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 static struct kmem_cache *kvm_vcpu_cache;
109 static __read_mostly struct preempt_ops kvm_preempt_ops;
111 struct dentry *kvm_debugfs_dir;
112 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
114 static int kvm_debugfs_num_entries;
115 static const struct file_operations stat_fops_per_vm;
117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
119 #ifdef CONFIG_KVM_COMPAT
120 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
122 #define KVM_COMPAT(c) .compat_ioctl = (c)
125 * For architectures that don't implement a compat infrastructure,
126 * adopt a double line of defense:
127 * - Prevent a compat task from opening /dev/kvm
128 * - If the open has been done by a 64bit task, and the KVM fd
129 * passed to a compat task, let the ioctls fail.
131 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
132 unsigned long arg) { return -EINVAL; }
134 static int kvm_no_compat_open(struct inode *inode, struct file *file)
136 return is_compat_task() ? -ENODEV : 0;
138 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
139 .open = kvm_no_compat_open
141 static int hardware_enable_all(void);
142 static void hardware_disable_all(void);
144 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
146 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
148 __visible bool kvm_rebooting;
149 EXPORT_SYMBOL_GPL(kvm_rebooting);
151 static bool largepages_enabled = true;
153 #define KVM_EVENT_CREATE_VM 0
154 #define KVM_EVENT_DESTROY_VM 1
155 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
156 static unsigned long long kvm_createvm_count;
157 static unsigned long long kvm_active_vms;
159 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
160 unsigned long start, unsigned long end, bool blockable)
165 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
168 * The metadata used by is_zone_device_page() to determine whether or
169 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
170 * the device has been pinned, e.g. by get_user_pages(). WARN if the
171 * page_count() is zero to help detect bad usage of this helper.
173 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
176 return is_zone_device_page(pfn_to_page(pfn));
179 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
182 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
183 * perspective they are "normal" pages, albeit with slightly different
187 return PageReserved(pfn_to_page(pfn)) &&
188 !kvm_is_zone_device_pfn(pfn);
194 * Switches to specified vcpu, until a matching vcpu_put()
196 void vcpu_load(struct kvm_vcpu *vcpu)
199 preempt_notifier_register(&vcpu->preempt_notifier);
200 kvm_arch_vcpu_load(vcpu, cpu);
203 EXPORT_SYMBOL_GPL(vcpu_load);
205 void vcpu_put(struct kvm_vcpu *vcpu)
208 kvm_arch_vcpu_put(vcpu);
209 preempt_notifier_unregister(&vcpu->preempt_notifier);
212 EXPORT_SYMBOL_GPL(vcpu_put);
214 /* TODO: merge with kvm_arch_vcpu_should_kick */
215 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
217 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
220 * We need to wait for the VCPU to reenable interrupts and get out of
221 * READING_SHADOW_PAGE_TABLES mode.
223 if (req & KVM_REQUEST_WAIT)
224 return mode != OUTSIDE_GUEST_MODE;
227 * Need to kick a running VCPU, but otherwise there is nothing to do.
229 return mode == IN_GUEST_MODE;
232 static void ack_flush(void *_completed)
236 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
239 cpus = cpu_online_mask;
241 if (cpumask_empty(cpus))
244 smp_call_function_many(cpus, ack_flush, NULL, wait);
248 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
249 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
252 struct kvm_vcpu *vcpu;
257 kvm_for_each_vcpu(i, vcpu, kvm) {
258 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
261 kvm_make_request(req, vcpu);
264 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
267 if (tmp != NULL && cpu != -1 && cpu != me &&
268 kvm_request_needs_ipi(vcpu, req))
269 __cpumask_set_cpu(cpu, tmp);
272 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
278 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
283 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
285 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
287 free_cpumask_var(cpus);
291 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
292 void kvm_flush_remote_tlbs(struct kvm *kvm)
295 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
296 * kvm_make_all_cpus_request.
298 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
301 * We want to publish modifications to the page tables before reading
302 * mode. Pairs with a memory barrier in arch-specific code.
303 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
304 * and smp_mb in walk_shadow_page_lockless_begin/end.
305 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
307 * There is already an smp_mb__after_atomic() before
308 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
311 if (!kvm_arch_flush_remote_tlb(kvm)
312 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
313 ++kvm->stat.remote_tlb_flush;
314 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
316 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
319 void kvm_reload_remote_mmus(struct kvm *kvm)
321 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
324 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
326 mutex_init(&vcpu->mutex);
331 init_swait_queue_head(&vcpu->wq);
332 kvm_async_pf_vcpu_init(vcpu);
335 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
337 kvm_vcpu_set_in_spin_loop(vcpu, false);
338 kvm_vcpu_set_dy_eligible(vcpu, false);
339 vcpu->preempted = false;
341 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
344 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
346 kvm_arch_vcpu_destroy(vcpu);
349 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
350 * the vcpu->pid pointer, and at destruction time all file descriptors
353 put_pid(rcu_dereference_protected(vcpu->pid, 1));
355 free_page((unsigned long)vcpu->run);
356 kmem_cache_free(kvm_vcpu_cache, vcpu);
358 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
360 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
361 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
363 return container_of(mn, struct kvm, mmu_notifier);
366 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
367 struct mm_struct *mm,
368 unsigned long address,
371 struct kvm *kvm = mmu_notifier_to_kvm(mn);
374 idx = srcu_read_lock(&kvm->srcu);
375 spin_lock(&kvm->mmu_lock);
376 kvm->mmu_notifier_seq++;
378 if (kvm_set_spte_hva(kvm, address, pte))
379 kvm_flush_remote_tlbs(kvm);
381 spin_unlock(&kvm->mmu_lock);
382 srcu_read_unlock(&kvm->srcu, idx);
385 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
386 const struct mmu_notifier_range *range)
388 struct kvm *kvm = mmu_notifier_to_kvm(mn);
389 int need_tlb_flush = 0, idx;
392 idx = srcu_read_lock(&kvm->srcu);
393 spin_lock(&kvm->mmu_lock);
395 * The count increase must become visible at unlock time as no
396 * spte can be established without taking the mmu_lock and
397 * count is also read inside the mmu_lock critical section.
399 kvm->mmu_notifier_count++;
400 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
401 need_tlb_flush |= kvm->tlbs_dirty;
402 /* we've to flush the tlb before the pages can be freed */
404 kvm_flush_remote_tlbs(kvm);
406 spin_unlock(&kvm->mmu_lock);
408 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
410 mmu_notifier_range_blockable(range));
412 srcu_read_unlock(&kvm->srcu, idx);
417 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
418 const struct mmu_notifier_range *range)
420 struct kvm *kvm = mmu_notifier_to_kvm(mn);
422 spin_lock(&kvm->mmu_lock);
424 * This sequence increase will notify the kvm page fault that
425 * the page that is going to be mapped in the spte could have
428 kvm->mmu_notifier_seq++;
431 * The above sequence increase must be visible before the
432 * below count decrease, which is ensured by the smp_wmb above
433 * in conjunction with the smp_rmb in mmu_notifier_retry().
435 kvm->mmu_notifier_count--;
436 spin_unlock(&kvm->mmu_lock);
438 BUG_ON(kvm->mmu_notifier_count < 0);
441 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
442 struct mm_struct *mm,
446 struct kvm *kvm = mmu_notifier_to_kvm(mn);
449 idx = srcu_read_lock(&kvm->srcu);
450 spin_lock(&kvm->mmu_lock);
452 young = kvm_age_hva(kvm, start, end);
454 kvm_flush_remote_tlbs(kvm);
456 spin_unlock(&kvm->mmu_lock);
457 srcu_read_unlock(&kvm->srcu, idx);
462 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
463 struct mm_struct *mm,
467 struct kvm *kvm = mmu_notifier_to_kvm(mn);
470 idx = srcu_read_lock(&kvm->srcu);
471 spin_lock(&kvm->mmu_lock);
473 * Even though we do not flush TLB, this will still adversely
474 * affect performance on pre-Haswell Intel EPT, where there is
475 * no EPT Access Bit to clear so that we have to tear down EPT
476 * tables instead. If we find this unacceptable, we can always
477 * add a parameter to kvm_age_hva so that it effectively doesn't
478 * do anything on clear_young.
480 * Also note that currently we never issue secondary TLB flushes
481 * from clear_young, leaving this job up to the regular system
482 * cadence. If we find this inaccurate, we might come up with a
483 * more sophisticated heuristic later.
485 young = kvm_age_hva(kvm, start, end);
486 spin_unlock(&kvm->mmu_lock);
487 srcu_read_unlock(&kvm->srcu, idx);
492 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
493 struct mm_struct *mm,
494 unsigned long address)
496 struct kvm *kvm = mmu_notifier_to_kvm(mn);
499 idx = srcu_read_lock(&kvm->srcu);
500 spin_lock(&kvm->mmu_lock);
501 young = kvm_test_age_hva(kvm, address);
502 spin_unlock(&kvm->mmu_lock);
503 srcu_read_unlock(&kvm->srcu, idx);
508 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
509 struct mm_struct *mm)
511 struct kvm *kvm = mmu_notifier_to_kvm(mn);
514 idx = srcu_read_lock(&kvm->srcu);
515 kvm_arch_flush_shadow_all(kvm);
516 srcu_read_unlock(&kvm->srcu, idx);
519 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
520 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
521 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
522 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
523 .clear_young = kvm_mmu_notifier_clear_young,
524 .test_young = kvm_mmu_notifier_test_young,
525 .change_pte = kvm_mmu_notifier_change_pte,
526 .release = kvm_mmu_notifier_release,
529 static int kvm_init_mmu_notifier(struct kvm *kvm)
531 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
532 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
535 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
537 static int kvm_init_mmu_notifier(struct kvm *kvm)
542 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
544 static struct kvm_memslots *kvm_alloc_memslots(void)
547 struct kvm_memslots *slots;
549 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
553 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
554 slots->id_to_index[i] = slots->memslots[i].id = i;
559 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
561 if (!memslot->dirty_bitmap)
564 kvfree(memslot->dirty_bitmap);
565 memslot->dirty_bitmap = NULL;
569 * Free any memory in @free but not in @dont.
