory_slot *memslot;
1315 unsigned long *dirty_bitmap;
1316 unsigned long *dirty_bitmap_buffer;
1318 as_id = log->slot >> 16;
1319 id = (u16)log->slot;
1320 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1323 if (log->first_page & 63)
1326 slots = __kvm_memslots(kvm, as_id);
1327 memslot = id_to_memslot(slots, id);
1329 dirty_bitmap = memslot->dirty_bitmap;
1333 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1335 if (log->first_page > memslot->npages ||
1336 log->num_pages > memslot->npages - log->first_page ||
1337 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1341 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1342 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1345 spin_lock(&kvm->mmu_lock);
1346 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1347 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1348 i++, offset += BITS_PER_LONG) {
1349 unsigned long mask = *dirty_bitmap_buffer++;
1350 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1354 mask &= atomic_long_fetch_andnot(mask, p);
1357 * mask contains the bits that really have been cleared. This
1358 * never includes any bits beyond the length of the memslot (if
1359 * the length is not aligned to 64 pages), therefore it is not
1360 * a problem if userspace sets them in log->dirty_bitmap.
1364 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1368 spin_unlock(&kvm->mmu_lock);
1372 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1375 bool kvm_largepages_enabled(void)
1377 return largepages_enabled;
1380 void kvm_disable_largepages(void)
1382 largepages_enabled = false;
1384 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1386 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1388 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1390 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1392 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1394 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1397 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1399 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1401 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1402 memslot->flags & KVM_MEMSLOT_INVALID)
1407 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1409 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1411 struct vm_area_struct *vma;
1412 unsigned long addr, size;
1416 addr = gfn_to_hva(kvm, gfn);
1417 if (kvm_is_error_hva(addr))
1420 down_read(¤t->mm->mmap_sem);
1421 vma = find_vma(current->mm, addr);
1425 size = vma_kernel_pagesize(vma);
1428 up_read(¤t->mm->mmap_sem);
1433 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1435 return slot->flags & KVM_MEM_READONLY;
1438 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1439 gfn_t *nr_pages, bool write)
1441 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1442 return KVM_HVA_ERR_BAD;
1444 if (memslot_is_readonly(slot) && write)
1445 return KVM_HVA_ERR_RO_BAD;
1448 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1450 return __gfn_to_hva_memslot(slot, gfn);
1453 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1456 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1459 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1462 return gfn_to_hva_many(slot, gfn, NULL);
1464 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1466 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1468 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1470 EXPORT_SYMBOL_GPL(gfn_to_hva);
1472 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1474 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1476 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1479 * Return the hva of a @gfn and the R/W attribute if possible.
1481 * @slot: the kvm_memory_slot which contains @gfn
1482 * @gfn: the gfn to be translated
1483 * @writable: used to return the read/write attribute of the @slot if the hva
1484 * is valid and @writable is not NULL
1486 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1487 gfn_t gfn, bool *writable)
1489 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1491 if (!kvm_is_error_hva(hva) && writable)
1492 *writable = !memslot_is_readonly(slot);
1497 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1499 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1501 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1504 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1506 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1508 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1511 static inline int check_user_page_hwpoison(unsigned long addr)
1513 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1515 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1516 return rc == -EHWPOISON;
1520 * The fast path to get the writable pfn which will be stored in @pfn,
1521 * true indicates success, otherwise false is returned. It's also the
1522 * only part that runs if we can are in atomic context.
1524 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1525 bool *writable, kvm_pfn_t *pfn)
1527 struct page *page[1];
1531 * Fast pin a writable pfn only if it is a write fault request
1532 * or the caller allows to map a writable pfn for a read fault
1535 if (!(write_fault || writable))
1538 npages = __get_user_pages_fast(addr, 1, 1, page);
1540 *pfn = page_to_pfn(page[0]);
1551 * The slow path to get the pfn of the specified host virtual address,
1552 * 1 indicates success, -errno is returned if error is detected.
1554 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1555 bool *writable, kvm_pfn_t *pfn)
1557 unsigned int flags = FOLL_HWPOISON;
1564 *writable = write_fault;
1567 flags |= FOLL_WRITE;
1569 flags |= FOLL_NOWAIT;
1571 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1575 /* map read fault as writable if possible */
1576 if (unlikely(!write_fault) && writable) {
1579 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1585 *pfn = page_to_pfn(page);
1589 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1591 if (unlikely(!(vma->vm_flags & VM_READ)))
1594 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1600 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1601 unsigned long addr, bool *async,
1602 bool write_fault, bool *writable,
1608 r = follow_pfn(vma, addr, &pfn);
1611 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1612 * not call the fault handler, so do it here.
1614 bool unlocked = false;
1615 r = fixup_user_fault(current, current->mm, addr,
1616 (write_fault ? FAULT_FLAG_WRITE : 0),
1623 r = follow_pfn(vma, addr, &pfn);
1633 * Get a reference here because callers of *hva_to_pfn* and
1634 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1635 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1636 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1637 * simply do nothing for reserved pfns.
1639 * Whoever called remap_pfn_range is also going to call e.g.
1640 * unmap_mapping_range before the underlying pages are freed,
1641 * causing a call to our MMU notifier.
1650 * Pin guest page in memory and return its pfn.
1651 * @addr: host virtual address which maps memory to the guest
1652 * @atomic: whether this function can sleep
1653 * @async: whether this function need to wait IO complete if the
1654 * host page is not in the memory
1655 * @write_fault: whether we should get a writable host page
1656 * @writable: whether it allows to map a writable host page for !@write_fault
1658 * The function will map a writable host page for these two cases:
1659 * 1): @write_fault = true
1660 * 2): @write_fault = false && @writable, @writable will tell the caller
1661 * whether the mapping is writable.
