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 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
329 mutex_init(&vcpu->mutex);
334 init_swait_queue_head(&vcpu->wq);
335 kvm_async_pf_vcpu_init(vcpu);
338 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
340 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
345 vcpu->run = page_address(page);
347 kvm_vcpu_set_in_spin_loop(vcpu, false);
348 kvm_vcpu_set_dy_eligible(vcpu, false);
349 vcpu->preempted = false;
351 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
353 r = kvm_arch_vcpu_init(vcpu);
359 free_page((unsigned long)vcpu->run);
364 static void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
367 * no need for rcu_read_lock as VCPU_RUN is the only place that
368 * will change the vcpu->pid pointer and on uninit all file
369 * descriptors are already gone.
371 put_pid(rcu_dereference_protected(vcpu->pid, 1));
372 kvm_arch_vcpu_uninit(vcpu);
373 free_page((unsigned long)vcpu->run);
376 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
378 kvm_arch_vcpu_destroy(vcpu);
380 kvm_vcpu_uninit(vcpu);
381 kmem_cache_free(kvm_vcpu_cache, vcpu);
383 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
385 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
386 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
388 return container_of(mn, struct kvm, mmu_notifier);
391 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
392 struct mm_struct *mm,
393 unsigned long address,
396 struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 idx = srcu_read_lock(&kvm->srcu);
400 spin_lock(&kvm->mmu_lock);
401 kvm->mmu_notifier_seq++;
403 if (kvm_set_spte_hva(kvm, address, pte))
404 kvm_flush_remote_tlbs(kvm);
406 spin_unlock(&kvm->mmu_lock);
407 srcu_read_unlock(&kvm->srcu, idx);
410 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
411 const struct mmu_notifier_range *range)
413 struct kvm *kvm = mmu_notifier_to_kvm(mn);
414 int need_tlb_flush = 0, idx;
417 idx = srcu_read_lock(&kvm->srcu);
418 spin_lock(&kvm->mmu_lock);
420 * The count increase must become visible at unlock time as no
421 * spte can be established without taking the mmu_lock and
422 * count is also read inside the mmu_lock critical section.
424 kvm->mmu_notifier_count++;
425 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
426 need_tlb_flush |= kvm->tlbs_dirty;
427 /* we've to flush the tlb before the pages can be freed */
429 kvm_flush_remote_tlbs(kvm);
431 spin_unlock(&kvm->mmu_lock);
433 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
435 mmu_notifier_range_blockable(range));
437 srcu_read_unlock(&kvm->srcu, idx);
442 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
443 const struct mmu_notifier_range *range)
445 struct kvm *kvm = mmu_notifier_to_kvm(mn);
447 spin_lock(&kvm->mmu_lock);
449 * This sequence increase will notify the kvm page fault that
450 * the page that is going to be mapped in the spte could have
453 kvm->mmu_notifier_seq++;
456 * The above sequence increase must be visible before the
457 * below count decrease, which is ensured by the smp_wmb above
458 * in conjunction with the smp_rmb in mmu_notifier_retry().
460 kvm->mmu_notifier_count--;
461 spin_unlock(&kvm->mmu_lock);
463 BUG_ON(kvm->mmu_notifier_count < 0);
466 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
467 struct mm_struct *mm,
471 struct kvm *kvm = mmu_notifier_to_kvm(mn);
474 idx = srcu_read_lock(&kvm->srcu);
475 spin_lock(&kvm->mmu_lock);
477 young = kvm_age_hva(kvm, start, end);
479 kvm_flush_remote_tlbs(kvm);
481 spin_unlock(&kvm->mmu_lock);
482 srcu_read_unlock(&kvm->srcu, idx);
487 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
488 struct mm_struct *mm,
492 struct kvm *kvm = mmu_notifier_to_kvm(mn);
495 idx = srcu_read_lock(&kvm->srcu);
496 spin_lock(&kvm->mmu_lock);
498 * Even though we do not flush TLB, this will still adversely
499 * affect performance on pre-Haswell Intel EPT, where there is
500 * no EPT Access Bit to clear so that we have to tear down EPT
501 * tables instead. If we find this unacceptable, we can always
502 * add a parameter to kvm_age_hva so that it effectively doesn't
503 * do anything on clear_young.
505 * Also note that currently we never issue secondary TLB flushes
506 * from clear_young, leaving this job up to the regular system
507 * cadence. If we find this inaccurate, we might come up with a
508 * more sophisticated heuristic later.
510 young = kvm_age_hva(kvm, start, end);
511 spin_unlock(&kvm->mmu_lock);
512 srcu_read_unlock(&kvm->srcu, idx);
517 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
518 struct mm_struct *mm,
519 unsigned long address)
521 struct kvm *kvm = mmu_notifier_to_kvm(mn);
524 idx = srcu_read_lock(&kvm->srcu);
525 spin_lock(&kvm->mmu_lock);
526 young = kvm_test_age_hva(kvm, address);
527 spin_unlock(&kvm->mmu_lock);
528 srcu_read_unlock(&kvm->srcu, idx);
533 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
534 struct mm_struct *mm)
536 struct kvm *kvm = mmu_notifier_to_kvm(mn);
539 idx = srcu_read_lock(&kvm->srcu);
540 kvm_arch_flush_shadow_all(kvm);
541 srcu_read_unlock(&kvm->srcu, idx);
544 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
545 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
546 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
547 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
548 .clear_young = kvm_mmu_notifier_clear_young,
549 .test_young = kvm_mmu_notifier_test_young,
550 .change_pte = kvm_mmu_notifier_change_pte,
551 .release = kvm_mmu_notifier_release,
554 static int kvm_init_mmu_notifier(struct kvm *kvm)
556 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
557 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
560 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
562 static int kvm_init_mmu_notifier(struct kvm *kvm)
567 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
569 static struct kvm_memslots *kvm_alloc_memslots(void)
572 struct kvm_memslots *slots;
574 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
578 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
579 slots->id_to_index[i] = slots->memslots[i].id = i;
584 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
586 if (!memslot->dirty_bitmap)
589 kvfree(memslot->dirty_bitmap);
590 memslot->dirty_bitmap = NULL;
594 * Free any memory in @free but not in @dont.
596 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
597 struct kvm_memory_slot *dont)
599 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
600 kvm_destroy_dirty_bitmap(free);
602 kvm_arch_free_memslot(kvm, free, dont);
607 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
609 struct kvm_memory_slot *memslot;
614 kvm_for_each_memslot(memslot, slots)
615 kvm_free_memslot(kvm, memslot, NULL);
620 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
624 if (!kvm->debugfs_dentry)
627 debugfs_remove_recursive(kvm->debugfs_dentry);
629 if (kvm->debugfs_stat_data) {
630 for (i = 0; i < kvm_debugfs_num_entries; i++)
631 kfree(kvm->debugfs_stat_data[i]);
632 kfree(kvm->debugfs_stat_data);
636 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
638 char dir_name[ITOA_MAX_LEN * 2];
639 struct kvm_stat_data *stat_data;
640 struct kvm_stats_debugfs_item *p;
642 if (!debugfs_initialized())
645 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
646 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
648 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
649 sizeof(*kvm->debugfs_stat_data),
651 if (!kvm->debugfs_stat_data)
654 for (p = debugfs_entries; p->name; p++) {
655 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
659 stat_data->kvm = kvm;
660 stat_data->dbgfs_item = p;
661 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
662 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
663 kvm->debugfs_dentry, stat_data,
670 * Called after the VM is otherwise initialized, but just before adding it to
673 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
679 * Called just after removing the VM from the vm_list, but before doing any
682 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
686 static struct kvm *kvm_create_vm(unsigned long type)
688 struct kvm *kvm = kvm_arch_alloc_vm();
693 return ERR_PTR(-ENOMEM);
695 spin_lock_init(&kvm->mmu_lock);
697 kvm->mm = current->mm;
698 kvm_eventfd_init(kvm);
699 mutex_init(&kvm->lock);
700 mutex_init(&kvm->irq_lock);
701 mutex_init(&kvm->slots_lock);
702 INIT_LIST_HEAD(&kvm->devices);
704 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
706 if (init_srcu_struct(&kvm->srcu))
707 goto out_err_no_srcu;
708 if (init_srcu_struct(&kvm->irq_srcu))
709 goto out_err_no_irq_srcu;
711 refcount_set(&kvm->users_count, 1);
712 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
713 struct kvm_memslots *slots = kvm_alloc_memslots();
716 goto out_err_no_arch_destroy_vm;
717 /* Generations must be different for each address space. */
718 slots->generation = i;
719 rcu_assign_pointer(kvm->memslots[i], slots);
722 for (i = 0; i < KVM_NR_BUSES; i++) {
723 rcu_assign_pointer(kvm->buses[i],
724 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
726 goto out_err_no_arch_destroy_vm;
729 r = kvm_arch_init_vm(kvm, type);
731 goto out_err_no_arch_destroy_vm;
733 r = hardware_enable_all();
735 goto out_err_no_disable;
737 #ifdef CONFIG_HAVE_KVM_IRQFD
738 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
741 r = kvm_init_mmu_notifier(kvm);
743 goto out_err_no_mmu_notifier;
745 r = kvm_arch_post_init_vm(kvm);
749 mutex_lock(&kvm_lock);
750 list_add(&kvm->vm_list, &vm_list);
751 mutex_unlock(&kvm_lock);
753 preempt_notifier_inc();
758 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
759 if (kvm->mmu_notifier.ops)
760 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
762 out_err_no_mmu_notifier:
763 hardware_disable_all();
765 kvm_arch_destroy_vm(kvm);
766 out_err_no_arch_destroy_vm:
767 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
768 for (i = 0; i < KVM_NR_BUSES; i++)
769 kfree(kvm_get_bus(kvm, i));
770 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
771 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
772 cleanup_srcu_struct(&kvm->irq_srcu);
774 cleanup_srcu_struct(&kvm->srcu);
776 kvm_arch_free_vm(kvm);
781 static void kvm_destroy_devices(struct kvm *kvm)
783 struct kvm_device *dev, *tmp;
786 * We do not need to take the kvm->lock here, because nobody else
787 * has a reference to the struct kvm at this point and therefore
788 * cannot access the devices list anyhow.