571 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
572 struct kvm_memory_slot *dont)
574 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
575 kvm_destroy_dirty_bitmap(free);
577 kvm_arch_free_memslot(kvm, free, dont);
582 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
584 struct kvm_memory_slot *memslot;
589 kvm_for_each_memslot(memslot, slots)
590 kvm_free_memslot(kvm, memslot, NULL);
595 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
599 if (!kvm->debugfs_dentry)
602 debugfs_remove_recursive(kvm->debugfs_dentry);
604 if (kvm->debugfs_stat_data) {
605 for (i = 0; i < kvm_debugfs_num_entries; i++)
606 kfree(kvm->debugfs_stat_data[i]);
607 kfree(kvm->debugfs_stat_data);
611 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
613 char dir_name[ITOA_MAX_LEN * 2];
614 struct kvm_stat_data *stat_data;
615 struct kvm_stats_debugfs_item *p;
617 if (!debugfs_initialized())
620 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
621 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
623 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
624 sizeof(*kvm->debugfs_stat_data),
626 if (!kvm->debugfs_stat_data)
629 for (p = debugfs_entries; p->name; p++) {
630 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
634 stat_data->kvm = kvm;
635 stat_data->dbgfs_item = p;
636 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
637 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
638 kvm->debugfs_dentry, stat_data,
645 * Called after the VM is otherwise initialized, but just before adding it to
648 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
654 * Called just after removing the VM from the vm_list, but before doing any
657 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
661 static struct kvm *kvm_create_vm(unsigned long type)
663 struct kvm *kvm = kvm_arch_alloc_vm();
668 return ERR_PTR(-ENOMEM);
670 spin_lock_init(&kvm->mmu_lock);
672 kvm->mm = current->mm;
673 kvm_eventfd_init(kvm);
674 mutex_init(&kvm->lock);
675 mutex_init(&kvm->irq_lock);
676 mutex_init(&kvm->slots_lock);
677 INIT_LIST_HEAD(&kvm->devices);
679 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
681 if (init_srcu_struct(&kvm->srcu))
682 goto out_err_no_srcu;
683 if (init_srcu_struct(&kvm->irq_srcu))
684 goto out_err_no_irq_srcu;
686 refcount_set(&kvm->users_count, 1);
687 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
688 struct kvm_memslots *slots = kvm_alloc_memslots();
691 goto out_err_no_arch_destroy_vm;
692 /* Generations must be different for each address space. */
693 slots->generation = i;
694 rcu_assign_pointer(kvm->memslots[i], slots);
697 for (i = 0; i < KVM_NR_BUSES; i++) {
698 rcu_assign_pointer(kvm->buses[i],
699 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
701 goto out_err_no_arch_destroy_vm;
704 r = kvm_arch_init_vm(kvm, type);
706 goto out_err_no_arch_destroy_vm;
708 r = hardware_enable_all();
710 goto out_err_no_disable;
712 #ifdef CONFIG_HAVE_KVM_IRQFD
713 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
716 r = kvm_init_mmu_notifier(kvm);
718 goto out_err_no_mmu_notifier;
720 r = kvm_arch_post_init_vm(kvm);
724 mutex_lock(&kvm_lock);
725 list_add(&kvm->vm_list, &vm_list);
726 mutex_unlock(&kvm_lock);
728 preempt_notifier_inc();
733 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
734 if (kvm->mmu_notifier.ops)
735 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
737 out_err_no_mmu_notifier:
738 hardware_disable_all();
740 kvm_arch_destroy_vm(kvm);
741 out_err_no_arch_destroy_vm:
742 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
743 for (i = 0; i < KVM_NR_BUSES; i++)
744 kfree(kvm_get_bus(kvm, i));
745 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
746 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
747 cleanup_srcu_struct(&kvm->irq_srcu);
749 cleanup_srcu_struct(&kvm->srcu);
751 kvm_arch_free_vm(kvm);
756 static void kvm_destroy_devices(struct kvm *kvm)
758 struct kvm_device *dev, *tmp;
761 * We do not need to take the kvm->lock here, because nobody else
762 * has a reference to the struct kvm at this point and therefore
763 * cannot access the devices list anyhow.
765 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
766 list_del(&dev->vm_node);
767 dev->ops->destroy(dev);
771 static void kvm_destroy_vm(struct kvm *kvm)
774 struct mm_struct *mm = kvm->mm;
776 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
777 kvm_destroy_vm_debugfs(kvm);
778 kvm_arch_sync_events(kvm);
779 mutex_lock(&kvm_lock);
780 list_del(&kvm->vm_list);
781 mutex_unlock(&kvm_lock);
782 kvm_arch_pre_destroy_vm(kvm);
784 kvm_free_irq_routing(kvm);
785 for (i = 0; i < KVM_NR_BUSES; i++) {
786 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
789 kvm_io_bus_destroy(bus);
790 kvm->buses[i] = NULL;
792 kvm_coalesced_mmio_free(kvm);
793 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
794 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
796 kvm_arch_flush_shadow_all(kvm);
798 kvm_arch_destroy_vm(kvm);
799 kvm_destroy_devices(kvm);
800 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
801 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
802 cleanup_srcu_struct(&kvm->irq_srcu);
803 cleanup_srcu_struct(&kvm->srcu);
804 kvm_arch_free_vm(kvm);
805 preempt_notifier_dec();
806 hardware_disable_all();
810 void kvm_get_kvm(struct kvm *kvm)
812 refcount_inc(&kvm->users_count);
814 EXPORT_SYMBOL_GPL(kvm_get_kvm);
816 void kvm_put_kvm(struct kvm *kvm)
818 if (refcount_dec_and_test(&kvm->users_count))
821 EXPORT_SYMBOL_GPL(kvm_put_kvm);
824 * Used to put a reference that was taken on behalf of an object associated
825 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
826 * of the new file descriptor fails and the reference cannot be transferred to
827 * its final owner. In such cases, the caller is still actively using @kvm and
828 * will fail miserably if the refcount unexpectedly hits zero.
830 void kvm_put_kvm_no_destroy(struct kvm *kvm)
832 WARN_ON(refcount_dec_and_test(&kvm->users_count));
834 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
836 static int kvm_vm_release(struct inode *inode, struct file *filp)
838 struct kvm *kvm = filp->private_data;
840 kvm_irqfd_release(kvm);
847 * Allocation size is twice as large as the actual dirty bitmap size.
848 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
850 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
852 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
854 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
855 if (!memslot->dirty_bitmap)
862 * Insert memslot and re-sort memslots based on their GFN,
863 * so binary search could be used to lookup GFN.
864 * Sorting algorithm takes advantage of having initially
865 * sorted array and known changed memslot position.
867 static void update_memslots(struct kvm_memslots *slots,
868 struct kvm_memory_slot *new,
869 enum kvm_mr_change change)
872 int i = slots->id_to_index[id];
873 struct kvm_memory_slot *mslots = slots->memslots;
875 WARN_ON(mslots[i].id != id);
879 WARN_ON(mslots[i].npages || !new->npages);
883 WARN_ON(new->npages || !mslots[i].npages);
889 while (i < KVM_MEM_SLOTS_NUM - 1 &&
890 new->base_gfn <= mslots[i + 1].base_gfn) {
891 if (!mslots[i + 1].npages)
893 mslots[i] = mslots[i + 1];
894 slots->id_to_index[mslots[i].id] = i;
899 * The ">=" is needed when creating a slot with base_gfn == 0,
900 * so that it moves before all those with base_gfn == npages == 0.
902 * On the other hand, if new->npages is zero, the above loop has
903 * already left i pointing to the beginning of the empty part of
904 * mslots, and the ">=" would move the hole backwards in this
905 * case---which is wrong. So skip the loop when deleting a slot.
909 new->base_gfn >= mslots[i - 1].base_gfn) {
910 mslots[i] = mslots[i - 1];
911 slots->id_to_index[mslots[i].id] = i;
915 WARN_ON_ONCE(i != slots->used_slots);
918 slots->id_to_index[mslots[i].id] = i;
921 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
923 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
925 #ifdef __KVM_HAVE_READONLY_MEM
926 valid_flags |= KVM_MEM_READONLY;
929 if (mem->flags & ~valid_flags)
935 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
936 int as_id, struct kvm_memslots *slots)
938 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
939 u64 gen = old_memslots->generation;
941 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
942 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
944 rcu_assign_pointer(kvm->memslots[as_id], slots);
945 synchronize_srcu_expedited(&kvm->srcu);
948 * Increment the new memslot generation a second time, dropping the
949 * update in-progress flag and incrementing the generation based on
950 * the number of address spaces. This provides a unique and easily
951 * identifiable generation number while the memslots are in flux.
953 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
956 * Generations must be unique even across address spaces. We do not need
957 * a global counter for that, instead the generation space is evenly split
958 * across address spaces. For example, with two address spaces, address
959 * space 0 will use generations 0, 2, 4, ... while address space 1 will
960 * use generations 1, 3, 5, ...
962 gen += KVM_ADDRESS_SPACE_NUM;
964 kvm_arch_memslots_updated(kvm, gen);
966 slots->generation = gen;
972 * Allocate some memory and give it an address in the guest physical address
975 * Discontiguous memory is allowed, mostly for framebuffers.
977 * Must be called holding kvm->slots_lock for write.