1663 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1664 bool write_fault, bool *writable)
1666 struct vm_area_struct *vma;
1670 /* we can do it either atomically or asynchronously, not both */
1671 BUG_ON(atomic && async);
1673 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1677 return KVM_PFN_ERR_FAULT;
1679 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1683 down_read(¤t->mm->mmap_sem);
1684 if (npages == -EHWPOISON ||
1685 (!async && check_user_page_hwpoison(addr))) {
1686 pfn = KVM_PFN_ERR_HWPOISON;
1691 vma = find_vma_intersection(current->mm, addr, addr + 1);
1694 pfn = KVM_PFN_ERR_FAULT;
1695 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1696 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1700 pfn = KVM_PFN_ERR_FAULT;
1702 if (async && vma_is_valid(vma, write_fault))
1704 pfn = KVM_PFN_ERR_FAULT;
1707 up_read(¤t->mm->mmap_sem);
1711 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1712 bool atomic, bool *async, bool write_fault,
1715 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1717 if (addr == KVM_HVA_ERR_RO_BAD) {
1720 return KVM_PFN_ERR_RO_FAULT;
1723 if (kvm_is_error_hva(addr)) {
1726 return KVM_PFN_NOSLOT;
1729 /* Do not map writable pfn in the readonly memslot. */
1730 if (writable && memslot_is_readonly(slot)) {
1735 return hva_to_pfn(addr, atomic, async, write_fault,
1738 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1740 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1743 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1744 write_fault, writable);
1746 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1748 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1750 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1752 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1754 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1756 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1758 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1760 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1762 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1764 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1766 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1768 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1770 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1772 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1774 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1776 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1778 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1780 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1782 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1784 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1785 struct page **pages, int nr_pages)
1790 addr = gfn_to_hva_many(slot, gfn, &entry);
1791 if (kvm_is_error_hva(addr))
1794 if (entry < nr_pages)
1797 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1799 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1801 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1803 if (is_error_noslot_pfn(pfn))
1804 return KVM_ERR_PTR_BAD_PAGE;
1806 if (kvm_is_reserved_pfn(pfn)) {
1808 return KVM_ERR_PTR_BAD_PAGE;
1811 return pfn_to_page(pfn);
1814 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1818 pfn = gfn_to_pfn(kvm, gfn);
1820 return kvm_pfn_to_page(pfn);
1822 EXPORT_SYMBOL_GPL(gfn_to_page);
1824 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1825 struct kvm_host_map *map)
1829 struct page *page = KVM_UNMAPPED_PAGE;
1834 pfn = gfn_to_pfn_memslot(slot, gfn);
1835 if (is_error_noslot_pfn(pfn))
1838 if (pfn_valid(pfn)) {
1839 page = pfn_to_page(pfn);
1841 #ifdef CONFIG_HAS_IOMEM
1843 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1858 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1860 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1862 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1864 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1873 if (map->page != KVM_UNMAPPED_PAGE)
1875 #ifdef CONFIG_HAS_IOMEM
1881 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1882 kvm_release_pfn_dirty(map->pfn);
1884 kvm_release_pfn_clean(map->pfn);
1890 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1892 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1896 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1898 return kvm_pfn_to_page(pfn);
1900 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1902 void kvm_release_page_clean(struct page *page)
1904 WARN_ON(is_error_page(page));
1906 kvm_release_pfn_clean(page_to_pfn(page));
1908 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1910 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1912 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1913 put_page(pfn_to_page(pfn));
1915 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1917 void kvm_release_page_dirty(struct page *page)
1919 WARN_ON(is_error_page(page));
1921 kvm_release_pfn_dirty(page_to_pfn(page));
1923 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1925 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1927 kvm_set_pfn_dirty(pfn);
1928 kvm_release_pfn_clean(pfn);
1930 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1932 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1934 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) {
1935 struct page *page = pfn_to_page(pfn);
1940 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1942 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1944 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1945 mark_page_accessed(pfn_to_page(pfn));
1947 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1949 void kvm_get_pfn(kvm_pfn_t pfn)
1951 if (!kvm_is_reserved_pfn(pfn))
1952 get_page(pfn_to_page(pfn));
1954 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1956 static int next_segment(unsigned long len, int offset)
1958 if (len > PAGE_SIZE - offset)
1959 return PAGE_SIZE - offset;
1964 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1965 void *data, int offset, int len)
1970 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1971 if (kvm_is_error_hva(addr))
1973 r = __copy_from_user(data, (void __user *)addr + offset, len);
1979 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1982 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1984 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1986 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1988 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1989 int offset, int len)
1991 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1993 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1995 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1997 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1999 gfn_t gfn = gpa >> PAGE_SHIFT;
2001 int offset = offset_in_page(gpa);
2004 while ((seg = next_segment(len, offset)) != 0) {
2005 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2015 EXPORT_SYMBOL_GPL(kvm_read_guest);
2017 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2019 gfn_t gfn = gpa >> PAGE_SHIFT;
2021 int offset = offset_in_page(gpa);
2024 while ((seg = next_segment(len, offset)) != 0) {
2025 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2035 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2037 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2038 void *data, int offset, unsigned long len)
2043 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2044 if (kvm_is_error_hva(addr))
2046 pagefault_disable();
2047 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2054 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
2057 gfn_t gfn = gpa >> PAGE_SHIFT;
2058 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2059 int offset = offset_in_page(gpa);
2061 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2063 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
2065 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2066 void *data, unsigned long len)
2068 gfn_t gfn = gpa >> PAGE_SHIFT;
2069 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2070 int offset = offset_in_page(gpa);
2072 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2074 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2076 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2077 const void *data, int offset, int len)
2082 addr = gfn_to_hva_memslot(memslot, gfn);
2083 if (kvm_is_error_hva(addr))
2085 r = __copy_to_user((void __user *)addr + offset, data, len);
2088 mark_page_dirty_in_slot(memslot, gfn);
2092 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2093 const void *data, int offset, int len)
2095 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2097 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2099 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2101 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2102 const void *data, int offset, int len)
2104 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2106 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2108 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2110 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2113 gfn_t gfn = gpa >> PAGE_SHIFT;
2115 int offset = offset_in_page(gpa);
2118 while ((seg = next_segment(len, offset)) != 0) {
2119 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2129 EXPORT_SYMBOL_GPL(kvm_write_guest);
2131 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2134 gfn_t gfn = gpa >> PAGE_SHIFT;
2136 int offset = offset_in_page(gpa);
2139 while ((seg = next_segment(len, offset)) != 0) {
2140 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2150 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2152 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2153 struct gfn_to_hva_cache *ghc,
2154 gpa_t gpa, unsigned long len)
2156 int offset = offset_in_page(gpa);
2157 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2158 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2159 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2160 gfn_t nr_pages_avail;
2161 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2164 ghc->generation = slots->generation;
2166 ghc->hva = KVM_HVA_ERR_BAD;
2169 * If the requested region crosses two memslots, we still
2170 * verify that the entire region is valid here.