790 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
791 list_del(&dev->vm_node);
792 dev->ops->destroy(dev);
796 static void kvm_destroy_vm(struct kvm *kvm)
799 struct mm_struct *mm = kvm->mm;
801 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
802 kvm_destroy_vm_debugfs(kvm);
803 kvm_arch_sync_events(kvm);
804 mutex_lock(&kvm_lock);
805 list_del(&kvm->vm_list);
806 mutex_unlock(&kvm_lock);
807 kvm_arch_pre_destroy_vm(kvm);
809 kvm_free_irq_routing(kvm);
810 for (i = 0; i < KVM_NR_BUSES; i++) {
811 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
814 kvm_io_bus_destroy(bus);
815 kvm->buses[i] = NULL;
817 kvm_coalesced_mmio_free(kvm);
818 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
819 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
821 kvm_arch_flush_shadow_all(kvm);
823 kvm_arch_destroy_vm(kvm);
824 kvm_destroy_devices(kvm);
825 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
826 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
827 cleanup_srcu_struct(&kvm->irq_srcu);
828 cleanup_srcu_struct(&kvm->srcu);
829 kvm_arch_free_vm(kvm);
830 preempt_notifier_dec();
831 hardware_disable_all();
835 void kvm_get_kvm(struct kvm *kvm)
837 refcount_inc(&kvm->users_count);
839 EXPORT_SYMBOL_GPL(kvm_get_kvm);
841 void kvm_put_kvm(struct kvm *kvm)
843 if (refcount_dec_and_test(&kvm->users_count))
846 EXPORT_SYMBOL_GPL(kvm_put_kvm);
849 * Used to put a reference that was taken on behalf of an object associated
850 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
851 * of the new file descriptor fails and the reference cannot be transferred to
852 * its final owner. In such cases, the caller is still actively using @kvm and
853 * will fail miserably if the refcount unexpectedly hits zero.
855 void kvm_put_kvm_no_destroy(struct kvm *kvm)
857 WARN_ON(refcount_dec_and_test(&kvm->users_count));
859 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
861 static int kvm_vm_release(struct inode *inode, struct file *filp)
863 struct kvm *kvm = filp->private_data;
865 kvm_irqfd_release(kvm);
872 * Allocation size is twice as large as the actual dirty bitmap size.
873 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
875 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
877 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
879 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
880 if (!memslot->dirty_bitmap)
887 * Insert memslot and re-sort memslots based on their GFN,
888 * so binary search could be used to lookup GFN.
889 * Sorting algorithm takes advantage of having initially
890 * sorted array and known changed memslot position.
892 static void update_memslots(struct kvm_memslots *slots,
893 struct kvm_memory_slot *new,
894 enum kvm_mr_change change)
897 int i = slots->id_to_index[id];
898 struct kvm_memory_slot *mslots = slots->memslots;
900 WARN_ON(mslots[i].id != id);
904 WARN_ON(mslots[i].npages || !new->npages);
908 WARN_ON(new->npages || !mslots[i].npages);
914 while (i < KVM_MEM_SLOTS_NUM - 1 &&
915 new->base_gfn <= mslots[i + 1].base_gfn) {
916 if (!mslots[i + 1].npages)
918 mslots[i] = mslots[i + 1];
919 slots->id_to_index[mslots[i].id] = i;
924 * The ">=" is needed when creating a slot with base_gfn == 0,
925 * so that it moves before all those with base_gfn == npages == 0.
927 * On the other hand, if new->npages is zero, the above loop has
928 * already left i pointing to the beginning of the empty part of
929 * mslots, and the ">=" would move the hole backwards in this
930 * case---which is wrong. So skip the loop when deleting a slot.
934 new->base_gfn >= mslots[i - 1].base_gfn) {
935 mslots[i] = mslots[i - 1];
936 slots->id_to_index[mslots[i].id] = i;
940 WARN_ON_ONCE(i != slots->used_slots);
943 slots->id_to_index[mslots[i].id] = i;
946 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
948 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
950 #ifdef __KVM_HAVE_READONLY_MEM
951 valid_flags |= KVM_MEM_READONLY;
954 if (mem->flags & ~valid_flags)
960 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
961 int as_id, struct kvm_memslots *slots)
963 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
964 u64 gen = old_memslots->generation;
966 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
967 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
969 rcu_assign_pointer(kvm->memslots[as_id], slots);
970 synchronize_srcu_expedited(&kvm->srcu);
973 * Increment the new memslot generation a second time, dropping the
974 * update in-progress flag and incrementing the generation based on
975 * the number of address spaces. This provides a unique and easily
976 * identifiable generation number while the memslots are in flux.
978 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
981 * Generations must be unique even across address spaces. We do not need
982 * a global counter for that, instead the generation space is evenly split
983 * across address spaces. For example, with two address spaces, address
984 * space 0 will use generations 0, 2, 4, ... while address space 1 will
985 * use generations 1, 3, 5, ...
987 gen += KVM_ADDRESS_SPACE_NUM;
989 kvm_arch_memslots_updated(kvm, gen);
991 slots->generation = gen;
997 * Allocate some memory and give it an address in the guest physical address
1000 * Discontiguous memory is allowed, mostly for framebuffers.
1002 * Must be called holding kvm->slots_lock for write.
1004 int __kvm_set_memory_region(struct kvm *kvm,
1005 const struct kvm_userspace_memory_region *mem)
1009 unsigned long npages;
1010 struct kvm_memory_slot *slot;
1011 struct kvm_memory_slot old, new;
1012 struct kvm_memslots *slots = NULL, *old_memslots;
1014 enum kvm_mr_change change;
1016 r = check_memory_region_flags(mem);
1021 as_id = mem->slot >> 16;
1022 id = (u16)mem->slot;
1024 /* General sanity checks */
1025 if (mem->memory_size & (PAGE_SIZE - 1))
1027 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1029 /* We can read the guest memory with __xxx_user() later on. */
1030 if ((id < KVM_USER_MEM_SLOTS) &&
1031 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1032 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1035 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1037 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1040 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1041 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1042 npages = mem->memory_size >> PAGE_SHIFT;
1044 if (npages > KVM_MEM_MAX_NR_PAGES)
1050 new.base_gfn = base_gfn;
1051 new.npages = npages;
1052 new.flags = mem->flags;
1056 change = KVM_MR_CREATE;
1057 else { /* Modify an existing slot. */
1058 if ((mem->userspace_addr != old.userspace_addr) ||
1059 (npages != old.npages) ||
1060 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1063 if (base_gfn != old.base_gfn)
1064 change = KVM_MR_MOVE;
1065 else if (new.flags != old.flags)
1066 change = KVM_MR_FLAGS_ONLY;
1067 else { /* Nothing to change. */
1076 change = KVM_MR_DELETE;
1081 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1082 /* Check for overlaps */
1084 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1087 if (!((base_gfn + npages <= slot->base_gfn) ||
1088 (base_gfn >= slot->base_gfn + slot->npages)))
1093 /* Free page dirty bitmap if unneeded */
1094 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1095 new.dirty_bitmap = NULL;
1098 if (change == KVM_MR_CREATE) {
1099 new.userspace_addr = mem->userspace_addr;
1101 if (kvm_arch_create_memslot(kvm, &new, npages))
1105 /* Allocate page dirty bitmap if needed */
1106 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1107 if (kvm_create_dirty_bitmap(&new) < 0)
1111 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1114 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1116 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1117 slot = id_to_memslot(slots, id);
1118 slot->flags |= KVM_MEMSLOT_INVALID;
1120 old_memslots = install_new_memslots(kvm, as_id, slots);
1122 /* From this point no new shadow pages pointing to a deleted,
1123 * or moved, memslot will be created.
1125 * validation of sp->gfn happens in:
1126 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1127 * - kvm_is_visible_gfn (mmu_check_root)
1129 kvm_arch_flush_shadow_memslot(kvm, slot);
1132 * We can re-use the old_memslots from above, the only difference
1133 * from the currently installed memslots is the invalid flag. This
1134 * will get overwritten by update_memslots anyway.
1136 slots = old_memslots;
1139 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1143 /* actual memory is freed via old in kvm_free_memslot below */
1144 if (change == KVM_MR_DELETE) {
1145 new.dirty_bitmap = NULL;
1146 memset(&new.arch, 0, sizeof(new.arch));
1149 update_memslots(slots, &new, change);
1150 old_memslots = install_new_memslots(kvm, as_id, slots);
1152 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1154 kvm_free_memslot(kvm, &old, &new);
1155 kvfree(old_memslots);
1161 kvm_free_memslot(kvm, &new, &old);
1165 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1167 int kvm_set_memory_region(struct kvm *kvm,
1168 const struct kvm_userspace_memory_region *mem)
1172 mutex_lock(&kvm->slots_lock);
1173 r = __kvm_set_memory_region(kvm, mem);
1174 mutex_unlock(&kvm->slots_lock);
1177 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1179 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1180 struct kvm_userspace_memory_region *mem)
1182 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1185 return kvm_set_memory_region(kvm, mem);
1188 int kvm_get_dirty_log(struct kvm *kvm,
1189 struct kvm_dirty_log *log, int *is_dirty)
1191 struct kvm_memslots *slots;
1192 struct kvm_memory_slot *memslot;
1195 unsigned long any = 0;
1197 as_id = log->slot >> 16;
1198 id = (u16)log->slot;
1199 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1202 slots = __kvm_memslots(kvm, as_id);
1203 memslot = id_to_memslot(slots, id);
1204 if (!memslot->dirty_bitmap)
1207 n = kvm_dirty_bitmap_bytes(memslot);
1209 for (i = 0; !any && i < n/sizeof(long); ++i)
1210 any = memslot->dirty_bitmap[i];
1212 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1219 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1221 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1223 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1224 * and reenable dirty page tracking for the corresponding pages.