979 int __kvm_set_memory_region(struct kvm *kvm,
980 const struct kvm_userspace_memory_region *mem)
984 unsigned long npages;
985 struct kvm_memory_slot *slot;
986 struct kvm_memory_slot old, new;
987 struct kvm_memslots *slots = NULL, *old_memslots;
989 enum kvm_mr_change change;
991 r = check_memory_region_flags(mem);
996 as_id = mem->slot >> 16;
999 /* General sanity checks */
1000 if (mem->memory_size & (PAGE_SIZE - 1))
1002 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1004 /* We can read the guest memory with __xxx_user() later on. */
1005 if ((id < KVM_USER_MEM_SLOTS) &&
1006 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1007 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1010 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1012 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1015 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1016 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1017 npages = mem->memory_size >> PAGE_SHIFT;
1019 if (npages > KVM_MEM_MAX_NR_PAGES)
1025 new.base_gfn = base_gfn;
1026 new.npages = npages;
1027 new.flags = mem->flags;
1031 change = KVM_MR_CREATE;
1032 else { /* Modify an existing slot. */
1033 if ((mem->userspace_addr != old.userspace_addr) ||
1034 (npages != old.npages) ||
1035 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1038 if (base_gfn != old.base_gfn)
1039 change = KVM_MR_MOVE;
1040 else if (new.flags != old.flags)
1041 change = KVM_MR_FLAGS_ONLY;
1042 else { /* Nothing to change. */
1051 change = KVM_MR_DELETE;
1056 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1057 /* Check for overlaps */
1059 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1062 if (!((base_gfn + npages <= slot->base_gfn) ||
1063 (base_gfn >= slot->base_gfn + slot->npages)))
1068 /* Free page dirty bitmap if unneeded */
1069 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1070 new.dirty_bitmap = NULL;
1073 if (change == KVM_MR_CREATE) {
1074 new.userspace_addr = mem->userspace_addr;
1076 if (kvm_arch_create_memslot(kvm, &new, npages))
1080 /* Allocate page dirty bitmap if needed */
1081 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1082 if (kvm_create_dirty_bitmap(&new) < 0)
1086 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1089 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1091 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1092 slot = id_to_memslot(slots, id);
1093 slot->flags |= KVM_MEMSLOT_INVALID;
1095 old_memslots = install_new_memslots(kvm, as_id, slots);
1097 /* From this point no new shadow pages pointing to a deleted,
1098 * or moved, memslot will be created.
1100 * validation of sp->gfn happens in:
1101 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1102 * - kvm_is_visible_gfn (mmu_check_root)
1104 kvm_arch_flush_shadow_memslot(kvm, slot);
1107 * We can re-use the old_memslots from above, the only difference
1108 * from the currently installed memslots is the invalid flag. This
1109 * will get overwritten by update_memslots anyway.
1111 slots = old_memslots;
1114 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1118 /* actual memory is freed via old in kvm_free_memslot below */
1119 if (change == KVM_MR_DELETE) {
1120 new.dirty_bitmap = NULL;
1121 memset(&new.arch, 0, sizeof(new.arch));
1124 update_memslots(slots, &new, change);
1125 old_memslots = install_new_memslots(kvm, as_id, slots);
1127 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1129 kvm_free_memslot(kvm, &old, &new);
1130 kvfree(old_memslots);
1136 kvm_free_memslot(kvm, &new, &old);
1140 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1142 int kvm_set_memory_region(struct kvm *kvm,
1143 const struct kvm_userspace_memory_region *mem)
1147 mutex_lock(&kvm->slots_lock);
1148 r = __kvm_set_memory_region(kvm, mem);
1149 mutex_unlock(&kvm->slots_lock);
1152 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1154 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1155 struct kvm_userspace_memory_region *mem)
1157 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1160 return kvm_set_memory_region(kvm, mem);
1163 int kvm_get_dirty_log(struct kvm *kvm,
1164 struct kvm_dirty_log *log, int *is_dirty)
1166 struct kvm_memslots *slots;
1167 struct kvm_memory_slot *memslot;
1170 unsigned long any = 0;
1172 as_id = log->slot >> 16;
1173 id = (u16)log->slot;
1174 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1177 slots = __kvm_memslots(kvm, as_id);
1178 memslot = id_to_memslot(slots, id);
1179 if (!memslot->dirty_bitmap)
1182 n = kvm_dirty_bitmap_bytes(memslot);
1184 for (i = 0; !any && i < n/sizeof(long); ++i)
1185 any = memslot->dirty_bitmap[i];
1187 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1194 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1196 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1198 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1199 * and reenable dirty page tracking for the corresponding pages.
1200 * @kvm: pointer to kvm instance
1201 * @log: slot id and address to which we copy the log
1202 * @flush: true if TLB flush is needed by caller
1204 * We need to keep it in mind that VCPU threads can write to the bitmap
1205 * concurrently. So, to avoid losing track of dirty pages we keep the
1208 * 1. Take a snapshot of the bit and clear it if needed.
1209 * 2. Write protect the corresponding page.
1210 * 3. Copy the snapshot to the userspace.
1211 * 4. Upon return caller flushes TLB's if needed.
1213 * Between 2 and 4, the guest may write to the page using the remaining TLB
1214 * entry. This is not a problem because the page is reported dirty using
1215 * the snapshot taken before and step 4 ensures that writes done after
1216 * exiting to userspace will be logged for the next call.
1219 int kvm_get_dirty_log_protect(struct kvm *kvm,
1220 struct kvm_dirty_log *log, bool *flush)
1222 struct kvm_memslots *slots;
1223 struct kvm_memory_slot *memslot;
1226 unsigned long *dirty_bitmap;
1227 unsigned long *dirty_bitmap_buffer;
1229 as_id = log->slot >> 16;
1230 id = (u16)log->slot;
1231 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1234 slots = __kvm_memslots(kvm, as_id);
1235 memslot = id_to_memslot(slots, id);
1237 dirty_bitmap = memslot->dirty_bitmap;
1241 n = kvm_dirty_bitmap_bytes(memslot);
1243 if (kvm->manual_dirty_log_protect) {
1245 * Unlike kvm_get_dirty_log, we always return false in *flush,
1246 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1247 * is some code duplication between this function and
1248 * kvm_get_dirty_log, but hopefully all architecture
1249 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1250 * can be eliminated.
1252 dirty_bitmap_buffer = dirty_bitmap;
1254 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1255 memset(dirty_bitmap_buffer, 0, n);
1257 spin_lock(&kvm->mmu_lock);
1258 for (i = 0; i < n / sizeof(long); i++) {
1262 if (!dirty_bitmap[i])
1266 mask = xchg(&dirty_bitmap[i], 0);
1267 dirty_bitmap_buffer[i] = mask;
1269 offset = i * BITS_PER_LONG;
1270 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1273 spin_unlock(&kvm->mmu_lock);
1276 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1280 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1283 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1284 * and reenable dirty page tracking for the corresponding pages.
1285 * @kvm: pointer to kvm instance
1286 * @log: slot id and address from which to fetch the bitmap of dirty pages
1287 * @flush: true if TLB flush is needed by caller
1289 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1290 struct kvm_clear_dirty_log *log, bool *flush)
1292 struct kvm_memslots *slots;
1293 struct kvm_memory_slot *memslot;
1297 unsigned long *dirty_bitmap;
1298 unsigned long *dirty_bitmap_buffer;
1300 as_id = log->slot >> 16;
1301 id = (u16)log->slot;
1302 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1305 if (log->first_page & 63)
1308 slots = __kvm_memslots(kvm, as_id);
1309 memslot = id_to_memslot(slots, id);
1311 dirty_bitmap = memslot->dirty_bitmap;
1315 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1317 if (log->first_page > memslot->npages ||
1318 log->num_pages > memslot->npages - log->first_page ||
1319 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1323 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1324 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1327 spin_lock(&kvm->mmu_lock);
1328 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1329 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1330 i++, offset += BITS_PER_LONG) {
1331 unsigned long mask = *dirty_bitmap_buffer++;
1332 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1336 mask &= atomic_long_fetch_andnot(mask, p);
1339 * mask contains the bits that really have been cleared. This
1340 * never includes any bits beyond the length of the memslot (if
1341 * the length is not aligned to 64 pages), therefore it is not
1342 * a problem if userspace sets them in log->dirty_bitmap.
1346 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1350 spin_unlock(&kvm->mmu_lock);
1354 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1357 bool kvm_largepages_enabled(void)
1359 return largepages_enabled;
1362 void kvm_disable_largepages(void)
1364 largepages_enabled = false;
1366 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1368 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1370 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1372 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1374 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1376 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1379 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1381 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1383 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1384 memslot->flags & KVM_MEMSLOT_INVALID)
1389 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1391 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1393 struct vm_area_struct *vma;
1394 unsigned long addr, size;
1398 addr = gfn_to_hva(kvm, gfn);
1399 if (kvm_is_error_hva(addr))
1402 down_read(¤t->mm->mmap_sem);
1403 vma = find_vma(current->mm, addr);
1407 size = vma_kernel_pagesize(vma);
1410 up_read(¤t->mm->mmap_sem);
1415 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1417 return slot->flags & KVM_MEM_READONLY;
1420 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1421 gfn_t *nr_pages, bool write)
1423 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1424 return KVM_HVA_ERR_BAD;
1426 if (memslot_is_readonly(slot) && write)
1427 return KVM_HVA_ERR_RO_BAD;
1430 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1432 return __gfn_to_hva_memslot(slot, gfn);
1435 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1438 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1441 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1444 return gfn_to_hva_many(slot, gfn, NULL);
1446 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1448 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1450 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1452 EXPORT_SYMBOL_GPL(gfn_to_hva);
1454 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1456 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1458 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1461 * Return the hva of a @gfn and the R/W attribute if possible.
1463 * @slot: the kvm_memory_slot which contains @gfn
1464 * @gfn: the gfn to be translated
1465 * @writable: used to return the read/write attribute of the @slot if the hva
1466 * is valid and @writable is not NULL
1468 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1469 gfn_t gfn, bool *writable)
1471 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1473 if (!kvm_is_error_hva(hva) && writable)
1474 *writable = !memslot_is_readonly(slot);
1479 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1481 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1483 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1486 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1488 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1490 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1493 static inline int check_user_page_hwpoison(unsigned long addr)
1495 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1497 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1498 return rc == -EHWPOISON;
1502 * The fast path to get the writable pfn which will be stored in @pfn,
1503 * true indicates success, otherwise false is returned. It's also the
1504 * only part that runs if we can in atomic context.
1506 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1507 bool *writable, kvm_pfn_t *pfn)
1509 struct page *page[1];
1513 * Fast pin a writable pfn only if it is a write fault request
1514 * or the caller allows to map a writable pfn for a read fault
1517 if (!(write_fault || writable))
1520 npages = __get_user_pages_fast(addr, 1, 1, page);
1522 *pfn = page_to_pfn(page[0]);
1533 * The slow path to get the pfn of the specified host virtual address,
1534 * 1 indicates success, -errno is returned if error is detected.
1536 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1537 bool *writable, kvm_pfn_t *pfn)
1539 unsigned int flags = FOLL_HWPOISON;
1546 *writable = write_fault;
1549 flags |= FOLL_WRITE;
1551 flags |= FOLL_NOWAIT;
1553 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1557 /* map read fault as writable if possible */
1558 if (unlikely(!write_fault) && writable) {
1561 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1567 *pfn = page_to_pfn(page);
1571 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1573 if (unlikely(!(vma->vm_flags & VM_READ)))
1576 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1582 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1583 unsigned long addr, bool *async,
1584 bool write_fault, bool *writable,
1590 r = follow_pfn(vma, addr, &pfn);
1593 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1594 * not call the fault handler, so do it here.