2172 while (!r && start_gfn <= end_gfn) {
2173 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2174 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2176 if (kvm_is_error_hva(ghc->hva))
2178 start_gfn += nr_pages_avail;
2181 /* Use the slow path for cross page reads and writes. */
2182 if (!r && nr_pages_needed == 1)
2185 ghc->memslot = NULL;
2190 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2191 gpa_t gpa, unsigned long len)
2193 struct kvm_memslots *slots = kvm_memslots(kvm);
2194 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2196 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2198 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2199 void *data, unsigned int offset,
2202 struct kvm_memslots *slots = kvm_memslots(kvm);
2204 gpa_t gpa = ghc->gpa + offset;
2206 BUG_ON(len + offset > ghc->len);
2208 if (slots->generation != ghc->generation)
2209 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2211 if (unlikely(!ghc->memslot))
2212 return kvm_write_guest(kvm, gpa, data, len);
2214 if (kvm_is_error_hva(ghc->hva))
2217 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2220 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2224 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2226 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2227 void *data, unsigned long len)
2229 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2231 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2233 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2234 void *data, unsigned long len)
2236 struct kvm_memslots *slots = kvm_memslots(kvm);
2239 BUG_ON(len > ghc->len);
2241 if (slots->generation != ghc->generation)
2242 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2244 if (unlikely(!ghc->memslot))
2245 return kvm_read_guest(kvm, ghc->gpa, data, len);
2247 if (kvm_is_error_hva(ghc->hva))
2250 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2256 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2258 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2260 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2262 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2264 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2266 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2268 gfn_t gfn = gpa >> PAGE_SHIFT;
2270 int offset = offset_in_page(gpa);
2273 while ((seg = next_segment(len, offset)) != 0) {
2274 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2283 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2285 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2288 if (memslot && memslot->dirty_bitmap) {
2289 unsigned long rel_gfn = gfn - memslot->base_gfn;
2291 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2295 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2297 struct kvm_memory_slot *memslot;
2299 memslot = gfn_to_memslot(kvm, gfn);
2300 mark_page_dirty_in_slot(memslot, gfn);
2302 EXPORT_SYMBOL_GPL(mark_page_dirty);
2304 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2306 struct kvm_memory_slot *memslot;
2308 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2309 mark_page_dirty_in_slot(memslot, gfn);
2311 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2313 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2315 if (!vcpu->sigset_active)
2319 * This does a lockless modification of ->real_blocked, which is fine
2320 * because, only current can change ->real_blocked and all readers of
2321 * ->real_blocked don't care as long ->real_blocked is always a subset
2324 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2327 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2329 if (!vcpu->sigset_active)
2332 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2333 sigemptyset(¤t->real_blocked);
2336 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2338 unsigned int old, val, grow, grow_start;
2340 old = val = vcpu->halt_poll_ns;
2341 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2342 grow = READ_ONCE(halt_poll_ns_grow);
2347 if (val < grow_start)
2350 if (val > halt_poll_ns)
2353 vcpu->halt_poll_ns = val;
2355 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2358 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2360 unsigned int old, val, shrink;
2362 old = val = vcpu->halt_poll_ns;
2363 shrink = READ_ONCE(halt_poll_ns_shrink);
2369 vcpu->halt_poll_ns = val;
2370 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2373 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2376 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2378 if (kvm_arch_vcpu_runnable(vcpu)) {
2379 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2382 if (kvm_cpu_has_pending_timer(vcpu))
2384 if (signal_pending(current))
2389 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2394 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2396 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2399 DECLARE_SWAITQUEUE(wait);
2400 bool waited = false;
2403 kvm_arch_vcpu_blocking(vcpu);
2405 start = cur = ktime_get();
2406 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2407 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2409 ++vcpu->stat.halt_attempted_poll;
2412 * This sets KVM_REQ_UNHALT if an interrupt
2415 if (kvm_vcpu_check_block(vcpu) < 0) {
2416 ++vcpu->stat.halt_successful_poll;
2417 if (!vcpu_valid_wakeup(vcpu))
2418 ++vcpu->stat.halt_poll_invalid;
2422 } while (single_task_running() && ktime_before(cur, stop));
2426 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2428 if (kvm_vcpu_check_block(vcpu) < 0)
2435 finish_swait(&vcpu->wq, &wait);
2438 kvm_arch_vcpu_unblocking(vcpu);
2439 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2441 if (!kvm_arch_no_poll(vcpu)) {
2442 if (!vcpu_valid_wakeup(vcpu)) {
2443 shrink_halt_poll_ns(vcpu);
2444 } else if (halt_poll_ns) {
2445 if (block_ns <= vcpu->halt_poll_ns)
2447 /* we had a long block, shrink polling */
2448 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2449 shrink_halt_poll_ns(vcpu);
2450 /* we had a short halt and our poll time is too small */
2451 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2452 block_ns < halt_poll_ns)
2453 grow_halt_poll_ns(vcpu);
2455 vcpu->halt_poll_ns = 0;
2459 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2460 kvm_arch_vcpu_block_finish(vcpu);
2462 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2464 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2466 struct swait_queue_head *wqp;
2468 wqp = kvm_arch_vcpu_wq(vcpu);
2469 if (swq_has_sleeper(wqp)) {
2471 WRITE_ONCE(vcpu->ready, true);
2472 ++vcpu->stat.halt_wakeup;
2478 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2482 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2484 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2487 int cpu = vcpu->cpu;
2489 if (kvm_vcpu_wake_up(vcpu))
2493 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2494 if (kvm_arch_vcpu_should_kick(vcpu))
2495 smp_send_reschedule(cpu);
2498 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2499 #endif /* !CONFIG_S390 */
2501 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2504 struct task_struct *task = NULL;
2508 pid = rcu_dereference(target->pid);
2510 task = get_pid_task(pid, PIDTYPE_PID);
2514 ret = yield_to(task, 1);
2515 put_task_struct(task);
2519 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2522 * Helper that checks whether a VCPU is eligible for directed yield.
2523 * Most eligible candidate to yield is decided by following heuristics:
2525 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2526 * (preempted lock holder), indicated by @in_spin_loop.
2527 * Set at the beiginning and cleared at the end of interception/PLE handler.
2529 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2530 * chance last time (mostly it has become eligible now since we have probably
2531 * yielded to lockholder in last iteration. This is done by toggling
2532 * @dy_eligible each time a VCPU checked for eligibility.)
2534 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2535 * to preempted lock-holder could result in wrong VCPU selection and CPU
2536 * burning. Giving priority for a potential lock-holder increases lock
2539 * Since algorithm is based on heuristics, accessing another VCPU data without
2540 * locking does not harm. It may result in trying to yield to same VCPU, fail
2541 * and continue with next VCPU and so on.
2543 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2545 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2548 eligible = !vcpu->spin_loop.in_spin_loop ||
2549 vcpu->spin_loop.dy_eligible;
2551 if (vcpu->spin_loop.in_spin_loop)
2552 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2561 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2562 * a vcpu_load/vcpu_put pair. However, for most architectures
2563 * kvm_arch_vcpu_runnable does not require vcpu_load.
2565 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2567 return kvm_arch_vcpu_runnable(vcpu);
2570 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2572 if (kvm_arch_dy_runnable(vcpu))
2575 #ifdef CONFIG_KVM_ASYNC_PF
2576 if (!list_empty_careful(&vcpu->async_pf.done))
2583 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2585 struct kvm *kvm = me->kvm;
2586 struct kvm_vcpu *vcpu;
2587 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2593 kvm_vcpu_set_in_spin_loop(me, true);
2595 * We boost the priority of a VCPU that is runnable but not
2596 * currently running, because it got preempted by something
2597 * else and called schedule in __vcpu_run. Hopefully that
2598 * VCPU is holding the lock that we need and will release it.
2599 * We approximate round-robin by starting at the last boosted VCPU.