1225 * @kvm: pointer to kvm instance
1226 * @log: slot id and address to which we copy the log
1227 * @flush: true if TLB flush is needed by caller
1229 * We need to keep it in mind that VCPU threads can write to the bitmap
1230 * concurrently. So, to avoid losing track of dirty pages we keep the
1233 * 1. Take a snapshot of the bit and clear it if needed.
1234 * 2. Write protect the corresponding page.
1235 * 3. Copy the snapshot to the userspace.
1236 * 4. Upon return caller flushes TLB's if needed.
1238 * Between 2 and 4, the guest may write to the page using the remaining TLB
1239 * entry. This is not a problem because the page is reported dirty using
1240 * the snapshot taken before and step 4 ensures that writes done after
1241 * exiting to userspace will be logged for the next call.
1244 int kvm_get_dirty_log_protect(struct kvm *kvm,
1245 struct kvm_dirty_log *log, bool *flush)
1247 struct kvm_memslots *slots;
1248 struct kvm_memory_slot *memslot;
1251 unsigned long *dirty_bitmap;
1252 unsigned long *dirty_bitmap_buffer;
1254 as_id = log->slot >> 16;
1255 id = (u16)log->slot;
1256 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1259 slots = __kvm_memslots(kvm, as_id);
1260 memslot = id_to_memslot(slots, id);
1262 dirty_bitmap = memslot->dirty_bitmap;
1266 n = kvm_dirty_bitmap_bytes(memslot);
1268 if (kvm->manual_dirty_log_protect) {
1270 * Unlike kvm_get_dirty_log, we always return false in *flush,
1271 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1272 * is some code duplication between this function and
1273 * kvm_get_dirty_log, but hopefully all architecture
1274 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1275 * can be eliminated.
1277 dirty_bitmap_buffer = dirty_bitmap;
1279 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1280 memset(dirty_bitmap_buffer, 0, n);
1282 spin_lock(&kvm->mmu_lock);
1283 for (i = 0; i < n / sizeof(long); i++) {
1287 if (!dirty_bitmap[i])
1291 mask = xchg(&dirty_bitmap[i], 0);
1292 dirty_bitmap_buffer[i] = mask;
1294 offset = i * BITS_PER_LONG;
1295 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1298 spin_unlock(&kvm->mmu_lock);
1301 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1305 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1308 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1309 * and reenable dirty page tracking for the corresponding pages.
1310 * @kvm: pointer to kvm instance
1311 * @log: slot id and address from which to fetch the bitmap of dirty pages
1312 * @flush: true if TLB flush is needed by caller
1314 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1315 struct kvm_clear_dirty_log *log, bool *flush)
1317 struct kvm_memslots *slots;
1318 struct kvm_memory_slot *memslot;
1322 unsigned long *dirty_bitmap;
1323 unsigned long *dirty_bitmap_buffer;
1325 as_id = log->slot >> 16;
1326 id = (u16)log->slot;
1327 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1330 if (log->first_page & 63)
1333 slots = __kvm_memslots(kvm, as_id);
1334 memslot = id_to_memslot(slots, id);
1336 dirty_bitmap = memslot->dirty_bitmap;
1340 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1342 if (log->first_page > memslot->npages ||
1343 log->num_pages > memslot->npages - log->first_page ||
1344 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1348 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1349 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1352 spin_lock(&kvm->mmu_lock);
1353 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1354 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1355 i++, offset += BITS_PER_LONG) {
1356 unsigned long mask = *dirty_bitmap_buffer++;
1357 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1361 mask &= atomic_long_fetch_andnot(mask, p);
1364 * mask contains the bits that really have been cleared. This
1365 * never includes any bits beyond the length of the memslot (if
1366 * the length is not aligned to 64 pages), therefore it is not
1367 * a problem if userspace sets them in log->dirty_bitmap.
1371 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1375 spin_unlock(&kvm->mmu_lock);
1379 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1382 bool kvm_largepages_enabled(void)
1384 return largepages_enabled;
1387 void kvm_disable_largepages(void)
1389 largepages_enabled = false;
1391 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1393 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1395 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1397 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1399 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1401 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1404 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1406 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1408 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1409 memslot->flags & KVM_MEMSLOT_INVALID)
1414 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1416 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1418 struct vm_area_struct *vma;
1419 unsigned long addr, size;
1423 addr = gfn_to_hva(kvm, gfn);
1424 if (kvm_is_error_hva(addr))
1427 down_read(¤t->mm->mmap_sem);
1428 vma = find_vma(current->mm, addr);
1432 size = vma_kernel_pagesize(vma);
1435 up_read(¤t->mm->mmap_sem);
1440 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1442 return slot->flags & KVM_MEM_READONLY;
1445 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1446 gfn_t *nr_pages, bool write)
1448 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1449 return KVM_HVA_ERR_BAD;
1451 if (memslot_is_readonly(slot) && write)
1452 return KVM_HVA_ERR_RO_BAD;
1455 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1457 return __gfn_to_hva_memslot(slot, gfn);
1460 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1463 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1466 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1469 return gfn_to_hva_many(slot, gfn, NULL);
1471 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1473 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1475 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1477 EXPORT_SYMBOL_GPL(gfn_to_hva);
1479 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1481 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1483 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1486 * Return the hva of a @gfn and the R/W attribute if possible.
1488 * @slot: the kvm_memory_slot which contains @gfn
1489 * @gfn: the gfn to be translated
1490 * @writable: used to return the read/write attribute of the @slot if the hva
1491 * is valid and @writable is not NULL
1493 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1494 gfn_t gfn, bool *writable)
1496 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1498 if (!kvm_is_error_hva(hva) && writable)
1499 *writable = !memslot_is_readonly(slot);
1504 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1506 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1508 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1511 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1513 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1515 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1518 static inline int check_user_page_hwpoison(unsigned long addr)
1520 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1522 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1523 return rc == -EHWPOISON;
1527 * The fast path to get the writable pfn which will be stored in @pfn,
1528 * true indicates success, otherwise false is returned. It's also the
1529 * only part that runs if we can in atomic context.
1531 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1532 bool *writable, kvm_pfn_t *pfn)
1534 struct page *page[1];
1538 * Fast pin a writable pfn only if it is a write fault request
1539 * or the caller allows to map a writable pfn for a read fault
1542 if (!(write_fault || writable))
1545 npages = __get_user_pages_fast(addr, 1, 1, page);
1547 *pfn = page_to_pfn(page[0]);
1558 * The slow path to get the pfn of the specified host virtual address,
1559 * 1 indicates success, -errno is returned if error is detected.
1561 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1562 bool *writable, kvm_pfn_t *pfn)
1564 unsigned int flags = FOLL_HWPOISON;
1571 *writable = write_fault;
1574 flags |= FOLL_WRITE;
1576 flags |= FOLL_NOWAIT;
1578 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1582 /* map read fault as writable if possible */
1583 if (unlikely(!write_fault) && writable) {
1586 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1592 *pfn = page_to_pfn(page);
1596 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1598 if (unlikely(!(vma->vm_flags & VM_READ)))
1601 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1607 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1608 unsigned long addr, bool *async,
1609 bool write_fault, bool *writable,
1615 r = follow_pfn(vma, addr, &pfn);
1618 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1619 * not call the fault handler, so do it here.
1621 bool unlocked = false;
1622 r = fixup_user_fault(current, current->mm, addr,
1623 (write_fault ? FAULT_FLAG_WRITE : 0),
1630 r = follow_pfn(vma, addr, &pfn);
1640 * Get a reference here because callers of *hva_to_pfn* and
1641 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1642 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1643 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1644 * simply do nothing for reserved pfns.
1646 * Whoever called remap_pfn_range is also going to call e.g.
1647 * unmap_mapping_range before the underlying pages are freed,
1648 * causing a call to our MMU notifier.
1657 * Pin guest page in memory and return its pfn.
1658 * @addr: host virtual address which maps memory to the guest
1659 * @atomic: whether this function can sleep
1660 * @async: whether this function need to wait IO complete if the
1661 * host page is not in the memory
1662 * @write_fault: whether we should get a writable host page
1663 * @writable: whether it allows to map a writable host page for !@write_fault
1665 * The function will map a writable host page for these two cases:
1666 * 1): @write_fault = true
1667 * 2): @write_fault = false && @writable, @writable will tell the caller
1668 * whether the mapping is writable.