1596 bool unlocked = false;
1597 r = fixup_user_fault(current, current->mm, addr,
1598 (write_fault ? FAULT_FLAG_WRITE : 0),
1605 r = follow_pfn(vma, addr, &pfn);
1615 * Get a reference here because callers of *hva_to_pfn* and
1616 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1617 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1618 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1619 * simply do nothing for reserved pfns.
1621 * Whoever called remap_pfn_range is also going to call e.g.
1622 * unmap_mapping_range before the underlying pages are freed,
1623 * causing a call to our MMU notifier.
1632 * Pin guest page in memory and return its pfn.
1633 * @addr: host virtual address which maps memory to the guest
1634 * @atomic: whether this function can sleep
1635 * @async: whether this function need to wait IO complete if the
1636 * host page is not in the memory
1637 * @write_fault: whether we should get a writable host page
1638 * @writable: whether it allows to map a writable host page for !@write_fault
1640 * The function will map a writable host page for these two cases:
1641 * 1): @write_fault = true
1642 * 2): @write_fault = false && @writable, @writable will tell the caller
1643 * whether the mapping is writable.
1645 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1646 bool write_fault, bool *writable)
1648 struct vm_area_struct *vma;
1652 /* we can do it either atomically or asynchronously, not both */
1653 BUG_ON(atomic && async);
1655 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1659 return KVM_PFN_ERR_FAULT;
1661 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1665 down_read(¤t->mm->mmap_sem);
1666 if (npages == -EHWPOISON ||
1667 (!async && check_user_page_hwpoison(addr))) {
1668 pfn = KVM_PFN_ERR_HWPOISON;
1673 vma = find_vma_intersection(current->mm, addr, addr + 1);
1676 pfn = KVM_PFN_ERR_FAULT;
1677 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1678 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1682 pfn = KVM_PFN_ERR_FAULT;
1684 if (async && vma_is_valid(vma, write_fault))
1686 pfn = KVM_PFN_ERR_FAULT;
1689 up_read(¤t->mm->mmap_sem);
1693 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1694 bool atomic, bool *async, bool write_fault,
1697 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1699 if (addr == KVM_HVA_ERR_RO_BAD) {
1702 return KVM_PFN_ERR_RO_FAULT;
1705 if (kvm_is_error_hva(addr)) {
1708 return KVM_PFN_NOSLOT;
1711 /* Do not map writable pfn in the readonly memslot. */
1712 if (writable && memslot_is_readonly(slot)) {
1717 return hva_to_pfn(addr, atomic, async, write_fault,
1720 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1722 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1725 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1726 write_fault, writable);
1728 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1730 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1732 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1734 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1736 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1738 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1740 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1742 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1744 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1746 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1748 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1750 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1752 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1754 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1756 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1758 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1760 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1762 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1764 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1766 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1767 struct page **pages, int nr_pages)
1772 addr = gfn_to_hva_many(slot, gfn, &entry);
1773 if (kvm_is_error_hva(addr))
1776 if (entry < nr_pages)
1779 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1781 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1783 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1785 if (is_error_noslot_pfn(pfn))
1786 return KVM_ERR_PTR_BAD_PAGE;
1788 if (kvm_is_reserved_pfn(pfn)) {
1790 return KVM_ERR_PTR_BAD_PAGE;
1793 return pfn_to_page(pfn);
1796 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1800 pfn = gfn_to_pfn(kvm, gfn);
1802 return kvm_pfn_to_page(pfn);
1804 EXPORT_SYMBOL_GPL(gfn_to_page);
1806 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1807 struct kvm_host_map *map)
1811 struct page *page = KVM_UNMAPPED_PAGE;
1816 pfn = gfn_to_pfn_memslot(slot, gfn);
1817 if (is_error_noslot_pfn(pfn))
1820 if (pfn_valid(pfn)) {
1821 page = pfn_to_page(pfn);
1823 #ifdef CONFIG_HAS_IOMEM
1825 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1840 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1842 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1844 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1846 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1855 if (map->page != KVM_UNMAPPED_PAGE)
1857 #ifdef CONFIG_HAS_IOMEM
1863 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1864 kvm_release_pfn_dirty(map->pfn);
1866 kvm_release_pfn_clean(map->pfn);
1872 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1874 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1878 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1880 return kvm_pfn_to_page(pfn);
1882 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1884 void kvm_release_page_clean(struct page *page)
1886 WARN_ON(is_error_page(page));
1888 kvm_release_pfn_clean(page_to_pfn(page));
1890 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1892 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1894 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1895 put_page(pfn_to_page(pfn));
1897 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1899 void kvm_release_page_dirty(struct page *page)
1901 WARN_ON(is_error_page(page));
1903 kvm_release_pfn_dirty(page_to_pfn(page));
1905 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1907 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1909 kvm_set_pfn_dirty(pfn);
1910 kvm_release_pfn_clean(pfn);
1912 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1914 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1916 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1917 SetPageDirty(pfn_to_page(pfn));
1919 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1921 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1923 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1924 mark_page_accessed(pfn_to_page(pfn));
1926 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1928 void kvm_get_pfn(kvm_pfn_t pfn)
1930 if (!kvm_is_reserved_pfn(pfn))
1931 get_page(pfn_to_page(pfn));
1933 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1935 static int next_segment(unsigned long len, int offset)
1937 if (len > PAGE_SIZE - offset)
1938 return PAGE_SIZE - offset;
1943 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1944 void *data, int offset, int len)
1949 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1950 if (kvm_is_error_hva(addr))
1952 r = __copy_from_user(data, (void __user *)addr + offset, len);
1958 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1961 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1963 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1965 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1967 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1968 int offset, int len)
1970 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1972 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1974 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1976 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1978 gfn_t gfn = gpa >> PAGE_SHIFT;
1980 int offset = offset_in_page(gpa);
1983 while ((seg = next_segment(len, offset)) != 0) {
1984 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1994 EXPORT_SYMBOL_GPL(kvm_read_guest);
1996 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1998 gfn_t gfn = gpa >> PAGE_SHIFT;
2000 int offset = offset_in_page(gpa);
2003 while ((seg = next_segment(len, offset)) != 0) {
2004 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2014 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2016 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2017 void *data, int offset, unsigned long len)
2022 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2023 if (kvm_is_error_hva(addr))
2025 pagefault_disable();
2026 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2033 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2034 void *data, unsigned long len)
2036 gfn_t gfn = gpa >> PAGE_SHIFT;
2037 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2038 int offset = offset_in_page(gpa);
2040 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2042 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2044 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2045 const void *data, int offset, int len)
2050 addr = gfn_to_hva_memslot(memslot, gfn);
2051 if (kvm_is_error_hva(addr))
2053 r = __copy_to_user((void __user *)addr + offset, data, len);
2056 mark_page_dirty_in_slot(memslot, gfn);
2060 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2061 const void *data, int offset, int len)
2063 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2065 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2067 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2069 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2070 const void *data, int offset, int len)
2072 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2074 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2076 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2078 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2081 gfn_t gfn = gpa >> PAGE_SHIFT;
2083 int offset = offset_in_page(gpa);
2086 while ((seg = next_segment(len, offset)) != 0) {
2087 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2097 EXPORT_SYMBOL_GPL(kvm_write_guest);
2099 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2102 gfn_t gfn = gpa >> PAGE_SHIFT;
2104 int offset = offset_in_page(gpa);
2107 while ((seg = next_segment(len, offset)) != 0) {
2108 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2118 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2120 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2121 struct gfn_to_hva_cache *ghc,
2122 gpa_t gpa, unsigned long len)
2124 int offset = offset_in_page(gpa);
2125 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2126 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2127 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2128 gfn_t nr_pages_avail;
2129 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2132 ghc->generation = slots->generation;
2134 ghc->hva = KVM_HVA_ERR_BAD;
2137 * If the requested region crosses two memslots, we still
2138 * verify that the entire region is valid here.