2601 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2602 kvm_for_each_vcpu(i, vcpu, kvm) {
2603 if (!pass && i <= last_boosted_vcpu) {
2604 i = last_boosted_vcpu;
2606 } else if (pass && i > last_boosted_vcpu)
2608 if (!READ_ONCE(vcpu->ready))
2612 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2614 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2615 !kvm_arch_vcpu_in_kernel(vcpu))
2617 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2620 yielded = kvm_vcpu_yield_to(vcpu);
2622 kvm->last_boosted_vcpu = i;
2624 } else if (yielded < 0) {
2631 kvm_vcpu_set_in_spin_loop(me, false);
2633 /* Ensure vcpu is not eligible during next spinloop */
2634 kvm_vcpu_set_dy_eligible(me, false);
2636 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2638 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2640 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2643 if (vmf->pgoff == 0)
2644 page = virt_to_page(vcpu->run);
2646 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2647 page = virt_to_page(vcpu->arch.pio_data);
2649 #ifdef CONFIG_KVM_MMIO
2650 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2651 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2654 return kvm_arch_vcpu_fault(vcpu, vmf);
2660 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2661 .fault = kvm_vcpu_fault,
2664 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2666 vma->vm_ops = &kvm_vcpu_vm_ops;
2670 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2672 struct kvm_vcpu *vcpu = filp->private_data;
2674 debugfs_remove_recursive(vcpu->debugfs_dentry);
2675 kvm_put_kvm(vcpu->kvm);
2679 static struct file_operations kvm_vcpu_fops = {
2680 .release = kvm_vcpu_release,
2681 .unlocked_ioctl = kvm_vcpu_ioctl,
2682 .mmap = kvm_vcpu_mmap,
2683 .llseek = noop_llseek,
2684 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2688 * Allocates an inode for the vcpu.
2690 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2692 char name[8 + 1 + ITOA_MAX_LEN + 1];
2694 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2695 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2698 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2700 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2701 char dir_name[ITOA_MAX_LEN * 2];
2703 if (!debugfs_initialized())
2706 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2707 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2708 vcpu->kvm->debugfs_dentry);
2710 kvm_arch_create_vcpu_debugfs(vcpu);
2715 * Creates some virtual cpus. Good luck creating more than one.
2717 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2720 struct kvm_vcpu *vcpu;
2722 if (id >= KVM_MAX_VCPU_ID)
2725 mutex_lock(&kvm->lock);
2726 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2727 mutex_unlock(&kvm->lock);
2731 kvm->created_vcpus++;
2732 mutex_unlock(&kvm->lock);
2734 vcpu = kvm_arch_vcpu_create(kvm, id);
2737 goto vcpu_decrement;
2740 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2742 r = kvm_arch_vcpu_setup(vcpu);
2746 kvm_create_vcpu_debugfs(vcpu);
2748 mutex_lock(&kvm->lock);
2749 if (kvm_get_vcpu_by_id(kvm, id)) {
2751 goto unlock_vcpu_destroy;
2754 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
2755 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
2757 /* Now it's all set up, let userspace reach it */
2759 r = create_vcpu_fd(vcpu);
2761 kvm_put_kvm_no_destroy(kvm);
2762 goto unlock_vcpu_destroy;
2765 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
2768 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2769 * before kvm->online_vcpu's incremented value.
2772 atomic_inc(&kvm->online_vcpus);
2774 mutex_unlock(&kvm->lock);
2775 kvm_arch_vcpu_postcreate(vcpu);
2778 unlock_vcpu_destroy:
2779 mutex_unlock(&kvm->lock);
2780 debugfs_remove_recursive(vcpu->debugfs_dentry);
2782 kvm_arch_vcpu_destroy(vcpu);
2784 mutex_lock(&kvm->lock);
2785 kvm->created_vcpus--;
2786 mutex_unlock(&kvm->lock);
2790 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2793 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2794 vcpu->sigset_active = 1;
2795 vcpu->sigset = *sigset;
2797 vcpu->sigset_active = 0;
2801 static long kvm_vcpu_ioctl(struct file *filp,
2802 unsigned int ioctl, unsigned long arg)
2804 struct kvm_vcpu *vcpu = filp->private_data;
2805 void __user *argp = (void __user *)arg;
2807 struct kvm_fpu *fpu = NULL;
2808 struct kvm_sregs *kvm_sregs = NULL;
2810 if (vcpu->kvm->mm != current->mm)
2813 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2817 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2818 * execution; mutex_lock() would break them.
2820 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2821 if (r != -ENOIOCTLCMD)
2824 if (mutex_lock_killable(&vcpu->mutex))
2832 oldpid = rcu_access_pointer(vcpu->pid);
2833 if (unlikely(oldpid != task_pid(current))) {
2834 /* The thread running this VCPU changed. */
2837 r = kvm_arch_vcpu_run_pid_change(vcpu);
2841 newpid = get_task_pid(current, PIDTYPE_PID);
2842 rcu_assign_pointer(vcpu->pid, newpid);
2847 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2848 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2851 case KVM_GET_REGS: {
2852 struct kvm_regs *kvm_regs;
2855 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2858 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2862 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2869 case KVM_SET_REGS: {
2870 struct kvm_regs *kvm_regs;
2873 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2874 if (IS_ERR(kvm_regs)) {
2875 r = PTR_ERR(kvm_regs);
2878 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2882 case KVM_GET_SREGS: {
2883 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2884 GFP_KERNEL_ACCOUNT);
2888 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2892 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2897 case KVM_SET_SREGS: {
2898 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2899 if (IS_ERR(kvm_sregs)) {
2900 r = PTR_ERR(kvm_sregs);
2904 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2907 case KVM_GET_MP_STATE: {
2908 struct kvm_mp_state mp_state;
2910 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2914 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2919 case KVM_SET_MP_STATE: {
2920 struct kvm_mp_state mp_state;
2923 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2925 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2928 case KVM_TRANSLATE: {
2929 struct kvm_translation tr;
2932 if (copy_from_user(&tr, argp, sizeof(tr)))
2934 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2938 if (copy_to_user(argp, &tr, sizeof(tr)))
2943 case KVM_SET_GUEST_DEBUG: {
2944 struct kvm_guest_debug dbg;
2947 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2949 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2952 case KVM_SET_SIGNAL_MASK: {
2953 struct kvm_signal_mask __user *sigmask_arg = argp;
2954 struct kvm_signal_mask kvm_sigmask;
2955 sigset_t sigset, *p;
2960 if (copy_from_user(&kvm_sigmask, argp,
2961 sizeof(kvm_sigmask)))
2964 if (kvm_sigmask.len != sizeof(sigset))