1670 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1671 bool write_fault, bool *writable)
1673 struct vm_area_struct *vma;
1677 /* we can do it either atomically or asynchronously, not both */
1678 BUG_ON(atomic && async);
1680 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1684 return KVM_PFN_ERR_FAULT;
1686 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1690 down_read(¤t->mm->mmap_sem);
1691 if (npages == -EHWPOISON ||
1692 (!async && check_user_page_hwpoison(addr))) {
1693 pfn = KVM_PFN_ERR_HWPOISON;
1698 vma = find_vma_intersection(current->mm, addr, addr + 1);
1701 pfn = KVM_PFN_ERR_FAULT;
1702 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1703 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1707 pfn = KVM_PFN_ERR_FAULT;
1709 if (async && vma_is_valid(vma, write_fault))
1711 pfn = KVM_PFN_ERR_FAULT;
1714 up_read(¤t->mm->mmap_sem);
1718 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1719 bool atomic, bool *async, bool write_fault,
1722 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1724 if (addr == KVM_HVA_ERR_RO_BAD) {
1727 return KVM_PFN_ERR_RO_FAULT;
1730 if (kvm_is_error_hva(addr)) {
1733 return KVM_PFN_NOSLOT;
1736 /* Do not map writable pfn in the readonly memslot. */
1737 if (writable && memslot_is_readonly(slot)) {
1742 return hva_to_pfn(addr, atomic, async, write_fault,
1745 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1747 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1750 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1751 write_fault, writable);
1753 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1755 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1757 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1759 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1761 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1763 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1765 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1767 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1769 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1771 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1773 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1775 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1777 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1779 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1781 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1783 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1785 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1787 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1789 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1791 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1792 struct page **pages, int nr_pages)
1797 addr = gfn_to_hva_many(slot, gfn, &entry);
1798 if (kvm_is_error_hva(addr))
1801 if (entry < nr_pages)
1804 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1806 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1808 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1810 if (is_error_noslot_pfn(pfn))
1811 return KVM_ERR_PTR_BAD_PAGE;
1813 if (kvm_is_reserved_pfn(pfn)) {
1815 return KVM_ERR_PTR_BAD_PAGE;
1818 return pfn_to_page(pfn);
1821 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1825 pfn = gfn_to_pfn(kvm, gfn);
1827 return kvm_pfn_to_page(pfn);
1829 EXPORT_SYMBOL_GPL(gfn_to_page);
1831 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1832 struct kvm_host_map *map)
1836 struct page *page = KVM_UNMAPPED_PAGE;
1841 pfn = gfn_to_pfn_memslot(slot, gfn);
1842 if (is_error_noslot_pfn(pfn))
1845 if (pfn_valid(pfn)) {
1846 page = pfn_to_page(pfn);
1848 #ifdef CONFIG_HAS_IOMEM
1850 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1865 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1867 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1869 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1871 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1880 if (map->page != KVM_UNMAPPED_PAGE)
1882 #ifdef CONFIG_HAS_IOMEM
1888 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1889 kvm_release_pfn_dirty(map->pfn);
1891 kvm_release_pfn_clean(map->pfn);
1897 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1899 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1903 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1905 return kvm_pfn_to_page(pfn);
1907 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1909 void kvm_release_page_clean(struct page *page)
1911 WARN_ON(is_error_page(page));
1913 kvm_release_pfn_clean(page_to_pfn(page));
1915 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1917 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1919 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1920 put_page(pfn_to_page(pfn));
1922 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1924 void kvm_release_page_dirty(struct page *page)
1926 WARN_ON(is_error_page(page));
1928 kvm_release_pfn_dirty(page_to_pfn(page));
1930 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1932 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1934 kvm_set_pfn_dirty(pfn);
1935 kvm_release_pfn_clean(pfn);
1937 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1939 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1941 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1942 SetPageDirty(pfn_to_page(pfn));
1944 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1946 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1948 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1949 mark_page_accessed(pfn_to_page(pfn));
1951 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1953 void kvm_get_pfn(kvm_pfn_t pfn)
1955 if (!kvm_is_reserved_pfn(pfn))
1956 get_page(pfn_to_page(pfn));
1958 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1960 static int next_segment(unsigned long len, int offset)
1962 if (len > PAGE_SIZE - offset)
1963 return PAGE_SIZE - offset;
1968 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1969 void *data, int offset, int len)
1974 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1975 if (kvm_is_error_hva(addr))
1977 r = __copy_from_user(data, (void __user *)addr + offset, len);
1983 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1986 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1988 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1990 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1992 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1993 int offset, int len)
1995 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1997 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1999 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2001 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2003 gfn_t gfn = gpa >> PAGE_SHIFT;
2005 int offset = offset_in_page(gpa);
2008 while ((seg = next_segment(len, offset)) != 0) {
2009 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2019 EXPORT_SYMBOL_GPL(kvm_read_guest);
2021 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2023 gfn_t gfn = gpa >> PAGE_SHIFT;
2025 int offset = offset_in_page(gpa);
2028 while ((seg = next_segment(len, offset)) != 0) {
2029 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2039 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2041 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2042 void *data, int offset, unsigned long len)
2047 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2048 if (kvm_is_error_hva(addr))
2050 pagefault_disable();
2051 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2058 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
2061 gfn_t gfn = gpa >> PAGE_SHIFT;
2062 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2063 int offset = offset_in_page(gpa);
2065 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2067 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
2069 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2070 void *data, unsigned long len)
2072 gfn_t gfn = gpa >> PAGE_SHIFT;
2073 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2074 int offset = offset_in_page(gpa);
2076 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2078 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2080 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2081 const void *data, int offset, int len)
2086 addr = gfn_to_hva_memslot(memslot, gfn);
2087 if (kvm_is_error_hva(addr))
2089 r = __copy_to_user((void __user *)addr + offset, data, len);
2092 mark_page_dirty_in_slot(memslot, gfn);
2096 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2097 const void *data, int offset, int len)
2099 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2101 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2103 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2105 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2106 const void *data, int offset, int len)
2108 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2110 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2112 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2114 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2117 gfn_t gfn = gpa >> PAGE_SHIFT;
2119 int offset = offset_in_page(gpa);
2122 while ((seg = next_segment(len, offset)) != 0) {
2123 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2133 EXPORT_SYMBOL_GPL(kvm_write_guest);
2135 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2138 gfn_t gfn = gpa >> PAGE_SHIFT;
2140 int offset = offset_in_page(gpa);
2143 while ((seg = next_segment(len, offset)) != 0) {
2144 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2154 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2156 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2157 struct gfn_to_hva_cache *ghc,
2158 gpa_t gpa, unsigned long len)
2160 int offset = offset_in_page(gpa);
2161 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2162 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2163 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2164 gfn_t nr_pages_avail;
2165 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2168 ghc->generation = slots->generation;
2170 ghc->hva = KVM_HVA_ERR_BAD;
2173 * If the requested region crosses two memslots, we still
2174 * verify that the entire region is valid here.
2176 while (!r && start_gfn <= end_gfn) {
2177 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2178 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2180 if (kvm_is_error_hva(ghc->hva))
2182 start_gfn += nr_pages_avail;
2185 /* Use the slow path for cross page reads and writes. */
2186 if (!r && nr_pages_needed == 1)
2189 ghc->memslot = NULL;
2194 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2195 gpa_t gpa, unsigned long len)
2197 struct kvm_memslots *slots = kvm_memslots(kvm);
2198 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2200 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2202 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2203 void *data, unsigned int offset,
2206 struct kvm_memslots *slots = kvm_memslots(kvm);
2208 gpa_t gpa = ghc->gpa + offset;
2210 BUG_ON(len + offset > ghc->len);
2212 if (slots->generation != ghc->generation)
2213 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2215 if (unlikely(!ghc->memslot))
2216 return kvm_write_guest(kvm, gpa, data, len);
2218 if (kvm_is_error_hva(ghc->hva))
2221 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2224 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2228 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2230 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2231 void *data, unsigned long len)
2233 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2235 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2237 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2238 void *data, unsigned long len)
2240 struct kvm_memslots *slots = kvm_memslots(kvm);
2243 BUG_ON(len > ghc->len);
2245 if (slots->generation != ghc->generation)
2246 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2248 if (unlikely(!ghc->memslot))
2249 return kvm_read_guest(kvm, ghc->gpa, data, len);
2251 if (kvm_is_error_hva(ghc->hva))
2254 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2260 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2262 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2264 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2266 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2268 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2270 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2272 gfn_t gfn = gpa >> PAGE_SHIFT;
2274 int offset = offset_in_page(gpa);
2277 while ((seg = next_segment(len, offset)) != 0) {
2278 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2287 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2289 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2292 if (memslot && memslot->dirty_bitmap) {
2293 unsigned long rel_gfn = gfn - memslot->base_gfn;
2295 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2299 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2301 struct kvm_memory_slot *memslot;
2303 memslot = gfn_to_memslot(kvm, gfn);
2304 mark_page_dirty_in_slot(memslot, gfn);
2306 EXPORT_SYMBOL_GPL(mark_page_dirty);
2308 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2310 struct kvm_memory_slot *memslot;
2312 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2313 mark_page_dirty_in_slot(memslot, gfn);
2315 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2317 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2319 if (!vcpu->sigset_active)
2323 * This does a lockless modification of ->real_blocked, which is fine
2324 * because, only current can change ->real_blocked and all readers of
2325 * ->real_blocked don't care as long ->real_blocked is always a subset
2328 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2331 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2333 if (!vcpu->sigset_active)
2336 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2337 sigemptyset(¤t->real_blocked);
2340 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2342 unsigned int old, val, grow, grow_start;
2344 old = val = vcpu->halt_poll_ns;
2345 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2346 grow = READ_ONCE(halt_poll_ns_grow);
2351 if (val < grow_start)
2354 if (val > halt_poll_ns)
2357 vcpu->halt_poll_ns = val;
2359 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2362 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2364 unsigned int old, val, shrink;
2366 old = val = vcpu->halt_poll_ns;
2367 shrink = READ_ONCE(halt_poll_ns_shrink);
2373 vcpu->halt_poll_ns = val;
2374 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2377 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2380 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2382 if (kvm_arch_vcpu_runnable(vcpu)) {
2383 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2386 if (kvm_cpu_has_pending_timer(vcpu))
2388 if (signal_pending(current))
2393 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2398 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2400 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2403 DECLARE_SWAITQUEUE(wait);
2404 bool waited = false;
2407 kvm_arch_vcpu_blocking(vcpu);
2409 start = cur = ktime_get();
2410 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2411 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2413 ++vcpu->stat.halt_attempted_poll;
2416 * This sets KVM_REQ_UNHALT if an interrupt
2419 if (kvm_vcpu_check_block(vcpu) < 0) {
2420 ++vcpu->stat.halt_successful_poll;
2421 if (!vcpu_valid_wakeup(vcpu))
2422 ++vcpu->stat.halt_poll_invalid;
2426 } while (single_task_running() && ktime_before(cur, stop));
2430 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2432 if (kvm_vcpu_check_block(vcpu) < 0)
2439 finish_swait(&vcpu->wq, &wait);
2442 kvm_arch_vcpu_unblocking(vcpu);
2443 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2445 if (!kvm_arch_no_poll(vcpu)) {
2446 if (!vcpu_valid_wakeup(vcpu)) {
2447 shrink_halt_poll_ns(vcpu);
2448 } else if (halt_poll_ns) {
2449 if (block_ns <= vcpu->halt_poll_ns)
2451 /* we had a long block, shrink polling */
2452 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2453 shrink_halt_poll_ns(vcpu);
2454 /* we had a short halt and our poll time is too small */
2455 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2456 block_ns < halt_poll_ns)
2457 grow_halt_poll_ns(vcpu);
2459 vcpu->halt_poll_ns = 0;
2463 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2464 kvm_arch_vcpu_block_finish(vcpu);
2466 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2468 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2470 struct swait_queue_head *wqp;
2472 wqp = kvm_arch_vcpu_wq(vcpu);
2473 if (swq_has_sleeper(wqp)) {
2475 WRITE_ONCE(vcpu->ready, true);
2476 ++vcpu->stat.halt_wakeup;
2482 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2486 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2488 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2491 int cpu = vcpu->cpu;
2493 if (kvm_vcpu_wake_up(vcpu))
2497 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2498 if (kvm_arch_vcpu_should_kick(vcpu))
2499 smp_send_reschedule(cpu);
2502 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2503 #endif /* !CONFIG_S390 */
2505 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2508 struct task_struct *task = NULL;
2512 pid = rcu_dereference(target->pid);
2514 task = get_pid_task(pid, PIDTYPE_PID);
2518 ret = yield_to(task, 1);
2519 put_task_struct(task);
2523 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2526 * Helper that checks whether a VCPU is eligible for directed yield.