2140 while (!r && start_gfn <= end_gfn) {
2141 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2142 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2144 if (kvm_is_error_hva(ghc->hva))
2146 start_gfn += nr_pages_avail;
2149 /* Use the slow path for cross page reads and writes. */
2150 if (!r && nr_pages_needed == 1)
2153 ghc->memslot = NULL;
2158 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2159 gpa_t gpa, unsigned long len)
2161 struct kvm_memslots *slots = kvm_memslots(kvm);
2162 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2164 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2166 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2167 void *data, unsigned int offset,
2170 struct kvm_memslots *slots = kvm_memslots(kvm);
2172 gpa_t gpa = ghc->gpa + offset;
2174 BUG_ON(len + offset > ghc->len);
2176 if (slots->generation != ghc->generation)
2177 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2179 if (unlikely(!ghc->memslot))
2180 return kvm_write_guest(kvm, gpa, data, len);
2182 if (kvm_is_error_hva(ghc->hva))
2185 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2188 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2192 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2194 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2195 void *data, unsigned long len)
2197 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2199 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2201 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2202 void *data, unsigned long len)
2204 struct kvm_memslots *slots = kvm_memslots(kvm);
2207 BUG_ON(len > ghc->len);
2209 if (slots->generation != ghc->generation)
2210 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2212 if (unlikely(!ghc->memslot))
2213 return kvm_read_guest(kvm, ghc->gpa, data, len);
2215 if (kvm_is_error_hva(ghc->hva))
2218 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2224 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2226 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2228 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2230 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2232 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2234 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2236 gfn_t gfn = gpa >> PAGE_SHIFT;
2238 int offset = offset_in_page(gpa);
2241 while ((seg = next_segment(len, offset)) != 0) {
2242 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2251 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2253 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2256 if (memslot && memslot->dirty_bitmap) {
2257 unsigned long rel_gfn = gfn - memslot->base_gfn;
2259 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2263 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2265 struct kvm_memory_slot *memslot;
2267 memslot = gfn_to_memslot(kvm, gfn);
2268 mark_page_dirty_in_slot(memslot, gfn);
2270 EXPORT_SYMBOL_GPL(mark_page_dirty);
2272 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2274 struct kvm_memory_slot *memslot;
2276 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2277 mark_page_dirty_in_slot(memslot, gfn);
2279 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2281 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2283 if (!vcpu->sigset_active)
2287 * This does a lockless modification of ->real_blocked, which is fine
2288 * because, only current can change ->real_blocked and all readers of
2289 * ->real_blocked don't care as long ->real_blocked is always a subset
2292 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2295 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2297 if (!vcpu->sigset_active)
2300 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2301 sigemptyset(¤t->real_blocked);
2304 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2306 unsigned int old, val, grow, grow_start;
2308 old = val = vcpu->halt_poll_ns;
2309 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2310 grow = READ_ONCE(halt_poll_ns_grow);
2315 if (val < grow_start)
2318 if (val > halt_poll_ns)
2321 vcpu->halt_poll_ns = val;
2323 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2326 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2328 unsigned int old, val, shrink;
2330 old = val = vcpu->halt_poll_ns;
2331 shrink = READ_ONCE(halt_poll_ns_shrink);
2337 vcpu->halt_poll_ns = val;
2338 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2341 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2344 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2346 if (kvm_arch_vcpu_runnable(vcpu)) {
2347 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2350 if (kvm_cpu_has_pending_timer(vcpu))
2352 if (signal_pending(current))
2357 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2362 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2364 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2367 DECLARE_SWAITQUEUE(wait);
2368 bool waited = false;
2371 kvm_arch_vcpu_blocking(vcpu);
2373 start = cur = ktime_get();
2374 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2375 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2377 ++vcpu->stat.halt_attempted_poll;
2380 * This sets KVM_REQ_UNHALT if an interrupt
2383 if (kvm_vcpu_check_block(vcpu) < 0) {
2384 ++vcpu->stat.halt_successful_poll;
2385 if (!vcpu_valid_wakeup(vcpu))
2386 ++vcpu->stat.halt_poll_invalid;
2390 } while (single_task_running() && ktime_before(cur, stop));
2394 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2396 if (kvm_vcpu_check_block(vcpu) < 0)
2403 finish_swait(&vcpu->wq, &wait);
2406 kvm_arch_vcpu_unblocking(vcpu);
2407 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2409 if (!kvm_arch_no_poll(vcpu)) {
2410 if (!vcpu_valid_wakeup(vcpu)) {
2411 shrink_halt_poll_ns(vcpu);
2412 } else if (halt_poll_ns) {
2413 if (block_ns <= vcpu->halt_poll_ns)
2415 /* we had a long block, shrink polling */
2416 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2417 shrink_halt_poll_ns(vcpu);
2418 /* we had a short halt and our poll time is too small */
2419 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2420 block_ns < halt_poll_ns)
2421 grow_halt_poll_ns(vcpu);
2423 vcpu->halt_poll_ns = 0;
2427 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2428 kvm_arch_vcpu_block_finish(vcpu);
2430 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2432 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2434 struct swait_queue_head *wqp;
2436 wqp = kvm_arch_vcpu_wq(vcpu);
2437 if (swq_has_sleeper(wqp)) {
2439 WRITE_ONCE(vcpu->ready, true);
2440 ++vcpu->stat.halt_wakeup;
2446 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2450 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2452 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2455 int cpu = vcpu->cpu;
2457 if (kvm_vcpu_wake_up(vcpu))
2461 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2462 if (kvm_arch_vcpu_should_kick(vcpu))
2463 smp_send_reschedule(cpu);
2466 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2467 #endif /* !CONFIG_S390 */
2469 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2472 struct task_struct *task = NULL;
2476 pid = rcu_dereference(target->pid);
2478 task = get_pid_task(pid, PIDTYPE_PID);
2482 ret = yield_to(task, 1);
2483 put_task_struct(task);
2487 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2490 * Helper that checks whether a VCPU is eligible for directed yield.
2491 * Most eligible candidate to yield is decided by following heuristics:
2493 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2494 * (preempted lock holder), indicated by @in_spin_loop.
2495 * Set at the beiginning and cleared at the end of interception/PLE handler.
2497 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2498 * chance last time (mostly it has become eligible now since we have probably
2499 * yielded to lockholder in last iteration. This is done by toggling
2500 * @dy_eligible each time a VCPU checked for eligibility.)
2502 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2503 * to preempted lock-holder could result in wrong VCPU selection and CPU
2504 * burning. Giving priority for a potential lock-holder increases lock
2507 * Since algorithm is based on heuristics, accessing another VCPU data without
2508 * locking does not harm. It may result in trying to yield to same VCPU, fail
2509 * and continue with next VCPU and so on.
2511 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2513 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2516 eligible = !vcpu->spin_loop.in_spin_loop ||
2517 vcpu->spin_loop.dy_eligible;
2519 if (vcpu->spin_loop.in_spin_loop)
2520 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2529 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2530 * a vcpu_load/vcpu_put pair. However, for most architectures
2531 * kvm_arch_vcpu_runnable does not require vcpu_load.
2533 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2535 return kvm_arch_vcpu_runnable(vcpu);
2538 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2540 if (kvm_arch_dy_runnable(vcpu))
2543 #ifdef CONFIG_KVM_ASYNC_PF
2544 if (!list_empty_careful(&vcpu->async_pf.done))
2551 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2553 struct kvm *kvm = me->kvm;
2554 struct kvm_vcpu *vcpu;
2555 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2561 kvm_vcpu_set_in_spin_loop(me, true);
2563 * We boost the priority of a VCPU that is runnable but not
2564 * currently running, because it got preempted by something
2565 * else and called schedule in __vcpu_run. Hopefully that
2566 * VCPU is holding the lock that we need and will release it.
2567 * We approximate round-robin by starting at the last boosted VCPU.
2569 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2570 kvm_for_each_vcpu(i, vcpu, kvm) {
2571 if (!pass && i <= last_boosted_vcpu) {
2572 i = last_boosted_vcpu;
2574 } else if (pass && i > last_boosted_vcpu)
2576 if (!READ_ONCE(vcpu->ready))
2580 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2582 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2583 !kvm_arch_vcpu_in_kernel(vcpu))
2585 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2588 yielded = kvm_vcpu_yield_to(vcpu);
2590 kvm->last_boosted_vcpu = i;
2592 } else if (yielded < 0) {
2599 kvm_vcpu_set_in_spin_loop(me, false);
2601 /* Ensure vcpu is not eligible during next spinloop */
2602 kvm_vcpu_set_dy_eligible(me, false);
2604 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2606 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2608 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2611 if (vmf->pgoff == 0)
2612 page = virt_to_page(vcpu->run);
2614 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2615 page = virt_to_page(vcpu->arch.pio_data);
2617 #ifdef CONFIG_KVM_MMIO
2618 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2619 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2622 return kvm_arch_vcpu_fault(vcpu, vmf);
2628 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2629 .fault = kvm_vcpu_fault,
2632 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2634 vma->vm_ops = &kvm_vcpu_vm_ops;
2638 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2640 struct kvm_vcpu *vcpu = filp->private_data;
2642 debugfs_remove_recursive(vcpu->debugfs_dentry);
2643 kvm_put_kvm(vcpu->kvm);
2647 static struct file_operations kvm_vcpu_fops = {
2648 .release = kvm_vcpu_release,
2649 .unlocked_ioctl = kvm_vcpu_ioctl,
2650 .mmap = kvm_vcpu_mmap,
2651 .llseek = noop_llseek,
2652 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2656 * Allocates an inode for the vcpu.
2658 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2660 char name[8 + 1 + ITOA_MAX_LEN + 1];
2662 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2663 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2666 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2668 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2669 char dir_name[ITOA_MAX_LEN * 2];
2671 if (!debugfs_initialized())
2674 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2675 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2676 vcpu->kvm->debugfs_dentry);
2678 kvm_arch_create_vcpu_debugfs(vcpu);
2683 * Creates some virtual cpus. Good luck creating more than one.
2685 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2688 struct kvm_vcpu *vcpu;
2691 if (id >= KVM_MAX_VCPU_ID)
2694 mutex_lock(&kvm->lock);
2695 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2696 mutex_unlock(&kvm->lock);
2700 kvm->created_vcpus++;
2701 mutex_unlock(&kvm->lock);
2703 r = kvm_arch_vcpu_precreate(kvm, id);
2705 goto vcpu_decrement;
2707 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2710 goto vcpu_decrement;
2713 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2718 vcpu->run = page_address(page);
2720 kvm_vcpu_init(vcpu, kvm, id);
2722 r = kvm_arch_vcpu_create(vcpu);
2724 goto vcpu_free_run_page;
2726 kvm_create_vcpu_debugfs(vcpu);
2728 mutex_lock(&kvm->lock);
2729 if (kvm_get_vcpu_by_id(kvm, id)) {
2731 goto unlock_vcpu_destroy;
2734 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
2735 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
2737 /* Now it's all set up, let userspace reach it */
2739 r = create_vcpu_fd(vcpu);
2741 kvm_put_kvm_no_destroy(kvm);
2742 goto unlock_vcpu_destroy;
2745 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
2748 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2749 * before kvm->online_vcpu's incremented value.
2752 atomic_inc(&kvm->online_vcpus);
2754 mutex_unlock(&kvm->lock);
2755 kvm_arch_vcpu_postcreate(vcpu);
2758 unlock_vcpu_destroy:
2759 mutex_unlock(&kvm->lock);
2760 debugfs_remove_recursive(vcpu->debugfs_dentry);
2761 kvm_arch_vcpu_destroy(vcpu);
2763 free_page((unsigned long)vcpu->run);
2765 kmem_cache_free(kvm_vcpu_cache, vcpu);
2767 mutex_lock(&kvm->lock);
2768 kvm->created_vcpus--;
2769 mutex_unlock(&kvm->lock);
2773 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2776 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2777 vcpu->sigset_active = 1;
2778 vcpu->sigset = *sigset;
2780 vcpu->sigset_active = 0;
2784 static long kvm_vcpu_ioctl(struct file *filp,
2785 unsigned int ioctl, unsigned long arg)
2787 struct kvm_vcpu *vcpu = filp->private_data;
2788 void __user *argp = (void __user *)arg;
2790 struct kvm_fpu *fpu = NULL;
2791 struct kvm_sregs *kvm_sregs = NULL;
2793 if (vcpu->kvm->mm != current->mm)
2796 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2800 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2801 * execution; mutex_lock() would break them.