2967 if (copy_from_user(&sigset, sigmask_arg->sigset,
2972 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2976 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2980 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2984 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2990 fpu = memdup_user(argp, sizeof(*fpu));
2996 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3000 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3003 mutex_unlock(&vcpu->mutex);
3009 #ifdef CONFIG_KVM_COMPAT
3010 static long kvm_vcpu_compat_ioctl(struct file *filp,
3011 unsigned int ioctl, unsigned long arg)
3013 struct kvm_vcpu *vcpu = filp->private_data;
3014 void __user *argp = compat_ptr(arg);
3017 if (vcpu->kvm->mm != current->mm)
3021 case KVM_SET_SIGNAL_MASK: {
3022 struct kvm_signal_mask __user *sigmask_arg = argp;
3023 struct kvm_signal_mask kvm_sigmask;
3028 if (copy_from_user(&kvm_sigmask, argp,
3029 sizeof(kvm_sigmask)))
3032 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3035 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3037 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3039 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3043 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3051 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3053 struct kvm_device *dev = filp->private_data;
3056 return dev->ops->mmap(dev, vma);
3061 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3062 int (*accessor)(struct kvm_device *dev,
3063 struct kvm_device_attr *attr),
3066 struct kvm_device_attr attr;
3071 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3074 return accessor(dev, &attr);
3077 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3080 struct kvm_device *dev = filp->private_data;
3082 if (dev->kvm->mm != current->mm)
3086 case KVM_SET_DEVICE_ATTR:
3087 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3088 case KVM_GET_DEVICE_ATTR:
3089 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3090 case KVM_HAS_DEVICE_ATTR:
3091 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3093 if (dev->ops->ioctl)
3094 return dev->ops->ioctl(dev, ioctl, arg);
3100 static int kvm_device_release(struct inode *inode, struct file *filp)
3102 struct kvm_device *dev = filp->private_data;
3103 struct kvm *kvm = dev->kvm;
3105 if (dev->ops->release) {
3106 mutex_lock(&kvm->lock);
3107 list_del(&dev->vm_node);
3108 dev->ops->release(dev);
3109 mutex_unlock(&kvm->lock);
3116 static const struct file_operations kvm_device_fops = {
3117 .unlocked_ioctl = kvm_device_ioctl,
3118 .release = kvm_device_release,
3119 KVM_COMPAT(kvm_device_ioctl),
3120 .mmap = kvm_device_mmap,
3123 struct kvm_device *kvm_device_from_filp(struct file *filp)
3125 if (filp->f_op != &kvm_device_fops)
3128 return filp->private_data;
3131 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3132 #ifdef CONFIG_KVM_MPIC
3133 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3134 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3138 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3140 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3143 if (kvm_device_ops_table[type] != NULL)
3146 kvm_device_ops_table[type] = ops;
3150 void kvm_unregister_device_ops(u32 type)
3152 if (kvm_device_ops_table[type] != NULL)
3153 kvm_device_ops_table[type] = NULL;
3156 static int kvm_ioctl_create_device(struct kvm *kvm,
3157 struct kvm_create_device *cd)
3159 const struct kvm_device_ops *ops = NULL;
3160 struct kvm_device *dev;
3161 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3165 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3168 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3169 ops = kvm_device_ops_table[type];
3176 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3183 mutex_lock(&kvm->lock);
3184 ret = ops->create(dev, type);
3186 mutex_unlock(&kvm->lock);
3190 list_add(&dev->vm_node, &kvm->devices);
3191 mutex_unlock(&kvm->lock);
3197 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3199 kvm_put_kvm_no_destroy(kvm);
3200 mutex_lock(&kvm->lock);
3201 list_del(&dev->vm_node);
3202 mutex_unlock(&kvm->lock);
3211 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3214 case KVM_CAP_USER_MEMORY:
3215 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3216 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3217 case KVM_CAP_INTERNAL_ERROR_DATA:
3218 #ifdef CONFIG_HAVE_KVM_MSI
3219 case KVM_CAP_SIGNAL_MSI:
3221 #ifdef CONFIG_HAVE_KVM_IRQFD
3223 case KVM_CAP_IRQFD_RESAMPLE:
3225 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3226 case KVM_CAP_CHECK_EXTENSION_VM:
3227 case KVM_CAP_ENABLE_CAP_VM:
3228 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3229 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3232 #ifdef CONFIG_KVM_MMIO
3233 case KVM_CAP_COALESCED_MMIO:
3234 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3235 case KVM_CAP_COALESCED_PIO:
3238 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3239 case KVM_CAP_IRQ_ROUTING:
3240 return KVM_MAX_IRQ_ROUTES;
3242 #if KVM_ADDRESS_SPACE_NUM > 1
3243 case KVM_CAP_MULTI_ADDRESS_SPACE:
3244 return KVM_ADDRESS_SPACE_NUM;
3246 case KVM_CAP_NR_MEMSLOTS:
3247 return KVM_USER_MEM_SLOTS;
3251 return kvm_vm_ioctl_check_extension(kvm, arg);
3254 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3255 struct kvm_enable_cap *cap)
3260 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3261 struct kvm_enable_cap *cap)
3264 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3265 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3266 if (cap->flags || (cap->args[0] & ~1))
3268 kvm->manual_dirty_log_protect = cap->args[0];
3272 return kvm_vm_ioctl_enable_cap(kvm, cap);
3276 static long kvm_vm_ioctl(struct file *filp,
3277 unsigned int ioctl, unsigned long arg)
3279 struct kvm *kvm = filp->private_data;
3280 void __user *argp = (void __user *)arg;
3283 if (kvm->mm != current->mm)
3286 case KVM_CREATE_VCPU:
3287 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3289 case KVM_ENABLE_CAP: {
3290 struct kvm_enable_cap cap;
3293 if (copy_from_user(&cap, argp, sizeof(cap)))
3295 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3298 case KVM_SET_USER_MEMORY_REGION: {
3299 struct kvm_userspace_memory_region kvm_userspace_mem;
3302 if (copy_from_user(&kvm_userspace_mem, argp,
3303 sizeof(kvm_userspace_mem)))
3306 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3309 case KVM_GET_DIRTY_LOG: {
3310 struct kvm_dirty_log log;
3313 if (copy_from_user(&log, argp, sizeof(log)))
3315 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3318 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3319 case KVM_CLEAR_DIRTY_LOG: {
3320 struct kvm_clear_dirty_log log;
3323 if (copy_from_user(&log, argp, sizeof(log)))
3325 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3329 #ifdef CONFIG_KVM_MMIO
3330 case KVM_REGISTER_COALESCED_MMIO: {
3331 struct kvm_coalesced_mmio_zone zone;
3334 if (copy_from_user(&zone, argp, sizeof(zone)))
3336 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3339 case KVM_UNREGISTER_COALESCED_MMIO: {