2527 * Most eligible candidate to yield is decided by following heuristics:
2529 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2530 * (preempted lock holder), indicated by @in_spin_loop.
2531 * Set at the beiginning and cleared at the end of interception/PLE handler.
2533 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2534 * chance last time (mostly it has become eligible now since we have probably
2535 * yielded to lockholder in last iteration. This is done by toggling
2536 * @dy_eligible each time a VCPU checked for eligibility.)
2538 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2539 * to preempted lock-holder could result in wrong VCPU selection and CPU
2540 * burning. Giving priority for a potential lock-holder increases lock
2543 * Since algorithm is based on heuristics, accessing another VCPU data without
2544 * locking does not harm. It may result in trying to yield to same VCPU, fail
2545 * and continue with next VCPU and so on.
2547 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2549 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2552 eligible = !vcpu->spin_loop.in_spin_loop ||
2553 vcpu->spin_loop.dy_eligible;
2555 if (vcpu->spin_loop.in_spin_loop)
2556 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2565 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2566 * a vcpu_load/vcpu_put pair. However, for most architectures
2567 * kvm_arch_vcpu_runnable does not require vcpu_load.
2569 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2571 return kvm_arch_vcpu_runnable(vcpu);
2574 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2576 if (kvm_arch_dy_runnable(vcpu))
2579 #ifdef CONFIG_KVM_ASYNC_PF
2580 if (!list_empty_careful(&vcpu->async_pf.done))
2587 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2589 struct kvm *kvm = me->kvm;
2590 struct kvm_vcpu *vcpu;
2591 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2597 kvm_vcpu_set_in_spin_loop(me, true);
2599 * We boost the priority of a VCPU that is runnable but not
2600 * currently running, because it got preempted by something
2601 * else and called schedule in __vcpu_run. Hopefully that
2602 * VCPU is holding the lock that we need and will release it.
2603 * We approximate round-robin by starting at the last boosted VCPU.
2605 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2606 kvm_for_each_vcpu(i, vcpu, kvm) {
2607 if (!pass && i <= last_boosted_vcpu) {
2608 i = last_boosted_vcpu;
2610 } else if (pass && i > last_boosted_vcpu)
2612 if (!READ_ONCE(vcpu->ready))
2616 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2618 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2619 !kvm_arch_vcpu_in_kernel(vcpu))
2621 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2624 yielded = kvm_vcpu_yield_to(vcpu);
2626 kvm->last_boosted_vcpu = i;
2628 } else if (yielded < 0) {
2635 kvm_vcpu_set_in_spin_loop(me, false);
2637 /* Ensure vcpu is not eligible during next spinloop */
2638 kvm_vcpu_set_dy_eligible(me, false);
2640 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2642 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2644 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2647 if (vmf->pgoff == 0)
2648 page = virt_to_page(vcpu->run);
2650 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2651 page = virt_to_page(vcpu->arch.pio_data);
2653 #ifdef CONFIG_KVM_MMIO
2654 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2655 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2658 return kvm_arch_vcpu_fault(vcpu, vmf);
2664 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2665 .fault = kvm_vcpu_fault,
2668 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2670 vma->vm_ops = &kvm_vcpu_vm_ops;
2674 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2676 struct kvm_vcpu *vcpu = filp->private_data;
2678 debugfs_remove_recursive(vcpu->debugfs_dentry);
2679 kvm_put_kvm(vcpu->kvm);
2683 static struct file_operations kvm_vcpu_fops = {
2684 .release = kvm_vcpu_release,
2685 .unlocked_ioctl = kvm_vcpu_ioctl,
2686 .mmap = kvm_vcpu_mmap,
2687 .llseek = noop_llseek,
2688 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2692 * Allocates an inode for the vcpu.
2694 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2696 char name[8 + 1 + ITOA_MAX_LEN + 1];
2698 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2699 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2702 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2704 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2705 char dir_name[ITOA_MAX_LEN * 2];
2707 if (!debugfs_initialized())
2710 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2711 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2712 vcpu->kvm->debugfs_dentry);
2714 kvm_arch_create_vcpu_debugfs(vcpu);
2719 * Creates some virtual cpus. Good luck creating more than one.
2721 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2724 struct kvm_vcpu *vcpu;
2726 if (id >= KVM_MAX_VCPU_ID)
2729 mutex_lock(&kvm->lock);
2730 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2731 mutex_unlock(&kvm->lock);
2735 kvm->created_vcpus++;
2736 mutex_unlock(&kvm->lock);
2738 r = kvm_arch_vcpu_precreate(kvm, id);
2740 goto vcpu_decrement;
2742 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2745 goto vcpu_decrement;
2748 r = kvm_vcpu_init(vcpu, kvm, id);
2752 r = kvm_arch_vcpu_create(vcpu);
2756 kvm_create_vcpu_debugfs(vcpu);
2758 mutex_lock(&kvm->lock);
2759 if (kvm_get_vcpu_by_id(kvm, id)) {
2761 goto unlock_vcpu_destroy;
2764 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
2765 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
2767 /* Now it's all set up, let userspace reach it */
2769 r = create_vcpu_fd(vcpu);
2771 kvm_put_kvm_no_destroy(kvm);
2772 goto unlock_vcpu_destroy;
2775 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
2778 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2779 * before kvm->online_vcpu's incremented value.
2782 atomic_inc(&kvm->online_vcpus);
2784 mutex_unlock(&kvm->lock);
2785 kvm_arch_vcpu_postcreate(vcpu);
2788 unlock_vcpu_destroy:
2789 mutex_unlock(&kvm->lock);
2790 debugfs_remove_recursive(vcpu->debugfs_dentry);
2791 kvm_arch_vcpu_destroy(vcpu);
2793 kvm_vcpu_uninit(vcpu);
2795 kmem_cache_free(kvm_vcpu_cache, vcpu);
2797 mutex_lock(&kvm->lock);
2798 kvm->created_vcpus--;
2799 mutex_unlock(&kvm->lock);
2803 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2806 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2807 vcpu->sigset_active = 1;
2808 vcpu->sigset = *sigset;
2810 vcpu->sigset_active = 0;
2814 static long kvm_vcpu_ioctl(struct file *filp,
2815 unsigned int ioctl, unsigned long arg)
2817 struct kvm_vcpu *vcpu = filp->private_data;
2818 void __user *argp = (void __user *)arg;
2820 struct kvm_fpu *fpu = NULL;
2821 struct kvm_sregs *kvm_sregs = NULL;
2823 if (vcpu->kvm->mm != current->mm)
2826 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2830 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2831 * execution; mutex_lock() would break them.