2803 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2804 if (r != -ENOIOCTLCMD)
2807 if (mutex_lock_killable(&vcpu->mutex))
2815 oldpid = rcu_access_pointer(vcpu->pid);
2816 if (unlikely(oldpid != task_pid(current))) {
2817 /* The thread running this VCPU changed. */
2820 r = kvm_arch_vcpu_run_pid_change(vcpu);
2824 newpid = get_task_pid(current, PIDTYPE_PID);
2825 rcu_assign_pointer(vcpu->pid, newpid);
2830 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2831 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2834 case KVM_GET_REGS: {
2835 struct kvm_regs *kvm_regs;
2838 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2841 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2845 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2852 case KVM_SET_REGS: {
2853 struct kvm_regs *kvm_regs;
2856 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2857 if (IS_ERR(kvm_regs)) {
2858 r = PTR_ERR(kvm_regs);
2861 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2865 case KVM_GET_SREGS: {
2866 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2867 GFP_KERNEL_ACCOUNT);
2871 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2875 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2880 case KVM_SET_SREGS: {
2881 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2882 if (IS_ERR(kvm_sregs)) {
2883 r = PTR_ERR(kvm_sregs);
2887 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2890 case KVM_GET_MP_STATE: {
2891 struct kvm_mp_state mp_state;
2893 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2897 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2902 case KVM_SET_MP_STATE: {
2903 struct kvm_mp_state mp_state;
2906 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2908 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2911 case KVM_TRANSLATE: {
2912 struct kvm_translation tr;
2915 if (copy_from_user(&tr, argp, sizeof(tr)))
2917 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2921 if (copy_to_user(argp, &tr, sizeof(tr)))
2926 case KVM_SET_GUEST_DEBUG: {
2927 struct kvm_guest_debug dbg;
2930 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2932 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2935 case KVM_SET_SIGNAL_MASK: {
2936 struct kvm_signal_mask __user *sigmask_arg = argp;
2937 struct kvm_signal_mask kvm_sigmask;
2938 sigset_t sigset, *p;
2943 if (copy_from_user(&kvm_sigmask, argp,
2944 sizeof(kvm_sigmask)))
2947 if (kvm_sigmask.len != sizeof(sigset))
2950 if (copy_from_user(&sigset, sigmask_arg->sigset,
2955 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2959 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2963 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2967 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2973 fpu = memdup_user(argp, sizeof(*fpu));
2979 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2983 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2986 mutex_unlock(&vcpu->mutex);
2992 #ifdef CONFIG_KVM_COMPAT
2993 static long kvm_vcpu_compat_ioctl(struct file *filp,
2994 unsigned int ioctl, unsigned long arg)
2996 struct kvm_vcpu *vcpu = filp->private_data;
2997 void __user *argp = compat_ptr(arg);
3000 if (vcpu->kvm->mm != current->mm)
3004 case KVM_SET_SIGNAL_MASK: {
3005 struct kvm_signal_mask __user *sigmask_arg = argp;
3006 struct kvm_signal_mask kvm_sigmask;
3011 if (copy_from_user(&kvm_sigmask, argp,
3012 sizeof(kvm_sigmask)))
3015 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3018 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3020 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3022 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3026 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3034 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3036 struct kvm_device *dev = filp->private_data;
3039 return dev->ops->mmap(dev, vma);
3044 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3045 int (*accessor)(struct kvm_device *dev,
3046 struct kvm_device_attr *attr),
3049 struct kvm_device_attr attr;
3054 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3057 return accessor(dev, &attr);
3060 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3063 struct kvm_device *dev = filp->private_data;
3065 if (dev->kvm->mm != current->mm)
3069 case KVM_SET_DEVICE_ATTR:
3070 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3071 case KVM_GET_DEVICE_ATTR:
3072 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3073 case KVM_HAS_DEVICE_ATTR:
3074 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3076 if (dev->ops->ioctl)
3077 return dev->ops->ioctl(dev, ioctl, arg);
3083 static int kvm_device_release(struct inode *inode, struct file *filp)
3085 struct kvm_device *dev = filp->private_data;
3086 struct kvm *kvm = dev->kvm;
3088 if (dev->ops->release) {
3089 mutex_lock(&kvm->lock);
3090 list_del(&dev->vm_node);
3091 dev->ops->release(dev);
3092 mutex_unlock(&kvm->lock);
3099 static const struct file_operations kvm_device_fops = {
3100 .unlocked_ioctl = kvm_device_ioctl,
3101 .release = kvm_device_release,
3102 KVM_COMPAT(kvm_device_ioctl),
3103 .mmap = kvm_device_mmap,
3106 struct kvm_device *kvm_device_from_filp(struct file *filp)
3108 if (filp->f_op != &kvm_device_fops)
3111 return filp->private_data;
3114 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3115 #ifdef CONFIG_KVM_MPIC
3116 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3117 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3121 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3123 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3126 if (kvm_device_ops_table[type] != NULL)
3129 kvm_device_ops_table[type] = ops;
3133 void kvm_unregister_device_ops(u32 type)
3135 if (kvm_device_ops_table[type] != NULL)
3136 kvm_device_ops_table[type] = NULL;
3139 static int kvm_ioctl_create_device(struct kvm *kvm,
3140 struct kvm_create_device *cd)
3142 const struct kvm_device_ops *ops = NULL;
3143 struct kvm_device *dev;
3144 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3148 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3151 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3152 ops = kvm_device_ops_table[type];
3159 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3166 mutex_lock(&kvm->lock);
3167 ret = ops->create(dev, type);
3169 mutex_unlock(&kvm->lock);
3173 list_add(&dev->vm_node, &kvm->devices);
3174 mutex_unlock(&kvm->lock);
3180 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3182 kvm_put_kvm_no_destroy(kvm);
3183 mutex_lock(&kvm->lock);
3184 list_del(&dev->vm_node);
3185 mutex_unlock(&kvm->lock);
3194 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3197 case KVM_CAP_USER_MEMORY:
3198 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3199 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3200 case KVM_CAP_INTERNAL_ERROR_DATA:
3201 #ifdef CONFIG_HAVE_KVM_MSI
3202 case KVM_CAP_SIGNAL_MSI:
3204 #ifdef CONFIG_HAVE_KVM_IRQFD
3206 case KVM_CAP_IRQFD_RESAMPLE:
3208 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3209 case KVM_CAP_CHECK_EXTENSION_VM:
3210 case KVM_CAP_ENABLE_CAP_VM:
3211 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3212 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3215 #ifdef CONFIG_KVM_MMIO
3216 case KVM_CAP_COALESCED_MMIO:
3217 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3218 case KVM_CAP_COALESCED_PIO:
3221 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3222 case KVM_CAP_IRQ_ROUTING:
3223 return KVM_MAX_IRQ_ROUTES;
3225 #if KVM_ADDRESS_SPACE_NUM > 1
3226 case KVM_CAP_MULTI_ADDRESS_SPACE:
3227 return KVM_ADDRESS_SPACE_NUM;
3229 case KVM_CAP_NR_MEMSLOTS:
3230 return KVM_USER_MEM_SLOTS;
3234 return kvm_vm_ioctl_check_extension(kvm, arg);
3237 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3238 struct kvm_enable_cap *cap)
3243 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3244 struct kvm_enable_cap *cap)
3247 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3248 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3249 if (cap->flags || (cap->args[0] & ~1))
3251 kvm->manual_dirty_log_protect = cap->args[0];
3255 return kvm_vm_ioctl_enable_cap(kvm, cap);
3259 static long kvm_vm_ioctl(struct file *filp,
3260 unsigned int ioctl, unsigned long arg)
3262 struct kvm *kvm = filp->private_data;
3263 void __user *argp = (void __user *)arg;
3266 if (kvm->mm != current->mm)
3269 case KVM_CREATE_VCPU:
3270 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3272 case KVM_ENABLE_CAP: {
3273 struct kvm_enable_cap cap;
3276 if (copy_from_user(&cap, argp, sizeof(cap)))
3278 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3281 case KVM_SET_USER_MEMORY_REGION: {
3282 struct kvm_userspace_memory_region kvm_userspace_mem;
3285 if (copy_from_user(&kvm_userspace_mem, argp,
3286 sizeof(kvm_userspace_mem)))
3289 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3292 case KVM_GET_DIRTY_LOG: {
3293 struct kvm_dirty_log log;
3296 if (copy_from_user(&log, argp, sizeof(log)))
3298 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3301 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3302 case KVM_CLEAR_DIRTY_LOG: {
3303 struct kvm_clear_dirty_log log;
3306 if (copy_from_user(&log, argp, sizeof(log)))
3308 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3312 #ifdef CONFIG_KVM_MMIO
3313 case KVM_REGISTER_COALESCED_MMIO: {
3314 struct kvm_coalesced_mmio_zone zone;
3317 if (copy_from_user(&zone, argp, sizeof(zone)))
3319 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3322 case KVM_UNREGISTER_COALESCED_MMIO: {
3323 struct kvm_coalesced_mmio_zone zone;
3326 if (copy_from_user(&zone, argp, sizeof(zone)))
3328 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3333 struct kvm_irqfd data;
3336 if (copy_from_user(&data, argp, sizeof(data)))
3338 r = kvm_irqfd(kvm, &data);
3341 case KVM_IOEVENTFD: {
3342 struct kvm_ioeventfd data;
3345 if (copy_from_user(&data, argp, sizeof(data)))
3347 r = kvm_ioeventfd(kvm, &data);
3350 #ifdef CONFIG_HAVE_KVM_MSI
3351 case KVM_SIGNAL_MSI: {
3355 if (copy_from_user(&msi, argp, sizeof(msi)))
3357 r = kvm_send_userspace_msi(kvm, &msi);
3361 #ifdef __KVM_HAVE_IRQ_LINE
3362 case KVM_IRQ_LINE_STATUS:
3363 case KVM_IRQ_LINE: {
3364 struct kvm_irq_level irq_event;
3367 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3370 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3371 ioctl == KVM_IRQ_LINE_STATUS);
3376 if (ioctl == KVM_IRQ_LINE_STATUS) {
3377 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3385 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3386 case KVM_SET_GSI_ROUTING: {
3387 struct kvm_irq_routing routing;
3388 struct kvm_irq_routing __user *urouting;
3389 struct kvm_irq_routing_entry *entries = NULL;
3392 if (copy_from_user(&routing, argp, sizeof(routing)))
3395 if (!kvm_arch_can_set_irq_routing(kvm))
3397 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3403 entries = vmalloc(array_size(sizeof(*entries),
3409 if (copy_from_user(entries, urouting->entries,
3410 routing.nr * sizeof(*entries)))
3411 goto out_free_irq_routing;
3413 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3415 out_free_irq_routing:
3419 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3420 case KVM_CREATE_DEVICE: {
3421 struct kvm_create_device cd;
3424 if (copy_from_user(&cd, argp, sizeof(cd)))
3427 r = kvm_ioctl_create_device(kvm, &cd);
3432 if (copy_to_user(argp, &cd, sizeof(cd)))
3438 case KVM_CHECK_EXTENSION:
3439 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3442 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3448 #ifdef CONFIG_KVM_COMPAT
3449 struct compat_kvm_dirty_log {
3453 compat_uptr_t dirty_bitmap; /* one bit per page */
3458 static long kvm_vm_compat_ioctl(struct file *filp,
3459 unsigned int ioctl, unsigned long arg)
3461 struct kvm *kvm = filp->private_data;
3464 if (kvm->mm != current->mm)
3467 case KVM_GET_DIRTY_LOG: {
3468 struct compat_kvm_dirty_log compat_log;
3469 struct kvm_dirty_log log;
3471 if (copy_from_user(&compat_log, (void __user *)arg,
3472 sizeof(compat_log)))
3474 log.slot = compat_log.slot;
3475 log.padding1 = compat_log.padding1;
3476 log.padding2 = compat_log.padding2;
3477 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3479 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3483 r = kvm_vm_ioctl(filp, ioctl, arg);
3489 static struct file_operations kvm_vm_fops = {
3490 .release = kvm_vm_release,
3491 .unlocked_ioctl = kvm_vm_ioctl,
3492 .llseek = noop_llseek,
3493 KVM_COMPAT(kvm_vm_compat_ioctl),
3496 static int kvm_dev_ioctl_create_vm(unsigned long type)
3502 kvm = kvm_create_vm(type);
3504 return PTR_ERR(kvm);
3505 #ifdef CONFIG_KVM_MMIO
3506 r = kvm_coalesced_mmio_init(kvm);
3510 r = get_unused_fd_flags(O_CLOEXEC);
3514 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3522 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3523 * already set, with ->release() being kvm_vm_release(). In error
3524 * cases it will be called by the final fput(file) and will take
3525 * care of doing kvm_put_kvm(kvm).