3340 struct kvm_coalesced_mmio_zone zone;
3343 if (copy_from_user(&zone, argp, sizeof(zone)))
3345 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3350 struct kvm_irqfd data;
3353 if (copy_from_user(&data, argp, sizeof(data)))
3355 r = kvm_irqfd(kvm, &data);
3358 case KVM_IOEVENTFD: {
3359 struct kvm_ioeventfd data;
3362 if (copy_from_user(&data, argp, sizeof(data)))
3364 r = kvm_ioeventfd(kvm, &data);
3367 #ifdef CONFIG_HAVE_KVM_MSI
3368 case KVM_SIGNAL_MSI: {
3372 if (copy_from_user(&msi, argp, sizeof(msi)))
3374 r = kvm_send_userspace_msi(kvm, &msi);
3378 #ifdef __KVM_HAVE_IRQ_LINE
3379 case KVM_IRQ_LINE_STATUS:
3380 case KVM_IRQ_LINE: {
3381 struct kvm_irq_level irq_event;
3384 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3387 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3388 ioctl == KVM_IRQ_LINE_STATUS);
3393 if (ioctl == KVM_IRQ_LINE_STATUS) {
3394 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3402 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3403 case KVM_SET_GSI_ROUTING: {
3404 struct kvm_irq_routing routing;
3405 struct kvm_irq_routing __user *urouting;
3406 struct kvm_irq_routing_entry *entries = NULL;
3409 if (copy_from_user(&routing, argp, sizeof(routing)))
3412 if (!kvm_arch_can_set_irq_routing(kvm))
3414 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3420 entries = vmalloc(array_size(sizeof(*entries),
3426 if (copy_from_user(entries, urouting->entries,
3427 routing.nr * sizeof(*entries)))
3428 goto out_free_irq_routing;
3430 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3432 out_free_irq_routing:
3436 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3437 case KVM_CREATE_DEVICE: {
3438 struct kvm_create_device cd;
3441 if (copy_from_user(&cd, argp, sizeof(cd)))
3444 r = kvm_ioctl_create_device(kvm, &cd);
3449 if (copy_to_user(argp, &cd, sizeof(cd)))
3455 case KVM_CHECK_EXTENSION:
3456 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3459 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3465 #ifdef CONFIG_KVM_COMPAT
3466 struct compat_kvm_dirty_log {
3470 compat_uptr_t dirty_bitmap; /* one bit per page */
3475 static long kvm_vm_compat_ioctl(struct file *filp,
3476 unsigned int ioctl, unsigned long arg)
3478 struct kvm *kvm = filp->private_data;
3481 if (kvm->mm != current->mm)
3484 case KVM_GET_DIRTY_LOG: {
3485 struct compat_kvm_dirty_log compat_log;
3486 struct kvm_dirty_log log;
3488 if (copy_from_user(&compat_log, (void __user *)arg,
3489 sizeof(compat_log)))
3491 log.slot = compat_log.slot;
3492 log.padding1 = compat_log.padding1;
3493 log.padding2 = compat_log.padding2;
3494 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3496 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3500 r = kvm_vm_ioctl(filp, ioctl, arg);
3506 static struct file_operations kvm_vm_fops = {
3507 .release = kvm_vm_release,
3508 .unlocked_ioctl = kvm_vm_ioctl,
3509 .llseek = noop_llseek,
3510 KVM_COMPAT(kvm_vm_compat_ioctl),
3513 static int kvm_dev_ioctl_create_vm(unsigned long type)
3519 kvm = kvm_create_vm(type);
3521 return PTR_ERR(kvm);
3522 #ifdef CONFIG_KVM_MMIO
3523 r = kvm_coalesced_mmio_init(kvm);
3527 r = get_unused_fd_flags(O_CLOEXEC);
3531 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3539 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3540 * already set, with ->release() being kvm_vm_release(). In error
3541 * cases it will be called by the final fput(file) and will take
3542 * care of doing kvm_put_kvm(kvm).
3544 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3549 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3551 fd_install(r, file);
3559 static long kvm_dev_ioctl(struct file *filp,
3560 unsigned int ioctl, unsigned long arg)
3565 case KVM_GET_API_VERSION:
3568 r = KVM_API_VERSION;
3571 r = kvm_dev_ioctl_create_vm(arg);
3573 case KVM_CHECK_EXTENSION:
3574 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3576 case KVM_GET_VCPU_MMAP_SIZE:
3579 r = PAGE_SIZE; /* struct kvm_run */
3581 r += PAGE_SIZE; /* pio data page */
3583 #ifdef CONFIG_KVM_MMIO
3584 r += PAGE_SIZE; /* coalesced mmio ring page */
3587 case KVM_TRACE_ENABLE:
3588 case KVM_TRACE_PAUSE:
3589 case KVM_TRACE_DISABLE:
3593 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3599 static struct file_operations kvm_chardev_ops = {
3600 .unlocked_ioctl = kvm_dev_ioctl,
3601 .llseek = noop_llseek,
3602 KVM_COMPAT(kvm_dev_ioctl),
3605 static struct miscdevice kvm_dev = {
3611 static void hardware_enable_nolock(void *junk)
3613 int cpu = raw_smp_processor_id();
3616 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3619 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3621 r = kvm_arch_hardware_enable();
3624 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3625 atomic_inc(&hardware_enable_failed);
3626 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3630 static int kvm_starting_cpu(unsigned int cpu)
3632 raw_spin_lock(&kvm_count_lock);
3633 if (kvm_usage_count)
3634 hardware_enable_nolock(NULL);
3635 raw_spin_unlock(&kvm_count_lock);
3639 static void hardware_disable_nolock(void *junk)
3641 int cpu = raw_smp_processor_id();
3643 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3645 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3646 kvm_arch_hardware_disable();
3649 static int kvm_dying_cpu(unsigned int cpu)
3651 raw_spin_lock(&kvm_count_lock);
3652 if (kvm_usage_count)
3653 hardware_disable_nolock(NULL);
3654 raw_spin_unlock(&kvm_count_lock);
3658 static void hardware_disable_all_nolock(void)
3660 BUG_ON(!kvm_usage_count);
3663 if (!kvm_usage_count)
3664 on_each_cpu(hardware_disable_nolock, NULL, 1);
3667 static void hardware_disable_all(void)
3669 raw_spin_lock(&kvm_count_lock);
3670 hardware_disable_all_nolock();
3671 raw_spin_unlock(&kvm_count_lock);
3674 static int hardware_enable_all(void)
3678 raw_spin_lock(&kvm_count_lock);
3681 if (kvm_usage_count == 1) {
3682 atomic_set(&hardware_enable_failed, 0);
3683 on_each_cpu(hardware_enable_nolock, NULL, 1);
3685 if (atomic_read(&hardware_enable_failed)) {
3686 hardware_disable_all_nolock();
3691 raw_spin_unlock(&kvm_count_lock);
3696 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3700 * Some (well, at least mine) BIOSes hang on reboot if
3703 * And Intel TXT required VMX off for all cpu when system shutdown.
3705 pr_info("kvm: exiting hardware virtualization\n");
3706 kvm_rebooting = true;
3707 on_each_cpu(hardware_disable_nolock, NULL, 1);
3711 static struct notifier_block kvm_reboot_notifier = {
3712 .notifier_call = kvm_reboot,
3716 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3720 for (i = 0; i < bus->dev_count; i++) {
3721 struct kvm_io_device *pos = bus->range[i].dev;
3723 kvm_iodevice_destructor(pos);
3728 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3729 const struct kvm_io_range *r2)
3731 gpa_t addr1 = r1->addr;
3732 gpa_t addr2 = r2->addr;
3737 /* If r2->len == 0, match the exact address. If r2->len != 0,
3738 * accept any overlapping write. Any order is acceptable for
3739 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3740 * we process all of them.