2833 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2834 if (r != -ENOIOCTLCMD)
2837 if (mutex_lock_killable(&vcpu->mutex))
2845 oldpid = rcu_access_pointer(vcpu->pid);
2846 if (unlikely(oldpid != task_pid(current))) {
2847 /* The thread running this VCPU changed. */
2850 r = kvm_arch_vcpu_run_pid_change(vcpu);
2854 newpid = get_task_pid(current, PIDTYPE_PID);
2855 rcu_assign_pointer(vcpu->pid, newpid);
2860 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2861 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2864 case KVM_GET_REGS: {
2865 struct kvm_regs *kvm_regs;
2868 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2871 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2875 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2882 case KVM_SET_REGS: {
2883 struct kvm_regs *kvm_regs;
2886 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2887 if (IS_ERR(kvm_regs)) {
2888 r = PTR_ERR(kvm_regs);
2891 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2895 case KVM_GET_SREGS: {
2896 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2897 GFP_KERNEL_ACCOUNT);
2901 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2905 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2910 case KVM_SET_SREGS: {
2911 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2912 if (IS_ERR(kvm_sregs)) {
2913 r = PTR_ERR(kvm_sregs);
2917 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2920 case KVM_GET_MP_STATE: {
2921 struct kvm_mp_state mp_state;
2923 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2927 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2932 case KVM_SET_MP_STATE: {
2933 struct kvm_mp_state mp_state;
2936 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2938 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2941 case KVM_TRANSLATE: {
2942 struct kvm_translation tr;
2945 if (copy_from_user(&tr, argp, sizeof(tr)))
2947 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2951 if (copy_to_user(argp, &tr, sizeof(tr)))
2956 case KVM_SET_GUEST_DEBUG: {
2957 struct kvm_guest_debug dbg;
2960 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2962 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2965 case KVM_SET_SIGNAL_MASK: {
2966 struct kvm_signal_mask __user *sigmask_arg = argp;
2967 struct kvm_signal_mask kvm_sigmask;
2968 sigset_t sigset, *p;
2973 if (copy_from_user(&kvm_sigmask, argp,
2974 sizeof(kvm_sigmask)))
2977 if (kvm_sigmask.len != sizeof(sigset))
2980 if (copy_from_user(&sigset, sigmask_arg->sigset,
2985 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2989 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2993 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2997 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3003 fpu = memdup_user(argp, sizeof(*fpu));
3009 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3013 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3016 mutex_unlock(&vcpu->mutex);
3022 #ifdef CONFIG_KVM_COMPAT
3023 static long kvm_vcpu_compat_ioctl(struct file *filp,
3024 unsigned int ioctl, unsigned long arg)
3026 struct kvm_vcpu *vcpu = filp->private_data;
3027 void __user *argp = compat_ptr(arg);
3030 if (vcpu->kvm->mm != current->mm)
3034 case KVM_SET_SIGNAL_MASK: {
3035 struct kvm_signal_mask __user *sigmask_arg = argp;
3036 struct kvm_signal_mask kvm_sigmask;
3041 if (copy_from_user(&kvm_sigmask, argp,
3042 sizeof(kvm_sigmask)))
3045 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3048 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3050 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3052 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3056 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3064 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3066 struct kvm_device *dev = filp->private_data;
3069 return dev->ops->mmap(dev, vma);
3074 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3075 int (*accessor)(struct kvm_device *dev,
3076 struct kvm_device_attr *attr),
3079 struct kvm_device_attr attr;
3084 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3087 return accessor(dev, &attr);
3090 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3093 struct kvm_device *dev = filp->private_data;
3095 if (dev->kvm->mm != current->mm)
3099 case KVM_SET_DEVICE_ATTR:
3100 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3101 case KVM_GET_DEVICE_ATTR:
3102 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3103 case KVM_HAS_DEVICE_ATTR:
3104 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3106 if (dev->ops->ioctl)
3107 return dev->ops->ioctl(dev, ioctl, arg);
3113 static int kvm_device_release(struct inode *inode, struct file *filp)
3115 struct kvm_device *dev = filp->private_data;
3116 struct kvm *kvm = dev->kvm;
3118 if (dev->ops->release) {
3119 mutex_lock(&kvm->lock);
3120 list_del(&dev->vm_node);
3121 dev->ops->release(dev);
3122 mutex_unlock(&kvm->lock);
3129 static const struct file_operations kvm_device_fops = {
3130 .unlocked_ioctl = kvm_device_ioctl,
3131 .release = kvm_device_release,
3132 KVM_COMPAT(kvm_device_ioctl),
3133 .mmap = kvm_device_mmap,
3136 struct kvm_device *kvm_device_from_filp(struct file *filp)
3138 if (filp->f_op != &kvm_device_fops)
3141 return filp->private_data;
3144 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3145 #ifdef CONFIG_KVM_MPIC
3146 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3147 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3151 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3153 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3156 if (kvm_device_ops_table[type] != NULL)
3159 kvm_device_ops_table[type] = ops;
3163 void kvm_unregister_device_ops(u32 type)
3165 if (kvm_device_ops_table[type] != NULL)
3166 kvm_device_ops_table[type] = NULL;
3169 static int kvm_ioctl_create_device(struct kvm *kvm,
3170 struct kvm_create_device *cd)
3172 const struct kvm_device_ops *ops = NULL;
3173 struct kvm_device *dev;
3174 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3178 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3181 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3182 ops = kvm_device_ops_table[type];
3189 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3196 mutex_lock(&kvm->lock);
3197 ret = ops->create(dev, type);
3199 mutex_unlock(&kvm->lock);
3203 list_add(&dev->vm_node, &kvm->devices);
3204 mutex_unlock(&kvm->lock);
3210 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3212 kvm_put_kvm_no_destroy(kvm);
3213 mutex_lock(&kvm->lock);
3214 list_del(&dev->vm_node);
3215 mutex_unlock(&kvm->lock);
3224 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3227 case KVM_CAP_USER_MEMORY:
3228 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3229 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3230 case KVM_CAP_INTERNAL_ERROR_DATA:
3231 #ifdef CONFIG_HAVE_KVM_MSI
3232 case KVM_CAP_SIGNAL_MSI:
3234 #ifdef CONFIG_HAVE_KVM_IRQFD
3236 case KVM_CAP_IRQFD_RESAMPLE:
3238 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3239 case KVM_CAP_CHECK_EXTENSION_VM:
3240 case KVM_CAP_ENABLE_CAP_VM:
3241 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3242 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3245 #ifdef CONFIG_KVM_MMIO
3246 case KVM_CAP_COALESCED_MMIO:
3247 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3248 case KVM_CAP_COALESCED_PIO:
3251 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3252 case KVM_CAP_IRQ_ROUTING:
3253 return KVM_MAX_IRQ_ROUTES;
3255 #if KVM_ADDRESS_SPACE_NUM > 1
3256 case KVM_CAP_MULTI_ADDRESS_SPACE:
3257 return KVM_ADDRESS_SPACE_NUM;
3259 case KVM_CAP_NR_MEMSLOTS:
3260 return KVM_USER_MEM_SLOTS;
3264 return kvm_vm_ioctl_check_extension(kvm, arg);
3267 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3268 struct kvm_enable_cap *cap)
3273 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3274 struct kvm_enable_cap *cap)
3277 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3278 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3279 if (cap->flags || (cap->args[0] & ~1))
3281 kvm->manual_dirty_log_protect = cap->args[0];
3285 return kvm_vm_ioctl_enable_cap(kvm, cap);
3289 static long kvm_vm_ioctl(struct file *filp,
3290 unsigned int ioctl, unsigned long arg)
3292 struct kvm *kvm = filp->private_data;
3293 void __user *argp = (void __user *)arg;
3296 if (kvm->mm != current->mm)
3299 case KVM_CREATE_VCPU:
3300 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3302 case KVM_ENABLE_CAP: {
3303 struct kvm_enable_cap cap;
3306 if (copy_from_user(&cap, argp, sizeof(cap)))
3308 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3311 case KVM_SET_USER_MEMORY_REGION: {
3312 struct kvm_userspace_memory_region kvm_userspace_mem;
3315 if (copy_from_user(&kvm_userspace_mem, argp,
3316 sizeof(kvm_userspace_mem)))
3319 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3322 case KVM_GET_DIRTY_LOG: {
3323 struct kvm_dirty_log log;
3326 if (copy_from_user(&log, argp, sizeof(log)))
3328 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3331 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3332 case KVM_CLEAR_DIRTY_LOG: {
3333 struct kvm_clear_dirty_log log;
3336 if (copy_from_user(&log, argp, sizeof(log)))
3338 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3342 #ifdef CONFIG_KVM_MMIO
3343 case KVM_REGISTER_COALESCED_MMIO: {
3344 struct kvm_coalesced_mmio_zone zone;
3347 if (copy_from_user(&zone, argp, sizeof(zone)))
3349 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3352 case KVM_UNREGISTER_COALESCED_MMIO: {
3353 struct kvm_coalesced_mmio_zone zone;
3356 if (copy_from_user(&zone, argp, sizeof(zone)))
3358 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3363 struct kvm_irqfd data;
3366 if (copy_from_user(&data, argp, sizeof(data)))
3368 r = kvm_irqfd(kvm, &data);
3371 case KVM_IOEVENTFD: {
3372 struct kvm_ioeventfd data;
3375 if (copy_from_user(&data, argp, sizeof(data)))
3377 r = kvm_ioeventfd(kvm, &data);
3380 #ifdef CONFIG_HAVE_KVM_MSI
3381 case KVM_SIGNAL_MSI: {
3385 if (copy_from_user(&msi, argp, sizeof(msi)))
3387 r = kvm_send_userspace_msi(kvm, &msi);
3391 #ifdef __KVM_HAVE_IRQ_LINE
3392 case KVM_IRQ_LINE_STATUS:
3393 case KVM_IRQ_LINE: {
3394 struct kvm_irq_level irq_event;
3397 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3400 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3401 ioctl == KVM_IRQ_LINE_STATUS);
3406 if (ioctl == KVM_IRQ_LINE_STATUS) {
3407 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3415 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3416 case KVM_SET_GSI_ROUTING: {
3417 struct kvm_irq_routing routing;
3418 struct kvm_irq_routing __user *urouting;
3419 struct kvm_irq_routing_entry *entries = NULL;
3422 if (copy_from_user(&routing, argp, sizeof(routing)))
3425 if (!kvm_arch_can_set_irq_routing(kvm))
3427 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3433 entries = vmalloc(array_size(sizeof(*entries),
3439 if (copy_from_user(entries, urouting->entries,
3440 routing.nr * sizeof(*entries)))
3441 goto out_free_irq_routing;
3443 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3445 out_free_irq_routing:
3449 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3450 case KVM_CREATE_DEVICE: {
3451 struct kvm_create_device cd;
3454 if (copy_from_user(&cd, argp, sizeof(cd)))
3457 r = kvm_ioctl_create_device(kvm, &cd);
3462 if (copy_to_user(argp, &cd, sizeof(cd)))
3468 case KVM_CHECK_EXTENSION:
3469 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3472 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3478 #ifdef CONFIG_KVM_COMPAT
3479 struct compat_kvm_dirty_log {
3483 compat_uptr_t dirty_bitmap; /* one bit per page */
3488 static long kvm_vm_compat_ioctl(struct file *filp,
3489 unsigned int ioctl, unsigned long arg)
3491 struct kvm *kvm = filp->private_data;
3494 if (kvm->mm != current->mm)
3497 case KVM_GET_DIRTY_LOG: {
3498 struct compat_kvm_dirty_log compat_log;
3499 struct kvm_dirty_log log;
3501 if (copy_from_user(&compat_log, (void __user *)arg,
3502 sizeof(compat_log)))
3504 log.slot = compat_log.slot;
3505 log.padding1 = compat_log.padding1;
3506 log.padding2 = compat_log.padding2;
3507 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3509 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3513 r = kvm_vm_ioctl(filp, ioctl, arg);
3519 static struct file_operations kvm_vm_fops = {
3520 .release = kvm_vm_release,
3521 .unlocked_ioctl = kvm_vm_ioctl,
3522 .llseek = noop_llseek,
3523 KVM_COMPAT(kvm_vm_compat_ioctl),
3526 static int kvm_dev_ioctl_create_vm(unsigned long type)
3532 kvm = kvm_create_vm(type);
3534 return PTR_ERR(kvm);
3535 #ifdef CONFIG_KVM_MMIO
3536 r = kvm_coalesced_mmio_init(kvm);
3540 r = get_unused_fd_flags(O_CLOEXEC);
3544 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3552 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3553 * already set, with ->release() being kvm_vm_release(). In error
3554 * cases it will be called by the final fput(file) and will take
3555 * care of doing kvm_put_kvm(kvm).