3527 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3532 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3534 fd_install(r, file);
3542 static long kvm_dev_ioctl(struct file *filp,
3543 unsigned int ioctl, unsigned long arg)
3548 case KVM_GET_API_VERSION:
3551 r = KVM_API_VERSION;
3554 r = kvm_dev_ioctl_create_vm(arg);
3556 case KVM_CHECK_EXTENSION:
3557 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3559 case KVM_GET_VCPU_MMAP_SIZE:
3562 r = PAGE_SIZE; /* struct kvm_run */
3564 r += PAGE_SIZE; /* pio data page */
3566 #ifdef CONFIG_KVM_MMIO
3567 r += PAGE_SIZE; /* coalesced mmio ring page */
3570 case KVM_TRACE_ENABLE:
3571 case KVM_TRACE_PAUSE:
3572 case KVM_TRACE_DISABLE:
3576 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3582 static struct file_operations kvm_chardev_ops = {
3583 .unlocked_ioctl = kvm_dev_ioctl,
3584 .llseek = noop_llseek,
3585 KVM_COMPAT(kvm_dev_ioctl),
3588 static struct miscdevice kvm_dev = {
3594 static void hardware_enable_nolock(void *junk)
3596 int cpu = raw_smp_processor_id();
3599 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3602 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3604 r = kvm_arch_hardware_enable();
3607 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3608 atomic_inc(&hardware_enable_failed);
3609 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3613 static int kvm_starting_cpu(unsigned int cpu)
3615 raw_spin_lock(&kvm_count_lock);
3616 if (kvm_usage_count)
3617 hardware_enable_nolock(NULL);
3618 raw_spin_unlock(&kvm_count_lock);
3622 static void hardware_disable_nolock(void *junk)
3624 int cpu = raw_smp_processor_id();
3626 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3628 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3629 kvm_arch_hardware_disable();
3632 static int kvm_dying_cpu(unsigned int cpu)
3634 raw_spin_lock(&kvm_count_lock);
3635 if (kvm_usage_count)
3636 hardware_disable_nolock(NULL);
3637 raw_spin_unlock(&kvm_count_lock);
3641 static void hardware_disable_all_nolock(void)
3643 BUG_ON(!kvm_usage_count);
3646 if (!kvm_usage_count)
3647 on_each_cpu(hardware_disable_nolock, NULL, 1);
3650 static void hardware_disable_all(void)
3652 raw_spin_lock(&kvm_count_lock);
3653 hardware_disable_all_nolock();
3654 raw_spin_unlock(&kvm_count_lock);
3657 static int hardware_enable_all(void)
3661 raw_spin_lock(&kvm_count_lock);
3664 if (kvm_usage_count == 1) {
3665 atomic_set(&hardware_enable_failed, 0);
3666 on_each_cpu(hardware_enable_nolock, NULL, 1);
3668 if (atomic_read(&hardware_enable_failed)) {
3669 hardware_disable_all_nolock();
3674 raw_spin_unlock(&kvm_count_lock);
3679 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3683 * Some (well, at least mine) BIOSes hang on reboot if
3686 * And Intel TXT required VMX off for all cpu when system shutdown.
3688 pr_info("kvm: exiting hardware virtualization\n");
3689 kvm_rebooting = true;
3690 on_each_cpu(hardware_disable_nolock, NULL, 1);
3694 static struct notifier_block kvm_reboot_notifier = {
3695 .notifier_call = kvm_reboot,
3699 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3703 for (i = 0; i < bus->dev_count; i++) {
3704 struct kvm_io_device *pos = bus->range[i].dev;
3706 kvm_iodevice_destructor(pos);
3711 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3712 const struct kvm_io_range *r2)
3714 gpa_t addr1 = r1->addr;
3715 gpa_t addr2 = r2->addr;
3720 /* If r2->len == 0, match the exact address. If r2->len != 0,
3721 * accept any overlapping write. Any order is acceptable for
3722 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3723 * we process all of them.
3736 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3738 return kvm_io_bus_cmp(p1, p2);
3741 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3742 gpa_t addr, int len)
3744 struct kvm_io_range *range, key;
3747 key = (struct kvm_io_range) {
3752 range = bsearch(&key, bus->range, bus->dev_count,
3753 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3757 off = range - bus->range;
3759 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3765 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3766 struct kvm_io_range *range, const void *val)
3770 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3774 while (idx < bus->dev_count &&
3775 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3776 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3785 /* kvm_io_bus_write - called under kvm->slots_lock */
3786 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3787 int len, const void *val)
3789 struct kvm_io_bus *bus;
3790 struct kvm_io_range range;
3793 range = (struct kvm_io_range) {
3798 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3801 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3802 return r < 0 ? r : 0;
3804 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3806 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3807 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3808 gpa_t addr, int len, const void *val, long cookie)
3810 struct kvm_io_bus *bus;
3811 struct kvm_io_range range;
3813 range = (struct kvm_io_range) {
3818 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3822 /* First try the device referenced by cookie. */
3823 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3824 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3825 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3830 * cookie contained garbage; fall back to search and return the
3831 * correct cookie value.
3833 return __kvm_io_bus_write(vcpu, bus, &range, val);
3836 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3837 struct kvm_io_range *range, void *val)
3841 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3845 while (idx < bus->dev_count &&
3846 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3847 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3856 /* kvm_io_bus_read - called under kvm->slots_lock */
3857 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3860 struct kvm_io_bus *bus;
3861 struct kvm_io_range range;
3864 range = (struct kvm_io_range) {
3869 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3872 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3873 return r < 0 ? r : 0;
3876 /* Caller must hold slots_lock. */
3877 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3878 int len, struct kvm_io_device *dev)
3881 struct kvm_io_bus *new_bus, *bus;
3882 struct kvm_io_range range;
3884 bus = kvm_get_bus(kvm, bus_idx);
3888 /* exclude ioeventfd which is limited by maximum fd */
3889 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3892 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3893 GFP_KERNEL_ACCOUNT);
3897 range = (struct kvm_io_range) {
3903 for (i = 0; i < bus->dev_count; i++)
3904 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3907 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3908 new_bus->dev_count++;
3909 new_bus->range[i] = range;
3910 memcpy(new_bus->range + i + 1, bus->range + i,
3911 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3912 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3913 synchronize_srcu_expedited(&kvm->srcu);
3919 /* Caller must hold slots_lock. */
3920 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3921 struct kvm_io_device *dev)
3924 struct kvm_io_bus *new_bus, *bus;
3926 bus = kvm_get_bus(kvm, bus_idx);
3930 for (i = 0; i < bus->dev_count; i++)
3931 if (bus->range[i].dev == dev) {
3935 if (i == bus->dev_count)
3938 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3939 GFP_KERNEL_ACCOUNT);
3941 pr_err("kvm: failed to shrink bus, removing it completely\n");
3945 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3946 new_bus->dev_count--;
3947 memcpy(new_bus->range + i, bus->range + i + 1,
3948 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3951 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3952 synchronize_srcu_expedited(&kvm->srcu);
3957 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3960 struct kvm_io_bus *bus;
3961 int dev_idx, srcu_idx;
3962 struct kvm_io_device *iodev = NULL;
3964 srcu_idx = srcu_read_lock(&kvm->srcu);
3966 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3970 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3974 iodev = bus->range[dev_idx].dev;
3977 srcu_read_unlock(&kvm->srcu, srcu_idx);
3981 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3983 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3984 int (*get)(void *, u64 *), int (*set)(void *, u64),
3987 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3990 /* The debugfs files are a reference to the kvm struct which
3991 * is still valid when kvm_destroy_vm is called.