3753 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3755 return kvm_io_bus_cmp(p1, p2);
3758 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3759 gpa_t addr, int len)
3761 struct kvm_io_range *range, key;
3764 key = (struct kvm_io_range) {
3769 range = bsearch(&key, bus->range, bus->dev_count,
3770 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3774 off = range - bus->range;
3776 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3782 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3783 struct kvm_io_range *range, const void *val)
3787 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3791 while (idx < bus->dev_count &&
3792 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3793 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3802 /* kvm_io_bus_write - called under kvm->slots_lock */
3803 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3804 int len, const void *val)
3806 struct kvm_io_bus *bus;
3807 struct kvm_io_range range;
3810 range = (struct kvm_io_range) {
3815 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3818 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3819 return r < 0 ? r : 0;
3821 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3823 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3824 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3825 gpa_t addr, int len, const void *val, long cookie)
3827 struct kvm_io_bus *bus;
3828 struct kvm_io_range range;
3830 range = (struct kvm_io_range) {
3835 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3839 /* First try the device referenced by cookie. */
3840 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3841 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3842 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3847 * cookie contained garbage; fall back to search and return the
3848 * correct cookie value.
3850 return __kvm_io_bus_write(vcpu, bus, &range, val);
3853 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3854 struct kvm_io_range *range, void *val)
3858 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3862 while (idx < bus->dev_count &&
3863 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3864 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3873 /* kvm_io_bus_read - called under kvm->slots_lock */
3874 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3877 struct kvm_io_bus *bus;
3878 struct kvm_io_range range;
3881 range = (struct kvm_io_range) {
3886 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3889 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3890 return r < 0 ? r : 0;
3893 /* Caller must hold slots_lock. */
3894 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3895 int len, struct kvm_io_device *dev)
3898 struct kvm_io_bus *new_bus, *bus;
3899 struct kvm_io_range range;
3901 bus = kvm_get_bus(kvm, bus_idx);
3905 /* exclude ioeventfd which is limited by maximum fd */
3906 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3909 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3910 GFP_KERNEL_ACCOUNT);
3914 range = (struct kvm_io_range) {
3920 for (i = 0; i < bus->dev_count; i++)
3921 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3924 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3925 new_bus->dev_count++;
3926 new_bus->range[i] = range;
3927 memcpy(new_bus->range + i + 1, bus->range + i,
3928 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3929 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3930 synchronize_srcu_expedited(&kvm->srcu);
3936 /* Caller must hold slots_lock. */
3937 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3938 struct kvm_io_device *dev)
3941 struct kvm_io_bus *new_bus, *bus;
3943 bus = kvm_get_bus(kvm, bus_idx);
3947 for (i = 0; i < bus->dev_count; i++)
3948 if (bus->range[i].dev == dev) {
3952 if (i == bus->dev_count)
3955 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3956 GFP_KERNEL_ACCOUNT);
3958 pr_err("kvm: failed to shrink bus, removing it completely\n");
3962 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3963 new_bus->dev_count--;
3964 memcpy(new_bus->range + i, bus->range + i + 1,
3965 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3968 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3969 synchronize_srcu_expedited(&kvm->srcu);
3974 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3977 struct kvm_io_bus *bus;
3978 int dev_idx, srcu_idx;
3979 struct kvm_io_device *iodev = NULL;
3981 srcu_idx = srcu_read_lock(&kvm->srcu);
3983 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3987 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3991 iodev = bus->range[dev_idx].dev;
3994 srcu_read_unlock(&kvm->srcu, srcu_idx);
3998 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4000 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4001 int (*get)(void *, u64 *), int (*set)(void *, u64),
4004 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4007 /* The debugfs files are a reference to the kvm struct which
4008 * is still valid when kvm_destroy_vm is called.
4009 * To avoid the race between open and the removal of the debugfs
4010 * directory we test against the users count.
4012 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4015 if (simple_attr_open(inode, file, get,
4016 stat_data->mode & S_IWUGO ? set : NULL,
4018 kvm_put_kvm(stat_data->kvm);
4025 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4027 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4030 simple_attr_release(inode, file);
4031 kvm_put_kvm(stat_data->kvm);
4036 static int vm_stat_get_per_vm(void *data, u64 *val)
4038 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4040 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
4045 static int vm_stat_clear_per_vm(void *data, u64 val)
4047 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4052 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
4057 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
4059 __simple_attr_check_format("%llu\n", 0ull);
4060 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
4061 vm_stat_clear_per_vm, "%llu\n");
4064 static const struct file_operations vm_stat_get_per_vm_fops = {
4065 .owner = THIS_MODULE,
4066 .open = vm_stat_get_per_vm_open,
4067 .release = kvm_debugfs_release,
4068 .read = simple_attr_read,
4069 .write = simple_attr_write,
4070 .llseek = no_llseek,
4073 static int vcpu_stat_get_per_vm(void *data, u64 *val)
4076 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4077 struct kvm_vcpu *vcpu;
4081 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4082 *val += *(u64 *)((void *)vcpu + stat_data->offset);
4087 static int vcpu_stat_clear_per_vm(void *data, u64 val)
4090 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4091 struct kvm_vcpu *vcpu;
4096 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4097 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4102 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4104 __simple_attr_check_format("%llu\n", 0ull);
4105 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4106 vcpu_stat_clear_per_vm, "%llu\n");
4109 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4110 .owner = THIS_MODULE,
4111 .open = vcpu_stat_get_per_vm_open,
4112 .release = kvm_debugfs_release,
4113 .read = simple_attr_read,
4114 .write = simple_attr_write,
4115 .llseek = no_llseek,
4118 static const struct file_operations *stat_fops_per_vm[] = {
4119 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4120 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4123 static int vm_stat_get(void *_offset, u64 *val)
4125 unsigned offset = (long)_offset;
4127 struct kvm_stat_data stat_tmp = {.offset = offset};
4131 mutex_lock(&kvm_lock);
4132 list_for_each_entry(kvm, &vm_list, vm_list) {
4134 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4137 mutex_unlock(&kvm_lock);
4141 static int vm_stat_clear(void *_offset, u64 val)
4143 unsigned offset = (long)_offset;
4145 struct kvm_stat_data stat_tmp = {.offset = offset};