3557 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3562 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3564 fd_install(r, file);
3572 static long kvm_dev_ioctl(struct file *filp,
3573 unsigned int ioctl, unsigned long arg)
3578 case KVM_GET_API_VERSION:
3581 r = KVM_API_VERSION;
3584 r = kvm_dev_ioctl_create_vm(arg);
3586 case KVM_CHECK_EXTENSION:
3587 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3589 case KVM_GET_VCPU_MMAP_SIZE:
3592 r = PAGE_SIZE; /* struct kvm_run */
3594 r += PAGE_SIZE; /* pio data page */
3596 #ifdef CONFIG_KVM_MMIO
3597 r += PAGE_SIZE; /* coalesced mmio ring page */
3600 case KVM_TRACE_ENABLE:
3601 case KVM_TRACE_PAUSE:
3602 case KVM_TRACE_DISABLE:
3606 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3612 static struct file_operations kvm_chardev_ops = {
3613 .unlocked_ioctl = kvm_dev_ioctl,
3614 .llseek = noop_llseek,
3615 KVM_COMPAT(kvm_dev_ioctl),
3618 static struct miscdevice kvm_dev = {
3624 static void hardware_enable_nolock(void *junk)
3626 int cpu = raw_smp_processor_id();
3629 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3632 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3634 r = kvm_arch_hardware_enable();
3637 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3638 atomic_inc(&hardware_enable_failed);
3639 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3643 static int kvm_starting_cpu(unsigned int cpu)
3645 raw_spin_lock(&kvm_count_lock);
3646 if (kvm_usage_count)
3647 hardware_enable_nolock(NULL);
3648 raw_spin_unlock(&kvm_count_lock);
3652 static void hardware_disable_nolock(void *junk)
3654 int cpu = raw_smp_processor_id();
3656 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3658 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3659 kvm_arch_hardware_disable();
3662 static int kvm_dying_cpu(unsigned int cpu)
3664 raw_spin_lock(&kvm_count_lock);
3665 if (kvm_usage_count)
3666 hardware_disable_nolock(NULL);
3667 raw_spin_unlock(&kvm_count_lock);
3671 static void hardware_disable_all_nolock(void)
3673 BUG_ON(!kvm_usage_count);
3676 if (!kvm_usage_count)
3677 on_each_cpu(hardware_disable_nolock, NULL, 1);
3680 static void hardware_disable_all(void)
3682 raw_spin_lock(&kvm_count_lock);
3683 hardware_disable_all_nolock();
3684 raw_spin_unlock(&kvm_count_lock);
3687 static int hardware_enable_all(void)
3691 raw_spin_lock(&kvm_count_lock);
3694 if (kvm_usage_count == 1) {
3695 atomic_set(&hardware_enable_failed, 0);
3696 on_each_cpu(hardware_enable_nolock, NULL, 1);
3698 if (atomic_read(&hardware_enable_failed)) {
3699 hardware_disable_all_nolock();
3704 raw_spin_unlock(&kvm_count_lock);
3709 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3713 * Some (well, at least mine) BIOSes hang on reboot if
3716 * And Intel TXT required VMX off for all cpu when system shutdown.
3718 pr_info("kvm: exiting hardware virtualization\n");
3719 kvm_rebooting = true;
3720 on_each_cpu(hardware_disable_nolock, NULL, 1);
3724 static struct notifier_block kvm_reboot_notifier = {
3725 .notifier_call = kvm_reboot,
3729 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3733 for (i = 0; i < bus->dev_count; i++) {
3734 struct kvm_io_device *pos = bus->range[i].dev;
3736 kvm_iodevice_destructor(pos);
3741 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3742 const struct kvm_io_range *r2)
3744 gpa_t addr1 = r1->addr;
3745 gpa_t addr2 = r2->addr;
3750 /* If r2->len == 0, match the exact address. If r2->len != 0,
3751 * accept any overlapping write. Any order is acceptable for
3752 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3753 * we process all of them.
3766 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3768 return kvm_io_bus_cmp(p1, p2);
3771 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3772 gpa_t addr, int len)
3774 struct kvm_io_range *range, key;
3777 key = (struct kvm_io_range) {
3782 range = bsearch(&key, bus->range, bus->dev_count,
3783 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3787 off = range - bus->range;
3789 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3795 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3796 struct kvm_io_range *range, const void *val)
3800 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3804 while (idx < bus->dev_count &&
3805 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3806 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3815 /* kvm_io_bus_write - called under kvm->slots_lock */
3816 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3817 int len, const void *val)
3819 struct kvm_io_bus *bus;
3820 struct kvm_io_range range;
3823 range = (struct kvm_io_range) {
3828 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3831 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3832 return r < 0 ? r : 0;
3834 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3836 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3837 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3838 gpa_t addr, int len, const void *val, long cookie)
3840 struct kvm_io_bus *bus;
3841 struct kvm_io_range range;
3843 range = (struct kvm_io_range) {
3848 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3852 /* First try the device referenced by cookie. */
3853 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3854 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3855 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3860 * cookie contained garbage; fall back to search and return the
3861 * correct cookie value.
3863 return __kvm_io_bus_write(vcpu, bus, &range, val);
3866 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3867 struct kvm_io_range *range, void *val)
3871 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3875 while (idx < bus->dev_count &&
3876 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3877 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3886 /* kvm_io_bus_read - called under kvm->slots_lock */
3887 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3890 struct kvm_io_bus *bus;
3891 struct kvm_io_range range;
3894 range = (struct kvm_io_range) {
3899 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3902 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3903 return r < 0 ? r : 0;
3906 /* Caller must hold slots_lock. */
3907 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3908 int len, struct kvm_io_device *dev)
3911 struct kvm_io_bus *new_bus, *bus;
3912 struct kvm_io_range range;
3914 bus = kvm_get_bus(kvm, bus_idx);
3918 /* exclude ioeventfd which is limited by maximum fd */
3919 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3922 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3923 GFP_KERNEL_ACCOUNT);
3927 range = (struct kvm_io_range) {
3933 for (i = 0; i < bus->dev_count; i++)
3934 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3937 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3938 new_bus->dev_count++;
3939 new_bus->range[i] = range;
3940 memcpy(new_bus->range + i + 1, bus->range + i,
3941 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3942 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3943 synchronize_srcu_expedited(&kvm->srcu);
3949 /* Caller must hold slots_lock. */
3950 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3951 struct kvm_io_device *dev)
3954 struct kvm_io_bus *new_bus, *bus;
3956 bus = kvm_get_bus(kvm, bus_idx);
3960 for (i = 0; i < bus->dev_count; i++)
3961 if (bus->range[i].dev == dev) {
3965 if (i == bus->dev_count)
3968 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3969 GFP_KERNEL_ACCOUNT);
3971 pr_err("kvm: failed to shrink bus, removing it completely\n");
3975 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3976 new_bus->dev_count--;
3977 memcpy(new_bus->range + i, bus->range + i + 1,
3978 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3981 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3982 synchronize_srcu_expedited(&kvm->srcu);
3987 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3990 struct kvm_io_bus *bus;
3991 int dev_idx, srcu_idx;
3992 struct kvm_io_device *iodev = NULL;
3994 srcu_idx = srcu_read_lock(&kvm->srcu);
3996 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4000 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4004 iodev = bus->range[dev_idx].dev;
4007 srcu_read_unlock(&kvm->srcu, srcu_idx);
4011 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4013 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4014 int (*get)(void *, u64 *), int (*set)(void *, u64),
4017 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4020 /* The debugfs files are a reference to the kvm struct which
4021 * is still valid when kvm_destroy_vm is called.
4022 * To avoid the race between open and the removal of the debugfs
4023 * directory we test against the users count.