3992 * To avoid the race between open and the removal of the debugfs
3993 * directory we test against the users count.
3995 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3998 if (simple_attr_open(inode, file, get,
3999 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4002 kvm_put_kvm(stat_data->kvm);
4009 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4011 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4014 simple_attr_release(inode, file);
4015 kvm_put_kvm(stat_data->kvm);
4020 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4022 *val = *(ulong *)((void *)kvm + offset);
4027 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4029 *(ulong *)((void *)kvm + offset) = 0;
4034 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4037 struct kvm_vcpu *vcpu;
4041 kvm_for_each_vcpu(i, vcpu, kvm)
4042 *val += *(u64 *)((void *)vcpu + offset);
4047 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4050 struct kvm_vcpu *vcpu;
4052 kvm_for_each_vcpu(i, vcpu, kvm)
4053 *(u64 *)((void *)vcpu + offset) = 0;
4058 static int kvm_stat_data_get(void *data, u64 *val)
4061 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4063 switch (stat_data->dbgfs_item->kind) {
4065 r = kvm_get_stat_per_vm(stat_data->kvm,
4066 stat_data->dbgfs_item->offset, val);
4069 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4070 stat_data->dbgfs_item->offset, val);
4077 static int kvm_stat_data_clear(void *data, u64 val)
4080 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4085 switch (stat_data->dbgfs_item->kind) {
4087 r = kvm_clear_stat_per_vm(stat_data->kvm,
4088 stat_data->dbgfs_item->offset);
4091 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4092 stat_data->dbgfs_item->offset);
4099 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4101 __simple_attr_check_format("%llu\n", 0ull);
4102 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4103 kvm_stat_data_clear, "%llu\n");
4106 static const struct file_operations stat_fops_per_vm = {
4107 .owner = THIS_MODULE,
4108 .open = kvm_stat_data_open,
4109 .release = kvm_debugfs_release,
4110 .read = simple_attr_read,
4111 .write = simple_attr_write,
4112 .llseek = no_llseek,
4115 static int vm_stat_get(void *_offset, u64 *val)
4117 unsigned offset = (long)_offset;
4122 mutex_lock(&kvm_lock);
4123 list_for_each_entry(kvm, &vm_list, vm_list) {
4124 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4127 mutex_unlock(&kvm_lock);
4131 static int vm_stat_clear(void *_offset, u64 val)
4133 unsigned offset = (long)_offset;
4139 mutex_lock(&kvm_lock);
4140 list_for_each_entry(kvm, &vm_list, vm_list) {
4141 kvm_clear_stat_per_vm(kvm, offset);
4143 mutex_unlock(&kvm_lock);
4148 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4150 static int vcpu_stat_get(void *_offset, u64 *val)
4152 unsigned offset = (long)_offset;
4157 mutex_lock(&kvm_lock);
4158 list_for_each_entry(kvm, &vm_list, vm_list) {
4159 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4162 mutex_unlock(&kvm_lock);
4166 static int vcpu_stat_clear(void *_offset, u64 val)
4168 unsigned offset = (long)_offset;
4174 mutex_lock(&kvm_lock);
4175 list_for_each_entry(kvm, &vm_list, vm_list) {
4176 kvm_clear_stat_per_vcpu(kvm, offset);
4178 mutex_unlock(&kvm_lock);
4183 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4186 static const struct file_operations *stat_fops[] = {
4187 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4188 [KVM_STAT_VM] = &vm_stat_fops,
4191 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4193 struct kobj_uevent_env *env;
4194 unsigned long long created, active;
4196 if (!kvm_dev.this_device || !kvm)
4199 mutex_lock(&kvm_lock);
4200 if (type == KVM_EVENT_CREATE_VM) {
4201 kvm_createvm_count++;
4203 } else if (type == KVM_EVENT_DESTROY_VM) {
4206 created = kvm_createvm_count;
4207 active = kvm_active_vms;
4208 mutex_unlock(&kvm_lock);
4210 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4214 add_uevent_var(env, "CREATED=%llu", created);
4215 add_uevent_var(env, "COUNT=%llu", active);
4217 if (type == KVM_EVENT_CREATE_VM) {
4218 add_uevent_var(env, "EVENT=create");
4219 kvm->userspace_pid = task_pid_nr(current);
4220 } else if (type == KVM_EVENT_DESTROY_VM) {
4221 add_uevent_var(env, "EVENT=destroy");
4223 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4225 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4226 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4229 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4231 add_uevent_var(env, "STATS_PATH=%s", tmp);
4235 /* no need for checks, since we are adding at most only 5 keys */
4236 env->envp[env->envp_idx++] = NULL;
4237 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4241 static void kvm_init_debug(void)
4243 struct kvm_stats_debugfs_item *p;
4245 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4247 kvm_debugfs_num_entries = 0;
4248 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4249 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4250 kvm_debugfs_dir, (void *)(long)p->offset,
4251 stat_fops[p->kind]);
4255 static int kvm_suspend(void)
4257 if (kvm_usage_count)
4258 hardware_disable_nolock(NULL);
4262 static void kvm_resume(void)
4264 if (kvm_usage_count) {
4265 #ifdef CONFIG_LOCKDEP
4266 WARN_ON(lockdep_is_held(&kvm_count_lock));
4268 hardware_enable_nolock(NULL);
4272 static struct syscore_ops kvm_syscore_ops = {
4273 .suspend = kvm_suspend,
4274 .resume = kvm_resume,
4278 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4280 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4283 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4285 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4287 WRITE_ONCE(vcpu->preempted, false);
4288 WRITE_ONCE(vcpu->ready, false);
4290 kvm_arch_sched_in(vcpu, cpu);
4292 kvm_arch_vcpu_load(vcpu, cpu);
4295 static void kvm_sched_out(struct preempt_notifier *pn,
4296 struct task_struct *next)
4298 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4300 if (current->state == TASK_RUNNING) {
4301 WRITE_ONCE(vcpu->preempted, true);
4302 WRITE_ONCE(vcpu->ready, true);
4304 kvm_arch_vcpu_put(vcpu);
4307 static void check_processor_compat(void *rtn)
4309 *(int *)rtn = kvm_arch_check_processor_compat();
4312 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4313 struct module *module)
4318 r = kvm_arch_init(opaque);
4323 * kvm_arch_init makes sure there's at most one caller
4324 * for architectures that support multiple implementations,
4325 * like intel and amd on x86.
4326 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4327 * conflicts in case kvm is already setup for another implementation.
4329 r = kvm_irqfd_init();
4333 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4338 r = kvm_arch_hardware_setup();
4342 for_each_online_cpu(cpu) {
4343 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4348 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4349 kvm_starting_cpu, kvm_dying_cpu);
4352 register_reboot_notifier(&kvm_reboot_notifier);
4354 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4356 vcpu_align = __alignof__(struct kvm_vcpu);
4358 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4360 offsetof(struct kvm_vcpu, arch),
4361 sizeof_field(struct kvm_vcpu, arch),
4363 if (!kvm_vcpu_cache) {
4368 r = kvm_async_pf_init();
4372 kvm_chardev_ops.owner = module;
4373 kvm_vm_fops.owner = module;
4374 kvm_vcpu_fops.owner = module;
4376 r = misc_register(&kvm_dev);
4378 pr_err("kvm: misc device register failed\n");
4382 register_syscore_ops(&kvm_syscore_ops);
4384 kvm_preempt_ops.sched_in = kvm_sched_in;
4385 kvm_preempt_ops.sched_out = kvm_sched_out;
4389 r = kvm_vfio_ops_init();
4395 kvm_async_pf_deinit();
4397 kmem_cache_destroy(kvm_vcpu_cache);
4399 unregister_reboot_notifier(&kvm_reboot_notifier);
4400 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4402 kvm_arch_hardware_unsetup();
4404 free_cpumask_var(cpus_hardware_enabled);
4412 EXPORT_SYMBOL_GPL(kvm_init);
4416 debugfs_remove_recursive(kvm_debugfs_dir);
4417 misc_deregister(&kvm_dev);
4418 kmem_cache_destroy(kvm_vcpu_cache);
4419 kvm_async_pf_deinit();
4420 unregister_syscore_ops(&kvm_syscore_ops);
4421 unregister_reboot_notifier(&kvm_reboot_notifier);
4422 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4423 on_each_cpu(hardware_disable_nolock, NULL, 1);
4424 kvm_arch_hardware_unsetup();
4427 free_cpumask_var(cpus_hardware_enabled);
4428 kvm_vfio_ops_exit();
4430 EXPORT_SYMBOL_GPL(kvm_exit);
4432 struct kvm_vm_worker_thread_context {
4434 struct task_struct *parent;
4435 struct completion init_done;
4436 kvm_vm_thread_fn_t thread_fn;
4441 static int kvm_vm_worker_thread(void *context)
4444 * The init_context is allocated on the stack of the parent thread, so
4445 * we have to locally copy anything that is needed beyond initialization
4447 struct kvm_vm_worker_thread_context *init_context = context;
4448 struct kvm *kvm = init_context->kvm;
4449 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4450 uintptr_t data = init_context->data;
4453 err = kthread_park(current);
4454 /* kthread_park(current) is never supposed to return an error */
4459 err = cgroup_attach_task_all(init_context->parent, current);
4461 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4466 set_user_nice(current, task_nice(init_context->parent));
4469 init_context->err = err;
4470 complete(&init_context->init_done);
4471 init_context = NULL;
4476 /* Wait to be woken up by the spawner before proceeding. */
4479 if (!kthread_should_stop())
4480 err = thread_fn(kvm, data);
4485 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4486 uintptr_t data, const char *name,
4487 struct task_struct **thread_ptr)
4489 struct kvm_vm_worker_thread_context init_context = {};
4490 struct task_struct *thread;
4493 init_context.kvm = kvm;
4494 init_context.parent = current;
4495 init_context.thread_fn = thread_fn;
4496 init_context.data = data;
4497 init_completion(&init_context.init_done);
4499 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4500 "%s-%d", name, task_pid_nr(current));
4502 return PTR_ERR(thread);
4504 /* kthread_run is never supposed to return NULL */
4505 WARN_ON(thread == NULL);
4507 wait_for_completion(&init_context.init_done);
4509 if (!init_context.err)
4510 *thread_ptr = thread;
4512 return init_context.err;