4150 mutex_lock(&kvm_lock);
4151 list_for_each_entry(kvm, &vm_list, vm_list) {
4153 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4155 mutex_unlock(&kvm_lock);
4160 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4162 static int vcpu_stat_get(void *_offset, u64 *val)
4164 unsigned offset = (long)_offset;
4166 struct kvm_stat_data stat_tmp = {.offset = offset};
4170 mutex_lock(&kvm_lock);
4171 list_for_each_entry(kvm, &vm_list, vm_list) {
4173 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4176 mutex_unlock(&kvm_lock);
4180 static int vcpu_stat_clear(void *_offset, u64 val)
4182 unsigned offset = (long)_offset;
4184 struct kvm_stat_data stat_tmp = {.offset = offset};
4189 mutex_lock(&kvm_lock);
4190 list_for_each_entry(kvm, &vm_list, vm_list) {
4192 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4194 mutex_unlock(&kvm_lock);
4199 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4202 static const struct file_operations *stat_fops[] = {
4203 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4204 [KVM_STAT_VM] = &vm_stat_fops,
4207 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4209 struct kobj_uevent_env *env;
4210 unsigned long long created, active;
4212 if (!kvm_dev.this_device || !kvm)
4215 mutex_lock(&kvm_lock);
4216 if (type == KVM_EVENT_CREATE_VM) {
4217 kvm_createvm_count++;
4219 } else if (type == KVM_EVENT_DESTROY_VM) {
4222 created = kvm_createvm_count;
4223 active = kvm_active_vms;
4224 mutex_unlock(&kvm_lock);
4226 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4230 add_uevent_var(env, "CREATED=%llu", created);
4231 add_uevent_var(env, "COUNT=%llu", active);
4233 if (type == KVM_EVENT_CREATE_VM) {
4234 add_uevent_var(env, "EVENT=create");
4235 kvm->userspace_pid = task_pid_nr(current);
4236 } else if (type == KVM_EVENT_DESTROY_VM) {
4237 add_uevent_var(env, "EVENT=destroy");
4239 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4241 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4242 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4245 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4247 add_uevent_var(env, "STATS_PATH=%s", tmp);
4251 /* no need for checks, since we are adding at most only 5 keys */
4252 env->envp[env->envp_idx++] = NULL;
4253 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4257 static void kvm_init_debug(void)
4259 struct kvm_stats_debugfs_item *p;
4261 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4263 kvm_debugfs_num_entries = 0;
4264 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4265 int mode = p->mode ? p->mode : 0644;
4266 debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4267 (void *)(long)p->offset,
4268 stat_fops[p->kind]);
4272 static int kvm_suspend(void)
4274 if (kvm_usage_count)
4275 hardware_disable_nolock(NULL);
4279 static void kvm_resume(void)
4281 if (kvm_usage_count) {
4282 #ifdef CONFIG_LOCKDEP
4283 WARN_ON(lockdep_is_held(&kvm_count_lock));
4285 hardware_enable_nolock(NULL);
4289 static struct syscore_ops kvm_syscore_ops = {
4290 .suspend = kvm_suspend,
4291 .resume = kvm_resume,
4295 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4297 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4300 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4302 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4304 WRITE_ONCE(vcpu->preempted, false);
4305 WRITE_ONCE(vcpu->ready, false);
4307 kvm_arch_sched_in(vcpu, cpu);
4309 kvm_arch_vcpu_load(vcpu, cpu);
4312 static void kvm_sched_out(struct preempt_notifier *pn,
4313 struct task_struct *next)
4315 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4317 if (current->state == TASK_RUNNING) {
4318 WRITE_ONCE(vcpu->preempted, true);
4319 WRITE_ONCE(vcpu->ready, true);
4321 kvm_arch_vcpu_put(vcpu);
4324 static void check_processor_compat(void *rtn)
4326 *(int *)rtn = kvm_arch_check_processor_compat();
4329 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4330 struct module *module)
4335 r = kvm_arch_init(opaque);
4340 * kvm_arch_init makes sure there's at most one caller
4341 * for architectures that support multiple implementations,
4342 * like intel and amd on x86.
4343 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4344 * conflicts in case kvm is already setup for another implementation.
4346 r = kvm_irqfd_init();
4350 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4355 r = kvm_arch_hardware_setup();
4359 for_each_online_cpu(cpu) {
4360 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4365 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4366 kvm_starting_cpu, kvm_dying_cpu);
4369 register_reboot_notifier(&kvm_reboot_notifier);
4371 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4373 vcpu_align = __alignof__(struct kvm_vcpu);
4375 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4377 offsetof(struct kvm_vcpu, arch),
4378 sizeof_field(struct kvm_vcpu, arch),
4380 if (!kvm_vcpu_cache) {
4385 r = kvm_async_pf_init();
4389 kvm_chardev_ops.owner = module;
4390 kvm_vm_fops.owner = module;
4391 kvm_vcpu_fops.owner = module;
4393 r = misc_register(&kvm_dev);
4395 pr_err("kvm: misc device register failed\n");
4399 register_syscore_ops(&kvm_syscore_ops);
4401 kvm_preempt_ops.sched_in = kvm_sched_in;
4402 kvm_preempt_ops.sched_out = kvm_sched_out;
4406 r = kvm_vfio_ops_init();
4412 kvm_async_pf_deinit();
4414 kmem_cache_destroy(kvm_vcpu_cache);
4416 unregister_reboot_notifier(&kvm_reboot_notifier);
4417 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4419 kvm_arch_hardware_unsetup();
4421 free_cpumask_var(cpus_hardware_enabled);
4429 EXPORT_SYMBOL_GPL(kvm_init);
4433 debugfs_remove_recursive(kvm_debugfs_dir);
4434 misc_deregister(&kvm_dev);
4435 kmem_cache_destroy(kvm_vcpu_cache);
4436 kvm_async_pf_deinit();
4437 unregister_syscore_ops(&kvm_syscore_ops);
4438 unregister_reboot_notifier(&kvm_reboot_notifier);
4439 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4440 on_each_cpu(hardware_disable_nolock, NULL, 1);
4441 kvm_arch_hardware_unsetup();
4444 free_cpumask_var(cpus_hardware_enabled);
4445 kvm_vfio_ops_exit();
4447 EXPORT_SYMBOL_GPL(kvm_exit);
4449 struct kvm_vm_worker_thread_context {
4451 struct task_struct *parent;
4452 struct completion init_done;
4453 kvm_vm_thread_fn_t thread_fn;
4458 static int kvm_vm_worker_thread(void *context)
4461 * The init_context is allocated on the stack of the parent thread, so
4462 * we have to locally copy anything that is needed beyond initialization
4464 struct kvm_vm_worker_thread_context *init_context = context;
4465 struct kvm *kvm = init_context->kvm;
4466 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4467 uintptr_t data = init_context->data;
4470 err = kthread_park(current);
4471 /* kthread_park(current) is never supposed to return an error */
4476 err = cgroup_attach_task_all(init_context->parent, current);
4478 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4483 set_user_nice(current, task_nice(init_context->parent));
4486 init_context->err = err;
4487 complete(&init_context->init_done);
4488 init_context = NULL;
4493 /* Wait to be woken up by the spawner before proceeding. */
4496 if (!kthread_should_stop())
4497 err = thread_fn(kvm, data);
4502 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4503 uintptr_t data, const char *name,
4504 struct task_struct **thread_ptr)
4506 struct kvm_vm_worker_thread_context init_context = {};
4507 struct task_struct *thread;
4510 init_context.kvm = kvm;
4511 init_context.parent = current;
4512 init_context.thread_fn = thread_fn;
4513 init_context.data = data;
4514 init_completion(&init_context.init_done);
4516 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4517 "%s-%d", name, task_pid_nr(current));
4519 return PTR_ERR(thread);
4521 /* kthread_run is never supposed to return NULL */
4522 WARN_ON(thread == NULL);
4524 wait_for_completion(&init_context.init_done);
4526 if (!init_context.err)
4527 *thread_ptr = thread;
4529 return init_context.err;