4025 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4028 if (simple_attr_open(inode, file, get,
4029 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4032 kvm_put_kvm(stat_data->kvm);
4039 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4041 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4044 simple_attr_release(inode, file);
4045 kvm_put_kvm(stat_data->kvm);
4050 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4052 *val = *(ulong *)((void *)kvm + offset);
4057 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4059 *(ulong *)((void *)kvm + offset) = 0;
4064 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4067 struct kvm_vcpu *vcpu;
4071 kvm_for_each_vcpu(i, vcpu, kvm)
4072 *val += *(u64 *)((void *)vcpu + offset);
4077 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4080 struct kvm_vcpu *vcpu;
4082 kvm_for_each_vcpu(i, vcpu, kvm)
4083 *(u64 *)((void *)vcpu + offset) = 0;
4088 static int kvm_stat_data_get(void *data, u64 *val)
4091 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4093 switch (stat_data->dbgfs_item->kind) {
4095 r = kvm_get_stat_per_vm(stat_data->kvm,
4096 stat_data->dbgfs_item->offset, val);
4099 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4100 stat_data->dbgfs_item->offset, val);
4107 static int kvm_stat_data_clear(void *data, u64 val)
4110 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4115 switch (stat_data->dbgfs_item->kind) {
4117 r = kvm_clear_stat_per_vm(stat_data->kvm,
4118 stat_data->dbgfs_item->offset);
4121 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4122 stat_data->dbgfs_item->offset);
4129 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4131 __simple_attr_check_format("%llu\n", 0ull);
4132 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4133 kvm_stat_data_clear, "%llu\n");
4136 static const struct file_operations stat_fops_per_vm = {
4137 .owner = THIS_MODULE,
4138 .open = kvm_stat_data_open,
4139 .release = kvm_debugfs_release,
4140 .read = simple_attr_read,
4141 .write = simple_attr_write,
4142 .llseek = no_llseek,
4145 static int vm_stat_get(void *_offset, u64 *val)
4147 unsigned offset = (long)_offset;
4152 mutex_lock(&kvm_lock);
4153 list_for_each_entry(kvm, &vm_list, vm_list) {
4154 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4157 mutex_unlock(&kvm_lock);
4161 static int vm_stat_clear(void *_offset, u64 val)
4163 unsigned offset = (long)_offset;
4169 mutex_lock(&kvm_lock);
4170 list_for_each_entry(kvm, &vm_list, vm_list) {
4171 kvm_clear_stat_per_vm(kvm, offset);
4173 mutex_unlock(&kvm_lock);
4178 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4180 static int vcpu_stat_get(void *_offset, u64 *val)
4182 unsigned offset = (long)_offset;
4187 mutex_lock(&kvm_lock);
4188 list_for_each_entry(kvm, &vm_list, vm_list) {
4189 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4192 mutex_unlock(&kvm_lock);
4196 static int vcpu_stat_clear(void *_offset, u64 val)
4198 unsigned offset = (long)_offset;
4204 mutex_lock(&kvm_lock);
4205 list_for_each_entry(kvm, &vm_list, vm_list) {
4206 kvm_clear_stat_per_vcpu(kvm, offset);
4208 mutex_unlock(&kvm_lock);
4213 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4216 static const struct file_operations *stat_fops[] = {
4217 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4218 [KVM_STAT_VM] = &vm_stat_fops,
4221 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4223 struct kobj_uevent_env *env;
4224 unsigned long long created, active;
4226 if (!kvm_dev.this_device || !kvm)
4229 mutex_lock(&kvm_lock);
4230 if (type == KVM_EVENT_CREATE_VM) {
4231 kvm_createvm_count++;
4233 } else if (type == KVM_EVENT_DESTROY_VM) {
4236 created = kvm_createvm_count;
4237 active = kvm_active_vms;
4238 mutex_unlock(&kvm_lock);
4240 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4244 add_uevent_var(env, "CREATED=%llu", created);
4245 add_uevent_var(env, "COUNT=%llu", active);
4247 if (type == KVM_EVENT_CREATE_VM) {
4248 add_uevent_var(env, "EVENT=create");
4249 kvm->userspace_pid = task_pid_nr(current);
4250 } else if (type == KVM_EVENT_DESTROY_VM) {
4251 add_uevent_var(env, "EVENT=destroy");
4253 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4255 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4256 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4259 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4261 add_uevent_var(env, "STATS_PATH=%s", tmp);
4265 /* no need for checks, since we are adding at most only 5 keys */
4266 env->envp[env->envp_idx++] = NULL;
4267 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4271 static void kvm_init_debug(void)
4273 struct kvm_stats_debugfs_item *p;
4275 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4277 kvm_debugfs_num_entries = 0;
4278 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4279 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4280 kvm_debugfs_dir, (void *)(long)p->offset,
4281 stat_fops[p->kind]);
4285 static int kvm_suspend(void)
4287 if (kvm_usage_count)
4288 hardware_disable_nolock(NULL);
4292 static void kvm_resume(void)
4294 if (kvm_usage_count) {
4295 #ifdef CONFIG_LOCKDEP
4296 WARN_ON(lockdep_is_held(&kvm_count_lock));
4298 hardware_enable_nolock(NULL);
4302 static struct syscore_ops kvm_syscore_ops = {
4303 .suspend = kvm_suspend,
4304 .resume = kvm_resume,
4308 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4310 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4313 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4315 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4317 WRITE_ONCE(vcpu->preempted, false);
4318 WRITE_ONCE(vcpu->ready, false);
4320 kvm_arch_sched_in(vcpu, cpu);
4322 kvm_arch_vcpu_load(vcpu, cpu);
4325 static void kvm_sched_out(struct preempt_notifier *pn,
4326 struct task_struct *next)
4328 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4330 if (current->state == TASK_RUNNING) {
4331 WRITE_ONCE(vcpu->preempted, true);
4332 WRITE_ONCE(vcpu->ready, true);
4334 kvm_arch_vcpu_put(vcpu);
4337 static void check_processor_compat(void *rtn)
4339 *(int *)rtn = kvm_arch_check_processor_compat();
4342 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4343 struct module *module)
4348 r = kvm_arch_init(opaque);
4353 * kvm_arch_init makes sure there's at most one caller
4354 * for architectures that support multiple implementations,
4355 * like intel and amd on x86.
4356 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4357 * conflicts in case kvm is already setup for another implementation.
4359 r = kvm_irqfd_init();
4363 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4368 r = kvm_arch_hardware_setup();
4372 for_each_online_cpu(cpu) {
4373 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4378 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4379 kvm_starting_cpu, kvm_dying_cpu);
4382 register_reboot_notifier(&kvm_reboot_notifier);
4384 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4386 vcpu_align = __alignof__(struct kvm_vcpu);
4388 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4390 offsetof(struct kvm_vcpu, arch),
4391 sizeof_field(struct kvm_vcpu, arch),
4393 if (!kvm_vcpu_cache) {
4398 r = kvm_async_pf_init();
4402 kvm_chardev_ops.owner = module;
4403 kvm_vm_fops.owner = module;
4404 kvm_vcpu_fops.owner = module;
4406 r = misc_register(&kvm_dev);
4408 pr_err("kvm: misc device register failed\n");
4412 register_syscore_ops(&kvm_syscore_ops);
4414 kvm_preempt_ops.sched_in = kvm_sched_in;
4415 kvm_preempt_ops.sched_out = kvm_sched_out;
4419 r = kvm_vfio_ops_init();
4425 kvm_async_pf_deinit();
4427 kmem_cache_destroy(kvm_vcpu_cache);
4429 unregister_reboot_notifier(&kvm_reboot_notifier);
4430 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4432 kvm_arch_hardware_unsetup();
4434 free_cpumask_var(cpus_hardware_enabled);
4442 EXPORT_SYMBOL_GPL(kvm_init);
4446 debugfs_remove_recursive(kvm_debugfs_dir);
4447 misc_deregister(&kvm_dev);
4448 kmem_cache_destroy(kvm_vcpu_cache);
4449 kvm_async_pf_deinit();
4450 unregister_syscore_ops(&kvm_syscore_ops);
4451 unregister_reboot_notifier(&kvm_reboot_notifier);
4452 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4453 on_each_cpu(hardware_disable_nolock, NULL, 1);
4454 kvm_arch_hardware_unsetup();
4457 free_cpumask_var(cpus_hardware_enabled);
4458 kvm_vfio_ops_exit();
4460 EXPORT_SYMBOL_GPL(kvm_exit);
4462 struct kvm_vm_worker_thread_context {
4464 struct task_struct *parent;
4465 struct completion init_done;
4466 kvm_vm_thread_fn_t thread_fn;
4471 static int kvm_vm_worker_thread(void *context)
4474 * The init_context is allocated on the stack of the parent thread, so
4475 * we have to locally copy anything that is needed beyond initialization
4477 struct kvm_vm_worker_thread_context *init_context = context;
4478 struct kvm *kvm = init_context->kvm;
4479 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4480 uintptr_t data = init_context->data;
4483 err = kthread_park(current);
4484 /* kthread_park(current) is never supposed to return an error */
4489 err = cgroup_attach_task_all(init_context->parent, current);
4491 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4496 set_user_nice(current, task_nice(init_context->parent));
4499 init_context->err = err;
4500 complete(&init_context->init_done);
4501 init_context = NULL;
4506 /* Wait to be woken up by the spawner before proceeding. */
4509 if (!kthread_should_stop())
4510 err = thread_fn(kvm, data);
4515 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4516 uintptr_t data, const char *name,
4517 struct task_struct **thread_ptr)
4519 struct kvm_vm_worker_thread_context init_context = {};
4520 struct task_struct *thread;
4523 init_context.kvm = kvm;
4524 init_context.parent = current;
4525 init_context.thread_fn = thread_fn;
4526 init_context.data = data;
4527 init_completion(&init_context.init_done);
4529 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4530 "%s-%d", name, task_pid_nr(current));
4532 return PTR_ERR(thread);
4534 /* kthread_run is never supposed to return NULL */
4535 WARN_ON(thread == NULL);
4537 wait_for_completion(&init_context.init_done);
4539 if (!init_context.err)
4540 *thread_ptr = thread;
4542 return init_context.err;